secondary.c revision 219864
1228753Smm/*- 2228753Smm * Copyright (c) 2009-2010 The FreeBSD Foundation 3228753Smm * Copyright (c) 2010 Pawel Jakub Dawidek <pjd@FreeBSD.org> 4228753Smm * All rights reserved. 5228753Smm * 6228753Smm * This software was developed by Pawel Jakub Dawidek under sponsorship from 7228753Smm * the FreeBSD Foundation. 8228753Smm * 9228753Smm * Redistribution and use in source and binary forms, with or without 10228753Smm * modification, are permitted provided that the following conditions 11228753Smm * are met: 12228753Smm * 1. Redistributions of source code must retain the above copyright 13228753Smm * notice, this list of conditions and the following disclaimer. 14228753Smm * 2. Redistributions in binary form must reproduce the above copyright 15228753Smm * notice, this list of conditions and the following disclaimer in the 16228753Smm * documentation and/or other materials provided with the distribution. 17228753Smm * 18228753Smm * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 19228753Smm * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20228753Smm * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21228753Smm * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 22228753Smm * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23228753Smm * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24228753Smm * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25228753Smm * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26228753Smm * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27229592Smm * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28228753Smm * SUCH DAMAGE. 29228753Smm */ 30228753Smm 31228753Smm#include <sys/cdefs.h> 32228753Smm__FBSDID("$FreeBSD: head/sbin/hastd/secondary.c 219864 2011-03-22 10:39:34Z pjd $"); 33228753Smm 34228753Smm#include <sys/param.h> 35228753Smm#include <sys/time.h> 36228753Smm#include <sys/bio.h> 37228753Smm#include <sys/disk.h> 38228753Smm#include <sys/stat.h> 39228753Smm 40228753Smm#include <err.h> 41228753Smm#include <errno.h> 42228753Smm#include <fcntl.h> 43228753Smm#include <libgeom.h> 44228753Smm#include <pthread.h> 45228753Smm#include <signal.h> 46228753Smm#include <stdint.h> 47228753Smm#include <stdio.h> 48228753Smm#include <string.h> 49228753Smm#include <sysexits.h> 50228753Smm#include <unistd.h> 51228753Smm 52228753Smm#include <activemap.h> 53228753Smm#include <nv.h> 54228753Smm#include <pjdlog.h> 55228753Smm 56228753Smm#include "control.h" 57228753Smm#include "event.h" 58228753Smm#include "hast.h" 59228753Smm#include "hast_proto.h" 60228753Smm#include "hastd.h" 61228753Smm#include "hooks.h" 62228753Smm#include "metadata.h" 63228753Smm#include "proto.h" 64228753Smm#include "subr.h" 65228753Smm#include "synch.h" 66228753Smm 67228753Smmstruct hio { 68228753Smm uint64_t hio_seq; 69228753Smm int hio_error; 70228753Smm struct nv *hio_nv; 71228753Smm void *hio_data; 72228753Smm uint8_t hio_cmd; 73228753Smm uint64_t hio_offset; 74228753Smm uint64_t hio_length; 75228753Smm TAILQ_ENTRY(hio) hio_next; 76228753Smm}; 77228753Smm 78228753Smmstatic struct hast_resource *gres; 79228753Smm 80228753Smm/* 81228753Smm * Free list holds unused structures. When free list is empty, we have to wait 82228753Smm * until some in-progress requests are freed. 83228753Smm */ 84228753Smmstatic TAILQ_HEAD(, hio) hio_free_list; 85228753Smmstatic pthread_mutex_t hio_free_list_lock; 86228753Smmstatic pthread_cond_t hio_free_list_cond; 87228753Smm/* 88228753Smm * Disk thread (the one that do I/O requests) takes requests from this list. 89228753Smm */ 90228753Smmstatic TAILQ_HEAD(, hio) hio_disk_list; 91228753Smmstatic pthread_mutex_t hio_disk_list_lock; 92228753Smmstatic pthread_cond_t hio_disk_list_cond; 93228753Smm/* 94228753Smm * There is one recv list for every component, although local components don't 95228753Smm * use recv lists as local requests are done synchronously. 96228753Smm */ 97228753Smmstatic TAILQ_HEAD(, hio) hio_send_list; 98228753Smmstatic pthread_mutex_t hio_send_list_lock; 99228753Smmstatic pthread_cond_t hio_send_list_cond; 100228753Smm 101228753Smm/* 102228753Smm * Maximum number of outstanding I/O requests. 103228753Smm */ 104228753Smm#define HAST_HIO_MAX 256 105228753Smm 106229592Smmstatic void *recv_thread(void *arg); 107228753Smmstatic void *disk_thread(void *arg); 108228753Smmstatic void *send_thread(void *arg); 109228753Smm 110228753Smm#define QUEUE_INSERT(name, hio) do { \ 111228753Smm bool _wakeup; \ 112228753Smm \ 113228753Smm mtx_lock(&hio_##name##_list_lock); \ 114228753Smm _wakeup = TAILQ_EMPTY(&hio_##name##_list); \ 115228753Smm TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_next); \ 116229592Smm mtx_unlock(&hio_##name##_list_lock); \ 117229592Smm if (_wakeup) \ 118229592Smm cv_signal(&hio_##name##_list_cond); \ 119229592Smm} while (0) 120228753Smm#define QUEUE_TAKE(name, hio) do { \ 121228753Smm mtx_lock(&hio_##name##_list_lock); \ 122229592Smm while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) { \ 123228753Smm cv_wait(&hio_##name##_list_cond, \ 124228753Smm &hio_##name##_list_lock); \ 125228753Smm } \ 126228753Smm TAILQ_REMOVE(&hio_##name##_list, (hio), hio_next); \ 127228753Smm mtx_unlock(&hio_##name##_list_lock); \ 128228753Smm} while (0) 129228753Smm 130228753Smmstatic void 131228753Smminit_environment(void) 132228753Smm{ 133228753Smm struct hio *hio; 134228753Smm unsigned int ii; 135228753Smm 136228753Smm /* 137228753Smm * Initialize lists, their locks and theirs condition variables. 138228753Smm */ 139228753Smm TAILQ_INIT(&hio_free_list); 140228753Smm mtx_init(&hio_free_list_lock); 141228753Smm cv_init(&hio_free_list_cond); 142228753Smm TAILQ_INIT(&hio_disk_list); 143228753Smm mtx_init(&hio_disk_list_lock); 144228753Smm cv_init(&hio_disk_list_cond); 145228753Smm TAILQ_INIT(&hio_send_list); 146228753Smm mtx_init(&hio_send_list_lock); 147228753Smm cv_init(&hio_send_list_cond); 148228753Smm 149228753Smm /* 150228753Smm * Allocate requests pool and initialize requests. 151228753Smm */ 152228753Smm for (ii = 0; ii < HAST_HIO_MAX; ii++) { 153228753Smm hio = malloc(sizeof(*hio)); 154228753Smm if (hio == NULL) { 155228753Smm pjdlog_exitx(EX_TEMPFAIL, 156228753Smm "Unable to allocate memory (%zu bytes) for hio request.", 157228753Smm sizeof(*hio)); 158228753Smm } 159228753Smm hio->hio_error = 0; 160228753Smm hio->hio_data = malloc(MAXPHYS); 161228753Smm if (hio->hio_data == NULL) { 162228753Smm pjdlog_exitx(EX_TEMPFAIL, 163228753Smm "Unable to allocate memory (%zu bytes) for gctl_data.", 164228753Smm (size_t)MAXPHYS); 165228753Smm } 166228753Smm TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_next); 167228753Smm } 168228753Smm} 169228753Smm 170228753Smmstatic void 171228753Smminit_local(struct hast_resource *res) 172228753Smm{ 173228753Smm 174228753Smm if (metadata_read(res, true) < 0) 175228753Smm exit(EX_NOINPUT); 176228753Smm} 177228753Smm 178228753Smmstatic void 179228753Smminit_remote(struct hast_resource *res, struct nv *nvin) 180228753Smm{ 181228753Smm uint64_t resuid; 182228753Smm struct nv *nvout; 183228753Smm unsigned char *map; 184228753Smm size_t mapsize; 185228753Smm 186228753Smm map = NULL; 187228753Smm mapsize = 0; 188228753Smm nvout = nv_alloc(); 189228753Smm nv_add_int64(nvout, (int64_t)res->hr_datasize, "datasize"); 190228753Smm nv_add_int32(nvout, (int32_t)res->hr_extentsize, "extentsize"); 191228753Smm resuid = nv_get_uint64(nvin, "resuid"); 192228753Smm res->hr_primary_localcnt = nv_get_uint64(nvin, "localcnt"); 193228753Smm res->hr_primary_remotecnt = nv_get_uint64(nvin, "remotecnt"); 194228753Smm nv_add_uint64(nvout, res->hr_secondary_localcnt, "localcnt"); 195228753Smm nv_add_uint64(nvout, res->hr_secondary_remotecnt, "remotecnt"); 196228753Smm mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 197228753Smm METADATA_SIZE, res->hr_extentsize, res->hr_local_sectorsize); 198228753Smm map = malloc(mapsize); 199228753Smm if (map == NULL) { 200228753Smm pjdlog_exitx(EX_TEMPFAIL, 201228753Smm "Unable to allocate memory (%zu bytes) for activemap.", 202228753Smm mapsize); 203228753Smm } 204228753Smm nv_add_uint32(nvout, (uint32_t)mapsize, "mapsize"); 205228753Smm /* 206228753Smm * When we work as primary and secondary is missing we will increase 207228753Smm * localcnt in our metadata. When secondary is connected and synced 208228753Smm * we make localcnt be equal to remotecnt, which means nodes are more 209228753Smm * or less in sync. 210228753Smm * Split-brain condition is when both nodes are not able to communicate 211228753Smm * and are both configured as primary nodes. In turn, they can both 212228753Smm * make incompatible changes to the data and we have to detect that. 213228753Smm * Under split-brain condition we will increase our localcnt on first 214228753Smm * write and remote node will increase its localcnt on first write. 215228753Smm * When we connect we can see that primary's localcnt is greater than 216228753Smm * our remotecnt (primary was modified while we weren't watching) and 217228753Smm * our localcnt is greater than primary's remotecnt (we were modified 218228753Smm * while primary wasn't watching). 219228753Smm * There are many possible combinations which are all gathered below. 220228753Smm * Don't pay too much attention to exact numbers, the more important 221228753Smm * is to compare them. We compare secondary's local with primary's 222229592Smm * remote and secondary's remote with primary's local. 223228753Smm * Note that every case where primary's localcnt is smaller than 224228753Smm * secondary's remotecnt and where secondary's localcnt is smaller than 225228753Smm * primary's remotecnt should be impossible in practise. We will perform 226228753Smm * full synchronization then. Those cases are marked with an asterisk. 227228753Smm * Regular synchronization means that only extents marked as dirty are 228228753Smm * synchronized (regular synchronization). 229228753Smm * 230228753Smm * SECONDARY METADATA PRIMARY METADATA 231228753Smm * local=3 remote=3 local=2 remote=2* ?! Full sync from secondary. 232228753Smm * local=3 remote=3 local=2 remote=3* ?! Full sync from primary. 233228753Smm * local=3 remote=3 local=2 remote=4* ?! Full sync from primary. 234228753Smm * local=3 remote=3 local=3 remote=2 Primary is out-of-date, 235228753Smm * regular sync from secondary. 236228753Smm * local=3 remote=3 local=3 remote=3 Regular sync just in case. 237228753Smm * local=3 remote=3 local=3 remote=4* ?! Full sync from primary. 238228753Smm * local=3 remote=3 local=4 remote=2 Split-brain condition. 239228753Smm * local=3 remote=3 local=4 remote=3 Secondary out-of-date, 240228753Smm * regular sync from primary. 241228753Smm * local=3 remote=3 local=4 remote=4* ?! Full sync from primary. 242228753Smm */ 243228753Smm if (res->hr_resuid == 0) { 244228753Smm /* 245228753Smm * Provider is used for the first time. If primary node done no 246228753Smm * writes yet as well (we will find "virgin" argument) then 247228753Smm * there is no need to synchronize anything. If primary node 248228753Smm * done any writes already we have to synchronize everything. 249228753Smm */ 250228753Smm PJDLOG_ASSERT(res->hr_secondary_localcnt == 0); 251228753Smm res->hr_resuid = resuid; 252228753Smm if (metadata_write(res) < 0) 253228753Smm exit(EX_NOINPUT); 254228753Smm if (nv_exists(nvin, "virgin")) { 255228753Smm free(map); 256228753Smm map = NULL; 257228753Smm mapsize = 0; 258228753Smm } else { 259228753Smm memset(map, 0xff, mapsize); 260228753Smm } 261228753Smm nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 262228753Smm } else if (res->hr_resuid != resuid) { 263228753Smm char errmsg[256]; 264228753Smm 265228753Smm (void)snprintf(errmsg, sizeof(errmsg), 266228753Smm "Resource unique ID mismatch (primary=%ju, secondary=%ju).", 267228753Smm (uintmax_t)resuid, (uintmax_t)res->hr_resuid); 268228753Smm pjdlog_error("%s", errmsg); 269228753Smm nv_add_string(nvout, errmsg, "errmsg"); 270228753Smm if (hast_proto_send(res, res->hr_remotein, nvout, NULL, 0) < 0) { 271228753Smm pjdlog_exit(EX_TEMPFAIL, "Unable to send response to %s", 272228753Smm res->hr_remoteaddr); 273228753Smm } 274228753Smm nv_free(nvout); 275228753Smm exit(EX_CONFIG); 276228753Smm } else if ( 277228753Smm /* Is primary is out-of-date? */ 278228753Smm (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 279228753Smm res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 280228753Smm /* Nodes are more or less in sync? */ 281228753Smm (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 282228753Smm res->hr_secondary_remotecnt == res->hr_primary_localcnt) || 283228753Smm /* Is secondary is out-of-date? */ 284228753Smm (res->hr_secondary_localcnt == res->hr_primary_remotecnt && 285228753Smm res->hr_secondary_remotecnt < res->hr_primary_localcnt)) { 286228753Smm /* 287228753Smm * Nodes are more or less in sync or one of the nodes is 288228753Smm * out-of-date. 289228753Smm * It doesn't matter at this point which one, we just have to 290228753Smm * send out local bitmap to the remote node. 291228753Smm */ 292228753Smm if (pread(res->hr_localfd, map, mapsize, METADATA_SIZE) != 293228753Smm (ssize_t)mapsize) { 294228753Smm pjdlog_exit(LOG_ERR, "Unable to read activemap"); 295228753Smm } 296228753Smm if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 297228753Smm res->hr_secondary_remotecnt == res->hr_primary_localcnt) { 298228753Smm /* Primary is out-of-date, sync from secondary. */ 299228753Smm nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 300228753Smm } else { 301228753Smm /* 302228753Smm * Secondary is out-of-date or counts match. 303228753Smm * Sync from primary. 304228753Smm */ 305228753Smm nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 306228753Smm } 307228753Smm } else if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 308228753Smm res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 309228753Smm /* 310228753Smm * Not good, we have split-brain condition. 311228753Smm */ 312228753Smm pjdlog_error("Split-brain detected, exiting."); 313228753Smm nv_add_string(nvout, "Split-brain condition!", "errmsg"); 314228753Smm free(map); 315228753Smm map = NULL; 316228753Smm mapsize = 0; 317228753Smm } else /* if (res->hr_secondary_localcnt < res->hr_primary_remotecnt || 318228753Smm res->hr_primary_localcnt < res->hr_secondary_remotecnt) */ { 319228753Smm /* 320228753Smm * This should never happen in practise, but we will perform 321228753Smm * full synchronization. 322228753Smm */ 323228753Smm PJDLOG_ASSERT(res->hr_secondary_localcnt < res->hr_primary_remotecnt || 324228753Smm res->hr_primary_localcnt < res->hr_secondary_remotecnt); 325228753Smm mapsize = activemap_calc_ondisk_size(res->hr_local_mediasize - 326228753Smm METADATA_SIZE, res->hr_extentsize, 327228753Smm res->hr_local_sectorsize); 328228753Smm memset(map, 0xff, mapsize); 329228753Smm if (res->hr_secondary_localcnt > res->hr_primary_remotecnt) { 330228753Smm /* In this one of five cases sync from secondary. */ 331228753Smm nv_add_uint8(nvout, HAST_SYNCSRC_SECONDARY, "syncsrc"); 332228753Smm } else { 333228753Smm /* For the rest four cases sync from primary. */ 334228753Smm nv_add_uint8(nvout, HAST_SYNCSRC_PRIMARY, "syncsrc"); 335228753Smm } 336228753Smm pjdlog_warning("This should never happen, asking for full synchronization (primary(local=%ju, remote=%ju), secondary(local=%ju, remote=%ju)).", 337228753Smm (uintmax_t)res->hr_primary_localcnt, 338228753Smm (uintmax_t)res->hr_primary_remotecnt, 339228753Smm (uintmax_t)res->hr_secondary_localcnt, 340228753Smm (uintmax_t)res->hr_secondary_remotecnt); 341228753Smm } 342228753Smm if (hast_proto_send(res, res->hr_remotein, nvout, map, mapsize) < 0) { 343228753Smm pjdlog_exit(EX_TEMPFAIL, "Unable to send activemap to %s", 344228753Smm res->hr_remoteaddr); 345228753Smm } 346228753Smm if (map != NULL) 347228753Smm free(map); 348228753Smm nv_free(nvout); 349228753Smm if (res->hr_secondary_localcnt > res->hr_primary_remotecnt && 350228753Smm res->hr_primary_localcnt > res->hr_secondary_remotecnt) { 351228753Smm /* Exit on split-brain. */ 352228753Smm event_send(res, EVENT_SPLITBRAIN); 353228753Smm exit(EX_CONFIG); 354228753Smm } 355228753Smm} 356228753Smm 357228753Smmvoid 358228753Smmhastd_secondary(struct hast_resource *res, struct nv *nvin) 359228753Smm{ 360228753Smm sigset_t mask; 361228753Smm pthread_t td; 362228753Smm pid_t pid; 363228753Smm int error, mode, debuglevel; 364228753Smm 365228753Smm /* 366228753Smm * Create communication channel between parent and child. 367228753Smm */ 368228753Smm if (proto_client(NULL, "socketpair://", &res->hr_ctrl) < 0) { 369228753Smm KEEP_ERRNO((void)pidfile_remove(pfh)); 370228753Smm pjdlog_exit(EX_OSERR, 371228753Smm "Unable to create control sockets between parent and child"); 372228753Smm } 373228753Smm /* 374228753Smm * Create communication channel between child and parent. 375228753Smm */ 376228753Smm if (proto_client(NULL, "socketpair://", &res->hr_event) < 0) { 377228753Smm KEEP_ERRNO((void)pidfile_remove(pfh)); 378228753Smm pjdlog_exit(EX_OSERR, 379228753Smm "Unable to create event sockets between child and parent"); 380228753Smm } 381228753Smm /* 382228753Smm * Create communication channel for sending connection requests from 383228753Smm * parent to child. 384228753Smm */ 385228753Smm if (proto_client(NULL, "socketpair://", &res->hr_conn) < 0) { 386228753Smm /* TODO: There's no need for this to be fatal error. */ 387228753Smm KEEP_ERRNO((void)pidfile_remove(pfh)); 388228753Smm pjdlog_exit(EX_OSERR, 389228753Smm "Unable to create connection sockets between parent and child"); 390228753Smm } 391228753Smm 392228753Smm pid = fork(); 393228753Smm if (pid < 0) { 394228753Smm KEEP_ERRNO((void)pidfile_remove(pfh)); 395228753Smm pjdlog_exit(EX_OSERR, "Unable to fork"); 396228753Smm } 397228753Smm 398228753Smm if (pid > 0) { 399228753Smm /* This is parent. */ 400228753Smm proto_close(res->hr_remotein); 401228753Smm res->hr_remotein = NULL; 402228753Smm proto_close(res->hr_remoteout); 403228753Smm res->hr_remoteout = NULL; 404228753Smm /* Declare that we are receiver. */ 405228753Smm proto_recv(res->hr_event, NULL, 0); 406228753Smm /* Declare that we are sender. */ 407228753Smm proto_send(res->hr_ctrl, NULL, 0); 408228753Smm proto_send(res->hr_conn, NULL, 0); 409228753Smm res->hr_workerpid = pid; 410228753Smm return; 411228753Smm } 412228753Smm 413228753Smm gres = res; 414228753Smm mode = pjdlog_mode_get(); 415228753Smm debuglevel = pjdlog_debug_get(); 416228753Smm 417228753Smm /* Declare that we are sender. */ 418228753Smm proto_send(res->hr_event, NULL, 0); 419228753Smm /* Declare that we are receiver. */ 420228753Smm proto_recv(res->hr_ctrl, NULL, 0); 421228753Smm proto_recv(res->hr_conn, NULL, 0); 422228753Smm descriptors_cleanup(res); 423228753Smm 424228753Smm descriptors_assert(res, mode); 425228753Smm 426228753Smm pjdlog_init(mode); 427228753Smm pjdlog_debug_set(debuglevel); 428228753Smm pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role)); 429228753Smm setproctitle("%s (secondary)", res->hr_name); 430228753Smm 431228753Smm PJDLOG_VERIFY(sigemptyset(&mask) == 0); 432228753Smm PJDLOG_VERIFY(sigprocmask(SIG_SETMASK, &mask, NULL) == 0); 433228753Smm 434228753Smm /* Error in setting timeout is not critical, but why should it fail? */ 435228753Smm if (proto_timeout(res->hr_remotein, 2 * HAST_KEEPALIVE) < 0) 436228753Smm pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 437228753Smm if (proto_timeout(res->hr_remoteout, res->hr_timeout) < 0) 438228753Smm pjdlog_errno(LOG_WARNING, "Unable to set connection timeout"); 439228753Smm 440228753Smm init_local(res); 441228753Smm init_environment(); 442228753Smm 443228753Smm if (drop_privs(true) != 0) 444228753Smm exit(EX_CONFIG); 445228753Smm pjdlog_info("Privileges successfully dropped."); 446228753Smm 447228753Smm /* 448228753Smm * Create the control thread before sending any event to the parent, 449228753Smm * as we can deadlock when parent sends control request to worker, 450228753Smm * but worker has no control thread started yet, so parent waits. 451 * In the meantime worker sends an event to the parent, but parent 452 * is unable to handle the event, because it waits for control 453 * request response. 454 */ 455 error = pthread_create(&td, NULL, ctrl_thread, res); 456 PJDLOG_ASSERT(error == 0); 457 458 init_remote(res, nvin); 459 event_send(res, EVENT_CONNECT); 460 461 error = pthread_create(&td, NULL, recv_thread, res); 462 PJDLOG_ASSERT(error == 0); 463 error = pthread_create(&td, NULL, disk_thread, res); 464 PJDLOG_ASSERT(error == 0); 465 (void)send_thread(res); 466} 467 468static void 469reqlog(int loglevel, int debuglevel, int error, struct hio *hio, const char *fmt, ...) 470{ 471 char msg[1024]; 472 va_list ap; 473 int len; 474 475 va_start(ap, fmt); 476 len = vsnprintf(msg, sizeof(msg), fmt, ap); 477 va_end(ap); 478 if ((size_t)len < sizeof(msg)) { 479 switch (hio->hio_cmd) { 480 case HIO_READ: 481 (void)snprintf(msg + len, sizeof(msg) - len, 482 "READ(%ju, %ju).", (uintmax_t)hio->hio_offset, 483 (uintmax_t)hio->hio_length); 484 break; 485 case HIO_DELETE: 486 (void)snprintf(msg + len, sizeof(msg) - len, 487 "DELETE(%ju, %ju).", (uintmax_t)hio->hio_offset, 488 (uintmax_t)hio->hio_length); 489 break; 490 case HIO_FLUSH: 491 (void)snprintf(msg + len, sizeof(msg) - len, "FLUSH."); 492 break; 493 case HIO_WRITE: 494 (void)snprintf(msg + len, sizeof(msg) - len, 495 "WRITE(%ju, %ju).", (uintmax_t)hio->hio_offset, 496 (uintmax_t)hio->hio_length); 497 break; 498 case HIO_KEEPALIVE: 499 (void)snprintf(msg + len, sizeof(msg) - len, "KEEPALIVE."); 500 break; 501 default: 502 (void)snprintf(msg + len, sizeof(msg) - len, 503 "UNKNOWN(%u).", (unsigned int)hio->hio_cmd); 504 break; 505 } 506 } 507 pjdlog_common(loglevel, debuglevel, error, "%s", msg); 508} 509 510static int 511requnpack(struct hast_resource *res, struct hio *hio) 512{ 513 514 hio->hio_cmd = nv_get_uint8(hio->hio_nv, "cmd"); 515 if (hio->hio_cmd == 0) { 516 pjdlog_error("Header contains no 'cmd' field."); 517 hio->hio_error = EINVAL; 518 goto end; 519 } 520 switch (hio->hio_cmd) { 521 case HIO_KEEPALIVE: 522 break; 523 case HIO_READ: 524 case HIO_WRITE: 525 case HIO_DELETE: 526 hio->hio_offset = nv_get_uint64(hio->hio_nv, "offset"); 527 if (nv_error(hio->hio_nv) != 0) { 528 pjdlog_error("Header is missing 'offset' field."); 529 hio->hio_error = EINVAL; 530 goto end; 531 } 532 hio->hio_length = nv_get_uint64(hio->hio_nv, "length"); 533 if (nv_error(hio->hio_nv) != 0) { 534 pjdlog_error("Header is missing 'length' field."); 535 hio->hio_error = EINVAL; 536 goto end; 537 } 538 if (hio->hio_length == 0) { 539 pjdlog_error("Data length is zero."); 540 hio->hio_error = EINVAL; 541 goto end; 542 } 543 if (hio->hio_length > MAXPHYS) { 544 pjdlog_error("Data length is too large (%ju > %ju).", 545 (uintmax_t)hio->hio_length, (uintmax_t)MAXPHYS); 546 hio->hio_error = EINVAL; 547 goto end; 548 } 549 if ((hio->hio_offset % res->hr_local_sectorsize) != 0) { 550 pjdlog_error("Offset %ju is not multiple of sector size.", 551 (uintmax_t)hio->hio_offset); 552 hio->hio_error = EINVAL; 553 goto end; 554 } 555 if ((hio->hio_length % res->hr_local_sectorsize) != 0) { 556 pjdlog_error("Length %ju is not multiple of sector size.", 557 (uintmax_t)hio->hio_length); 558 hio->hio_error = EINVAL; 559 goto end; 560 } 561 if (hio->hio_offset + hio->hio_length > 562 (uint64_t)res->hr_datasize) { 563 pjdlog_error("Data offset is too large (%ju > %ju).", 564 (uintmax_t)(hio->hio_offset + hio->hio_length), 565 (uintmax_t)res->hr_datasize); 566 hio->hio_error = EINVAL; 567 goto end; 568 } 569 break; 570 default: 571 pjdlog_error("Header contains invalid 'cmd' (%hhu).", 572 hio->hio_cmd); 573 hio->hio_error = EINVAL; 574 goto end; 575 } 576 hio->hio_error = 0; 577end: 578 return (hio->hio_error); 579} 580 581static __dead2 void 582secondary_exit(int exitcode, const char *fmt, ...) 583{ 584 va_list ap; 585 586 PJDLOG_ASSERT(exitcode != EX_OK); 587 va_start(ap, fmt); 588 pjdlogv_errno(LOG_ERR, fmt, ap); 589 va_end(ap); 590 event_send(gres, EVENT_DISCONNECT); 591 exit(exitcode); 592} 593 594/* 595 * Thread receives requests from the primary node. 596 */ 597static void * 598recv_thread(void *arg) 599{ 600 struct hast_resource *res = arg; 601 struct hio *hio; 602 603 for (;;) { 604 pjdlog_debug(2, "recv: Taking free request."); 605 QUEUE_TAKE(free, hio); 606 pjdlog_debug(2, "recv: (%p) Got request.", hio); 607 if (hast_proto_recv_hdr(res->hr_remotein, &hio->hio_nv) < 0) { 608 secondary_exit(EX_TEMPFAIL, 609 "Unable to receive request header"); 610 } 611 if (requnpack(res, hio) != 0) { 612 pjdlog_debug(2, 613 "recv: (%p) Moving request to the send queue.", 614 hio); 615 QUEUE_INSERT(send, hio); 616 continue; 617 } 618 reqlog(LOG_DEBUG, 2, -1, hio, 619 "recv: (%p) Got request header: ", hio); 620 if (hio->hio_cmd == HIO_KEEPALIVE) { 621 pjdlog_debug(2, 622 "recv: (%p) Moving request to the free queue.", 623 hio); 624 nv_free(hio->hio_nv); 625 QUEUE_INSERT(free, hio); 626 continue; 627 } else if (hio->hio_cmd == HIO_WRITE) { 628 if (hast_proto_recv_data(res, res->hr_remotein, 629 hio->hio_nv, hio->hio_data, MAXPHYS) < 0) { 630 secondary_exit(EX_TEMPFAIL, 631 "Unable to receive request data"); 632 } 633 } 634 pjdlog_debug(2, "recv: (%p) Moving request to the disk queue.", 635 hio); 636 QUEUE_INSERT(disk, hio); 637 } 638 /* NOTREACHED */ 639 return (NULL); 640} 641 642/* 643 * Thread reads from or writes to local component and also handles DELETE and 644 * FLUSH requests. 645 */ 646static void * 647disk_thread(void *arg) 648{ 649 struct hast_resource *res = arg; 650 struct hio *hio; 651 ssize_t ret; 652 bool clear_activemap; 653 654 clear_activemap = true; 655 656 for (;;) { 657 pjdlog_debug(2, "disk: Taking request."); 658 QUEUE_TAKE(disk, hio); 659 while (clear_activemap) { 660 unsigned char *map; 661 size_t mapsize; 662 663 /* 664 * When first request is received, it means that primary 665 * already received our activemap, merged it and stored 666 * locally. We can now safely clear our activemap. 667 */ 668 mapsize = 669 activemap_calc_ondisk_size(res->hr_local_mediasize - 670 METADATA_SIZE, res->hr_extentsize, 671 res->hr_local_sectorsize); 672 map = calloc(1, mapsize); 673 if (map == NULL) { 674 pjdlog_warning("Unable to allocate memory to clear local activemap."); 675 break; 676 } 677 if (pwrite(res->hr_localfd, map, mapsize, 678 METADATA_SIZE) != (ssize_t)mapsize) { 679 pjdlog_errno(LOG_WARNING, 680 "Unable to store cleared activemap"); 681 free(map); 682 break; 683 } 684 free(map); 685 clear_activemap = false; 686 pjdlog_debug(1, "Local activemap cleared."); 687 } 688 reqlog(LOG_DEBUG, 2, -1, hio, "disk: (%p) Got request: ", hio); 689 /* Handle the actual request. */ 690 switch (hio->hio_cmd) { 691 case HIO_READ: 692 ret = pread(res->hr_localfd, hio->hio_data, 693 hio->hio_length, 694 hio->hio_offset + res->hr_localoff); 695 if (ret < 0) 696 hio->hio_error = errno; 697 else if (ret != (int64_t)hio->hio_length) 698 hio->hio_error = EIO; 699 else 700 hio->hio_error = 0; 701 break; 702 case HIO_WRITE: 703 ret = pwrite(res->hr_localfd, hio->hio_data, 704 hio->hio_length, 705 hio->hio_offset + res->hr_localoff); 706 if (ret < 0) 707 hio->hio_error = errno; 708 else if (ret != (int64_t)hio->hio_length) 709 hio->hio_error = EIO; 710 else 711 hio->hio_error = 0; 712 break; 713 case HIO_DELETE: 714 ret = g_delete(res->hr_localfd, 715 hio->hio_offset + res->hr_localoff, 716 hio->hio_length); 717 if (ret < 0) 718 hio->hio_error = errno; 719 else 720 hio->hio_error = 0; 721 break; 722 case HIO_FLUSH: 723 ret = g_flush(res->hr_localfd); 724 if (ret < 0) 725 hio->hio_error = errno; 726 else 727 hio->hio_error = 0; 728 break; 729 } 730 if (hio->hio_error != 0) { 731 reqlog(LOG_ERR, 0, hio->hio_error, hio, 732 "Request failed: "); 733 } 734 pjdlog_debug(2, "disk: (%p) Moving request to the send queue.", 735 hio); 736 QUEUE_INSERT(send, hio); 737 } 738 /* NOTREACHED */ 739 return (NULL); 740} 741 742/* 743 * Thread sends requests back to primary node. 744 */ 745static void * 746send_thread(void *arg) 747{ 748 struct hast_resource *res = arg; 749 struct nv *nvout; 750 struct hio *hio; 751 void *data; 752 size_t length; 753 754 for (;;) { 755 pjdlog_debug(2, "send: Taking request."); 756 QUEUE_TAKE(send, hio); 757 reqlog(LOG_DEBUG, 2, -1, hio, "send: (%p) Got request: ", hio); 758 nvout = nv_alloc(); 759 /* Copy sequence number. */ 760 nv_add_uint64(nvout, nv_get_uint64(hio->hio_nv, "seq"), "seq"); 761 switch (hio->hio_cmd) { 762 case HIO_READ: 763 if (hio->hio_error == 0) { 764 data = hio->hio_data; 765 length = hio->hio_length; 766 break; 767 } 768 /* 769 * We send no data in case of an error. 770 */ 771 /* FALLTHROUGH */ 772 case HIO_DELETE: 773 case HIO_FLUSH: 774 case HIO_WRITE: 775 data = NULL; 776 length = 0; 777 break; 778 default: 779 abort(); 780 break; 781 } 782 if (hio->hio_error != 0) 783 nv_add_int16(nvout, hio->hio_error, "error"); 784 if (hast_proto_send(res, res->hr_remoteout, nvout, data, 785 length) < 0) { 786 secondary_exit(EX_TEMPFAIL, "Unable to send reply."); 787 } 788 nv_free(nvout); 789 pjdlog_debug(2, "send: (%p) Moving request to the free queue.", 790 hio); 791 nv_free(hio->hio_nv); 792 hio->hio_error = 0; 793 QUEUE_INSERT(free, hio); 794 } 795 /* NOTREACHED */ 796 return (NULL); 797} 798