g_raid.c revision 234603
1219974Smav/*- 2219974Smav * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org> 3219974Smav * All rights reserved. 4219974Smav * 5219974Smav * Redistribution and use in source and binary forms, with or without 6219974Smav * modification, are permitted provided that the following conditions 7219974Smav * are met: 8219974Smav * 1. Redistributions of source code must retain the above copyright 9219974Smav * notice, this list of conditions and the following disclaimer. 10219974Smav * 2. Redistributions in binary form must reproduce the above copyright 11219974Smav * notice, this list of conditions and the following disclaimer in the 12219974Smav * documentation and/or other materials provided with the distribution. 13219974Smav * 14219974Smav * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15219974Smav * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16219974Smav * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17219974Smav * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18219974Smav * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19219974Smav * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20219974Smav * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21219974Smav * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22219974Smav * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23219974Smav * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24219974Smav * SUCH DAMAGE. 25219974Smav */ 26219974Smav 27219974Smav#include <sys/cdefs.h> 28219974Smav__FBSDID("$FreeBSD: head/sys/geom/raid/g_raid.c 234603 2012-04-23 13:04:02Z mav $"); 29219974Smav 30219974Smav#include <sys/param.h> 31219974Smav#include <sys/systm.h> 32219974Smav#include <sys/kernel.h> 33219974Smav#include <sys/module.h> 34219974Smav#include <sys/limits.h> 35219974Smav#include <sys/lock.h> 36219974Smav#include <sys/mutex.h> 37219974Smav#include <sys/bio.h> 38223921Sae#include <sys/sbuf.h> 39219974Smav#include <sys/sysctl.h> 40219974Smav#include <sys/malloc.h> 41219974Smav#include <sys/eventhandler.h> 42219974Smav#include <vm/uma.h> 43219974Smav#include <geom/geom.h> 44219974Smav#include <sys/proc.h> 45219974Smav#include <sys/kthread.h> 46219974Smav#include <sys/sched.h> 47219974Smav#include <geom/raid/g_raid.h> 48219974Smav#include "g_raid_md_if.h" 49219974Smav#include "g_raid_tr_if.h" 50219974Smav 51219974Smavstatic MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data"); 52219974Smav 53219974SmavSYSCTL_DECL(_kern_geom); 54219974SmavSYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff"); 55219974Smavu_int g_raid_aggressive_spare = 0; 56219974SmavTUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare); 57219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW, 58219974Smav &g_raid_aggressive_spare, 0, "Use disks without metadata as spare"); 59220790Smavu_int g_raid_debug = 0; 60219974SmavTUNABLE_INT("kern.geom.raid.debug", &g_raid_debug); 61219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0, 62219974Smav "Debug level"); 63219974Smavint g_raid_read_err_thresh = 10; 64219974SmavTUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh); 65219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW, 66219974Smav &g_raid_read_err_thresh, 0, 67219974Smav "Number of read errors equated to disk failure"); 68219974Smavu_int g_raid_start_timeout = 30; 69219974SmavTUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout); 70219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW, 71219974Smav &g_raid_start_timeout, 0, 72219974Smav "Time to wait for all array components"); 73219974Smavstatic u_int g_raid_clean_time = 5; 74219974SmavTUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time); 75219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW, 76219974Smav &g_raid_clean_time, 0, "Mark volume as clean when idling"); 77219974Smavstatic u_int g_raid_disconnect_on_failure = 1; 78219974SmavTUNABLE_INT("kern.geom.raid.disconnect_on_failure", 79219974Smav &g_raid_disconnect_on_failure); 80219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, 81219974Smav &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 82219974Smavstatic u_int g_raid_name_format = 0; 83219974SmavTUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format); 84219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW, 85219974Smav &g_raid_name_format, 0, "Providers name format."); 86219974Smavstatic u_int g_raid_idle_threshold = 1000000; 87219974SmavTUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold); 88219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW, 89219974Smav &g_raid_idle_threshold, 1000000, 90219974Smav "Time in microseconds to consider a volume idle."); 91219974Smav 92219974Smav#define MSLEEP(rv, ident, mtx, priority, wmesg, timeout) do { \ 93219974Smav G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 94219974Smav rv = msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 95219974Smav G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 96219974Smav} while (0) 97219974Smav 98219974SmavLIST_HEAD(, g_raid_md_class) g_raid_md_classes = 99219974Smav LIST_HEAD_INITIALIZER(g_raid_md_classes); 100219974Smav 101219974SmavLIST_HEAD(, g_raid_tr_class) g_raid_tr_classes = 102219974Smav LIST_HEAD_INITIALIZER(g_raid_tr_classes); 103219974Smav 104219974SmavLIST_HEAD(, g_raid_volume) g_raid_volumes = 105219974Smav LIST_HEAD_INITIALIZER(g_raid_volumes); 106219974Smav 107219974Smavstatic eventhandler_tag g_raid_pre_sync = NULL; 108219974Smavstatic int g_raid_started = 0; 109219974Smav 110219974Smavstatic int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp, 111219974Smav struct g_geom *gp); 112219974Smavstatic g_taste_t g_raid_taste; 113219974Smavstatic void g_raid_init(struct g_class *mp); 114219974Smavstatic void g_raid_fini(struct g_class *mp); 115219974Smav 116219974Smavstruct g_class g_raid_class = { 117219974Smav .name = G_RAID_CLASS_NAME, 118219974Smav .version = G_VERSION, 119219974Smav .ctlreq = g_raid_ctl, 120219974Smav .taste = g_raid_taste, 121219974Smav .destroy_geom = g_raid_destroy_geom, 122219974Smav .init = g_raid_init, 123219974Smav .fini = g_raid_fini 124219974Smav}; 125219974Smav 126219974Smavstatic void g_raid_destroy_provider(struct g_raid_volume *vol); 127219974Smavstatic int g_raid_update_disk(struct g_raid_disk *disk, u_int event); 128219974Smavstatic int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event); 129219974Smavstatic int g_raid_update_volume(struct g_raid_volume *vol, u_int event); 130219974Smavstatic int g_raid_update_node(struct g_raid_softc *sc, u_int event); 131219974Smavstatic void g_raid_dumpconf(struct sbuf *sb, const char *indent, 132219974Smav struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 133219974Smavstatic void g_raid_start(struct bio *bp); 134219974Smavstatic void g_raid_start_request(struct bio *bp); 135219974Smavstatic void g_raid_disk_done(struct bio *bp); 136219974Smavstatic void g_raid_poll(struct g_raid_softc *sc); 137219974Smav 138219974Smavstatic const char * 139219974Smavg_raid_node_event2str(int event) 140219974Smav{ 141219974Smav 142219974Smav switch (event) { 143219974Smav case G_RAID_NODE_E_WAKE: 144219974Smav return ("WAKE"); 145219974Smav case G_RAID_NODE_E_START: 146219974Smav return ("START"); 147219974Smav default: 148219974Smav return ("INVALID"); 149219974Smav } 150219974Smav} 151219974Smav 152219974Smavconst char * 153219974Smavg_raid_disk_state2str(int state) 154219974Smav{ 155219974Smav 156219974Smav switch (state) { 157219974Smav case G_RAID_DISK_S_NONE: 158219974Smav return ("NONE"); 159219974Smav case G_RAID_DISK_S_OFFLINE: 160219974Smav return ("OFFLINE"); 161219974Smav case G_RAID_DISK_S_FAILED: 162219974Smav return ("FAILED"); 163219974Smav case G_RAID_DISK_S_STALE_FAILED: 164219974Smav return ("STALE_FAILED"); 165219974Smav case G_RAID_DISK_S_SPARE: 166219974Smav return ("SPARE"); 167219974Smav case G_RAID_DISK_S_STALE: 168219974Smav return ("STALE"); 169219974Smav case G_RAID_DISK_S_ACTIVE: 170219974Smav return ("ACTIVE"); 171219974Smav default: 172219974Smav return ("INVALID"); 173219974Smav } 174219974Smav} 175219974Smav 176219974Smavstatic const char * 177219974Smavg_raid_disk_event2str(int event) 178219974Smav{ 179219974Smav 180219974Smav switch (event) { 181219974Smav case G_RAID_DISK_E_DISCONNECTED: 182219974Smav return ("DISCONNECTED"); 183219974Smav default: 184219974Smav return ("INVALID"); 185219974Smav } 186219974Smav} 187219974Smav 188219974Smavconst char * 189219974Smavg_raid_subdisk_state2str(int state) 190219974Smav{ 191219974Smav 192219974Smav switch (state) { 193219974Smav case G_RAID_SUBDISK_S_NONE: 194219974Smav return ("NONE"); 195219974Smav case G_RAID_SUBDISK_S_FAILED: 196219974Smav return ("FAILED"); 197219974Smav case G_RAID_SUBDISK_S_NEW: 198219974Smav return ("NEW"); 199219974Smav case G_RAID_SUBDISK_S_REBUILD: 200219974Smav return ("REBUILD"); 201219974Smav case G_RAID_SUBDISK_S_UNINITIALIZED: 202219974Smav return ("UNINITIALIZED"); 203219974Smav case G_RAID_SUBDISK_S_STALE: 204219974Smav return ("STALE"); 205219974Smav case G_RAID_SUBDISK_S_RESYNC: 206219974Smav return ("RESYNC"); 207219974Smav case G_RAID_SUBDISK_S_ACTIVE: 208219974Smav return ("ACTIVE"); 209219974Smav default: 210219974Smav return ("INVALID"); 211219974Smav } 212219974Smav} 213219974Smav 214219974Smavstatic const char * 215219974Smavg_raid_subdisk_event2str(int event) 216219974Smav{ 217219974Smav 218219974Smav switch (event) { 219219974Smav case G_RAID_SUBDISK_E_NEW: 220219974Smav return ("NEW"); 221219974Smav case G_RAID_SUBDISK_E_DISCONNECTED: 222219974Smav return ("DISCONNECTED"); 223219974Smav default: 224219974Smav return ("INVALID"); 225219974Smav } 226219974Smav} 227219974Smav 228219974Smavconst char * 229219974Smavg_raid_volume_state2str(int state) 230219974Smav{ 231219974Smav 232219974Smav switch (state) { 233219974Smav case G_RAID_VOLUME_S_STARTING: 234219974Smav return ("STARTING"); 235219974Smav case G_RAID_VOLUME_S_BROKEN: 236219974Smav return ("BROKEN"); 237219974Smav case G_RAID_VOLUME_S_DEGRADED: 238219974Smav return ("DEGRADED"); 239219974Smav case G_RAID_VOLUME_S_SUBOPTIMAL: 240219974Smav return ("SUBOPTIMAL"); 241219974Smav case G_RAID_VOLUME_S_OPTIMAL: 242219974Smav return ("OPTIMAL"); 243219974Smav case G_RAID_VOLUME_S_UNSUPPORTED: 244219974Smav return ("UNSUPPORTED"); 245219974Smav case G_RAID_VOLUME_S_STOPPED: 246219974Smav return ("STOPPED"); 247219974Smav default: 248219974Smav return ("INVALID"); 249219974Smav } 250219974Smav} 251219974Smav 252219974Smavstatic const char * 253219974Smavg_raid_volume_event2str(int event) 254219974Smav{ 255219974Smav 256219974Smav switch (event) { 257219974Smav case G_RAID_VOLUME_E_UP: 258219974Smav return ("UP"); 259219974Smav case G_RAID_VOLUME_E_DOWN: 260219974Smav return ("DOWN"); 261219974Smav case G_RAID_VOLUME_E_START: 262219974Smav return ("START"); 263219974Smav case G_RAID_VOLUME_E_STARTMD: 264219974Smav return ("STARTMD"); 265219974Smav default: 266219974Smav return ("INVALID"); 267219974Smav } 268219974Smav} 269219974Smav 270219974Smavconst char * 271219974Smavg_raid_volume_level2str(int level, int qual) 272219974Smav{ 273219974Smav 274219974Smav switch (level) { 275219974Smav case G_RAID_VOLUME_RL_RAID0: 276219974Smav return ("RAID0"); 277219974Smav case G_RAID_VOLUME_RL_RAID1: 278219974Smav return ("RAID1"); 279219974Smav case G_RAID_VOLUME_RL_RAID3: 280234603Smav if (qual == G_RAID_VOLUME_RLQ_R3P0) 281234603Smav return ("RAID3-P0"); 282234603Smav if (qual == G_RAID_VOLUME_RLQ_R3PN) 283234603Smav return ("RAID3-PN"); 284219974Smav return ("RAID3"); 285219974Smav case G_RAID_VOLUME_RL_RAID4: 286234603Smav if (qual == G_RAID_VOLUME_RLQ_R4P0) 287234603Smav return ("RAID3-P0"); 288234603Smav if (qual == G_RAID_VOLUME_RLQ_R4PN) 289234603Smav return ("RAID3-PN"); 290219974Smav return ("RAID4"); 291219974Smav case G_RAID_VOLUME_RL_RAID5: 292234458Smav if (qual == G_RAID_VOLUME_RLQ_R5RA) 293234603Smav return ("RAID5-RA"); 294234458Smav if (qual == G_RAID_VOLUME_RLQ_R5RS) 295234603Smav return ("RAID5-RS"); 296234458Smav if (qual == G_RAID_VOLUME_RLQ_R5LA) 297234603Smav return ("RAID5-LA"); 298234458Smav if (qual == G_RAID_VOLUME_RLQ_R5LS) 299234603Smav return ("RAID5-LS"); 300219974Smav return ("RAID5"); 301219974Smav case G_RAID_VOLUME_RL_RAID6: 302234603Smav if (qual == G_RAID_VOLUME_RLQ_R6RA) 303234603Smav return ("RAID6-RA"); 304234603Smav if (qual == G_RAID_VOLUME_RLQ_R6RS) 305234603Smav return ("RAID6-RS"); 306234603Smav if (qual == G_RAID_VOLUME_RLQ_R6LA) 307234603Smav return ("RAID6-LA"); 308234603Smav if (qual == G_RAID_VOLUME_RLQ_R6LS) 309234603Smav return ("RAID6-LS"); 310219974Smav return ("RAID6"); 311234603Smav case G_RAID_VOLUME_RL_RAIDMDF: 312234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFRA) 313234603Smav return ("RAIDMDF-RA"); 314234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFRS) 315234603Smav return ("RAIDMDF-RS"); 316234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFLA) 317234603Smav return ("RAIDMDF-LA"); 318234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFLS) 319234603Smav return ("RAIDMDF-LS"); 320234603Smav return ("RAIDMDF"); 321219974Smav case G_RAID_VOLUME_RL_RAID1E: 322234603Smav if (qual == G_RAID_VOLUME_RLQ_R1EA) 323234603Smav return ("RAID1E-A"); 324234603Smav if (qual == G_RAID_VOLUME_RLQ_R1EO) 325234603Smav return ("RAID1E-O"); 326219974Smav return ("RAID1E"); 327219974Smav case G_RAID_VOLUME_RL_SINGLE: 328219974Smav return ("SINGLE"); 329219974Smav case G_RAID_VOLUME_RL_CONCAT: 330219974Smav return ("CONCAT"); 331219974Smav case G_RAID_VOLUME_RL_RAID5E: 332234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ERA) 333234603Smav return ("RAID5E-RA"); 334234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ERS) 335234603Smav return ("RAID5E-RS"); 336234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ELA) 337234603Smav return ("RAID5E-LA"); 338234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ELS) 339234603Smav return ("RAID5E-LS"); 340219974Smav return ("RAID5E"); 341219974Smav case G_RAID_VOLUME_RL_RAID5EE: 342234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EERA) 343234603Smav return ("RAID5EE-RA"); 344234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EERS) 345234603Smav return ("RAID5EE-RS"); 346234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EELA) 347234603Smav return ("RAID5EE-LA"); 348234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EELS) 349234603Smav return ("RAID5EE-LS"); 350219974Smav return ("RAID5EE"); 351234603Smav case G_RAID_VOLUME_RL_RAID5R: 352234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RRA) 353234603Smav return ("RAID5R-RA"); 354234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RRS) 355234603Smav return ("RAID5R-RS"); 356234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RLA) 357234603Smav return ("RAID5R-LA"); 358234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RLS) 359234603Smav return ("RAID5R-LS"); 360234603Smav return ("RAID5E"); 361219974Smav default: 362219974Smav return ("UNKNOWN"); 363219974Smav } 364219974Smav} 365219974Smav 366219974Smavint 367219974Smavg_raid_volume_str2level(const char *str, int *level, int *qual) 368219974Smav{ 369219974Smav 370219974Smav *level = G_RAID_VOLUME_RL_UNKNOWN; 371219974Smav *qual = G_RAID_VOLUME_RLQ_NONE; 372219974Smav if (strcasecmp(str, "RAID0") == 0) 373219974Smav *level = G_RAID_VOLUME_RL_RAID0; 374219974Smav else if (strcasecmp(str, "RAID1") == 0) 375219974Smav *level = G_RAID_VOLUME_RL_RAID1; 376234603Smav else if (strcasecmp(str, "RAID3-P0") == 0) { 377219974Smav *level = G_RAID_VOLUME_RL_RAID3; 378234603Smav *qual = G_RAID_VOLUME_RLQ_R3P0; 379234603Smav } else if (strcasecmp(str, "RAID3-PN") == 0 && 380234603Smav strcasecmp(str, "RAID3") == 0) { 381234603Smav *level = G_RAID_VOLUME_RL_RAID3; 382234603Smav *qual = G_RAID_VOLUME_RLQ_R3P0; 383234603Smav } else if (strcasecmp(str, "RAID4-P0") == 0) { 384219974Smav *level = G_RAID_VOLUME_RL_RAID4; 385234603Smav *qual = G_RAID_VOLUME_RLQ_R4P0; 386234603Smav } else if (strcasecmp(str, "RAID4-PN") == 0 && 387234603Smav strcasecmp(str, "RAID4") == 0) { 388234603Smav *level = G_RAID_VOLUME_RL_RAID4; 389234603Smav *qual = G_RAID_VOLUME_RLQ_R4P0; 390234603Smav } else if (strcasecmp(str, "RAID5-RA") == 0) { 391219974Smav *level = G_RAID_VOLUME_RL_RAID5; 392234458Smav *qual = G_RAID_VOLUME_RLQ_R5RA; 393234603Smav } else if (strcasecmp(str, "RAID5-RS") == 0) { 394234458Smav *level = G_RAID_VOLUME_RL_RAID5; 395234458Smav *qual = G_RAID_VOLUME_RLQ_R5RS; 396234458Smav } else if (strcasecmp(str, "RAID5") == 0 || 397234603Smav strcasecmp(str, "RAID5-LA") == 0) { 398234458Smav *level = G_RAID_VOLUME_RL_RAID5; 399234458Smav *qual = G_RAID_VOLUME_RLQ_R5LA; 400234603Smav } else if (strcasecmp(str, "RAID5-LS") == 0) { 401234458Smav *level = G_RAID_VOLUME_RL_RAID5; 402234458Smav *qual = G_RAID_VOLUME_RLQ_R5LS; 403234603Smav } else if (strcasecmp(str, "RAID6-RA") == 0) { 404219974Smav *level = G_RAID_VOLUME_RL_RAID6; 405234603Smav *qual = G_RAID_VOLUME_RLQ_R6RA; 406234603Smav } else if (strcasecmp(str, "RAID6-RS") == 0) { 407234603Smav *level = G_RAID_VOLUME_RL_RAID6; 408234603Smav *qual = G_RAID_VOLUME_RLQ_R6RS; 409234603Smav } else if (strcasecmp(str, "RAID6") == 0 || 410234603Smav strcasecmp(str, "RAID6-LA") == 0) { 411234603Smav *level = G_RAID_VOLUME_RL_RAID6; 412234603Smav *qual = G_RAID_VOLUME_RLQ_R6LA; 413234603Smav } else if (strcasecmp(str, "RAID6-LS") == 0) { 414234603Smav *level = G_RAID_VOLUME_RL_RAID6; 415234603Smav *qual = G_RAID_VOLUME_RLQ_R6LS; 416234603Smav } else if (strcasecmp(str, "RAIDMDF-RA") == 0) { 417234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 418234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFRA; 419234603Smav } else if (strcasecmp(str, "RAIDMDF-RS") == 0) { 420234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 421234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFRS; 422234603Smav } else if (strcasecmp(str, "RAIDMDF") == 0 || 423234603Smav strcasecmp(str, "RAIDMDF-LA") == 0) { 424234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 425234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFLA; 426234603Smav } else if (strcasecmp(str, "RAIDMDF-LS") == 0) { 427234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 428234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFLS; 429234603Smav } else if (strcasecmp(str, "RAID10") == 0 || 430234603Smav strcasecmp(str, "RAID1E") == 0 || 431234603Smav strcasecmp(str, "RAID1E-A") == 0) { 432219974Smav *level = G_RAID_VOLUME_RL_RAID1E; 433234603Smav *qual = G_RAID_VOLUME_RLQ_R1EA; 434234603Smav } else if (strcasecmp(str, "RAID1E-O") == 0) { 435234603Smav *level = G_RAID_VOLUME_RL_RAID1E; 436234603Smav *qual = G_RAID_VOLUME_RLQ_R1EO; 437234603Smav } else if (strcasecmp(str, "SINGLE") == 0) 438219974Smav *level = G_RAID_VOLUME_RL_SINGLE; 439219974Smav else if (strcasecmp(str, "CONCAT") == 0) 440219974Smav *level = G_RAID_VOLUME_RL_CONCAT; 441234603Smav else if (strcasecmp(str, "RAID5E-RA") == 0) { 442219974Smav *level = G_RAID_VOLUME_RL_RAID5E; 443234603Smav *qual = G_RAID_VOLUME_RLQ_R5ERA; 444234603Smav } else if (strcasecmp(str, "RAID5E-RS") == 0) { 445234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 446234603Smav *qual = G_RAID_VOLUME_RLQ_R5ERS; 447234603Smav } else if (strcasecmp(str, "RAID5E") == 0 || 448234603Smav strcasecmp(str, "RAID5E-LA") == 0) { 449234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 450234603Smav *qual = G_RAID_VOLUME_RLQ_R5ELA; 451234603Smav } else if (strcasecmp(str, "RAID5E-LS") == 0) { 452234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 453234603Smav *qual = G_RAID_VOLUME_RLQ_R5ELS; 454234603Smav } else if (strcasecmp(str, "RAID5EE-RA") == 0) { 455219974Smav *level = G_RAID_VOLUME_RL_RAID5EE; 456234603Smav *qual = G_RAID_VOLUME_RLQ_R5EERA; 457234603Smav } else if (strcasecmp(str, "RAID5EE-RS") == 0) { 458234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 459234603Smav *qual = G_RAID_VOLUME_RLQ_R5EERS; 460234603Smav } else if (strcasecmp(str, "RAID5EE") == 0 || 461234603Smav strcasecmp(str, "RAID5EE-LA") == 0) { 462234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 463234603Smav *qual = G_RAID_VOLUME_RLQ_R5EELA; 464234603Smav } else if (strcasecmp(str, "RAID5EE-LS") == 0) { 465234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 466234603Smav *qual = G_RAID_VOLUME_RLQ_R5EELS; 467234603Smav } else if (strcasecmp(str, "RAID5R-RA") == 0) { 468234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 469234603Smav *qual = G_RAID_VOLUME_RLQ_R5RRA; 470234603Smav } else if (strcasecmp(str, "RAID5R-RS") == 0) { 471234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 472234603Smav *qual = G_RAID_VOLUME_RLQ_R5RRS; 473234603Smav } else if (strcasecmp(str, "RAID5R") == 0 || 474234603Smav strcasecmp(str, "RAID5R-LA") == 0) { 475234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 476234603Smav *qual = G_RAID_VOLUME_RLQ_R5RLA; 477234603Smav } else if (strcasecmp(str, "RAID5R-LS") == 0) { 478234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 479234603Smav *qual = G_RAID_VOLUME_RLQ_R5RLS; 480234603Smav } else 481219974Smav return (-1); 482219974Smav return (0); 483219974Smav} 484219974Smav 485219974Smavconst char * 486219974Smavg_raid_get_diskname(struct g_raid_disk *disk) 487219974Smav{ 488219974Smav 489219974Smav if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 490219974Smav return ("[unknown]"); 491219974Smav return (disk->d_consumer->provider->name); 492219974Smav} 493219974Smav 494219974Smavvoid 495219974Smavg_raid_report_disk_state(struct g_raid_disk *disk) 496219974Smav{ 497219974Smav struct g_raid_subdisk *sd; 498219974Smav int len, state; 499219974Smav uint32_t s; 500219974Smav 501219974Smav if (disk->d_consumer == NULL) 502219974Smav return; 503219974Smav if (disk->d_state == G_RAID_DISK_S_FAILED || 504219974Smav disk->d_state == G_RAID_DISK_S_STALE_FAILED) { 505219974Smav s = G_STATE_FAILED; 506219974Smav } else { 507219974Smav state = G_RAID_SUBDISK_S_ACTIVE; 508219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 509219974Smav if (sd->sd_state < state) 510219974Smav state = sd->sd_state; 511219974Smav } 512219974Smav if (state == G_RAID_SUBDISK_S_FAILED) 513219974Smav s = G_STATE_FAILED; 514219974Smav else if (state == G_RAID_SUBDISK_S_NEW || 515219974Smav state == G_RAID_SUBDISK_S_REBUILD) 516219974Smav s = G_STATE_REBUILD; 517219974Smav else if (state == G_RAID_SUBDISK_S_STALE || 518219974Smav state == G_RAID_SUBDISK_S_RESYNC) 519219974Smav s = G_STATE_RESYNC; 520219974Smav else 521219974Smav s = G_STATE_ACTIVE; 522219974Smav } 523219974Smav len = sizeof(s); 524219974Smav g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s); 525219974Smav G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.", 526219974Smav g_raid_get_diskname(disk), s); 527219974Smav} 528219974Smav 529219974Smavvoid 530219974Smavg_raid_change_disk_state(struct g_raid_disk *disk, int state) 531219974Smav{ 532219974Smav 533219974Smav G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.", 534219974Smav g_raid_get_diskname(disk), 535219974Smav g_raid_disk_state2str(disk->d_state), 536219974Smav g_raid_disk_state2str(state)); 537219974Smav disk->d_state = state; 538219974Smav g_raid_report_disk_state(disk); 539219974Smav} 540219974Smav 541219974Smavvoid 542219974Smavg_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state) 543219974Smav{ 544219974Smav 545219974Smav G_RAID_DEBUG1(0, sd->sd_softc, 546219974Smav "Subdisk %s:%d-%s state changed from %s to %s.", 547219974Smav sd->sd_volume->v_name, sd->sd_pos, 548219974Smav sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", 549219974Smav g_raid_subdisk_state2str(sd->sd_state), 550219974Smav g_raid_subdisk_state2str(state)); 551219974Smav sd->sd_state = state; 552219974Smav if (sd->sd_disk) 553219974Smav g_raid_report_disk_state(sd->sd_disk); 554219974Smav} 555219974Smav 556219974Smavvoid 557219974Smavg_raid_change_volume_state(struct g_raid_volume *vol, int state) 558219974Smav{ 559219974Smav 560219974Smav G_RAID_DEBUG1(0, vol->v_softc, 561219974Smav "Volume %s state changed from %s to %s.", 562219974Smav vol->v_name, 563219974Smav g_raid_volume_state2str(vol->v_state), 564219974Smav g_raid_volume_state2str(state)); 565219974Smav vol->v_state = state; 566219974Smav} 567219974Smav 568219974Smav/* 569219974Smav * --- Events handling functions --- 570219974Smav * Events in geom_raid are used to maintain subdisks and volumes status 571219974Smav * from one thread to simplify locking. 572219974Smav */ 573219974Smavstatic void 574219974Smavg_raid_event_free(struct g_raid_event *ep) 575219974Smav{ 576219974Smav 577219974Smav free(ep, M_RAID); 578219974Smav} 579219974Smav 580219974Smavint 581219974Smavg_raid_event_send(void *arg, int event, int flags) 582219974Smav{ 583219974Smav struct g_raid_softc *sc; 584219974Smav struct g_raid_event *ep; 585219974Smav int error; 586219974Smav 587219974Smav if ((flags & G_RAID_EVENT_VOLUME) != 0) { 588219974Smav sc = ((struct g_raid_volume *)arg)->v_softc; 589219974Smav } else if ((flags & G_RAID_EVENT_DISK) != 0) { 590219974Smav sc = ((struct g_raid_disk *)arg)->d_softc; 591219974Smav } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) { 592219974Smav sc = ((struct g_raid_subdisk *)arg)->sd_softc; 593219974Smav } else { 594219974Smav sc = arg; 595219974Smav } 596219974Smav ep = malloc(sizeof(*ep), M_RAID, 597219974Smav sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT); 598219974Smav if (ep == NULL) 599219974Smav return (ENOMEM); 600219974Smav ep->e_tgt = arg; 601219974Smav ep->e_event = event; 602219974Smav ep->e_flags = flags; 603219974Smav ep->e_error = 0; 604219974Smav G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc); 605219974Smav mtx_lock(&sc->sc_queue_mtx); 606219974Smav TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 607219974Smav mtx_unlock(&sc->sc_queue_mtx); 608219974Smav wakeup(sc); 609219974Smav 610219974Smav if ((flags & G_RAID_EVENT_WAIT) == 0) 611219974Smav return (0); 612219974Smav 613219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 614219974Smav G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep); 615219974Smav sx_xunlock(&sc->sc_lock); 616219974Smav while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) { 617219974Smav mtx_lock(&sc->sc_queue_mtx); 618219974Smav MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event", 619219974Smav hz * 5); 620219974Smav } 621219974Smav error = ep->e_error; 622219974Smav g_raid_event_free(ep); 623219974Smav sx_xlock(&sc->sc_lock); 624219974Smav return (error); 625219974Smav} 626219974Smav 627219974Smavstatic void 628219974Smavg_raid_event_cancel(struct g_raid_softc *sc, void *tgt) 629219974Smav{ 630219974Smav struct g_raid_event *ep, *tmpep; 631219974Smav 632219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 633219974Smav 634219974Smav mtx_lock(&sc->sc_queue_mtx); 635219974Smav TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 636219974Smav if (ep->e_tgt != tgt) 637219974Smav continue; 638219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 639219974Smav if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) 640219974Smav g_raid_event_free(ep); 641219974Smav else { 642219974Smav ep->e_error = ECANCELED; 643219974Smav wakeup(ep); 644219974Smav } 645219974Smav } 646219974Smav mtx_unlock(&sc->sc_queue_mtx); 647219974Smav} 648219974Smav 649219974Smavstatic int 650219974Smavg_raid_event_check(struct g_raid_softc *sc, void *tgt) 651219974Smav{ 652219974Smav struct g_raid_event *ep; 653219974Smav int res = 0; 654219974Smav 655219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 656219974Smav 657219974Smav mtx_lock(&sc->sc_queue_mtx); 658219974Smav TAILQ_FOREACH(ep, &sc->sc_events, e_next) { 659219974Smav if (ep->e_tgt != tgt) 660219974Smav continue; 661219974Smav res = 1; 662219974Smav break; 663219974Smav } 664219974Smav mtx_unlock(&sc->sc_queue_mtx); 665219974Smav return (res); 666219974Smav} 667219974Smav 668219974Smav/* 669219974Smav * Return the number of disks in given state. 670219974Smav * If state is equal to -1, count all connected disks. 671219974Smav */ 672219974Smavu_int 673219974Smavg_raid_ndisks(struct g_raid_softc *sc, int state) 674219974Smav{ 675219974Smav struct g_raid_disk *disk; 676219974Smav u_int n; 677219974Smav 678219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 679219974Smav 680219974Smav n = 0; 681219974Smav TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 682219974Smav if (disk->d_state == state || state == -1) 683219974Smav n++; 684219974Smav } 685219974Smav return (n); 686219974Smav} 687219974Smav 688219974Smav/* 689219974Smav * Return the number of subdisks in given state. 690219974Smav * If state is equal to -1, count all connected disks. 691219974Smav */ 692219974Smavu_int 693219974Smavg_raid_nsubdisks(struct g_raid_volume *vol, int state) 694219974Smav{ 695219974Smav struct g_raid_subdisk *subdisk; 696219974Smav struct g_raid_softc *sc; 697219974Smav u_int i, n ; 698219974Smav 699219974Smav sc = vol->v_softc; 700219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 701219974Smav 702219974Smav n = 0; 703219974Smav for (i = 0; i < vol->v_disks_count; i++) { 704219974Smav subdisk = &vol->v_subdisks[i]; 705219974Smav if ((state == -1 && 706219974Smav subdisk->sd_state != G_RAID_SUBDISK_S_NONE) || 707219974Smav subdisk->sd_state == state) 708219974Smav n++; 709219974Smav } 710219974Smav return (n); 711219974Smav} 712219974Smav 713219974Smav/* 714219974Smav * Return the first subdisk in given state. 715219974Smav * If state is equal to -1, then the first connected disks. 716219974Smav */ 717219974Smavstruct g_raid_subdisk * 718219974Smavg_raid_get_subdisk(struct g_raid_volume *vol, int state) 719219974Smav{ 720219974Smav struct g_raid_subdisk *sd; 721219974Smav struct g_raid_softc *sc; 722219974Smav u_int i; 723219974Smav 724219974Smav sc = vol->v_softc; 725219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 726219974Smav 727219974Smav for (i = 0; i < vol->v_disks_count; i++) { 728219974Smav sd = &vol->v_subdisks[i]; 729219974Smav if ((state == -1 && 730219974Smav sd->sd_state != G_RAID_SUBDISK_S_NONE) || 731219974Smav sd->sd_state == state) 732219974Smav return (sd); 733219974Smav } 734219974Smav return (NULL); 735219974Smav} 736219974Smav 737219974Smavstruct g_consumer * 738219974Smavg_raid_open_consumer(struct g_raid_softc *sc, const char *name) 739219974Smav{ 740219974Smav struct g_consumer *cp; 741219974Smav struct g_provider *pp; 742219974Smav 743219974Smav g_topology_assert(); 744219974Smav 745219974Smav if (strncmp(name, "/dev/", 5) == 0) 746219974Smav name += 5; 747219974Smav pp = g_provider_by_name(name); 748219974Smav if (pp == NULL) 749219974Smav return (NULL); 750219974Smav cp = g_new_consumer(sc->sc_geom); 751219974Smav if (g_attach(cp, pp) != 0) { 752219974Smav g_destroy_consumer(cp); 753219974Smav return (NULL); 754219974Smav } 755219974Smav if (g_access(cp, 1, 1, 1) != 0) { 756219974Smav g_detach(cp); 757219974Smav g_destroy_consumer(cp); 758219974Smav return (NULL); 759219974Smav } 760219974Smav return (cp); 761219974Smav} 762219974Smav 763219974Smavstatic u_int 764219974Smavg_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) 765219974Smav{ 766219974Smav struct bio *bp; 767219974Smav u_int nreqs = 0; 768219974Smav 769219974Smav mtx_lock(&sc->sc_queue_mtx); 770219974Smav TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 771219974Smav if (bp->bio_from == cp) 772219974Smav nreqs++; 773219974Smav } 774219974Smav mtx_unlock(&sc->sc_queue_mtx); 775219974Smav return (nreqs); 776219974Smav} 777219974Smav 778219974Smavu_int 779219974Smavg_raid_nopens(struct g_raid_softc *sc) 780219974Smav{ 781219974Smav struct g_raid_volume *vol; 782219974Smav u_int opens; 783219974Smav 784219974Smav opens = 0; 785219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 786219974Smav if (vol->v_provider_open != 0) 787219974Smav opens++; 788219974Smav } 789219974Smav return (opens); 790219974Smav} 791219974Smav 792219974Smavstatic int 793219974Smavg_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) 794219974Smav{ 795219974Smav 796219974Smav if (cp->index > 0) { 797219974Smav G_RAID_DEBUG1(2, sc, 798219974Smav "I/O requests for %s exist, can't destroy it now.", 799219974Smav cp->provider->name); 800219974Smav return (1); 801219974Smav } 802219974Smav if (g_raid_nrequests(sc, cp) > 0) { 803219974Smav G_RAID_DEBUG1(2, sc, 804219974Smav "I/O requests for %s in queue, can't destroy it now.", 805219974Smav cp->provider->name); 806219974Smav return (1); 807219974Smav } 808219974Smav return (0); 809219974Smav} 810219974Smav 811219974Smavstatic void 812219974Smavg_raid_destroy_consumer(void *arg, int flags __unused) 813219974Smav{ 814219974Smav struct g_consumer *cp; 815219974Smav 816219974Smav g_topology_assert(); 817219974Smav 818219974Smav cp = arg; 819219974Smav G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 820219974Smav g_detach(cp); 821219974Smav g_destroy_consumer(cp); 822219974Smav} 823219974Smav 824219974Smavvoid 825219974Smavg_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) 826219974Smav{ 827219974Smav struct g_provider *pp; 828219974Smav int retaste_wait; 829219974Smav 830219974Smav g_topology_assert_not(); 831219974Smav 832219974Smav g_topology_lock(); 833219974Smav cp->private = NULL; 834219974Smav if (g_raid_consumer_is_busy(sc, cp)) 835219974Smav goto out; 836219974Smav pp = cp->provider; 837219974Smav retaste_wait = 0; 838219974Smav if (cp->acw == 1) { 839219974Smav if ((pp->geom->flags & G_GEOM_WITHER) == 0) 840219974Smav retaste_wait = 1; 841219974Smav } 842219974Smav if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 843219974Smav g_access(cp, -cp->acr, -cp->acw, -cp->ace); 844219974Smav if (retaste_wait) { 845219974Smav /* 846219974Smav * After retaste event was send (inside g_access()), we can send 847219974Smav * event to detach and destroy consumer. 848219974Smav * A class, which has consumer to the given provider connected 849219974Smav * will not receive retaste event for the provider. 850219974Smav * This is the way how I ignore retaste events when I close 851219974Smav * consumers opened for write: I detach and destroy consumer 852219974Smav * after retaste event is sent. 853219974Smav */ 854219974Smav g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); 855219974Smav goto out; 856219974Smav } 857219974Smav G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); 858219974Smav g_detach(cp); 859219974Smav g_destroy_consumer(cp); 860219974Smavout: 861219974Smav g_topology_unlock(); 862219974Smav} 863219974Smav 864219974Smavstatic void 865219974Smavg_raid_orphan(struct g_consumer *cp) 866219974Smav{ 867219974Smav struct g_raid_disk *disk; 868219974Smav 869219974Smav g_topology_assert(); 870219974Smav 871219974Smav disk = cp->private; 872219974Smav if (disk == NULL) 873219974Smav return; 874219974Smav g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, 875219974Smav G_RAID_EVENT_DISK); 876219974Smav} 877219974Smav 878219974Smavstatic int 879219974Smavg_raid_clean(struct g_raid_volume *vol, int acw) 880219974Smav{ 881219974Smav struct g_raid_softc *sc; 882219974Smav int timeout; 883219974Smav 884219974Smav sc = vol->v_softc; 885219974Smav g_topology_assert_not(); 886219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 887219974Smav 888219974Smav// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 889219974Smav// return (0); 890219974Smav if (!vol->v_dirty) 891219974Smav return (0); 892219974Smav if (vol->v_writes > 0) 893219974Smav return (0); 894219974Smav if (acw > 0 || (acw == -1 && 895219974Smav vol->v_provider != NULL && vol->v_provider->acw > 0)) { 896219974Smav timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); 897219974Smav if (timeout > 0) 898219974Smav return (timeout); 899219974Smav } 900219974Smav vol->v_dirty = 0; 901219974Smav G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", 902219974Smav vol->v_name); 903219974Smav g_raid_write_metadata(sc, vol, NULL, NULL); 904219974Smav return (0); 905219974Smav} 906219974Smav 907219974Smavstatic void 908219974Smavg_raid_dirty(struct g_raid_volume *vol) 909219974Smav{ 910219974Smav struct g_raid_softc *sc; 911219974Smav 912219974Smav sc = vol->v_softc; 913219974Smav g_topology_assert_not(); 914219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 915219974Smav 916219974Smav// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 917219974Smav// return; 918219974Smav vol->v_dirty = 1; 919219974Smav G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", 920219974Smav vol->v_name); 921219974Smav g_raid_write_metadata(sc, vol, NULL, NULL); 922219974Smav} 923219974Smav 924219974Smavvoid 925219974Smavg_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) 926219974Smav{ 927219974Smav struct g_raid_softc *sc; 928219974Smav struct g_raid_volume *vol; 929219974Smav struct g_raid_subdisk *sd; 930219974Smav struct bio_queue_head queue; 931219974Smav struct bio *cbp; 932219974Smav int i; 933219974Smav 934219974Smav vol = tr->tro_volume; 935219974Smav sc = vol->v_softc; 936219974Smav 937219974Smav /* 938219974Smav * Allocate all bios before sending any request, so we can return 939219974Smav * ENOMEM in nice and clean way. 940219974Smav */ 941219974Smav bioq_init(&queue); 942219974Smav for (i = 0; i < vol->v_disks_count; i++) { 943219974Smav sd = &vol->v_subdisks[i]; 944219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_NONE || 945219974Smav sd->sd_state == G_RAID_SUBDISK_S_FAILED) 946219974Smav continue; 947219974Smav cbp = g_clone_bio(bp); 948219974Smav if (cbp == NULL) 949219974Smav goto failure; 950219974Smav cbp->bio_caller1 = sd; 951219974Smav bioq_insert_tail(&queue, cbp); 952219974Smav } 953219974Smav for (cbp = bioq_first(&queue); cbp != NULL; 954219974Smav cbp = bioq_first(&queue)) { 955219974Smav bioq_remove(&queue, cbp); 956219974Smav sd = cbp->bio_caller1; 957219974Smav cbp->bio_caller1 = NULL; 958219974Smav g_raid_subdisk_iostart(sd, cbp); 959219974Smav } 960219974Smav return; 961219974Smavfailure: 962219974Smav for (cbp = bioq_first(&queue); cbp != NULL; 963219974Smav cbp = bioq_first(&queue)) { 964219974Smav bioq_remove(&queue, cbp); 965219974Smav g_destroy_bio(cbp); 966219974Smav } 967219974Smav if (bp->bio_error == 0) 968219974Smav bp->bio_error = ENOMEM; 969219974Smav g_raid_iodone(bp, bp->bio_error); 970219974Smav} 971219974Smav 972219974Smavstatic void 973219974Smavg_raid_tr_kerneldump_common_done(struct bio *bp) 974219974Smav{ 975219974Smav 976219974Smav bp->bio_flags |= BIO_DONE; 977219974Smav} 978219974Smav 979219974Smavint 980219974Smavg_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, 981219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 982219974Smav{ 983219974Smav struct g_raid_softc *sc; 984219974Smav struct g_raid_volume *vol; 985219974Smav struct bio bp; 986219974Smav 987219974Smav vol = tr->tro_volume; 988219974Smav sc = vol->v_softc; 989219974Smav 990219974Smav bzero(&bp, sizeof(bp)); 991219974Smav bp.bio_cmd = BIO_WRITE; 992219974Smav bp.bio_done = g_raid_tr_kerneldump_common_done; 993219974Smav bp.bio_attribute = NULL; 994219974Smav bp.bio_offset = offset; 995219974Smav bp.bio_length = length; 996219974Smav bp.bio_data = virtual; 997219974Smav bp.bio_to = vol->v_provider; 998219974Smav 999219974Smav g_raid_start(&bp); 1000219974Smav while (!(bp.bio_flags & BIO_DONE)) { 1001219974Smav G_RAID_DEBUG1(4, sc, "Poll..."); 1002219974Smav g_raid_poll(sc); 1003219974Smav DELAY(10); 1004219974Smav } 1005219974Smav 1006219974Smav return (bp.bio_error != 0 ? EIO : 0); 1007219974Smav} 1008219974Smav 1009219974Smavstatic int 1010219974Smavg_raid_dump(void *arg, 1011219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1012219974Smav{ 1013219974Smav struct g_raid_volume *vol; 1014219974Smav int error; 1015219974Smav 1016219974Smav vol = (struct g_raid_volume *)arg; 1017219974Smav G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", 1018219974Smav (long long unsigned)offset, (long long unsigned)length); 1019219974Smav 1020219974Smav error = G_RAID_TR_KERNELDUMP(vol->v_tr, 1021219974Smav virtual, physical, offset, length); 1022219974Smav return (error); 1023219974Smav} 1024219974Smav 1025219974Smavstatic void 1026219974Smavg_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) 1027219974Smav{ 1028219974Smav struct g_kerneldump *gkd; 1029219974Smav struct g_provider *pp; 1030219974Smav struct g_raid_volume *vol; 1031219974Smav 1032219974Smav gkd = (struct g_kerneldump*)bp->bio_data; 1033219974Smav pp = bp->bio_to; 1034219974Smav vol = pp->private; 1035219974Smav g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", 1036219974Smav pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 1037219974Smav gkd->di.dumper = g_raid_dump; 1038219974Smav gkd->di.priv = vol; 1039219974Smav gkd->di.blocksize = vol->v_sectorsize; 1040219974Smav gkd->di.maxiosize = DFLTPHYS; 1041219974Smav gkd->di.mediaoffset = gkd->offset; 1042219974Smav if ((gkd->offset + gkd->length) > vol->v_mediasize) 1043219974Smav gkd->length = vol->v_mediasize - gkd->offset; 1044219974Smav gkd->di.mediasize = gkd->length; 1045219974Smav g_io_deliver(bp, 0); 1046219974Smav} 1047219974Smav 1048219974Smavstatic void 1049219974Smavg_raid_start(struct bio *bp) 1050219974Smav{ 1051219974Smav struct g_raid_softc *sc; 1052219974Smav 1053219974Smav sc = bp->bio_to->geom->softc; 1054219974Smav /* 1055219974Smav * If sc == NULL or there are no valid disks, provider's error 1056219974Smav * should be set and g_raid_start() should not be called at all. 1057219974Smav */ 1058219974Smav// KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, 1059219974Smav// ("Provider's error should be set (error=%d)(mirror=%s).", 1060219974Smav// bp->bio_to->error, bp->bio_to->name)); 1061219974Smav G_RAID_LOGREQ(3, bp, "Request received."); 1062219974Smav 1063219974Smav switch (bp->bio_cmd) { 1064219974Smav case BIO_READ: 1065219974Smav case BIO_WRITE: 1066219974Smav case BIO_DELETE: 1067219974Smav case BIO_FLUSH: 1068219974Smav break; 1069219974Smav case BIO_GETATTR: 1070219974Smav if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 1071219974Smav g_raid_kerneldump(sc, bp); 1072219974Smav else 1073219974Smav g_io_deliver(bp, EOPNOTSUPP); 1074219974Smav return; 1075219974Smav default: 1076219974Smav g_io_deliver(bp, EOPNOTSUPP); 1077219974Smav return; 1078219974Smav } 1079219974Smav mtx_lock(&sc->sc_queue_mtx); 1080219974Smav bioq_disksort(&sc->sc_queue, bp); 1081219974Smav mtx_unlock(&sc->sc_queue_mtx); 1082219974Smav if (!dumping) { 1083219974Smav G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); 1084219974Smav wakeup(sc); 1085219974Smav } 1086219974Smav} 1087219974Smav 1088219974Smavstatic int 1089219974Smavg_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) 1090219974Smav{ 1091219974Smav /* 1092219974Smav * 5 cases: 1093219974Smav * (1) bp entirely below NO 1094219974Smav * (2) bp entirely above NO 1095219974Smav * (3) bp start below, but end in range YES 1096219974Smav * (4) bp entirely within YES 1097219974Smav * (5) bp starts within, ends above YES 1098219974Smav * 1099219974Smav * lock range 10-19 (offset 10 length 10) 1100219974Smav * (1) 1-5: first if kicks it out 1101219974Smav * (2) 30-35: second if kicks it out 1102219974Smav * (3) 5-15: passes both ifs 1103219974Smav * (4) 12-14: passes both ifs 1104219974Smav * (5) 19-20: passes both 1105219974Smav */ 1106219974Smav off_t lend = lstart + len - 1; 1107219974Smav off_t bstart = bp->bio_offset; 1108219974Smav off_t bend = bp->bio_offset + bp->bio_length - 1; 1109219974Smav 1110219974Smav if (bend < lstart) 1111219974Smav return (0); 1112219974Smav if (lend < bstart) 1113219974Smav return (0); 1114219974Smav return (1); 1115219974Smav} 1116219974Smav 1117219974Smavstatic int 1118219974Smavg_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) 1119219974Smav{ 1120219974Smav struct g_raid_lock *lp; 1121219974Smav 1122219974Smav sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); 1123219974Smav 1124219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1125219974Smav if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) 1126219974Smav return (1); 1127219974Smav } 1128219974Smav return (0); 1129219974Smav} 1130219974Smav 1131219974Smavstatic void 1132219974Smavg_raid_start_request(struct bio *bp) 1133219974Smav{ 1134219974Smav struct g_raid_softc *sc; 1135219974Smav struct g_raid_volume *vol; 1136219974Smav 1137219974Smav sc = bp->bio_to->geom->softc; 1138219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1139219974Smav vol = bp->bio_to->private; 1140219974Smav 1141219974Smav /* 1142219974Smav * Check to see if this item is in a locked range. If so, 1143219974Smav * queue it to our locked queue and return. We'll requeue 1144219974Smav * it when the range is unlocked. Internal I/O for the 1145219974Smav * rebuild/rescan/recovery process is excluded from this 1146219974Smav * check so we can actually do the recovery. 1147219974Smav */ 1148219974Smav if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && 1149219974Smav g_raid_is_in_locked_range(vol, bp)) { 1150219974Smav G_RAID_LOGREQ(3, bp, "Defer request."); 1151219974Smav bioq_insert_tail(&vol->v_locked, bp); 1152219974Smav return; 1153219974Smav } 1154219974Smav 1155219974Smav /* 1156219974Smav * If we're actually going to do the write/delete, then 1157219974Smav * update the idle stats for the volume. 1158219974Smav */ 1159219974Smav if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1160219974Smav if (!vol->v_dirty) 1161219974Smav g_raid_dirty(vol); 1162219974Smav vol->v_writes++; 1163219974Smav } 1164219974Smav 1165219974Smav /* 1166219974Smav * Put request onto inflight queue, so we can check if new 1167219974Smav * synchronization requests don't collide with it. Then tell 1168219974Smav * the transformation layer to start the I/O. 1169219974Smav */ 1170219974Smav bioq_insert_tail(&vol->v_inflight, bp); 1171219974Smav G_RAID_LOGREQ(4, bp, "Request started"); 1172219974Smav G_RAID_TR_IOSTART(vol->v_tr, bp); 1173219974Smav} 1174219974Smav 1175219974Smavstatic void 1176219974Smavg_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) 1177219974Smav{ 1178219974Smav off_t off, len; 1179219974Smav struct bio *nbp; 1180219974Smav struct g_raid_lock *lp; 1181219974Smav 1182219974Smav vol->v_pending_lock = 0; 1183219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1184219974Smav if (lp->l_pending) { 1185219974Smav off = lp->l_offset; 1186219974Smav len = lp->l_length; 1187219974Smav lp->l_pending = 0; 1188219974Smav TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { 1189219974Smav if (g_raid_bio_overlaps(nbp, off, len)) 1190219974Smav lp->l_pending++; 1191219974Smav } 1192219974Smav if (lp->l_pending) { 1193219974Smav vol->v_pending_lock = 1; 1194219974Smav G_RAID_DEBUG1(4, vol->v_softc, 1195219974Smav "Deferred lock(%jd, %jd) has %d pending", 1196219974Smav (intmax_t)off, (intmax_t)(off + len), 1197219974Smav lp->l_pending); 1198219974Smav continue; 1199219974Smav } 1200219974Smav G_RAID_DEBUG1(4, vol->v_softc, 1201219974Smav "Deferred lock of %jd to %jd completed", 1202219974Smav (intmax_t)off, (intmax_t)(off + len)); 1203219974Smav G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1204219974Smav } 1205219974Smav } 1206219974Smav} 1207219974Smav 1208219974Smavvoid 1209219974Smavg_raid_iodone(struct bio *bp, int error) 1210219974Smav{ 1211219974Smav struct g_raid_softc *sc; 1212219974Smav struct g_raid_volume *vol; 1213219974Smav 1214219974Smav sc = bp->bio_to->geom->softc; 1215219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1216219974Smav vol = bp->bio_to->private; 1217219974Smav G_RAID_LOGREQ(3, bp, "Request done: %d.", error); 1218219974Smav 1219219974Smav /* Update stats if we done write/delete. */ 1220219974Smav if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1221219974Smav vol->v_writes--; 1222219974Smav vol->v_last_write = time_uptime; 1223219974Smav } 1224219974Smav 1225219974Smav bioq_remove(&vol->v_inflight, bp); 1226219974Smav if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) 1227219974Smav g_raid_finish_with_locked_ranges(vol, bp); 1228219974Smav getmicrouptime(&vol->v_last_done); 1229219974Smav g_io_deliver(bp, error); 1230219974Smav} 1231219974Smav 1232219974Smavint 1233219974Smavg_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, 1234219974Smav struct bio *ignore, void *argp) 1235219974Smav{ 1236219974Smav struct g_raid_softc *sc; 1237219974Smav struct g_raid_lock *lp; 1238219974Smav struct bio *bp; 1239219974Smav 1240219974Smav sc = vol->v_softc; 1241219974Smav lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); 1242219974Smav LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); 1243219974Smav lp->l_offset = off; 1244219974Smav lp->l_length = len; 1245219974Smav lp->l_callback_arg = argp; 1246219974Smav 1247219974Smav lp->l_pending = 0; 1248219974Smav TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { 1249219974Smav if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) 1250219974Smav lp->l_pending++; 1251219974Smav } 1252219974Smav 1253219974Smav /* 1254219974Smav * If there are any writes that are pending, we return EBUSY. All 1255219974Smav * callers will have to wait until all pending writes clear. 1256219974Smav */ 1257219974Smav if (lp->l_pending > 0) { 1258219974Smav vol->v_pending_lock = 1; 1259219974Smav G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", 1260219974Smav (intmax_t)off, (intmax_t)(off+len), lp->l_pending); 1261219974Smav return (EBUSY); 1262219974Smav } 1263219974Smav G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", 1264219974Smav (intmax_t)off, (intmax_t)(off+len)); 1265219974Smav G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1266219974Smav return (0); 1267219974Smav} 1268219974Smav 1269219974Smavint 1270219974Smavg_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) 1271219974Smav{ 1272219974Smav struct g_raid_lock *lp; 1273219974Smav struct g_raid_softc *sc; 1274219974Smav struct bio *bp; 1275219974Smav 1276219974Smav sc = vol->v_softc; 1277219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1278219974Smav if (lp->l_offset == off && lp->l_length == len) { 1279219974Smav LIST_REMOVE(lp, l_next); 1280219974Smav /* XXX 1281219974Smav * Right now we just put them all back on the queue 1282219974Smav * and hope for the best. We hope this because any 1283219974Smav * locked ranges will go right back on this list 1284219974Smav * when the worker thread runs. 1285219974Smav * XXX 1286219974Smav */ 1287219974Smav G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", 1288219974Smav (intmax_t)lp->l_offset, 1289219974Smav (intmax_t)(lp->l_offset+lp->l_length)); 1290219974Smav mtx_lock(&sc->sc_queue_mtx); 1291219974Smav while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) 1292219974Smav bioq_disksort(&sc->sc_queue, bp); 1293219974Smav mtx_unlock(&sc->sc_queue_mtx); 1294219974Smav free(lp, M_RAID); 1295219974Smav return (0); 1296219974Smav } 1297219974Smav } 1298219974Smav return (EINVAL); 1299219974Smav} 1300219974Smav 1301219974Smavvoid 1302219974Smavg_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) 1303219974Smav{ 1304219974Smav struct g_consumer *cp; 1305219974Smav struct g_raid_disk *disk, *tdisk; 1306219974Smav 1307219974Smav bp->bio_caller1 = sd; 1308219974Smav 1309219974Smav /* 1310219974Smav * Make sure that the disk is present. Generally it is a task of 1311219974Smav * transformation layers to not send requests to absent disks, but 1312219974Smav * it is better to be safe and report situation then sorry. 1313219974Smav */ 1314219974Smav if (sd->sd_disk == NULL) { 1315219974Smav G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); 1316219974Smavnodisk: 1317219974Smav bp->bio_from = NULL; 1318219974Smav bp->bio_to = NULL; 1319219974Smav bp->bio_error = ENXIO; 1320219974Smav g_raid_disk_done(bp); 1321219974Smav return; 1322219974Smav } 1323219974Smav disk = sd->sd_disk; 1324219974Smav if (disk->d_state != G_RAID_DISK_S_ACTIVE && 1325219974Smav disk->d_state != G_RAID_DISK_S_FAILED) { 1326219974Smav G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " 1327219974Smav "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 1328219974Smav goto nodisk; 1329219974Smav } 1330219974Smav 1331219974Smav cp = disk->d_consumer; 1332219974Smav bp->bio_from = cp; 1333219974Smav bp->bio_to = cp->provider; 1334219974Smav cp->index++; 1335219974Smav 1336219974Smav /* Update average disks load. */ 1337219974Smav TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { 1338219974Smav if (tdisk->d_consumer == NULL) 1339219974Smav tdisk->d_load = 0; 1340219974Smav else 1341219974Smav tdisk->d_load = (tdisk->d_consumer->index * 1342219974Smav G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; 1343219974Smav } 1344219974Smav 1345219974Smav disk->d_last_offset = bp->bio_offset + bp->bio_length; 1346219974Smav if (dumping) { 1347219974Smav G_RAID_LOGREQ(3, bp, "Sending dumping request."); 1348219974Smav if (bp->bio_cmd == BIO_WRITE) { 1349219974Smav bp->bio_error = g_raid_subdisk_kerneldump(sd, 1350219974Smav bp->bio_data, 0, bp->bio_offset, bp->bio_length); 1351219974Smav } else 1352219974Smav bp->bio_error = EOPNOTSUPP; 1353219974Smav g_raid_disk_done(bp); 1354219974Smav } else { 1355219974Smav bp->bio_done = g_raid_disk_done; 1356219974Smav bp->bio_offset += sd->sd_offset; 1357219974Smav G_RAID_LOGREQ(3, bp, "Sending request."); 1358219974Smav g_io_request(bp, cp); 1359219974Smav } 1360219974Smav} 1361219974Smav 1362219974Smavint 1363219974Smavg_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, 1364219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1365219974Smav{ 1366219974Smav 1367219974Smav if (sd->sd_disk == NULL) 1368219974Smav return (ENXIO); 1369219974Smav if (sd->sd_disk->d_kd.di.dumper == NULL) 1370219974Smav return (EOPNOTSUPP); 1371219974Smav return (dump_write(&sd->sd_disk->d_kd.di, 1372219974Smav virtual, physical, 1373219974Smav sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, 1374219974Smav length)); 1375219974Smav} 1376219974Smav 1377219974Smavstatic void 1378219974Smavg_raid_disk_done(struct bio *bp) 1379219974Smav{ 1380219974Smav struct g_raid_softc *sc; 1381219974Smav struct g_raid_subdisk *sd; 1382219974Smav 1383219974Smav sd = bp->bio_caller1; 1384219974Smav sc = sd->sd_softc; 1385219974Smav mtx_lock(&sc->sc_queue_mtx); 1386219974Smav bioq_disksort(&sc->sc_queue, bp); 1387219974Smav mtx_unlock(&sc->sc_queue_mtx); 1388219974Smav if (!dumping) 1389219974Smav wakeup(sc); 1390219974Smav} 1391219974Smav 1392219974Smavstatic void 1393219974Smavg_raid_disk_done_request(struct bio *bp) 1394219974Smav{ 1395219974Smav struct g_raid_softc *sc; 1396219974Smav struct g_raid_disk *disk; 1397219974Smav struct g_raid_subdisk *sd; 1398219974Smav struct g_raid_volume *vol; 1399219974Smav 1400219974Smav g_topology_assert_not(); 1401219974Smav 1402219974Smav G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); 1403219974Smav sd = bp->bio_caller1; 1404219974Smav sc = sd->sd_softc; 1405219974Smav vol = sd->sd_volume; 1406219974Smav if (bp->bio_from != NULL) { 1407219974Smav bp->bio_from->index--; 1408219974Smav disk = bp->bio_from->private; 1409219974Smav if (disk == NULL) 1410219974Smav g_raid_kill_consumer(sc, bp->bio_from); 1411219974Smav } 1412219974Smav bp->bio_offset -= sd->sd_offset; 1413219974Smav 1414219974Smav G_RAID_TR_IODONE(vol->v_tr, sd, bp); 1415219974Smav} 1416219974Smav 1417219974Smavstatic void 1418219974Smavg_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) 1419219974Smav{ 1420219974Smav 1421219974Smav if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) 1422219974Smav ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); 1423219974Smav else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) 1424219974Smav ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); 1425219974Smav else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) 1426219974Smav ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); 1427219974Smav else 1428219974Smav ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); 1429219974Smav if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { 1430219974Smav KASSERT(ep->e_error == 0, 1431219974Smav ("Error cannot be handled.")); 1432219974Smav g_raid_event_free(ep); 1433219974Smav } else { 1434219974Smav ep->e_flags |= G_RAID_EVENT_DONE; 1435219974Smav G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); 1436219974Smav mtx_lock(&sc->sc_queue_mtx); 1437219974Smav wakeup(ep); 1438219974Smav mtx_unlock(&sc->sc_queue_mtx); 1439219974Smav } 1440219974Smav} 1441219974Smav 1442219974Smav/* 1443219974Smav * Worker thread. 1444219974Smav */ 1445219974Smavstatic void 1446219974Smavg_raid_worker(void *arg) 1447219974Smav{ 1448219974Smav struct g_raid_softc *sc; 1449219974Smav struct g_raid_event *ep; 1450219974Smav struct g_raid_volume *vol; 1451219974Smav struct bio *bp; 1452219974Smav struct timeval now, t; 1453219974Smav int timeout, rv; 1454219974Smav 1455219974Smav sc = arg; 1456219974Smav thread_lock(curthread); 1457219974Smav sched_prio(curthread, PRIBIO); 1458219974Smav thread_unlock(curthread); 1459219974Smav 1460219974Smav sx_xlock(&sc->sc_lock); 1461219974Smav for (;;) { 1462219974Smav mtx_lock(&sc->sc_queue_mtx); 1463219974Smav /* 1464219974Smav * First take a look at events. 1465219974Smav * This is important to handle events before any I/O requests. 1466219974Smav */ 1467219974Smav bp = NULL; 1468219974Smav vol = NULL; 1469219974Smav rv = 0; 1470219974Smav ep = TAILQ_FIRST(&sc->sc_events); 1471219974Smav if (ep != NULL) 1472219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1473219974Smav else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) 1474219974Smav ; 1475219974Smav else { 1476219974Smav getmicrouptime(&now); 1477219974Smav t = now; 1478219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1479219974Smav if (bioq_first(&vol->v_inflight) == NULL && 1480219974Smav vol->v_tr && 1481219974Smav timevalcmp(&vol->v_last_done, &t, < )) 1482219974Smav t = vol->v_last_done; 1483219974Smav } 1484219974Smav timevalsub(&t, &now); 1485219974Smav timeout = g_raid_idle_threshold + 1486219974Smav t.tv_sec * 1000000 + t.tv_usec; 1487219974Smav if (timeout > 0) { 1488219974Smav /* 1489219974Smav * Two steps to avoid overflows at HZ=1000 1490219974Smav * and idle timeouts > 2.1s. Some rounding 1491219974Smav * errors can occur, but they are < 1tick, 1492219974Smav * which is deemed to be close enough for 1493219974Smav * this purpose. 1494219974Smav */ 1495219974Smav int micpertic = 1000000 / hz; 1496219974Smav timeout = (timeout + micpertic - 1) / micpertic; 1497219974Smav sx_xunlock(&sc->sc_lock); 1498219974Smav MSLEEP(rv, sc, &sc->sc_queue_mtx, 1499219974Smav PRIBIO | PDROP, "-", timeout); 1500219974Smav sx_xlock(&sc->sc_lock); 1501219974Smav goto process; 1502219974Smav } else 1503219974Smav rv = EWOULDBLOCK; 1504219974Smav } 1505219974Smav mtx_unlock(&sc->sc_queue_mtx); 1506219974Smavprocess: 1507219974Smav if (ep != NULL) { 1508219974Smav g_raid_handle_event(sc, ep); 1509219974Smav } else if (bp != NULL) { 1510219974Smav if (bp->bio_to != NULL && 1511219974Smav bp->bio_to->geom == sc->sc_geom) 1512219974Smav g_raid_start_request(bp); 1513219974Smav else 1514219974Smav g_raid_disk_done_request(bp); 1515219974Smav } else if (rv == EWOULDBLOCK) { 1516219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1517219974Smav if (vol->v_writes == 0 && vol->v_dirty) 1518219974Smav g_raid_clean(vol, -1); 1519219974Smav if (bioq_first(&vol->v_inflight) == NULL && 1520219974Smav vol->v_tr) { 1521219974Smav t.tv_sec = g_raid_idle_threshold / 1000000; 1522219974Smav t.tv_usec = g_raid_idle_threshold % 1000000; 1523219974Smav timevaladd(&t, &vol->v_last_done); 1524219974Smav getmicrouptime(&now); 1525219974Smav if (timevalcmp(&t, &now, <= )) { 1526219974Smav G_RAID_TR_IDLE(vol->v_tr); 1527219974Smav vol->v_last_done = now; 1528219974Smav } 1529219974Smav } 1530219974Smav } 1531219974Smav } 1532219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) 1533219974Smav g_raid_destroy_node(sc, 1); /* May not return. */ 1534219974Smav } 1535219974Smav} 1536219974Smav 1537219974Smavstatic void 1538219974Smavg_raid_poll(struct g_raid_softc *sc) 1539219974Smav{ 1540219974Smav struct g_raid_event *ep; 1541219974Smav struct bio *bp; 1542219974Smav 1543219974Smav sx_xlock(&sc->sc_lock); 1544219974Smav mtx_lock(&sc->sc_queue_mtx); 1545219974Smav /* 1546219974Smav * First take a look at events. 1547219974Smav * This is important to handle events before any I/O requests. 1548219974Smav */ 1549219974Smav ep = TAILQ_FIRST(&sc->sc_events); 1550219974Smav if (ep != NULL) { 1551219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1552219974Smav mtx_unlock(&sc->sc_queue_mtx); 1553219974Smav g_raid_handle_event(sc, ep); 1554219974Smav goto out; 1555219974Smav } 1556219974Smav bp = bioq_takefirst(&sc->sc_queue); 1557219974Smav if (bp != NULL) { 1558219974Smav mtx_unlock(&sc->sc_queue_mtx); 1559219974Smav if (bp->bio_from == NULL || 1560219974Smav bp->bio_from->geom != sc->sc_geom) 1561219974Smav g_raid_start_request(bp); 1562219974Smav else 1563219974Smav g_raid_disk_done_request(bp); 1564219974Smav } 1565219974Smavout: 1566219974Smav sx_xunlock(&sc->sc_lock); 1567219974Smav} 1568219974Smav 1569219974Smavstatic void 1570219974Smavg_raid_launch_provider(struct g_raid_volume *vol) 1571219974Smav{ 1572219974Smav struct g_raid_disk *disk; 1573219974Smav struct g_raid_softc *sc; 1574219974Smav struct g_provider *pp; 1575219974Smav char name[G_RAID_MAX_VOLUMENAME]; 1576219974Smav off_t off; 1577219974Smav 1578219974Smav sc = vol->v_softc; 1579219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1580219974Smav 1581219974Smav g_topology_lock(); 1582219974Smav /* Try to name provider with volume name. */ 1583219974Smav snprintf(name, sizeof(name), "raid/%s", vol->v_name); 1584219974Smav if (g_raid_name_format == 0 || vol->v_name[0] == 0 || 1585219974Smav g_provider_by_name(name) != NULL) { 1586219974Smav /* Otherwise use sequential volume number. */ 1587219974Smav snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); 1588219974Smav } 1589219974Smav pp = g_new_providerf(sc->sc_geom, "%s", name); 1590219974Smav pp->private = vol; 1591219974Smav pp->mediasize = vol->v_mediasize; 1592219974Smav pp->sectorsize = vol->v_sectorsize; 1593219974Smav pp->stripesize = 0; 1594219974Smav pp->stripeoffset = 0; 1595219974Smav if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 1596219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || 1597219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || 1598219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { 1599219974Smav if ((disk = vol->v_subdisks[0].sd_disk) != NULL && 1600219974Smav disk->d_consumer != NULL && 1601219974Smav disk->d_consumer->provider != NULL) { 1602219974Smav pp->stripesize = disk->d_consumer->provider->stripesize; 1603219974Smav off = disk->d_consumer->provider->stripeoffset; 1604219974Smav pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; 1605219974Smav if (off > 0) 1606219974Smav pp->stripeoffset %= off; 1607219974Smav } 1608219974Smav if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { 1609219974Smav pp->stripesize *= (vol->v_disks_count - 1); 1610219974Smav pp->stripeoffset *= (vol->v_disks_count - 1); 1611219974Smav } 1612219974Smav } else 1613219974Smav pp->stripesize = vol->v_strip_size; 1614219974Smav vol->v_provider = pp; 1615219974Smav g_error_provider(pp, 0); 1616219974Smav g_topology_unlock(); 1617219974Smav G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", 1618219974Smav pp->name, vol->v_name); 1619219974Smav} 1620219974Smav 1621219974Smavstatic void 1622219974Smavg_raid_destroy_provider(struct g_raid_volume *vol) 1623219974Smav{ 1624219974Smav struct g_raid_softc *sc; 1625219974Smav struct g_provider *pp; 1626219974Smav struct bio *bp, *tmp; 1627219974Smav 1628219974Smav g_topology_assert_not(); 1629219974Smav sc = vol->v_softc; 1630219974Smav pp = vol->v_provider; 1631219974Smav KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); 1632219974Smav 1633219974Smav g_topology_lock(); 1634219974Smav g_error_provider(pp, ENXIO); 1635219974Smav mtx_lock(&sc->sc_queue_mtx); 1636219974Smav TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { 1637219974Smav if (bp->bio_to != pp) 1638219974Smav continue; 1639219974Smav bioq_remove(&sc->sc_queue, bp); 1640219974Smav g_io_deliver(bp, ENXIO); 1641219974Smav } 1642219974Smav mtx_unlock(&sc->sc_queue_mtx); 1643219974Smav G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", 1644219974Smav pp->name, vol->v_name); 1645219974Smav g_wither_provider(pp, ENXIO); 1646219974Smav g_topology_unlock(); 1647219974Smav vol->v_provider = NULL; 1648219974Smav} 1649219974Smav 1650219974Smav/* 1651219974Smav * Update device state. 1652219974Smav */ 1653219974Smavstatic int 1654219974Smavg_raid_update_volume(struct g_raid_volume *vol, u_int event) 1655219974Smav{ 1656219974Smav struct g_raid_softc *sc; 1657219974Smav 1658219974Smav sc = vol->v_softc; 1659219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1660219974Smav 1661219974Smav G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", 1662219974Smav g_raid_volume_event2str(event), 1663219974Smav vol->v_name); 1664219974Smav switch (event) { 1665219974Smav case G_RAID_VOLUME_E_DOWN: 1666219974Smav if (vol->v_provider != NULL) 1667219974Smav g_raid_destroy_provider(vol); 1668219974Smav break; 1669219974Smav case G_RAID_VOLUME_E_UP: 1670219974Smav if (vol->v_provider == NULL) 1671219974Smav g_raid_launch_provider(vol); 1672219974Smav break; 1673219974Smav case G_RAID_VOLUME_E_START: 1674219974Smav if (vol->v_tr) 1675219974Smav G_RAID_TR_START(vol->v_tr); 1676219974Smav return (0); 1677219974Smav default: 1678219974Smav if (sc->sc_md) 1679219974Smav G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); 1680219974Smav return (0); 1681219974Smav } 1682219974Smav 1683219974Smav /* Manage root mount release. */ 1684219974Smav if (vol->v_starting) { 1685219974Smav vol->v_starting = 0; 1686219974Smav G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); 1687219974Smav root_mount_rel(vol->v_rootmount); 1688219974Smav vol->v_rootmount = NULL; 1689219974Smav } 1690219974Smav if (vol->v_stopping && vol->v_provider_open == 0) 1691219974Smav g_raid_destroy_volume(vol); 1692219974Smav return (0); 1693219974Smav} 1694219974Smav 1695219974Smav/* 1696219974Smav * Update subdisk state. 1697219974Smav */ 1698219974Smavstatic int 1699219974Smavg_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) 1700219974Smav{ 1701219974Smav struct g_raid_softc *sc; 1702219974Smav struct g_raid_volume *vol; 1703219974Smav 1704219974Smav sc = sd->sd_softc; 1705219974Smav vol = sd->sd_volume; 1706219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1707219974Smav 1708219974Smav G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", 1709219974Smav g_raid_subdisk_event2str(event), 1710219974Smav vol->v_name, sd->sd_pos, 1711219974Smav sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); 1712219974Smav if (vol->v_tr) 1713219974Smav G_RAID_TR_EVENT(vol->v_tr, sd, event); 1714219974Smav 1715219974Smav return (0); 1716219974Smav} 1717219974Smav 1718219974Smav/* 1719219974Smav * Update disk state. 1720219974Smav */ 1721219974Smavstatic int 1722219974Smavg_raid_update_disk(struct g_raid_disk *disk, u_int event) 1723219974Smav{ 1724219974Smav struct g_raid_softc *sc; 1725219974Smav 1726219974Smav sc = disk->d_softc; 1727219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1728219974Smav 1729219974Smav G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", 1730219974Smav g_raid_disk_event2str(event), 1731219974Smav g_raid_get_diskname(disk)); 1732219974Smav 1733219974Smav if (sc->sc_md) 1734219974Smav G_RAID_MD_EVENT(sc->sc_md, disk, event); 1735219974Smav return (0); 1736219974Smav} 1737219974Smav 1738219974Smav/* 1739219974Smav * Node event. 1740219974Smav */ 1741219974Smavstatic int 1742219974Smavg_raid_update_node(struct g_raid_softc *sc, u_int event) 1743219974Smav{ 1744219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1745219974Smav 1746219974Smav G_RAID_DEBUG1(2, sc, "Event %s for the array.", 1747219974Smav g_raid_node_event2str(event)); 1748219974Smav 1749219974Smav if (event == G_RAID_NODE_E_WAKE) 1750219974Smav return (0); 1751219974Smav if (sc->sc_md) 1752219974Smav G_RAID_MD_EVENT(sc->sc_md, NULL, event); 1753219974Smav return (0); 1754219974Smav} 1755219974Smav 1756219974Smavstatic int 1757219974Smavg_raid_access(struct g_provider *pp, int acr, int acw, int ace) 1758219974Smav{ 1759219974Smav struct g_raid_volume *vol; 1760219974Smav struct g_raid_softc *sc; 1761220210Smav int dcw, opens, error = 0; 1762219974Smav 1763219974Smav g_topology_assert(); 1764219974Smav sc = pp->geom->softc; 1765219974Smav vol = pp->private; 1766219974Smav KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 1767219974Smav KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); 1768219974Smav 1769219974Smav G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, 1770219974Smav acr, acw, ace); 1771219974Smav dcw = pp->acw + acw; 1772219974Smav 1773219974Smav g_topology_unlock(); 1774219974Smav sx_xlock(&sc->sc_lock); 1775219974Smav /* Deny new opens while dying. */ 1776219974Smav if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { 1777219974Smav error = ENXIO; 1778219974Smav goto out; 1779219974Smav } 1780219974Smav if (dcw == 0 && vol->v_dirty) 1781219974Smav g_raid_clean(vol, dcw); 1782219974Smav vol->v_provider_open += acr + acw + ace; 1783219974Smav /* Handle delayed node destruction. */ 1784219974Smav if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && 1785219974Smav vol->v_provider_open == 0) { 1786219974Smav /* Count open volumes. */ 1787219974Smav opens = g_raid_nopens(sc); 1788219974Smav if (opens == 0) { 1789219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 1790219974Smav /* Wake up worker to make it selfdestruct. */ 1791219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1792219974Smav } 1793219974Smav } 1794219974Smav /* Handle open volume destruction. */ 1795219974Smav if (vol->v_stopping && vol->v_provider_open == 0) 1796219974Smav g_raid_destroy_volume(vol); 1797219974Smavout: 1798219974Smav sx_xunlock(&sc->sc_lock); 1799219974Smav g_topology_lock(); 1800219974Smav return (error); 1801219974Smav} 1802219974Smav 1803219974Smavstruct g_raid_softc * 1804219974Smavg_raid_create_node(struct g_class *mp, 1805219974Smav const char *name, struct g_raid_md_object *md) 1806219974Smav{ 1807219974Smav struct g_raid_softc *sc; 1808219974Smav struct g_geom *gp; 1809219974Smav int error; 1810219974Smav 1811219974Smav g_topology_assert(); 1812219974Smav G_RAID_DEBUG(1, "Creating array %s.", name); 1813219974Smav 1814219974Smav gp = g_new_geomf(mp, "%s", name); 1815219974Smav sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); 1816219974Smav gp->start = g_raid_start; 1817219974Smav gp->orphan = g_raid_orphan; 1818219974Smav gp->access = g_raid_access; 1819219974Smav gp->dumpconf = g_raid_dumpconf; 1820219974Smav 1821219974Smav sc->sc_md = md; 1822219974Smav sc->sc_geom = gp; 1823219974Smav sc->sc_flags = 0; 1824219974Smav TAILQ_INIT(&sc->sc_volumes); 1825219974Smav TAILQ_INIT(&sc->sc_disks); 1826219974Smav sx_init(&sc->sc_lock, "gmirror:lock"); 1827219974Smav mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 1828219974Smav TAILQ_INIT(&sc->sc_events); 1829219974Smav bioq_init(&sc->sc_queue); 1830219974Smav gp->softc = sc; 1831219974Smav error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, 1832219974Smav "g_raid %s", name); 1833219974Smav if (error != 0) { 1834219974Smav G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); 1835219974Smav mtx_destroy(&sc->sc_queue_mtx); 1836219974Smav sx_destroy(&sc->sc_lock); 1837219974Smav g_destroy_geom(sc->sc_geom); 1838219974Smav free(sc, M_RAID); 1839219974Smav return (NULL); 1840219974Smav } 1841219974Smav 1842219974Smav G_RAID_DEBUG1(0, sc, "Array %s created.", name); 1843219974Smav return (sc); 1844219974Smav} 1845219974Smav 1846219974Smavstruct g_raid_volume * 1847219974Smavg_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) 1848219974Smav{ 1849219974Smav struct g_raid_volume *vol, *vol1; 1850219974Smav int i; 1851219974Smav 1852219974Smav G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); 1853219974Smav vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); 1854219974Smav vol->v_softc = sc; 1855219974Smav strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); 1856219974Smav vol->v_state = G_RAID_VOLUME_S_STARTING; 1857219974Smav vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 1858219974Smav vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; 1859219974Smav bioq_init(&vol->v_inflight); 1860219974Smav bioq_init(&vol->v_locked); 1861219974Smav LIST_INIT(&vol->v_locks); 1862219974Smav for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 1863219974Smav vol->v_subdisks[i].sd_softc = sc; 1864219974Smav vol->v_subdisks[i].sd_volume = vol; 1865219974Smav vol->v_subdisks[i].sd_pos = i; 1866219974Smav vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; 1867219974Smav } 1868219974Smav 1869219974Smav /* Find free ID for this volume. */ 1870219974Smav g_topology_lock(); 1871219974Smav vol1 = vol; 1872219974Smav if (id >= 0) { 1873219974Smav LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1874219974Smav if (vol1->v_global_id == id) 1875219974Smav break; 1876219974Smav } 1877219974Smav } 1878219974Smav if (vol1 != NULL) { 1879219974Smav for (id = 0; ; id++) { 1880219974Smav LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1881219974Smav if (vol1->v_global_id == id) 1882219974Smav break; 1883219974Smav } 1884219974Smav if (vol1 == NULL) 1885219974Smav break; 1886219974Smav } 1887219974Smav } 1888219974Smav vol->v_global_id = id; 1889219974Smav LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); 1890219974Smav g_topology_unlock(); 1891219974Smav 1892219974Smav /* Delay root mounting. */ 1893219974Smav vol->v_rootmount = root_mount_hold("GRAID"); 1894219974Smav G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); 1895219974Smav vol->v_starting = 1; 1896219974Smav TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); 1897219974Smav return (vol); 1898219974Smav} 1899219974Smav 1900219974Smavstruct g_raid_disk * 1901219974Smavg_raid_create_disk(struct g_raid_softc *sc) 1902219974Smav{ 1903219974Smav struct g_raid_disk *disk; 1904219974Smav 1905219974Smav G_RAID_DEBUG1(1, sc, "Creating disk."); 1906219974Smav disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); 1907219974Smav disk->d_softc = sc; 1908219974Smav disk->d_state = G_RAID_DISK_S_NONE; 1909219974Smav TAILQ_INIT(&disk->d_subdisks); 1910219974Smav TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); 1911219974Smav return (disk); 1912219974Smav} 1913219974Smav 1914219974Smavint g_raid_start_volume(struct g_raid_volume *vol) 1915219974Smav{ 1916219974Smav struct g_raid_tr_class *class; 1917219974Smav struct g_raid_tr_object *obj; 1918219974Smav int status; 1919219974Smav 1920219974Smav G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); 1921219974Smav LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { 1922219974Smav G_RAID_DEBUG1(2, vol->v_softc, 1923219974Smav "Tasting volume %s for %s transformation.", 1924219974Smav vol->v_name, class->name); 1925219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 1926219974Smav M_WAITOK); 1927219974Smav obj->tro_class = class; 1928219974Smav obj->tro_volume = vol; 1929219974Smav status = G_RAID_TR_TASTE(obj, vol); 1930219974Smav if (status != G_RAID_TR_TASTE_FAIL) 1931219974Smav break; 1932219974Smav kobj_delete((kobj_t)obj, M_RAID); 1933219974Smav } 1934219974Smav if (class == NULL) { 1935219974Smav G_RAID_DEBUG1(0, vol->v_softc, 1936219974Smav "No transformation module found for %s.", 1937219974Smav vol->v_name); 1938219974Smav vol->v_tr = NULL; 1939219974Smav g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); 1940219974Smav g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, 1941219974Smav G_RAID_EVENT_VOLUME); 1942219974Smav return (-1); 1943219974Smav } 1944219974Smav G_RAID_DEBUG1(2, vol->v_softc, 1945219974Smav "Transformation module %s chosen for %s.", 1946219974Smav class->name, vol->v_name); 1947219974Smav vol->v_tr = obj; 1948219974Smav return (0); 1949219974Smav} 1950219974Smav 1951219974Smavint 1952219974Smavg_raid_destroy_node(struct g_raid_softc *sc, int worker) 1953219974Smav{ 1954219974Smav struct g_raid_volume *vol, *tmpv; 1955219974Smav struct g_raid_disk *disk, *tmpd; 1956219974Smav int error = 0; 1957219974Smav 1958219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 1959219974Smav TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { 1960219974Smav if (g_raid_destroy_volume(vol)) 1961219974Smav error = EBUSY; 1962219974Smav } 1963219974Smav if (error) 1964219974Smav return (error); 1965219974Smav TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { 1966219974Smav if (g_raid_destroy_disk(disk)) 1967219974Smav error = EBUSY; 1968219974Smav } 1969219974Smav if (error) 1970219974Smav return (error); 1971219974Smav if (sc->sc_md) { 1972219974Smav G_RAID_MD_FREE(sc->sc_md); 1973219974Smav kobj_delete((kobj_t)sc->sc_md, M_RAID); 1974219974Smav sc->sc_md = NULL; 1975219974Smav } 1976219974Smav if (sc->sc_geom != NULL) { 1977219974Smav G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); 1978219974Smav g_topology_lock(); 1979219974Smav sc->sc_geom->softc = NULL; 1980219974Smav g_wither_geom(sc->sc_geom, ENXIO); 1981219974Smav g_topology_unlock(); 1982219974Smav sc->sc_geom = NULL; 1983219974Smav } else 1984219974Smav G_RAID_DEBUG(1, "Array destroyed."); 1985219974Smav if (worker) { 1986219974Smav g_raid_event_cancel(sc, sc); 1987219974Smav mtx_destroy(&sc->sc_queue_mtx); 1988219974Smav sx_xunlock(&sc->sc_lock); 1989219974Smav sx_destroy(&sc->sc_lock); 1990219974Smav wakeup(&sc->sc_stopping); 1991219974Smav free(sc, M_RAID); 1992219974Smav curthread->td_pflags &= ~TDP_GEOM; 1993219974Smav G_RAID_DEBUG(1, "Thread exiting."); 1994219974Smav kproc_exit(0); 1995219974Smav } else { 1996219974Smav /* Wake up worker to make it selfdestruct. */ 1997219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1998219974Smav } 1999219974Smav return (0); 2000219974Smav} 2001219974Smav 2002219974Smavint 2003219974Smavg_raid_destroy_volume(struct g_raid_volume *vol) 2004219974Smav{ 2005219974Smav struct g_raid_softc *sc; 2006219974Smav struct g_raid_disk *disk; 2007219974Smav int i; 2008219974Smav 2009219974Smav sc = vol->v_softc; 2010219974Smav G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); 2011219974Smav vol->v_stopping = 1; 2012219974Smav if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { 2013219974Smav if (vol->v_tr) { 2014219974Smav G_RAID_TR_STOP(vol->v_tr); 2015219974Smav return (EBUSY); 2016219974Smav } else 2017219974Smav vol->v_state = G_RAID_VOLUME_S_STOPPED; 2018219974Smav } 2019219974Smav if (g_raid_event_check(sc, vol) != 0) 2020219974Smav return (EBUSY); 2021219974Smav if (vol->v_provider != NULL) 2022219974Smav return (EBUSY); 2023219974Smav if (vol->v_provider_open != 0) 2024219974Smav return (EBUSY); 2025219974Smav if (vol->v_tr) { 2026219974Smav G_RAID_TR_FREE(vol->v_tr); 2027219974Smav kobj_delete((kobj_t)vol->v_tr, M_RAID); 2028219974Smav vol->v_tr = NULL; 2029219974Smav } 2030219974Smav if (vol->v_rootmount) 2031219974Smav root_mount_rel(vol->v_rootmount); 2032219974Smav g_topology_lock(); 2033219974Smav LIST_REMOVE(vol, v_global_next); 2034219974Smav g_topology_unlock(); 2035219974Smav TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); 2036219974Smav for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 2037219974Smav g_raid_event_cancel(sc, &vol->v_subdisks[i]); 2038219974Smav disk = vol->v_subdisks[i].sd_disk; 2039219974Smav if (disk == NULL) 2040219974Smav continue; 2041219974Smav TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); 2042219974Smav } 2043219974Smav G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); 2044219974Smav if (sc->sc_md) 2045219974Smav G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); 2046219974Smav g_raid_event_cancel(sc, vol); 2047219974Smav free(vol, M_RAID); 2048219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) { 2049219974Smav /* Wake up worker to let it selfdestruct. */ 2050219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2051219974Smav } 2052219974Smav return (0); 2053219974Smav} 2054219974Smav 2055219974Smavint 2056219974Smavg_raid_destroy_disk(struct g_raid_disk *disk) 2057219974Smav{ 2058219974Smav struct g_raid_softc *sc; 2059219974Smav struct g_raid_subdisk *sd, *tmp; 2060219974Smav 2061219974Smav sc = disk->d_softc; 2062219974Smav G_RAID_DEBUG1(2, sc, "Destroying disk."); 2063219974Smav if (disk->d_consumer) { 2064219974Smav g_raid_kill_consumer(sc, disk->d_consumer); 2065219974Smav disk->d_consumer = NULL; 2066219974Smav } 2067219974Smav TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { 2068219974Smav g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); 2069219974Smav g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 2070219974Smav G_RAID_EVENT_SUBDISK); 2071219974Smav TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); 2072219974Smav sd->sd_disk = NULL; 2073219974Smav } 2074219974Smav TAILQ_REMOVE(&sc->sc_disks, disk, d_next); 2075219974Smav if (sc->sc_md) 2076219974Smav G_RAID_MD_FREE_DISK(sc->sc_md, disk); 2077219974Smav g_raid_event_cancel(sc, disk); 2078219974Smav free(disk, M_RAID); 2079219974Smav return (0); 2080219974Smav} 2081219974Smav 2082219974Smavint 2083219974Smavg_raid_destroy(struct g_raid_softc *sc, int how) 2084219974Smav{ 2085219974Smav int opens; 2086219974Smav 2087219974Smav g_topology_assert_not(); 2088219974Smav if (sc == NULL) 2089219974Smav return (ENXIO); 2090219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 2091219974Smav 2092219974Smav /* Count open volumes. */ 2093219974Smav opens = g_raid_nopens(sc); 2094219974Smav 2095219974Smav /* React on some opened volumes. */ 2096219974Smav if (opens > 0) { 2097219974Smav switch (how) { 2098219974Smav case G_RAID_DESTROY_SOFT: 2099219974Smav G_RAID_DEBUG1(1, sc, 2100219974Smav "%d volumes are still open.", 2101219974Smav opens); 2102219974Smav return (EBUSY); 2103219974Smav case G_RAID_DESTROY_DELAYED: 2104219974Smav G_RAID_DEBUG1(1, sc, 2105219974Smav "Array will be destroyed on last close."); 2106219974Smav sc->sc_stopping = G_RAID_DESTROY_DELAYED; 2107219974Smav return (EBUSY); 2108219974Smav case G_RAID_DESTROY_HARD: 2109219974Smav G_RAID_DEBUG1(1, sc, 2110219974Smav "%d volumes are still open.", 2111219974Smav opens); 2112219974Smav } 2113219974Smav } 2114219974Smav 2115219974Smav /* Mark node for destruction. */ 2116219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 2117219974Smav /* Wake up worker to let it selfdestruct. */ 2118219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2119219974Smav /* Sleep until node destroyed. */ 2120219974Smav sx_sleep(&sc->sc_stopping, &sc->sc_lock, 2121219974Smav PRIBIO | PDROP, "r:destroy", 0); 2122219974Smav return (0); 2123219974Smav} 2124219974Smav 2125219974Smavstatic void 2126219974Smavg_raid_taste_orphan(struct g_consumer *cp) 2127219974Smav{ 2128219974Smav 2129219974Smav KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2130219974Smav cp->provider->name)); 2131219974Smav} 2132219974Smav 2133219974Smavstatic struct g_geom * 2134219974Smavg_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2135219974Smav{ 2136219974Smav struct g_consumer *cp; 2137219974Smav struct g_geom *gp, *geom; 2138219974Smav struct g_raid_md_class *class; 2139219974Smav struct g_raid_md_object *obj; 2140219974Smav int status; 2141219974Smav 2142219974Smav g_topology_assert(); 2143219974Smav g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2144219974Smav G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); 2145219974Smav 2146219974Smav gp = g_new_geomf(mp, "mirror:taste"); 2147219974Smav /* 2148219974Smav * This orphan function should be never called. 2149219974Smav */ 2150219974Smav gp->orphan = g_raid_taste_orphan; 2151219974Smav cp = g_new_consumer(gp); 2152219974Smav g_attach(cp, pp); 2153219974Smav 2154219974Smav geom = NULL; 2155219974Smav LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2156219974Smav G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", 2157219974Smav pp->name, class->name); 2158219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2159219974Smav M_WAITOK); 2160219974Smav obj->mdo_class = class; 2161219974Smav status = G_RAID_MD_TASTE(obj, mp, cp, &geom); 2162219974Smav if (status != G_RAID_MD_TASTE_NEW) 2163219974Smav kobj_delete((kobj_t)obj, M_RAID); 2164219974Smav if (status != G_RAID_MD_TASTE_FAIL) 2165219974Smav break; 2166219974Smav } 2167219974Smav 2168219974Smav g_detach(cp); 2169219974Smav g_destroy_consumer(cp); 2170219974Smav g_destroy_geom(gp); 2171219974Smav G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); 2172219974Smav return (geom); 2173219974Smav} 2174219974Smav 2175219974Smavint 2176219974Smavg_raid_create_node_format(const char *format, struct g_geom **gp) 2177219974Smav{ 2178219974Smav struct g_raid_md_class *class; 2179219974Smav struct g_raid_md_object *obj; 2180219974Smav int status; 2181219974Smav 2182219974Smav G_RAID_DEBUG(2, "Creating array for %s metadata.", format); 2183219974Smav LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2184219974Smav if (strcasecmp(class->name, format) == 0) 2185219974Smav break; 2186219974Smav } 2187219974Smav if (class == NULL) { 2188219974Smav G_RAID_DEBUG(1, "No support for %s metadata.", format); 2189219974Smav return (G_RAID_MD_TASTE_FAIL); 2190219974Smav } 2191219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2192219974Smav M_WAITOK); 2193219974Smav obj->mdo_class = class; 2194219974Smav status = G_RAID_MD_CREATE(obj, &g_raid_class, gp); 2195219974Smav if (status != G_RAID_MD_TASTE_NEW) 2196219974Smav kobj_delete((kobj_t)obj, M_RAID); 2197219974Smav return (status); 2198219974Smav} 2199219974Smav 2200219974Smavstatic int 2201219974Smavg_raid_destroy_geom(struct gctl_req *req __unused, 2202219974Smav struct g_class *mp __unused, struct g_geom *gp) 2203219974Smav{ 2204219974Smav struct g_raid_softc *sc; 2205219974Smav int error; 2206219974Smav 2207219974Smav g_topology_unlock(); 2208219974Smav sc = gp->softc; 2209219974Smav sx_xlock(&sc->sc_lock); 2210219974Smav g_cancel_event(sc); 2211219974Smav error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); 2212219974Smav if (error != 0) 2213219974Smav sx_xunlock(&sc->sc_lock); 2214219974Smav g_topology_lock(); 2215219974Smav return (error); 2216219974Smav} 2217219974Smav 2218219974Smavvoid g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, 2219219974Smav struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2220219974Smav{ 2221219974Smav 2222219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) 2223219974Smav return; 2224219974Smav if (sc->sc_md) 2225219974Smav G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); 2226219974Smav} 2227219974Smav 2228219974Smavvoid g_raid_fail_disk(struct g_raid_softc *sc, 2229219974Smav struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2230219974Smav{ 2231219974Smav 2232219974Smav if (disk == NULL) 2233219974Smav disk = sd->sd_disk; 2234219974Smav if (disk == NULL) { 2235219974Smav G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); 2236219974Smav return; 2237219974Smav } 2238219974Smav if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 2239219974Smav G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " 2240219974Smav "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 2241219974Smav return; 2242219974Smav } 2243219974Smav if (sc->sc_md) 2244219974Smav G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); 2245219974Smav} 2246219974Smav 2247219974Smavstatic void 2248219974Smavg_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2249219974Smav struct g_consumer *cp, struct g_provider *pp) 2250219974Smav{ 2251219974Smav struct g_raid_softc *sc; 2252219974Smav struct g_raid_volume *vol; 2253219974Smav struct g_raid_subdisk *sd; 2254219974Smav struct g_raid_disk *disk; 2255219974Smav int i, s; 2256219974Smav 2257219974Smav g_topology_assert(); 2258219974Smav 2259219974Smav sc = gp->softc; 2260219974Smav if (sc == NULL) 2261219974Smav return; 2262219974Smav if (pp != NULL) { 2263219974Smav vol = pp->private; 2264219974Smav g_topology_unlock(); 2265219974Smav sx_xlock(&sc->sc_lock); 2266219974Smav sbuf_printf(sb, "%s<Label>%s</Label>\n", indent, 2267219974Smav vol->v_name); 2268219974Smav sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent, 2269219974Smav g_raid_volume_level2str(vol->v_raid_level, 2270219974Smav vol->v_raid_level_qualifier)); 2271219974Smav sbuf_printf(sb, 2272219974Smav "%s<Transformation>%s</Transformation>\n", indent, 2273219974Smav vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); 2274219974Smav sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2275219974Smav vol->v_disks_count); 2276219974Smav sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent, 2277219974Smav vol->v_strip_size); 2278219974Smav sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2279219974Smav g_raid_volume_state2str(vol->v_state)); 2280219974Smav sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent, 2281219974Smav vol->v_dirty ? "Yes" : "No"); 2282219974Smav sbuf_printf(sb, "%s<Subdisks>", indent); 2283219974Smav for (i = 0; i < vol->v_disks_count; i++) { 2284219974Smav sd = &vol->v_subdisks[i]; 2285219974Smav if (sd->sd_disk != NULL && 2286219974Smav sd->sd_disk->d_consumer != NULL) { 2287219974Smav sbuf_printf(sb, "%s ", 2288219974Smav g_raid_get_diskname(sd->sd_disk)); 2289219974Smav } else { 2290219974Smav sbuf_printf(sb, "NONE "); 2291219974Smav } 2292219974Smav sbuf_printf(sb, "(%s", 2293219974Smav g_raid_subdisk_state2str(sd->sd_state)); 2294219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2295219974Smav sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2296219974Smav sbuf_printf(sb, " %d%%", 2297219974Smav (int)(sd->sd_rebuild_pos * 100 / 2298219974Smav sd->sd_size)); 2299219974Smav } 2300219974Smav sbuf_printf(sb, ")"); 2301219974Smav if (i + 1 < vol->v_disks_count) 2302219974Smav sbuf_printf(sb, ", "); 2303219974Smav } 2304219974Smav sbuf_printf(sb, "</Subdisks>\n"); 2305219974Smav sx_xunlock(&sc->sc_lock); 2306219974Smav g_topology_lock(); 2307219974Smav } else if (cp != NULL) { 2308219974Smav disk = cp->private; 2309219974Smav if (disk == NULL) 2310219974Smav return; 2311219974Smav g_topology_unlock(); 2312219974Smav sx_xlock(&sc->sc_lock); 2313219974Smav sbuf_printf(sb, "%s<State>%s", indent, 2314219974Smav g_raid_disk_state2str(disk->d_state)); 2315219974Smav if (!TAILQ_EMPTY(&disk->d_subdisks)) { 2316219974Smav sbuf_printf(sb, " ("); 2317219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2318219974Smav sbuf_printf(sb, "%s", 2319219974Smav g_raid_subdisk_state2str(sd->sd_state)); 2320219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2321219974Smav sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2322219974Smav sbuf_printf(sb, " %d%%", 2323219974Smav (int)(sd->sd_rebuild_pos * 100 / 2324219974Smav sd->sd_size)); 2325219974Smav } 2326219974Smav if (TAILQ_NEXT(sd, sd_next)) 2327219974Smav sbuf_printf(sb, ", "); 2328219974Smav } 2329219974Smav sbuf_printf(sb, ")"); 2330219974Smav } 2331219974Smav sbuf_printf(sb, "</State>\n"); 2332219974Smav sbuf_printf(sb, "%s<Subdisks>", indent); 2333219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2334219974Smav sbuf_printf(sb, "r%d(%s):%d@%ju", 2335219974Smav sd->sd_volume->v_global_id, 2336219974Smav sd->sd_volume->v_name, 2337219974Smav sd->sd_pos, sd->sd_offset); 2338219974Smav if (TAILQ_NEXT(sd, sd_next)) 2339219974Smav sbuf_printf(sb, ", "); 2340219974Smav } 2341219974Smav sbuf_printf(sb, "</Subdisks>\n"); 2342219974Smav sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent, 2343219974Smav disk->d_read_errs); 2344219974Smav sx_xunlock(&sc->sc_lock); 2345219974Smav g_topology_lock(); 2346219974Smav } else { 2347219974Smav g_topology_unlock(); 2348219974Smav sx_xlock(&sc->sc_lock); 2349219974Smav if (sc->sc_md) { 2350219974Smav sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent, 2351219974Smav sc->sc_md->mdo_class->name); 2352219974Smav } 2353219974Smav if (!TAILQ_EMPTY(&sc->sc_volumes)) { 2354219974Smav s = 0xff; 2355219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 2356219974Smav if (vol->v_state < s) 2357219974Smav s = vol->v_state; 2358219974Smav } 2359219974Smav sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2360219974Smav g_raid_volume_state2str(s)); 2361219974Smav } 2362219974Smav sx_xunlock(&sc->sc_lock); 2363219974Smav g_topology_lock(); 2364219974Smav } 2365219974Smav} 2366219974Smav 2367219974Smavstatic void 2368219974Smavg_raid_shutdown_pre_sync(void *arg, int howto) 2369219974Smav{ 2370219974Smav struct g_class *mp; 2371219974Smav struct g_geom *gp, *gp2; 2372219974Smav struct g_raid_softc *sc; 2373219974Smav int error; 2374219974Smav 2375219974Smav mp = arg; 2376219974Smav DROP_GIANT(); 2377219974Smav g_topology_lock(); 2378219974Smav LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2379219974Smav if ((sc = gp->softc) == NULL) 2380219974Smav continue; 2381219974Smav g_topology_unlock(); 2382219974Smav sx_xlock(&sc->sc_lock); 2383219974Smav g_cancel_event(sc); 2384219974Smav error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); 2385219974Smav if (error != 0) 2386219974Smav sx_xunlock(&sc->sc_lock); 2387219974Smav g_topology_lock(); 2388219974Smav } 2389219974Smav g_topology_unlock(); 2390219974Smav PICKUP_GIANT(); 2391219974Smav} 2392219974Smav 2393219974Smavstatic void 2394219974Smavg_raid_init(struct g_class *mp) 2395219974Smav{ 2396219974Smav 2397219974Smav g_raid_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync, 2398219974Smav g_raid_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST); 2399219974Smav if (g_raid_pre_sync == NULL) 2400219974Smav G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); 2401219974Smav g_raid_started = 1; 2402219974Smav} 2403219974Smav 2404219974Smavstatic void 2405219974Smavg_raid_fini(struct g_class *mp) 2406219974Smav{ 2407219974Smav 2408219974Smav if (g_raid_pre_sync != NULL) 2409219974Smav EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_raid_pre_sync); 2410219974Smav g_raid_started = 0; 2411219974Smav} 2412219974Smav 2413219974Smavint 2414219974Smavg_raid_md_modevent(module_t mod, int type, void *arg) 2415219974Smav{ 2416219974Smav struct g_raid_md_class *class, *c, *nc; 2417219974Smav int error; 2418219974Smav 2419219974Smav error = 0; 2420219974Smav class = arg; 2421219974Smav switch (type) { 2422219974Smav case MOD_LOAD: 2423219974Smav c = LIST_FIRST(&g_raid_md_classes); 2424219974Smav if (c == NULL || c->mdc_priority > class->mdc_priority) 2425219974Smav LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); 2426219974Smav else { 2427219974Smav while ((nc = LIST_NEXT(c, mdc_list)) != NULL && 2428219974Smav nc->mdc_priority < class->mdc_priority) 2429219974Smav c = nc; 2430219974Smav LIST_INSERT_AFTER(c, class, mdc_list); 2431219974Smav } 2432219974Smav if (g_raid_started) 2433219974Smav g_retaste(&g_raid_class); 2434219974Smav break; 2435219974Smav case MOD_UNLOAD: 2436219974Smav LIST_REMOVE(class, mdc_list); 2437219974Smav break; 2438219974Smav default: 2439219974Smav error = EOPNOTSUPP; 2440219974Smav break; 2441219974Smav } 2442219974Smav 2443219974Smav return (error); 2444219974Smav} 2445219974Smav 2446219974Smavint 2447219974Smavg_raid_tr_modevent(module_t mod, int type, void *arg) 2448219974Smav{ 2449219974Smav struct g_raid_tr_class *class, *c, *nc; 2450219974Smav int error; 2451219974Smav 2452219974Smav error = 0; 2453219974Smav class = arg; 2454219974Smav switch (type) { 2455219974Smav case MOD_LOAD: 2456219974Smav c = LIST_FIRST(&g_raid_tr_classes); 2457219974Smav if (c == NULL || c->trc_priority > class->trc_priority) 2458219974Smav LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); 2459219974Smav else { 2460219974Smav while ((nc = LIST_NEXT(c, trc_list)) != NULL && 2461219974Smav nc->trc_priority < class->trc_priority) 2462219974Smav c = nc; 2463219974Smav LIST_INSERT_AFTER(c, class, trc_list); 2464219974Smav } 2465219974Smav break; 2466219974Smav case MOD_UNLOAD: 2467219974Smav LIST_REMOVE(class, trc_list); 2468219974Smav break; 2469219974Smav default: 2470219974Smav error = EOPNOTSUPP; 2471219974Smav break; 2472219974Smav } 2473219974Smav 2474219974Smav return (error); 2475219974Smav} 2476219974Smav 2477219974Smav/* 2478219974Smav * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) 2479219974Smav * to reduce module priority, allowing submodules to register them first. 2480219974Smav */ 2481219974Smavstatic moduledata_t g_raid_mod = { 2482219974Smav "g_raid", 2483219974Smav g_modevent, 2484219974Smav &g_raid_class 2485219974Smav}; 2486219974SmavDECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD); 2487219974SmavMODULE_VERSION(geom_raid, 0); 2488