g_raid.c revision 239175
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 239175 2012-08-10 13:36:33Z 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"); 221239175Smav case G_RAID_SUBDISK_E_FAILED: 222239175Smav return ("FAILED"); 223219974Smav case G_RAID_SUBDISK_E_DISCONNECTED: 224219974Smav return ("DISCONNECTED"); 225219974Smav default: 226219974Smav return ("INVALID"); 227219974Smav } 228219974Smav} 229219974Smav 230219974Smavconst char * 231219974Smavg_raid_volume_state2str(int state) 232219974Smav{ 233219974Smav 234219974Smav switch (state) { 235219974Smav case G_RAID_VOLUME_S_STARTING: 236219974Smav return ("STARTING"); 237219974Smav case G_RAID_VOLUME_S_BROKEN: 238219974Smav return ("BROKEN"); 239219974Smav case G_RAID_VOLUME_S_DEGRADED: 240219974Smav return ("DEGRADED"); 241219974Smav case G_RAID_VOLUME_S_SUBOPTIMAL: 242219974Smav return ("SUBOPTIMAL"); 243219974Smav case G_RAID_VOLUME_S_OPTIMAL: 244219974Smav return ("OPTIMAL"); 245219974Smav case G_RAID_VOLUME_S_UNSUPPORTED: 246219974Smav return ("UNSUPPORTED"); 247219974Smav case G_RAID_VOLUME_S_STOPPED: 248219974Smav return ("STOPPED"); 249219974Smav default: 250219974Smav return ("INVALID"); 251219974Smav } 252219974Smav} 253219974Smav 254219974Smavstatic const char * 255219974Smavg_raid_volume_event2str(int event) 256219974Smav{ 257219974Smav 258219974Smav switch (event) { 259219974Smav case G_RAID_VOLUME_E_UP: 260219974Smav return ("UP"); 261219974Smav case G_RAID_VOLUME_E_DOWN: 262219974Smav return ("DOWN"); 263219974Smav case G_RAID_VOLUME_E_START: 264219974Smav return ("START"); 265219974Smav case G_RAID_VOLUME_E_STARTMD: 266219974Smav return ("STARTMD"); 267219974Smav default: 268219974Smav return ("INVALID"); 269219974Smav } 270219974Smav} 271219974Smav 272219974Smavconst char * 273219974Smavg_raid_volume_level2str(int level, int qual) 274219974Smav{ 275219974Smav 276219974Smav switch (level) { 277219974Smav case G_RAID_VOLUME_RL_RAID0: 278219974Smav return ("RAID0"); 279219974Smav case G_RAID_VOLUME_RL_RAID1: 280219974Smav return ("RAID1"); 281219974Smav case G_RAID_VOLUME_RL_RAID3: 282234603Smav if (qual == G_RAID_VOLUME_RLQ_R3P0) 283234603Smav return ("RAID3-P0"); 284234603Smav if (qual == G_RAID_VOLUME_RLQ_R3PN) 285234603Smav return ("RAID3-PN"); 286219974Smav return ("RAID3"); 287219974Smav case G_RAID_VOLUME_RL_RAID4: 288234603Smav if (qual == G_RAID_VOLUME_RLQ_R4P0) 289234610Smav return ("RAID4-P0"); 290234603Smav if (qual == G_RAID_VOLUME_RLQ_R4PN) 291234610Smav return ("RAID4-PN"); 292219974Smav return ("RAID4"); 293219974Smav case G_RAID_VOLUME_RL_RAID5: 294234458Smav if (qual == G_RAID_VOLUME_RLQ_R5RA) 295234603Smav return ("RAID5-RA"); 296234458Smav if (qual == G_RAID_VOLUME_RLQ_R5RS) 297234603Smav return ("RAID5-RS"); 298234458Smav if (qual == G_RAID_VOLUME_RLQ_R5LA) 299234603Smav return ("RAID5-LA"); 300234458Smav if (qual == G_RAID_VOLUME_RLQ_R5LS) 301234603Smav return ("RAID5-LS"); 302219974Smav return ("RAID5"); 303219974Smav case G_RAID_VOLUME_RL_RAID6: 304234603Smav if (qual == G_RAID_VOLUME_RLQ_R6RA) 305234603Smav return ("RAID6-RA"); 306234603Smav if (qual == G_RAID_VOLUME_RLQ_R6RS) 307234603Smav return ("RAID6-RS"); 308234603Smav if (qual == G_RAID_VOLUME_RLQ_R6LA) 309234603Smav return ("RAID6-LA"); 310234603Smav if (qual == G_RAID_VOLUME_RLQ_R6LS) 311234603Smav return ("RAID6-LS"); 312219974Smav return ("RAID6"); 313234603Smav case G_RAID_VOLUME_RL_RAIDMDF: 314234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFRA) 315234603Smav return ("RAIDMDF-RA"); 316234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFRS) 317234603Smav return ("RAIDMDF-RS"); 318234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFLA) 319234603Smav return ("RAIDMDF-LA"); 320234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFLS) 321234603Smav return ("RAIDMDF-LS"); 322234603Smav return ("RAIDMDF"); 323219974Smav case G_RAID_VOLUME_RL_RAID1E: 324234603Smav if (qual == G_RAID_VOLUME_RLQ_R1EA) 325234603Smav return ("RAID1E-A"); 326234603Smav if (qual == G_RAID_VOLUME_RLQ_R1EO) 327234603Smav return ("RAID1E-O"); 328219974Smav return ("RAID1E"); 329219974Smav case G_RAID_VOLUME_RL_SINGLE: 330219974Smav return ("SINGLE"); 331219974Smav case G_RAID_VOLUME_RL_CONCAT: 332219974Smav return ("CONCAT"); 333219974Smav case G_RAID_VOLUME_RL_RAID5E: 334234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ERA) 335234603Smav return ("RAID5E-RA"); 336234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ERS) 337234603Smav return ("RAID5E-RS"); 338234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ELA) 339234603Smav return ("RAID5E-LA"); 340234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ELS) 341234603Smav return ("RAID5E-LS"); 342219974Smav return ("RAID5E"); 343219974Smav case G_RAID_VOLUME_RL_RAID5EE: 344234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EERA) 345234603Smav return ("RAID5EE-RA"); 346234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EERS) 347234603Smav return ("RAID5EE-RS"); 348234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EELA) 349234603Smav return ("RAID5EE-LA"); 350234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EELS) 351234603Smav return ("RAID5EE-LS"); 352219974Smav return ("RAID5EE"); 353234603Smav case G_RAID_VOLUME_RL_RAID5R: 354234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RRA) 355234603Smav return ("RAID5R-RA"); 356234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RRS) 357234603Smav return ("RAID5R-RS"); 358234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RLA) 359234603Smav return ("RAID5R-LA"); 360234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RLS) 361234603Smav return ("RAID5R-LS"); 362234603Smav return ("RAID5E"); 363219974Smav default: 364219974Smav return ("UNKNOWN"); 365219974Smav } 366219974Smav} 367219974Smav 368219974Smavint 369219974Smavg_raid_volume_str2level(const char *str, int *level, int *qual) 370219974Smav{ 371219974Smav 372219974Smav *level = G_RAID_VOLUME_RL_UNKNOWN; 373219974Smav *qual = G_RAID_VOLUME_RLQ_NONE; 374219974Smav if (strcasecmp(str, "RAID0") == 0) 375219974Smav *level = G_RAID_VOLUME_RL_RAID0; 376219974Smav else if (strcasecmp(str, "RAID1") == 0) 377219974Smav *level = G_RAID_VOLUME_RL_RAID1; 378234603Smav else if (strcasecmp(str, "RAID3-P0") == 0) { 379219974Smav *level = G_RAID_VOLUME_RL_RAID3; 380234603Smav *qual = G_RAID_VOLUME_RLQ_R3P0; 381234993Smav } else if (strcasecmp(str, "RAID3-PN") == 0 || 382234603Smav strcasecmp(str, "RAID3") == 0) { 383234603Smav *level = G_RAID_VOLUME_RL_RAID3; 384234993Smav *qual = G_RAID_VOLUME_RLQ_R3PN; 385234603Smav } else if (strcasecmp(str, "RAID4-P0") == 0) { 386219974Smav *level = G_RAID_VOLUME_RL_RAID4; 387234603Smav *qual = G_RAID_VOLUME_RLQ_R4P0; 388234993Smav } else if (strcasecmp(str, "RAID4-PN") == 0 || 389234603Smav strcasecmp(str, "RAID4") == 0) { 390234603Smav *level = G_RAID_VOLUME_RL_RAID4; 391234993Smav *qual = G_RAID_VOLUME_RLQ_R4PN; 392234603Smav } else if (strcasecmp(str, "RAID5-RA") == 0) { 393219974Smav *level = G_RAID_VOLUME_RL_RAID5; 394234458Smav *qual = G_RAID_VOLUME_RLQ_R5RA; 395234603Smav } else if (strcasecmp(str, "RAID5-RS") == 0) { 396234458Smav *level = G_RAID_VOLUME_RL_RAID5; 397234458Smav *qual = G_RAID_VOLUME_RLQ_R5RS; 398234458Smav } else if (strcasecmp(str, "RAID5") == 0 || 399234603Smav strcasecmp(str, "RAID5-LA") == 0) { 400234458Smav *level = G_RAID_VOLUME_RL_RAID5; 401234458Smav *qual = G_RAID_VOLUME_RLQ_R5LA; 402234603Smav } else if (strcasecmp(str, "RAID5-LS") == 0) { 403234458Smav *level = G_RAID_VOLUME_RL_RAID5; 404234458Smav *qual = G_RAID_VOLUME_RLQ_R5LS; 405234603Smav } else if (strcasecmp(str, "RAID6-RA") == 0) { 406219974Smav *level = G_RAID_VOLUME_RL_RAID6; 407234603Smav *qual = G_RAID_VOLUME_RLQ_R6RA; 408234603Smav } else if (strcasecmp(str, "RAID6-RS") == 0) { 409234603Smav *level = G_RAID_VOLUME_RL_RAID6; 410234603Smav *qual = G_RAID_VOLUME_RLQ_R6RS; 411234603Smav } else if (strcasecmp(str, "RAID6") == 0 || 412234603Smav strcasecmp(str, "RAID6-LA") == 0) { 413234603Smav *level = G_RAID_VOLUME_RL_RAID6; 414234603Smav *qual = G_RAID_VOLUME_RLQ_R6LA; 415234603Smav } else if (strcasecmp(str, "RAID6-LS") == 0) { 416234603Smav *level = G_RAID_VOLUME_RL_RAID6; 417234603Smav *qual = G_RAID_VOLUME_RLQ_R6LS; 418234603Smav } else if (strcasecmp(str, "RAIDMDF-RA") == 0) { 419234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 420234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFRA; 421234603Smav } else if (strcasecmp(str, "RAIDMDF-RS") == 0) { 422234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 423234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFRS; 424234603Smav } else if (strcasecmp(str, "RAIDMDF") == 0 || 425234603Smav strcasecmp(str, "RAIDMDF-LA") == 0) { 426234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 427234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFLA; 428234603Smav } else if (strcasecmp(str, "RAIDMDF-LS") == 0) { 429234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 430234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFLS; 431234603Smav } else if (strcasecmp(str, "RAID10") == 0 || 432234603Smav strcasecmp(str, "RAID1E") == 0 || 433234603Smav strcasecmp(str, "RAID1E-A") == 0) { 434219974Smav *level = G_RAID_VOLUME_RL_RAID1E; 435234603Smav *qual = G_RAID_VOLUME_RLQ_R1EA; 436234603Smav } else if (strcasecmp(str, "RAID1E-O") == 0) { 437234603Smav *level = G_RAID_VOLUME_RL_RAID1E; 438234603Smav *qual = G_RAID_VOLUME_RLQ_R1EO; 439234603Smav } else if (strcasecmp(str, "SINGLE") == 0) 440219974Smav *level = G_RAID_VOLUME_RL_SINGLE; 441219974Smav else if (strcasecmp(str, "CONCAT") == 0) 442219974Smav *level = G_RAID_VOLUME_RL_CONCAT; 443234603Smav else if (strcasecmp(str, "RAID5E-RA") == 0) { 444219974Smav *level = G_RAID_VOLUME_RL_RAID5E; 445234603Smav *qual = G_RAID_VOLUME_RLQ_R5ERA; 446234603Smav } else if (strcasecmp(str, "RAID5E-RS") == 0) { 447234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 448234603Smav *qual = G_RAID_VOLUME_RLQ_R5ERS; 449234603Smav } else if (strcasecmp(str, "RAID5E") == 0 || 450234603Smav strcasecmp(str, "RAID5E-LA") == 0) { 451234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 452234603Smav *qual = G_RAID_VOLUME_RLQ_R5ELA; 453234603Smav } else if (strcasecmp(str, "RAID5E-LS") == 0) { 454234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 455234603Smav *qual = G_RAID_VOLUME_RLQ_R5ELS; 456234603Smav } else if (strcasecmp(str, "RAID5EE-RA") == 0) { 457219974Smav *level = G_RAID_VOLUME_RL_RAID5EE; 458234603Smav *qual = G_RAID_VOLUME_RLQ_R5EERA; 459234603Smav } else if (strcasecmp(str, "RAID5EE-RS") == 0) { 460234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 461234603Smav *qual = G_RAID_VOLUME_RLQ_R5EERS; 462234603Smav } else if (strcasecmp(str, "RAID5EE") == 0 || 463234603Smav strcasecmp(str, "RAID5EE-LA") == 0) { 464234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 465234603Smav *qual = G_RAID_VOLUME_RLQ_R5EELA; 466234603Smav } else if (strcasecmp(str, "RAID5EE-LS") == 0) { 467234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 468234603Smav *qual = G_RAID_VOLUME_RLQ_R5EELS; 469234603Smav } else if (strcasecmp(str, "RAID5R-RA") == 0) { 470234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 471234603Smav *qual = G_RAID_VOLUME_RLQ_R5RRA; 472234603Smav } else if (strcasecmp(str, "RAID5R-RS") == 0) { 473234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 474234603Smav *qual = G_RAID_VOLUME_RLQ_R5RRS; 475234603Smav } else if (strcasecmp(str, "RAID5R") == 0 || 476234603Smav strcasecmp(str, "RAID5R-LA") == 0) { 477234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 478234603Smav *qual = G_RAID_VOLUME_RLQ_R5RLA; 479234603Smav } else if (strcasecmp(str, "RAID5R-LS") == 0) { 480234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 481234603Smav *qual = G_RAID_VOLUME_RLQ_R5RLS; 482234603Smav } else 483219974Smav return (-1); 484219974Smav return (0); 485219974Smav} 486219974Smav 487219974Smavconst char * 488219974Smavg_raid_get_diskname(struct g_raid_disk *disk) 489219974Smav{ 490219974Smav 491219974Smav if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 492219974Smav return ("[unknown]"); 493219974Smav return (disk->d_consumer->provider->name); 494219974Smav} 495219974Smav 496219974Smavvoid 497219974Smavg_raid_report_disk_state(struct g_raid_disk *disk) 498219974Smav{ 499219974Smav struct g_raid_subdisk *sd; 500219974Smav int len, state; 501219974Smav uint32_t s; 502219974Smav 503219974Smav if (disk->d_consumer == NULL) 504219974Smav return; 505219974Smav if (disk->d_state == G_RAID_DISK_S_FAILED || 506219974Smav disk->d_state == G_RAID_DISK_S_STALE_FAILED) { 507219974Smav s = G_STATE_FAILED; 508219974Smav } else { 509219974Smav state = G_RAID_SUBDISK_S_ACTIVE; 510219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 511219974Smav if (sd->sd_state < state) 512219974Smav state = sd->sd_state; 513219974Smav } 514219974Smav if (state == G_RAID_SUBDISK_S_FAILED) 515219974Smav s = G_STATE_FAILED; 516219974Smav else if (state == G_RAID_SUBDISK_S_NEW || 517219974Smav state == G_RAID_SUBDISK_S_REBUILD) 518219974Smav s = G_STATE_REBUILD; 519219974Smav else if (state == G_RAID_SUBDISK_S_STALE || 520219974Smav state == G_RAID_SUBDISK_S_RESYNC) 521219974Smav s = G_STATE_RESYNC; 522219974Smav else 523219974Smav s = G_STATE_ACTIVE; 524219974Smav } 525219974Smav len = sizeof(s); 526219974Smav g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s); 527219974Smav G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.", 528219974Smav g_raid_get_diskname(disk), s); 529219974Smav} 530219974Smav 531219974Smavvoid 532219974Smavg_raid_change_disk_state(struct g_raid_disk *disk, int state) 533219974Smav{ 534219974Smav 535219974Smav G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.", 536219974Smav g_raid_get_diskname(disk), 537219974Smav g_raid_disk_state2str(disk->d_state), 538219974Smav g_raid_disk_state2str(state)); 539219974Smav disk->d_state = state; 540219974Smav g_raid_report_disk_state(disk); 541219974Smav} 542219974Smav 543219974Smavvoid 544219974Smavg_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state) 545219974Smav{ 546219974Smav 547219974Smav G_RAID_DEBUG1(0, sd->sd_softc, 548219974Smav "Subdisk %s:%d-%s state changed from %s to %s.", 549219974Smav sd->sd_volume->v_name, sd->sd_pos, 550219974Smav sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", 551219974Smav g_raid_subdisk_state2str(sd->sd_state), 552219974Smav g_raid_subdisk_state2str(state)); 553219974Smav sd->sd_state = state; 554219974Smav if (sd->sd_disk) 555219974Smav g_raid_report_disk_state(sd->sd_disk); 556219974Smav} 557219974Smav 558219974Smavvoid 559219974Smavg_raid_change_volume_state(struct g_raid_volume *vol, int state) 560219974Smav{ 561219974Smav 562219974Smav G_RAID_DEBUG1(0, vol->v_softc, 563219974Smav "Volume %s state changed from %s to %s.", 564219974Smav vol->v_name, 565219974Smav g_raid_volume_state2str(vol->v_state), 566219974Smav g_raid_volume_state2str(state)); 567219974Smav vol->v_state = state; 568219974Smav} 569219974Smav 570219974Smav/* 571219974Smav * --- Events handling functions --- 572219974Smav * Events in geom_raid are used to maintain subdisks and volumes status 573219974Smav * from one thread to simplify locking. 574219974Smav */ 575219974Smavstatic void 576219974Smavg_raid_event_free(struct g_raid_event *ep) 577219974Smav{ 578219974Smav 579219974Smav free(ep, M_RAID); 580219974Smav} 581219974Smav 582219974Smavint 583219974Smavg_raid_event_send(void *arg, int event, int flags) 584219974Smav{ 585219974Smav struct g_raid_softc *sc; 586219974Smav struct g_raid_event *ep; 587219974Smav int error; 588219974Smav 589219974Smav if ((flags & G_RAID_EVENT_VOLUME) != 0) { 590219974Smav sc = ((struct g_raid_volume *)arg)->v_softc; 591219974Smav } else if ((flags & G_RAID_EVENT_DISK) != 0) { 592219974Smav sc = ((struct g_raid_disk *)arg)->d_softc; 593219974Smav } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) { 594219974Smav sc = ((struct g_raid_subdisk *)arg)->sd_softc; 595219974Smav } else { 596219974Smav sc = arg; 597219974Smav } 598219974Smav ep = malloc(sizeof(*ep), M_RAID, 599219974Smav sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT); 600219974Smav if (ep == NULL) 601219974Smav return (ENOMEM); 602219974Smav ep->e_tgt = arg; 603219974Smav ep->e_event = event; 604219974Smav ep->e_flags = flags; 605219974Smav ep->e_error = 0; 606219974Smav G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc); 607219974Smav mtx_lock(&sc->sc_queue_mtx); 608219974Smav TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 609219974Smav mtx_unlock(&sc->sc_queue_mtx); 610219974Smav wakeup(sc); 611219974Smav 612219974Smav if ((flags & G_RAID_EVENT_WAIT) == 0) 613219974Smav return (0); 614219974Smav 615219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 616219974Smav G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep); 617219974Smav sx_xunlock(&sc->sc_lock); 618219974Smav while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) { 619219974Smav mtx_lock(&sc->sc_queue_mtx); 620219974Smav MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event", 621219974Smav hz * 5); 622219974Smav } 623219974Smav error = ep->e_error; 624219974Smav g_raid_event_free(ep); 625219974Smav sx_xlock(&sc->sc_lock); 626219974Smav return (error); 627219974Smav} 628219974Smav 629219974Smavstatic void 630219974Smavg_raid_event_cancel(struct g_raid_softc *sc, void *tgt) 631219974Smav{ 632219974Smav struct g_raid_event *ep, *tmpep; 633219974Smav 634219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 635219974Smav 636219974Smav mtx_lock(&sc->sc_queue_mtx); 637219974Smav TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 638219974Smav if (ep->e_tgt != tgt) 639219974Smav continue; 640219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 641219974Smav if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) 642219974Smav g_raid_event_free(ep); 643219974Smav else { 644219974Smav ep->e_error = ECANCELED; 645219974Smav wakeup(ep); 646219974Smav } 647219974Smav } 648219974Smav mtx_unlock(&sc->sc_queue_mtx); 649219974Smav} 650219974Smav 651219974Smavstatic int 652219974Smavg_raid_event_check(struct g_raid_softc *sc, void *tgt) 653219974Smav{ 654219974Smav struct g_raid_event *ep; 655219974Smav int res = 0; 656219974Smav 657219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 658219974Smav 659219974Smav mtx_lock(&sc->sc_queue_mtx); 660219974Smav TAILQ_FOREACH(ep, &sc->sc_events, e_next) { 661219974Smav if (ep->e_tgt != tgt) 662219974Smav continue; 663219974Smav res = 1; 664219974Smav break; 665219974Smav } 666219974Smav mtx_unlock(&sc->sc_queue_mtx); 667219974Smav return (res); 668219974Smav} 669219974Smav 670219974Smav/* 671219974Smav * Return the number of disks in given state. 672219974Smav * If state is equal to -1, count all connected disks. 673219974Smav */ 674219974Smavu_int 675219974Smavg_raid_ndisks(struct g_raid_softc *sc, int state) 676219974Smav{ 677219974Smav struct g_raid_disk *disk; 678219974Smav u_int n; 679219974Smav 680219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 681219974Smav 682219974Smav n = 0; 683219974Smav TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 684219974Smav if (disk->d_state == state || state == -1) 685219974Smav n++; 686219974Smav } 687219974Smav return (n); 688219974Smav} 689219974Smav 690219974Smav/* 691219974Smav * Return the number of subdisks in given state. 692219974Smav * If state is equal to -1, count all connected disks. 693219974Smav */ 694219974Smavu_int 695219974Smavg_raid_nsubdisks(struct g_raid_volume *vol, int state) 696219974Smav{ 697219974Smav struct g_raid_subdisk *subdisk; 698219974Smav struct g_raid_softc *sc; 699219974Smav u_int i, n ; 700219974Smav 701219974Smav sc = vol->v_softc; 702219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 703219974Smav 704219974Smav n = 0; 705219974Smav for (i = 0; i < vol->v_disks_count; i++) { 706219974Smav subdisk = &vol->v_subdisks[i]; 707219974Smav if ((state == -1 && 708219974Smav subdisk->sd_state != G_RAID_SUBDISK_S_NONE) || 709219974Smav subdisk->sd_state == state) 710219974Smav n++; 711219974Smav } 712219974Smav return (n); 713219974Smav} 714219974Smav 715219974Smav/* 716219974Smav * Return the first subdisk in given state. 717219974Smav * If state is equal to -1, then the first connected disks. 718219974Smav */ 719219974Smavstruct g_raid_subdisk * 720219974Smavg_raid_get_subdisk(struct g_raid_volume *vol, int state) 721219974Smav{ 722219974Smav struct g_raid_subdisk *sd; 723219974Smav struct g_raid_softc *sc; 724219974Smav u_int i; 725219974Smav 726219974Smav sc = vol->v_softc; 727219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 728219974Smav 729219974Smav for (i = 0; i < vol->v_disks_count; i++) { 730219974Smav sd = &vol->v_subdisks[i]; 731219974Smav if ((state == -1 && 732219974Smav sd->sd_state != G_RAID_SUBDISK_S_NONE) || 733219974Smav sd->sd_state == state) 734219974Smav return (sd); 735219974Smav } 736219974Smav return (NULL); 737219974Smav} 738219974Smav 739219974Smavstruct g_consumer * 740219974Smavg_raid_open_consumer(struct g_raid_softc *sc, const char *name) 741219974Smav{ 742219974Smav struct g_consumer *cp; 743219974Smav struct g_provider *pp; 744219974Smav 745219974Smav g_topology_assert(); 746219974Smav 747219974Smav if (strncmp(name, "/dev/", 5) == 0) 748219974Smav name += 5; 749219974Smav pp = g_provider_by_name(name); 750219974Smav if (pp == NULL) 751219974Smav return (NULL); 752219974Smav cp = g_new_consumer(sc->sc_geom); 753219974Smav if (g_attach(cp, pp) != 0) { 754219974Smav g_destroy_consumer(cp); 755219974Smav return (NULL); 756219974Smav } 757219974Smav if (g_access(cp, 1, 1, 1) != 0) { 758219974Smav g_detach(cp); 759219974Smav g_destroy_consumer(cp); 760219974Smav return (NULL); 761219974Smav } 762219974Smav return (cp); 763219974Smav} 764219974Smav 765219974Smavstatic u_int 766219974Smavg_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) 767219974Smav{ 768219974Smav struct bio *bp; 769219974Smav u_int nreqs = 0; 770219974Smav 771219974Smav mtx_lock(&sc->sc_queue_mtx); 772219974Smav TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 773219974Smav if (bp->bio_from == cp) 774219974Smav nreqs++; 775219974Smav } 776219974Smav mtx_unlock(&sc->sc_queue_mtx); 777219974Smav return (nreqs); 778219974Smav} 779219974Smav 780219974Smavu_int 781219974Smavg_raid_nopens(struct g_raid_softc *sc) 782219974Smav{ 783219974Smav struct g_raid_volume *vol; 784219974Smav u_int opens; 785219974Smav 786219974Smav opens = 0; 787219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 788219974Smav if (vol->v_provider_open != 0) 789219974Smav opens++; 790219974Smav } 791219974Smav return (opens); 792219974Smav} 793219974Smav 794219974Smavstatic int 795219974Smavg_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) 796219974Smav{ 797219974Smav 798219974Smav if (cp->index > 0) { 799219974Smav G_RAID_DEBUG1(2, sc, 800219974Smav "I/O requests for %s exist, can't destroy it now.", 801219974Smav cp->provider->name); 802219974Smav return (1); 803219974Smav } 804219974Smav if (g_raid_nrequests(sc, cp) > 0) { 805219974Smav G_RAID_DEBUG1(2, sc, 806219974Smav "I/O requests for %s in queue, can't destroy it now.", 807219974Smav cp->provider->name); 808219974Smav return (1); 809219974Smav } 810219974Smav return (0); 811219974Smav} 812219974Smav 813219974Smavstatic void 814219974Smavg_raid_destroy_consumer(void *arg, int flags __unused) 815219974Smav{ 816219974Smav struct g_consumer *cp; 817219974Smav 818219974Smav g_topology_assert(); 819219974Smav 820219974Smav cp = arg; 821219974Smav G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 822219974Smav g_detach(cp); 823219974Smav g_destroy_consumer(cp); 824219974Smav} 825219974Smav 826219974Smavvoid 827219974Smavg_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) 828219974Smav{ 829219974Smav struct g_provider *pp; 830219974Smav int retaste_wait; 831219974Smav 832219974Smav g_topology_assert_not(); 833219974Smav 834219974Smav g_topology_lock(); 835219974Smav cp->private = NULL; 836219974Smav if (g_raid_consumer_is_busy(sc, cp)) 837219974Smav goto out; 838219974Smav pp = cp->provider; 839219974Smav retaste_wait = 0; 840219974Smav if (cp->acw == 1) { 841219974Smav if ((pp->geom->flags & G_GEOM_WITHER) == 0) 842219974Smav retaste_wait = 1; 843219974Smav } 844219974Smav if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 845219974Smav g_access(cp, -cp->acr, -cp->acw, -cp->ace); 846219974Smav if (retaste_wait) { 847219974Smav /* 848219974Smav * After retaste event was send (inside g_access()), we can send 849219974Smav * event to detach and destroy consumer. 850219974Smav * A class, which has consumer to the given provider connected 851219974Smav * will not receive retaste event for the provider. 852219974Smav * This is the way how I ignore retaste events when I close 853219974Smav * consumers opened for write: I detach and destroy consumer 854219974Smav * after retaste event is sent. 855219974Smav */ 856219974Smav g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); 857219974Smav goto out; 858219974Smav } 859219974Smav G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); 860219974Smav g_detach(cp); 861219974Smav g_destroy_consumer(cp); 862219974Smavout: 863219974Smav g_topology_unlock(); 864219974Smav} 865219974Smav 866219974Smavstatic void 867219974Smavg_raid_orphan(struct g_consumer *cp) 868219974Smav{ 869219974Smav struct g_raid_disk *disk; 870219974Smav 871219974Smav g_topology_assert(); 872219974Smav 873219974Smav disk = cp->private; 874219974Smav if (disk == NULL) 875219974Smav return; 876219974Smav g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, 877219974Smav G_RAID_EVENT_DISK); 878219974Smav} 879219974Smav 880219974Smavstatic int 881219974Smavg_raid_clean(struct g_raid_volume *vol, int acw) 882219974Smav{ 883219974Smav struct g_raid_softc *sc; 884219974Smav int timeout; 885219974Smav 886219974Smav sc = vol->v_softc; 887219974Smav g_topology_assert_not(); 888219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 889219974Smav 890219974Smav// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 891219974Smav// return (0); 892219974Smav if (!vol->v_dirty) 893219974Smav return (0); 894219974Smav if (vol->v_writes > 0) 895219974Smav return (0); 896219974Smav if (acw > 0 || (acw == -1 && 897219974Smav vol->v_provider != NULL && vol->v_provider->acw > 0)) { 898219974Smav timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); 899219974Smav if (timeout > 0) 900219974Smav return (timeout); 901219974Smav } 902219974Smav vol->v_dirty = 0; 903219974Smav G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", 904219974Smav vol->v_name); 905219974Smav g_raid_write_metadata(sc, vol, NULL, NULL); 906219974Smav return (0); 907219974Smav} 908219974Smav 909219974Smavstatic void 910219974Smavg_raid_dirty(struct g_raid_volume *vol) 911219974Smav{ 912219974Smav struct g_raid_softc *sc; 913219974Smav 914219974Smav sc = vol->v_softc; 915219974Smav g_topology_assert_not(); 916219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 917219974Smav 918219974Smav// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 919219974Smav// return; 920219974Smav vol->v_dirty = 1; 921219974Smav G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", 922219974Smav vol->v_name); 923219974Smav g_raid_write_metadata(sc, vol, NULL, NULL); 924219974Smav} 925219974Smav 926219974Smavvoid 927219974Smavg_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) 928219974Smav{ 929219974Smav struct g_raid_softc *sc; 930219974Smav struct g_raid_volume *vol; 931219974Smav struct g_raid_subdisk *sd; 932219974Smav struct bio_queue_head queue; 933219974Smav struct bio *cbp; 934219974Smav int i; 935219974Smav 936219974Smav vol = tr->tro_volume; 937219974Smav sc = vol->v_softc; 938219974Smav 939219974Smav /* 940219974Smav * Allocate all bios before sending any request, so we can return 941219974Smav * ENOMEM in nice and clean way. 942219974Smav */ 943219974Smav bioq_init(&queue); 944219974Smav for (i = 0; i < vol->v_disks_count; i++) { 945219974Smav sd = &vol->v_subdisks[i]; 946219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_NONE || 947219974Smav sd->sd_state == G_RAID_SUBDISK_S_FAILED) 948219974Smav continue; 949219974Smav cbp = g_clone_bio(bp); 950219974Smav if (cbp == NULL) 951219974Smav goto failure; 952219974Smav cbp->bio_caller1 = sd; 953219974Smav bioq_insert_tail(&queue, cbp); 954219974Smav } 955219974Smav for (cbp = bioq_first(&queue); cbp != NULL; 956219974Smav cbp = bioq_first(&queue)) { 957219974Smav bioq_remove(&queue, cbp); 958219974Smav sd = cbp->bio_caller1; 959219974Smav cbp->bio_caller1 = NULL; 960219974Smav g_raid_subdisk_iostart(sd, cbp); 961219974Smav } 962219974Smav return; 963219974Smavfailure: 964219974Smav for (cbp = bioq_first(&queue); cbp != NULL; 965219974Smav cbp = bioq_first(&queue)) { 966219974Smav bioq_remove(&queue, cbp); 967219974Smav g_destroy_bio(cbp); 968219974Smav } 969219974Smav if (bp->bio_error == 0) 970219974Smav bp->bio_error = ENOMEM; 971219974Smav g_raid_iodone(bp, bp->bio_error); 972219974Smav} 973219974Smav 974219974Smavstatic void 975219974Smavg_raid_tr_kerneldump_common_done(struct bio *bp) 976219974Smav{ 977219974Smav 978219974Smav bp->bio_flags |= BIO_DONE; 979219974Smav} 980219974Smav 981219974Smavint 982219974Smavg_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, 983219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 984219974Smav{ 985219974Smav struct g_raid_softc *sc; 986219974Smav struct g_raid_volume *vol; 987219974Smav struct bio bp; 988219974Smav 989219974Smav vol = tr->tro_volume; 990219974Smav sc = vol->v_softc; 991219974Smav 992219974Smav bzero(&bp, sizeof(bp)); 993219974Smav bp.bio_cmd = BIO_WRITE; 994219974Smav bp.bio_done = g_raid_tr_kerneldump_common_done; 995219974Smav bp.bio_attribute = NULL; 996219974Smav bp.bio_offset = offset; 997219974Smav bp.bio_length = length; 998219974Smav bp.bio_data = virtual; 999219974Smav bp.bio_to = vol->v_provider; 1000219974Smav 1001219974Smav g_raid_start(&bp); 1002219974Smav while (!(bp.bio_flags & BIO_DONE)) { 1003219974Smav G_RAID_DEBUG1(4, sc, "Poll..."); 1004219974Smav g_raid_poll(sc); 1005219974Smav DELAY(10); 1006219974Smav } 1007219974Smav 1008219974Smav return (bp.bio_error != 0 ? EIO : 0); 1009219974Smav} 1010219974Smav 1011219974Smavstatic int 1012219974Smavg_raid_dump(void *arg, 1013219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1014219974Smav{ 1015219974Smav struct g_raid_volume *vol; 1016219974Smav int error; 1017219974Smav 1018219974Smav vol = (struct g_raid_volume *)arg; 1019219974Smav G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", 1020219974Smav (long long unsigned)offset, (long long unsigned)length); 1021219974Smav 1022219974Smav error = G_RAID_TR_KERNELDUMP(vol->v_tr, 1023219974Smav virtual, physical, offset, length); 1024219974Smav return (error); 1025219974Smav} 1026219974Smav 1027219974Smavstatic void 1028219974Smavg_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) 1029219974Smav{ 1030219974Smav struct g_kerneldump *gkd; 1031219974Smav struct g_provider *pp; 1032219974Smav struct g_raid_volume *vol; 1033219974Smav 1034219974Smav gkd = (struct g_kerneldump*)bp->bio_data; 1035219974Smav pp = bp->bio_to; 1036219974Smav vol = pp->private; 1037219974Smav g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", 1038219974Smav pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 1039219974Smav gkd->di.dumper = g_raid_dump; 1040219974Smav gkd->di.priv = vol; 1041219974Smav gkd->di.blocksize = vol->v_sectorsize; 1042219974Smav gkd->di.maxiosize = DFLTPHYS; 1043219974Smav gkd->di.mediaoffset = gkd->offset; 1044219974Smav if ((gkd->offset + gkd->length) > vol->v_mediasize) 1045219974Smav gkd->length = vol->v_mediasize - gkd->offset; 1046219974Smav gkd->di.mediasize = gkd->length; 1047219974Smav g_io_deliver(bp, 0); 1048219974Smav} 1049219974Smav 1050219974Smavstatic void 1051219974Smavg_raid_start(struct bio *bp) 1052219974Smav{ 1053219974Smav struct g_raid_softc *sc; 1054219974Smav 1055219974Smav sc = bp->bio_to->geom->softc; 1056219974Smav /* 1057219974Smav * If sc == NULL or there are no valid disks, provider's error 1058219974Smav * should be set and g_raid_start() should not be called at all. 1059219974Smav */ 1060219974Smav// KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, 1061219974Smav// ("Provider's error should be set (error=%d)(mirror=%s).", 1062219974Smav// bp->bio_to->error, bp->bio_to->name)); 1063219974Smav G_RAID_LOGREQ(3, bp, "Request received."); 1064219974Smav 1065219974Smav switch (bp->bio_cmd) { 1066219974Smav case BIO_READ: 1067219974Smav case BIO_WRITE: 1068219974Smav case BIO_DELETE: 1069219974Smav case BIO_FLUSH: 1070219974Smav break; 1071219974Smav case BIO_GETATTR: 1072219974Smav if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 1073219974Smav g_raid_kerneldump(sc, bp); 1074219974Smav else 1075219974Smav g_io_deliver(bp, EOPNOTSUPP); 1076219974Smav return; 1077219974Smav default: 1078219974Smav g_io_deliver(bp, EOPNOTSUPP); 1079219974Smav return; 1080219974Smav } 1081219974Smav mtx_lock(&sc->sc_queue_mtx); 1082219974Smav bioq_disksort(&sc->sc_queue, bp); 1083219974Smav mtx_unlock(&sc->sc_queue_mtx); 1084219974Smav if (!dumping) { 1085219974Smav G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); 1086219974Smav wakeup(sc); 1087219974Smav } 1088219974Smav} 1089219974Smav 1090219974Smavstatic int 1091219974Smavg_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) 1092219974Smav{ 1093219974Smav /* 1094219974Smav * 5 cases: 1095219974Smav * (1) bp entirely below NO 1096219974Smav * (2) bp entirely above NO 1097219974Smav * (3) bp start below, but end in range YES 1098219974Smav * (4) bp entirely within YES 1099219974Smav * (5) bp starts within, ends above YES 1100219974Smav * 1101219974Smav * lock range 10-19 (offset 10 length 10) 1102219974Smav * (1) 1-5: first if kicks it out 1103219974Smav * (2) 30-35: second if kicks it out 1104219974Smav * (3) 5-15: passes both ifs 1105219974Smav * (4) 12-14: passes both ifs 1106219974Smav * (5) 19-20: passes both 1107219974Smav */ 1108219974Smav off_t lend = lstart + len - 1; 1109219974Smav off_t bstart = bp->bio_offset; 1110219974Smav off_t bend = bp->bio_offset + bp->bio_length - 1; 1111219974Smav 1112219974Smav if (bend < lstart) 1113219974Smav return (0); 1114219974Smav if (lend < bstart) 1115219974Smav return (0); 1116219974Smav return (1); 1117219974Smav} 1118219974Smav 1119219974Smavstatic int 1120219974Smavg_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) 1121219974Smav{ 1122219974Smav struct g_raid_lock *lp; 1123219974Smav 1124219974Smav sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); 1125219974Smav 1126219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1127219974Smav if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) 1128219974Smav return (1); 1129219974Smav } 1130219974Smav return (0); 1131219974Smav} 1132219974Smav 1133219974Smavstatic void 1134219974Smavg_raid_start_request(struct bio *bp) 1135219974Smav{ 1136219974Smav struct g_raid_softc *sc; 1137219974Smav struct g_raid_volume *vol; 1138219974Smav 1139219974Smav sc = bp->bio_to->geom->softc; 1140219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1141219974Smav vol = bp->bio_to->private; 1142219974Smav 1143219974Smav /* 1144219974Smav * Check to see if this item is in a locked range. If so, 1145219974Smav * queue it to our locked queue and return. We'll requeue 1146219974Smav * it when the range is unlocked. Internal I/O for the 1147219974Smav * rebuild/rescan/recovery process is excluded from this 1148219974Smav * check so we can actually do the recovery. 1149219974Smav */ 1150219974Smav if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && 1151219974Smav g_raid_is_in_locked_range(vol, bp)) { 1152219974Smav G_RAID_LOGREQ(3, bp, "Defer request."); 1153219974Smav bioq_insert_tail(&vol->v_locked, bp); 1154219974Smav return; 1155219974Smav } 1156219974Smav 1157219974Smav /* 1158219974Smav * If we're actually going to do the write/delete, then 1159219974Smav * update the idle stats for the volume. 1160219974Smav */ 1161219974Smav if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1162219974Smav if (!vol->v_dirty) 1163219974Smav g_raid_dirty(vol); 1164219974Smav vol->v_writes++; 1165219974Smav } 1166219974Smav 1167219974Smav /* 1168219974Smav * Put request onto inflight queue, so we can check if new 1169219974Smav * synchronization requests don't collide with it. Then tell 1170219974Smav * the transformation layer to start the I/O. 1171219974Smav */ 1172219974Smav bioq_insert_tail(&vol->v_inflight, bp); 1173219974Smav G_RAID_LOGREQ(4, bp, "Request started"); 1174219974Smav G_RAID_TR_IOSTART(vol->v_tr, bp); 1175219974Smav} 1176219974Smav 1177219974Smavstatic void 1178219974Smavg_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) 1179219974Smav{ 1180219974Smav off_t off, len; 1181219974Smav struct bio *nbp; 1182219974Smav struct g_raid_lock *lp; 1183219974Smav 1184219974Smav vol->v_pending_lock = 0; 1185219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1186219974Smav if (lp->l_pending) { 1187219974Smav off = lp->l_offset; 1188219974Smav len = lp->l_length; 1189219974Smav lp->l_pending = 0; 1190219974Smav TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { 1191219974Smav if (g_raid_bio_overlaps(nbp, off, len)) 1192219974Smav lp->l_pending++; 1193219974Smav } 1194219974Smav if (lp->l_pending) { 1195219974Smav vol->v_pending_lock = 1; 1196219974Smav G_RAID_DEBUG1(4, vol->v_softc, 1197219974Smav "Deferred lock(%jd, %jd) has %d pending", 1198219974Smav (intmax_t)off, (intmax_t)(off + len), 1199219974Smav lp->l_pending); 1200219974Smav continue; 1201219974Smav } 1202219974Smav G_RAID_DEBUG1(4, vol->v_softc, 1203219974Smav "Deferred lock of %jd to %jd completed", 1204219974Smav (intmax_t)off, (intmax_t)(off + len)); 1205219974Smav G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1206219974Smav } 1207219974Smav } 1208219974Smav} 1209219974Smav 1210219974Smavvoid 1211219974Smavg_raid_iodone(struct bio *bp, int error) 1212219974Smav{ 1213219974Smav struct g_raid_softc *sc; 1214219974Smav struct g_raid_volume *vol; 1215219974Smav 1216219974Smav sc = bp->bio_to->geom->softc; 1217219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1218219974Smav vol = bp->bio_to->private; 1219219974Smav G_RAID_LOGREQ(3, bp, "Request done: %d.", error); 1220219974Smav 1221219974Smav /* Update stats if we done write/delete. */ 1222219974Smav if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1223219974Smav vol->v_writes--; 1224219974Smav vol->v_last_write = time_uptime; 1225219974Smav } 1226219974Smav 1227219974Smav bioq_remove(&vol->v_inflight, bp); 1228219974Smav if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) 1229219974Smav g_raid_finish_with_locked_ranges(vol, bp); 1230219974Smav getmicrouptime(&vol->v_last_done); 1231219974Smav g_io_deliver(bp, error); 1232219974Smav} 1233219974Smav 1234219974Smavint 1235219974Smavg_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, 1236219974Smav struct bio *ignore, void *argp) 1237219974Smav{ 1238219974Smav struct g_raid_softc *sc; 1239219974Smav struct g_raid_lock *lp; 1240219974Smav struct bio *bp; 1241219974Smav 1242219974Smav sc = vol->v_softc; 1243219974Smav lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); 1244219974Smav LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); 1245219974Smav lp->l_offset = off; 1246219974Smav lp->l_length = len; 1247219974Smav lp->l_callback_arg = argp; 1248219974Smav 1249219974Smav lp->l_pending = 0; 1250219974Smav TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { 1251219974Smav if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) 1252219974Smav lp->l_pending++; 1253219974Smav } 1254219974Smav 1255219974Smav /* 1256219974Smav * If there are any writes that are pending, we return EBUSY. All 1257219974Smav * callers will have to wait until all pending writes clear. 1258219974Smav */ 1259219974Smav if (lp->l_pending > 0) { 1260219974Smav vol->v_pending_lock = 1; 1261219974Smav G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", 1262219974Smav (intmax_t)off, (intmax_t)(off+len), lp->l_pending); 1263219974Smav return (EBUSY); 1264219974Smav } 1265219974Smav G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", 1266219974Smav (intmax_t)off, (intmax_t)(off+len)); 1267219974Smav G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1268219974Smav return (0); 1269219974Smav} 1270219974Smav 1271219974Smavint 1272219974Smavg_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) 1273219974Smav{ 1274219974Smav struct g_raid_lock *lp; 1275219974Smav struct g_raid_softc *sc; 1276219974Smav struct bio *bp; 1277219974Smav 1278219974Smav sc = vol->v_softc; 1279219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1280219974Smav if (lp->l_offset == off && lp->l_length == len) { 1281219974Smav LIST_REMOVE(lp, l_next); 1282219974Smav /* XXX 1283219974Smav * Right now we just put them all back on the queue 1284219974Smav * and hope for the best. We hope this because any 1285219974Smav * locked ranges will go right back on this list 1286219974Smav * when the worker thread runs. 1287219974Smav * XXX 1288219974Smav */ 1289219974Smav G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", 1290219974Smav (intmax_t)lp->l_offset, 1291219974Smav (intmax_t)(lp->l_offset+lp->l_length)); 1292219974Smav mtx_lock(&sc->sc_queue_mtx); 1293219974Smav while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) 1294219974Smav bioq_disksort(&sc->sc_queue, bp); 1295219974Smav mtx_unlock(&sc->sc_queue_mtx); 1296219974Smav free(lp, M_RAID); 1297219974Smav return (0); 1298219974Smav } 1299219974Smav } 1300219974Smav return (EINVAL); 1301219974Smav} 1302219974Smav 1303219974Smavvoid 1304219974Smavg_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) 1305219974Smav{ 1306219974Smav struct g_consumer *cp; 1307219974Smav struct g_raid_disk *disk, *tdisk; 1308219974Smav 1309219974Smav bp->bio_caller1 = sd; 1310219974Smav 1311219974Smav /* 1312219974Smav * Make sure that the disk is present. Generally it is a task of 1313219974Smav * transformation layers to not send requests to absent disks, but 1314219974Smav * it is better to be safe and report situation then sorry. 1315219974Smav */ 1316219974Smav if (sd->sd_disk == NULL) { 1317219974Smav G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); 1318219974Smavnodisk: 1319219974Smav bp->bio_from = NULL; 1320219974Smav bp->bio_to = NULL; 1321219974Smav bp->bio_error = ENXIO; 1322219974Smav g_raid_disk_done(bp); 1323219974Smav return; 1324219974Smav } 1325219974Smav disk = sd->sd_disk; 1326219974Smav if (disk->d_state != G_RAID_DISK_S_ACTIVE && 1327219974Smav disk->d_state != G_RAID_DISK_S_FAILED) { 1328219974Smav G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " 1329219974Smav "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 1330219974Smav goto nodisk; 1331219974Smav } 1332219974Smav 1333219974Smav cp = disk->d_consumer; 1334219974Smav bp->bio_from = cp; 1335219974Smav bp->bio_to = cp->provider; 1336219974Smav cp->index++; 1337219974Smav 1338219974Smav /* Update average disks load. */ 1339219974Smav TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { 1340219974Smav if (tdisk->d_consumer == NULL) 1341219974Smav tdisk->d_load = 0; 1342219974Smav else 1343219974Smav tdisk->d_load = (tdisk->d_consumer->index * 1344219974Smav G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; 1345219974Smav } 1346219974Smav 1347219974Smav disk->d_last_offset = bp->bio_offset + bp->bio_length; 1348219974Smav if (dumping) { 1349219974Smav G_RAID_LOGREQ(3, bp, "Sending dumping request."); 1350219974Smav if (bp->bio_cmd == BIO_WRITE) { 1351219974Smav bp->bio_error = g_raid_subdisk_kerneldump(sd, 1352219974Smav bp->bio_data, 0, bp->bio_offset, bp->bio_length); 1353219974Smav } else 1354219974Smav bp->bio_error = EOPNOTSUPP; 1355219974Smav g_raid_disk_done(bp); 1356219974Smav } else { 1357219974Smav bp->bio_done = g_raid_disk_done; 1358219974Smav bp->bio_offset += sd->sd_offset; 1359219974Smav G_RAID_LOGREQ(3, bp, "Sending request."); 1360219974Smav g_io_request(bp, cp); 1361219974Smav } 1362219974Smav} 1363219974Smav 1364219974Smavint 1365219974Smavg_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, 1366219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1367219974Smav{ 1368219974Smav 1369219974Smav if (sd->sd_disk == NULL) 1370219974Smav return (ENXIO); 1371219974Smav if (sd->sd_disk->d_kd.di.dumper == NULL) 1372219974Smav return (EOPNOTSUPP); 1373219974Smav return (dump_write(&sd->sd_disk->d_kd.di, 1374219974Smav virtual, physical, 1375219974Smav sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, 1376219974Smav length)); 1377219974Smav} 1378219974Smav 1379219974Smavstatic void 1380219974Smavg_raid_disk_done(struct bio *bp) 1381219974Smav{ 1382219974Smav struct g_raid_softc *sc; 1383219974Smav struct g_raid_subdisk *sd; 1384219974Smav 1385219974Smav sd = bp->bio_caller1; 1386219974Smav sc = sd->sd_softc; 1387219974Smav mtx_lock(&sc->sc_queue_mtx); 1388219974Smav bioq_disksort(&sc->sc_queue, bp); 1389219974Smav mtx_unlock(&sc->sc_queue_mtx); 1390219974Smav if (!dumping) 1391219974Smav wakeup(sc); 1392219974Smav} 1393219974Smav 1394219974Smavstatic void 1395219974Smavg_raid_disk_done_request(struct bio *bp) 1396219974Smav{ 1397219974Smav struct g_raid_softc *sc; 1398219974Smav struct g_raid_disk *disk; 1399219974Smav struct g_raid_subdisk *sd; 1400219974Smav struct g_raid_volume *vol; 1401219974Smav 1402219974Smav g_topology_assert_not(); 1403219974Smav 1404219974Smav G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); 1405219974Smav sd = bp->bio_caller1; 1406219974Smav sc = sd->sd_softc; 1407219974Smav vol = sd->sd_volume; 1408219974Smav if (bp->bio_from != NULL) { 1409219974Smav bp->bio_from->index--; 1410219974Smav disk = bp->bio_from->private; 1411219974Smav if (disk == NULL) 1412219974Smav g_raid_kill_consumer(sc, bp->bio_from); 1413219974Smav } 1414219974Smav bp->bio_offset -= sd->sd_offset; 1415219974Smav 1416219974Smav G_RAID_TR_IODONE(vol->v_tr, sd, bp); 1417219974Smav} 1418219974Smav 1419219974Smavstatic void 1420219974Smavg_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) 1421219974Smav{ 1422219974Smav 1423219974Smav if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) 1424219974Smav ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); 1425219974Smav else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) 1426219974Smav ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); 1427219974Smav else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) 1428219974Smav ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); 1429219974Smav else 1430219974Smav ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); 1431219974Smav if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { 1432219974Smav KASSERT(ep->e_error == 0, 1433219974Smav ("Error cannot be handled.")); 1434219974Smav g_raid_event_free(ep); 1435219974Smav } else { 1436219974Smav ep->e_flags |= G_RAID_EVENT_DONE; 1437219974Smav G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); 1438219974Smav mtx_lock(&sc->sc_queue_mtx); 1439219974Smav wakeup(ep); 1440219974Smav mtx_unlock(&sc->sc_queue_mtx); 1441219974Smav } 1442219974Smav} 1443219974Smav 1444219974Smav/* 1445219974Smav * Worker thread. 1446219974Smav */ 1447219974Smavstatic void 1448219974Smavg_raid_worker(void *arg) 1449219974Smav{ 1450219974Smav struct g_raid_softc *sc; 1451219974Smav struct g_raid_event *ep; 1452219974Smav struct g_raid_volume *vol; 1453219974Smav struct bio *bp; 1454219974Smav struct timeval now, t; 1455219974Smav int timeout, rv; 1456219974Smav 1457219974Smav sc = arg; 1458219974Smav thread_lock(curthread); 1459219974Smav sched_prio(curthread, PRIBIO); 1460219974Smav thread_unlock(curthread); 1461219974Smav 1462219974Smav sx_xlock(&sc->sc_lock); 1463219974Smav for (;;) { 1464219974Smav mtx_lock(&sc->sc_queue_mtx); 1465219974Smav /* 1466219974Smav * First take a look at events. 1467219974Smav * This is important to handle events before any I/O requests. 1468219974Smav */ 1469219974Smav bp = NULL; 1470219974Smav vol = NULL; 1471219974Smav rv = 0; 1472219974Smav ep = TAILQ_FIRST(&sc->sc_events); 1473219974Smav if (ep != NULL) 1474219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1475219974Smav else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) 1476219974Smav ; 1477219974Smav else { 1478219974Smav getmicrouptime(&now); 1479219974Smav t = now; 1480219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1481219974Smav if (bioq_first(&vol->v_inflight) == NULL && 1482219974Smav vol->v_tr && 1483219974Smav timevalcmp(&vol->v_last_done, &t, < )) 1484219974Smav t = vol->v_last_done; 1485219974Smav } 1486219974Smav timevalsub(&t, &now); 1487219974Smav timeout = g_raid_idle_threshold + 1488219974Smav t.tv_sec * 1000000 + t.tv_usec; 1489219974Smav if (timeout > 0) { 1490219974Smav /* 1491219974Smav * Two steps to avoid overflows at HZ=1000 1492219974Smav * and idle timeouts > 2.1s. Some rounding 1493219974Smav * errors can occur, but they are < 1tick, 1494219974Smav * which is deemed to be close enough for 1495219974Smav * this purpose. 1496219974Smav */ 1497219974Smav int micpertic = 1000000 / hz; 1498219974Smav timeout = (timeout + micpertic - 1) / micpertic; 1499219974Smav sx_xunlock(&sc->sc_lock); 1500219974Smav MSLEEP(rv, sc, &sc->sc_queue_mtx, 1501219974Smav PRIBIO | PDROP, "-", timeout); 1502219974Smav sx_xlock(&sc->sc_lock); 1503219974Smav goto process; 1504219974Smav } else 1505219974Smav rv = EWOULDBLOCK; 1506219974Smav } 1507219974Smav mtx_unlock(&sc->sc_queue_mtx); 1508219974Smavprocess: 1509219974Smav if (ep != NULL) { 1510219974Smav g_raid_handle_event(sc, ep); 1511219974Smav } else if (bp != NULL) { 1512219974Smav if (bp->bio_to != NULL && 1513219974Smav bp->bio_to->geom == sc->sc_geom) 1514219974Smav g_raid_start_request(bp); 1515219974Smav else 1516219974Smav g_raid_disk_done_request(bp); 1517219974Smav } else if (rv == EWOULDBLOCK) { 1518219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1519219974Smav if (vol->v_writes == 0 && vol->v_dirty) 1520219974Smav g_raid_clean(vol, -1); 1521219974Smav if (bioq_first(&vol->v_inflight) == NULL && 1522219974Smav vol->v_tr) { 1523219974Smav t.tv_sec = g_raid_idle_threshold / 1000000; 1524219974Smav t.tv_usec = g_raid_idle_threshold % 1000000; 1525219974Smav timevaladd(&t, &vol->v_last_done); 1526219974Smav getmicrouptime(&now); 1527219974Smav if (timevalcmp(&t, &now, <= )) { 1528219974Smav G_RAID_TR_IDLE(vol->v_tr); 1529219974Smav vol->v_last_done = now; 1530219974Smav } 1531219974Smav } 1532219974Smav } 1533219974Smav } 1534219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) 1535219974Smav g_raid_destroy_node(sc, 1); /* May not return. */ 1536219974Smav } 1537219974Smav} 1538219974Smav 1539219974Smavstatic void 1540219974Smavg_raid_poll(struct g_raid_softc *sc) 1541219974Smav{ 1542219974Smav struct g_raid_event *ep; 1543219974Smav struct bio *bp; 1544219974Smav 1545219974Smav sx_xlock(&sc->sc_lock); 1546219974Smav mtx_lock(&sc->sc_queue_mtx); 1547219974Smav /* 1548219974Smav * First take a look at events. 1549219974Smav * This is important to handle events before any I/O requests. 1550219974Smav */ 1551219974Smav ep = TAILQ_FIRST(&sc->sc_events); 1552219974Smav if (ep != NULL) { 1553219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1554219974Smav mtx_unlock(&sc->sc_queue_mtx); 1555219974Smav g_raid_handle_event(sc, ep); 1556219974Smav goto out; 1557219974Smav } 1558219974Smav bp = bioq_takefirst(&sc->sc_queue); 1559219974Smav if (bp != NULL) { 1560219974Smav mtx_unlock(&sc->sc_queue_mtx); 1561219974Smav if (bp->bio_from == NULL || 1562219974Smav bp->bio_from->geom != sc->sc_geom) 1563219974Smav g_raid_start_request(bp); 1564219974Smav else 1565219974Smav g_raid_disk_done_request(bp); 1566219974Smav } 1567219974Smavout: 1568219974Smav sx_xunlock(&sc->sc_lock); 1569219974Smav} 1570219974Smav 1571219974Smavstatic void 1572219974Smavg_raid_launch_provider(struct g_raid_volume *vol) 1573219974Smav{ 1574219974Smav struct g_raid_disk *disk; 1575219974Smav struct g_raid_softc *sc; 1576219974Smav struct g_provider *pp; 1577219974Smav char name[G_RAID_MAX_VOLUMENAME]; 1578219974Smav off_t off; 1579219974Smav 1580219974Smav sc = vol->v_softc; 1581219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1582219974Smav 1583219974Smav g_topology_lock(); 1584219974Smav /* Try to name provider with volume name. */ 1585219974Smav snprintf(name, sizeof(name), "raid/%s", vol->v_name); 1586219974Smav if (g_raid_name_format == 0 || vol->v_name[0] == 0 || 1587219974Smav g_provider_by_name(name) != NULL) { 1588219974Smav /* Otherwise use sequential volume number. */ 1589219974Smav snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); 1590219974Smav } 1591219974Smav pp = g_new_providerf(sc->sc_geom, "%s", name); 1592219974Smav pp->private = vol; 1593219974Smav pp->mediasize = vol->v_mediasize; 1594219974Smav pp->sectorsize = vol->v_sectorsize; 1595219974Smav pp->stripesize = 0; 1596219974Smav pp->stripeoffset = 0; 1597219974Smav if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 1598219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || 1599219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || 1600219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { 1601219974Smav if ((disk = vol->v_subdisks[0].sd_disk) != NULL && 1602219974Smav disk->d_consumer != NULL && 1603219974Smav disk->d_consumer->provider != NULL) { 1604219974Smav pp->stripesize = disk->d_consumer->provider->stripesize; 1605219974Smav off = disk->d_consumer->provider->stripeoffset; 1606219974Smav pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; 1607219974Smav if (off > 0) 1608219974Smav pp->stripeoffset %= off; 1609219974Smav } 1610219974Smav if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { 1611219974Smav pp->stripesize *= (vol->v_disks_count - 1); 1612219974Smav pp->stripeoffset *= (vol->v_disks_count - 1); 1613219974Smav } 1614219974Smav } else 1615219974Smav pp->stripesize = vol->v_strip_size; 1616219974Smav vol->v_provider = pp; 1617219974Smav g_error_provider(pp, 0); 1618219974Smav g_topology_unlock(); 1619219974Smav G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", 1620219974Smav pp->name, vol->v_name); 1621219974Smav} 1622219974Smav 1623219974Smavstatic void 1624219974Smavg_raid_destroy_provider(struct g_raid_volume *vol) 1625219974Smav{ 1626219974Smav struct g_raid_softc *sc; 1627219974Smav struct g_provider *pp; 1628219974Smav struct bio *bp, *tmp; 1629219974Smav 1630219974Smav g_topology_assert_not(); 1631219974Smav sc = vol->v_softc; 1632219974Smav pp = vol->v_provider; 1633219974Smav KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); 1634219974Smav 1635219974Smav g_topology_lock(); 1636219974Smav g_error_provider(pp, ENXIO); 1637219974Smav mtx_lock(&sc->sc_queue_mtx); 1638219974Smav TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { 1639219974Smav if (bp->bio_to != pp) 1640219974Smav continue; 1641219974Smav bioq_remove(&sc->sc_queue, bp); 1642219974Smav g_io_deliver(bp, ENXIO); 1643219974Smav } 1644219974Smav mtx_unlock(&sc->sc_queue_mtx); 1645219974Smav G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", 1646219974Smav pp->name, vol->v_name); 1647219974Smav g_wither_provider(pp, ENXIO); 1648219974Smav g_topology_unlock(); 1649219974Smav vol->v_provider = NULL; 1650219974Smav} 1651219974Smav 1652219974Smav/* 1653219974Smav * Update device state. 1654219974Smav */ 1655219974Smavstatic int 1656219974Smavg_raid_update_volume(struct g_raid_volume *vol, u_int event) 1657219974Smav{ 1658219974Smav struct g_raid_softc *sc; 1659219974Smav 1660219974Smav sc = vol->v_softc; 1661219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1662219974Smav 1663219974Smav G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", 1664219974Smav g_raid_volume_event2str(event), 1665219974Smav vol->v_name); 1666219974Smav switch (event) { 1667219974Smav case G_RAID_VOLUME_E_DOWN: 1668219974Smav if (vol->v_provider != NULL) 1669219974Smav g_raid_destroy_provider(vol); 1670219974Smav break; 1671219974Smav case G_RAID_VOLUME_E_UP: 1672219974Smav if (vol->v_provider == NULL) 1673219974Smav g_raid_launch_provider(vol); 1674219974Smav break; 1675219974Smav case G_RAID_VOLUME_E_START: 1676219974Smav if (vol->v_tr) 1677219974Smav G_RAID_TR_START(vol->v_tr); 1678219974Smav return (0); 1679219974Smav default: 1680219974Smav if (sc->sc_md) 1681219974Smav G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); 1682219974Smav return (0); 1683219974Smav } 1684219974Smav 1685219974Smav /* Manage root mount release. */ 1686219974Smav if (vol->v_starting) { 1687219974Smav vol->v_starting = 0; 1688219974Smav G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); 1689219974Smav root_mount_rel(vol->v_rootmount); 1690219974Smav vol->v_rootmount = NULL; 1691219974Smav } 1692219974Smav if (vol->v_stopping && vol->v_provider_open == 0) 1693219974Smav g_raid_destroy_volume(vol); 1694219974Smav return (0); 1695219974Smav} 1696219974Smav 1697219974Smav/* 1698219974Smav * Update subdisk state. 1699219974Smav */ 1700219974Smavstatic int 1701219974Smavg_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) 1702219974Smav{ 1703219974Smav struct g_raid_softc *sc; 1704219974Smav struct g_raid_volume *vol; 1705219974Smav 1706219974Smav sc = sd->sd_softc; 1707219974Smav vol = sd->sd_volume; 1708219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1709219974Smav 1710219974Smav G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", 1711219974Smav g_raid_subdisk_event2str(event), 1712219974Smav vol->v_name, sd->sd_pos, 1713219974Smav sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); 1714219974Smav if (vol->v_tr) 1715219974Smav G_RAID_TR_EVENT(vol->v_tr, sd, event); 1716219974Smav 1717219974Smav return (0); 1718219974Smav} 1719219974Smav 1720219974Smav/* 1721219974Smav * Update disk state. 1722219974Smav */ 1723219974Smavstatic int 1724219974Smavg_raid_update_disk(struct g_raid_disk *disk, u_int event) 1725219974Smav{ 1726219974Smav struct g_raid_softc *sc; 1727219974Smav 1728219974Smav sc = disk->d_softc; 1729219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1730219974Smav 1731219974Smav G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", 1732219974Smav g_raid_disk_event2str(event), 1733219974Smav g_raid_get_diskname(disk)); 1734219974Smav 1735219974Smav if (sc->sc_md) 1736219974Smav G_RAID_MD_EVENT(sc->sc_md, disk, event); 1737219974Smav return (0); 1738219974Smav} 1739219974Smav 1740219974Smav/* 1741219974Smav * Node event. 1742219974Smav */ 1743219974Smavstatic int 1744219974Smavg_raid_update_node(struct g_raid_softc *sc, u_int event) 1745219974Smav{ 1746219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1747219974Smav 1748219974Smav G_RAID_DEBUG1(2, sc, "Event %s for the array.", 1749219974Smav g_raid_node_event2str(event)); 1750219974Smav 1751219974Smav if (event == G_RAID_NODE_E_WAKE) 1752219974Smav return (0); 1753219974Smav if (sc->sc_md) 1754219974Smav G_RAID_MD_EVENT(sc->sc_md, NULL, event); 1755219974Smav return (0); 1756219974Smav} 1757219974Smav 1758219974Smavstatic int 1759219974Smavg_raid_access(struct g_provider *pp, int acr, int acw, int ace) 1760219974Smav{ 1761219974Smav struct g_raid_volume *vol; 1762219974Smav struct g_raid_softc *sc; 1763220210Smav int dcw, opens, error = 0; 1764219974Smav 1765219974Smav g_topology_assert(); 1766219974Smav sc = pp->geom->softc; 1767219974Smav vol = pp->private; 1768219974Smav KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 1769219974Smav KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); 1770219974Smav 1771219974Smav G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, 1772219974Smav acr, acw, ace); 1773219974Smav dcw = pp->acw + acw; 1774219974Smav 1775219974Smav g_topology_unlock(); 1776219974Smav sx_xlock(&sc->sc_lock); 1777219974Smav /* Deny new opens while dying. */ 1778219974Smav if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { 1779219974Smav error = ENXIO; 1780219974Smav goto out; 1781219974Smav } 1782219974Smav if (dcw == 0 && vol->v_dirty) 1783219974Smav g_raid_clean(vol, dcw); 1784219974Smav vol->v_provider_open += acr + acw + ace; 1785219974Smav /* Handle delayed node destruction. */ 1786219974Smav if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && 1787219974Smav vol->v_provider_open == 0) { 1788219974Smav /* Count open volumes. */ 1789219974Smav opens = g_raid_nopens(sc); 1790219974Smav if (opens == 0) { 1791219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 1792219974Smav /* Wake up worker to make it selfdestruct. */ 1793219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1794219974Smav } 1795219974Smav } 1796219974Smav /* Handle open volume destruction. */ 1797219974Smav if (vol->v_stopping && vol->v_provider_open == 0) 1798219974Smav g_raid_destroy_volume(vol); 1799219974Smavout: 1800219974Smav sx_xunlock(&sc->sc_lock); 1801219974Smav g_topology_lock(); 1802219974Smav return (error); 1803219974Smav} 1804219974Smav 1805219974Smavstruct g_raid_softc * 1806219974Smavg_raid_create_node(struct g_class *mp, 1807219974Smav const char *name, struct g_raid_md_object *md) 1808219974Smav{ 1809219974Smav struct g_raid_softc *sc; 1810219974Smav struct g_geom *gp; 1811219974Smav int error; 1812219974Smav 1813219974Smav g_topology_assert(); 1814219974Smav G_RAID_DEBUG(1, "Creating array %s.", name); 1815219974Smav 1816219974Smav gp = g_new_geomf(mp, "%s", name); 1817219974Smav sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); 1818219974Smav gp->start = g_raid_start; 1819219974Smav gp->orphan = g_raid_orphan; 1820219974Smav gp->access = g_raid_access; 1821219974Smav gp->dumpconf = g_raid_dumpconf; 1822219974Smav 1823219974Smav sc->sc_md = md; 1824219974Smav sc->sc_geom = gp; 1825219974Smav sc->sc_flags = 0; 1826219974Smav TAILQ_INIT(&sc->sc_volumes); 1827219974Smav TAILQ_INIT(&sc->sc_disks); 1828234816Smav sx_init(&sc->sc_lock, "graid:lock"); 1829234816Smav mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF); 1830219974Smav TAILQ_INIT(&sc->sc_events); 1831219974Smav bioq_init(&sc->sc_queue); 1832219974Smav gp->softc = sc; 1833219974Smav error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, 1834219974Smav "g_raid %s", name); 1835219974Smav if (error != 0) { 1836219974Smav G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); 1837219974Smav mtx_destroy(&sc->sc_queue_mtx); 1838219974Smav sx_destroy(&sc->sc_lock); 1839219974Smav g_destroy_geom(sc->sc_geom); 1840219974Smav free(sc, M_RAID); 1841219974Smav return (NULL); 1842219974Smav } 1843219974Smav 1844219974Smav G_RAID_DEBUG1(0, sc, "Array %s created.", name); 1845219974Smav return (sc); 1846219974Smav} 1847219974Smav 1848219974Smavstruct g_raid_volume * 1849219974Smavg_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) 1850219974Smav{ 1851219974Smav struct g_raid_volume *vol, *vol1; 1852219974Smav int i; 1853219974Smav 1854219974Smav G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); 1855219974Smav vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); 1856219974Smav vol->v_softc = sc; 1857219974Smav strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); 1858219974Smav vol->v_state = G_RAID_VOLUME_S_STARTING; 1859219974Smav vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 1860219974Smav vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; 1861235076Smav vol->v_rotate_parity = 1; 1862219974Smav bioq_init(&vol->v_inflight); 1863219974Smav bioq_init(&vol->v_locked); 1864219974Smav LIST_INIT(&vol->v_locks); 1865219974Smav for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 1866219974Smav vol->v_subdisks[i].sd_softc = sc; 1867219974Smav vol->v_subdisks[i].sd_volume = vol; 1868219974Smav vol->v_subdisks[i].sd_pos = i; 1869219974Smav vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; 1870219974Smav } 1871219974Smav 1872219974Smav /* Find free ID for this volume. */ 1873219974Smav g_topology_lock(); 1874219974Smav vol1 = vol; 1875219974Smav if (id >= 0) { 1876219974Smav LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1877219974Smav if (vol1->v_global_id == id) 1878219974Smav break; 1879219974Smav } 1880219974Smav } 1881219974Smav if (vol1 != NULL) { 1882219974Smav for (id = 0; ; id++) { 1883219974Smav LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1884219974Smav if (vol1->v_global_id == id) 1885219974Smav break; 1886219974Smav } 1887219974Smav if (vol1 == NULL) 1888219974Smav break; 1889219974Smav } 1890219974Smav } 1891219974Smav vol->v_global_id = id; 1892219974Smav LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); 1893219974Smav g_topology_unlock(); 1894219974Smav 1895219974Smav /* Delay root mounting. */ 1896219974Smav vol->v_rootmount = root_mount_hold("GRAID"); 1897219974Smav G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); 1898219974Smav vol->v_starting = 1; 1899219974Smav TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); 1900219974Smav return (vol); 1901219974Smav} 1902219974Smav 1903219974Smavstruct g_raid_disk * 1904219974Smavg_raid_create_disk(struct g_raid_softc *sc) 1905219974Smav{ 1906219974Smav struct g_raid_disk *disk; 1907219974Smav 1908219974Smav G_RAID_DEBUG1(1, sc, "Creating disk."); 1909219974Smav disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); 1910219974Smav disk->d_softc = sc; 1911219974Smav disk->d_state = G_RAID_DISK_S_NONE; 1912219974Smav TAILQ_INIT(&disk->d_subdisks); 1913219974Smav TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); 1914219974Smav return (disk); 1915219974Smav} 1916219974Smav 1917219974Smavint g_raid_start_volume(struct g_raid_volume *vol) 1918219974Smav{ 1919219974Smav struct g_raid_tr_class *class; 1920219974Smav struct g_raid_tr_object *obj; 1921219974Smav int status; 1922219974Smav 1923219974Smav G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); 1924219974Smav LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { 1925219974Smav G_RAID_DEBUG1(2, vol->v_softc, 1926219974Smav "Tasting volume %s for %s transformation.", 1927219974Smav vol->v_name, class->name); 1928219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 1929219974Smav M_WAITOK); 1930219974Smav obj->tro_class = class; 1931219974Smav obj->tro_volume = vol; 1932219974Smav status = G_RAID_TR_TASTE(obj, vol); 1933219974Smav if (status != G_RAID_TR_TASTE_FAIL) 1934219974Smav break; 1935219974Smav kobj_delete((kobj_t)obj, M_RAID); 1936219974Smav } 1937219974Smav if (class == NULL) { 1938219974Smav G_RAID_DEBUG1(0, vol->v_softc, 1939219974Smav "No transformation module found for %s.", 1940219974Smav vol->v_name); 1941219974Smav vol->v_tr = NULL; 1942219974Smav g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); 1943219974Smav g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, 1944219974Smav G_RAID_EVENT_VOLUME); 1945219974Smav return (-1); 1946219974Smav } 1947219974Smav G_RAID_DEBUG1(2, vol->v_softc, 1948219974Smav "Transformation module %s chosen for %s.", 1949219974Smav class->name, vol->v_name); 1950219974Smav vol->v_tr = obj; 1951219974Smav return (0); 1952219974Smav} 1953219974Smav 1954219974Smavint 1955219974Smavg_raid_destroy_node(struct g_raid_softc *sc, int worker) 1956219974Smav{ 1957219974Smav struct g_raid_volume *vol, *tmpv; 1958219974Smav struct g_raid_disk *disk, *tmpd; 1959219974Smav int error = 0; 1960219974Smav 1961219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 1962219974Smav TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { 1963219974Smav if (g_raid_destroy_volume(vol)) 1964219974Smav error = EBUSY; 1965219974Smav } 1966219974Smav if (error) 1967219974Smav return (error); 1968219974Smav TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { 1969219974Smav if (g_raid_destroy_disk(disk)) 1970219974Smav error = EBUSY; 1971219974Smav } 1972219974Smav if (error) 1973219974Smav return (error); 1974219974Smav if (sc->sc_md) { 1975219974Smav G_RAID_MD_FREE(sc->sc_md); 1976219974Smav kobj_delete((kobj_t)sc->sc_md, M_RAID); 1977219974Smav sc->sc_md = NULL; 1978219974Smav } 1979219974Smav if (sc->sc_geom != NULL) { 1980219974Smav G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); 1981219974Smav g_topology_lock(); 1982219974Smav sc->sc_geom->softc = NULL; 1983219974Smav g_wither_geom(sc->sc_geom, ENXIO); 1984219974Smav g_topology_unlock(); 1985219974Smav sc->sc_geom = NULL; 1986219974Smav } else 1987219974Smav G_RAID_DEBUG(1, "Array destroyed."); 1988219974Smav if (worker) { 1989219974Smav g_raid_event_cancel(sc, sc); 1990219974Smav mtx_destroy(&sc->sc_queue_mtx); 1991219974Smav sx_xunlock(&sc->sc_lock); 1992219974Smav sx_destroy(&sc->sc_lock); 1993219974Smav wakeup(&sc->sc_stopping); 1994219974Smav free(sc, M_RAID); 1995219974Smav curthread->td_pflags &= ~TDP_GEOM; 1996219974Smav G_RAID_DEBUG(1, "Thread exiting."); 1997219974Smav kproc_exit(0); 1998219974Smav } else { 1999219974Smav /* Wake up worker to make it selfdestruct. */ 2000219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2001219974Smav } 2002219974Smav return (0); 2003219974Smav} 2004219974Smav 2005219974Smavint 2006219974Smavg_raid_destroy_volume(struct g_raid_volume *vol) 2007219974Smav{ 2008219974Smav struct g_raid_softc *sc; 2009219974Smav struct g_raid_disk *disk; 2010219974Smav int i; 2011219974Smav 2012219974Smav sc = vol->v_softc; 2013219974Smav G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); 2014219974Smav vol->v_stopping = 1; 2015219974Smav if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { 2016219974Smav if (vol->v_tr) { 2017219974Smav G_RAID_TR_STOP(vol->v_tr); 2018219974Smav return (EBUSY); 2019219974Smav } else 2020219974Smav vol->v_state = G_RAID_VOLUME_S_STOPPED; 2021219974Smav } 2022219974Smav if (g_raid_event_check(sc, vol) != 0) 2023219974Smav return (EBUSY); 2024219974Smav if (vol->v_provider != NULL) 2025219974Smav return (EBUSY); 2026219974Smav if (vol->v_provider_open != 0) 2027219974Smav return (EBUSY); 2028219974Smav if (vol->v_tr) { 2029219974Smav G_RAID_TR_FREE(vol->v_tr); 2030219974Smav kobj_delete((kobj_t)vol->v_tr, M_RAID); 2031219974Smav vol->v_tr = NULL; 2032219974Smav } 2033219974Smav if (vol->v_rootmount) 2034219974Smav root_mount_rel(vol->v_rootmount); 2035219974Smav g_topology_lock(); 2036219974Smav LIST_REMOVE(vol, v_global_next); 2037219974Smav g_topology_unlock(); 2038219974Smav TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); 2039219974Smav for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 2040219974Smav g_raid_event_cancel(sc, &vol->v_subdisks[i]); 2041219974Smav disk = vol->v_subdisks[i].sd_disk; 2042219974Smav if (disk == NULL) 2043219974Smav continue; 2044219974Smav TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); 2045219974Smav } 2046219974Smav G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); 2047219974Smav if (sc->sc_md) 2048219974Smav G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); 2049219974Smav g_raid_event_cancel(sc, vol); 2050219974Smav free(vol, M_RAID); 2051219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) { 2052219974Smav /* Wake up worker to let it selfdestruct. */ 2053219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2054219974Smav } 2055219974Smav return (0); 2056219974Smav} 2057219974Smav 2058219974Smavint 2059219974Smavg_raid_destroy_disk(struct g_raid_disk *disk) 2060219974Smav{ 2061219974Smav struct g_raid_softc *sc; 2062219974Smav struct g_raid_subdisk *sd, *tmp; 2063219974Smav 2064219974Smav sc = disk->d_softc; 2065219974Smav G_RAID_DEBUG1(2, sc, "Destroying disk."); 2066219974Smav if (disk->d_consumer) { 2067219974Smav g_raid_kill_consumer(sc, disk->d_consumer); 2068219974Smav disk->d_consumer = NULL; 2069219974Smav } 2070219974Smav TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { 2071219974Smav g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); 2072219974Smav g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 2073219974Smav G_RAID_EVENT_SUBDISK); 2074219974Smav TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); 2075219974Smav sd->sd_disk = NULL; 2076219974Smav } 2077219974Smav TAILQ_REMOVE(&sc->sc_disks, disk, d_next); 2078219974Smav if (sc->sc_md) 2079219974Smav G_RAID_MD_FREE_DISK(sc->sc_md, disk); 2080219974Smav g_raid_event_cancel(sc, disk); 2081219974Smav free(disk, M_RAID); 2082219974Smav return (0); 2083219974Smav} 2084219974Smav 2085219974Smavint 2086219974Smavg_raid_destroy(struct g_raid_softc *sc, int how) 2087219974Smav{ 2088219974Smav int opens; 2089219974Smav 2090219974Smav g_topology_assert_not(); 2091219974Smav if (sc == NULL) 2092219974Smav return (ENXIO); 2093219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 2094219974Smav 2095219974Smav /* Count open volumes. */ 2096219974Smav opens = g_raid_nopens(sc); 2097219974Smav 2098219974Smav /* React on some opened volumes. */ 2099219974Smav if (opens > 0) { 2100219974Smav switch (how) { 2101219974Smav case G_RAID_DESTROY_SOFT: 2102219974Smav G_RAID_DEBUG1(1, sc, 2103219974Smav "%d volumes are still open.", 2104219974Smav opens); 2105219974Smav return (EBUSY); 2106219974Smav case G_RAID_DESTROY_DELAYED: 2107219974Smav G_RAID_DEBUG1(1, sc, 2108219974Smav "Array will be destroyed on last close."); 2109219974Smav sc->sc_stopping = G_RAID_DESTROY_DELAYED; 2110219974Smav return (EBUSY); 2111219974Smav case G_RAID_DESTROY_HARD: 2112219974Smav G_RAID_DEBUG1(1, sc, 2113219974Smav "%d volumes are still open.", 2114219974Smav opens); 2115219974Smav } 2116219974Smav } 2117219974Smav 2118219974Smav /* Mark node for destruction. */ 2119219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 2120219974Smav /* Wake up worker to let it selfdestruct. */ 2121219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2122219974Smav /* Sleep until node destroyed. */ 2123219974Smav sx_sleep(&sc->sc_stopping, &sc->sc_lock, 2124219974Smav PRIBIO | PDROP, "r:destroy", 0); 2125219974Smav return (0); 2126219974Smav} 2127219974Smav 2128219974Smavstatic void 2129219974Smavg_raid_taste_orphan(struct g_consumer *cp) 2130219974Smav{ 2131219974Smav 2132219974Smav KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2133219974Smav cp->provider->name)); 2134219974Smav} 2135219974Smav 2136219974Smavstatic struct g_geom * 2137219974Smavg_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2138219974Smav{ 2139219974Smav struct g_consumer *cp; 2140219974Smav struct g_geom *gp, *geom; 2141219974Smav struct g_raid_md_class *class; 2142219974Smav struct g_raid_md_object *obj; 2143219974Smav int status; 2144219974Smav 2145219974Smav g_topology_assert(); 2146219974Smav g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2147219974Smav G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); 2148219974Smav 2149234940Smav gp = g_new_geomf(mp, "raid:taste"); 2150219974Smav /* 2151219974Smav * This orphan function should be never called. 2152219974Smav */ 2153219974Smav gp->orphan = g_raid_taste_orphan; 2154219974Smav cp = g_new_consumer(gp); 2155219974Smav g_attach(cp, pp); 2156219974Smav 2157219974Smav geom = NULL; 2158219974Smav LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2159219974Smav G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", 2160219974Smav pp->name, class->name); 2161219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2162219974Smav M_WAITOK); 2163219974Smav obj->mdo_class = class; 2164219974Smav status = G_RAID_MD_TASTE(obj, mp, cp, &geom); 2165219974Smav if (status != G_RAID_MD_TASTE_NEW) 2166219974Smav kobj_delete((kobj_t)obj, M_RAID); 2167219974Smav if (status != G_RAID_MD_TASTE_FAIL) 2168219974Smav break; 2169219974Smav } 2170219974Smav 2171219974Smav g_detach(cp); 2172219974Smav g_destroy_consumer(cp); 2173219974Smav g_destroy_geom(gp); 2174219974Smav G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); 2175219974Smav return (geom); 2176219974Smav} 2177219974Smav 2178219974Smavint 2179234940Smavg_raid_create_node_format(const char *format, struct gctl_req *req, 2180234940Smav struct g_geom **gp) 2181219974Smav{ 2182219974Smav struct g_raid_md_class *class; 2183219974Smav struct g_raid_md_object *obj; 2184219974Smav int status; 2185219974Smav 2186219974Smav G_RAID_DEBUG(2, "Creating array for %s metadata.", format); 2187219974Smav LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2188219974Smav if (strcasecmp(class->name, format) == 0) 2189219974Smav break; 2190219974Smav } 2191219974Smav if (class == NULL) { 2192219974Smav G_RAID_DEBUG(1, "No support for %s metadata.", format); 2193219974Smav return (G_RAID_MD_TASTE_FAIL); 2194219974Smav } 2195219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2196219974Smav M_WAITOK); 2197219974Smav obj->mdo_class = class; 2198234940Smav status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp); 2199219974Smav if (status != G_RAID_MD_TASTE_NEW) 2200219974Smav kobj_delete((kobj_t)obj, M_RAID); 2201219974Smav return (status); 2202219974Smav} 2203219974Smav 2204219974Smavstatic int 2205219974Smavg_raid_destroy_geom(struct gctl_req *req __unused, 2206219974Smav struct g_class *mp __unused, struct g_geom *gp) 2207219974Smav{ 2208219974Smav struct g_raid_softc *sc; 2209219974Smav int error; 2210219974Smav 2211219974Smav g_topology_unlock(); 2212219974Smav sc = gp->softc; 2213219974Smav sx_xlock(&sc->sc_lock); 2214219974Smav g_cancel_event(sc); 2215219974Smav error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); 2216219974Smav if (error != 0) 2217219974Smav sx_xunlock(&sc->sc_lock); 2218219974Smav g_topology_lock(); 2219219974Smav return (error); 2220219974Smav} 2221219974Smav 2222219974Smavvoid g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, 2223219974Smav struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2224219974Smav{ 2225219974Smav 2226219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) 2227219974Smav return; 2228219974Smav if (sc->sc_md) 2229219974Smav G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); 2230219974Smav} 2231219974Smav 2232219974Smavvoid g_raid_fail_disk(struct g_raid_softc *sc, 2233219974Smav struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2234219974Smav{ 2235219974Smav 2236219974Smav if (disk == NULL) 2237219974Smav disk = sd->sd_disk; 2238219974Smav if (disk == NULL) { 2239219974Smav G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); 2240219974Smav return; 2241219974Smav } 2242219974Smav if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 2243219974Smav G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " 2244219974Smav "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 2245219974Smav return; 2246219974Smav } 2247219974Smav if (sc->sc_md) 2248219974Smav G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); 2249219974Smav} 2250219974Smav 2251219974Smavstatic void 2252219974Smavg_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2253219974Smav struct g_consumer *cp, struct g_provider *pp) 2254219974Smav{ 2255219974Smav struct g_raid_softc *sc; 2256219974Smav struct g_raid_volume *vol; 2257219974Smav struct g_raid_subdisk *sd; 2258219974Smav struct g_raid_disk *disk; 2259219974Smav int i, s; 2260219974Smav 2261219974Smav g_topology_assert(); 2262219974Smav 2263219974Smav sc = gp->softc; 2264219974Smav if (sc == NULL) 2265219974Smav return; 2266219974Smav if (pp != NULL) { 2267219974Smav vol = pp->private; 2268219974Smav g_topology_unlock(); 2269219974Smav sx_xlock(&sc->sc_lock); 2270219974Smav sbuf_printf(sb, "%s<Label>%s</Label>\n", indent, 2271219974Smav vol->v_name); 2272219974Smav sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent, 2273219974Smav g_raid_volume_level2str(vol->v_raid_level, 2274219974Smav vol->v_raid_level_qualifier)); 2275219974Smav sbuf_printf(sb, 2276219974Smav "%s<Transformation>%s</Transformation>\n", indent, 2277219974Smav vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); 2278219974Smav sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2279219974Smav vol->v_disks_count); 2280219974Smav sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent, 2281219974Smav vol->v_strip_size); 2282219974Smav sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2283219974Smav g_raid_volume_state2str(vol->v_state)); 2284219974Smav sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent, 2285219974Smav vol->v_dirty ? "Yes" : "No"); 2286219974Smav sbuf_printf(sb, "%s<Subdisks>", indent); 2287219974Smav for (i = 0; i < vol->v_disks_count; i++) { 2288219974Smav sd = &vol->v_subdisks[i]; 2289219974Smav if (sd->sd_disk != NULL && 2290219974Smav sd->sd_disk->d_consumer != NULL) { 2291219974Smav sbuf_printf(sb, "%s ", 2292219974Smav g_raid_get_diskname(sd->sd_disk)); 2293219974Smav } else { 2294219974Smav sbuf_printf(sb, "NONE "); 2295219974Smav } 2296219974Smav sbuf_printf(sb, "(%s", 2297219974Smav g_raid_subdisk_state2str(sd->sd_state)); 2298219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2299219974Smav sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2300219974Smav sbuf_printf(sb, " %d%%", 2301219974Smav (int)(sd->sd_rebuild_pos * 100 / 2302219974Smav sd->sd_size)); 2303219974Smav } 2304219974Smav sbuf_printf(sb, ")"); 2305219974Smav if (i + 1 < vol->v_disks_count) 2306219974Smav sbuf_printf(sb, ", "); 2307219974Smav } 2308219974Smav sbuf_printf(sb, "</Subdisks>\n"); 2309219974Smav sx_xunlock(&sc->sc_lock); 2310219974Smav g_topology_lock(); 2311219974Smav } else if (cp != NULL) { 2312219974Smav disk = cp->private; 2313219974Smav if (disk == NULL) 2314219974Smav return; 2315219974Smav g_topology_unlock(); 2316219974Smav sx_xlock(&sc->sc_lock); 2317219974Smav sbuf_printf(sb, "%s<State>%s", indent, 2318219974Smav g_raid_disk_state2str(disk->d_state)); 2319219974Smav if (!TAILQ_EMPTY(&disk->d_subdisks)) { 2320219974Smav sbuf_printf(sb, " ("); 2321219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2322219974Smav sbuf_printf(sb, "%s", 2323219974Smav g_raid_subdisk_state2str(sd->sd_state)); 2324219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2325219974Smav sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2326219974Smav sbuf_printf(sb, " %d%%", 2327219974Smav (int)(sd->sd_rebuild_pos * 100 / 2328219974Smav sd->sd_size)); 2329219974Smav } 2330219974Smav if (TAILQ_NEXT(sd, sd_next)) 2331219974Smav sbuf_printf(sb, ", "); 2332219974Smav } 2333219974Smav sbuf_printf(sb, ")"); 2334219974Smav } 2335219974Smav sbuf_printf(sb, "</State>\n"); 2336219974Smav sbuf_printf(sb, "%s<Subdisks>", indent); 2337219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2338219974Smav sbuf_printf(sb, "r%d(%s):%d@%ju", 2339219974Smav sd->sd_volume->v_global_id, 2340219974Smav sd->sd_volume->v_name, 2341219974Smav sd->sd_pos, sd->sd_offset); 2342219974Smav if (TAILQ_NEXT(sd, sd_next)) 2343219974Smav sbuf_printf(sb, ", "); 2344219974Smav } 2345219974Smav sbuf_printf(sb, "</Subdisks>\n"); 2346219974Smav sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent, 2347219974Smav disk->d_read_errs); 2348219974Smav sx_xunlock(&sc->sc_lock); 2349219974Smav g_topology_lock(); 2350219974Smav } else { 2351219974Smav g_topology_unlock(); 2352219974Smav sx_xlock(&sc->sc_lock); 2353219974Smav if (sc->sc_md) { 2354219974Smav sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent, 2355219974Smav sc->sc_md->mdo_class->name); 2356219974Smav } 2357219974Smav if (!TAILQ_EMPTY(&sc->sc_volumes)) { 2358219974Smav s = 0xff; 2359219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 2360219974Smav if (vol->v_state < s) 2361219974Smav s = vol->v_state; 2362219974Smav } 2363219974Smav sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2364219974Smav g_raid_volume_state2str(s)); 2365219974Smav } 2366219974Smav sx_xunlock(&sc->sc_lock); 2367219974Smav g_topology_lock(); 2368219974Smav } 2369219974Smav} 2370219974Smav 2371219974Smavstatic void 2372219974Smavg_raid_shutdown_pre_sync(void *arg, int howto) 2373219974Smav{ 2374219974Smav struct g_class *mp; 2375219974Smav struct g_geom *gp, *gp2; 2376219974Smav struct g_raid_softc *sc; 2377219974Smav int error; 2378219974Smav 2379219974Smav mp = arg; 2380219974Smav DROP_GIANT(); 2381219974Smav g_topology_lock(); 2382219974Smav LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2383219974Smav if ((sc = gp->softc) == NULL) 2384219974Smav continue; 2385219974Smav g_topology_unlock(); 2386219974Smav sx_xlock(&sc->sc_lock); 2387219974Smav g_cancel_event(sc); 2388219974Smav error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); 2389219974Smav if (error != 0) 2390219974Smav sx_xunlock(&sc->sc_lock); 2391219974Smav g_topology_lock(); 2392219974Smav } 2393219974Smav g_topology_unlock(); 2394219974Smav PICKUP_GIANT(); 2395219974Smav} 2396219974Smav 2397219974Smavstatic void 2398219974Smavg_raid_init(struct g_class *mp) 2399219974Smav{ 2400219974Smav 2401219974Smav g_raid_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync, 2402219974Smav g_raid_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST); 2403219974Smav if (g_raid_pre_sync == NULL) 2404219974Smav G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); 2405219974Smav g_raid_started = 1; 2406219974Smav} 2407219974Smav 2408219974Smavstatic void 2409219974Smavg_raid_fini(struct g_class *mp) 2410219974Smav{ 2411219974Smav 2412219974Smav if (g_raid_pre_sync != NULL) 2413219974Smav EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_raid_pre_sync); 2414219974Smav g_raid_started = 0; 2415219974Smav} 2416219974Smav 2417219974Smavint 2418219974Smavg_raid_md_modevent(module_t mod, int type, void *arg) 2419219974Smav{ 2420219974Smav struct g_raid_md_class *class, *c, *nc; 2421219974Smav int error; 2422219974Smav 2423219974Smav error = 0; 2424219974Smav class = arg; 2425219974Smav switch (type) { 2426219974Smav case MOD_LOAD: 2427219974Smav c = LIST_FIRST(&g_raid_md_classes); 2428219974Smav if (c == NULL || c->mdc_priority > class->mdc_priority) 2429219974Smav LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); 2430219974Smav else { 2431219974Smav while ((nc = LIST_NEXT(c, mdc_list)) != NULL && 2432219974Smav nc->mdc_priority < class->mdc_priority) 2433219974Smav c = nc; 2434219974Smav LIST_INSERT_AFTER(c, class, mdc_list); 2435219974Smav } 2436219974Smav if (g_raid_started) 2437219974Smav g_retaste(&g_raid_class); 2438219974Smav break; 2439219974Smav case MOD_UNLOAD: 2440219974Smav LIST_REMOVE(class, mdc_list); 2441219974Smav break; 2442219974Smav default: 2443219974Smav error = EOPNOTSUPP; 2444219974Smav break; 2445219974Smav } 2446219974Smav 2447219974Smav return (error); 2448219974Smav} 2449219974Smav 2450219974Smavint 2451219974Smavg_raid_tr_modevent(module_t mod, int type, void *arg) 2452219974Smav{ 2453219974Smav struct g_raid_tr_class *class, *c, *nc; 2454219974Smav int error; 2455219974Smav 2456219974Smav error = 0; 2457219974Smav class = arg; 2458219974Smav switch (type) { 2459219974Smav case MOD_LOAD: 2460219974Smav c = LIST_FIRST(&g_raid_tr_classes); 2461219974Smav if (c == NULL || c->trc_priority > class->trc_priority) 2462219974Smav LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); 2463219974Smav else { 2464219974Smav while ((nc = LIST_NEXT(c, trc_list)) != NULL && 2465219974Smav nc->trc_priority < class->trc_priority) 2466219974Smav c = nc; 2467219974Smav LIST_INSERT_AFTER(c, class, trc_list); 2468219974Smav } 2469219974Smav break; 2470219974Smav case MOD_UNLOAD: 2471219974Smav LIST_REMOVE(class, trc_list); 2472219974Smav break; 2473219974Smav default: 2474219974Smav error = EOPNOTSUPP; 2475219974Smav break; 2476219974Smav } 2477219974Smav 2478219974Smav return (error); 2479219974Smav} 2480219974Smav 2481219974Smav/* 2482219974Smav * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) 2483219974Smav * to reduce module priority, allowing submodules to register them first. 2484219974Smav */ 2485219974Smavstatic moduledata_t g_raid_mod = { 2486219974Smav "g_raid", 2487219974Smav g_modevent, 2488219974Smav &g_raid_class 2489219974Smav}; 2490219974SmavDECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD); 2491219974SmavMODULE_VERSION(geom_raid, 0); 2492