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("$FreeBSD$"); 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"); 55240465Smavint g_raid_enable = 1; 56267992ShselaskySYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RWTUN, 57240465Smav &g_raid_enable, 0, "Enable on-disk metadata taste"); 58219974Smavu_int g_raid_aggressive_spare = 0; 59267992ShselaskySYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RWTUN, 60219974Smav &g_raid_aggressive_spare, 0, "Use disks without metadata as spare"); 61220790Smavu_int g_raid_debug = 0; 62267992ShselaskySYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid_debug, 0, 63219974Smav "Debug level"); 64219974Smavint g_raid_read_err_thresh = 10; 65267992ShselaskySYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RWTUN, 66219974Smav &g_raid_read_err_thresh, 0, 67219974Smav "Number of read errors equated to disk failure"); 68219974Smavu_int g_raid_start_timeout = 30; 69267992ShselaskySYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RWTUN, 70219974Smav &g_raid_start_timeout, 0, 71219974Smav "Time to wait for all array components"); 72219974Smavstatic u_int g_raid_clean_time = 5; 73267992ShselaskySYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RWTUN, 74219974Smav &g_raid_clean_time, 0, "Mark volume as clean when idling"); 75219974Smavstatic u_int g_raid_disconnect_on_failure = 1; 76267992ShselaskySYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 77219974Smav &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 78219974Smavstatic u_int g_raid_name_format = 0; 79267992ShselaskySYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RWTUN, 80219974Smav &g_raid_name_format, 0, "Providers name format."); 81219974Smavstatic u_int g_raid_idle_threshold = 1000000; 82267992ShselaskySYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RWTUN, 83219974Smav &g_raid_idle_threshold, 1000000, 84219974Smav "Time in microseconds to consider a volume idle."); 85219974Smav 86219974Smav#define MSLEEP(rv, ident, mtx, priority, wmesg, timeout) do { \ 87219974Smav G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 88219974Smav rv = msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 89219974Smav G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 90219974Smav} while (0) 91219974Smav 92219974SmavLIST_HEAD(, g_raid_md_class) g_raid_md_classes = 93219974Smav LIST_HEAD_INITIALIZER(g_raid_md_classes); 94219974Smav 95219974SmavLIST_HEAD(, g_raid_tr_class) g_raid_tr_classes = 96219974Smav LIST_HEAD_INITIALIZER(g_raid_tr_classes); 97219974Smav 98219974SmavLIST_HEAD(, g_raid_volume) g_raid_volumes = 99219974Smav LIST_HEAD_INITIALIZER(g_raid_volumes); 100219974Smav 101242314Smavstatic eventhandler_tag g_raid_post_sync = NULL; 102219974Smavstatic int g_raid_started = 0; 103242314Smavstatic int g_raid_shutdown = 0; 104219974Smav 105219974Smavstatic int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp, 106219974Smav struct g_geom *gp); 107219974Smavstatic g_taste_t g_raid_taste; 108219974Smavstatic void g_raid_init(struct g_class *mp); 109219974Smavstatic void g_raid_fini(struct g_class *mp); 110219974Smav 111219974Smavstruct g_class g_raid_class = { 112219974Smav .name = G_RAID_CLASS_NAME, 113219974Smav .version = G_VERSION, 114219974Smav .ctlreq = g_raid_ctl, 115219974Smav .taste = g_raid_taste, 116219974Smav .destroy_geom = g_raid_destroy_geom, 117219974Smav .init = g_raid_init, 118219974Smav .fini = g_raid_fini 119219974Smav}; 120219974Smav 121219974Smavstatic void g_raid_destroy_provider(struct g_raid_volume *vol); 122219974Smavstatic int g_raid_update_disk(struct g_raid_disk *disk, u_int event); 123219974Smavstatic int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event); 124219974Smavstatic int g_raid_update_volume(struct g_raid_volume *vol, u_int event); 125219974Smavstatic int g_raid_update_node(struct g_raid_softc *sc, u_int event); 126219974Smavstatic void g_raid_dumpconf(struct sbuf *sb, const char *indent, 127219974Smav struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 128219974Smavstatic void g_raid_start(struct bio *bp); 129219974Smavstatic void g_raid_start_request(struct bio *bp); 130219974Smavstatic void g_raid_disk_done(struct bio *bp); 131219974Smavstatic void g_raid_poll(struct g_raid_softc *sc); 132219974Smav 133219974Smavstatic const char * 134219974Smavg_raid_node_event2str(int event) 135219974Smav{ 136219974Smav 137219974Smav switch (event) { 138219974Smav case G_RAID_NODE_E_WAKE: 139219974Smav return ("WAKE"); 140219974Smav case G_RAID_NODE_E_START: 141219974Smav return ("START"); 142219974Smav default: 143219974Smav return ("INVALID"); 144219974Smav } 145219974Smav} 146219974Smav 147219974Smavconst char * 148219974Smavg_raid_disk_state2str(int state) 149219974Smav{ 150219974Smav 151219974Smav switch (state) { 152219974Smav case G_RAID_DISK_S_NONE: 153219974Smav return ("NONE"); 154219974Smav case G_RAID_DISK_S_OFFLINE: 155219974Smav return ("OFFLINE"); 156245326Smav case G_RAID_DISK_S_DISABLED: 157245326Smav return ("DISABLED"); 158219974Smav case G_RAID_DISK_S_FAILED: 159219974Smav return ("FAILED"); 160219974Smav case G_RAID_DISK_S_STALE_FAILED: 161219974Smav return ("STALE_FAILED"); 162219974Smav case G_RAID_DISK_S_SPARE: 163219974Smav return ("SPARE"); 164219974Smav case G_RAID_DISK_S_STALE: 165219974Smav return ("STALE"); 166219974Smav case G_RAID_DISK_S_ACTIVE: 167219974Smav return ("ACTIVE"); 168219974Smav default: 169219974Smav return ("INVALID"); 170219974Smav } 171219974Smav} 172219974Smav 173219974Smavstatic const char * 174219974Smavg_raid_disk_event2str(int event) 175219974Smav{ 176219974Smav 177219974Smav switch (event) { 178219974Smav case G_RAID_DISK_E_DISCONNECTED: 179219974Smav return ("DISCONNECTED"); 180219974Smav default: 181219974Smav return ("INVALID"); 182219974Smav } 183219974Smav} 184219974Smav 185219974Smavconst char * 186219974Smavg_raid_subdisk_state2str(int state) 187219974Smav{ 188219974Smav 189219974Smav switch (state) { 190219974Smav case G_RAID_SUBDISK_S_NONE: 191219974Smav return ("NONE"); 192219974Smav case G_RAID_SUBDISK_S_FAILED: 193219974Smav return ("FAILED"); 194219974Smav case G_RAID_SUBDISK_S_NEW: 195219974Smav return ("NEW"); 196219974Smav case G_RAID_SUBDISK_S_REBUILD: 197219974Smav return ("REBUILD"); 198219974Smav case G_RAID_SUBDISK_S_UNINITIALIZED: 199219974Smav return ("UNINITIALIZED"); 200219974Smav case G_RAID_SUBDISK_S_STALE: 201219974Smav return ("STALE"); 202219974Smav case G_RAID_SUBDISK_S_RESYNC: 203219974Smav return ("RESYNC"); 204219974Smav case G_RAID_SUBDISK_S_ACTIVE: 205219974Smav return ("ACTIVE"); 206219974Smav default: 207219974Smav return ("INVALID"); 208219974Smav } 209219974Smav} 210219974Smav 211219974Smavstatic const char * 212219974Smavg_raid_subdisk_event2str(int event) 213219974Smav{ 214219974Smav 215219974Smav switch (event) { 216219974Smav case G_RAID_SUBDISK_E_NEW: 217219974Smav return ("NEW"); 218239175Smav case G_RAID_SUBDISK_E_FAILED: 219239175Smav return ("FAILED"); 220219974Smav case G_RAID_SUBDISK_E_DISCONNECTED: 221219974Smav return ("DISCONNECTED"); 222219974Smav default: 223219974Smav return ("INVALID"); 224219974Smav } 225219974Smav} 226219974Smav 227219974Smavconst char * 228219974Smavg_raid_volume_state2str(int state) 229219974Smav{ 230219974Smav 231219974Smav switch (state) { 232219974Smav case G_RAID_VOLUME_S_STARTING: 233219974Smav return ("STARTING"); 234219974Smav case G_RAID_VOLUME_S_BROKEN: 235219974Smav return ("BROKEN"); 236219974Smav case G_RAID_VOLUME_S_DEGRADED: 237219974Smav return ("DEGRADED"); 238219974Smav case G_RAID_VOLUME_S_SUBOPTIMAL: 239219974Smav return ("SUBOPTIMAL"); 240219974Smav case G_RAID_VOLUME_S_OPTIMAL: 241219974Smav return ("OPTIMAL"); 242219974Smav case G_RAID_VOLUME_S_UNSUPPORTED: 243219974Smav return ("UNSUPPORTED"); 244219974Smav case G_RAID_VOLUME_S_STOPPED: 245219974Smav return ("STOPPED"); 246219974Smav default: 247219974Smav return ("INVALID"); 248219974Smav } 249219974Smav} 250219974Smav 251219974Smavstatic const char * 252219974Smavg_raid_volume_event2str(int event) 253219974Smav{ 254219974Smav 255219974Smav switch (event) { 256219974Smav case G_RAID_VOLUME_E_UP: 257219974Smav return ("UP"); 258219974Smav case G_RAID_VOLUME_E_DOWN: 259219974Smav return ("DOWN"); 260219974Smav case G_RAID_VOLUME_E_START: 261219974Smav return ("START"); 262219974Smav case G_RAID_VOLUME_E_STARTMD: 263219974Smav return ("STARTMD"); 264219974Smav default: 265219974Smav return ("INVALID"); 266219974Smav } 267219974Smav} 268219974Smav 269219974Smavconst char * 270219974Smavg_raid_volume_level2str(int level, int qual) 271219974Smav{ 272219974Smav 273219974Smav switch (level) { 274219974Smav case G_RAID_VOLUME_RL_RAID0: 275219974Smav return ("RAID0"); 276219974Smav case G_RAID_VOLUME_RL_RAID1: 277219974Smav return ("RAID1"); 278219974Smav case G_RAID_VOLUME_RL_RAID3: 279234603Smav if (qual == G_RAID_VOLUME_RLQ_R3P0) 280234603Smav return ("RAID3-P0"); 281234603Smav if (qual == G_RAID_VOLUME_RLQ_R3PN) 282234603Smav return ("RAID3-PN"); 283219974Smav return ("RAID3"); 284219974Smav case G_RAID_VOLUME_RL_RAID4: 285234603Smav if (qual == G_RAID_VOLUME_RLQ_R4P0) 286234610Smav return ("RAID4-P0"); 287234603Smav if (qual == G_RAID_VOLUME_RLQ_R4PN) 288234610Smav return ("RAID4-PN"); 289219974Smav return ("RAID4"); 290219974Smav case G_RAID_VOLUME_RL_RAID5: 291234458Smav if (qual == G_RAID_VOLUME_RLQ_R5RA) 292234603Smav return ("RAID5-RA"); 293234458Smav if (qual == G_RAID_VOLUME_RLQ_R5RS) 294234603Smav return ("RAID5-RS"); 295234458Smav if (qual == G_RAID_VOLUME_RLQ_R5LA) 296234603Smav return ("RAID5-LA"); 297234458Smav if (qual == G_RAID_VOLUME_RLQ_R5LS) 298234603Smav return ("RAID5-LS"); 299219974Smav return ("RAID5"); 300219974Smav case G_RAID_VOLUME_RL_RAID6: 301234603Smav if (qual == G_RAID_VOLUME_RLQ_R6RA) 302234603Smav return ("RAID6-RA"); 303234603Smav if (qual == G_RAID_VOLUME_RLQ_R6RS) 304234603Smav return ("RAID6-RS"); 305234603Smav if (qual == G_RAID_VOLUME_RLQ_R6LA) 306234603Smav return ("RAID6-LA"); 307234603Smav if (qual == G_RAID_VOLUME_RLQ_R6LS) 308234603Smav return ("RAID6-LS"); 309219974Smav return ("RAID6"); 310234603Smav case G_RAID_VOLUME_RL_RAIDMDF: 311234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFRA) 312234603Smav return ("RAIDMDF-RA"); 313234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFRS) 314234603Smav return ("RAIDMDF-RS"); 315234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFLA) 316234603Smav return ("RAIDMDF-LA"); 317234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFLS) 318234603Smav return ("RAIDMDF-LS"); 319234603Smav return ("RAIDMDF"); 320219974Smav case G_RAID_VOLUME_RL_RAID1E: 321234603Smav if (qual == G_RAID_VOLUME_RLQ_R1EA) 322234603Smav return ("RAID1E-A"); 323234603Smav if (qual == G_RAID_VOLUME_RLQ_R1EO) 324234603Smav return ("RAID1E-O"); 325219974Smav return ("RAID1E"); 326219974Smav case G_RAID_VOLUME_RL_SINGLE: 327219974Smav return ("SINGLE"); 328219974Smav case G_RAID_VOLUME_RL_CONCAT: 329219974Smav return ("CONCAT"); 330219974Smav case G_RAID_VOLUME_RL_RAID5E: 331234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ERA) 332234603Smav return ("RAID5E-RA"); 333234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ERS) 334234603Smav return ("RAID5E-RS"); 335234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ELA) 336234603Smav return ("RAID5E-LA"); 337234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ELS) 338234603Smav return ("RAID5E-LS"); 339219974Smav return ("RAID5E"); 340219974Smav case G_RAID_VOLUME_RL_RAID5EE: 341234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EERA) 342234603Smav return ("RAID5EE-RA"); 343234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EERS) 344234603Smav return ("RAID5EE-RS"); 345234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EELA) 346234603Smav return ("RAID5EE-LA"); 347234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EELS) 348234603Smav return ("RAID5EE-LS"); 349219974Smav return ("RAID5EE"); 350234603Smav case G_RAID_VOLUME_RL_RAID5R: 351234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RRA) 352234603Smav return ("RAID5R-RA"); 353234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RRS) 354234603Smav return ("RAID5R-RS"); 355234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RLA) 356234603Smav return ("RAID5R-LA"); 357234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RLS) 358234603Smav return ("RAID5R-LS"); 359234603Smav return ("RAID5E"); 360219974Smav default: 361219974Smav return ("UNKNOWN"); 362219974Smav } 363219974Smav} 364219974Smav 365219974Smavint 366219974Smavg_raid_volume_str2level(const char *str, int *level, int *qual) 367219974Smav{ 368219974Smav 369219974Smav *level = G_RAID_VOLUME_RL_UNKNOWN; 370219974Smav *qual = G_RAID_VOLUME_RLQ_NONE; 371219974Smav if (strcasecmp(str, "RAID0") == 0) 372219974Smav *level = G_RAID_VOLUME_RL_RAID0; 373219974Smav else if (strcasecmp(str, "RAID1") == 0) 374219974Smav *level = G_RAID_VOLUME_RL_RAID1; 375234603Smav else if (strcasecmp(str, "RAID3-P0") == 0) { 376219974Smav *level = G_RAID_VOLUME_RL_RAID3; 377234603Smav *qual = G_RAID_VOLUME_RLQ_R3P0; 378234993Smav } else if (strcasecmp(str, "RAID3-PN") == 0 || 379234603Smav strcasecmp(str, "RAID3") == 0) { 380234603Smav *level = G_RAID_VOLUME_RL_RAID3; 381234993Smav *qual = G_RAID_VOLUME_RLQ_R3PN; 382234603Smav } else if (strcasecmp(str, "RAID4-P0") == 0) { 383219974Smav *level = G_RAID_VOLUME_RL_RAID4; 384234603Smav *qual = G_RAID_VOLUME_RLQ_R4P0; 385234993Smav } else if (strcasecmp(str, "RAID4-PN") == 0 || 386234603Smav strcasecmp(str, "RAID4") == 0) { 387234603Smav *level = G_RAID_VOLUME_RL_RAID4; 388234993Smav *qual = G_RAID_VOLUME_RLQ_R4PN; 389234603Smav } else if (strcasecmp(str, "RAID5-RA") == 0) { 390219974Smav *level = G_RAID_VOLUME_RL_RAID5; 391234458Smav *qual = G_RAID_VOLUME_RLQ_R5RA; 392234603Smav } else if (strcasecmp(str, "RAID5-RS") == 0) { 393234458Smav *level = G_RAID_VOLUME_RL_RAID5; 394234458Smav *qual = G_RAID_VOLUME_RLQ_R5RS; 395234458Smav } else if (strcasecmp(str, "RAID5") == 0 || 396234603Smav strcasecmp(str, "RAID5-LA") == 0) { 397234458Smav *level = G_RAID_VOLUME_RL_RAID5; 398234458Smav *qual = G_RAID_VOLUME_RLQ_R5LA; 399234603Smav } else if (strcasecmp(str, "RAID5-LS") == 0) { 400234458Smav *level = G_RAID_VOLUME_RL_RAID5; 401234458Smav *qual = G_RAID_VOLUME_RLQ_R5LS; 402234603Smav } else if (strcasecmp(str, "RAID6-RA") == 0) { 403219974Smav *level = G_RAID_VOLUME_RL_RAID6; 404234603Smav *qual = G_RAID_VOLUME_RLQ_R6RA; 405234603Smav } else if (strcasecmp(str, "RAID6-RS") == 0) { 406234603Smav *level = G_RAID_VOLUME_RL_RAID6; 407234603Smav *qual = G_RAID_VOLUME_RLQ_R6RS; 408234603Smav } else if (strcasecmp(str, "RAID6") == 0 || 409234603Smav strcasecmp(str, "RAID6-LA") == 0) { 410234603Smav *level = G_RAID_VOLUME_RL_RAID6; 411234603Smav *qual = G_RAID_VOLUME_RLQ_R6LA; 412234603Smav } else if (strcasecmp(str, "RAID6-LS") == 0) { 413234603Smav *level = G_RAID_VOLUME_RL_RAID6; 414234603Smav *qual = G_RAID_VOLUME_RLQ_R6LS; 415234603Smav } else if (strcasecmp(str, "RAIDMDF-RA") == 0) { 416234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 417234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFRA; 418234603Smav } else if (strcasecmp(str, "RAIDMDF-RS") == 0) { 419234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 420234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFRS; 421234603Smav } else if (strcasecmp(str, "RAIDMDF") == 0 || 422234603Smav strcasecmp(str, "RAIDMDF-LA") == 0) { 423234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 424234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFLA; 425234603Smav } else if (strcasecmp(str, "RAIDMDF-LS") == 0) { 426234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 427234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFLS; 428234603Smav } else if (strcasecmp(str, "RAID10") == 0 || 429234603Smav strcasecmp(str, "RAID1E") == 0 || 430234603Smav strcasecmp(str, "RAID1E-A") == 0) { 431219974Smav *level = G_RAID_VOLUME_RL_RAID1E; 432234603Smav *qual = G_RAID_VOLUME_RLQ_R1EA; 433234603Smav } else if (strcasecmp(str, "RAID1E-O") == 0) { 434234603Smav *level = G_RAID_VOLUME_RL_RAID1E; 435234603Smav *qual = G_RAID_VOLUME_RLQ_R1EO; 436234603Smav } else if (strcasecmp(str, "SINGLE") == 0) 437219974Smav *level = G_RAID_VOLUME_RL_SINGLE; 438219974Smav else if (strcasecmp(str, "CONCAT") == 0) 439219974Smav *level = G_RAID_VOLUME_RL_CONCAT; 440234603Smav else if (strcasecmp(str, "RAID5E-RA") == 0) { 441219974Smav *level = G_RAID_VOLUME_RL_RAID5E; 442234603Smav *qual = G_RAID_VOLUME_RLQ_R5ERA; 443234603Smav } else if (strcasecmp(str, "RAID5E-RS") == 0) { 444234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 445234603Smav *qual = G_RAID_VOLUME_RLQ_R5ERS; 446234603Smav } else if (strcasecmp(str, "RAID5E") == 0 || 447234603Smav strcasecmp(str, "RAID5E-LA") == 0) { 448234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 449234603Smav *qual = G_RAID_VOLUME_RLQ_R5ELA; 450234603Smav } else if (strcasecmp(str, "RAID5E-LS") == 0) { 451234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 452234603Smav *qual = G_RAID_VOLUME_RLQ_R5ELS; 453234603Smav } else if (strcasecmp(str, "RAID5EE-RA") == 0) { 454219974Smav *level = G_RAID_VOLUME_RL_RAID5EE; 455234603Smav *qual = G_RAID_VOLUME_RLQ_R5EERA; 456234603Smav } else if (strcasecmp(str, "RAID5EE-RS") == 0) { 457234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 458234603Smav *qual = G_RAID_VOLUME_RLQ_R5EERS; 459234603Smav } else if (strcasecmp(str, "RAID5EE") == 0 || 460234603Smav strcasecmp(str, "RAID5EE-LA") == 0) { 461234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 462234603Smav *qual = G_RAID_VOLUME_RLQ_R5EELA; 463234603Smav } else if (strcasecmp(str, "RAID5EE-LS") == 0) { 464234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 465234603Smav *qual = G_RAID_VOLUME_RLQ_R5EELS; 466234603Smav } else if (strcasecmp(str, "RAID5R-RA") == 0) { 467234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 468234603Smav *qual = G_RAID_VOLUME_RLQ_R5RRA; 469234603Smav } else if (strcasecmp(str, "RAID5R-RS") == 0) { 470234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 471234603Smav *qual = G_RAID_VOLUME_RLQ_R5RRS; 472234603Smav } else if (strcasecmp(str, "RAID5R") == 0 || 473234603Smav strcasecmp(str, "RAID5R-LA") == 0) { 474234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 475234603Smav *qual = G_RAID_VOLUME_RLQ_R5RLA; 476234603Smav } else if (strcasecmp(str, "RAID5R-LS") == 0) { 477234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 478234603Smav *qual = G_RAID_VOLUME_RLQ_R5RLS; 479234603Smav } else 480219974Smav return (-1); 481219974Smav return (0); 482219974Smav} 483219974Smav 484219974Smavconst char * 485219974Smavg_raid_get_diskname(struct g_raid_disk *disk) 486219974Smav{ 487219974Smav 488219974Smav if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 489219974Smav return ("[unknown]"); 490219974Smav return (disk->d_consumer->provider->name); 491219974Smav} 492219974Smav 493219974Smavvoid 494242323Smavg_raid_get_disk_info(struct g_raid_disk *disk) 495242323Smav{ 496242323Smav struct g_consumer *cp = disk->d_consumer; 497242323Smav int error, len; 498242323Smav 499242323Smav /* Read kernel dumping information. */ 500242323Smav disk->d_kd.offset = 0; 501242323Smav disk->d_kd.length = OFF_MAX; 502242323Smav len = sizeof(disk->d_kd); 503242323Smav error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); 504242323Smav if (error) 505242323Smav disk->d_kd.di.dumper = NULL; 506242323Smav if (disk->d_kd.di.dumper == NULL) 507242323Smav G_RAID_DEBUG1(2, disk->d_softc, 508242323Smav "Dumping not supported by %s: %d.", 509242323Smav cp->provider->name, error); 510242323Smav 511242323Smav /* Read BIO_DELETE support. */ 512242323Smav error = g_getattr("GEOM::candelete", cp, &disk->d_candelete); 513242323Smav if (error) 514242323Smav disk->d_candelete = 0; 515242323Smav if (!disk->d_candelete) 516242323Smav G_RAID_DEBUG1(2, disk->d_softc, 517242323Smav "BIO_DELETE not supported by %s: %d.", 518242323Smav cp->provider->name, error); 519242323Smav} 520242323Smav 521242323Smavvoid 522219974Smavg_raid_report_disk_state(struct g_raid_disk *disk) 523219974Smav{ 524219974Smav struct g_raid_subdisk *sd; 525219974Smav int len, state; 526219974Smav uint32_t s; 527219974Smav 528219974Smav if (disk->d_consumer == NULL) 529219974Smav return; 530245326Smav if (disk->d_state == G_RAID_DISK_S_DISABLED) { 531245363Smav s = G_STATE_ACTIVE; /* XXX */ 532245326Smav } else if (disk->d_state == G_RAID_DISK_S_FAILED || 533219974Smav disk->d_state == G_RAID_DISK_S_STALE_FAILED) { 534219974Smav s = G_STATE_FAILED; 535219974Smav } else { 536219974Smav state = G_RAID_SUBDISK_S_ACTIVE; 537219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 538219974Smav if (sd->sd_state < state) 539219974Smav state = sd->sd_state; 540219974Smav } 541219974Smav if (state == G_RAID_SUBDISK_S_FAILED) 542219974Smav s = G_STATE_FAILED; 543219974Smav else if (state == G_RAID_SUBDISK_S_NEW || 544219974Smav state == G_RAID_SUBDISK_S_REBUILD) 545219974Smav s = G_STATE_REBUILD; 546219974Smav else if (state == G_RAID_SUBDISK_S_STALE || 547219974Smav state == G_RAID_SUBDISK_S_RESYNC) 548219974Smav s = G_STATE_RESYNC; 549219974Smav else 550219974Smav s = G_STATE_ACTIVE; 551219974Smav } 552219974Smav len = sizeof(s); 553219974Smav g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s); 554219974Smav G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.", 555219974Smav g_raid_get_diskname(disk), s); 556219974Smav} 557219974Smav 558219974Smavvoid 559219974Smavg_raid_change_disk_state(struct g_raid_disk *disk, int state) 560219974Smav{ 561219974Smav 562219974Smav G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.", 563219974Smav g_raid_get_diskname(disk), 564219974Smav g_raid_disk_state2str(disk->d_state), 565219974Smav g_raid_disk_state2str(state)); 566219974Smav disk->d_state = state; 567219974Smav g_raid_report_disk_state(disk); 568219974Smav} 569219974Smav 570219974Smavvoid 571219974Smavg_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state) 572219974Smav{ 573219974Smav 574219974Smav G_RAID_DEBUG1(0, sd->sd_softc, 575219974Smav "Subdisk %s:%d-%s state changed from %s to %s.", 576219974Smav sd->sd_volume->v_name, sd->sd_pos, 577219974Smav sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", 578219974Smav g_raid_subdisk_state2str(sd->sd_state), 579219974Smav g_raid_subdisk_state2str(state)); 580219974Smav sd->sd_state = state; 581219974Smav if (sd->sd_disk) 582219974Smav g_raid_report_disk_state(sd->sd_disk); 583219974Smav} 584219974Smav 585219974Smavvoid 586219974Smavg_raid_change_volume_state(struct g_raid_volume *vol, int state) 587219974Smav{ 588219974Smav 589219974Smav G_RAID_DEBUG1(0, vol->v_softc, 590219974Smav "Volume %s state changed from %s to %s.", 591219974Smav vol->v_name, 592219974Smav g_raid_volume_state2str(vol->v_state), 593219974Smav g_raid_volume_state2str(state)); 594219974Smav vol->v_state = state; 595219974Smav} 596219974Smav 597219974Smav/* 598219974Smav * --- Events handling functions --- 599219974Smav * Events in geom_raid are used to maintain subdisks and volumes status 600219974Smav * from one thread to simplify locking. 601219974Smav */ 602219974Smavstatic void 603219974Smavg_raid_event_free(struct g_raid_event *ep) 604219974Smav{ 605219974Smav 606219974Smav free(ep, M_RAID); 607219974Smav} 608219974Smav 609219974Smavint 610219974Smavg_raid_event_send(void *arg, int event, int flags) 611219974Smav{ 612219974Smav struct g_raid_softc *sc; 613219974Smav struct g_raid_event *ep; 614219974Smav int error; 615219974Smav 616219974Smav if ((flags & G_RAID_EVENT_VOLUME) != 0) { 617219974Smav sc = ((struct g_raid_volume *)arg)->v_softc; 618219974Smav } else if ((flags & G_RAID_EVENT_DISK) != 0) { 619219974Smav sc = ((struct g_raid_disk *)arg)->d_softc; 620219974Smav } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) { 621219974Smav sc = ((struct g_raid_subdisk *)arg)->sd_softc; 622219974Smav } else { 623219974Smav sc = arg; 624219974Smav } 625219974Smav ep = malloc(sizeof(*ep), M_RAID, 626219974Smav sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT); 627219974Smav if (ep == NULL) 628219974Smav return (ENOMEM); 629219974Smav ep->e_tgt = arg; 630219974Smav ep->e_event = event; 631219974Smav ep->e_flags = flags; 632219974Smav ep->e_error = 0; 633219974Smav G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc); 634219974Smav mtx_lock(&sc->sc_queue_mtx); 635219974Smav TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 636219974Smav mtx_unlock(&sc->sc_queue_mtx); 637219974Smav wakeup(sc); 638219974Smav 639219974Smav if ((flags & G_RAID_EVENT_WAIT) == 0) 640219974Smav return (0); 641219974Smav 642219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 643219974Smav G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep); 644219974Smav sx_xunlock(&sc->sc_lock); 645219974Smav while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) { 646219974Smav mtx_lock(&sc->sc_queue_mtx); 647219974Smav MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event", 648219974Smav hz * 5); 649219974Smav } 650219974Smav error = ep->e_error; 651219974Smav g_raid_event_free(ep); 652219974Smav sx_xlock(&sc->sc_lock); 653219974Smav return (error); 654219974Smav} 655219974Smav 656219974Smavstatic void 657219974Smavg_raid_event_cancel(struct g_raid_softc *sc, void *tgt) 658219974Smav{ 659219974Smav struct g_raid_event *ep, *tmpep; 660219974Smav 661219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 662219974Smav 663219974Smav mtx_lock(&sc->sc_queue_mtx); 664219974Smav TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 665219974Smav if (ep->e_tgt != tgt) 666219974Smav continue; 667219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 668219974Smav if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) 669219974Smav g_raid_event_free(ep); 670219974Smav else { 671219974Smav ep->e_error = ECANCELED; 672219974Smav wakeup(ep); 673219974Smav } 674219974Smav } 675219974Smav mtx_unlock(&sc->sc_queue_mtx); 676219974Smav} 677219974Smav 678219974Smavstatic int 679219974Smavg_raid_event_check(struct g_raid_softc *sc, void *tgt) 680219974Smav{ 681219974Smav struct g_raid_event *ep; 682219974Smav int res = 0; 683219974Smav 684219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 685219974Smav 686219974Smav mtx_lock(&sc->sc_queue_mtx); 687219974Smav TAILQ_FOREACH(ep, &sc->sc_events, e_next) { 688219974Smav if (ep->e_tgt != tgt) 689219974Smav continue; 690219974Smav res = 1; 691219974Smav break; 692219974Smav } 693219974Smav mtx_unlock(&sc->sc_queue_mtx); 694219974Smav return (res); 695219974Smav} 696219974Smav 697219974Smav/* 698219974Smav * Return the number of disks in given state. 699219974Smav * If state is equal to -1, count all connected disks. 700219974Smav */ 701219974Smavu_int 702219974Smavg_raid_ndisks(struct g_raid_softc *sc, int state) 703219974Smav{ 704219974Smav struct g_raid_disk *disk; 705219974Smav u_int n; 706219974Smav 707219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 708219974Smav 709219974Smav n = 0; 710219974Smav TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 711219974Smav if (disk->d_state == state || state == -1) 712219974Smav n++; 713219974Smav } 714219974Smav return (n); 715219974Smav} 716219974Smav 717219974Smav/* 718219974Smav * Return the number of subdisks in given state. 719219974Smav * If state is equal to -1, count all connected disks. 720219974Smav */ 721219974Smavu_int 722219974Smavg_raid_nsubdisks(struct g_raid_volume *vol, int state) 723219974Smav{ 724219974Smav struct g_raid_subdisk *subdisk; 725219974Smav struct g_raid_softc *sc; 726219974Smav u_int i, n ; 727219974Smav 728219974Smav sc = vol->v_softc; 729219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 730219974Smav 731219974Smav n = 0; 732219974Smav for (i = 0; i < vol->v_disks_count; i++) { 733219974Smav subdisk = &vol->v_subdisks[i]; 734219974Smav if ((state == -1 && 735219974Smav subdisk->sd_state != G_RAID_SUBDISK_S_NONE) || 736219974Smav subdisk->sd_state == state) 737219974Smav n++; 738219974Smav } 739219974Smav return (n); 740219974Smav} 741219974Smav 742219974Smav/* 743219974Smav * Return the first subdisk in given state. 744219974Smav * If state is equal to -1, then the first connected disks. 745219974Smav */ 746219974Smavstruct g_raid_subdisk * 747219974Smavg_raid_get_subdisk(struct g_raid_volume *vol, int state) 748219974Smav{ 749219974Smav struct g_raid_subdisk *sd; 750219974Smav struct g_raid_softc *sc; 751219974Smav u_int i; 752219974Smav 753219974Smav sc = vol->v_softc; 754219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 755219974Smav 756219974Smav for (i = 0; i < vol->v_disks_count; i++) { 757219974Smav sd = &vol->v_subdisks[i]; 758219974Smav if ((state == -1 && 759219974Smav sd->sd_state != G_RAID_SUBDISK_S_NONE) || 760219974Smav sd->sd_state == state) 761219974Smav return (sd); 762219974Smav } 763219974Smav return (NULL); 764219974Smav} 765219974Smav 766219974Smavstruct g_consumer * 767219974Smavg_raid_open_consumer(struct g_raid_softc *sc, const char *name) 768219974Smav{ 769219974Smav struct g_consumer *cp; 770219974Smav struct g_provider *pp; 771219974Smav 772219974Smav g_topology_assert(); 773219974Smav 774219974Smav if (strncmp(name, "/dev/", 5) == 0) 775219974Smav name += 5; 776219974Smav pp = g_provider_by_name(name); 777219974Smav if (pp == NULL) 778219974Smav return (NULL); 779219974Smav cp = g_new_consumer(sc->sc_geom); 780256880Smav cp->flags |= G_CF_DIRECT_RECEIVE; 781219974Smav if (g_attach(cp, pp) != 0) { 782219974Smav g_destroy_consumer(cp); 783219974Smav return (NULL); 784219974Smav } 785219974Smav if (g_access(cp, 1, 1, 1) != 0) { 786219974Smav g_detach(cp); 787219974Smav g_destroy_consumer(cp); 788219974Smav return (NULL); 789219974Smav } 790219974Smav return (cp); 791219974Smav} 792219974Smav 793219974Smavstatic u_int 794219974Smavg_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) 795219974Smav{ 796219974Smav struct bio *bp; 797219974Smav u_int nreqs = 0; 798219974Smav 799219974Smav mtx_lock(&sc->sc_queue_mtx); 800219974Smav TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 801219974Smav if (bp->bio_from == cp) 802219974Smav nreqs++; 803219974Smav } 804219974Smav mtx_unlock(&sc->sc_queue_mtx); 805219974Smav return (nreqs); 806219974Smav} 807219974Smav 808219974Smavu_int 809219974Smavg_raid_nopens(struct g_raid_softc *sc) 810219974Smav{ 811219974Smav struct g_raid_volume *vol; 812219974Smav u_int opens; 813219974Smav 814219974Smav opens = 0; 815219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 816219974Smav if (vol->v_provider_open != 0) 817219974Smav opens++; 818219974Smav } 819219974Smav return (opens); 820219974Smav} 821219974Smav 822219974Smavstatic int 823219974Smavg_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) 824219974Smav{ 825219974Smav 826219974Smav if (cp->index > 0) { 827219974Smav G_RAID_DEBUG1(2, sc, 828219974Smav "I/O requests for %s exist, can't destroy it now.", 829219974Smav cp->provider->name); 830219974Smav return (1); 831219974Smav } 832219974Smav if (g_raid_nrequests(sc, cp) > 0) { 833219974Smav G_RAID_DEBUG1(2, sc, 834219974Smav "I/O requests for %s in queue, can't destroy it now.", 835219974Smav cp->provider->name); 836219974Smav return (1); 837219974Smav } 838219974Smav return (0); 839219974Smav} 840219974Smav 841219974Smavstatic void 842219974Smavg_raid_destroy_consumer(void *arg, int flags __unused) 843219974Smav{ 844219974Smav struct g_consumer *cp; 845219974Smav 846219974Smav g_topology_assert(); 847219974Smav 848219974Smav cp = arg; 849219974Smav G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 850219974Smav g_detach(cp); 851219974Smav g_destroy_consumer(cp); 852219974Smav} 853219974Smav 854219974Smavvoid 855219974Smavg_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) 856219974Smav{ 857219974Smav struct g_provider *pp; 858219974Smav int retaste_wait; 859219974Smav 860219974Smav g_topology_assert_not(); 861219974Smav 862219974Smav g_topology_lock(); 863219974Smav cp->private = NULL; 864219974Smav if (g_raid_consumer_is_busy(sc, cp)) 865219974Smav goto out; 866219974Smav pp = cp->provider; 867219974Smav retaste_wait = 0; 868219974Smav if (cp->acw == 1) { 869219974Smav if ((pp->geom->flags & G_GEOM_WITHER) == 0) 870219974Smav retaste_wait = 1; 871219974Smav } 872219974Smav if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 873219974Smav g_access(cp, -cp->acr, -cp->acw, -cp->ace); 874219974Smav if (retaste_wait) { 875219974Smav /* 876219974Smav * After retaste event was send (inside g_access()), we can send 877219974Smav * event to detach and destroy consumer. 878219974Smav * A class, which has consumer to the given provider connected 879219974Smav * will not receive retaste event for the provider. 880219974Smav * This is the way how I ignore retaste events when I close 881219974Smav * consumers opened for write: I detach and destroy consumer 882219974Smav * after retaste event is sent. 883219974Smav */ 884219974Smav g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); 885219974Smav goto out; 886219974Smav } 887219974Smav G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); 888219974Smav g_detach(cp); 889219974Smav g_destroy_consumer(cp); 890219974Smavout: 891219974Smav g_topology_unlock(); 892219974Smav} 893219974Smav 894219974Smavstatic void 895219974Smavg_raid_orphan(struct g_consumer *cp) 896219974Smav{ 897219974Smav struct g_raid_disk *disk; 898219974Smav 899219974Smav g_topology_assert(); 900219974Smav 901219974Smav disk = cp->private; 902219974Smav if (disk == NULL) 903219974Smav return; 904219974Smav g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, 905219974Smav G_RAID_EVENT_DISK); 906219974Smav} 907219974Smav 908242314Smavstatic void 909219974Smavg_raid_clean(struct g_raid_volume *vol, int acw) 910219974Smav{ 911219974Smav struct g_raid_softc *sc; 912219974Smav int timeout; 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) 919242314Smav// return; 920219974Smav if (!vol->v_dirty) 921242314Smav return; 922219974Smav if (vol->v_writes > 0) 923242314Smav return; 924219974Smav if (acw > 0 || (acw == -1 && 925219974Smav vol->v_provider != NULL && vol->v_provider->acw > 0)) { 926219974Smav timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); 927242314Smav if (!g_raid_shutdown && timeout > 0) 928242314Smav return; 929219974Smav } 930219974Smav vol->v_dirty = 0; 931219974Smav G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", 932219974Smav vol->v_name); 933219974Smav g_raid_write_metadata(sc, vol, NULL, NULL); 934219974Smav} 935219974Smav 936219974Smavstatic void 937219974Smavg_raid_dirty(struct g_raid_volume *vol) 938219974Smav{ 939219974Smav struct g_raid_softc *sc; 940219974Smav 941219974Smav sc = vol->v_softc; 942219974Smav g_topology_assert_not(); 943219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 944219974Smav 945219974Smav// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 946219974Smav// return; 947219974Smav vol->v_dirty = 1; 948219974Smav G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", 949219974Smav vol->v_name); 950219974Smav g_raid_write_metadata(sc, vol, NULL, NULL); 951219974Smav} 952219974Smav 953219974Smavvoid 954219974Smavg_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) 955219974Smav{ 956219974Smav struct g_raid_softc *sc; 957219974Smav struct g_raid_volume *vol; 958219974Smav struct g_raid_subdisk *sd; 959219974Smav struct bio_queue_head queue; 960219974Smav struct bio *cbp; 961219974Smav int i; 962219974Smav 963219974Smav vol = tr->tro_volume; 964219974Smav sc = vol->v_softc; 965219974Smav 966219974Smav /* 967219974Smav * Allocate all bios before sending any request, so we can return 968219974Smav * ENOMEM in nice and clean way. 969219974Smav */ 970219974Smav bioq_init(&queue); 971219974Smav for (i = 0; i < vol->v_disks_count; i++) { 972219974Smav sd = &vol->v_subdisks[i]; 973219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_NONE || 974219974Smav sd->sd_state == G_RAID_SUBDISK_S_FAILED) 975219974Smav continue; 976219974Smav cbp = g_clone_bio(bp); 977219974Smav if (cbp == NULL) 978219974Smav goto failure; 979219974Smav cbp->bio_caller1 = sd; 980219974Smav bioq_insert_tail(&queue, cbp); 981219974Smav } 982256610Smav while ((cbp = bioq_takefirst(&queue)) != NULL) { 983219974Smav sd = cbp->bio_caller1; 984219974Smav cbp->bio_caller1 = NULL; 985219974Smav g_raid_subdisk_iostart(sd, cbp); 986219974Smav } 987219974Smav return; 988219974Smavfailure: 989256610Smav while ((cbp = bioq_takefirst(&queue)) != NULL) 990219974Smav g_destroy_bio(cbp); 991219974Smav if (bp->bio_error == 0) 992219974Smav bp->bio_error = ENOMEM; 993219974Smav g_raid_iodone(bp, bp->bio_error); 994219974Smav} 995219974Smav 996219974Smavstatic void 997219974Smavg_raid_tr_kerneldump_common_done(struct bio *bp) 998219974Smav{ 999219974Smav 1000219974Smav bp->bio_flags |= BIO_DONE; 1001219974Smav} 1002219974Smav 1003219974Smavint 1004219974Smavg_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, 1005219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1006219974Smav{ 1007219974Smav struct g_raid_softc *sc; 1008219974Smav struct g_raid_volume *vol; 1009219974Smav struct bio bp; 1010219974Smav 1011219974Smav vol = tr->tro_volume; 1012219974Smav sc = vol->v_softc; 1013219974Smav 1014295707Simp g_reset_bio(&bp); 1015219974Smav bp.bio_cmd = BIO_WRITE; 1016219974Smav bp.bio_done = g_raid_tr_kerneldump_common_done; 1017219974Smav bp.bio_attribute = NULL; 1018219974Smav bp.bio_offset = offset; 1019219974Smav bp.bio_length = length; 1020219974Smav bp.bio_data = virtual; 1021219974Smav bp.bio_to = vol->v_provider; 1022219974Smav 1023219974Smav g_raid_start(&bp); 1024219974Smav while (!(bp.bio_flags & BIO_DONE)) { 1025219974Smav G_RAID_DEBUG1(4, sc, "Poll..."); 1026219974Smav g_raid_poll(sc); 1027219974Smav DELAY(10); 1028219974Smav } 1029219974Smav 1030219974Smav return (bp.bio_error != 0 ? EIO : 0); 1031219974Smav} 1032219974Smav 1033219974Smavstatic int 1034219974Smavg_raid_dump(void *arg, 1035219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1036219974Smav{ 1037219974Smav struct g_raid_volume *vol; 1038219974Smav int error; 1039219974Smav 1040219974Smav vol = (struct g_raid_volume *)arg; 1041219974Smav G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", 1042219974Smav (long long unsigned)offset, (long long unsigned)length); 1043219974Smav 1044219974Smav error = G_RAID_TR_KERNELDUMP(vol->v_tr, 1045219974Smav virtual, physical, offset, length); 1046219974Smav return (error); 1047219974Smav} 1048219974Smav 1049219974Smavstatic void 1050219974Smavg_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) 1051219974Smav{ 1052219974Smav struct g_kerneldump *gkd; 1053219974Smav struct g_provider *pp; 1054219974Smav struct g_raid_volume *vol; 1055219974Smav 1056219974Smav gkd = (struct g_kerneldump*)bp->bio_data; 1057219974Smav pp = bp->bio_to; 1058219974Smav vol = pp->private; 1059219974Smav g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", 1060219974Smav pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 1061219974Smav gkd->di.dumper = g_raid_dump; 1062219974Smav gkd->di.priv = vol; 1063219974Smav gkd->di.blocksize = vol->v_sectorsize; 1064219974Smav gkd->di.maxiosize = DFLTPHYS; 1065219974Smav gkd->di.mediaoffset = gkd->offset; 1066219974Smav if ((gkd->offset + gkd->length) > vol->v_mediasize) 1067219974Smav gkd->length = vol->v_mediasize - gkd->offset; 1068219974Smav gkd->di.mediasize = gkd->length; 1069219974Smav g_io_deliver(bp, 0); 1070219974Smav} 1071219974Smav 1072219974Smavstatic void 1073242323Smavg_raid_candelete(struct g_raid_softc *sc, struct bio *bp) 1074242323Smav{ 1075242323Smav struct g_provider *pp; 1076242323Smav struct g_raid_volume *vol; 1077242323Smav struct g_raid_subdisk *sd; 1078242323Smav int *val; 1079242323Smav int i; 1080242323Smav 1081242323Smav val = (int *)bp->bio_data; 1082242323Smav pp = bp->bio_to; 1083242323Smav vol = pp->private; 1084242323Smav *val = 0; 1085242323Smav for (i = 0; i < vol->v_disks_count; i++) { 1086242323Smav sd = &vol->v_subdisks[i]; 1087242323Smav if (sd->sd_state == G_RAID_SUBDISK_S_NONE) 1088242323Smav continue; 1089242323Smav if (sd->sd_disk->d_candelete) { 1090242323Smav *val = 1; 1091242323Smav break; 1092242323Smav } 1093242323Smav } 1094242323Smav g_io_deliver(bp, 0); 1095242323Smav} 1096242323Smav 1097242323Smavstatic void 1098219974Smavg_raid_start(struct bio *bp) 1099219974Smav{ 1100219974Smav struct g_raid_softc *sc; 1101219974Smav 1102219974Smav sc = bp->bio_to->geom->softc; 1103219974Smav /* 1104219974Smav * If sc == NULL or there are no valid disks, provider's error 1105219974Smav * should be set and g_raid_start() should not be called at all. 1106219974Smav */ 1107219974Smav// KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, 1108219974Smav// ("Provider's error should be set (error=%d)(mirror=%s).", 1109219974Smav// bp->bio_to->error, bp->bio_to->name)); 1110219974Smav G_RAID_LOGREQ(3, bp, "Request received."); 1111219974Smav 1112219974Smav switch (bp->bio_cmd) { 1113219974Smav case BIO_READ: 1114219974Smav case BIO_WRITE: 1115219974Smav case BIO_DELETE: 1116219974Smav case BIO_FLUSH: 1117219974Smav break; 1118219974Smav case BIO_GETATTR: 1119242323Smav if (!strcmp(bp->bio_attribute, "GEOM::candelete")) 1120242323Smav g_raid_candelete(sc, bp); 1121242323Smav else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 1122219974Smav g_raid_kerneldump(sc, bp); 1123219974Smav else 1124219974Smav g_io_deliver(bp, EOPNOTSUPP); 1125219974Smav return; 1126219974Smav default: 1127219974Smav g_io_deliver(bp, EOPNOTSUPP); 1128219974Smav return; 1129219974Smav } 1130219974Smav mtx_lock(&sc->sc_queue_mtx); 1131280757Smav bioq_insert_tail(&sc->sc_queue, bp); 1132219974Smav mtx_unlock(&sc->sc_queue_mtx); 1133219974Smav if (!dumping) { 1134219974Smav G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); 1135219974Smav wakeup(sc); 1136219974Smav } 1137219974Smav} 1138219974Smav 1139219974Smavstatic int 1140219974Smavg_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) 1141219974Smav{ 1142219974Smav /* 1143219974Smav * 5 cases: 1144219974Smav * (1) bp entirely below NO 1145219974Smav * (2) bp entirely above NO 1146219974Smav * (3) bp start below, but end in range YES 1147219974Smav * (4) bp entirely within YES 1148219974Smav * (5) bp starts within, ends above YES 1149219974Smav * 1150219974Smav * lock range 10-19 (offset 10 length 10) 1151219974Smav * (1) 1-5: first if kicks it out 1152219974Smav * (2) 30-35: second if kicks it out 1153219974Smav * (3) 5-15: passes both ifs 1154219974Smav * (4) 12-14: passes both ifs 1155219974Smav * (5) 19-20: passes both 1156219974Smav */ 1157219974Smav off_t lend = lstart + len - 1; 1158219974Smav off_t bstart = bp->bio_offset; 1159219974Smav off_t bend = bp->bio_offset + bp->bio_length - 1; 1160219974Smav 1161219974Smav if (bend < lstart) 1162219974Smav return (0); 1163219974Smav if (lend < bstart) 1164219974Smav return (0); 1165219974Smav return (1); 1166219974Smav} 1167219974Smav 1168219974Smavstatic int 1169219974Smavg_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) 1170219974Smav{ 1171219974Smav struct g_raid_lock *lp; 1172219974Smav 1173219974Smav sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); 1174219974Smav 1175219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1176219974Smav if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) 1177219974Smav return (1); 1178219974Smav } 1179219974Smav return (0); 1180219974Smav} 1181219974Smav 1182219974Smavstatic void 1183219974Smavg_raid_start_request(struct bio *bp) 1184219974Smav{ 1185219974Smav struct g_raid_softc *sc; 1186219974Smav struct g_raid_volume *vol; 1187219974Smav 1188219974Smav sc = bp->bio_to->geom->softc; 1189219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1190219974Smav vol = bp->bio_to->private; 1191219974Smav 1192219974Smav /* 1193219974Smav * Check to see if this item is in a locked range. If so, 1194219974Smav * queue it to our locked queue and return. We'll requeue 1195219974Smav * it when the range is unlocked. Internal I/O for the 1196219974Smav * rebuild/rescan/recovery process is excluded from this 1197219974Smav * check so we can actually do the recovery. 1198219974Smav */ 1199219974Smav if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && 1200219974Smav g_raid_is_in_locked_range(vol, bp)) { 1201219974Smav G_RAID_LOGREQ(3, bp, "Defer request."); 1202219974Smav bioq_insert_tail(&vol->v_locked, bp); 1203219974Smav return; 1204219974Smav } 1205219974Smav 1206219974Smav /* 1207219974Smav * If we're actually going to do the write/delete, then 1208219974Smav * update the idle stats for the volume. 1209219974Smav */ 1210219974Smav if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1211219974Smav if (!vol->v_dirty) 1212219974Smav g_raid_dirty(vol); 1213219974Smav vol->v_writes++; 1214219974Smav } 1215219974Smav 1216219974Smav /* 1217219974Smav * Put request onto inflight queue, so we can check if new 1218219974Smav * synchronization requests don't collide with it. Then tell 1219219974Smav * the transformation layer to start the I/O. 1220219974Smav */ 1221219974Smav bioq_insert_tail(&vol->v_inflight, bp); 1222219974Smav G_RAID_LOGREQ(4, bp, "Request started"); 1223219974Smav G_RAID_TR_IOSTART(vol->v_tr, bp); 1224219974Smav} 1225219974Smav 1226219974Smavstatic void 1227219974Smavg_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) 1228219974Smav{ 1229219974Smav off_t off, len; 1230219974Smav struct bio *nbp; 1231219974Smav struct g_raid_lock *lp; 1232219974Smav 1233219974Smav vol->v_pending_lock = 0; 1234219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1235219974Smav if (lp->l_pending) { 1236219974Smav off = lp->l_offset; 1237219974Smav len = lp->l_length; 1238219974Smav lp->l_pending = 0; 1239219974Smav TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { 1240219974Smav if (g_raid_bio_overlaps(nbp, off, len)) 1241219974Smav lp->l_pending++; 1242219974Smav } 1243219974Smav if (lp->l_pending) { 1244219974Smav vol->v_pending_lock = 1; 1245219974Smav G_RAID_DEBUG1(4, vol->v_softc, 1246219974Smav "Deferred lock(%jd, %jd) has %d pending", 1247219974Smav (intmax_t)off, (intmax_t)(off + len), 1248219974Smav lp->l_pending); 1249219974Smav continue; 1250219974Smav } 1251219974Smav G_RAID_DEBUG1(4, vol->v_softc, 1252219974Smav "Deferred lock of %jd to %jd completed", 1253219974Smav (intmax_t)off, (intmax_t)(off + len)); 1254219974Smav G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1255219974Smav } 1256219974Smav } 1257219974Smav} 1258219974Smav 1259219974Smavvoid 1260219974Smavg_raid_iodone(struct bio *bp, int error) 1261219974Smav{ 1262219974Smav struct g_raid_softc *sc; 1263219974Smav struct g_raid_volume *vol; 1264219974Smav 1265219974Smav sc = bp->bio_to->geom->softc; 1266219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1267219974Smav vol = bp->bio_to->private; 1268219974Smav G_RAID_LOGREQ(3, bp, "Request done: %d.", error); 1269219974Smav 1270219974Smav /* Update stats if we done write/delete. */ 1271219974Smav if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1272219974Smav vol->v_writes--; 1273219974Smav vol->v_last_write = time_uptime; 1274219974Smav } 1275219974Smav 1276219974Smav bioq_remove(&vol->v_inflight, bp); 1277219974Smav if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) 1278219974Smav g_raid_finish_with_locked_ranges(vol, bp); 1279219974Smav getmicrouptime(&vol->v_last_done); 1280219974Smav g_io_deliver(bp, error); 1281219974Smav} 1282219974Smav 1283219974Smavint 1284219974Smavg_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, 1285219974Smav struct bio *ignore, void *argp) 1286219974Smav{ 1287219974Smav struct g_raid_softc *sc; 1288219974Smav struct g_raid_lock *lp; 1289219974Smav struct bio *bp; 1290219974Smav 1291219974Smav sc = vol->v_softc; 1292219974Smav lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); 1293219974Smav LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); 1294219974Smav lp->l_offset = off; 1295219974Smav lp->l_length = len; 1296219974Smav lp->l_callback_arg = argp; 1297219974Smav 1298219974Smav lp->l_pending = 0; 1299219974Smav TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { 1300219974Smav if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) 1301219974Smav lp->l_pending++; 1302219974Smav } 1303219974Smav 1304219974Smav /* 1305219974Smav * If there are any writes that are pending, we return EBUSY. All 1306219974Smav * callers will have to wait until all pending writes clear. 1307219974Smav */ 1308219974Smav if (lp->l_pending > 0) { 1309219974Smav vol->v_pending_lock = 1; 1310219974Smav G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", 1311219974Smav (intmax_t)off, (intmax_t)(off+len), lp->l_pending); 1312219974Smav return (EBUSY); 1313219974Smav } 1314219974Smav G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", 1315219974Smav (intmax_t)off, (intmax_t)(off+len)); 1316219974Smav G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1317219974Smav return (0); 1318219974Smav} 1319219974Smav 1320219974Smavint 1321219974Smavg_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) 1322219974Smav{ 1323219974Smav struct g_raid_lock *lp; 1324219974Smav struct g_raid_softc *sc; 1325219974Smav struct bio *bp; 1326219974Smav 1327219974Smav sc = vol->v_softc; 1328219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1329219974Smav if (lp->l_offset == off && lp->l_length == len) { 1330219974Smav LIST_REMOVE(lp, l_next); 1331219974Smav /* XXX 1332219974Smav * Right now we just put them all back on the queue 1333219974Smav * and hope for the best. We hope this because any 1334219974Smav * locked ranges will go right back on this list 1335219974Smav * when the worker thread runs. 1336219974Smav * XXX 1337219974Smav */ 1338219974Smav G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", 1339219974Smav (intmax_t)lp->l_offset, 1340219974Smav (intmax_t)(lp->l_offset+lp->l_length)); 1341219974Smav mtx_lock(&sc->sc_queue_mtx); 1342219974Smav while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) 1343280757Smav bioq_insert_tail(&sc->sc_queue, bp); 1344219974Smav mtx_unlock(&sc->sc_queue_mtx); 1345219974Smav free(lp, M_RAID); 1346219974Smav return (0); 1347219974Smav } 1348219974Smav } 1349219974Smav return (EINVAL); 1350219974Smav} 1351219974Smav 1352219974Smavvoid 1353219974Smavg_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) 1354219974Smav{ 1355219974Smav struct g_consumer *cp; 1356219974Smav struct g_raid_disk *disk, *tdisk; 1357219974Smav 1358219974Smav bp->bio_caller1 = sd; 1359219974Smav 1360219974Smav /* 1361219974Smav * Make sure that the disk is present. Generally it is a task of 1362219974Smav * transformation layers to not send requests to absent disks, but 1363219974Smav * it is better to be safe and report situation then sorry. 1364219974Smav */ 1365219974Smav if (sd->sd_disk == NULL) { 1366219974Smav G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); 1367219974Smavnodisk: 1368219974Smav bp->bio_from = NULL; 1369219974Smav bp->bio_to = NULL; 1370219974Smav bp->bio_error = ENXIO; 1371219974Smav g_raid_disk_done(bp); 1372219974Smav return; 1373219974Smav } 1374219974Smav disk = sd->sd_disk; 1375219974Smav if (disk->d_state != G_RAID_DISK_S_ACTIVE && 1376219974Smav disk->d_state != G_RAID_DISK_S_FAILED) { 1377219974Smav G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " 1378219974Smav "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 1379219974Smav goto nodisk; 1380219974Smav } 1381219974Smav 1382219974Smav cp = disk->d_consumer; 1383219974Smav bp->bio_from = cp; 1384219974Smav bp->bio_to = cp->provider; 1385219974Smav cp->index++; 1386219974Smav 1387219974Smav /* Update average disks load. */ 1388219974Smav TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { 1389219974Smav if (tdisk->d_consumer == NULL) 1390219974Smav tdisk->d_load = 0; 1391219974Smav else 1392219974Smav tdisk->d_load = (tdisk->d_consumer->index * 1393219974Smav G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; 1394219974Smav } 1395219974Smav 1396219974Smav disk->d_last_offset = bp->bio_offset + bp->bio_length; 1397219974Smav if (dumping) { 1398219974Smav G_RAID_LOGREQ(3, bp, "Sending dumping request."); 1399219974Smav if (bp->bio_cmd == BIO_WRITE) { 1400219974Smav bp->bio_error = g_raid_subdisk_kerneldump(sd, 1401219974Smav bp->bio_data, 0, bp->bio_offset, bp->bio_length); 1402219974Smav } else 1403219974Smav bp->bio_error = EOPNOTSUPP; 1404219974Smav g_raid_disk_done(bp); 1405219974Smav } else { 1406219974Smav bp->bio_done = g_raid_disk_done; 1407219974Smav bp->bio_offset += sd->sd_offset; 1408219974Smav G_RAID_LOGREQ(3, bp, "Sending request."); 1409219974Smav g_io_request(bp, cp); 1410219974Smav } 1411219974Smav} 1412219974Smav 1413219974Smavint 1414219974Smavg_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, 1415219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1416219974Smav{ 1417219974Smav 1418219974Smav if (sd->sd_disk == NULL) 1419219974Smav return (ENXIO); 1420219974Smav if (sd->sd_disk->d_kd.di.dumper == NULL) 1421219974Smav return (EOPNOTSUPP); 1422219974Smav return (dump_write(&sd->sd_disk->d_kd.di, 1423219974Smav virtual, physical, 1424219974Smav sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, 1425219974Smav length)); 1426219974Smav} 1427219974Smav 1428219974Smavstatic void 1429219974Smavg_raid_disk_done(struct bio *bp) 1430219974Smav{ 1431219974Smav struct g_raid_softc *sc; 1432219974Smav struct g_raid_subdisk *sd; 1433219974Smav 1434219974Smav sd = bp->bio_caller1; 1435219974Smav sc = sd->sd_softc; 1436219974Smav mtx_lock(&sc->sc_queue_mtx); 1437280757Smav bioq_insert_tail(&sc->sc_queue, bp); 1438219974Smav mtx_unlock(&sc->sc_queue_mtx); 1439219974Smav if (!dumping) 1440219974Smav wakeup(sc); 1441219974Smav} 1442219974Smav 1443219974Smavstatic void 1444219974Smavg_raid_disk_done_request(struct bio *bp) 1445219974Smav{ 1446219974Smav struct g_raid_softc *sc; 1447219974Smav struct g_raid_disk *disk; 1448219974Smav struct g_raid_subdisk *sd; 1449219974Smav struct g_raid_volume *vol; 1450219974Smav 1451219974Smav g_topology_assert_not(); 1452219974Smav 1453219974Smav G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); 1454219974Smav sd = bp->bio_caller1; 1455219974Smav sc = sd->sd_softc; 1456219974Smav vol = sd->sd_volume; 1457219974Smav if (bp->bio_from != NULL) { 1458219974Smav bp->bio_from->index--; 1459219974Smav disk = bp->bio_from->private; 1460219974Smav if (disk == NULL) 1461219974Smav g_raid_kill_consumer(sc, bp->bio_from); 1462219974Smav } 1463219974Smav bp->bio_offset -= sd->sd_offset; 1464219974Smav 1465219974Smav G_RAID_TR_IODONE(vol->v_tr, sd, bp); 1466219974Smav} 1467219974Smav 1468219974Smavstatic void 1469219974Smavg_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) 1470219974Smav{ 1471219974Smav 1472219974Smav if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) 1473219974Smav ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); 1474219974Smav else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) 1475219974Smav ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); 1476219974Smav else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) 1477219974Smav ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); 1478219974Smav else 1479219974Smav ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); 1480219974Smav if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { 1481219974Smav KASSERT(ep->e_error == 0, 1482219974Smav ("Error cannot be handled.")); 1483219974Smav g_raid_event_free(ep); 1484219974Smav } else { 1485219974Smav ep->e_flags |= G_RAID_EVENT_DONE; 1486219974Smav G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); 1487219974Smav mtx_lock(&sc->sc_queue_mtx); 1488219974Smav wakeup(ep); 1489219974Smav mtx_unlock(&sc->sc_queue_mtx); 1490219974Smav } 1491219974Smav} 1492219974Smav 1493219974Smav/* 1494219974Smav * Worker thread. 1495219974Smav */ 1496219974Smavstatic void 1497219974Smavg_raid_worker(void *arg) 1498219974Smav{ 1499219974Smav struct g_raid_softc *sc; 1500219974Smav struct g_raid_event *ep; 1501219974Smav struct g_raid_volume *vol; 1502219974Smav struct bio *bp; 1503219974Smav struct timeval now, t; 1504219974Smav int timeout, rv; 1505219974Smav 1506219974Smav sc = arg; 1507219974Smav thread_lock(curthread); 1508219974Smav sched_prio(curthread, PRIBIO); 1509219974Smav thread_unlock(curthread); 1510219974Smav 1511219974Smav sx_xlock(&sc->sc_lock); 1512219974Smav for (;;) { 1513219974Smav mtx_lock(&sc->sc_queue_mtx); 1514219974Smav /* 1515219974Smav * First take a look at events. 1516219974Smav * This is important to handle events before any I/O requests. 1517219974Smav */ 1518219974Smav bp = NULL; 1519219974Smav vol = NULL; 1520219974Smav rv = 0; 1521219974Smav ep = TAILQ_FIRST(&sc->sc_events); 1522219974Smav if (ep != NULL) 1523219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1524219974Smav else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) 1525219974Smav ; 1526219974Smav else { 1527219974Smav getmicrouptime(&now); 1528219974Smav t = now; 1529219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1530219974Smav if (bioq_first(&vol->v_inflight) == NULL && 1531219974Smav vol->v_tr && 1532219974Smav timevalcmp(&vol->v_last_done, &t, < )) 1533219974Smav t = vol->v_last_done; 1534219974Smav } 1535219974Smav timevalsub(&t, &now); 1536219974Smav timeout = g_raid_idle_threshold + 1537219974Smav t.tv_sec * 1000000 + t.tv_usec; 1538219974Smav if (timeout > 0) { 1539219974Smav /* 1540219974Smav * Two steps to avoid overflows at HZ=1000 1541219974Smav * and idle timeouts > 2.1s. Some rounding 1542219974Smav * errors can occur, but they are < 1tick, 1543219974Smav * which is deemed to be close enough for 1544219974Smav * this purpose. 1545219974Smav */ 1546219974Smav int micpertic = 1000000 / hz; 1547219974Smav timeout = (timeout + micpertic - 1) / micpertic; 1548219974Smav sx_xunlock(&sc->sc_lock); 1549219974Smav MSLEEP(rv, sc, &sc->sc_queue_mtx, 1550219974Smav PRIBIO | PDROP, "-", timeout); 1551219974Smav sx_xlock(&sc->sc_lock); 1552219974Smav goto process; 1553219974Smav } else 1554219974Smav rv = EWOULDBLOCK; 1555219974Smav } 1556219974Smav mtx_unlock(&sc->sc_queue_mtx); 1557219974Smavprocess: 1558219974Smav if (ep != NULL) { 1559219974Smav g_raid_handle_event(sc, ep); 1560219974Smav } else if (bp != NULL) { 1561219974Smav if (bp->bio_to != NULL && 1562219974Smav bp->bio_to->geom == sc->sc_geom) 1563219974Smav g_raid_start_request(bp); 1564219974Smav else 1565219974Smav g_raid_disk_done_request(bp); 1566219974Smav } else if (rv == EWOULDBLOCK) { 1567219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1568242314Smav g_raid_clean(vol, -1); 1569219974Smav if (bioq_first(&vol->v_inflight) == NULL && 1570219974Smav vol->v_tr) { 1571219974Smav t.tv_sec = g_raid_idle_threshold / 1000000; 1572219974Smav t.tv_usec = g_raid_idle_threshold % 1000000; 1573219974Smav timevaladd(&t, &vol->v_last_done); 1574219974Smav getmicrouptime(&now); 1575219974Smav if (timevalcmp(&t, &now, <= )) { 1576219974Smav G_RAID_TR_IDLE(vol->v_tr); 1577219974Smav vol->v_last_done = now; 1578219974Smav } 1579219974Smav } 1580219974Smav } 1581219974Smav } 1582219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) 1583219974Smav g_raid_destroy_node(sc, 1); /* May not return. */ 1584219974Smav } 1585219974Smav} 1586219974Smav 1587219974Smavstatic void 1588219974Smavg_raid_poll(struct g_raid_softc *sc) 1589219974Smav{ 1590219974Smav struct g_raid_event *ep; 1591219974Smav struct bio *bp; 1592219974Smav 1593219974Smav sx_xlock(&sc->sc_lock); 1594219974Smav mtx_lock(&sc->sc_queue_mtx); 1595219974Smav /* 1596219974Smav * First take a look at events. 1597219974Smav * This is important to handle events before any I/O requests. 1598219974Smav */ 1599219974Smav ep = TAILQ_FIRST(&sc->sc_events); 1600219974Smav if (ep != NULL) { 1601219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1602219974Smav mtx_unlock(&sc->sc_queue_mtx); 1603219974Smav g_raid_handle_event(sc, ep); 1604219974Smav goto out; 1605219974Smav } 1606219974Smav bp = bioq_takefirst(&sc->sc_queue); 1607219974Smav if (bp != NULL) { 1608219974Smav mtx_unlock(&sc->sc_queue_mtx); 1609219974Smav if (bp->bio_from == NULL || 1610219974Smav bp->bio_from->geom != sc->sc_geom) 1611219974Smav g_raid_start_request(bp); 1612219974Smav else 1613219974Smav g_raid_disk_done_request(bp); 1614219974Smav } 1615219974Smavout: 1616219974Smav sx_xunlock(&sc->sc_lock); 1617219974Smav} 1618219974Smav 1619219974Smavstatic void 1620219974Smavg_raid_launch_provider(struct g_raid_volume *vol) 1621219974Smav{ 1622219974Smav struct g_raid_disk *disk; 1623256610Smav struct g_raid_subdisk *sd; 1624219974Smav struct g_raid_softc *sc; 1625219974Smav struct g_provider *pp; 1626219974Smav char name[G_RAID_MAX_VOLUMENAME]; 1627219974Smav off_t off; 1628256610Smav int i; 1629219974Smav 1630219974Smav sc = vol->v_softc; 1631219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1632219974Smav 1633219974Smav g_topology_lock(); 1634219974Smav /* Try to name provider with volume name. */ 1635219974Smav snprintf(name, sizeof(name), "raid/%s", vol->v_name); 1636219974Smav if (g_raid_name_format == 0 || vol->v_name[0] == 0 || 1637219974Smav g_provider_by_name(name) != NULL) { 1638219974Smav /* Otherwise use sequential volume number. */ 1639219974Smav snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); 1640219974Smav } 1641248068Ssbruno 1642219974Smav pp = g_new_providerf(sc->sc_geom, "%s", name); 1643256880Smav pp->flags |= G_PF_DIRECT_RECEIVE; 1644256610Smav if (vol->v_tr->tro_class->trc_accept_unmapped) { 1645256610Smav pp->flags |= G_PF_ACCEPT_UNMAPPED; 1646256610Smav for (i = 0; i < vol->v_disks_count; i++) { 1647256610Smav sd = &vol->v_subdisks[i]; 1648256610Smav if (sd->sd_state == G_RAID_SUBDISK_S_NONE) 1649256610Smav continue; 1650256610Smav if ((sd->sd_disk->d_consumer->provider->flags & 1651256610Smav G_PF_ACCEPT_UNMAPPED) == 0) 1652256610Smav pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 1653256610Smav } 1654256610Smav } 1655219974Smav pp->private = vol; 1656219974Smav pp->mediasize = vol->v_mediasize; 1657219974Smav pp->sectorsize = vol->v_sectorsize; 1658219974Smav pp->stripesize = 0; 1659219974Smav pp->stripeoffset = 0; 1660219974Smav if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 1661219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || 1662219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || 1663219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { 1664219974Smav if ((disk = vol->v_subdisks[0].sd_disk) != NULL && 1665219974Smav disk->d_consumer != NULL && 1666219974Smav disk->d_consumer->provider != NULL) { 1667219974Smav pp->stripesize = disk->d_consumer->provider->stripesize; 1668219974Smav off = disk->d_consumer->provider->stripeoffset; 1669219974Smav pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; 1670219974Smav if (off > 0) 1671219974Smav pp->stripeoffset %= off; 1672219974Smav } 1673219974Smav if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { 1674219974Smav pp->stripesize *= (vol->v_disks_count - 1); 1675219974Smav pp->stripeoffset *= (vol->v_disks_count - 1); 1676219974Smav } 1677219974Smav } else 1678219974Smav pp->stripesize = vol->v_strip_size; 1679219974Smav vol->v_provider = pp; 1680219974Smav g_error_provider(pp, 0); 1681219974Smav g_topology_unlock(); 1682219974Smav G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", 1683219974Smav pp->name, vol->v_name); 1684219974Smav} 1685219974Smav 1686219974Smavstatic void 1687219974Smavg_raid_destroy_provider(struct g_raid_volume *vol) 1688219974Smav{ 1689219974Smav struct g_raid_softc *sc; 1690219974Smav struct g_provider *pp; 1691219974Smav struct bio *bp, *tmp; 1692219974Smav 1693219974Smav g_topology_assert_not(); 1694219974Smav sc = vol->v_softc; 1695219974Smav pp = vol->v_provider; 1696219974Smav KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); 1697219974Smav 1698219974Smav g_topology_lock(); 1699219974Smav g_error_provider(pp, ENXIO); 1700219974Smav mtx_lock(&sc->sc_queue_mtx); 1701219974Smav TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { 1702219974Smav if (bp->bio_to != pp) 1703219974Smav continue; 1704219974Smav bioq_remove(&sc->sc_queue, bp); 1705219974Smav g_io_deliver(bp, ENXIO); 1706219974Smav } 1707219974Smav mtx_unlock(&sc->sc_queue_mtx); 1708219974Smav G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", 1709219974Smav pp->name, vol->v_name); 1710219974Smav g_wither_provider(pp, ENXIO); 1711219974Smav g_topology_unlock(); 1712219974Smav vol->v_provider = NULL; 1713219974Smav} 1714219974Smav 1715219974Smav/* 1716219974Smav * Update device state. 1717219974Smav */ 1718219974Smavstatic int 1719219974Smavg_raid_update_volume(struct g_raid_volume *vol, u_int event) 1720219974Smav{ 1721219974Smav struct g_raid_softc *sc; 1722219974Smav 1723219974Smav sc = vol->v_softc; 1724219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1725219974Smav 1726219974Smav G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", 1727219974Smav g_raid_volume_event2str(event), 1728219974Smav vol->v_name); 1729219974Smav switch (event) { 1730219974Smav case G_RAID_VOLUME_E_DOWN: 1731219974Smav if (vol->v_provider != NULL) 1732219974Smav g_raid_destroy_provider(vol); 1733219974Smav break; 1734219974Smav case G_RAID_VOLUME_E_UP: 1735219974Smav if (vol->v_provider == NULL) 1736219974Smav g_raid_launch_provider(vol); 1737219974Smav break; 1738219974Smav case G_RAID_VOLUME_E_START: 1739219974Smav if (vol->v_tr) 1740219974Smav G_RAID_TR_START(vol->v_tr); 1741219974Smav return (0); 1742219974Smav default: 1743219974Smav if (sc->sc_md) 1744219974Smav G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); 1745219974Smav return (0); 1746219974Smav } 1747219974Smav 1748219974Smav /* Manage root mount release. */ 1749219974Smav if (vol->v_starting) { 1750219974Smav vol->v_starting = 0; 1751219974Smav G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); 1752219974Smav root_mount_rel(vol->v_rootmount); 1753219974Smav vol->v_rootmount = NULL; 1754219974Smav } 1755219974Smav if (vol->v_stopping && vol->v_provider_open == 0) 1756219974Smav g_raid_destroy_volume(vol); 1757219974Smav return (0); 1758219974Smav} 1759219974Smav 1760219974Smav/* 1761219974Smav * Update subdisk state. 1762219974Smav */ 1763219974Smavstatic int 1764219974Smavg_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) 1765219974Smav{ 1766219974Smav struct g_raid_softc *sc; 1767219974Smav struct g_raid_volume *vol; 1768219974Smav 1769219974Smav sc = sd->sd_softc; 1770219974Smav vol = sd->sd_volume; 1771219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1772219974Smav 1773219974Smav G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", 1774219974Smav g_raid_subdisk_event2str(event), 1775219974Smav vol->v_name, sd->sd_pos, 1776219974Smav sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); 1777219974Smav if (vol->v_tr) 1778219974Smav G_RAID_TR_EVENT(vol->v_tr, sd, event); 1779219974Smav 1780219974Smav return (0); 1781219974Smav} 1782219974Smav 1783219974Smav/* 1784219974Smav * Update disk state. 1785219974Smav */ 1786219974Smavstatic int 1787219974Smavg_raid_update_disk(struct g_raid_disk *disk, u_int event) 1788219974Smav{ 1789219974Smav struct g_raid_softc *sc; 1790219974Smav 1791219974Smav sc = disk->d_softc; 1792219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1793219974Smav 1794219974Smav G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", 1795219974Smav g_raid_disk_event2str(event), 1796219974Smav g_raid_get_diskname(disk)); 1797219974Smav 1798219974Smav if (sc->sc_md) 1799219974Smav G_RAID_MD_EVENT(sc->sc_md, disk, event); 1800219974Smav return (0); 1801219974Smav} 1802219974Smav 1803219974Smav/* 1804219974Smav * Node event. 1805219974Smav */ 1806219974Smavstatic int 1807219974Smavg_raid_update_node(struct g_raid_softc *sc, u_int event) 1808219974Smav{ 1809219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1810219974Smav 1811219974Smav G_RAID_DEBUG1(2, sc, "Event %s for the array.", 1812219974Smav g_raid_node_event2str(event)); 1813219974Smav 1814219974Smav if (event == G_RAID_NODE_E_WAKE) 1815219974Smav return (0); 1816219974Smav if (sc->sc_md) 1817219974Smav G_RAID_MD_EVENT(sc->sc_md, NULL, event); 1818219974Smav return (0); 1819219974Smav} 1820219974Smav 1821219974Smavstatic int 1822219974Smavg_raid_access(struct g_provider *pp, int acr, int acw, int ace) 1823219974Smav{ 1824219974Smav struct g_raid_volume *vol; 1825219974Smav struct g_raid_softc *sc; 1826220210Smav int dcw, opens, error = 0; 1827219974Smav 1828219974Smav g_topology_assert(); 1829219974Smav sc = pp->geom->softc; 1830219974Smav vol = pp->private; 1831219974Smav KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 1832219974Smav KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); 1833219974Smav 1834219974Smav G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, 1835219974Smav acr, acw, ace); 1836219974Smav dcw = pp->acw + acw; 1837219974Smav 1838219974Smav g_topology_unlock(); 1839219974Smav sx_xlock(&sc->sc_lock); 1840219974Smav /* Deny new opens while dying. */ 1841219974Smav if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { 1842219974Smav error = ENXIO; 1843219974Smav goto out; 1844219974Smav } 1845254275Smav /* Deny write opens for read-only volumes. */ 1846254275Smav if (vol->v_read_only && acw > 0) { 1847254275Smav error = EROFS; 1848254275Smav goto out; 1849254275Smav } 1850242314Smav if (dcw == 0) 1851219974Smav g_raid_clean(vol, dcw); 1852219974Smav vol->v_provider_open += acr + acw + ace; 1853219974Smav /* Handle delayed node destruction. */ 1854219974Smav if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && 1855219974Smav vol->v_provider_open == 0) { 1856219974Smav /* Count open volumes. */ 1857219974Smav opens = g_raid_nopens(sc); 1858219974Smav if (opens == 0) { 1859219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 1860219974Smav /* Wake up worker to make it selfdestruct. */ 1861219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1862219974Smav } 1863219974Smav } 1864219974Smav /* Handle open volume destruction. */ 1865219974Smav if (vol->v_stopping && vol->v_provider_open == 0) 1866219974Smav g_raid_destroy_volume(vol); 1867219974Smavout: 1868219974Smav sx_xunlock(&sc->sc_lock); 1869219974Smav g_topology_lock(); 1870219974Smav return (error); 1871219974Smav} 1872219974Smav 1873219974Smavstruct g_raid_softc * 1874219974Smavg_raid_create_node(struct g_class *mp, 1875219974Smav const char *name, struct g_raid_md_object *md) 1876219974Smav{ 1877219974Smav struct g_raid_softc *sc; 1878219974Smav struct g_geom *gp; 1879219974Smav int error; 1880219974Smav 1881219974Smav g_topology_assert(); 1882219974Smav G_RAID_DEBUG(1, "Creating array %s.", name); 1883219974Smav 1884219974Smav gp = g_new_geomf(mp, "%s", name); 1885219974Smav sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); 1886219974Smav gp->start = g_raid_start; 1887219974Smav gp->orphan = g_raid_orphan; 1888219974Smav gp->access = g_raid_access; 1889219974Smav gp->dumpconf = g_raid_dumpconf; 1890219974Smav 1891219974Smav sc->sc_md = md; 1892219974Smav sc->sc_geom = gp; 1893219974Smav sc->sc_flags = 0; 1894219974Smav TAILQ_INIT(&sc->sc_volumes); 1895219974Smav TAILQ_INIT(&sc->sc_disks); 1896234816Smav sx_init(&sc->sc_lock, "graid:lock"); 1897234816Smav mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF); 1898219974Smav TAILQ_INIT(&sc->sc_events); 1899219974Smav bioq_init(&sc->sc_queue); 1900219974Smav gp->softc = sc; 1901219974Smav error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, 1902219974Smav "g_raid %s", name); 1903219974Smav if (error != 0) { 1904219974Smav G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); 1905219974Smav mtx_destroy(&sc->sc_queue_mtx); 1906219974Smav sx_destroy(&sc->sc_lock); 1907219974Smav g_destroy_geom(sc->sc_geom); 1908219974Smav free(sc, M_RAID); 1909219974Smav return (NULL); 1910219974Smav } 1911219974Smav 1912219974Smav G_RAID_DEBUG1(0, sc, "Array %s created.", name); 1913219974Smav return (sc); 1914219974Smav} 1915219974Smav 1916219974Smavstruct g_raid_volume * 1917219974Smavg_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) 1918219974Smav{ 1919219974Smav struct g_raid_volume *vol, *vol1; 1920219974Smav int i; 1921219974Smav 1922219974Smav G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); 1923219974Smav vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); 1924219974Smav vol->v_softc = sc; 1925219974Smav strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); 1926219974Smav vol->v_state = G_RAID_VOLUME_S_STARTING; 1927219974Smav vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 1928219974Smav vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; 1929235076Smav vol->v_rotate_parity = 1; 1930219974Smav bioq_init(&vol->v_inflight); 1931219974Smav bioq_init(&vol->v_locked); 1932219974Smav LIST_INIT(&vol->v_locks); 1933219974Smav for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 1934219974Smav vol->v_subdisks[i].sd_softc = sc; 1935219974Smav vol->v_subdisks[i].sd_volume = vol; 1936219974Smav vol->v_subdisks[i].sd_pos = i; 1937219974Smav vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; 1938219974Smav } 1939219974Smav 1940219974Smav /* Find free ID for this volume. */ 1941219974Smav g_topology_lock(); 1942219974Smav vol1 = vol; 1943219974Smav if (id >= 0) { 1944219974Smav LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1945219974Smav if (vol1->v_global_id == id) 1946219974Smav break; 1947219974Smav } 1948219974Smav } 1949219974Smav if (vol1 != NULL) { 1950219974Smav for (id = 0; ; id++) { 1951219974Smav LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1952219974Smav if (vol1->v_global_id == id) 1953219974Smav break; 1954219974Smav } 1955219974Smav if (vol1 == NULL) 1956219974Smav break; 1957219974Smav } 1958219974Smav } 1959219974Smav vol->v_global_id = id; 1960219974Smav LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); 1961219974Smav g_topology_unlock(); 1962219974Smav 1963219974Smav /* Delay root mounting. */ 1964219974Smav vol->v_rootmount = root_mount_hold("GRAID"); 1965219974Smav G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); 1966219974Smav vol->v_starting = 1; 1967219974Smav TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); 1968219974Smav return (vol); 1969219974Smav} 1970219974Smav 1971219974Smavstruct g_raid_disk * 1972219974Smavg_raid_create_disk(struct g_raid_softc *sc) 1973219974Smav{ 1974219974Smav struct g_raid_disk *disk; 1975219974Smav 1976219974Smav G_RAID_DEBUG1(1, sc, "Creating disk."); 1977219974Smav disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); 1978219974Smav disk->d_softc = sc; 1979219974Smav disk->d_state = G_RAID_DISK_S_NONE; 1980219974Smav TAILQ_INIT(&disk->d_subdisks); 1981219974Smav TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); 1982219974Smav return (disk); 1983219974Smav} 1984219974Smav 1985219974Smavint g_raid_start_volume(struct g_raid_volume *vol) 1986219974Smav{ 1987219974Smav struct g_raid_tr_class *class; 1988219974Smav struct g_raid_tr_object *obj; 1989219974Smav int status; 1990219974Smav 1991219974Smav G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); 1992219974Smav LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { 1993240465Smav if (!class->trc_enable) 1994240465Smav continue; 1995219974Smav G_RAID_DEBUG1(2, vol->v_softc, 1996219974Smav "Tasting volume %s for %s transformation.", 1997219974Smav vol->v_name, class->name); 1998219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 1999219974Smav M_WAITOK); 2000219974Smav obj->tro_class = class; 2001219974Smav obj->tro_volume = vol; 2002219974Smav status = G_RAID_TR_TASTE(obj, vol); 2003219974Smav if (status != G_RAID_TR_TASTE_FAIL) 2004219974Smav break; 2005219974Smav kobj_delete((kobj_t)obj, M_RAID); 2006219974Smav } 2007219974Smav if (class == NULL) { 2008219974Smav G_RAID_DEBUG1(0, vol->v_softc, 2009219974Smav "No transformation module found for %s.", 2010219974Smav vol->v_name); 2011219974Smav vol->v_tr = NULL; 2012219974Smav g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); 2013219974Smav g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, 2014219974Smav G_RAID_EVENT_VOLUME); 2015219974Smav return (-1); 2016219974Smav } 2017219974Smav G_RAID_DEBUG1(2, vol->v_softc, 2018219974Smav "Transformation module %s chosen for %s.", 2019219974Smav class->name, vol->v_name); 2020219974Smav vol->v_tr = obj; 2021219974Smav return (0); 2022219974Smav} 2023219974Smav 2024219974Smavint 2025219974Smavg_raid_destroy_node(struct g_raid_softc *sc, int worker) 2026219974Smav{ 2027219974Smav struct g_raid_volume *vol, *tmpv; 2028219974Smav struct g_raid_disk *disk, *tmpd; 2029219974Smav int error = 0; 2030219974Smav 2031219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 2032219974Smav TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { 2033219974Smav if (g_raid_destroy_volume(vol)) 2034219974Smav error = EBUSY; 2035219974Smav } 2036219974Smav if (error) 2037219974Smav return (error); 2038219974Smav TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { 2039219974Smav if (g_raid_destroy_disk(disk)) 2040219974Smav error = EBUSY; 2041219974Smav } 2042219974Smav if (error) 2043219974Smav return (error); 2044219974Smav if (sc->sc_md) { 2045219974Smav G_RAID_MD_FREE(sc->sc_md); 2046219974Smav kobj_delete((kobj_t)sc->sc_md, M_RAID); 2047219974Smav sc->sc_md = NULL; 2048219974Smav } 2049219974Smav if (sc->sc_geom != NULL) { 2050219974Smav G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); 2051219974Smav g_topology_lock(); 2052219974Smav sc->sc_geom->softc = NULL; 2053219974Smav g_wither_geom(sc->sc_geom, ENXIO); 2054219974Smav g_topology_unlock(); 2055219974Smav sc->sc_geom = NULL; 2056219974Smav } else 2057219974Smav G_RAID_DEBUG(1, "Array destroyed."); 2058219974Smav if (worker) { 2059219974Smav g_raid_event_cancel(sc, sc); 2060219974Smav mtx_destroy(&sc->sc_queue_mtx); 2061219974Smav sx_xunlock(&sc->sc_lock); 2062219974Smav sx_destroy(&sc->sc_lock); 2063219974Smav wakeup(&sc->sc_stopping); 2064219974Smav free(sc, M_RAID); 2065219974Smav curthread->td_pflags &= ~TDP_GEOM; 2066219974Smav G_RAID_DEBUG(1, "Thread exiting."); 2067219974Smav kproc_exit(0); 2068219974Smav } else { 2069219974Smav /* Wake up worker to make it selfdestruct. */ 2070219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2071219974Smav } 2072219974Smav return (0); 2073219974Smav} 2074219974Smav 2075219974Smavint 2076219974Smavg_raid_destroy_volume(struct g_raid_volume *vol) 2077219974Smav{ 2078219974Smav struct g_raid_softc *sc; 2079219974Smav struct g_raid_disk *disk; 2080219974Smav int i; 2081219974Smav 2082219974Smav sc = vol->v_softc; 2083219974Smav G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); 2084219974Smav vol->v_stopping = 1; 2085219974Smav if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { 2086219974Smav if (vol->v_tr) { 2087219974Smav G_RAID_TR_STOP(vol->v_tr); 2088219974Smav return (EBUSY); 2089219974Smav } else 2090219974Smav vol->v_state = G_RAID_VOLUME_S_STOPPED; 2091219974Smav } 2092219974Smav if (g_raid_event_check(sc, vol) != 0) 2093219974Smav return (EBUSY); 2094219974Smav if (vol->v_provider != NULL) 2095219974Smav return (EBUSY); 2096219974Smav if (vol->v_provider_open != 0) 2097219974Smav return (EBUSY); 2098219974Smav if (vol->v_tr) { 2099219974Smav G_RAID_TR_FREE(vol->v_tr); 2100219974Smav kobj_delete((kobj_t)vol->v_tr, M_RAID); 2101219974Smav vol->v_tr = NULL; 2102219974Smav } 2103219974Smav if (vol->v_rootmount) 2104219974Smav root_mount_rel(vol->v_rootmount); 2105219974Smav g_topology_lock(); 2106219974Smav LIST_REMOVE(vol, v_global_next); 2107219974Smav g_topology_unlock(); 2108219974Smav TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); 2109219974Smav for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 2110219974Smav g_raid_event_cancel(sc, &vol->v_subdisks[i]); 2111219974Smav disk = vol->v_subdisks[i].sd_disk; 2112219974Smav if (disk == NULL) 2113219974Smav continue; 2114219974Smav TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); 2115219974Smav } 2116219974Smav G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); 2117219974Smav if (sc->sc_md) 2118219974Smav G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); 2119219974Smav g_raid_event_cancel(sc, vol); 2120219974Smav free(vol, M_RAID); 2121219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) { 2122219974Smav /* Wake up worker to let it selfdestruct. */ 2123219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2124219974Smav } 2125219974Smav return (0); 2126219974Smav} 2127219974Smav 2128219974Smavint 2129219974Smavg_raid_destroy_disk(struct g_raid_disk *disk) 2130219974Smav{ 2131219974Smav struct g_raid_softc *sc; 2132219974Smav struct g_raid_subdisk *sd, *tmp; 2133219974Smav 2134219974Smav sc = disk->d_softc; 2135219974Smav G_RAID_DEBUG1(2, sc, "Destroying disk."); 2136219974Smav if (disk->d_consumer) { 2137219974Smav g_raid_kill_consumer(sc, disk->d_consumer); 2138219974Smav disk->d_consumer = NULL; 2139219974Smav } 2140219974Smav TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { 2141219974Smav g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); 2142219974Smav g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 2143219974Smav G_RAID_EVENT_SUBDISK); 2144219974Smav TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); 2145219974Smav sd->sd_disk = NULL; 2146219974Smav } 2147219974Smav TAILQ_REMOVE(&sc->sc_disks, disk, d_next); 2148219974Smav if (sc->sc_md) 2149219974Smav G_RAID_MD_FREE_DISK(sc->sc_md, disk); 2150219974Smav g_raid_event_cancel(sc, disk); 2151219974Smav free(disk, M_RAID); 2152219974Smav return (0); 2153219974Smav} 2154219974Smav 2155219974Smavint 2156219974Smavg_raid_destroy(struct g_raid_softc *sc, int how) 2157219974Smav{ 2158253706Smav int error, opens; 2159219974Smav 2160219974Smav g_topology_assert_not(); 2161219974Smav if (sc == NULL) 2162219974Smav return (ENXIO); 2163219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 2164219974Smav 2165219974Smav /* Count open volumes. */ 2166219974Smav opens = g_raid_nopens(sc); 2167219974Smav 2168219974Smav /* React on some opened volumes. */ 2169219974Smav if (opens > 0) { 2170219974Smav switch (how) { 2171219974Smav case G_RAID_DESTROY_SOFT: 2172219974Smav G_RAID_DEBUG1(1, sc, 2173219974Smav "%d volumes are still open.", 2174219974Smav opens); 2175253706Smav sx_xunlock(&sc->sc_lock); 2176219974Smav return (EBUSY); 2177219974Smav case G_RAID_DESTROY_DELAYED: 2178219974Smav G_RAID_DEBUG1(1, sc, 2179219974Smav "Array will be destroyed on last close."); 2180219974Smav sc->sc_stopping = G_RAID_DESTROY_DELAYED; 2181253706Smav sx_xunlock(&sc->sc_lock); 2182219974Smav return (EBUSY); 2183219974Smav case G_RAID_DESTROY_HARD: 2184219974Smav G_RAID_DEBUG1(1, sc, 2185219974Smav "%d volumes are still open.", 2186219974Smav opens); 2187219974Smav } 2188219974Smav } 2189219974Smav 2190219974Smav /* Mark node for destruction. */ 2191219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 2192219974Smav /* Wake up worker to let it selfdestruct. */ 2193219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2194219974Smav /* Sleep until node destroyed. */ 2195253706Smav error = sx_sleep(&sc->sc_stopping, &sc->sc_lock, 2196253706Smav PRIBIO | PDROP, "r:destroy", hz * 3); 2197253706Smav return (error == EWOULDBLOCK ? EBUSY : 0); 2198219974Smav} 2199219974Smav 2200219974Smavstatic void 2201219974Smavg_raid_taste_orphan(struct g_consumer *cp) 2202219974Smav{ 2203219974Smav 2204219974Smav KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2205219974Smav cp->provider->name)); 2206219974Smav} 2207219974Smav 2208219974Smavstatic struct g_geom * 2209219974Smavg_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2210219974Smav{ 2211219974Smav struct g_consumer *cp; 2212219974Smav struct g_geom *gp, *geom; 2213219974Smav struct g_raid_md_class *class; 2214219974Smav struct g_raid_md_object *obj; 2215219974Smav int status; 2216219974Smav 2217219974Smav g_topology_assert(); 2218219974Smav g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2219240465Smav if (!g_raid_enable) 2220240465Smav return (NULL); 2221219974Smav G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); 2222219974Smav 2223265054Smav geom = NULL; 2224265054Smav status = G_RAID_MD_TASTE_FAIL; 2225234940Smav gp = g_new_geomf(mp, "raid:taste"); 2226219974Smav /* 2227219974Smav * This orphan function should be never called. 2228219974Smav */ 2229219974Smav gp->orphan = g_raid_taste_orphan; 2230219974Smav cp = g_new_consumer(gp); 2231256880Smav cp->flags |= G_CF_DIRECT_RECEIVE; 2232219974Smav g_attach(cp, pp); 2233265054Smav if (g_access(cp, 1, 0, 0) != 0) 2234265054Smav goto ofail; 2235219974Smav 2236219974Smav LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2237240465Smav if (!class->mdc_enable) 2238240465Smav continue; 2239219974Smav G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", 2240219974Smav pp->name, class->name); 2241219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2242219974Smav M_WAITOK); 2243219974Smav obj->mdo_class = class; 2244219974Smav status = G_RAID_MD_TASTE(obj, mp, cp, &geom); 2245219974Smav if (status != G_RAID_MD_TASTE_NEW) 2246219974Smav kobj_delete((kobj_t)obj, M_RAID); 2247219974Smav if (status != G_RAID_MD_TASTE_FAIL) 2248219974Smav break; 2249219974Smav } 2250219974Smav 2251265054Smav if (status == G_RAID_MD_TASTE_FAIL) 2252265054Smav (void)g_access(cp, -1, 0, 0); 2253265054Smavofail: 2254219974Smav g_detach(cp); 2255219974Smav g_destroy_consumer(cp); 2256219974Smav g_destroy_geom(gp); 2257219974Smav G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); 2258219974Smav return (geom); 2259219974Smav} 2260219974Smav 2261219974Smavint 2262234940Smavg_raid_create_node_format(const char *format, struct gctl_req *req, 2263234940Smav struct g_geom **gp) 2264219974Smav{ 2265219974Smav struct g_raid_md_class *class; 2266219974Smav struct g_raid_md_object *obj; 2267219974Smav int status; 2268219974Smav 2269219974Smav G_RAID_DEBUG(2, "Creating array for %s metadata.", format); 2270219974Smav LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2271219974Smav if (strcasecmp(class->name, format) == 0) 2272219974Smav break; 2273219974Smav } 2274219974Smav if (class == NULL) { 2275219974Smav G_RAID_DEBUG(1, "No support for %s metadata.", format); 2276219974Smav return (G_RAID_MD_TASTE_FAIL); 2277219974Smav } 2278219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2279219974Smav M_WAITOK); 2280219974Smav obj->mdo_class = class; 2281234940Smav status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp); 2282219974Smav if (status != G_RAID_MD_TASTE_NEW) 2283219974Smav kobj_delete((kobj_t)obj, M_RAID); 2284219974Smav return (status); 2285219974Smav} 2286219974Smav 2287219974Smavstatic int 2288219974Smavg_raid_destroy_geom(struct gctl_req *req __unused, 2289219974Smav struct g_class *mp __unused, struct g_geom *gp) 2290219974Smav{ 2291219974Smav struct g_raid_softc *sc; 2292219974Smav int error; 2293219974Smav 2294219974Smav g_topology_unlock(); 2295219974Smav sc = gp->softc; 2296219974Smav sx_xlock(&sc->sc_lock); 2297219974Smav g_cancel_event(sc); 2298219974Smav error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); 2299219974Smav g_topology_lock(); 2300219974Smav return (error); 2301219974Smav} 2302219974Smav 2303219974Smavvoid g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, 2304219974Smav struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2305219974Smav{ 2306219974Smav 2307219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) 2308219974Smav return; 2309219974Smav if (sc->sc_md) 2310219974Smav G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); 2311219974Smav} 2312219974Smav 2313219974Smavvoid g_raid_fail_disk(struct g_raid_softc *sc, 2314219974Smav struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2315219974Smav{ 2316219974Smav 2317219974Smav if (disk == NULL) 2318219974Smav disk = sd->sd_disk; 2319219974Smav if (disk == NULL) { 2320219974Smav G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); 2321219974Smav return; 2322219974Smav } 2323219974Smav if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 2324219974Smav G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " 2325219974Smav "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 2326219974Smav return; 2327219974Smav } 2328219974Smav if (sc->sc_md) 2329219974Smav G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); 2330219974Smav} 2331219974Smav 2332219974Smavstatic void 2333219974Smavg_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2334219974Smav struct g_consumer *cp, struct g_provider *pp) 2335219974Smav{ 2336219974Smav struct g_raid_softc *sc; 2337219974Smav struct g_raid_volume *vol; 2338219974Smav struct g_raid_subdisk *sd; 2339219974Smav struct g_raid_disk *disk; 2340219974Smav int i, s; 2341219974Smav 2342219974Smav g_topology_assert(); 2343219974Smav 2344219974Smav sc = gp->softc; 2345219974Smav if (sc == NULL) 2346219974Smav return; 2347219974Smav if (pp != NULL) { 2348219974Smav vol = pp->private; 2349219974Smav g_topology_unlock(); 2350219974Smav sx_xlock(&sc->sc_lock); 2351249974Smav sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent, 2352249974Smav sc->sc_md->mdo_class->name, 2353249974Smav g_raid_volume_level2str(vol->v_raid_level, 2354249974Smav vol->v_raid_level_qualifier)); 2355219974Smav sbuf_printf(sb, "%s<Label>%s</Label>\n", indent, 2356219974Smav vol->v_name); 2357219974Smav sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent, 2358219974Smav g_raid_volume_level2str(vol->v_raid_level, 2359219974Smav vol->v_raid_level_qualifier)); 2360219974Smav sbuf_printf(sb, 2361219974Smav "%s<Transformation>%s</Transformation>\n", indent, 2362219974Smav vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); 2363219974Smav sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2364219974Smav vol->v_disks_count); 2365219974Smav sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent, 2366219974Smav vol->v_strip_size); 2367219974Smav sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2368219974Smav g_raid_volume_state2str(vol->v_state)); 2369219974Smav sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent, 2370219974Smav vol->v_dirty ? "Yes" : "No"); 2371219974Smav sbuf_printf(sb, "%s<Subdisks>", indent); 2372219974Smav for (i = 0; i < vol->v_disks_count; i++) { 2373219974Smav sd = &vol->v_subdisks[i]; 2374219974Smav if (sd->sd_disk != NULL && 2375219974Smav sd->sd_disk->d_consumer != NULL) { 2376219974Smav sbuf_printf(sb, "%s ", 2377219974Smav g_raid_get_diskname(sd->sd_disk)); 2378219974Smav } else { 2379219974Smav sbuf_printf(sb, "NONE "); 2380219974Smav } 2381219974Smav sbuf_printf(sb, "(%s", 2382219974Smav g_raid_subdisk_state2str(sd->sd_state)); 2383219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2384219974Smav sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2385219974Smav sbuf_printf(sb, " %d%%", 2386219974Smav (int)(sd->sd_rebuild_pos * 100 / 2387219974Smav sd->sd_size)); 2388219974Smav } 2389219974Smav sbuf_printf(sb, ")"); 2390219974Smav if (i + 1 < vol->v_disks_count) 2391219974Smav sbuf_printf(sb, ", "); 2392219974Smav } 2393219974Smav sbuf_printf(sb, "</Subdisks>\n"); 2394219974Smav sx_xunlock(&sc->sc_lock); 2395219974Smav g_topology_lock(); 2396219974Smav } else if (cp != NULL) { 2397219974Smav disk = cp->private; 2398219974Smav if (disk == NULL) 2399219974Smav return; 2400219974Smav g_topology_unlock(); 2401219974Smav sx_xlock(&sc->sc_lock); 2402219974Smav sbuf_printf(sb, "%s<State>%s", indent, 2403219974Smav g_raid_disk_state2str(disk->d_state)); 2404219974Smav if (!TAILQ_EMPTY(&disk->d_subdisks)) { 2405219974Smav sbuf_printf(sb, " ("); 2406219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2407219974Smav sbuf_printf(sb, "%s", 2408219974Smav g_raid_subdisk_state2str(sd->sd_state)); 2409219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2410219974Smav sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2411219974Smav sbuf_printf(sb, " %d%%", 2412219974Smav (int)(sd->sd_rebuild_pos * 100 / 2413219974Smav sd->sd_size)); 2414219974Smav } 2415219974Smav if (TAILQ_NEXT(sd, sd_next)) 2416219974Smav sbuf_printf(sb, ", "); 2417219974Smav } 2418219974Smav sbuf_printf(sb, ")"); 2419219974Smav } 2420219974Smav sbuf_printf(sb, "</State>\n"); 2421219974Smav sbuf_printf(sb, "%s<Subdisks>", indent); 2422219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2423219974Smav sbuf_printf(sb, "r%d(%s):%d@%ju", 2424219974Smav sd->sd_volume->v_global_id, 2425219974Smav sd->sd_volume->v_name, 2426219974Smav sd->sd_pos, sd->sd_offset); 2427219974Smav if (TAILQ_NEXT(sd, sd_next)) 2428219974Smav sbuf_printf(sb, ", "); 2429219974Smav } 2430219974Smav sbuf_printf(sb, "</Subdisks>\n"); 2431219974Smav sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent, 2432219974Smav disk->d_read_errs); 2433219974Smav sx_xunlock(&sc->sc_lock); 2434219974Smav g_topology_lock(); 2435219974Smav } else { 2436219974Smav g_topology_unlock(); 2437219974Smav sx_xlock(&sc->sc_lock); 2438219974Smav if (sc->sc_md) { 2439219974Smav sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent, 2440219974Smav sc->sc_md->mdo_class->name); 2441219974Smav } 2442219974Smav if (!TAILQ_EMPTY(&sc->sc_volumes)) { 2443219974Smav s = 0xff; 2444219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 2445219974Smav if (vol->v_state < s) 2446219974Smav s = vol->v_state; 2447219974Smav } 2448219974Smav sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2449219974Smav g_raid_volume_state2str(s)); 2450219974Smav } 2451219974Smav sx_xunlock(&sc->sc_lock); 2452219974Smav g_topology_lock(); 2453219974Smav } 2454219974Smav} 2455219974Smav 2456219974Smavstatic void 2457242314Smavg_raid_shutdown_post_sync(void *arg, int howto) 2458219974Smav{ 2459219974Smav struct g_class *mp; 2460219974Smav struct g_geom *gp, *gp2; 2461219974Smav struct g_raid_softc *sc; 2462242314Smav struct g_raid_volume *vol; 2463219974Smav 2464219974Smav mp = arg; 2465219974Smav g_topology_lock(); 2466242314Smav g_raid_shutdown = 1; 2467219974Smav LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2468219974Smav if ((sc = gp->softc) == NULL) 2469219974Smav continue; 2470219974Smav g_topology_unlock(); 2471219974Smav sx_xlock(&sc->sc_lock); 2472242314Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) 2473242314Smav g_raid_clean(vol, -1); 2474219974Smav g_cancel_event(sc); 2475253706Smav g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); 2476219974Smav g_topology_lock(); 2477219974Smav } 2478219974Smav g_topology_unlock(); 2479219974Smav} 2480219974Smav 2481219974Smavstatic void 2482219974Smavg_raid_init(struct g_class *mp) 2483219974Smav{ 2484219974Smav 2485242314Smav g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 2486242314Smav g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 2487242314Smav if (g_raid_post_sync == NULL) 2488219974Smav G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); 2489219974Smav g_raid_started = 1; 2490219974Smav} 2491219974Smav 2492219974Smavstatic void 2493219974Smavg_raid_fini(struct g_class *mp) 2494219974Smav{ 2495219974Smav 2496242314Smav if (g_raid_post_sync != NULL) 2497242314Smav EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync); 2498219974Smav g_raid_started = 0; 2499219974Smav} 2500219974Smav 2501219974Smavint 2502219974Smavg_raid_md_modevent(module_t mod, int type, void *arg) 2503219974Smav{ 2504219974Smav struct g_raid_md_class *class, *c, *nc; 2505219974Smav int error; 2506219974Smav 2507219974Smav error = 0; 2508219974Smav class = arg; 2509219974Smav switch (type) { 2510219974Smav case MOD_LOAD: 2511219974Smav c = LIST_FIRST(&g_raid_md_classes); 2512219974Smav if (c == NULL || c->mdc_priority > class->mdc_priority) 2513219974Smav LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); 2514219974Smav else { 2515219974Smav while ((nc = LIST_NEXT(c, mdc_list)) != NULL && 2516219974Smav nc->mdc_priority < class->mdc_priority) 2517219974Smav c = nc; 2518219974Smav LIST_INSERT_AFTER(c, class, mdc_list); 2519219974Smav } 2520219974Smav if (g_raid_started) 2521219974Smav g_retaste(&g_raid_class); 2522219974Smav break; 2523219974Smav case MOD_UNLOAD: 2524219974Smav LIST_REMOVE(class, mdc_list); 2525219974Smav break; 2526219974Smav default: 2527219974Smav error = EOPNOTSUPP; 2528219974Smav break; 2529219974Smav } 2530219974Smav 2531219974Smav return (error); 2532219974Smav} 2533219974Smav 2534219974Smavint 2535219974Smavg_raid_tr_modevent(module_t mod, int type, void *arg) 2536219974Smav{ 2537219974Smav struct g_raid_tr_class *class, *c, *nc; 2538219974Smav int error; 2539219974Smav 2540219974Smav error = 0; 2541219974Smav class = arg; 2542219974Smav switch (type) { 2543219974Smav case MOD_LOAD: 2544219974Smav c = LIST_FIRST(&g_raid_tr_classes); 2545219974Smav if (c == NULL || c->trc_priority > class->trc_priority) 2546219974Smav LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); 2547219974Smav else { 2548219974Smav while ((nc = LIST_NEXT(c, trc_list)) != NULL && 2549219974Smav nc->trc_priority < class->trc_priority) 2550219974Smav c = nc; 2551219974Smav LIST_INSERT_AFTER(c, class, trc_list); 2552219974Smav } 2553219974Smav break; 2554219974Smav case MOD_UNLOAD: 2555219974Smav LIST_REMOVE(class, trc_list); 2556219974Smav break; 2557219974Smav default: 2558219974Smav error = EOPNOTSUPP; 2559219974Smav break; 2560219974Smav } 2561219974Smav 2562219974Smav return (error); 2563219974Smav} 2564219974Smav 2565219974Smav/* 2566219974Smav * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) 2567219974Smav * to reduce module priority, allowing submodules to register them first. 2568219974Smav */ 2569219974Smavstatic moduledata_t g_raid_mod = { 2570219974Smav "g_raid", 2571219974Smav g_modevent, 2572219974Smav &g_raid_class 2573219974Smav}; 2574219974SmavDECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD); 2575219974SmavMODULE_VERSION(geom_raid, 0); 2576