g_raid.c revision 249974
1219974Smav/*- 2219974Smav * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org> 3219974Smav * All rights reserved. 4219974Smav * 5219974Smav * Redistribution and use in source and binary forms, with or without 6219974Smav * modification, are permitted provided that the following conditions 7219974Smav * are met: 8219974Smav * 1. Redistributions of source code must retain the above copyright 9219974Smav * notice, this list of conditions and the following disclaimer. 10219974Smav * 2. Redistributions in binary form must reproduce the above copyright 11219974Smav * notice, this list of conditions and the following disclaimer in the 12219974Smav * documentation and/or other materials provided with the distribution. 13219974Smav * 14219974Smav * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15219974Smav * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16219974Smav * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17219974Smav * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18219974Smav * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19219974Smav * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20219974Smav * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21219974Smav * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22219974Smav * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23219974Smav * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24219974Smav * SUCH DAMAGE. 25219974Smav */ 26219974Smav 27219974Smav#include <sys/cdefs.h> 28219974Smav__FBSDID("$FreeBSD: head/sys/geom/raid/g_raid.c 249974 2013-04-27 06:57:39Z mav $"); 29219974Smav 30219974Smav#include <sys/param.h> 31219974Smav#include <sys/systm.h> 32219974Smav#include <sys/kernel.h> 33219974Smav#include <sys/module.h> 34219974Smav#include <sys/limits.h> 35219974Smav#include <sys/lock.h> 36219974Smav#include <sys/mutex.h> 37219974Smav#include <sys/bio.h> 38223921Sae#include <sys/sbuf.h> 39219974Smav#include <sys/sysctl.h> 40219974Smav#include <sys/malloc.h> 41219974Smav#include <sys/eventhandler.h> 42219974Smav#include <vm/uma.h> 43219974Smav#include <geom/geom.h> 44219974Smav#include <sys/proc.h> 45219974Smav#include <sys/kthread.h> 46219974Smav#include <sys/sched.h> 47219974Smav#include <geom/raid/g_raid.h> 48219974Smav#include "g_raid_md_if.h" 49219974Smav#include "g_raid_tr_if.h" 50219974Smav 51219974Smavstatic MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data"); 52219974Smav 53219974SmavSYSCTL_DECL(_kern_geom); 54219974SmavSYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff"); 55240465Smavint g_raid_enable = 1; 56240465SmavTUNABLE_INT("kern.geom.raid.enable", &g_raid_enable); 57240465SmavSYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RW, 58240465Smav &g_raid_enable, 0, "Enable on-disk metadata taste"); 59219974Smavu_int g_raid_aggressive_spare = 0; 60219974SmavTUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare); 61219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW, 62219974Smav &g_raid_aggressive_spare, 0, "Use disks without metadata as spare"); 63220790Smavu_int g_raid_debug = 0; 64219974SmavTUNABLE_INT("kern.geom.raid.debug", &g_raid_debug); 65219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0, 66219974Smav "Debug level"); 67219974Smavint g_raid_read_err_thresh = 10; 68219974SmavTUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh); 69219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW, 70219974Smav &g_raid_read_err_thresh, 0, 71219974Smav "Number of read errors equated to disk failure"); 72219974Smavu_int g_raid_start_timeout = 30; 73219974SmavTUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout); 74219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW, 75219974Smav &g_raid_start_timeout, 0, 76219974Smav "Time to wait for all array components"); 77219974Smavstatic u_int g_raid_clean_time = 5; 78219974SmavTUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time); 79219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW, 80219974Smav &g_raid_clean_time, 0, "Mark volume as clean when idling"); 81219974Smavstatic u_int g_raid_disconnect_on_failure = 1; 82219974SmavTUNABLE_INT("kern.geom.raid.disconnect_on_failure", 83219974Smav &g_raid_disconnect_on_failure); 84219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, 85219974Smav &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 86219974Smavstatic u_int g_raid_name_format = 0; 87219974SmavTUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format); 88219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW, 89219974Smav &g_raid_name_format, 0, "Providers name format."); 90219974Smavstatic u_int g_raid_idle_threshold = 1000000; 91219974SmavTUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold); 92219974SmavSYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW, 93219974Smav &g_raid_idle_threshold, 1000000, 94219974Smav "Time in microseconds to consider a volume idle."); 95248068Ssbrunostatic u_int ar_legacy_aliases = 1; 96248068SsbrunoSYSCTL_INT(_kern_geom_raid, OID_AUTO, legacy_aliases, CTLFLAG_RW, 97248068Ssbruno &ar_legacy_aliases, 0, "Create aliases named as the legacy ataraid style."); 98248068SsbrunoTUNABLE_INT("kern.geom_raid.legacy_aliases", &ar_legacy_aliases); 99219974Smav 100248068Ssbruno 101219974Smav#define MSLEEP(rv, ident, mtx, priority, wmesg, timeout) do { \ 102219974Smav G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 103219974Smav rv = msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 104219974Smav G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 105219974Smav} while (0) 106219974Smav 107219974SmavLIST_HEAD(, g_raid_md_class) g_raid_md_classes = 108219974Smav LIST_HEAD_INITIALIZER(g_raid_md_classes); 109219974Smav 110219974SmavLIST_HEAD(, g_raid_tr_class) g_raid_tr_classes = 111219974Smav LIST_HEAD_INITIALIZER(g_raid_tr_classes); 112219974Smav 113219974SmavLIST_HEAD(, g_raid_volume) g_raid_volumes = 114219974Smav LIST_HEAD_INITIALIZER(g_raid_volumes); 115219974Smav 116242314Smavstatic eventhandler_tag g_raid_post_sync = NULL; 117219974Smavstatic int g_raid_started = 0; 118242314Smavstatic int g_raid_shutdown = 0; 119219974Smav 120219974Smavstatic int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp, 121219974Smav struct g_geom *gp); 122219974Smavstatic g_taste_t g_raid_taste; 123219974Smavstatic void g_raid_init(struct g_class *mp); 124219974Smavstatic void g_raid_fini(struct g_class *mp); 125219974Smav 126219974Smavstruct g_class g_raid_class = { 127219974Smav .name = G_RAID_CLASS_NAME, 128219974Smav .version = G_VERSION, 129219974Smav .ctlreq = g_raid_ctl, 130219974Smav .taste = g_raid_taste, 131219974Smav .destroy_geom = g_raid_destroy_geom, 132219974Smav .init = g_raid_init, 133219974Smav .fini = g_raid_fini 134219974Smav}; 135219974Smav 136219974Smavstatic void g_raid_destroy_provider(struct g_raid_volume *vol); 137219974Smavstatic int g_raid_update_disk(struct g_raid_disk *disk, u_int event); 138219974Smavstatic int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event); 139219974Smavstatic int g_raid_update_volume(struct g_raid_volume *vol, u_int event); 140219974Smavstatic int g_raid_update_node(struct g_raid_softc *sc, u_int event); 141219974Smavstatic void g_raid_dumpconf(struct sbuf *sb, const char *indent, 142219974Smav struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 143219974Smavstatic void g_raid_start(struct bio *bp); 144219974Smavstatic void g_raid_start_request(struct bio *bp); 145219974Smavstatic void g_raid_disk_done(struct bio *bp); 146219974Smavstatic void g_raid_poll(struct g_raid_softc *sc); 147219974Smav 148219974Smavstatic const char * 149219974Smavg_raid_node_event2str(int event) 150219974Smav{ 151219974Smav 152219974Smav switch (event) { 153219974Smav case G_RAID_NODE_E_WAKE: 154219974Smav return ("WAKE"); 155219974Smav case G_RAID_NODE_E_START: 156219974Smav return ("START"); 157219974Smav default: 158219974Smav return ("INVALID"); 159219974Smav } 160219974Smav} 161219974Smav 162219974Smavconst char * 163219974Smavg_raid_disk_state2str(int state) 164219974Smav{ 165219974Smav 166219974Smav switch (state) { 167219974Smav case G_RAID_DISK_S_NONE: 168219974Smav return ("NONE"); 169219974Smav case G_RAID_DISK_S_OFFLINE: 170219974Smav return ("OFFLINE"); 171245326Smav case G_RAID_DISK_S_DISABLED: 172245326Smav return ("DISABLED"); 173219974Smav case G_RAID_DISK_S_FAILED: 174219974Smav return ("FAILED"); 175219974Smav case G_RAID_DISK_S_STALE_FAILED: 176219974Smav return ("STALE_FAILED"); 177219974Smav case G_RAID_DISK_S_SPARE: 178219974Smav return ("SPARE"); 179219974Smav case G_RAID_DISK_S_STALE: 180219974Smav return ("STALE"); 181219974Smav case G_RAID_DISK_S_ACTIVE: 182219974Smav return ("ACTIVE"); 183219974Smav default: 184219974Smav return ("INVALID"); 185219974Smav } 186219974Smav} 187219974Smav 188219974Smavstatic const char * 189219974Smavg_raid_disk_event2str(int event) 190219974Smav{ 191219974Smav 192219974Smav switch (event) { 193219974Smav case G_RAID_DISK_E_DISCONNECTED: 194219974Smav return ("DISCONNECTED"); 195219974Smav default: 196219974Smav return ("INVALID"); 197219974Smav } 198219974Smav} 199219974Smav 200219974Smavconst char * 201219974Smavg_raid_subdisk_state2str(int state) 202219974Smav{ 203219974Smav 204219974Smav switch (state) { 205219974Smav case G_RAID_SUBDISK_S_NONE: 206219974Smav return ("NONE"); 207219974Smav case G_RAID_SUBDISK_S_FAILED: 208219974Smav return ("FAILED"); 209219974Smav case G_RAID_SUBDISK_S_NEW: 210219974Smav return ("NEW"); 211219974Smav case G_RAID_SUBDISK_S_REBUILD: 212219974Smav return ("REBUILD"); 213219974Smav case G_RAID_SUBDISK_S_UNINITIALIZED: 214219974Smav return ("UNINITIALIZED"); 215219974Smav case G_RAID_SUBDISK_S_STALE: 216219974Smav return ("STALE"); 217219974Smav case G_RAID_SUBDISK_S_RESYNC: 218219974Smav return ("RESYNC"); 219219974Smav case G_RAID_SUBDISK_S_ACTIVE: 220219974Smav return ("ACTIVE"); 221219974Smav default: 222219974Smav return ("INVALID"); 223219974Smav } 224219974Smav} 225219974Smav 226219974Smavstatic const char * 227219974Smavg_raid_subdisk_event2str(int event) 228219974Smav{ 229219974Smav 230219974Smav switch (event) { 231219974Smav case G_RAID_SUBDISK_E_NEW: 232219974Smav return ("NEW"); 233239175Smav case G_RAID_SUBDISK_E_FAILED: 234239175Smav return ("FAILED"); 235219974Smav case G_RAID_SUBDISK_E_DISCONNECTED: 236219974Smav return ("DISCONNECTED"); 237219974Smav default: 238219974Smav return ("INVALID"); 239219974Smav } 240219974Smav} 241219974Smav 242219974Smavconst char * 243219974Smavg_raid_volume_state2str(int state) 244219974Smav{ 245219974Smav 246219974Smav switch (state) { 247219974Smav case G_RAID_VOLUME_S_STARTING: 248219974Smav return ("STARTING"); 249219974Smav case G_RAID_VOLUME_S_BROKEN: 250219974Smav return ("BROKEN"); 251219974Smav case G_RAID_VOLUME_S_DEGRADED: 252219974Smav return ("DEGRADED"); 253219974Smav case G_RAID_VOLUME_S_SUBOPTIMAL: 254219974Smav return ("SUBOPTIMAL"); 255219974Smav case G_RAID_VOLUME_S_OPTIMAL: 256219974Smav return ("OPTIMAL"); 257219974Smav case G_RAID_VOLUME_S_UNSUPPORTED: 258219974Smav return ("UNSUPPORTED"); 259219974Smav case G_RAID_VOLUME_S_STOPPED: 260219974Smav return ("STOPPED"); 261219974Smav default: 262219974Smav return ("INVALID"); 263219974Smav } 264219974Smav} 265219974Smav 266219974Smavstatic const char * 267219974Smavg_raid_volume_event2str(int event) 268219974Smav{ 269219974Smav 270219974Smav switch (event) { 271219974Smav case G_RAID_VOLUME_E_UP: 272219974Smav return ("UP"); 273219974Smav case G_RAID_VOLUME_E_DOWN: 274219974Smav return ("DOWN"); 275219974Smav case G_RAID_VOLUME_E_START: 276219974Smav return ("START"); 277219974Smav case G_RAID_VOLUME_E_STARTMD: 278219974Smav return ("STARTMD"); 279219974Smav default: 280219974Smav return ("INVALID"); 281219974Smav } 282219974Smav} 283219974Smav 284219974Smavconst char * 285219974Smavg_raid_volume_level2str(int level, int qual) 286219974Smav{ 287219974Smav 288219974Smav switch (level) { 289219974Smav case G_RAID_VOLUME_RL_RAID0: 290219974Smav return ("RAID0"); 291219974Smav case G_RAID_VOLUME_RL_RAID1: 292219974Smav return ("RAID1"); 293219974Smav case G_RAID_VOLUME_RL_RAID3: 294234603Smav if (qual == G_RAID_VOLUME_RLQ_R3P0) 295234603Smav return ("RAID3-P0"); 296234603Smav if (qual == G_RAID_VOLUME_RLQ_R3PN) 297234603Smav return ("RAID3-PN"); 298219974Smav return ("RAID3"); 299219974Smav case G_RAID_VOLUME_RL_RAID4: 300234603Smav if (qual == G_RAID_VOLUME_RLQ_R4P0) 301234610Smav return ("RAID4-P0"); 302234603Smav if (qual == G_RAID_VOLUME_RLQ_R4PN) 303234610Smav return ("RAID4-PN"); 304219974Smav return ("RAID4"); 305219974Smav case G_RAID_VOLUME_RL_RAID5: 306234458Smav if (qual == G_RAID_VOLUME_RLQ_R5RA) 307234603Smav return ("RAID5-RA"); 308234458Smav if (qual == G_RAID_VOLUME_RLQ_R5RS) 309234603Smav return ("RAID5-RS"); 310234458Smav if (qual == G_RAID_VOLUME_RLQ_R5LA) 311234603Smav return ("RAID5-LA"); 312234458Smav if (qual == G_RAID_VOLUME_RLQ_R5LS) 313234603Smav return ("RAID5-LS"); 314219974Smav return ("RAID5"); 315219974Smav case G_RAID_VOLUME_RL_RAID6: 316234603Smav if (qual == G_RAID_VOLUME_RLQ_R6RA) 317234603Smav return ("RAID6-RA"); 318234603Smav if (qual == G_RAID_VOLUME_RLQ_R6RS) 319234603Smav return ("RAID6-RS"); 320234603Smav if (qual == G_RAID_VOLUME_RLQ_R6LA) 321234603Smav return ("RAID6-LA"); 322234603Smav if (qual == G_RAID_VOLUME_RLQ_R6LS) 323234603Smav return ("RAID6-LS"); 324219974Smav return ("RAID6"); 325234603Smav case G_RAID_VOLUME_RL_RAIDMDF: 326234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFRA) 327234603Smav return ("RAIDMDF-RA"); 328234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFRS) 329234603Smav return ("RAIDMDF-RS"); 330234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFLA) 331234603Smav return ("RAIDMDF-LA"); 332234603Smav if (qual == G_RAID_VOLUME_RLQ_RMDFLS) 333234603Smav return ("RAIDMDF-LS"); 334234603Smav return ("RAIDMDF"); 335219974Smav case G_RAID_VOLUME_RL_RAID1E: 336234603Smav if (qual == G_RAID_VOLUME_RLQ_R1EA) 337234603Smav return ("RAID1E-A"); 338234603Smav if (qual == G_RAID_VOLUME_RLQ_R1EO) 339234603Smav return ("RAID1E-O"); 340219974Smav return ("RAID1E"); 341219974Smav case G_RAID_VOLUME_RL_SINGLE: 342219974Smav return ("SINGLE"); 343219974Smav case G_RAID_VOLUME_RL_CONCAT: 344219974Smav return ("CONCAT"); 345219974Smav case G_RAID_VOLUME_RL_RAID5E: 346234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ERA) 347234603Smav return ("RAID5E-RA"); 348234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ERS) 349234603Smav return ("RAID5E-RS"); 350234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ELA) 351234603Smav return ("RAID5E-LA"); 352234603Smav if (qual == G_RAID_VOLUME_RLQ_R5ELS) 353234603Smav return ("RAID5E-LS"); 354219974Smav return ("RAID5E"); 355219974Smav case G_RAID_VOLUME_RL_RAID5EE: 356234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EERA) 357234603Smav return ("RAID5EE-RA"); 358234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EERS) 359234603Smav return ("RAID5EE-RS"); 360234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EELA) 361234603Smav return ("RAID5EE-LA"); 362234603Smav if (qual == G_RAID_VOLUME_RLQ_R5EELS) 363234603Smav return ("RAID5EE-LS"); 364219974Smav return ("RAID5EE"); 365234603Smav case G_RAID_VOLUME_RL_RAID5R: 366234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RRA) 367234603Smav return ("RAID5R-RA"); 368234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RRS) 369234603Smav return ("RAID5R-RS"); 370234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RLA) 371234603Smav return ("RAID5R-LA"); 372234603Smav if (qual == G_RAID_VOLUME_RLQ_R5RLS) 373234603Smav return ("RAID5R-LS"); 374234603Smav return ("RAID5E"); 375219974Smav default: 376219974Smav return ("UNKNOWN"); 377219974Smav } 378219974Smav} 379219974Smav 380219974Smavint 381219974Smavg_raid_volume_str2level(const char *str, int *level, int *qual) 382219974Smav{ 383219974Smav 384219974Smav *level = G_RAID_VOLUME_RL_UNKNOWN; 385219974Smav *qual = G_RAID_VOLUME_RLQ_NONE; 386219974Smav if (strcasecmp(str, "RAID0") == 0) 387219974Smav *level = G_RAID_VOLUME_RL_RAID0; 388219974Smav else if (strcasecmp(str, "RAID1") == 0) 389219974Smav *level = G_RAID_VOLUME_RL_RAID1; 390234603Smav else if (strcasecmp(str, "RAID3-P0") == 0) { 391219974Smav *level = G_RAID_VOLUME_RL_RAID3; 392234603Smav *qual = G_RAID_VOLUME_RLQ_R3P0; 393234993Smav } else if (strcasecmp(str, "RAID3-PN") == 0 || 394234603Smav strcasecmp(str, "RAID3") == 0) { 395234603Smav *level = G_RAID_VOLUME_RL_RAID3; 396234993Smav *qual = G_RAID_VOLUME_RLQ_R3PN; 397234603Smav } else if (strcasecmp(str, "RAID4-P0") == 0) { 398219974Smav *level = G_RAID_VOLUME_RL_RAID4; 399234603Smav *qual = G_RAID_VOLUME_RLQ_R4P0; 400234993Smav } else if (strcasecmp(str, "RAID4-PN") == 0 || 401234603Smav strcasecmp(str, "RAID4") == 0) { 402234603Smav *level = G_RAID_VOLUME_RL_RAID4; 403234993Smav *qual = G_RAID_VOLUME_RLQ_R4PN; 404234603Smav } else if (strcasecmp(str, "RAID5-RA") == 0) { 405219974Smav *level = G_RAID_VOLUME_RL_RAID5; 406234458Smav *qual = G_RAID_VOLUME_RLQ_R5RA; 407234603Smav } else if (strcasecmp(str, "RAID5-RS") == 0) { 408234458Smav *level = G_RAID_VOLUME_RL_RAID5; 409234458Smav *qual = G_RAID_VOLUME_RLQ_R5RS; 410234458Smav } else if (strcasecmp(str, "RAID5") == 0 || 411234603Smav strcasecmp(str, "RAID5-LA") == 0) { 412234458Smav *level = G_RAID_VOLUME_RL_RAID5; 413234458Smav *qual = G_RAID_VOLUME_RLQ_R5LA; 414234603Smav } else if (strcasecmp(str, "RAID5-LS") == 0) { 415234458Smav *level = G_RAID_VOLUME_RL_RAID5; 416234458Smav *qual = G_RAID_VOLUME_RLQ_R5LS; 417234603Smav } else if (strcasecmp(str, "RAID6-RA") == 0) { 418219974Smav *level = G_RAID_VOLUME_RL_RAID6; 419234603Smav *qual = G_RAID_VOLUME_RLQ_R6RA; 420234603Smav } else if (strcasecmp(str, "RAID6-RS") == 0) { 421234603Smav *level = G_RAID_VOLUME_RL_RAID6; 422234603Smav *qual = G_RAID_VOLUME_RLQ_R6RS; 423234603Smav } else if (strcasecmp(str, "RAID6") == 0 || 424234603Smav strcasecmp(str, "RAID6-LA") == 0) { 425234603Smav *level = G_RAID_VOLUME_RL_RAID6; 426234603Smav *qual = G_RAID_VOLUME_RLQ_R6LA; 427234603Smav } else if (strcasecmp(str, "RAID6-LS") == 0) { 428234603Smav *level = G_RAID_VOLUME_RL_RAID6; 429234603Smav *qual = G_RAID_VOLUME_RLQ_R6LS; 430234603Smav } else if (strcasecmp(str, "RAIDMDF-RA") == 0) { 431234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 432234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFRA; 433234603Smav } else if (strcasecmp(str, "RAIDMDF-RS") == 0) { 434234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 435234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFRS; 436234603Smav } else if (strcasecmp(str, "RAIDMDF") == 0 || 437234603Smav strcasecmp(str, "RAIDMDF-LA") == 0) { 438234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 439234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFLA; 440234603Smav } else if (strcasecmp(str, "RAIDMDF-LS") == 0) { 441234603Smav *level = G_RAID_VOLUME_RL_RAIDMDF; 442234603Smav *qual = G_RAID_VOLUME_RLQ_RMDFLS; 443234603Smav } else if (strcasecmp(str, "RAID10") == 0 || 444234603Smav strcasecmp(str, "RAID1E") == 0 || 445234603Smav strcasecmp(str, "RAID1E-A") == 0) { 446219974Smav *level = G_RAID_VOLUME_RL_RAID1E; 447234603Smav *qual = G_RAID_VOLUME_RLQ_R1EA; 448234603Smav } else if (strcasecmp(str, "RAID1E-O") == 0) { 449234603Smav *level = G_RAID_VOLUME_RL_RAID1E; 450234603Smav *qual = G_RAID_VOLUME_RLQ_R1EO; 451234603Smav } else if (strcasecmp(str, "SINGLE") == 0) 452219974Smav *level = G_RAID_VOLUME_RL_SINGLE; 453219974Smav else if (strcasecmp(str, "CONCAT") == 0) 454219974Smav *level = G_RAID_VOLUME_RL_CONCAT; 455234603Smav else if (strcasecmp(str, "RAID5E-RA") == 0) { 456219974Smav *level = G_RAID_VOLUME_RL_RAID5E; 457234603Smav *qual = G_RAID_VOLUME_RLQ_R5ERA; 458234603Smav } else if (strcasecmp(str, "RAID5E-RS") == 0) { 459234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 460234603Smav *qual = G_RAID_VOLUME_RLQ_R5ERS; 461234603Smav } else if (strcasecmp(str, "RAID5E") == 0 || 462234603Smav strcasecmp(str, "RAID5E-LA") == 0) { 463234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 464234603Smav *qual = G_RAID_VOLUME_RLQ_R5ELA; 465234603Smav } else if (strcasecmp(str, "RAID5E-LS") == 0) { 466234603Smav *level = G_RAID_VOLUME_RL_RAID5E; 467234603Smav *qual = G_RAID_VOLUME_RLQ_R5ELS; 468234603Smav } else if (strcasecmp(str, "RAID5EE-RA") == 0) { 469219974Smav *level = G_RAID_VOLUME_RL_RAID5EE; 470234603Smav *qual = G_RAID_VOLUME_RLQ_R5EERA; 471234603Smav } else if (strcasecmp(str, "RAID5EE-RS") == 0) { 472234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 473234603Smav *qual = G_RAID_VOLUME_RLQ_R5EERS; 474234603Smav } else if (strcasecmp(str, "RAID5EE") == 0 || 475234603Smav strcasecmp(str, "RAID5EE-LA") == 0) { 476234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 477234603Smav *qual = G_RAID_VOLUME_RLQ_R5EELA; 478234603Smav } else if (strcasecmp(str, "RAID5EE-LS") == 0) { 479234603Smav *level = G_RAID_VOLUME_RL_RAID5EE; 480234603Smav *qual = G_RAID_VOLUME_RLQ_R5EELS; 481234603Smav } else if (strcasecmp(str, "RAID5R-RA") == 0) { 482234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 483234603Smav *qual = G_RAID_VOLUME_RLQ_R5RRA; 484234603Smav } else if (strcasecmp(str, "RAID5R-RS") == 0) { 485234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 486234603Smav *qual = G_RAID_VOLUME_RLQ_R5RRS; 487234603Smav } else if (strcasecmp(str, "RAID5R") == 0 || 488234603Smav strcasecmp(str, "RAID5R-LA") == 0) { 489234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 490234603Smav *qual = G_RAID_VOLUME_RLQ_R5RLA; 491234603Smav } else if (strcasecmp(str, "RAID5R-LS") == 0) { 492234603Smav *level = G_RAID_VOLUME_RL_RAID5R; 493234603Smav *qual = G_RAID_VOLUME_RLQ_R5RLS; 494234603Smav } else 495219974Smav return (-1); 496219974Smav return (0); 497219974Smav} 498219974Smav 499219974Smavconst char * 500219974Smavg_raid_get_diskname(struct g_raid_disk *disk) 501219974Smav{ 502219974Smav 503219974Smav if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 504219974Smav return ("[unknown]"); 505219974Smav return (disk->d_consumer->provider->name); 506219974Smav} 507219974Smav 508219974Smavvoid 509242323Smavg_raid_get_disk_info(struct g_raid_disk *disk) 510242323Smav{ 511242323Smav struct g_consumer *cp = disk->d_consumer; 512242323Smav int error, len; 513242323Smav 514242323Smav /* Read kernel dumping information. */ 515242323Smav disk->d_kd.offset = 0; 516242323Smav disk->d_kd.length = OFF_MAX; 517242323Smav len = sizeof(disk->d_kd); 518242323Smav error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); 519242323Smav if (error) 520242323Smav disk->d_kd.di.dumper = NULL; 521242323Smav if (disk->d_kd.di.dumper == NULL) 522242323Smav G_RAID_DEBUG1(2, disk->d_softc, 523242323Smav "Dumping not supported by %s: %d.", 524242323Smav cp->provider->name, error); 525242323Smav 526242323Smav /* Read BIO_DELETE support. */ 527242323Smav error = g_getattr("GEOM::candelete", cp, &disk->d_candelete); 528242323Smav if (error) 529242323Smav disk->d_candelete = 0; 530242323Smav if (!disk->d_candelete) 531242323Smav G_RAID_DEBUG1(2, disk->d_softc, 532242323Smav "BIO_DELETE not supported by %s: %d.", 533242323Smav cp->provider->name, error); 534242323Smav} 535242323Smav 536242323Smavvoid 537219974Smavg_raid_report_disk_state(struct g_raid_disk *disk) 538219974Smav{ 539219974Smav struct g_raid_subdisk *sd; 540219974Smav int len, state; 541219974Smav uint32_t s; 542219974Smav 543219974Smav if (disk->d_consumer == NULL) 544219974Smav return; 545245326Smav if (disk->d_state == G_RAID_DISK_S_DISABLED) { 546245363Smav s = G_STATE_ACTIVE; /* XXX */ 547245326Smav } else if (disk->d_state == G_RAID_DISK_S_FAILED || 548219974Smav disk->d_state == G_RAID_DISK_S_STALE_FAILED) { 549219974Smav s = G_STATE_FAILED; 550219974Smav } else { 551219974Smav state = G_RAID_SUBDISK_S_ACTIVE; 552219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 553219974Smav if (sd->sd_state < state) 554219974Smav state = sd->sd_state; 555219974Smav } 556219974Smav if (state == G_RAID_SUBDISK_S_FAILED) 557219974Smav s = G_STATE_FAILED; 558219974Smav else if (state == G_RAID_SUBDISK_S_NEW || 559219974Smav state == G_RAID_SUBDISK_S_REBUILD) 560219974Smav s = G_STATE_REBUILD; 561219974Smav else if (state == G_RAID_SUBDISK_S_STALE || 562219974Smav state == G_RAID_SUBDISK_S_RESYNC) 563219974Smav s = G_STATE_RESYNC; 564219974Smav else 565219974Smav s = G_STATE_ACTIVE; 566219974Smav } 567219974Smav len = sizeof(s); 568219974Smav g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s); 569219974Smav G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.", 570219974Smav g_raid_get_diskname(disk), s); 571219974Smav} 572219974Smav 573219974Smavvoid 574219974Smavg_raid_change_disk_state(struct g_raid_disk *disk, int state) 575219974Smav{ 576219974Smav 577219974Smav G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.", 578219974Smav g_raid_get_diskname(disk), 579219974Smav g_raid_disk_state2str(disk->d_state), 580219974Smav g_raid_disk_state2str(state)); 581219974Smav disk->d_state = state; 582219974Smav g_raid_report_disk_state(disk); 583219974Smav} 584219974Smav 585219974Smavvoid 586219974Smavg_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state) 587219974Smav{ 588219974Smav 589219974Smav G_RAID_DEBUG1(0, sd->sd_softc, 590219974Smav "Subdisk %s:%d-%s state changed from %s to %s.", 591219974Smav sd->sd_volume->v_name, sd->sd_pos, 592219974Smav sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", 593219974Smav g_raid_subdisk_state2str(sd->sd_state), 594219974Smav g_raid_subdisk_state2str(state)); 595219974Smav sd->sd_state = state; 596219974Smav if (sd->sd_disk) 597219974Smav g_raid_report_disk_state(sd->sd_disk); 598219974Smav} 599219974Smav 600219974Smavvoid 601219974Smavg_raid_change_volume_state(struct g_raid_volume *vol, int state) 602219974Smav{ 603219974Smav 604219974Smav G_RAID_DEBUG1(0, vol->v_softc, 605219974Smav "Volume %s state changed from %s to %s.", 606219974Smav vol->v_name, 607219974Smav g_raid_volume_state2str(vol->v_state), 608219974Smav g_raid_volume_state2str(state)); 609219974Smav vol->v_state = state; 610219974Smav} 611219974Smav 612219974Smav/* 613219974Smav * --- Events handling functions --- 614219974Smav * Events in geom_raid are used to maintain subdisks and volumes status 615219974Smav * from one thread to simplify locking. 616219974Smav */ 617219974Smavstatic void 618219974Smavg_raid_event_free(struct g_raid_event *ep) 619219974Smav{ 620219974Smav 621219974Smav free(ep, M_RAID); 622219974Smav} 623219974Smav 624219974Smavint 625219974Smavg_raid_event_send(void *arg, int event, int flags) 626219974Smav{ 627219974Smav struct g_raid_softc *sc; 628219974Smav struct g_raid_event *ep; 629219974Smav int error; 630219974Smav 631219974Smav if ((flags & G_RAID_EVENT_VOLUME) != 0) { 632219974Smav sc = ((struct g_raid_volume *)arg)->v_softc; 633219974Smav } else if ((flags & G_RAID_EVENT_DISK) != 0) { 634219974Smav sc = ((struct g_raid_disk *)arg)->d_softc; 635219974Smav } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) { 636219974Smav sc = ((struct g_raid_subdisk *)arg)->sd_softc; 637219974Smav } else { 638219974Smav sc = arg; 639219974Smav } 640219974Smav ep = malloc(sizeof(*ep), M_RAID, 641219974Smav sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT); 642219974Smav if (ep == NULL) 643219974Smav return (ENOMEM); 644219974Smav ep->e_tgt = arg; 645219974Smav ep->e_event = event; 646219974Smav ep->e_flags = flags; 647219974Smav ep->e_error = 0; 648219974Smav G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc); 649219974Smav mtx_lock(&sc->sc_queue_mtx); 650219974Smav TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 651219974Smav mtx_unlock(&sc->sc_queue_mtx); 652219974Smav wakeup(sc); 653219974Smav 654219974Smav if ((flags & G_RAID_EVENT_WAIT) == 0) 655219974Smav return (0); 656219974Smav 657219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 658219974Smav G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep); 659219974Smav sx_xunlock(&sc->sc_lock); 660219974Smav while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) { 661219974Smav mtx_lock(&sc->sc_queue_mtx); 662219974Smav MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event", 663219974Smav hz * 5); 664219974Smav } 665219974Smav error = ep->e_error; 666219974Smav g_raid_event_free(ep); 667219974Smav sx_xlock(&sc->sc_lock); 668219974Smav return (error); 669219974Smav} 670219974Smav 671219974Smavstatic void 672219974Smavg_raid_event_cancel(struct g_raid_softc *sc, void *tgt) 673219974Smav{ 674219974Smav struct g_raid_event *ep, *tmpep; 675219974Smav 676219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 677219974Smav 678219974Smav mtx_lock(&sc->sc_queue_mtx); 679219974Smav TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 680219974Smav if (ep->e_tgt != tgt) 681219974Smav continue; 682219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 683219974Smav if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) 684219974Smav g_raid_event_free(ep); 685219974Smav else { 686219974Smav ep->e_error = ECANCELED; 687219974Smav wakeup(ep); 688219974Smav } 689219974Smav } 690219974Smav mtx_unlock(&sc->sc_queue_mtx); 691219974Smav} 692219974Smav 693219974Smavstatic int 694219974Smavg_raid_event_check(struct g_raid_softc *sc, void *tgt) 695219974Smav{ 696219974Smav struct g_raid_event *ep; 697219974Smav int res = 0; 698219974Smav 699219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 700219974Smav 701219974Smav mtx_lock(&sc->sc_queue_mtx); 702219974Smav TAILQ_FOREACH(ep, &sc->sc_events, e_next) { 703219974Smav if (ep->e_tgt != tgt) 704219974Smav continue; 705219974Smav res = 1; 706219974Smav break; 707219974Smav } 708219974Smav mtx_unlock(&sc->sc_queue_mtx); 709219974Smav return (res); 710219974Smav} 711219974Smav 712219974Smav/* 713219974Smav * Return the number of disks in given state. 714219974Smav * If state is equal to -1, count all connected disks. 715219974Smav */ 716219974Smavu_int 717219974Smavg_raid_ndisks(struct g_raid_softc *sc, int state) 718219974Smav{ 719219974Smav struct g_raid_disk *disk; 720219974Smav u_int n; 721219974Smav 722219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 723219974Smav 724219974Smav n = 0; 725219974Smav TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 726219974Smav if (disk->d_state == state || state == -1) 727219974Smav n++; 728219974Smav } 729219974Smav return (n); 730219974Smav} 731219974Smav 732219974Smav/* 733219974Smav * Return the number of subdisks in given state. 734219974Smav * If state is equal to -1, count all connected disks. 735219974Smav */ 736219974Smavu_int 737219974Smavg_raid_nsubdisks(struct g_raid_volume *vol, int state) 738219974Smav{ 739219974Smav struct g_raid_subdisk *subdisk; 740219974Smav struct g_raid_softc *sc; 741219974Smav u_int i, n ; 742219974Smav 743219974Smav sc = vol->v_softc; 744219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 745219974Smav 746219974Smav n = 0; 747219974Smav for (i = 0; i < vol->v_disks_count; i++) { 748219974Smav subdisk = &vol->v_subdisks[i]; 749219974Smav if ((state == -1 && 750219974Smav subdisk->sd_state != G_RAID_SUBDISK_S_NONE) || 751219974Smav subdisk->sd_state == state) 752219974Smav n++; 753219974Smav } 754219974Smav return (n); 755219974Smav} 756219974Smav 757219974Smav/* 758219974Smav * Return the first subdisk in given state. 759219974Smav * If state is equal to -1, then the first connected disks. 760219974Smav */ 761219974Smavstruct g_raid_subdisk * 762219974Smavg_raid_get_subdisk(struct g_raid_volume *vol, int state) 763219974Smav{ 764219974Smav struct g_raid_subdisk *sd; 765219974Smav struct g_raid_softc *sc; 766219974Smav u_int i; 767219974Smav 768219974Smav sc = vol->v_softc; 769219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 770219974Smav 771219974Smav for (i = 0; i < vol->v_disks_count; i++) { 772219974Smav sd = &vol->v_subdisks[i]; 773219974Smav if ((state == -1 && 774219974Smav sd->sd_state != G_RAID_SUBDISK_S_NONE) || 775219974Smav sd->sd_state == state) 776219974Smav return (sd); 777219974Smav } 778219974Smav return (NULL); 779219974Smav} 780219974Smav 781219974Smavstruct g_consumer * 782219974Smavg_raid_open_consumer(struct g_raid_softc *sc, const char *name) 783219974Smav{ 784219974Smav struct g_consumer *cp; 785219974Smav struct g_provider *pp; 786219974Smav 787219974Smav g_topology_assert(); 788219974Smav 789219974Smav if (strncmp(name, "/dev/", 5) == 0) 790219974Smav name += 5; 791219974Smav pp = g_provider_by_name(name); 792219974Smav if (pp == NULL) 793219974Smav return (NULL); 794219974Smav cp = g_new_consumer(sc->sc_geom); 795219974Smav if (g_attach(cp, pp) != 0) { 796219974Smav g_destroy_consumer(cp); 797219974Smav return (NULL); 798219974Smav } 799219974Smav if (g_access(cp, 1, 1, 1) != 0) { 800219974Smav g_detach(cp); 801219974Smav g_destroy_consumer(cp); 802219974Smav return (NULL); 803219974Smav } 804219974Smav return (cp); 805219974Smav} 806219974Smav 807219974Smavstatic u_int 808219974Smavg_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) 809219974Smav{ 810219974Smav struct bio *bp; 811219974Smav u_int nreqs = 0; 812219974Smav 813219974Smav mtx_lock(&sc->sc_queue_mtx); 814219974Smav TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 815219974Smav if (bp->bio_from == cp) 816219974Smav nreqs++; 817219974Smav } 818219974Smav mtx_unlock(&sc->sc_queue_mtx); 819219974Smav return (nreqs); 820219974Smav} 821219974Smav 822219974Smavu_int 823219974Smavg_raid_nopens(struct g_raid_softc *sc) 824219974Smav{ 825219974Smav struct g_raid_volume *vol; 826219974Smav u_int opens; 827219974Smav 828219974Smav opens = 0; 829219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 830219974Smav if (vol->v_provider_open != 0) 831219974Smav opens++; 832219974Smav } 833219974Smav return (opens); 834219974Smav} 835219974Smav 836219974Smavstatic int 837219974Smavg_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) 838219974Smav{ 839219974Smav 840219974Smav if (cp->index > 0) { 841219974Smav G_RAID_DEBUG1(2, sc, 842219974Smav "I/O requests for %s exist, can't destroy it now.", 843219974Smav cp->provider->name); 844219974Smav return (1); 845219974Smav } 846219974Smav if (g_raid_nrequests(sc, cp) > 0) { 847219974Smav G_RAID_DEBUG1(2, sc, 848219974Smav "I/O requests for %s in queue, can't destroy it now.", 849219974Smav cp->provider->name); 850219974Smav return (1); 851219974Smav } 852219974Smav return (0); 853219974Smav} 854219974Smav 855219974Smavstatic void 856219974Smavg_raid_destroy_consumer(void *arg, int flags __unused) 857219974Smav{ 858219974Smav struct g_consumer *cp; 859219974Smav 860219974Smav g_topology_assert(); 861219974Smav 862219974Smav cp = arg; 863219974Smav G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 864219974Smav g_detach(cp); 865219974Smav g_destroy_consumer(cp); 866219974Smav} 867219974Smav 868219974Smavvoid 869219974Smavg_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) 870219974Smav{ 871219974Smav struct g_provider *pp; 872219974Smav int retaste_wait; 873219974Smav 874219974Smav g_topology_assert_not(); 875219974Smav 876219974Smav g_topology_lock(); 877219974Smav cp->private = NULL; 878219974Smav if (g_raid_consumer_is_busy(sc, cp)) 879219974Smav goto out; 880219974Smav pp = cp->provider; 881219974Smav retaste_wait = 0; 882219974Smav if (cp->acw == 1) { 883219974Smav if ((pp->geom->flags & G_GEOM_WITHER) == 0) 884219974Smav retaste_wait = 1; 885219974Smav } 886219974Smav if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 887219974Smav g_access(cp, -cp->acr, -cp->acw, -cp->ace); 888219974Smav if (retaste_wait) { 889219974Smav /* 890219974Smav * After retaste event was send (inside g_access()), we can send 891219974Smav * event to detach and destroy consumer. 892219974Smav * A class, which has consumer to the given provider connected 893219974Smav * will not receive retaste event for the provider. 894219974Smav * This is the way how I ignore retaste events when I close 895219974Smav * consumers opened for write: I detach and destroy consumer 896219974Smav * after retaste event is sent. 897219974Smav */ 898219974Smav g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); 899219974Smav goto out; 900219974Smav } 901219974Smav G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); 902219974Smav g_detach(cp); 903219974Smav g_destroy_consumer(cp); 904219974Smavout: 905219974Smav g_topology_unlock(); 906219974Smav} 907219974Smav 908219974Smavstatic void 909219974Smavg_raid_orphan(struct g_consumer *cp) 910219974Smav{ 911219974Smav struct g_raid_disk *disk; 912219974Smav 913219974Smav g_topology_assert(); 914219974Smav 915219974Smav disk = cp->private; 916219974Smav if (disk == NULL) 917219974Smav return; 918219974Smav g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, 919219974Smav G_RAID_EVENT_DISK); 920219974Smav} 921219974Smav 922242314Smavstatic void 923219974Smavg_raid_clean(struct g_raid_volume *vol, int acw) 924219974Smav{ 925219974Smav struct g_raid_softc *sc; 926219974Smav int timeout; 927219974Smav 928219974Smav sc = vol->v_softc; 929219974Smav g_topology_assert_not(); 930219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 931219974Smav 932219974Smav// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 933242314Smav// return; 934219974Smav if (!vol->v_dirty) 935242314Smav return; 936219974Smav if (vol->v_writes > 0) 937242314Smav return; 938219974Smav if (acw > 0 || (acw == -1 && 939219974Smav vol->v_provider != NULL && vol->v_provider->acw > 0)) { 940219974Smav timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); 941242314Smav if (!g_raid_shutdown && timeout > 0) 942242314Smav return; 943219974Smav } 944219974Smav vol->v_dirty = 0; 945219974Smav G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", 946219974Smav vol->v_name); 947219974Smav g_raid_write_metadata(sc, vol, NULL, NULL); 948219974Smav} 949219974Smav 950219974Smavstatic void 951219974Smavg_raid_dirty(struct g_raid_volume *vol) 952219974Smav{ 953219974Smav struct g_raid_softc *sc; 954219974Smav 955219974Smav sc = vol->v_softc; 956219974Smav g_topology_assert_not(); 957219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 958219974Smav 959219974Smav// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 960219974Smav// return; 961219974Smav vol->v_dirty = 1; 962219974Smav G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", 963219974Smav vol->v_name); 964219974Smav g_raid_write_metadata(sc, vol, NULL, NULL); 965219974Smav} 966219974Smav 967219974Smavvoid 968219974Smavg_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) 969219974Smav{ 970219974Smav struct g_raid_softc *sc; 971219974Smav struct g_raid_volume *vol; 972219974Smav struct g_raid_subdisk *sd; 973219974Smav struct bio_queue_head queue; 974219974Smav struct bio *cbp; 975219974Smav int i; 976219974Smav 977219974Smav vol = tr->tro_volume; 978219974Smav sc = vol->v_softc; 979219974Smav 980219974Smav /* 981219974Smav * Allocate all bios before sending any request, so we can return 982219974Smav * ENOMEM in nice and clean way. 983219974Smav */ 984219974Smav bioq_init(&queue); 985219974Smav for (i = 0; i < vol->v_disks_count; i++) { 986219974Smav sd = &vol->v_subdisks[i]; 987219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_NONE || 988219974Smav sd->sd_state == G_RAID_SUBDISK_S_FAILED) 989219974Smav continue; 990219974Smav cbp = g_clone_bio(bp); 991219974Smav if (cbp == NULL) 992219974Smav goto failure; 993219974Smav cbp->bio_caller1 = sd; 994219974Smav bioq_insert_tail(&queue, cbp); 995219974Smav } 996219974Smav for (cbp = bioq_first(&queue); cbp != NULL; 997219974Smav cbp = bioq_first(&queue)) { 998219974Smav bioq_remove(&queue, cbp); 999219974Smav sd = cbp->bio_caller1; 1000219974Smav cbp->bio_caller1 = NULL; 1001219974Smav g_raid_subdisk_iostart(sd, cbp); 1002219974Smav } 1003219974Smav return; 1004219974Smavfailure: 1005219974Smav for (cbp = bioq_first(&queue); cbp != NULL; 1006219974Smav cbp = bioq_first(&queue)) { 1007219974Smav bioq_remove(&queue, cbp); 1008219974Smav g_destroy_bio(cbp); 1009219974Smav } 1010219974Smav if (bp->bio_error == 0) 1011219974Smav bp->bio_error = ENOMEM; 1012219974Smav g_raid_iodone(bp, bp->bio_error); 1013219974Smav} 1014219974Smav 1015219974Smavstatic void 1016219974Smavg_raid_tr_kerneldump_common_done(struct bio *bp) 1017219974Smav{ 1018219974Smav 1019219974Smav bp->bio_flags |= BIO_DONE; 1020219974Smav} 1021219974Smav 1022219974Smavint 1023219974Smavg_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, 1024219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1025219974Smav{ 1026219974Smav struct g_raid_softc *sc; 1027219974Smav struct g_raid_volume *vol; 1028219974Smav struct bio bp; 1029219974Smav 1030219974Smav vol = tr->tro_volume; 1031219974Smav sc = vol->v_softc; 1032219974Smav 1033219974Smav bzero(&bp, sizeof(bp)); 1034219974Smav bp.bio_cmd = BIO_WRITE; 1035219974Smav bp.bio_done = g_raid_tr_kerneldump_common_done; 1036219974Smav bp.bio_attribute = NULL; 1037219974Smav bp.bio_offset = offset; 1038219974Smav bp.bio_length = length; 1039219974Smav bp.bio_data = virtual; 1040219974Smav bp.bio_to = vol->v_provider; 1041219974Smav 1042219974Smav g_raid_start(&bp); 1043219974Smav while (!(bp.bio_flags & BIO_DONE)) { 1044219974Smav G_RAID_DEBUG1(4, sc, "Poll..."); 1045219974Smav g_raid_poll(sc); 1046219974Smav DELAY(10); 1047219974Smav } 1048219974Smav 1049219974Smav return (bp.bio_error != 0 ? EIO : 0); 1050219974Smav} 1051219974Smav 1052219974Smavstatic int 1053219974Smavg_raid_dump(void *arg, 1054219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1055219974Smav{ 1056219974Smav struct g_raid_volume *vol; 1057219974Smav int error; 1058219974Smav 1059219974Smav vol = (struct g_raid_volume *)arg; 1060219974Smav G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", 1061219974Smav (long long unsigned)offset, (long long unsigned)length); 1062219974Smav 1063219974Smav error = G_RAID_TR_KERNELDUMP(vol->v_tr, 1064219974Smav virtual, physical, offset, length); 1065219974Smav return (error); 1066219974Smav} 1067219974Smav 1068219974Smavstatic void 1069219974Smavg_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) 1070219974Smav{ 1071219974Smav struct g_kerneldump *gkd; 1072219974Smav struct g_provider *pp; 1073219974Smav struct g_raid_volume *vol; 1074219974Smav 1075219974Smav gkd = (struct g_kerneldump*)bp->bio_data; 1076219974Smav pp = bp->bio_to; 1077219974Smav vol = pp->private; 1078219974Smav g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", 1079219974Smav pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 1080219974Smav gkd->di.dumper = g_raid_dump; 1081219974Smav gkd->di.priv = vol; 1082219974Smav gkd->di.blocksize = vol->v_sectorsize; 1083219974Smav gkd->di.maxiosize = DFLTPHYS; 1084219974Smav gkd->di.mediaoffset = gkd->offset; 1085219974Smav if ((gkd->offset + gkd->length) > vol->v_mediasize) 1086219974Smav gkd->length = vol->v_mediasize - gkd->offset; 1087219974Smav gkd->di.mediasize = gkd->length; 1088219974Smav g_io_deliver(bp, 0); 1089219974Smav} 1090219974Smav 1091219974Smavstatic void 1092242323Smavg_raid_candelete(struct g_raid_softc *sc, struct bio *bp) 1093242323Smav{ 1094242323Smav struct g_provider *pp; 1095242323Smav struct g_raid_volume *vol; 1096242323Smav struct g_raid_subdisk *sd; 1097242323Smav int *val; 1098242323Smav int i; 1099242323Smav 1100242323Smav val = (int *)bp->bio_data; 1101242323Smav pp = bp->bio_to; 1102242323Smav vol = pp->private; 1103242323Smav *val = 0; 1104242323Smav for (i = 0; i < vol->v_disks_count; i++) { 1105242323Smav sd = &vol->v_subdisks[i]; 1106242323Smav if (sd->sd_state == G_RAID_SUBDISK_S_NONE) 1107242323Smav continue; 1108242323Smav if (sd->sd_disk->d_candelete) { 1109242323Smav *val = 1; 1110242323Smav break; 1111242323Smav } 1112242323Smav } 1113242323Smav g_io_deliver(bp, 0); 1114242323Smav} 1115242323Smav 1116242323Smavstatic void 1117219974Smavg_raid_start(struct bio *bp) 1118219974Smav{ 1119219974Smav struct g_raid_softc *sc; 1120219974Smav 1121219974Smav sc = bp->bio_to->geom->softc; 1122219974Smav /* 1123219974Smav * If sc == NULL or there are no valid disks, provider's error 1124219974Smav * should be set and g_raid_start() should not be called at all. 1125219974Smav */ 1126219974Smav// KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, 1127219974Smav// ("Provider's error should be set (error=%d)(mirror=%s).", 1128219974Smav// bp->bio_to->error, bp->bio_to->name)); 1129219974Smav G_RAID_LOGREQ(3, bp, "Request received."); 1130219974Smav 1131219974Smav switch (bp->bio_cmd) { 1132219974Smav case BIO_READ: 1133219974Smav case BIO_WRITE: 1134219974Smav case BIO_DELETE: 1135219974Smav case BIO_FLUSH: 1136219974Smav break; 1137219974Smav case BIO_GETATTR: 1138242323Smav if (!strcmp(bp->bio_attribute, "GEOM::candelete")) 1139242323Smav g_raid_candelete(sc, bp); 1140242323Smav else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 1141219974Smav g_raid_kerneldump(sc, bp); 1142219974Smav else 1143219974Smav g_io_deliver(bp, EOPNOTSUPP); 1144219974Smav return; 1145219974Smav default: 1146219974Smav g_io_deliver(bp, EOPNOTSUPP); 1147219974Smav return; 1148219974Smav } 1149219974Smav mtx_lock(&sc->sc_queue_mtx); 1150219974Smav bioq_disksort(&sc->sc_queue, bp); 1151219974Smav mtx_unlock(&sc->sc_queue_mtx); 1152219974Smav if (!dumping) { 1153219974Smav G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); 1154219974Smav wakeup(sc); 1155219974Smav } 1156219974Smav} 1157219974Smav 1158219974Smavstatic int 1159219974Smavg_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) 1160219974Smav{ 1161219974Smav /* 1162219974Smav * 5 cases: 1163219974Smav * (1) bp entirely below NO 1164219974Smav * (2) bp entirely above NO 1165219974Smav * (3) bp start below, but end in range YES 1166219974Smav * (4) bp entirely within YES 1167219974Smav * (5) bp starts within, ends above YES 1168219974Smav * 1169219974Smav * lock range 10-19 (offset 10 length 10) 1170219974Smav * (1) 1-5: first if kicks it out 1171219974Smav * (2) 30-35: second if kicks it out 1172219974Smav * (3) 5-15: passes both ifs 1173219974Smav * (4) 12-14: passes both ifs 1174219974Smav * (5) 19-20: passes both 1175219974Smav */ 1176219974Smav off_t lend = lstart + len - 1; 1177219974Smav off_t bstart = bp->bio_offset; 1178219974Smav off_t bend = bp->bio_offset + bp->bio_length - 1; 1179219974Smav 1180219974Smav if (bend < lstart) 1181219974Smav return (0); 1182219974Smav if (lend < bstart) 1183219974Smav return (0); 1184219974Smav return (1); 1185219974Smav} 1186219974Smav 1187219974Smavstatic int 1188219974Smavg_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) 1189219974Smav{ 1190219974Smav struct g_raid_lock *lp; 1191219974Smav 1192219974Smav sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); 1193219974Smav 1194219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1195219974Smav if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) 1196219974Smav return (1); 1197219974Smav } 1198219974Smav return (0); 1199219974Smav} 1200219974Smav 1201219974Smavstatic void 1202219974Smavg_raid_start_request(struct bio *bp) 1203219974Smav{ 1204219974Smav struct g_raid_softc *sc; 1205219974Smav struct g_raid_volume *vol; 1206219974Smav 1207219974Smav sc = bp->bio_to->geom->softc; 1208219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1209219974Smav vol = bp->bio_to->private; 1210219974Smav 1211219974Smav /* 1212219974Smav * Check to see if this item is in a locked range. If so, 1213219974Smav * queue it to our locked queue and return. We'll requeue 1214219974Smav * it when the range is unlocked. Internal I/O for the 1215219974Smav * rebuild/rescan/recovery process is excluded from this 1216219974Smav * check so we can actually do the recovery. 1217219974Smav */ 1218219974Smav if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && 1219219974Smav g_raid_is_in_locked_range(vol, bp)) { 1220219974Smav G_RAID_LOGREQ(3, bp, "Defer request."); 1221219974Smav bioq_insert_tail(&vol->v_locked, bp); 1222219974Smav return; 1223219974Smav } 1224219974Smav 1225219974Smav /* 1226219974Smav * If we're actually going to do the write/delete, then 1227219974Smav * update the idle stats for the volume. 1228219974Smav */ 1229219974Smav if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1230219974Smav if (!vol->v_dirty) 1231219974Smav g_raid_dirty(vol); 1232219974Smav vol->v_writes++; 1233219974Smav } 1234219974Smav 1235219974Smav /* 1236219974Smav * Put request onto inflight queue, so we can check if new 1237219974Smav * synchronization requests don't collide with it. Then tell 1238219974Smav * the transformation layer to start the I/O. 1239219974Smav */ 1240219974Smav bioq_insert_tail(&vol->v_inflight, bp); 1241219974Smav G_RAID_LOGREQ(4, bp, "Request started"); 1242219974Smav G_RAID_TR_IOSTART(vol->v_tr, bp); 1243219974Smav} 1244219974Smav 1245219974Smavstatic void 1246219974Smavg_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) 1247219974Smav{ 1248219974Smav off_t off, len; 1249219974Smav struct bio *nbp; 1250219974Smav struct g_raid_lock *lp; 1251219974Smav 1252219974Smav vol->v_pending_lock = 0; 1253219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1254219974Smav if (lp->l_pending) { 1255219974Smav off = lp->l_offset; 1256219974Smav len = lp->l_length; 1257219974Smav lp->l_pending = 0; 1258219974Smav TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { 1259219974Smav if (g_raid_bio_overlaps(nbp, off, len)) 1260219974Smav lp->l_pending++; 1261219974Smav } 1262219974Smav if (lp->l_pending) { 1263219974Smav vol->v_pending_lock = 1; 1264219974Smav G_RAID_DEBUG1(4, vol->v_softc, 1265219974Smav "Deferred lock(%jd, %jd) has %d pending", 1266219974Smav (intmax_t)off, (intmax_t)(off + len), 1267219974Smav lp->l_pending); 1268219974Smav continue; 1269219974Smav } 1270219974Smav G_RAID_DEBUG1(4, vol->v_softc, 1271219974Smav "Deferred lock of %jd to %jd completed", 1272219974Smav (intmax_t)off, (intmax_t)(off + len)); 1273219974Smav G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1274219974Smav } 1275219974Smav } 1276219974Smav} 1277219974Smav 1278219974Smavvoid 1279219974Smavg_raid_iodone(struct bio *bp, int error) 1280219974Smav{ 1281219974Smav struct g_raid_softc *sc; 1282219974Smav struct g_raid_volume *vol; 1283219974Smav 1284219974Smav sc = bp->bio_to->geom->softc; 1285219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1286219974Smav vol = bp->bio_to->private; 1287219974Smav G_RAID_LOGREQ(3, bp, "Request done: %d.", error); 1288219974Smav 1289219974Smav /* Update stats if we done write/delete. */ 1290219974Smav if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1291219974Smav vol->v_writes--; 1292219974Smav vol->v_last_write = time_uptime; 1293219974Smav } 1294219974Smav 1295219974Smav bioq_remove(&vol->v_inflight, bp); 1296219974Smav if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) 1297219974Smav g_raid_finish_with_locked_ranges(vol, bp); 1298219974Smav getmicrouptime(&vol->v_last_done); 1299219974Smav g_io_deliver(bp, error); 1300219974Smav} 1301219974Smav 1302219974Smavint 1303219974Smavg_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, 1304219974Smav struct bio *ignore, void *argp) 1305219974Smav{ 1306219974Smav struct g_raid_softc *sc; 1307219974Smav struct g_raid_lock *lp; 1308219974Smav struct bio *bp; 1309219974Smav 1310219974Smav sc = vol->v_softc; 1311219974Smav lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); 1312219974Smav LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); 1313219974Smav lp->l_offset = off; 1314219974Smav lp->l_length = len; 1315219974Smav lp->l_callback_arg = argp; 1316219974Smav 1317219974Smav lp->l_pending = 0; 1318219974Smav TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { 1319219974Smav if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) 1320219974Smav lp->l_pending++; 1321219974Smav } 1322219974Smav 1323219974Smav /* 1324219974Smav * If there are any writes that are pending, we return EBUSY. All 1325219974Smav * callers will have to wait until all pending writes clear. 1326219974Smav */ 1327219974Smav if (lp->l_pending > 0) { 1328219974Smav vol->v_pending_lock = 1; 1329219974Smav G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", 1330219974Smav (intmax_t)off, (intmax_t)(off+len), lp->l_pending); 1331219974Smav return (EBUSY); 1332219974Smav } 1333219974Smav G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", 1334219974Smav (intmax_t)off, (intmax_t)(off+len)); 1335219974Smav G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1336219974Smav return (0); 1337219974Smav} 1338219974Smav 1339219974Smavint 1340219974Smavg_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) 1341219974Smav{ 1342219974Smav struct g_raid_lock *lp; 1343219974Smav struct g_raid_softc *sc; 1344219974Smav struct bio *bp; 1345219974Smav 1346219974Smav sc = vol->v_softc; 1347219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1348219974Smav if (lp->l_offset == off && lp->l_length == len) { 1349219974Smav LIST_REMOVE(lp, l_next); 1350219974Smav /* XXX 1351219974Smav * Right now we just put them all back on the queue 1352219974Smav * and hope for the best. We hope this because any 1353219974Smav * locked ranges will go right back on this list 1354219974Smav * when the worker thread runs. 1355219974Smav * XXX 1356219974Smav */ 1357219974Smav G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", 1358219974Smav (intmax_t)lp->l_offset, 1359219974Smav (intmax_t)(lp->l_offset+lp->l_length)); 1360219974Smav mtx_lock(&sc->sc_queue_mtx); 1361219974Smav while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) 1362219974Smav bioq_disksort(&sc->sc_queue, bp); 1363219974Smav mtx_unlock(&sc->sc_queue_mtx); 1364219974Smav free(lp, M_RAID); 1365219974Smav return (0); 1366219974Smav } 1367219974Smav } 1368219974Smav return (EINVAL); 1369219974Smav} 1370219974Smav 1371219974Smavvoid 1372219974Smavg_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) 1373219974Smav{ 1374219974Smav struct g_consumer *cp; 1375219974Smav struct g_raid_disk *disk, *tdisk; 1376219974Smav 1377219974Smav bp->bio_caller1 = sd; 1378219974Smav 1379219974Smav /* 1380219974Smav * Make sure that the disk is present. Generally it is a task of 1381219974Smav * transformation layers to not send requests to absent disks, but 1382219974Smav * it is better to be safe and report situation then sorry. 1383219974Smav */ 1384219974Smav if (sd->sd_disk == NULL) { 1385219974Smav G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); 1386219974Smavnodisk: 1387219974Smav bp->bio_from = NULL; 1388219974Smav bp->bio_to = NULL; 1389219974Smav bp->bio_error = ENXIO; 1390219974Smav g_raid_disk_done(bp); 1391219974Smav return; 1392219974Smav } 1393219974Smav disk = sd->sd_disk; 1394219974Smav if (disk->d_state != G_RAID_DISK_S_ACTIVE && 1395219974Smav disk->d_state != G_RAID_DISK_S_FAILED) { 1396219974Smav G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " 1397219974Smav "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 1398219974Smav goto nodisk; 1399219974Smav } 1400219974Smav 1401219974Smav cp = disk->d_consumer; 1402219974Smav bp->bio_from = cp; 1403219974Smav bp->bio_to = cp->provider; 1404219974Smav cp->index++; 1405219974Smav 1406219974Smav /* Update average disks load. */ 1407219974Smav TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { 1408219974Smav if (tdisk->d_consumer == NULL) 1409219974Smav tdisk->d_load = 0; 1410219974Smav else 1411219974Smav tdisk->d_load = (tdisk->d_consumer->index * 1412219974Smav G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; 1413219974Smav } 1414219974Smav 1415219974Smav disk->d_last_offset = bp->bio_offset + bp->bio_length; 1416219974Smav if (dumping) { 1417219974Smav G_RAID_LOGREQ(3, bp, "Sending dumping request."); 1418219974Smav if (bp->bio_cmd == BIO_WRITE) { 1419219974Smav bp->bio_error = g_raid_subdisk_kerneldump(sd, 1420219974Smav bp->bio_data, 0, bp->bio_offset, bp->bio_length); 1421219974Smav } else 1422219974Smav bp->bio_error = EOPNOTSUPP; 1423219974Smav g_raid_disk_done(bp); 1424219974Smav } else { 1425219974Smav bp->bio_done = g_raid_disk_done; 1426219974Smav bp->bio_offset += sd->sd_offset; 1427219974Smav G_RAID_LOGREQ(3, bp, "Sending request."); 1428219974Smav g_io_request(bp, cp); 1429219974Smav } 1430219974Smav} 1431219974Smav 1432219974Smavint 1433219974Smavg_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, 1434219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1435219974Smav{ 1436219974Smav 1437219974Smav if (sd->sd_disk == NULL) 1438219974Smav return (ENXIO); 1439219974Smav if (sd->sd_disk->d_kd.di.dumper == NULL) 1440219974Smav return (EOPNOTSUPP); 1441219974Smav return (dump_write(&sd->sd_disk->d_kd.di, 1442219974Smav virtual, physical, 1443219974Smav sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, 1444219974Smav length)); 1445219974Smav} 1446219974Smav 1447219974Smavstatic void 1448219974Smavg_raid_disk_done(struct bio *bp) 1449219974Smav{ 1450219974Smav struct g_raid_softc *sc; 1451219974Smav struct g_raid_subdisk *sd; 1452219974Smav 1453219974Smav sd = bp->bio_caller1; 1454219974Smav sc = sd->sd_softc; 1455219974Smav mtx_lock(&sc->sc_queue_mtx); 1456219974Smav bioq_disksort(&sc->sc_queue, bp); 1457219974Smav mtx_unlock(&sc->sc_queue_mtx); 1458219974Smav if (!dumping) 1459219974Smav wakeup(sc); 1460219974Smav} 1461219974Smav 1462219974Smavstatic void 1463219974Smavg_raid_disk_done_request(struct bio *bp) 1464219974Smav{ 1465219974Smav struct g_raid_softc *sc; 1466219974Smav struct g_raid_disk *disk; 1467219974Smav struct g_raid_subdisk *sd; 1468219974Smav struct g_raid_volume *vol; 1469219974Smav 1470219974Smav g_topology_assert_not(); 1471219974Smav 1472219974Smav G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); 1473219974Smav sd = bp->bio_caller1; 1474219974Smav sc = sd->sd_softc; 1475219974Smav vol = sd->sd_volume; 1476219974Smav if (bp->bio_from != NULL) { 1477219974Smav bp->bio_from->index--; 1478219974Smav disk = bp->bio_from->private; 1479219974Smav if (disk == NULL) 1480219974Smav g_raid_kill_consumer(sc, bp->bio_from); 1481219974Smav } 1482219974Smav bp->bio_offset -= sd->sd_offset; 1483219974Smav 1484219974Smav G_RAID_TR_IODONE(vol->v_tr, sd, bp); 1485219974Smav} 1486219974Smav 1487219974Smavstatic void 1488219974Smavg_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) 1489219974Smav{ 1490219974Smav 1491219974Smav if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) 1492219974Smav ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); 1493219974Smav else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) 1494219974Smav ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); 1495219974Smav else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) 1496219974Smav ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); 1497219974Smav else 1498219974Smav ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); 1499219974Smav if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { 1500219974Smav KASSERT(ep->e_error == 0, 1501219974Smav ("Error cannot be handled.")); 1502219974Smav g_raid_event_free(ep); 1503219974Smav } else { 1504219974Smav ep->e_flags |= G_RAID_EVENT_DONE; 1505219974Smav G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); 1506219974Smav mtx_lock(&sc->sc_queue_mtx); 1507219974Smav wakeup(ep); 1508219974Smav mtx_unlock(&sc->sc_queue_mtx); 1509219974Smav } 1510219974Smav} 1511219974Smav 1512219974Smav/* 1513219974Smav * Worker thread. 1514219974Smav */ 1515219974Smavstatic void 1516219974Smavg_raid_worker(void *arg) 1517219974Smav{ 1518219974Smav struct g_raid_softc *sc; 1519219974Smav struct g_raid_event *ep; 1520219974Smav struct g_raid_volume *vol; 1521219974Smav struct bio *bp; 1522219974Smav struct timeval now, t; 1523219974Smav int timeout, rv; 1524219974Smav 1525219974Smav sc = arg; 1526219974Smav thread_lock(curthread); 1527219974Smav sched_prio(curthread, PRIBIO); 1528219974Smav thread_unlock(curthread); 1529219974Smav 1530219974Smav sx_xlock(&sc->sc_lock); 1531219974Smav for (;;) { 1532219974Smav mtx_lock(&sc->sc_queue_mtx); 1533219974Smav /* 1534219974Smav * First take a look at events. 1535219974Smav * This is important to handle events before any I/O requests. 1536219974Smav */ 1537219974Smav bp = NULL; 1538219974Smav vol = NULL; 1539219974Smav rv = 0; 1540219974Smav ep = TAILQ_FIRST(&sc->sc_events); 1541219974Smav if (ep != NULL) 1542219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1543219974Smav else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) 1544219974Smav ; 1545219974Smav else { 1546219974Smav getmicrouptime(&now); 1547219974Smav t = now; 1548219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1549219974Smav if (bioq_first(&vol->v_inflight) == NULL && 1550219974Smav vol->v_tr && 1551219974Smav timevalcmp(&vol->v_last_done, &t, < )) 1552219974Smav t = vol->v_last_done; 1553219974Smav } 1554219974Smav timevalsub(&t, &now); 1555219974Smav timeout = g_raid_idle_threshold + 1556219974Smav t.tv_sec * 1000000 + t.tv_usec; 1557219974Smav if (timeout > 0) { 1558219974Smav /* 1559219974Smav * Two steps to avoid overflows at HZ=1000 1560219974Smav * and idle timeouts > 2.1s. Some rounding 1561219974Smav * errors can occur, but they are < 1tick, 1562219974Smav * which is deemed to be close enough for 1563219974Smav * this purpose. 1564219974Smav */ 1565219974Smav int micpertic = 1000000 / hz; 1566219974Smav timeout = (timeout + micpertic - 1) / micpertic; 1567219974Smav sx_xunlock(&sc->sc_lock); 1568219974Smav MSLEEP(rv, sc, &sc->sc_queue_mtx, 1569219974Smav PRIBIO | PDROP, "-", timeout); 1570219974Smav sx_xlock(&sc->sc_lock); 1571219974Smav goto process; 1572219974Smav } else 1573219974Smav rv = EWOULDBLOCK; 1574219974Smav } 1575219974Smav mtx_unlock(&sc->sc_queue_mtx); 1576219974Smavprocess: 1577219974Smav if (ep != NULL) { 1578219974Smav g_raid_handle_event(sc, ep); 1579219974Smav } else if (bp != NULL) { 1580219974Smav if (bp->bio_to != NULL && 1581219974Smav bp->bio_to->geom == sc->sc_geom) 1582219974Smav g_raid_start_request(bp); 1583219974Smav else 1584219974Smav g_raid_disk_done_request(bp); 1585219974Smav } else if (rv == EWOULDBLOCK) { 1586219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1587242314Smav g_raid_clean(vol, -1); 1588219974Smav if (bioq_first(&vol->v_inflight) == NULL && 1589219974Smav vol->v_tr) { 1590219974Smav t.tv_sec = g_raid_idle_threshold / 1000000; 1591219974Smav t.tv_usec = g_raid_idle_threshold % 1000000; 1592219974Smav timevaladd(&t, &vol->v_last_done); 1593219974Smav getmicrouptime(&now); 1594219974Smav if (timevalcmp(&t, &now, <= )) { 1595219974Smav G_RAID_TR_IDLE(vol->v_tr); 1596219974Smav vol->v_last_done = now; 1597219974Smav } 1598219974Smav } 1599219974Smav } 1600219974Smav } 1601219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) 1602219974Smav g_raid_destroy_node(sc, 1); /* May not return. */ 1603219974Smav } 1604219974Smav} 1605219974Smav 1606219974Smavstatic void 1607219974Smavg_raid_poll(struct g_raid_softc *sc) 1608219974Smav{ 1609219974Smav struct g_raid_event *ep; 1610219974Smav struct bio *bp; 1611219974Smav 1612219974Smav sx_xlock(&sc->sc_lock); 1613219974Smav mtx_lock(&sc->sc_queue_mtx); 1614219974Smav /* 1615219974Smav * First take a look at events. 1616219974Smav * This is important to handle events before any I/O requests. 1617219974Smav */ 1618219974Smav ep = TAILQ_FIRST(&sc->sc_events); 1619219974Smav if (ep != NULL) { 1620219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1621219974Smav mtx_unlock(&sc->sc_queue_mtx); 1622219974Smav g_raid_handle_event(sc, ep); 1623219974Smav goto out; 1624219974Smav } 1625219974Smav bp = bioq_takefirst(&sc->sc_queue); 1626219974Smav if (bp != NULL) { 1627219974Smav mtx_unlock(&sc->sc_queue_mtx); 1628219974Smav if (bp->bio_from == NULL || 1629219974Smav bp->bio_from->geom != sc->sc_geom) 1630219974Smav g_raid_start_request(bp); 1631219974Smav else 1632219974Smav g_raid_disk_done_request(bp); 1633219974Smav } 1634219974Smavout: 1635219974Smav sx_xunlock(&sc->sc_lock); 1636219974Smav} 1637219974Smav 1638219974Smavstatic void 1639219974Smavg_raid_launch_provider(struct g_raid_volume *vol) 1640219974Smav{ 1641219974Smav struct g_raid_disk *disk; 1642219974Smav struct g_raid_softc *sc; 1643219974Smav struct g_provider *pp; 1644219974Smav char name[G_RAID_MAX_VOLUMENAME]; 1645248068Ssbruno char announce_buf[80], buf1[32]; 1646219974Smav off_t off; 1647219974Smav 1648219974Smav sc = vol->v_softc; 1649219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1650219974Smav 1651219974Smav g_topology_lock(); 1652219974Smav /* Try to name provider with volume name. */ 1653219974Smav snprintf(name, sizeof(name), "raid/%s", vol->v_name); 1654219974Smav if (g_raid_name_format == 0 || vol->v_name[0] == 0 || 1655219974Smav g_provider_by_name(name) != NULL) { 1656219974Smav /* Otherwise use sequential volume number. */ 1657219974Smav snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); 1658219974Smav } 1659248068Ssbruno 1660248068Ssbruno /* 1661248068Ssbruno * Create a /dev/ar%d that the old ataraid(4) stack once 1662248068Ssbruno * created as an alias for /dev/raid/r%d if requested. 1663248068Ssbruno * This helps going from stable/7 ataraid devices to newer 1664248068Ssbruno * FreeBSD releases. sbruno 07 MAY 2013 1665248068Ssbruno */ 1666248068Ssbruno 1667248068Ssbruno if (ar_legacy_aliases) { 1668248068Ssbruno snprintf(announce_buf, sizeof(announce_buf), 1669248068Ssbruno "kern.devalias.%s", name); 1670248068Ssbruno snprintf(buf1, sizeof(buf1), 1671248068Ssbruno "ar%d", vol->v_global_id); 1672248068Ssbruno setenv(announce_buf, buf1); 1673248068Ssbruno } 1674248068Ssbruno 1675219974Smav pp = g_new_providerf(sc->sc_geom, "%s", name); 1676219974Smav pp->private = vol; 1677219974Smav pp->mediasize = vol->v_mediasize; 1678219974Smav pp->sectorsize = vol->v_sectorsize; 1679219974Smav pp->stripesize = 0; 1680219974Smav pp->stripeoffset = 0; 1681219974Smav if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 1682219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || 1683219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || 1684219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { 1685219974Smav if ((disk = vol->v_subdisks[0].sd_disk) != NULL && 1686219974Smav disk->d_consumer != NULL && 1687219974Smav disk->d_consumer->provider != NULL) { 1688219974Smav pp->stripesize = disk->d_consumer->provider->stripesize; 1689219974Smav off = disk->d_consumer->provider->stripeoffset; 1690219974Smav pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; 1691219974Smav if (off > 0) 1692219974Smav pp->stripeoffset %= off; 1693219974Smav } 1694219974Smav if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { 1695219974Smav pp->stripesize *= (vol->v_disks_count - 1); 1696219974Smav pp->stripeoffset *= (vol->v_disks_count - 1); 1697219974Smav } 1698219974Smav } else 1699219974Smav pp->stripesize = vol->v_strip_size; 1700219974Smav vol->v_provider = pp; 1701219974Smav g_error_provider(pp, 0); 1702219974Smav g_topology_unlock(); 1703219974Smav G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", 1704219974Smav pp->name, vol->v_name); 1705219974Smav} 1706219974Smav 1707219974Smavstatic void 1708219974Smavg_raid_destroy_provider(struct g_raid_volume *vol) 1709219974Smav{ 1710219974Smav struct g_raid_softc *sc; 1711219974Smav struct g_provider *pp; 1712219974Smav struct bio *bp, *tmp; 1713219974Smav 1714219974Smav g_topology_assert_not(); 1715219974Smav sc = vol->v_softc; 1716219974Smav pp = vol->v_provider; 1717219974Smav KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); 1718219974Smav 1719219974Smav g_topology_lock(); 1720219974Smav g_error_provider(pp, ENXIO); 1721219974Smav mtx_lock(&sc->sc_queue_mtx); 1722219974Smav TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { 1723219974Smav if (bp->bio_to != pp) 1724219974Smav continue; 1725219974Smav bioq_remove(&sc->sc_queue, bp); 1726219974Smav g_io_deliver(bp, ENXIO); 1727219974Smav } 1728219974Smav mtx_unlock(&sc->sc_queue_mtx); 1729219974Smav G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", 1730219974Smav pp->name, vol->v_name); 1731219974Smav g_wither_provider(pp, ENXIO); 1732219974Smav g_topology_unlock(); 1733219974Smav vol->v_provider = NULL; 1734219974Smav} 1735219974Smav 1736219974Smav/* 1737219974Smav * Update device state. 1738219974Smav */ 1739219974Smavstatic int 1740219974Smavg_raid_update_volume(struct g_raid_volume *vol, u_int event) 1741219974Smav{ 1742219974Smav struct g_raid_softc *sc; 1743219974Smav 1744219974Smav sc = vol->v_softc; 1745219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1746219974Smav 1747219974Smav G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", 1748219974Smav g_raid_volume_event2str(event), 1749219974Smav vol->v_name); 1750219974Smav switch (event) { 1751219974Smav case G_RAID_VOLUME_E_DOWN: 1752219974Smav if (vol->v_provider != NULL) 1753219974Smav g_raid_destroy_provider(vol); 1754219974Smav break; 1755219974Smav case G_RAID_VOLUME_E_UP: 1756219974Smav if (vol->v_provider == NULL) 1757219974Smav g_raid_launch_provider(vol); 1758219974Smav break; 1759219974Smav case G_RAID_VOLUME_E_START: 1760219974Smav if (vol->v_tr) 1761219974Smav G_RAID_TR_START(vol->v_tr); 1762219974Smav return (0); 1763219974Smav default: 1764219974Smav if (sc->sc_md) 1765219974Smav G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); 1766219974Smav return (0); 1767219974Smav } 1768219974Smav 1769219974Smav /* Manage root mount release. */ 1770219974Smav if (vol->v_starting) { 1771219974Smav vol->v_starting = 0; 1772219974Smav G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); 1773219974Smav root_mount_rel(vol->v_rootmount); 1774219974Smav vol->v_rootmount = NULL; 1775219974Smav } 1776219974Smav if (vol->v_stopping && vol->v_provider_open == 0) 1777219974Smav g_raid_destroy_volume(vol); 1778219974Smav return (0); 1779219974Smav} 1780219974Smav 1781219974Smav/* 1782219974Smav * Update subdisk state. 1783219974Smav */ 1784219974Smavstatic int 1785219974Smavg_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) 1786219974Smav{ 1787219974Smav struct g_raid_softc *sc; 1788219974Smav struct g_raid_volume *vol; 1789219974Smav 1790219974Smav sc = sd->sd_softc; 1791219974Smav vol = sd->sd_volume; 1792219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1793219974Smav 1794219974Smav G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", 1795219974Smav g_raid_subdisk_event2str(event), 1796219974Smav vol->v_name, sd->sd_pos, 1797219974Smav sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); 1798219974Smav if (vol->v_tr) 1799219974Smav G_RAID_TR_EVENT(vol->v_tr, sd, event); 1800219974Smav 1801219974Smav return (0); 1802219974Smav} 1803219974Smav 1804219974Smav/* 1805219974Smav * Update disk state. 1806219974Smav */ 1807219974Smavstatic int 1808219974Smavg_raid_update_disk(struct g_raid_disk *disk, u_int event) 1809219974Smav{ 1810219974Smav struct g_raid_softc *sc; 1811219974Smav 1812219974Smav sc = disk->d_softc; 1813219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1814219974Smav 1815219974Smav G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", 1816219974Smav g_raid_disk_event2str(event), 1817219974Smav g_raid_get_diskname(disk)); 1818219974Smav 1819219974Smav if (sc->sc_md) 1820219974Smav G_RAID_MD_EVENT(sc->sc_md, disk, event); 1821219974Smav return (0); 1822219974Smav} 1823219974Smav 1824219974Smav/* 1825219974Smav * Node event. 1826219974Smav */ 1827219974Smavstatic int 1828219974Smavg_raid_update_node(struct g_raid_softc *sc, u_int event) 1829219974Smav{ 1830219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1831219974Smav 1832219974Smav G_RAID_DEBUG1(2, sc, "Event %s for the array.", 1833219974Smav g_raid_node_event2str(event)); 1834219974Smav 1835219974Smav if (event == G_RAID_NODE_E_WAKE) 1836219974Smav return (0); 1837219974Smav if (sc->sc_md) 1838219974Smav G_RAID_MD_EVENT(sc->sc_md, NULL, event); 1839219974Smav return (0); 1840219974Smav} 1841219974Smav 1842219974Smavstatic int 1843219974Smavg_raid_access(struct g_provider *pp, int acr, int acw, int ace) 1844219974Smav{ 1845219974Smav struct g_raid_volume *vol; 1846219974Smav struct g_raid_softc *sc; 1847220210Smav int dcw, opens, error = 0; 1848219974Smav 1849219974Smav g_topology_assert(); 1850219974Smav sc = pp->geom->softc; 1851219974Smav vol = pp->private; 1852219974Smav KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 1853219974Smav KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); 1854219974Smav 1855219974Smav G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, 1856219974Smav acr, acw, ace); 1857219974Smav dcw = pp->acw + acw; 1858219974Smav 1859219974Smav g_topology_unlock(); 1860219974Smav sx_xlock(&sc->sc_lock); 1861219974Smav /* Deny new opens while dying. */ 1862219974Smav if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { 1863219974Smav error = ENXIO; 1864219974Smav goto out; 1865219974Smav } 1866242314Smav if (dcw == 0) 1867219974Smav g_raid_clean(vol, dcw); 1868219974Smav vol->v_provider_open += acr + acw + ace; 1869219974Smav /* Handle delayed node destruction. */ 1870219974Smav if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && 1871219974Smav vol->v_provider_open == 0) { 1872219974Smav /* Count open volumes. */ 1873219974Smav opens = g_raid_nopens(sc); 1874219974Smav if (opens == 0) { 1875219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 1876219974Smav /* Wake up worker to make it selfdestruct. */ 1877219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1878219974Smav } 1879219974Smav } 1880219974Smav /* Handle open volume destruction. */ 1881219974Smav if (vol->v_stopping && vol->v_provider_open == 0) 1882219974Smav g_raid_destroy_volume(vol); 1883219974Smavout: 1884219974Smav sx_xunlock(&sc->sc_lock); 1885219974Smav g_topology_lock(); 1886219974Smav return (error); 1887219974Smav} 1888219974Smav 1889219974Smavstruct g_raid_softc * 1890219974Smavg_raid_create_node(struct g_class *mp, 1891219974Smav const char *name, struct g_raid_md_object *md) 1892219974Smav{ 1893219974Smav struct g_raid_softc *sc; 1894219974Smav struct g_geom *gp; 1895219974Smav int error; 1896219974Smav 1897219974Smav g_topology_assert(); 1898219974Smav G_RAID_DEBUG(1, "Creating array %s.", name); 1899219974Smav 1900219974Smav gp = g_new_geomf(mp, "%s", name); 1901219974Smav sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); 1902219974Smav gp->start = g_raid_start; 1903219974Smav gp->orphan = g_raid_orphan; 1904219974Smav gp->access = g_raid_access; 1905219974Smav gp->dumpconf = g_raid_dumpconf; 1906219974Smav 1907219974Smav sc->sc_md = md; 1908219974Smav sc->sc_geom = gp; 1909219974Smav sc->sc_flags = 0; 1910219974Smav TAILQ_INIT(&sc->sc_volumes); 1911219974Smav TAILQ_INIT(&sc->sc_disks); 1912234816Smav sx_init(&sc->sc_lock, "graid:lock"); 1913234816Smav mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF); 1914219974Smav TAILQ_INIT(&sc->sc_events); 1915219974Smav bioq_init(&sc->sc_queue); 1916219974Smav gp->softc = sc; 1917219974Smav error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, 1918219974Smav "g_raid %s", name); 1919219974Smav if (error != 0) { 1920219974Smav G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); 1921219974Smav mtx_destroy(&sc->sc_queue_mtx); 1922219974Smav sx_destroy(&sc->sc_lock); 1923219974Smav g_destroy_geom(sc->sc_geom); 1924219974Smav free(sc, M_RAID); 1925219974Smav return (NULL); 1926219974Smav } 1927219974Smav 1928219974Smav G_RAID_DEBUG1(0, sc, "Array %s created.", name); 1929219974Smav return (sc); 1930219974Smav} 1931219974Smav 1932219974Smavstruct g_raid_volume * 1933219974Smavg_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) 1934219974Smav{ 1935219974Smav struct g_raid_volume *vol, *vol1; 1936219974Smav int i; 1937219974Smav 1938219974Smav G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); 1939219974Smav vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); 1940219974Smav vol->v_softc = sc; 1941219974Smav strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); 1942219974Smav vol->v_state = G_RAID_VOLUME_S_STARTING; 1943219974Smav vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 1944219974Smav vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; 1945235076Smav vol->v_rotate_parity = 1; 1946219974Smav bioq_init(&vol->v_inflight); 1947219974Smav bioq_init(&vol->v_locked); 1948219974Smav LIST_INIT(&vol->v_locks); 1949219974Smav for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 1950219974Smav vol->v_subdisks[i].sd_softc = sc; 1951219974Smav vol->v_subdisks[i].sd_volume = vol; 1952219974Smav vol->v_subdisks[i].sd_pos = i; 1953219974Smav vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; 1954219974Smav } 1955219974Smav 1956219974Smav /* Find free ID for this volume. */ 1957219974Smav g_topology_lock(); 1958219974Smav vol1 = vol; 1959219974Smav if (id >= 0) { 1960219974Smav LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1961219974Smav if (vol1->v_global_id == id) 1962219974Smav break; 1963219974Smav } 1964219974Smav } 1965219974Smav if (vol1 != NULL) { 1966219974Smav for (id = 0; ; id++) { 1967219974Smav LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1968219974Smav if (vol1->v_global_id == id) 1969219974Smav break; 1970219974Smav } 1971219974Smav if (vol1 == NULL) 1972219974Smav break; 1973219974Smav } 1974219974Smav } 1975219974Smav vol->v_global_id = id; 1976219974Smav LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); 1977219974Smav g_topology_unlock(); 1978219974Smav 1979219974Smav /* Delay root mounting. */ 1980219974Smav vol->v_rootmount = root_mount_hold("GRAID"); 1981219974Smav G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); 1982219974Smav vol->v_starting = 1; 1983219974Smav TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); 1984219974Smav return (vol); 1985219974Smav} 1986219974Smav 1987219974Smavstruct g_raid_disk * 1988219974Smavg_raid_create_disk(struct g_raid_softc *sc) 1989219974Smav{ 1990219974Smav struct g_raid_disk *disk; 1991219974Smav 1992219974Smav G_RAID_DEBUG1(1, sc, "Creating disk."); 1993219974Smav disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); 1994219974Smav disk->d_softc = sc; 1995219974Smav disk->d_state = G_RAID_DISK_S_NONE; 1996219974Smav TAILQ_INIT(&disk->d_subdisks); 1997219974Smav TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); 1998219974Smav return (disk); 1999219974Smav} 2000219974Smav 2001219974Smavint g_raid_start_volume(struct g_raid_volume *vol) 2002219974Smav{ 2003219974Smav struct g_raid_tr_class *class; 2004219974Smav struct g_raid_tr_object *obj; 2005219974Smav int status; 2006219974Smav 2007219974Smav G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); 2008219974Smav LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { 2009240465Smav if (!class->trc_enable) 2010240465Smav continue; 2011219974Smav G_RAID_DEBUG1(2, vol->v_softc, 2012219974Smav "Tasting volume %s for %s transformation.", 2013219974Smav vol->v_name, class->name); 2014219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2015219974Smav M_WAITOK); 2016219974Smav obj->tro_class = class; 2017219974Smav obj->tro_volume = vol; 2018219974Smav status = G_RAID_TR_TASTE(obj, vol); 2019219974Smav if (status != G_RAID_TR_TASTE_FAIL) 2020219974Smav break; 2021219974Smav kobj_delete((kobj_t)obj, M_RAID); 2022219974Smav } 2023219974Smav if (class == NULL) { 2024219974Smav G_RAID_DEBUG1(0, vol->v_softc, 2025219974Smav "No transformation module found for %s.", 2026219974Smav vol->v_name); 2027219974Smav vol->v_tr = NULL; 2028219974Smav g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); 2029219974Smav g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, 2030219974Smav G_RAID_EVENT_VOLUME); 2031219974Smav return (-1); 2032219974Smav } 2033219974Smav G_RAID_DEBUG1(2, vol->v_softc, 2034219974Smav "Transformation module %s chosen for %s.", 2035219974Smav class->name, vol->v_name); 2036219974Smav vol->v_tr = obj; 2037219974Smav return (0); 2038219974Smav} 2039219974Smav 2040219974Smavint 2041219974Smavg_raid_destroy_node(struct g_raid_softc *sc, int worker) 2042219974Smav{ 2043219974Smav struct g_raid_volume *vol, *tmpv; 2044219974Smav struct g_raid_disk *disk, *tmpd; 2045219974Smav int error = 0; 2046219974Smav 2047219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 2048219974Smav TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { 2049219974Smav if (g_raid_destroy_volume(vol)) 2050219974Smav error = EBUSY; 2051219974Smav } 2052219974Smav if (error) 2053219974Smav return (error); 2054219974Smav TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { 2055219974Smav if (g_raid_destroy_disk(disk)) 2056219974Smav error = EBUSY; 2057219974Smav } 2058219974Smav if (error) 2059219974Smav return (error); 2060219974Smav if (sc->sc_md) { 2061219974Smav G_RAID_MD_FREE(sc->sc_md); 2062219974Smav kobj_delete((kobj_t)sc->sc_md, M_RAID); 2063219974Smav sc->sc_md = NULL; 2064219974Smav } 2065219974Smav if (sc->sc_geom != NULL) { 2066219974Smav G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); 2067219974Smav g_topology_lock(); 2068219974Smav sc->sc_geom->softc = NULL; 2069219974Smav g_wither_geom(sc->sc_geom, ENXIO); 2070219974Smav g_topology_unlock(); 2071219974Smav sc->sc_geom = NULL; 2072219974Smav } else 2073219974Smav G_RAID_DEBUG(1, "Array destroyed."); 2074219974Smav if (worker) { 2075219974Smav g_raid_event_cancel(sc, sc); 2076219974Smav mtx_destroy(&sc->sc_queue_mtx); 2077219974Smav sx_xunlock(&sc->sc_lock); 2078219974Smav sx_destroy(&sc->sc_lock); 2079219974Smav wakeup(&sc->sc_stopping); 2080219974Smav free(sc, M_RAID); 2081219974Smav curthread->td_pflags &= ~TDP_GEOM; 2082219974Smav G_RAID_DEBUG(1, "Thread exiting."); 2083219974Smav kproc_exit(0); 2084219974Smav } else { 2085219974Smav /* Wake up worker to make it selfdestruct. */ 2086219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2087219974Smav } 2088219974Smav return (0); 2089219974Smav} 2090219974Smav 2091219974Smavint 2092219974Smavg_raid_destroy_volume(struct g_raid_volume *vol) 2093219974Smav{ 2094219974Smav struct g_raid_softc *sc; 2095219974Smav struct g_raid_disk *disk; 2096219974Smav int i; 2097219974Smav 2098219974Smav sc = vol->v_softc; 2099219974Smav G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); 2100219974Smav vol->v_stopping = 1; 2101219974Smav if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { 2102219974Smav if (vol->v_tr) { 2103219974Smav G_RAID_TR_STOP(vol->v_tr); 2104219974Smav return (EBUSY); 2105219974Smav } else 2106219974Smav vol->v_state = G_RAID_VOLUME_S_STOPPED; 2107219974Smav } 2108219974Smav if (g_raid_event_check(sc, vol) != 0) 2109219974Smav return (EBUSY); 2110219974Smav if (vol->v_provider != NULL) 2111219974Smav return (EBUSY); 2112219974Smav if (vol->v_provider_open != 0) 2113219974Smav return (EBUSY); 2114219974Smav if (vol->v_tr) { 2115219974Smav G_RAID_TR_FREE(vol->v_tr); 2116219974Smav kobj_delete((kobj_t)vol->v_tr, M_RAID); 2117219974Smav vol->v_tr = NULL; 2118219974Smav } 2119219974Smav if (vol->v_rootmount) 2120219974Smav root_mount_rel(vol->v_rootmount); 2121219974Smav g_topology_lock(); 2122219974Smav LIST_REMOVE(vol, v_global_next); 2123219974Smav g_topology_unlock(); 2124219974Smav TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); 2125219974Smav for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 2126219974Smav g_raid_event_cancel(sc, &vol->v_subdisks[i]); 2127219974Smav disk = vol->v_subdisks[i].sd_disk; 2128219974Smav if (disk == NULL) 2129219974Smav continue; 2130219974Smav TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); 2131219974Smav } 2132219974Smav G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); 2133219974Smav if (sc->sc_md) 2134219974Smav G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); 2135219974Smav g_raid_event_cancel(sc, vol); 2136219974Smav free(vol, M_RAID); 2137219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) { 2138219974Smav /* Wake up worker to let it selfdestruct. */ 2139219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2140219974Smav } 2141219974Smav return (0); 2142219974Smav} 2143219974Smav 2144219974Smavint 2145219974Smavg_raid_destroy_disk(struct g_raid_disk *disk) 2146219974Smav{ 2147219974Smav struct g_raid_softc *sc; 2148219974Smav struct g_raid_subdisk *sd, *tmp; 2149219974Smav 2150219974Smav sc = disk->d_softc; 2151219974Smav G_RAID_DEBUG1(2, sc, "Destroying disk."); 2152219974Smav if (disk->d_consumer) { 2153219974Smav g_raid_kill_consumer(sc, disk->d_consumer); 2154219974Smav disk->d_consumer = NULL; 2155219974Smav } 2156219974Smav TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { 2157219974Smav g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); 2158219974Smav g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 2159219974Smav G_RAID_EVENT_SUBDISK); 2160219974Smav TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); 2161219974Smav sd->sd_disk = NULL; 2162219974Smav } 2163219974Smav TAILQ_REMOVE(&sc->sc_disks, disk, d_next); 2164219974Smav if (sc->sc_md) 2165219974Smav G_RAID_MD_FREE_DISK(sc->sc_md, disk); 2166219974Smav g_raid_event_cancel(sc, disk); 2167219974Smav free(disk, M_RAID); 2168219974Smav return (0); 2169219974Smav} 2170219974Smav 2171219974Smavint 2172219974Smavg_raid_destroy(struct g_raid_softc *sc, int how) 2173219974Smav{ 2174219974Smav int opens; 2175219974Smav 2176219974Smav g_topology_assert_not(); 2177219974Smav if (sc == NULL) 2178219974Smav return (ENXIO); 2179219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 2180219974Smav 2181219974Smav /* Count open volumes. */ 2182219974Smav opens = g_raid_nopens(sc); 2183219974Smav 2184219974Smav /* React on some opened volumes. */ 2185219974Smav if (opens > 0) { 2186219974Smav switch (how) { 2187219974Smav case G_RAID_DESTROY_SOFT: 2188219974Smav G_RAID_DEBUG1(1, sc, 2189219974Smav "%d volumes are still open.", 2190219974Smav opens); 2191219974Smav return (EBUSY); 2192219974Smav case G_RAID_DESTROY_DELAYED: 2193219974Smav G_RAID_DEBUG1(1, sc, 2194219974Smav "Array will be destroyed on last close."); 2195219974Smav sc->sc_stopping = G_RAID_DESTROY_DELAYED; 2196219974Smav return (EBUSY); 2197219974Smav case G_RAID_DESTROY_HARD: 2198219974Smav G_RAID_DEBUG1(1, sc, 2199219974Smav "%d volumes are still open.", 2200219974Smav opens); 2201219974Smav } 2202219974Smav } 2203219974Smav 2204219974Smav /* Mark node for destruction. */ 2205219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 2206219974Smav /* Wake up worker to let it selfdestruct. */ 2207219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2208219974Smav /* Sleep until node destroyed. */ 2209219974Smav sx_sleep(&sc->sc_stopping, &sc->sc_lock, 2210219974Smav PRIBIO | PDROP, "r:destroy", 0); 2211219974Smav return (0); 2212219974Smav} 2213219974Smav 2214219974Smavstatic void 2215219974Smavg_raid_taste_orphan(struct g_consumer *cp) 2216219974Smav{ 2217219974Smav 2218219974Smav KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2219219974Smav cp->provider->name)); 2220219974Smav} 2221219974Smav 2222219974Smavstatic struct g_geom * 2223219974Smavg_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2224219974Smav{ 2225219974Smav struct g_consumer *cp; 2226219974Smav struct g_geom *gp, *geom; 2227219974Smav struct g_raid_md_class *class; 2228219974Smav struct g_raid_md_object *obj; 2229219974Smav int status; 2230219974Smav 2231219974Smav g_topology_assert(); 2232219974Smav g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2233240465Smav if (!g_raid_enable) 2234240465Smav return (NULL); 2235219974Smav G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); 2236219974Smav 2237234940Smav gp = g_new_geomf(mp, "raid:taste"); 2238219974Smav /* 2239219974Smav * This orphan function should be never called. 2240219974Smav */ 2241219974Smav gp->orphan = g_raid_taste_orphan; 2242219974Smav cp = g_new_consumer(gp); 2243219974Smav g_attach(cp, pp); 2244219974Smav 2245219974Smav geom = NULL; 2246219974Smav LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2247240465Smav if (!class->mdc_enable) 2248240465Smav continue; 2249219974Smav G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", 2250219974Smav pp->name, class->name); 2251219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2252219974Smav M_WAITOK); 2253219974Smav obj->mdo_class = class; 2254219974Smav status = G_RAID_MD_TASTE(obj, mp, cp, &geom); 2255219974Smav if (status != G_RAID_MD_TASTE_NEW) 2256219974Smav kobj_delete((kobj_t)obj, M_RAID); 2257219974Smav if (status != G_RAID_MD_TASTE_FAIL) 2258219974Smav break; 2259219974Smav } 2260219974Smav 2261219974Smav g_detach(cp); 2262219974Smav g_destroy_consumer(cp); 2263219974Smav g_destroy_geom(gp); 2264219974Smav G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); 2265219974Smav return (geom); 2266219974Smav} 2267219974Smav 2268219974Smavint 2269234940Smavg_raid_create_node_format(const char *format, struct gctl_req *req, 2270234940Smav struct g_geom **gp) 2271219974Smav{ 2272219974Smav struct g_raid_md_class *class; 2273219974Smav struct g_raid_md_object *obj; 2274219974Smav int status; 2275219974Smav 2276219974Smav G_RAID_DEBUG(2, "Creating array for %s metadata.", format); 2277219974Smav LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2278219974Smav if (strcasecmp(class->name, format) == 0) 2279219974Smav break; 2280219974Smav } 2281219974Smav if (class == NULL) { 2282219974Smav G_RAID_DEBUG(1, "No support for %s metadata.", format); 2283219974Smav return (G_RAID_MD_TASTE_FAIL); 2284219974Smav } 2285219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2286219974Smav M_WAITOK); 2287219974Smav obj->mdo_class = class; 2288234940Smav status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp); 2289219974Smav if (status != G_RAID_MD_TASTE_NEW) 2290219974Smav kobj_delete((kobj_t)obj, M_RAID); 2291219974Smav return (status); 2292219974Smav} 2293219974Smav 2294219974Smavstatic int 2295219974Smavg_raid_destroy_geom(struct gctl_req *req __unused, 2296219974Smav struct g_class *mp __unused, struct g_geom *gp) 2297219974Smav{ 2298219974Smav struct g_raid_softc *sc; 2299219974Smav int error; 2300219974Smav 2301219974Smav g_topology_unlock(); 2302219974Smav sc = gp->softc; 2303219974Smav sx_xlock(&sc->sc_lock); 2304219974Smav g_cancel_event(sc); 2305219974Smav error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); 2306219974Smav if (error != 0) 2307219974Smav sx_xunlock(&sc->sc_lock); 2308219974Smav g_topology_lock(); 2309219974Smav return (error); 2310219974Smav} 2311219974Smav 2312219974Smavvoid g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, 2313219974Smav struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2314219974Smav{ 2315219974Smav 2316219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) 2317219974Smav return; 2318219974Smav if (sc->sc_md) 2319219974Smav G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); 2320219974Smav} 2321219974Smav 2322219974Smavvoid g_raid_fail_disk(struct g_raid_softc *sc, 2323219974Smav struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2324219974Smav{ 2325219974Smav 2326219974Smav if (disk == NULL) 2327219974Smav disk = sd->sd_disk; 2328219974Smav if (disk == NULL) { 2329219974Smav G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); 2330219974Smav return; 2331219974Smav } 2332219974Smav if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 2333219974Smav G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " 2334219974Smav "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 2335219974Smav return; 2336219974Smav } 2337219974Smav if (sc->sc_md) 2338219974Smav G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); 2339219974Smav} 2340219974Smav 2341219974Smavstatic void 2342219974Smavg_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2343219974Smav struct g_consumer *cp, struct g_provider *pp) 2344219974Smav{ 2345219974Smav struct g_raid_softc *sc; 2346219974Smav struct g_raid_volume *vol; 2347219974Smav struct g_raid_subdisk *sd; 2348219974Smav struct g_raid_disk *disk; 2349219974Smav int i, s; 2350219974Smav 2351219974Smav g_topology_assert(); 2352219974Smav 2353219974Smav sc = gp->softc; 2354219974Smav if (sc == NULL) 2355219974Smav return; 2356219974Smav if (pp != NULL) { 2357219974Smav vol = pp->private; 2358219974Smav g_topology_unlock(); 2359219974Smav sx_xlock(&sc->sc_lock); 2360249974Smav sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent, 2361249974Smav sc->sc_md->mdo_class->name, 2362249974Smav g_raid_volume_level2str(vol->v_raid_level, 2363249974Smav vol->v_raid_level_qualifier)); 2364219974Smav sbuf_printf(sb, "%s<Label>%s</Label>\n", indent, 2365219974Smav vol->v_name); 2366219974Smav sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent, 2367219974Smav g_raid_volume_level2str(vol->v_raid_level, 2368219974Smav vol->v_raid_level_qualifier)); 2369219974Smav sbuf_printf(sb, 2370219974Smav "%s<Transformation>%s</Transformation>\n", indent, 2371219974Smav vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); 2372219974Smav sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2373219974Smav vol->v_disks_count); 2374219974Smav sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent, 2375219974Smav vol->v_strip_size); 2376219974Smav sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2377219974Smav g_raid_volume_state2str(vol->v_state)); 2378219974Smav sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent, 2379219974Smav vol->v_dirty ? "Yes" : "No"); 2380219974Smav sbuf_printf(sb, "%s<Subdisks>", indent); 2381219974Smav for (i = 0; i < vol->v_disks_count; i++) { 2382219974Smav sd = &vol->v_subdisks[i]; 2383219974Smav if (sd->sd_disk != NULL && 2384219974Smav sd->sd_disk->d_consumer != NULL) { 2385219974Smav sbuf_printf(sb, "%s ", 2386219974Smav g_raid_get_diskname(sd->sd_disk)); 2387219974Smav } else { 2388219974Smav sbuf_printf(sb, "NONE "); 2389219974Smav } 2390219974Smav sbuf_printf(sb, "(%s", 2391219974Smav g_raid_subdisk_state2str(sd->sd_state)); 2392219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2393219974Smav sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2394219974Smav sbuf_printf(sb, " %d%%", 2395219974Smav (int)(sd->sd_rebuild_pos * 100 / 2396219974Smav sd->sd_size)); 2397219974Smav } 2398219974Smav sbuf_printf(sb, ")"); 2399219974Smav if (i + 1 < vol->v_disks_count) 2400219974Smav sbuf_printf(sb, ", "); 2401219974Smav } 2402219974Smav sbuf_printf(sb, "</Subdisks>\n"); 2403219974Smav sx_xunlock(&sc->sc_lock); 2404219974Smav g_topology_lock(); 2405219974Smav } else if (cp != NULL) { 2406219974Smav disk = cp->private; 2407219974Smav if (disk == NULL) 2408219974Smav return; 2409219974Smav g_topology_unlock(); 2410219974Smav sx_xlock(&sc->sc_lock); 2411219974Smav sbuf_printf(sb, "%s<State>%s", indent, 2412219974Smav g_raid_disk_state2str(disk->d_state)); 2413219974Smav if (!TAILQ_EMPTY(&disk->d_subdisks)) { 2414219974Smav sbuf_printf(sb, " ("); 2415219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2416219974Smav sbuf_printf(sb, "%s", 2417219974Smav g_raid_subdisk_state2str(sd->sd_state)); 2418219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2419219974Smav sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2420219974Smav sbuf_printf(sb, " %d%%", 2421219974Smav (int)(sd->sd_rebuild_pos * 100 / 2422219974Smav sd->sd_size)); 2423219974Smav } 2424219974Smav if (TAILQ_NEXT(sd, sd_next)) 2425219974Smav sbuf_printf(sb, ", "); 2426219974Smav } 2427219974Smav sbuf_printf(sb, ")"); 2428219974Smav } 2429219974Smav sbuf_printf(sb, "</State>\n"); 2430219974Smav sbuf_printf(sb, "%s<Subdisks>", indent); 2431219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2432219974Smav sbuf_printf(sb, "r%d(%s):%d@%ju", 2433219974Smav sd->sd_volume->v_global_id, 2434219974Smav sd->sd_volume->v_name, 2435219974Smav sd->sd_pos, sd->sd_offset); 2436219974Smav if (TAILQ_NEXT(sd, sd_next)) 2437219974Smav sbuf_printf(sb, ", "); 2438219974Smav } 2439219974Smav sbuf_printf(sb, "</Subdisks>\n"); 2440219974Smav sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent, 2441219974Smav disk->d_read_errs); 2442219974Smav sx_xunlock(&sc->sc_lock); 2443219974Smav g_topology_lock(); 2444219974Smav } else { 2445219974Smav g_topology_unlock(); 2446219974Smav sx_xlock(&sc->sc_lock); 2447219974Smav if (sc->sc_md) { 2448219974Smav sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent, 2449219974Smav sc->sc_md->mdo_class->name); 2450219974Smav } 2451219974Smav if (!TAILQ_EMPTY(&sc->sc_volumes)) { 2452219974Smav s = 0xff; 2453219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 2454219974Smav if (vol->v_state < s) 2455219974Smav s = vol->v_state; 2456219974Smav } 2457219974Smav sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2458219974Smav g_raid_volume_state2str(s)); 2459219974Smav } 2460219974Smav sx_xunlock(&sc->sc_lock); 2461219974Smav g_topology_lock(); 2462219974Smav } 2463219974Smav} 2464219974Smav 2465219974Smavstatic void 2466242314Smavg_raid_shutdown_post_sync(void *arg, int howto) 2467219974Smav{ 2468219974Smav struct g_class *mp; 2469219974Smav struct g_geom *gp, *gp2; 2470219974Smav struct g_raid_softc *sc; 2471242314Smav struct g_raid_volume *vol; 2472219974Smav int error; 2473219974Smav 2474219974Smav mp = arg; 2475219974Smav DROP_GIANT(); 2476219974Smav g_topology_lock(); 2477242314Smav g_raid_shutdown = 1; 2478219974Smav LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2479219974Smav if ((sc = gp->softc) == NULL) 2480219974Smav continue; 2481219974Smav g_topology_unlock(); 2482219974Smav sx_xlock(&sc->sc_lock); 2483242314Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) 2484242314Smav g_raid_clean(vol, -1); 2485219974Smav g_cancel_event(sc); 2486219974Smav error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); 2487219974Smav if (error != 0) 2488219974Smav sx_xunlock(&sc->sc_lock); 2489219974Smav g_topology_lock(); 2490219974Smav } 2491219974Smav g_topology_unlock(); 2492219974Smav PICKUP_GIANT(); 2493219974Smav} 2494219974Smav 2495219974Smavstatic void 2496219974Smavg_raid_init(struct g_class *mp) 2497219974Smav{ 2498219974Smav 2499242314Smav g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 2500242314Smav g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 2501242314Smav if (g_raid_post_sync == NULL) 2502219974Smav G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); 2503219974Smav g_raid_started = 1; 2504219974Smav} 2505219974Smav 2506219974Smavstatic void 2507219974Smavg_raid_fini(struct g_class *mp) 2508219974Smav{ 2509219974Smav 2510242314Smav if (g_raid_post_sync != NULL) 2511242314Smav EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync); 2512219974Smav g_raid_started = 0; 2513219974Smav} 2514219974Smav 2515219974Smavint 2516219974Smavg_raid_md_modevent(module_t mod, int type, void *arg) 2517219974Smav{ 2518219974Smav struct g_raid_md_class *class, *c, *nc; 2519219974Smav int error; 2520219974Smav 2521219974Smav error = 0; 2522219974Smav class = arg; 2523219974Smav switch (type) { 2524219974Smav case MOD_LOAD: 2525219974Smav c = LIST_FIRST(&g_raid_md_classes); 2526219974Smav if (c == NULL || c->mdc_priority > class->mdc_priority) 2527219974Smav LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); 2528219974Smav else { 2529219974Smav while ((nc = LIST_NEXT(c, mdc_list)) != NULL && 2530219974Smav nc->mdc_priority < class->mdc_priority) 2531219974Smav c = nc; 2532219974Smav LIST_INSERT_AFTER(c, class, mdc_list); 2533219974Smav } 2534219974Smav if (g_raid_started) 2535219974Smav g_retaste(&g_raid_class); 2536219974Smav break; 2537219974Smav case MOD_UNLOAD: 2538219974Smav LIST_REMOVE(class, mdc_list); 2539219974Smav break; 2540219974Smav default: 2541219974Smav error = EOPNOTSUPP; 2542219974Smav break; 2543219974Smav } 2544219974Smav 2545219974Smav return (error); 2546219974Smav} 2547219974Smav 2548219974Smavint 2549219974Smavg_raid_tr_modevent(module_t mod, int type, void *arg) 2550219974Smav{ 2551219974Smav struct g_raid_tr_class *class, *c, *nc; 2552219974Smav int error; 2553219974Smav 2554219974Smav error = 0; 2555219974Smav class = arg; 2556219974Smav switch (type) { 2557219974Smav case MOD_LOAD: 2558219974Smav c = LIST_FIRST(&g_raid_tr_classes); 2559219974Smav if (c == NULL || c->trc_priority > class->trc_priority) 2560219974Smav LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); 2561219974Smav else { 2562219974Smav while ((nc = LIST_NEXT(c, trc_list)) != NULL && 2563219974Smav nc->trc_priority < class->trc_priority) 2564219974Smav c = nc; 2565219974Smav LIST_INSERT_AFTER(c, class, trc_list); 2566219974Smav } 2567219974Smav break; 2568219974Smav case MOD_UNLOAD: 2569219974Smav LIST_REMOVE(class, trc_list); 2570219974Smav break; 2571219974Smav default: 2572219974Smav error = EOPNOTSUPP; 2573219974Smav break; 2574219974Smav } 2575219974Smav 2576219974Smav return (error); 2577219974Smav} 2578219974Smav 2579219974Smav/* 2580219974Smav * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) 2581219974Smav * to reduce module priority, allowing submodules to register them first. 2582219974Smav */ 2583219974Smavstatic moduledata_t g_raid_mod = { 2584219974Smav "g_raid", 2585219974Smav g_modevent, 2586219974Smav &g_raid_class 2587219974Smav}; 2588219974SmavDECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD); 2589219974SmavMODULE_VERSION(geom_raid, 0); 2590