1219974Smav/*- 2219974Smav * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org> 3219974Smav * All rights reserved. 4219974Smav * 5219974Smav * Redistribution and use in source and binary forms, with or without 6219974Smav * modification, are permitted provided that the following conditions 7219974Smav * are met: 8219974Smav * 1. Redistributions of source code must retain the above copyright 9219974Smav * notice, this list of conditions and the following disclaimer. 10219974Smav * 2. Redistributions in binary form must reproduce the above copyright 11219974Smav * notice, this list of conditions and the following disclaimer in the 12219974Smav * documentation and/or other materials provided with the distribution. 13219974Smav * 14219974Smav * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 15219974Smav * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16219974Smav * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17219974Smav * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 18219974Smav * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19219974Smav * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20219974Smav * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21219974Smav * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22219974Smav * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23219974Smav * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24219974Smav * SUCH DAMAGE. 25219974Smav */ 26219974Smav 27219974Smav#include <sys/cdefs.h> 28219974Smav__FBSDID("$FreeBSD$"); 29219974Smav 30219974Smav#include <sys/param.h> 31219974Smav#include <sys/systm.h> 32219974Smav#include <sys/kernel.h> 33219974Smav#include <sys/module.h> 34219974Smav#include <sys/limits.h> 35219974Smav#include <sys/lock.h> 36219974Smav#include <sys/mutex.h> 37219974Smav#include <sys/bio.h> 38223921Sae#include <sys/sbuf.h> 39219974Smav#include <sys/sysctl.h> 40219974Smav#include <sys/malloc.h> 41219974Smav#include <sys/eventhandler.h> 42219974Smav#include <vm/uma.h> 43219974Smav#include <geom/geom.h> 44219974Smav#include <sys/proc.h> 45219974Smav#include <sys/kthread.h> 46219974Smav#include <sys/sched.h> 47219974Smav#include <geom/raid/g_raid.h> 48219974Smav#include "g_raid_md_if.h" 49219974Smav#include "g_raid_tr_if.h" 50219974Smav 51219974Smavstatic MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data"); 52219974Smav 53219974SmavSYSCTL_DECL(_kern_geom); 54219974SmavSYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff"); 55240465Smavint g_raid_enable = 1; 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); 795260385Sscottl cp->flags |= G_CF_DIRECT_RECEIVE; 796219974Smav if (g_attach(cp, pp) != 0) { 797219974Smav g_destroy_consumer(cp); 798219974Smav return (NULL); 799219974Smav } 800219974Smav if (g_access(cp, 1, 1, 1) != 0) { 801219974Smav g_detach(cp); 802219974Smav g_destroy_consumer(cp); 803219974Smav return (NULL); 804219974Smav } 805219974Smav return (cp); 806219974Smav} 807219974Smav 808219974Smavstatic u_int 809219974Smavg_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) 810219974Smav{ 811219974Smav struct bio *bp; 812219974Smav u_int nreqs = 0; 813219974Smav 814219974Smav mtx_lock(&sc->sc_queue_mtx); 815219974Smav TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 816219974Smav if (bp->bio_from == cp) 817219974Smav nreqs++; 818219974Smav } 819219974Smav mtx_unlock(&sc->sc_queue_mtx); 820219974Smav return (nreqs); 821219974Smav} 822219974Smav 823219974Smavu_int 824219974Smavg_raid_nopens(struct g_raid_softc *sc) 825219974Smav{ 826219974Smav struct g_raid_volume *vol; 827219974Smav u_int opens; 828219974Smav 829219974Smav opens = 0; 830219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 831219974Smav if (vol->v_provider_open != 0) 832219974Smav opens++; 833219974Smav } 834219974Smav return (opens); 835219974Smav} 836219974Smav 837219974Smavstatic int 838219974Smavg_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) 839219974Smav{ 840219974Smav 841219974Smav if (cp->index > 0) { 842219974Smav G_RAID_DEBUG1(2, sc, 843219974Smav "I/O requests for %s exist, can't destroy it now.", 844219974Smav cp->provider->name); 845219974Smav return (1); 846219974Smav } 847219974Smav if (g_raid_nrequests(sc, cp) > 0) { 848219974Smav G_RAID_DEBUG1(2, sc, 849219974Smav "I/O requests for %s in queue, can't destroy it now.", 850219974Smav cp->provider->name); 851219974Smav return (1); 852219974Smav } 853219974Smav return (0); 854219974Smav} 855219974Smav 856219974Smavstatic void 857219974Smavg_raid_destroy_consumer(void *arg, int flags __unused) 858219974Smav{ 859219974Smav struct g_consumer *cp; 860219974Smav 861219974Smav g_topology_assert(); 862219974Smav 863219974Smav cp = arg; 864219974Smav G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 865219974Smav g_detach(cp); 866219974Smav g_destroy_consumer(cp); 867219974Smav} 868219974Smav 869219974Smavvoid 870219974Smavg_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) 871219974Smav{ 872219974Smav struct g_provider *pp; 873219974Smav int retaste_wait; 874219974Smav 875219974Smav g_topology_assert_not(); 876219974Smav 877219974Smav g_topology_lock(); 878219974Smav cp->private = NULL; 879219974Smav if (g_raid_consumer_is_busy(sc, cp)) 880219974Smav goto out; 881219974Smav pp = cp->provider; 882219974Smav retaste_wait = 0; 883219974Smav if (cp->acw == 1) { 884219974Smav if ((pp->geom->flags & G_GEOM_WITHER) == 0) 885219974Smav retaste_wait = 1; 886219974Smav } 887219974Smav if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 888219974Smav g_access(cp, -cp->acr, -cp->acw, -cp->ace); 889219974Smav if (retaste_wait) { 890219974Smav /* 891219974Smav * After retaste event was send (inside g_access()), we can send 892219974Smav * event to detach and destroy consumer. 893219974Smav * A class, which has consumer to the given provider connected 894219974Smav * will not receive retaste event for the provider. 895219974Smav * This is the way how I ignore retaste events when I close 896219974Smav * consumers opened for write: I detach and destroy consumer 897219974Smav * after retaste event is sent. 898219974Smav */ 899219974Smav g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); 900219974Smav goto out; 901219974Smav } 902219974Smav G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); 903219974Smav g_detach(cp); 904219974Smav g_destroy_consumer(cp); 905219974Smavout: 906219974Smav g_topology_unlock(); 907219974Smav} 908219974Smav 909219974Smavstatic void 910219974Smavg_raid_orphan(struct g_consumer *cp) 911219974Smav{ 912219974Smav struct g_raid_disk *disk; 913219974Smav 914219974Smav g_topology_assert(); 915219974Smav 916219974Smav disk = cp->private; 917219974Smav if (disk == NULL) 918219974Smav return; 919219974Smav g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, 920219974Smav G_RAID_EVENT_DISK); 921219974Smav} 922219974Smav 923242314Smavstatic void 924219974Smavg_raid_clean(struct g_raid_volume *vol, int acw) 925219974Smav{ 926219974Smav struct g_raid_softc *sc; 927219974Smav int timeout; 928219974Smav 929219974Smav sc = vol->v_softc; 930219974Smav g_topology_assert_not(); 931219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 932219974Smav 933219974Smav// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 934242314Smav// return; 935219974Smav if (!vol->v_dirty) 936242314Smav return; 937219974Smav if (vol->v_writes > 0) 938242314Smav return; 939219974Smav if (acw > 0 || (acw == -1 && 940219974Smav vol->v_provider != NULL && vol->v_provider->acw > 0)) { 941219974Smav timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); 942242314Smav if (!g_raid_shutdown && timeout > 0) 943242314Smav return; 944219974Smav } 945219974Smav vol->v_dirty = 0; 946219974Smav G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", 947219974Smav vol->v_name); 948219974Smav g_raid_write_metadata(sc, vol, NULL, NULL); 949219974Smav} 950219974Smav 951219974Smavstatic void 952219974Smavg_raid_dirty(struct g_raid_volume *vol) 953219974Smav{ 954219974Smav struct g_raid_softc *sc; 955219974Smav 956219974Smav sc = vol->v_softc; 957219974Smav g_topology_assert_not(); 958219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 959219974Smav 960219974Smav// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 961219974Smav// return; 962219974Smav vol->v_dirty = 1; 963219974Smav G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", 964219974Smav vol->v_name); 965219974Smav g_raid_write_metadata(sc, vol, NULL, NULL); 966219974Smav} 967219974Smav 968219974Smavvoid 969219974Smavg_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) 970219974Smav{ 971219974Smav struct g_raid_softc *sc; 972219974Smav struct g_raid_volume *vol; 973219974Smav struct g_raid_subdisk *sd; 974219974Smav struct bio_queue_head queue; 975219974Smav struct bio *cbp; 976219974Smav int i; 977219974Smav 978219974Smav vol = tr->tro_volume; 979219974Smav sc = vol->v_softc; 980219974Smav 981219974Smav /* 982219974Smav * Allocate all bios before sending any request, so we can return 983219974Smav * ENOMEM in nice and clean way. 984219974Smav */ 985219974Smav bioq_init(&queue); 986219974Smav for (i = 0; i < vol->v_disks_count; i++) { 987219974Smav sd = &vol->v_subdisks[i]; 988219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_NONE || 989219974Smav sd->sd_state == G_RAID_SUBDISK_S_FAILED) 990219974Smav continue; 991219974Smav cbp = g_clone_bio(bp); 992219974Smav if (cbp == NULL) 993219974Smav goto failure; 994219974Smav cbp->bio_caller1 = sd; 995219974Smav bioq_insert_tail(&queue, cbp); 996219974Smav } 997260385Sscottl while ((cbp = bioq_takefirst(&queue)) != NULL) { 998219974Smav sd = cbp->bio_caller1; 999219974Smav cbp->bio_caller1 = NULL; 1000219974Smav g_raid_subdisk_iostart(sd, cbp); 1001219974Smav } 1002219974Smav return; 1003219974Smavfailure: 1004260385Sscottl while ((cbp = bioq_takefirst(&queue)) != NULL) 1005219974Smav g_destroy_bio(cbp); 1006219974Smav if (bp->bio_error == 0) 1007219974Smav bp->bio_error = ENOMEM; 1008219974Smav g_raid_iodone(bp, bp->bio_error); 1009219974Smav} 1010219974Smav 1011219974Smavstatic void 1012219974Smavg_raid_tr_kerneldump_common_done(struct bio *bp) 1013219974Smav{ 1014219974Smav 1015219974Smav bp->bio_flags |= BIO_DONE; 1016219974Smav} 1017219974Smav 1018219974Smavint 1019219974Smavg_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, 1020219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1021219974Smav{ 1022219974Smav struct g_raid_softc *sc; 1023219974Smav struct g_raid_volume *vol; 1024219974Smav struct bio bp; 1025219974Smav 1026219974Smav vol = tr->tro_volume; 1027219974Smav sc = vol->v_softc; 1028219974Smav 1029219974Smav bzero(&bp, sizeof(bp)); 1030219974Smav bp.bio_cmd = BIO_WRITE; 1031219974Smav bp.bio_done = g_raid_tr_kerneldump_common_done; 1032219974Smav bp.bio_attribute = NULL; 1033219974Smav bp.bio_offset = offset; 1034219974Smav bp.bio_length = length; 1035219974Smav bp.bio_data = virtual; 1036219974Smav bp.bio_to = vol->v_provider; 1037219974Smav 1038219974Smav g_raid_start(&bp); 1039219974Smav while (!(bp.bio_flags & BIO_DONE)) { 1040219974Smav G_RAID_DEBUG1(4, sc, "Poll..."); 1041219974Smav g_raid_poll(sc); 1042219974Smav DELAY(10); 1043219974Smav } 1044219974Smav 1045219974Smav return (bp.bio_error != 0 ? EIO : 0); 1046219974Smav} 1047219974Smav 1048219974Smavstatic int 1049219974Smavg_raid_dump(void *arg, 1050219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1051219974Smav{ 1052219974Smav struct g_raid_volume *vol; 1053219974Smav int error; 1054219974Smav 1055219974Smav vol = (struct g_raid_volume *)arg; 1056219974Smav G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", 1057219974Smav (long long unsigned)offset, (long long unsigned)length); 1058219974Smav 1059219974Smav error = G_RAID_TR_KERNELDUMP(vol->v_tr, 1060219974Smav virtual, physical, offset, length); 1061219974Smav return (error); 1062219974Smav} 1063219974Smav 1064219974Smavstatic void 1065219974Smavg_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) 1066219974Smav{ 1067219974Smav struct g_kerneldump *gkd; 1068219974Smav struct g_provider *pp; 1069219974Smav struct g_raid_volume *vol; 1070219974Smav 1071219974Smav gkd = (struct g_kerneldump*)bp->bio_data; 1072219974Smav pp = bp->bio_to; 1073219974Smav vol = pp->private; 1074219974Smav g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", 1075219974Smav pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 1076219974Smav gkd->di.dumper = g_raid_dump; 1077219974Smav gkd->di.priv = vol; 1078219974Smav gkd->di.blocksize = vol->v_sectorsize; 1079219974Smav gkd->di.maxiosize = DFLTPHYS; 1080219974Smav gkd->di.mediaoffset = gkd->offset; 1081219974Smav if ((gkd->offset + gkd->length) > vol->v_mediasize) 1082219974Smav gkd->length = vol->v_mediasize - gkd->offset; 1083219974Smav gkd->di.mediasize = gkd->length; 1084219974Smav g_io_deliver(bp, 0); 1085219974Smav} 1086219974Smav 1087219974Smavstatic void 1088242323Smavg_raid_candelete(struct g_raid_softc *sc, struct bio *bp) 1089242323Smav{ 1090242323Smav struct g_provider *pp; 1091242323Smav struct g_raid_volume *vol; 1092242323Smav struct g_raid_subdisk *sd; 1093242323Smav int *val; 1094242323Smav int i; 1095242323Smav 1096242323Smav val = (int *)bp->bio_data; 1097242323Smav pp = bp->bio_to; 1098242323Smav vol = pp->private; 1099242323Smav *val = 0; 1100242323Smav for (i = 0; i < vol->v_disks_count; i++) { 1101242323Smav sd = &vol->v_subdisks[i]; 1102242323Smav if (sd->sd_state == G_RAID_SUBDISK_S_NONE) 1103242323Smav continue; 1104242323Smav if (sd->sd_disk->d_candelete) { 1105242323Smav *val = 1; 1106242323Smav break; 1107242323Smav } 1108242323Smav } 1109242323Smav g_io_deliver(bp, 0); 1110242323Smav} 1111242323Smav 1112242323Smavstatic void 1113219974Smavg_raid_start(struct bio *bp) 1114219974Smav{ 1115219974Smav struct g_raid_softc *sc; 1116219974Smav 1117219974Smav sc = bp->bio_to->geom->softc; 1118219974Smav /* 1119219974Smav * If sc == NULL or there are no valid disks, provider's error 1120219974Smav * should be set and g_raid_start() should not be called at all. 1121219974Smav */ 1122219974Smav// KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, 1123219974Smav// ("Provider's error should be set (error=%d)(mirror=%s).", 1124219974Smav// bp->bio_to->error, bp->bio_to->name)); 1125219974Smav G_RAID_LOGREQ(3, bp, "Request received."); 1126219974Smav 1127219974Smav switch (bp->bio_cmd) { 1128219974Smav case BIO_READ: 1129219974Smav case BIO_WRITE: 1130219974Smav case BIO_DELETE: 1131219974Smav case BIO_FLUSH: 1132219974Smav break; 1133219974Smav case BIO_GETATTR: 1134242323Smav if (!strcmp(bp->bio_attribute, "GEOM::candelete")) 1135242323Smav g_raid_candelete(sc, bp); 1136242323Smav else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 1137219974Smav g_raid_kerneldump(sc, bp); 1138219974Smav else 1139219974Smav g_io_deliver(bp, EOPNOTSUPP); 1140219974Smav return; 1141219974Smav default: 1142219974Smav g_io_deliver(bp, EOPNOTSUPP); 1143219974Smav return; 1144219974Smav } 1145219974Smav mtx_lock(&sc->sc_queue_mtx); 1146219974Smav bioq_disksort(&sc->sc_queue, bp); 1147219974Smav mtx_unlock(&sc->sc_queue_mtx); 1148219974Smav if (!dumping) { 1149219974Smav G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); 1150219974Smav wakeup(sc); 1151219974Smav } 1152219974Smav} 1153219974Smav 1154219974Smavstatic int 1155219974Smavg_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) 1156219974Smav{ 1157219974Smav /* 1158219974Smav * 5 cases: 1159219974Smav * (1) bp entirely below NO 1160219974Smav * (2) bp entirely above NO 1161219974Smav * (3) bp start below, but end in range YES 1162219974Smav * (4) bp entirely within YES 1163219974Smav * (5) bp starts within, ends above YES 1164219974Smav * 1165219974Smav * lock range 10-19 (offset 10 length 10) 1166219974Smav * (1) 1-5: first if kicks it out 1167219974Smav * (2) 30-35: second if kicks it out 1168219974Smav * (3) 5-15: passes both ifs 1169219974Smav * (4) 12-14: passes both ifs 1170219974Smav * (5) 19-20: passes both 1171219974Smav */ 1172219974Smav off_t lend = lstart + len - 1; 1173219974Smav off_t bstart = bp->bio_offset; 1174219974Smav off_t bend = bp->bio_offset + bp->bio_length - 1; 1175219974Smav 1176219974Smav if (bend < lstart) 1177219974Smav return (0); 1178219974Smav if (lend < bstart) 1179219974Smav return (0); 1180219974Smav return (1); 1181219974Smav} 1182219974Smav 1183219974Smavstatic int 1184219974Smavg_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) 1185219974Smav{ 1186219974Smav struct g_raid_lock *lp; 1187219974Smav 1188219974Smav sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); 1189219974Smav 1190219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1191219974Smav if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) 1192219974Smav return (1); 1193219974Smav } 1194219974Smav return (0); 1195219974Smav} 1196219974Smav 1197219974Smavstatic void 1198219974Smavg_raid_start_request(struct bio *bp) 1199219974Smav{ 1200219974Smav struct g_raid_softc *sc; 1201219974Smav struct g_raid_volume *vol; 1202219974Smav 1203219974Smav sc = bp->bio_to->geom->softc; 1204219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1205219974Smav vol = bp->bio_to->private; 1206219974Smav 1207219974Smav /* 1208219974Smav * Check to see if this item is in a locked range. If so, 1209219974Smav * queue it to our locked queue and return. We'll requeue 1210219974Smav * it when the range is unlocked. Internal I/O for the 1211219974Smav * rebuild/rescan/recovery process is excluded from this 1212219974Smav * check so we can actually do the recovery. 1213219974Smav */ 1214219974Smav if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && 1215219974Smav g_raid_is_in_locked_range(vol, bp)) { 1216219974Smav G_RAID_LOGREQ(3, bp, "Defer request."); 1217219974Smav bioq_insert_tail(&vol->v_locked, bp); 1218219974Smav return; 1219219974Smav } 1220219974Smav 1221219974Smav /* 1222219974Smav * If we're actually going to do the write/delete, then 1223219974Smav * update the idle stats for the volume. 1224219974Smav */ 1225219974Smav if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1226219974Smav if (!vol->v_dirty) 1227219974Smav g_raid_dirty(vol); 1228219974Smav vol->v_writes++; 1229219974Smav } 1230219974Smav 1231219974Smav /* 1232219974Smav * Put request onto inflight queue, so we can check if new 1233219974Smav * synchronization requests don't collide with it. Then tell 1234219974Smav * the transformation layer to start the I/O. 1235219974Smav */ 1236219974Smav bioq_insert_tail(&vol->v_inflight, bp); 1237219974Smav G_RAID_LOGREQ(4, bp, "Request started"); 1238219974Smav G_RAID_TR_IOSTART(vol->v_tr, bp); 1239219974Smav} 1240219974Smav 1241219974Smavstatic void 1242219974Smavg_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) 1243219974Smav{ 1244219974Smav off_t off, len; 1245219974Smav struct bio *nbp; 1246219974Smav struct g_raid_lock *lp; 1247219974Smav 1248219974Smav vol->v_pending_lock = 0; 1249219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1250219974Smav if (lp->l_pending) { 1251219974Smav off = lp->l_offset; 1252219974Smav len = lp->l_length; 1253219974Smav lp->l_pending = 0; 1254219974Smav TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { 1255219974Smav if (g_raid_bio_overlaps(nbp, off, len)) 1256219974Smav lp->l_pending++; 1257219974Smav } 1258219974Smav if (lp->l_pending) { 1259219974Smav vol->v_pending_lock = 1; 1260219974Smav G_RAID_DEBUG1(4, vol->v_softc, 1261219974Smav "Deferred lock(%jd, %jd) has %d pending", 1262219974Smav (intmax_t)off, (intmax_t)(off + len), 1263219974Smav lp->l_pending); 1264219974Smav continue; 1265219974Smav } 1266219974Smav G_RAID_DEBUG1(4, vol->v_softc, 1267219974Smav "Deferred lock of %jd to %jd completed", 1268219974Smav (intmax_t)off, (intmax_t)(off + len)); 1269219974Smav G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1270219974Smav } 1271219974Smav } 1272219974Smav} 1273219974Smav 1274219974Smavvoid 1275219974Smavg_raid_iodone(struct bio *bp, int error) 1276219974Smav{ 1277219974Smav struct g_raid_softc *sc; 1278219974Smav struct g_raid_volume *vol; 1279219974Smav 1280219974Smav sc = bp->bio_to->geom->softc; 1281219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1282219974Smav vol = bp->bio_to->private; 1283219974Smav G_RAID_LOGREQ(3, bp, "Request done: %d.", error); 1284219974Smav 1285219974Smav /* Update stats if we done write/delete. */ 1286219974Smav if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 1287219974Smav vol->v_writes--; 1288219974Smav vol->v_last_write = time_uptime; 1289219974Smav } 1290219974Smav 1291219974Smav bioq_remove(&vol->v_inflight, bp); 1292219974Smav if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) 1293219974Smav g_raid_finish_with_locked_ranges(vol, bp); 1294219974Smav getmicrouptime(&vol->v_last_done); 1295219974Smav g_io_deliver(bp, error); 1296219974Smav} 1297219974Smav 1298219974Smavint 1299219974Smavg_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, 1300219974Smav struct bio *ignore, void *argp) 1301219974Smav{ 1302219974Smav struct g_raid_softc *sc; 1303219974Smav struct g_raid_lock *lp; 1304219974Smav struct bio *bp; 1305219974Smav 1306219974Smav sc = vol->v_softc; 1307219974Smav lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); 1308219974Smav LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); 1309219974Smav lp->l_offset = off; 1310219974Smav lp->l_length = len; 1311219974Smav lp->l_callback_arg = argp; 1312219974Smav 1313219974Smav lp->l_pending = 0; 1314219974Smav TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { 1315219974Smav if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) 1316219974Smav lp->l_pending++; 1317219974Smav } 1318219974Smav 1319219974Smav /* 1320219974Smav * If there are any writes that are pending, we return EBUSY. All 1321219974Smav * callers will have to wait until all pending writes clear. 1322219974Smav */ 1323219974Smav if (lp->l_pending > 0) { 1324219974Smav vol->v_pending_lock = 1; 1325219974Smav G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", 1326219974Smav (intmax_t)off, (intmax_t)(off+len), lp->l_pending); 1327219974Smav return (EBUSY); 1328219974Smav } 1329219974Smav G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", 1330219974Smav (intmax_t)off, (intmax_t)(off+len)); 1331219974Smav G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 1332219974Smav return (0); 1333219974Smav} 1334219974Smav 1335219974Smavint 1336219974Smavg_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) 1337219974Smav{ 1338219974Smav struct g_raid_lock *lp; 1339219974Smav struct g_raid_softc *sc; 1340219974Smav struct bio *bp; 1341219974Smav 1342219974Smav sc = vol->v_softc; 1343219974Smav LIST_FOREACH(lp, &vol->v_locks, l_next) { 1344219974Smav if (lp->l_offset == off && lp->l_length == len) { 1345219974Smav LIST_REMOVE(lp, l_next); 1346219974Smav /* XXX 1347219974Smav * Right now we just put them all back on the queue 1348219974Smav * and hope for the best. We hope this because any 1349219974Smav * locked ranges will go right back on this list 1350219974Smav * when the worker thread runs. 1351219974Smav * XXX 1352219974Smav */ 1353219974Smav G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", 1354219974Smav (intmax_t)lp->l_offset, 1355219974Smav (intmax_t)(lp->l_offset+lp->l_length)); 1356219974Smav mtx_lock(&sc->sc_queue_mtx); 1357219974Smav while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) 1358219974Smav bioq_disksort(&sc->sc_queue, bp); 1359219974Smav mtx_unlock(&sc->sc_queue_mtx); 1360219974Smav free(lp, M_RAID); 1361219974Smav return (0); 1362219974Smav } 1363219974Smav } 1364219974Smav return (EINVAL); 1365219974Smav} 1366219974Smav 1367219974Smavvoid 1368219974Smavg_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) 1369219974Smav{ 1370219974Smav struct g_consumer *cp; 1371219974Smav struct g_raid_disk *disk, *tdisk; 1372219974Smav 1373219974Smav bp->bio_caller1 = sd; 1374219974Smav 1375219974Smav /* 1376219974Smav * Make sure that the disk is present. Generally it is a task of 1377219974Smav * transformation layers to not send requests to absent disks, but 1378219974Smav * it is better to be safe and report situation then sorry. 1379219974Smav */ 1380219974Smav if (sd->sd_disk == NULL) { 1381219974Smav G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); 1382219974Smavnodisk: 1383219974Smav bp->bio_from = NULL; 1384219974Smav bp->bio_to = NULL; 1385219974Smav bp->bio_error = ENXIO; 1386219974Smav g_raid_disk_done(bp); 1387219974Smav return; 1388219974Smav } 1389219974Smav disk = sd->sd_disk; 1390219974Smav if (disk->d_state != G_RAID_DISK_S_ACTIVE && 1391219974Smav disk->d_state != G_RAID_DISK_S_FAILED) { 1392219974Smav G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " 1393219974Smav "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 1394219974Smav goto nodisk; 1395219974Smav } 1396219974Smav 1397219974Smav cp = disk->d_consumer; 1398219974Smav bp->bio_from = cp; 1399219974Smav bp->bio_to = cp->provider; 1400219974Smav cp->index++; 1401219974Smav 1402219974Smav /* Update average disks load. */ 1403219974Smav TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { 1404219974Smav if (tdisk->d_consumer == NULL) 1405219974Smav tdisk->d_load = 0; 1406219974Smav else 1407219974Smav tdisk->d_load = (tdisk->d_consumer->index * 1408219974Smav G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; 1409219974Smav } 1410219974Smav 1411219974Smav disk->d_last_offset = bp->bio_offset + bp->bio_length; 1412219974Smav if (dumping) { 1413219974Smav G_RAID_LOGREQ(3, bp, "Sending dumping request."); 1414219974Smav if (bp->bio_cmd == BIO_WRITE) { 1415219974Smav bp->bio_error = g_raid_subdisk_kerneldump(sd, 1416219974Smav bp->bio_data, 0, bp->bio_offset, bp->bio_length); 1417219974Smav } else 1418219974Smav bp->bio_error = EOPNOTSUPP; 1419219974Smav g_raid_disk_done(bp); 1420219974Smav } else { 1421219974Smav bp->bio_done = g_raid_disk_done; 1422219974Smav bp->bio_offset += sd->sd_offset; 1423219974Smav G_RAID_LOGREQ(3, bp, "Sending request."); 1424219974Smav g_io_request(bp, cp); 1425219974Smav } 1426219974Smav} 1427219974Smav 1428219974Smavint 1429219974Smavg_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, 1430219974Smav void *virtual, vm_offset_t physical, off_t offset, size_t length) 1431219974Smav{ 1432219974Smav 1433219974Smav if (sd->sd_disk == NULL) 1434219974Smav return (ENXIO); 1435219974Smav if (sd->sd_disk->d_kd.di.dumper == NULL) 1436219974Smav return (EOPNOTSUPP); 1437219974Smav return (dump_write(&sd->sd_disk->d_kd.di, 1438219974Smav virtual, physical, 1439219974Smav sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, 1440219974Smav length)); 1441219974Smav} 1442219974Smav 1443219974Smavstatic void 1444219974Smavg_raid_disk_done(struct bio *bp) 1445219974Smav{ 1446219974Smav struct g_raid_softc *sc; 1447219974Smav struct g_raid_subdisk *sd; 1448219974Smav 1449219974Smav sd = bp->bio_caller1; 1450219974Smav sc = sd->sd_softc; 1451219974Smav mtx_lock(&sc->sc_queue_mtx); 1452219974Smav bioq_disksort(&sc->sc_queue, bp); 1453219974Smav mtx_unlock(&sc->sc_queue_mtx); 1454219974Smav if (!dumping) 1455219974Smav wakeup(sc); 1456219974Smav} 1457219974Smav 1458219974Smavstatic void 1459219974Smavg_raid_disk_done_request(struct bio *bp) 1460219974Smav{ 1461219974Smav struct g_raid_softc *sc; 1462219974Smav struct g_raid_disk *disk; 1463219974Smav struct g_raid_subdisk *sd; 1464219974Smav struct g_raid_volume *vol; 1465219974Smav 1466219974Smav g_topology_assert_not(); 1467219974Smav 1468219974Smav G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); 1469219974Smav sd = bp->bio_caller1; 1470219974Smav sc = sd->sd_softc; 1471219974Smav vol = sd->sd_volume; 1472219974Smav if (bp->bio_from != NULL) { 1473219974Smav bp->bio_from->index--; 1474219974Smav disk = bp->bio_from->private; 1475219974Smav if (disk == NULL) 1476219974Smav g_raid_kill_consumer(sc, bp->bio_from); 1477219974Smav } 1478219974Smav bp->bio_offset -= sd->sd_offset; 1479219974Smav 1480219974Smav G_RAID_TR_IODONE(vol->v_tr, sd, bp); 1481219974Smav} 1482219974Smav 1483219974Smavstatic void 1484219974Smavg_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) 1485219974Smav{ 1486219974Smav 1487219974Smav if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) 1488219974Smav ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); 1489219974Smav else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) 1490219974Smav ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); 1491219974Smav else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) 1492219974Smav ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); 1493219974Smav else 1494219974Smav ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); 1495219974Smav if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { 1496219974Smav KASSERT(ep->e_error == 0, 1497219974Smav ("Error cannot be handled.")); 1498219974Smav g_raid_event_free(ep); 1499219974Smav } else { 1500219974Smav ep->e_flags |= G_RAID_EVENT_DONE; 1501219974Smav G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); 1502219974Smav mtx_lock(&sc->sc_queue_mtx); 1503219974Smav wakeup(ep); 1504219974Smav mtx_unlock(&sc->sc_queue_mtx); 1505219974Smav } 1506219974Smav} 1507219974Smav 1508219974Smav/* 1509219974Smav * Worker thread. 1510219974Smav */ 1511219974Smavstatic void 1512219974Smavg_raid_worker(void *arg) 1513219974Smav{ 1514219974Smav struct g_raid_softc *sc; 1515219974Smav struct g_raid_event *ep; 1516219974Smav struct g_raid_volume *vol; 1517219974Smav struct bio *bp; 1518219974Smav struct timeval now, t; 1519219974Smav int timeout, rv; 1520219974Smav 1521219974Smav sc = arg; 1522219974Smav thread_lock(curthread); 1523219974Smav sched_prio(curthread, PRIBIO); 1524219974Smav thread_unlock(curthread); 1525219974Smav 1526219974Smav sx_xlock(&sc->sc_lock); 1527219974Smav for (;;) { 1528219974Smav mtx_lock(&sc->sc_queue_mtx); 1529219974Smav /* 1530219974Smav * First take a look at events. 1531219974Smav * This is important to handle events before any I/O requests. 1532219974Smav */ 1533219974Smav bp = NULL; 1534219974Smav vol = NULL; 1535219974Smav rv = 0; 1536219974Smav ep = TAILQ_FIRST(&sc->sc_events); 1537219974Smav if (ep != NULL) 1538219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1539219974Smav else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) 1540219974Smav ; 1541219974Smav else { 1542219974Smav getmicrouptime(&now); 1543219974Smav t = now; 1544219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1545219974Smav if (bioq_first(&vol->v_inflight) == NULL && 1546219974Smav vol->v_tr && 1547219974Smav timevalcmp(&vol->v_last_done, &t, < )) 1548219974Smav t = vol->v_last_done; 1549219974Smav } 1550219974Smav timevalsub(&t, &now); 1551219974Smav timeout = g_raid_idle_threshold + 1552219974Smav t.tv_sec * 1000000 + t.tv_usec; 1553219974Smav if (timeout > 0) { 1554219974Smav /* 1555219974Smav * Two steps to avoid overflows at HZ=1000 1556219974Smav * and idle timeouts > 2.1s. Some rounding 1557219974Smav * errors can occur, but they are < 1tick, 1558219974Smav * which is deemed to be close enough for 1559219974Smav * this purpose. 1560219974Smav */ 1561219974Smav int micpertic = 1000000 / hz; 1562219974Smav timeout = (timeout + micpertic - 1) / micpertic; 1563219974Smav sx_xunlock(&sc->sc_lock); 1564219974Smav MSLEEP(rv, sc, &sc->sc_queue_mtx, 1565219974Smav PRIBIO | PDROP, "-", timeout); 1566219974Smav sx_xlock(&sc->sc_lock); 1567219974Smav goto process; 1568219974Smav } else 1569219974Smav rv = EWOULDBLOCK; 1570219974Smav } 1571219974Smav mtx_unlock(&sc->sc_queue_mtx); 1572219974Smavprocess: 1573219974Smav if (ep != NULL) { 1574219974Smav g_raid_handle_event(sc, ep); 1575219974Smav } else if (bp != NULL) { 1576219974Smav if (bp->bio_to != NULL && 1577219974Smav bp->bio_to->geom == sc->sc_geom) 1578219974Smav g_raid_start_request(bp); 1579219974Smav else 1580219974Smav g_raid_disk_done_request(bp); 1581219974Smav } else if (rv == EWOULDBLOCK) { 1582219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 1583242314Smav g_raid_clean(vol, -1); 1584219974Smav if (bioq_first(&vol->v_inflight) == NULL && 1585219974Smav vol->v_tr) { 1586219974Smav t.tv_sec = g_raid_idle_threshold / 1000000; 1587219974Smav t.tv_usec = g_raid_idle_threshold % 1000000; 1588219974Smav timevaladd(&t, &vol->v_last_done); 1589219974Smav getmicrouptime(&now); 1590219974Smav if (timevalcmp(&t, &now, <= )) { 1591219974Smav G_RAID_TR_IDLE(vol->v_tr); 1592219974Smav vol->v_last_done = now; 1593219974Smav } 1594219974Smav } 1595219974Smav } 1596219974Smav } 1597219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) 1598219974Smav g_raid_destroy_node(sc, 1); /* May not return. */ 1599219974Smav } 1600219974Smav} 1601219974Smav 1602219974Smavstatic void 1603219974Smavg_raid_poll(struct g_raid_softc *sc) 1604219974Smav{ 1605219974Smav struct g_raid_event *ep; 1606219974Smav struct bio *bp; 1607219974Smav 1608219974Smav sx_xlock(&sc->sc_lock); 1609219974Smav mtx_lock(&sc->sc_queue_mtx); 1610219974Smav /* 1611219974Smav * First take a look at events. 1612219974Smav * This is important to handle events before any I/O requests. 1613219974Smav */ 1614219974Smav ep = TAILQ_FIRST(&sc->sc_events); 1615219974Smav if (ep != NULL) { 1616219974Smav TAILQ_REMOVE(&sc->sc_events, ep, e_next); 1617219974Smav mtx_unlock(&sc->sc_queue_mtx); 1618219974Smav g_raid_handle_event(sc, ep); 1619219974Smav goto out; 1620219974Smav } 1621219974Smav bp = bioq_takefirst(&sc->sc_queue); 1622219974Smav if (bp != NULL) { 1623219974Smav mtx_unlock(&sc->sc_queue_mtx); 1624219974Smav if (bp->bio_from == NULL || 1625219974Smav bp->bio_from->geom != sc->sc_geom) 1626219974Smav g_raid_start_request(bp); 1627219974Smav else 1628219974Smav g_raid_disk_done_request(bp); 1629219974Smav } 1630219974Smavout: 1631219974Smav sx_xunlock(&sc->sc_lock); 1632219974Smav} 1633219974Smav 1634219974Smavstatic void 1635219974Smavg_raid_launch_provider(struct g_raid_volume *vol) 1636219974Smav{ 1637219974Smav struct g_raid_disk *disk; 1638260385Sscottl struct g_raid_subdisk *sd; 1639219974Smav struct g_raid_softc *sc; 1640219974Smav struct g_provider *pp; 1641219974Smav char name[G_RAID_MAX_VOLUMENAME]; 1642248068Ssbruno char announce_buf[80], buf1[32]; 1643219974Smav off_t off; 1644260385Sscottl int i; 1645219974Smav 1646219974Smav sc = vol->v_softc; 1647219974Smav sx_assert(&sc->sc_lock, SX_LOCKED); 1648219974Smav 1649219974Smav g_topology_lock(); 1650219974Smav /* Try to name provider with volume name. */ 1651219974Smav snprintf(name, sizeof(name), "raid/%s", vol->v_name); 1652219974Smav if (g_raid_name_format == 0 || vol->v_name[0] == 0 || 1653219974Smav g_provider_by_name(name) != NULL) { 1654219974Smav /* Otherwise use sequential volume number. */ 1655219974Smav snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); 1656219974Smav } 1657248068Ssbruno 1658248068Ssbruno /* 1659248068Ssbruno * Create a /dev/ar%d that the old ataraid(4) stack once 1660248068Ssbruno * created as an alias for /dev/raid/r%d if requested. 1661248068Ssbruno * This helps going from stable/7 ataraid devices to newer 1662248068Ssbruno * FreeBSD releases. sbruno 07 MAY 2013 1663248068Ssbruno */ 1664248068Ssbruno 1665248068Ssbruno if (ar_legacy_aliases) { 1666248068Ssbruno snprintf(announce_buf, sizeof(announce_buf), 1667248068Ssbruno "kern.devalias.%s", name); 1668248068Ssbruno snprintf(buf1, sizeof(buf1), 1669248068Ssbruno "ar%d", vol->v_global_id); 1670248068Ssbruno setenv(announce_buf, buf1); 1671248068Ssbruno } 1672248068Ssbruno 1673219974Smav pp = g_new_providerf(sc->sc_geom, "%s", name); 1674260385Sscottl pp->flags |= G_PF_DIRECT_RECEIVE; 1675260385Sscottl if (vol->v_tr->tro_class->trc_accept_unmapped) { 1676260385Sscottl pp->flags |= G_PF_ACCEPT_UNMAPPED; 1677260385Sscottl for (i = 0; i < vol->v_disks_count; i++) { 1678260385Sscottl sd = &vol->v_subdisks[i]; 1679260385Sscottl if (sd->sd_state == G_RAID_SUBDISK_S_NONE) 1680260385Sscottl continue; 1681260385Sscottl if ((sd->sd_disk->d_consumer->provider->flags & 1682260385Sscottl G_PF_ACCEPT_UNMAPPED) == 0) 1683260385Sscottl pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 1684260385Sscottl } 1685260385Sscottl } 1686219974Smav pp->private = vol; 1687219974Smav pp->mediasize = vol->v_mediasize; 1688219974Smav pp->sectorsize = vol->v_sectorsize; 1689219974Smav pp->stripesize = 0; 1690219974Smav pp->stripeoffset = 0; 1691219974Smav if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 1692219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || 1693219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || 1694219974Smav vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { 1695219974Smav if ((disk = vol->v_subdisks[0].sd_disk) != NULL && 1696219974Smav disk->d_consumer != NULL && 1697219974Smav disk->d_consumer->provider != NULL) { 1698219974Smav pp->stripesize = disk->d_consumer->provider->stripesize; 1699219974Smav off = disk->d_consumer->provider->stripeoffset; 1700219974Smav pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; 1701219974Smav if (off > 0) 1702219974Smav pp->stripeoffset %= off; 1703219974Smav } 1704219974Smav if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { 1705219974Smav pp->stripesize *= (vol->v_disks_count - 1); 1706219974Smav pp->stripeoffset *= (vol->v_disks_count - 1); 1707219974Smav } 1708219974Smav } else 1709219974Smav pp->stripesize = vol->v_strip_size; 1710219974Smav vol->v_provider = pp; 1711219974Smav g_error_provider(pp, 0); 1712219974Smav g_topology_unlock(); 1713219974Smav G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", 1714219974Smav pp->name, vol->v_name); 1715219974Smav} 1716219974Smav 1717219974Smavstatic void 1718219974Smavg_raid_destroy_provider(struct g_raid_volume *vol) 1719219974Smav{ 1720219974Smav struct g_raid_softc *sc; 1721219974Smav struct g_provider *pp; 1722219974Smav struct bio *bp, *tmp; 1723219974Smav 1724219974Smav g_topology_assert_not(); 1725219974Smav sc = vol->v_softc; 1726219974Smav pp = vol->v_provider; 1727219974Smav KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); 1728219974Smav 1729219974Smav g_topology_lock(); 1730219974Smav g_error_provider(pp, ENXIO); 1731219974Smav mtx_lock(&sc->sc_queue_mtx); 1732219974Smav TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { 1733219974Smav if (bp->bio_to != pp) 1734219974Smav continue; 1735219974Smav bioq_remove(&sc->sc_queue, bp); 1736219974Smav g_io_deliver(bp, ENXIO); 1737219974Smav } 1738219974Smav mtx_unlock(&sc->sc_queue_mtx); 1739219974Smav G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", 1740219974Smav pp->name, vol->v_name); 1741219974Smav g_wither_provider(pp, ENXIO); 1742219974Smav g_topology_unlock(); 1743219974Smav vol->v_provider = NULL; 1744219974Smav} 1745219974Smav 1746219974Smav/* 1747219974Smav * Update device state. 1748219974Smav */ 1749219974Smavstatic int 1750219974Smavg_raid_update_volume(struct g_raid_volume *vol, u_int event) 1751219974Smav{ 1752219974Smav struct g_raid_softc *sc; 1753219974Smav 1754219974Smav sc = vol->v_softc; 1755219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1756219974Smav 1757219974Smav G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", 1758219974Smav g_raid_volume_event2str(event), 1759219974Smav vol->v_name); 1760219974Smav switch (event) { 1761219974Smav case G_RAID_VOLUME_E_DOWN: 1762219974Smav if (vol->v_provider != NULL) 1763219974Smav g_raid_destroy_provider(vol); 1764219974Smav break; 1765219974Smav case G_RAID_VOLUME_E_UP: 1766219974Smav if (vol->v_provider == NULL) 1767219974Smav g_raid_launch_provider(vol); 1768219974Smav break; 1769219974Smav case G_RAID_VOLUME_E_START: 1770219974Smav if (vol->v_tr) 1771219974Smav G_RAID_TR_START(vol->v_tr); 1772219974Smav return (0); 1773219974Smav default: 1774219974Smav if (sc->sc_md) 1775219974Smav G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); 1776219974Smav return (0); 1777219974Smav } 1778219974Smav 1779219974Smav /* Manage root mount release. */ 1780219974Smav if (vol->v_starting) { 1781219974Smav vol->v_starting = 0; 1782219974Smav G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); 1783219974Smav root_mount_rel(vol->v_rootmount); 1784219974Smav vol->v_rootmount = NULL; 1785219974Smav } 1786219974Smav if (vol->v_stopping && vol->v_provider_open == 0) 1787219974Smav g_raid_destroy_volume(vol); 1788219974Smav return (0); 1789219974Smav} 1790219974Smav 1791219974Smav/* 1792219974Smav * Update subdisk state. 1793219974Smav */ 1794219974Smavstatic int 1795219974Smavg_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) 1796219974Smav{ 1797219974Smav struct g_raid_softc *sc; 1798219974Smav struct g_raid_volume *vol; 1799219974Smav 1800219974Smav sc = sd->sd_softc; 1801219974Smav vol = sd->sd_volume; 1802219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1803219974Smav 1804219974Smav G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", 1805219974Smav g_raid_subdisk_event2str(event), 1806219974Smav vol->v_name, sd->sd_pos, 1807219974Smav sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); 1808219974Smav if (vol->v_tr) 1809219974Smav G_RAID_TR_EVENT(vol->v_tr, sd, event); 1810219974Smav 1811219974Smav return (0); 1812219974Smav} 1813219974Smav 1814219974Smav/* 1815219974Smav * Update disk state. 1816219974Smav */ 1817219974Smavstatic int 1818219974Smavg_raid_update_disk(struct g_raid_disk *disk, u_int event) 1819219974Smav{ 1820219974Smav struct g_raid_softc *sc; 1821219974Smav 1822219974Smav sc = disk->d_softc; 1823219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1824219974Smav 1825219974Smav G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", 1826219974Smav g_raid_disk_event2str(event), 1827219974Smav g_raid_get_diskname(disk)); 1828219974Smav 1829219974Smav if (sc->sc_md) 1830219974Smav G_RAID_MD_EVENT(sc->sc_md, disk, event); 1831219974Smav return (0); 1832219974Smav} 1833219974Smav 1834219974Smav/* 1835219974Smav * Node event. 1836219974Smav */ 1837219974Smavstatic int 1838219974Smavg_raid_update_node(struct g_raid_softc *sc, u_int event) 1839219974Smav{ 1840219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 1841219974Smav 1842219974Smav G_RAID_DEBUG1(2, sc, "Event %s for the array.", 1843219974Smav g_raid_node_event2str(event)); 1844219974Smav 1845219974Smav if (event == G_RAID_NODE_E_WAKE) 1846219974Smav return (0); 1847219974Smav if (sc->sc_md) 1848219974Smav G_RAID_MD_EVENT(sc->sc_md, NULL, event); 1849219974Smav return (0); 1850219974Smav} 1851219974Smav 1852219974Smavstatic int 1853219974Smavg_raid_access(struct g_provider *pp, int acr, int acw, int ace) 1854219974Smav{ 1855219974Smav struct g_raid_volume *vol; 1856219974Smav struct g_raid_softc *sc; 1857220210Smav int dcw, opens, error = 0; 1858219974Smav 1859219974Smav g_topology_assert(); 1860219974Smav sc = pp->geom->softc; 1861219974Smav vol = pp->private; 1862219974Smav KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 1863219974Smav KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); 1864219974Smav 1865219974Smav G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, 1866219974Smav acr, acw, ace); 1867219974Smav dcw = pp->acw + acw; 1868219974Smav 1869219974Smav g_topology_unlock(); 1870219974Smav sx_xlock(&sc->sc_lock); 1871219974Smav /* Deny new opens while dying. */ 1872219974Smav if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { 1873219974Smav error = ENXIO; 1874219974Smav goto out; 1875219974Smav } 1876254275Smav /* Deny write opens for read-only volumes. */ 1877254275Smav if (vol->v_read_only && acw > 0) { 1878254275Smav error = EROFS; 1879254275Smav goto out; 1880254275Smav } 1881242314Smav if (dcw == 0) 1882219974Smav g_raid_clean(vol, dcw); 1883219974Smav vol->v_provider_open += acr + acw + ace; 1884219974Smav /* Handle delayed node destruction. */ 1885219974Smav if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && 1886219974Smav vol->v_provider_open == 0) { 1887219974Smav /* Count open volumes. */ 1888219974Smav opens = g_raid_nopens(sc); 1889219974Smav if (opens == 0) { 1890219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 1891219974Smav /* Wake up worker to make it selfdestruct. */ 1892219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 1893219974Smav } 1894219974Smav } 1895219974Smav /* Handle open volume destruction. */ 1896219974Smav if (vol->v_stopping && vol->v_provider_open == 0) 1897219974Smav g_raid_destroy_volume(vol); 1898219974Smavout: 1899219974Smav sx_xunlock(&sc->sc_lock); 1900219974Smav g_topology_lock(); 1901219974Smav return (error); 1902219974Smav} 1903219974Smav 1904219974Smavstruct g_raid_softc * 1905219974Smavg_raid_create_node(struct g_class *mp, 1906219974Smav const char *name, struct g_raid_md_object *md) 1907219974Smav{ 1908219974Smav struct g_raid_softc *sc; 1909219974Smav struct g_geom *gp; 1910219974Smav int error; 1911219974Smav 1912219974Smav g_topology_assert(); 1913219974Smav G_RAID_DEBUG(1, "Creating array %s.", name); 1914219974Smav 1915219974Smav gp = g_new_geomf(mp, "%s", name); 1916219974Smav sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); 1917219974Smav gp->start = g_raid_start; 1918219974Smav gp->orphan = g_raid_orphan; 1919219974Smav gp->access = g_raid_access; 1920219974Smav gp->dumpconf = g_raid_dumpconf; 1921219974Smav 1922219974Smav sc->sc_md = md; 1923219974Smav sc->sc_geom = gp; 1924219974Smav sc->sc_flags = 0; 1925219974Smav TAILQ_INIT(&sc->sc_volumes); 1926219974Smav TAILQ_INIT(&sc->sc_disks); 1927234816Smav sx_init(&sc->sc_lock, "graid:lock"); 1928234816Smav mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF); 1929219974Smav TAILQ_INIT(&sc->sc_events); 1930219974Smav bioq_init(&sc->sc_queue); 1931219974Smav gp->softc = sc; 1932219974Smav error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, 1933219974Smav "g_raid %s", name); 1934219974Smav if (error != 0) { 1935219974Smav G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); 1936219974Smav mtx_destroy(&sc->sc_queue_mtx); 1937219974Smav sx_destroy(&sc->sc_lock); 1938219974Smav g_destroy_geom(sc->sc_geom); 1939219974Smav free(sc, M_RAID); 1940219974Smav return (NULL); 1941219974Smav } 1942219974Smav 1943219974Smav G_RAID_DEBUG1(0, sc, "Array %s created.", name); 1944219974Smav return (sc); 1945219974Smav} 1946219974Smav 1947219974Smavstruct g_raid_volume * 1948219974Smavg_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) 1949219974Smav{ 1950219974Smav struct g_raid_volume *vol, *vol1; 1951219974Smav int i; 1952219974Smav 1953219974Smav G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); 1954219974Smav vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); 1955219974Smav vol->v_softc = sc; 1956219974Smav strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); 1957219974Smav vol->v_state = G_RAID_VOLUME_S_STARTING; 1958219974Smav vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 1959219974Smav vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; 1960235076Smav vol->v_rotate_parity = 1; 1961219974Smav bioq_init(&vol->v_inflight); 1962219974Smav bioq_init(&vol->v_locked); 1963219974Smav LIST_INIT(&vol->v_locks); 1964219974Smav for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 1965219974Smav vol->v_subdisks[i].sd_softc = sc; 1966219974Smav vol->v_subdisks[i].sd_volume = vol; 1967219974Smav vol->v_subdisks[i].sd_pos = i; 1968219974Smav vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; 1969219974Smav } 1970219974Smav 1971219974Smav /* Find free ID for this volume. */ 1972219974Smav g_topology_lock(); 1973219974Smav vol1 = vol; 1974219974Smav if (id >= 0) { 1975219974Smav LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1976219974Smav if (vol1->v_global_id == id) 1977219974Smav break; 1978219974Smav } 1979219974Smav } 1980219974Smav if (vol1 != NULL) { 1981219974Smav for (id = 0; ; id++) { 1982219974Smav LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 1983219974Smav if (vol1->v_global_id == id) 1984219974Smav break; 1985219974Smav } 1986219974Smav if (vol1 == NULL) 1987219974Smav break; 1988219974Smav } 1989219974Smav } 1990219974Smav vol->v_global_id = id; 1991219974Smav LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); 1992219974Smav g_topology_unlock(); 1993219974Smav 1994219974Smav /* Delay root mounting. */ 1995219974Smav vol->v_rootmount = root_mount_hold("GRAID"); 1996219974Smav G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); 1997219974Smav vol->v_starting = 1; 1998219974Smav TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); 1999219974Smav return (vol); 2000219974Smav} 2001219974Smav 2002219974Smavstruct g_raid_disk * 2003219974Smavg_raid_create_disk(struct g_raid_softc *sc) 2004219974Smav{ 2005219974Smav struct g_raid_disk *disk; 2006219974Smav 2007219974Smav G_RAID_DEBUG1(1, sc, "Creating disk."); 2008219974Smav disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); 2009219974Smav disk->d_softc = sc; 2010219974Smav disk->d_state = G_RAID_DISK_S_NONE; 2011219974Smav TAILQ_INIT(&disk->d_subdisks); 2012219974Smav TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); 2013219974Smav return (disk); 2014219974Smav} 2015219974Smav 2016219974Smavint g_raid_start_volume(struct g_raid_volume *vol) 2017219974Smav{ 2018219974Smav struct g_raid_tr_class *class; 2019219974Smav struct g_raid_tr_object *obj; 2020219974Smav int status; 2021219974Smav 2022219974Smav G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); 2023219974Smav LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { 2024240465Smav if (!class->trc_enable) 2025240465Smav continue; 2026219974Smav G_RAID_DEBUG1(2, vol->v_softc, 2027219974Smav "Tasting volume %s for %s transformation.", 2028219974Smav vol->v_name, class->name); 2029219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2030219974Smav M_WAITOK); 2031219974Smav obj->tro_class = class; 2032219974Smav obj->tro_volume = vol; 2033219974Smav status = G_RAID_TR_TASTE(obj, vol); 2034219974Smav if (status != G_RAID_TR_TASTE_FAIL) 2035219974Smav break; 2036219974Smav kobj_delete((kobj_t)obj, M_RAID); 2037219974Smav } 2038219974Smav if (class == NULL) { 2039219974Smav G_RAID_DEBUG1(0, vol->v_softc, 2040219974Smav "No transformation module found for %s.", 2041219974Smav vol->v_name); 2042219974Smav vol->v_tr = NULL; 2043219974Smav g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); 2044219974Smav g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, 2045219974Smav G_RAID_EVENT_VOLUME); 2046219974Smav return (-1); 2047219974Smav } 2048219974Smav G_RAID_DEBUG1(2, vol->v_softc, 2049219974Smav "Transformation module %s chosen for %s.", 2050219974Smav class->name, vol->v_name); 2051219974Smav vol->v_tr = obj; 2052219974Smav return (0); 2053219974Smav} 2054219974Smav 2055219974Smavint 2056219974Smavg_raid_destroy_node(struct g_raid_softc *sc, int worker) 2057219974Smav{ 2058219974Smav struct g_raid_volume *vol, *tmpv; 2059219974Smav struct g_raid_disk *disk, *tmpd; 2060219974Smav int error = 0; 2061219974Smav 2062219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 2063219974Smav TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { 2064219974Smav if (g_raid_destroy_volume(vol)) 2065219974Smav error = EBUSY; 2066219974Smav } 2067219974Smav if (error) 2068219974Smav return (error); 2069219974Smav TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { 2070219974Smav if (g_raid_destroy_disk(disk)) 2071219974Smav error = EBUSY; 2072219974Smav } 2073219974Smav if (error) 2074219974Smav return (error); 2075219974Smav if (sc->sc_md) { 2076219974Smav G_RAID_MD_FREE(sc->sc_md); 2077219974Smav kobj_delete((kobj_t)sc->sc_md, M_RAID); 2078219974Smav sc->sc_md = NULL; 2079219974Smav } 2080219974Smav if (sc->sc_geom != NULL) { 2081219974Smav G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); 2082219974Smav g_topology_lock(); 2083219974Smav sc->sc_geom->softc = NULL; 2084219974Smav g_wither_geom(sc->sc_geom, ENXIO); 2085219974Smav g_topology_unlock(); 2086219974Smav sc->sc_geom = NULL; 2087219974Smav } else 2088219974Smav G_RAID_DEBUG(1, "Array destroyed."); 2089219974Smav if (worker) { 2090219974Smav g_raid_event_cancel(sc, sc); 2091219974Smav mtx_destroy(&sc->sc_queue_mtx); 2092219974Smav sx_xunlock(&sc->sc_lock); 2093219974Smav sx_destroy(&sc->sc_lock); 2094219974Smav wakeup(&sc->sc_stopping); 2095219974Smav free(sc, M_RAID); 2096219974Smav curthread->td_pflags &= ~TDP_GEOM; 2097219974Smav G_RAID_DEBUG(1, "Thread exiting."); 2098219974Smav kproc_exit(0); 2099219974Smav } else { 2100219974Smav /* Wake up worker to make it selfdestruct. */ 2101219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2102219974Smav } 2103219974Smav return (0); 2104219974Smav} 2105219974Smav 2106219974Smavint 2107219974Smavg_raid_destroy_volume(struct g_raid_volume *vol) 2108219974Smav{ 2109219974Smav struct g_raid_softc *sc; 2110219974Smav struct g_raid_disk *disk; 2111219974Smav int i; 2112219974Smav 2113219974Smav sc = vol->v_softc; 2114219974Smav G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); 2115219974Smav vol->v_stopping = 1; 2116219974Smav if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { 2117219974Smav if (vol->v_tr) { 2118219974Smav G_RAID_TR_STOP(vol->v_tr); 2119219974Smav return (EBUSY); 2120219974Smav } else 2121219974Smav vol->v_state = G_RAID_VOLUME_S_STOPPED; 2122219974Smav } 2123219974Smav if (g_raid_event_check(sc, vol) != 0) 2124219974Smav return (EBUSY); 2125219974Smav if (vol->v_provider != NULL) 2126219974Smav return (EBUSY); 2127219974Smav if (vol->v_provider_open != 0) 2128219974Smav return (EBUSY); 2129219974Smav if (vol->v_tr) { 2130219974Smav G_RAID_TR_FREE(vol->v_tr); 2131219974Smav kobj_delete((kobj_t)vol->v_tr, M_RAID); 2132219974Smav vol->v_tr = NULL; 2133219974Smav } 2134219974Smav if (vol->v_rootmount) 2135219974Smav root_mount_rel(vol->v_rootmount); 2136219974Smav g_topology_lock(); 2137219974Smav LIST_REMOVE(vol, v_global_next); 2138219974Smav g_topology_unlock(); 2139219974Smav TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); 2140219974Smav for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 2141219974Smav g_raid_event_cancel(sc, &vol->v_subdisks[i]); 2142219974Smav disk = vol->v_subdisks[i].sd_disk; 2143219974Smav if (disk == NULL) 2144219974Smav continue; 2145219974Smav TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); 2146219974Smav } 2147219974Smav G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); 2148219974Smav if (sc->sc_md) 2149219974Smav G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); 2150219974Smav g_raid_event_cancel(sc, vol); 2151219974Smav free(vol, M_RAID); 2152219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) { 2153219974Smav /* Wake up worker to let it selfdestruct. */ 2154219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2155219974Smav } 2156219974Smav return (0); 2157219974Smav} 2158219974Smav 2159219974Smavint 2160219974Smavg_raid_destroy_disk(struct g_raid_disk *disk) 2161219974Smav{ 2162219974Smav struct g_raid_softc *sc; 2163219974Smav struct g_raid_subdisk *sd, *tmp; 2164219974Smav 2165219974Smav sc = disk->d_softc; 2166219974Smav G_RAID_DEBUG1(2, sc, "Destroying disk."); 2167219974Smav if (disk->d_consumer) { 2168219974Smav g_raid_kill_consumer(sc, disk->d_consumer); 2169219974Smav disk->d_consumer = NULL; 2170219974Smav } 2171219974Smav TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { 2172219974Smav g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); 2173219974Smav g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 2174219974Smav G_RAID_EVENT_SUBDISK); 2175219974Smav TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); 2176219974Smav sd->sd_disk = NULL; 2177219974Smav } 2178219974Smav TAILQ_REMOVE(&sc->sc_disks, disk, d_next); 2179219974Smav if (sc->sc_md) 2180219974Smav G_RAID_MD_FREE_DISK(sc->sc_md, disk); 2181219974Smav g_raid_event_cancel(sc, disk); 2182219974Smav free(disk, M_RAID); 2183219974Smav return (0); 2184219974Smav} 2185219974Smav 2186219974Smavint 2187219974Smavg_raid_destroy(struct g_raid_softc *sc, int how) 2188219974Smav{ 2189253706Smav int error, opens; 2190219974Smav 2191219974Smav g_topology_assert_not(); 2192219974Smav if (sc == NULL) 2193219974Smav return (ENXIO); 2194219974Smav sx_assert(&sc->sc_lock, SX_XLOCKED); 2195219974Smav 2196219974Smav /* Count open volumes. */ 2197219974Smav opens = g_raid_nopens(sc); 2198219974Smav 2199219974Smav /* React on some opened volumes. */ 2200219974Smav if (opens > 0) { 2201219974Smav switch (how) { 2202219974Smav case G_RAID_DESTROY_SOFT: 2203219974Smav G_RAID_DEBUG1(1, sc, 2204219974Smav "%d volumes are still open.", 2205219974Smav opens); 2206253706Smav sx_xunlock(&sc->sc_lock); 2207219974Smav return (EBUSY); 2208219974Smav case G_RAID_DESTROY_DELAYED: 2209219974Smav G_RAID_DEBUG1(1, sc, 2210219974Smav "Array will be destroyed on last close."); 2211219974Smav sc->sc_stopping = G_RAID_DESTROY_DELAYED; 2212253706Smav sx_xunlock(&sc->sc_lock); 2213219974Smav return (EBUSY); 2214219974Smav case G_RAID_DESTROY_HARD: 2215219974Smav G_RAID_DEBUG1(1, sc, 2216219974Smav "%d volumes are still open.", 2217219974Smav opens); 2218219974Smav } 2219219974Smav } 2220219974Smav 2221219974Smav /* Mark node for destruction. */ 2222219974Smav sc->sc_stopping = G_RAID_DESTROY_HARD; 2223219974Smav /* Wake up worker to let it selfdestruct. */ 2224219974Smav g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 2225219974Smav /* Sleep until node destroyed. */ 2226253706Smav error = sx_sleep(&sc->sc_stopping, &sc->sc_lock, 2227253706Smav PRIBIO | PDROP, "r:destroy", hz * 3); 2228253706Smav return (error == EWOULDBLOCK ? EBUSY : 0); 2229219974Smav} 2230219974Smav 2231219974Smavstatic void 2232219974Smavg_raid_taste_orphan(struct g_consumer *cp) 2233219974Smav{ 2234219974Smav 2235219974Smav KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 2236219974Smav cp->provider->name)); 2237219974Smav} 2238219974Smav 2239219974Smavstatic struct g_geom * 2240219974Smavg_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 2241219974Smav{ 2242219974Smav struct g_consumer *cp; 2243219974Smav struct g_geom *gp, *geom; 2244219974Smav struct g_raid_md_class *class; 2245219974Smav struct g_raid_md_object *obj; 2246219974Smav int status; 2247219974Smav 2248219974Smav g_topology_assert(); 2249219974Smav g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2250240465Smav if (!g_raid_enable) 2251240465Smav return (NULL); 2252219974Smav G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); 2253219974Smav 2254265669Smav geom = NULL; 2255265669Smav status = G_RAID_MD_TASTE_FAIL; 2256234940Smav gp = g_new_geomf(mp, "raid:taste"); 2257219974Smav /* 2258219974Smav * This orphan function should be never called. 2259219974Smav */ 2260219974Smav gp->orphan = g_raid_taste_orphan; 2261219974Smav cp = g_new_consumer(gp); 2262260385Sscottl cp->flags |= G_CF_DIRECT_RECEIVE; 2263219974Smav g_attach(cp, pp); 2264265669Smav if (g_access(cp, 1, 0, 0) != 0) 2265265669Smav goto ofail; 2266219974Smav 2267219974Smav LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2268240465Smav if (!class->mdc_enable) 2269240465Smav continue; 2270219974Smav G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", 2271219974Smav pp->name, class->name); 2272219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2273219974Smav M_WAITOK); 2274219974Smav obj->mdo_class = class; 2275219974Smav status = G_RAID_MD_TASTE(obj, mp, cp, &geom); 2276219974Smav if (status != G_RAID_MD_TASTE_NEW) 2277219974Smav kobj_delete((kobj_t)obj, M_RAID); 2278219974Smav if (status != G_RAID_MD_TASTE_FAIL) 2279219974Smav break; 2280219974Smav } 2281219974Smav 2282265669Smav if (status == G_RAID_MD_TASTE_FAIL) 2283265669Smav (void)g_access(cp, -1, 0, 0); 2284265669Smavofail: 2285219974Smav g_detach(cp); 2286219974Smav g_destroy_consumer(cp); 2287219974Smav g_destroy_geom(gp); 2288219974Smav G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); 2289219974Smav return (geom); 2290219974Smav} 2291219974Smav 2292219974Smavint 2293234940Smavg_raid_create_node_format(const char *format, struct gctl_req *req, 2294234940Smav struct g_geom **gp) 2295219974Smav{ 2296219974Smav struct g_raid_md_class *class; 2297219974Smav struct g_raid_md_object *obj; 2298219974Smav int status; 2299219974Smav 2300219974Smav G_RAID_DEBUG(2, "Creating array for %s metadata.", format); 2301219974Smav LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2302219974Smav if (strcasecmp(class->name, format) == 0) 2303219974Smav break; 2304219974Smav } 2305219974Smav if (class == NULL) { 2306219974Smav G_RAID_DEBUG(1, "No support for %s metadata.", format); 2307219974Smav return (G_RAID_MD_TASTE_FAIL); 2308219974Smav } 2309219974Smav obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 2310219974Smav M_WAITOK); 2311219974Smav obj->mdo_class = class; 2312234940Smav status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp); 2313219974Smav if (status != G_RAID_MD_TASTE_NEW) 2314219974Smav kobj_delete((kobj_t)obj, M_RAID); 2315219974Smav return (status); 2316219974Smav} 2317219974Smav 2318219974Smavstatic int 2319219974Smavg_raid_destroy_geom(struct gctl_req *req __unused, 2320219974Smav struct g_class *mp __unused, struct g_geom *gp) 2321219974Smav{ 2322219974Smav struct g_raid_softc *sc; 2323219974Smav int error; 2324219974Smav 2325219974Smav g_topology_unlock(); 2326219974Smav sc = gp->softc; 2327219974Smav sx_xlock(&sc->sc_lock); 2328219974Smav g_cancel_event(sc); 2329219974Smav error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); 2330219974Smav g_topology_lock(); 2331219974Smav return (error); 2332219974Smav} 2333219974Smav 2334219974Smavvoid g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, 2335219974Smav struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2336219974Smav{ 2337219974Smav 2338219974Smav if (sc->sc_stopping == G_RAID_DESTROY_HARD) 2339219974Smav return; 2340219974Smav if (sc->sc_md) 2341219974Smav G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); 2342219974Smav} 2343219974Smav 2344219974Smavvoid g_raid_fail_disk(struct g_raid_softc *sc, 2345219974Smav struct g_raid_subdisk *sd, struct g_raid_disk *disk) 2346219974Smav{ 2347219974Smav 2348219974Smav if (disk == NULL) 2349219974Smav disk = sd->sd_disk; 2350219974Smav if (disk == NULL) { 2351219974Smav G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); 2352219974Smav return; 2353219974Smav } 2354219974Smav if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 2355219974Smav G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " 2356219974Smav "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 2357219974Smav return; 2358219974Smav } 2359219974Smav if (sc->sc_md) 2360219974Smav G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); 2361219974Smav} 2362219974Smav 2363219974Smavstatic void 2364219974Smavg_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 2365219974Smav struct g_consumer *cp, struct g_provider *pp) 2366219974Smav{ 2367219974Smav struct g_raid_softc *sc; 2368219974Smav struct g_raid_volume *vol; 2369219974Smav struct g_raid_subdisk *sd; 2370219974Smav struct g_raid_disk *disk; 2371219974Smav int i, s; 2372219974Smav 2373219974Smav g_topology_assert(); 2374219974Smav 2375219974Smav sc = gp->softc; 2376219974Smav if (sc == NULL) 2377219974Smav return; 2378219974Smav if (pp != NULL) { 2379219974Smav vol = pp->private; 2380219974Smav g_topology_unlock(); 2381219974Smav sx_xlock(&sc->sc_lock); 2382249974Smav sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent, 2383249974Smav sc->sc_md->mdo_class->name, 2384249974Smav g_raid_volume_level2str(vol->v_raid_level, 2385249974Smav vol->v_raid_level_qualifier)); 2386219974Smav sbuf_printf(sb, "%s<Label>%s</Label>\n", indent, 2387219974Smav vol->v_name); 2388219974Smav sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent, 2389219974Smav g_raid_volume_level2str(vol->v_raid_level, 2390219974Smav vol->v_raid_level_qualifier)); 2391219974Smav sbuf_printf(sb, 2392219974Smav "%s<Transformation>%s</Transformation>\n", indent, 2393219974Smav vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); 2394219974Smav sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 2395219974Smav vol->v_disks_count); 2396219974Smav sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent, 2397219974Smav vol->v_strip_size); 2398219974Smav sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2399219974Smav g_raid_volume_state2str(vol->v_state)); 2400219974Smav sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent, 2401219974Smav vol->v_dirty ? "Yes" : "No"); 2402219974Smav sbuf_printf(sb, "%s<Subdisks>", indent); 2403219974Smav for (i = 0; i < vol->v_disks_count; i++) { 2404219974Smav sd = &vol->v_subdisks[i]; 2405219974Smav if (sd->sd_disk != NULL && 2406219974Smav sd->sd_disk->d_consumer != NULL) { 2407219974Smav sbuf_printf(sb, "%s ", 2408219974Smav g_raid_get_diskname(sd->sd_disk)); 2409219974Smav } else { 2410219974Smav sbuf_printf(sb, "NONE "); 2411219974Smav } 2412219974Smav sbuf_printf(sb, "(%s", 2413219974Smav g_raid_subdisk_state2str(sd->sd_state)); 2414219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2415219974Smav sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2416219974Smav sbuf_printf(sb, " %d%%", 2417219974Smav (int)(sd->sd_rebuild_pos * 100 / 2418219974Smav sd->sd_size)); 2419219974Smav } 2420219974Smav sbuf_printf(sb, ")"); 2421219974Smav if (i + 1 < vol->v_disks_count) 2422219974Smav sbuf_printf(sb, ", "); 2423219974Smav } 2424219974Smav sbuf_printf(sb, "</Subdisks>\n"); 2425219974Smav sx_xunlock(&sc->sc_lock); 2426219974Smav g_topology_lock(); 2427219974Smav } else if (cp != NULL) { 2428219974Smav disk = cp->private; 2429219974Smav if (disk == NULL) 2430219974Smav return; 2431219974Smav g_topology_unlock(); 2432219974Smav sx_xlock(&sc->sc_lock); 2433219974Smav sbuf_printf(sb, "%s<State>%s", indent, 2434219974Smav g_raid_disk_state2str(disk->d_state)); 2435219974Smav if (!TAILQ_EMPTY(&disk->d_subdisks)) { 2436219974Smav sbuf_printf(sb, " ("); 2437219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2438219974Smav sbuf_printf(sb, "%s", 2439219974Smav g_raid_subdisk_state2str(sd->sd_state)); 2440219974Smav if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 2441219974Smav sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 2442219974Smav sbuf_printf(sb, " %d%%", 2443219974Smav (int)(sd->sd_rebuild_pos * 100 / 2444219974Smav sd->sd_size)); 2445219974Smav } 2446219974Smav if (TAILQ_NEXT(sd, sd_next)) 2447219974Smav sbuf_printf(sb, ", "); 2448219974Smav } 2449219974Smav sbuf_printf(sb, ")"); 2450219974Smav } 2451219974Smav sbuf_printf(sb, "</State>\n"); 2452219974Smav sbuf_printf(sb, "%s<Subdisks>", indent); 2453219974Smav TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 2454219974Smav sbuf_printf(sb, "r%d(%s):%d@%ju", 2455219974Smav sd->sd_volume->v_global_id, 2456219974Smav sd->sd_volume->v_name, 2457219974Smav sd->sd_pos, sd->sd_offset); 2458219974Smav if (TAILQ_NEXT(sd, sd_next)) 2459219974Smav sbuf_printf(sb, ", "); 2460219974Smav } 2461219974Smav sbuf_printf(sb, "</Subdisks>\n"); 2462219974Smav sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent, 2463219974Smav disk->d_read_errs); 2464219974Smav sx_xunlock(&sc->sc_lock); 2465219974Smav g_topology_lock(); 2466219974Smav } else { 2467219974Smav g_topology_unlock(); 2468219974Smav sx_xlock(&sc->sc_lock); 2469219974Smav if (sc->sc_md) { 2470219974Smav sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent, 2471219974Smav sc->sc_md->mdo_class->name); 2472219974Smav } 2473219974Smav if (!TAILQ_EMPTY(&sc->sc_volumes)) { 2474219974Smav s = 0xff; 2475219974Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 2476219974Smav if (vol->v_state < s) 2477219974Smav s = vol->v_state; 2478219974Smav } 2479219974Smav sbuf_printf(sb, "%s<State>%s</State>\n", indent, 2480219974Smav g_raid_volume_state2str(s)); 2481219974Smav } 2482219974Smav sx_xunlock(&sc->sc_lock); 2483219974Smav g_topology_lock(); 2484219974Smav } 2485219974Smav} 2486219974Smav 2487219974Smavstatic void 2488242314Smavg_raid_shutdown_post_sync(void *arg, int howto) 2489219974Smav{ 2490219974Smav struct g_class *mp; 2491219974Smav struct g_geom *gp, *gp2; 2492219974Smav struct g_raid_softc *sc; 2493242314Smav struct g_raid_volume *vol; 2494219974Smav 2495219974Smav mp = arg; 2496219974Smav DROP_GIANT(); 2497219974Smav g_topology_lock(); 2498242314Smav g_raid_shutdown = 1; 2499219974Smav LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 2500219974Smav if ((sc = gp->softc) == NULL) 2501219974Smav continue; 2502219974Smav g_topology_unlock(); 2503219974Smav sx_xlock(&sc->sc_lock); 2504242314Smav TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) 2505242314Smav g_raid_clean(vol, -1); 2506219974Smav g_cancel_event(sc); 2507253706Smav g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); 2508219974Smav g_topology_lock(); 2509219974Smav } 2510219974Smav g_topology_unlock(); 2511219974Smav PICKUP_GIANT(); 2512219974Smav} 2513219974Smav 2514219974Smavstatic void 2515219974Smavg_raid_init(struct g_class *mp) 2516219974Smav{ 2517219974Smav 2518242314Smav g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 2519242314Smav g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 2520242314Smav if (g_raid_post_sync == NULL) 2521219974Smav G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); 2522219974Smav g_raid_started = 1; 2523219974Smav} 2524219974Smav 2525219974Smavstatic void 2526219974Smavg_raid_fini(struct g_class *mp) 2527219974Smav{ 2528219974Smav 2529242314Smav if (g_raid_post_sync != NULL) 2530242314Smav EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync); 2531219974Smav g_raid_started = 0; 2532219974Smav} 2533219974Smav 2534219974Smavint 2535219974Smavg_raid_md_modevent(module_t mod, int type, void *arg) 2536219974Smav{ 2537219974Smav struct g_raid_md_class *class, *c, *nc; 2538219974Smav int error; 2539219974Smav 2540219974Smav error = 0; 2541219974Smav class = arg; 2542219974Smav switch (type) { 2543219974Smav case MOD_LOAD: 2544219974Smav c = LIST_FIRST(&g_raid_md_classes); 2545219974Smav if (c == NULL || c->mdc_priority > class->mdc_priority) 2546219974Smav LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); 2547219974Smav else { 2548219974Smav while ((nc = LIST_NEXT(c, mdc_list)) != NULL && 2549219974Smav nc->mdc_priority < class->mdc_priority) 2550219974Smav c = nc; 2551219974Smav LIST_INSERT_AFTER(c, class, mdc_list); 2552219974Smav } 2553219974Smav if (g_raid_started) 2554219974Smav g_retaste(&g_raid_class); 2555219974Smav break; 2556219974Smav case MOD_UNLOAD: 2557219974Smav LIST_REMOVE(class, mdc_list); 2558219974Smav break; 2559219974Smav default: 2560219974Smav error = EOPNOTSUPP; 2561219974Smav break; 2562219974Smav } 2563219974Smav 2564219974Smav return (error); 2565219974Smav} 2566219974Smav 2567219974Smavint 2568219974Smavg_raid_tr_modevent(module_t mod, int type, void *arg) 2569219974Smav{ 2570219974Smav struct g_raid_tr_class *class, *c, *nc; 2571219974Smav int error; 2572219974Smav 2573219974Smav error = 0; 2574219974Smav class = arg; 2575219974Smav switch (type) { 2576219974Smav case MOD_LOAD: 2577219974Smav c = LIST_FIRST(&g_raid_tr_classes); 2578219974Smav if (c == NULL || c->trc_priority > class->trc_priority) 2579219974Smav LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); 2580219974Smav else { 2581219974Smav while ((nc = LIST_NEXT(c, trc_list)) != NULL && 2582219974Smav nc->trc_priority < class->trc_priority) 2583219974Smav c = nc; 2584219974Smav LIST_INSERT_AFTER(c, class, trc_list); 2585219974Smav } 2586219974Smav break; 2587219974Smav case MOD_UNLOAD: 2588219974Smav LIST_REMOVE(class, trc_list); 2589219974Smav break; 2590219974Smav default: 2591219974Smav error = EOPNOTSUPP; 2592219974Smav break; 2593219974Smav } 2594219974Smav 2595219974Smav return (error); 2596219974Smav} 2597219974Smav 2598219974Smav/* 2599219974Smav * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) 2600219974Smav * to reduce module priority, allowing submodules to register them first. 2601219974Smav */ 2602219974Smavstatic moduledata_t g_raid_mod = { 2603219974Smav "g_raid", 2604219974Smav g_modevent, 2605219974Smav &g_raid_class 2606219974Smav}; 2607219974SmavDECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD); 2608219974SmavMODULE_VERSION(geom_raid, 0); 2609