189b17223SAlexander Motin /*- 23728855aSPedro F. Giffuni * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 33728855aSPedro F. Giffuni * 489b17223SAlexander Motin * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org> 589b17223SAlexander Motin * All rights reserved. 689b17223SAlexander Motin * 789b17223SAlexander Motin * Redistribution and use in source and binary forms, with or without 889b17223SAlexander Motin * modification, are permitted provided that the following conditions 989b17223SAlexander Motin * are met: 1089b17223SAlexander Motin * 1. Redistributions of source code must retain the above copyright 1189b17223SAlexander Motin * notice, this list of conditions and the following disclaimer. 1289b17223SAlexander Motin * 2. Redistributions in binary form must reproduce the above copyright 1389b17223SAlexander Motin * notice, this list of conditions and the following disclaimer in the 1489b17223SAlexander Motin * documentation and/or other materials provided with the distribution. 1589b17223SAlexander Motin * 1689b17223SAlexander Motin * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 1789b17223SAlexander Motin * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1889b17223SAlexander Motin * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1989b17223SAlexander Motin * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 2089b17223SAlexander Motin * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2189b17223SAlexander Motin * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2289b17223SAlexander Motin * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2389b17223SAlexander Motin * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2489b17223SAlexander Motin * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2589b17223SAlexander Motin * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2689b17223SAlexander Motin * SUCH DAMAGE. 2789b17223SAlexander Motin */ 2889b17223SAlexander Motin 2989b17223SAlexander Motin #include <sys/cdefs.h> 3089b17223SAlexander Motin __FBSDID("$FreeBSD$"); 3189b17223SAlexander Motin 3289b17223SAlexander Motin #include <sys/param.h> 3389b17223SAlexander Motin #include <sys/systm.h> 3489b17223SAlexander Motin #include <sys/kernel.h> 3589b17223SAlexander Motin #include <sys/module.h> 3689b17223SAlexander Motin #include <sys/limits.h> 3789b17223SAlexander Motin #include <sys/lock.h> 3889b17223SAlexander Motin #include <sys/mutex.h> 3989b17223SAlexander Motin #include <sys/bio.h> 405d807a0eSAndrey V. Elsukov #include <sys/sbuf.h> 4189b17223SAlexander Motin #include <sys/sysctl.h> 4289b17223SAlexander Motin #include <sys/malloc.h> 4389b17223SAlexander Motin #include <sys/eventhandler.h> 4489b17223SAlexander Motin #include <vm/uma.h> 4589b17223SAlexander Motin #include <geom/geom.h> 46ac03832eSConrad Meyer #include <geom/geom_dbg.h> 4789b17223SAlexander Motin #include <sys/proc.h> 4889b17223SAlexander Motin #include <sys/kthread.h> 4989b17223SAlexander Motin #include <sys/sched.h> 5089b17223SAlexander Motin #include <geom/raid/g_raid.h> 5189b17223SAlexander Motin #include "g_raid_md_if.h" 5289b17223SAlexander Motin #include "g_raid_tr_if.h" 5389b17223SAlexander Motin 5489b17223SAlexander Motin static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data"); 5589b17223SAlexander Motin 5689b17223SAlexander Motin SYSCTL_DECL(_kern_geom); 5789b17223SAlexander Motin SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff"); 58c89d2fbeSAlexander Motin int g_raid_enable = 1; 59af3b2549SHans Petter Selasky SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RWTUN, 60c89d2fbeSAlexander Motin &g_raid_enable, 0, "Enable on-disk metadata taste"); 6189b17223SAlexander Motin u_int g_raid_aggressive_spare = 0; 62af3b2549SHans Petter Selasky SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RWTUN, 6389b17223SAlexander Motin &g_raid_aggressive_spare, 0, "Use disks without metadata as spare"); 64fe51d6c1SAlexander Motin u_int g_raid_debug = 0; 65af3b2549SHans Petter Selasky SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid_debug, 0, 6689b17223SAlexander Motin "Debug level"); 6789b17223SAlexander Motin int g_raid_read_err_thresh = 10; 68af3b2549SHans Petter Selasky SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RWTUN, 6989b17223SAlexander Motin &g_raid_read_err_thresh, 0, 7089b17223SAlexander Motin "Number of read errors equated to disk failure"); 7189b17223SAlexander Motin u_int g_raid_start_timeout = 30; 72af3b2549SHans Petter Selasky SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RWTUN, 7389b17223SAlexander Motin &g_raid_start_timeout, 0, 7489b17223SAlexander Motin "Time to wait for all array components"); 7589b17223SAlexander Motin static u_int g_raid_clean_time = 5; 76af3b2549SHans Petter Selasky SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RWTUN, 7789b17223SAlexander Motin &g_raid_clean_time, 0, "Mark volume as clean when idling"); 7889b17223SAlexander Motin static u_int g_raid_disconnect_on_failure = 1; 79af3b2549SHans Petter Selasky SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN, 8089b17223SAlexander Motin &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 8189b17223SAlexander Motin static u_int g_raid_name_format = 0; 82af3b2549SHans Petter Selasky SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RWTUN, 8389b17223SAlexander Motin &g_raid_name_format, 0, "Providers name format."); 8489b17223SAlexander Motin static u_int g_raid_idle_threshold = 1000000; 85af3b2549SHans Petter Selasky SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RWTUN, 8689b17223SAlexander Motin &g_raid_idle_threshold, 1000000, 8789b17223SAlexander Motin "Time in microseconds to consider a volume idle."); 8889b17223SAlexander Motin 8989b17223SAlexander Motin #define MSLEEP(rv, ident, mtx, priority, wmesg, timeout) do { \ 9089b17223SAlexander Motin G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 9189b17223SAlexander Motin rv = msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 9289b17223SAlexander Motin G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 9389b17223SAlexander Motin } while (0) 9489b17223SAlexander Motin 9589b17223SAlexander Motin LIST_HEAD(, g_raid_md_class) g_raid_md_classes = 9689b17223SAlexander Motin LIST_HEAD_INITIALIZER(g_raid_md_classes); 9789b17223SAlexander Motin 9889b17223SAlexander Motin LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes = 9989b17223SAlexander Motin LIST_HEAD_INITIALIZER(g_raid_tr_classes); 10089b17223SAlexander Motin 10189b17223SAlexander Motin LIST_HEAD(, g_raid_volume) g_raid_volumes = 10289b17223SAlexander Motin LIST_HEAD_INITIALIZER(g_raid_volumes); 10389b17223SAlexander Motin 104a479c51bSAlexander Motin static eventhandler_tag g_raid_post_sync = NULL; 10589b17223SAlexander Motin static int g_raid_started = 0; 106a479c51bSAlexander Motin static int g_raid_shutdown = 0; 10789b17223SAlexander Motin 10889b17223SAlexander Motin static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp, 10989b17223SAlexander Motin struct g_geom *gp); 11089b17223SAlexander Motin static g_taste_t g_raid_taste; 11189b17223SAlexander Motin static void g_raid_init(struct g_class *mp); 11289b17223SAlexander Motin static void g_raid_fini(struct g_class *mp); 11389b17223SAlexander Motin 11489b17223SAlexander Motin struct g_class g_raid_class = { 11589b17223SAlexander Motin .name = G_RAID_CLASS_NAME, 11689b17223SAlexander Motin .version = G_VERSION, 11789b17223SAlexander Motin .ctlreq = g_raid_ctl, 11889b17223SAlexander Motin .taste = g_raid_taste, 11989b17223SAlexander Motin .destroy_geom = g_raid_destroy_geom, 12089b17223SAlexander Motin .init = g_raid_init, 12189b17223SAlexander Motin .fini = g_raid_fini 12289b17223SAlexander Motin }; 12389b17223SAlexander Motin 12489b17223SAlexander Motin static void g_raid_destroy_provider(struct g_raid_volume *vol); 12589b17223SAlexander Motin static int g_raid_update_disk(struct g_raid_disk *disk, u_int event); 12689b17223SAlexander Motin static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event); 12789b17223SAlexander Motin static int g_raid_update_volume(struct g_raid_volume *vol, u_int event); 12889b17223SAlexander Motin static int g_raid_update_node(struct g_raid_softc *sc, u_int event); 12989b17223SAlexander Motin static void g_raid_dumpconf(struct sbuf *sb, const char *indent, 13089b17223SAlexander Motin struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 13189b17223SAlexander Motin static void g_raid_start(struct bio *bp); 13289b17223SAlexander Motin static void g_raid_start_request(struct bio *bp); 13389b17223SAlexander Motin static void g_raid_disk_done(struct bio *bp); 13489b17223SAlexander Motin static void g_raid_poll(struct g_raid_softc *sc); 13589b17223SAlexander Motin 13689b17223SAlexander Motin static const char * 13789b17223SAlexander Motin g_raid_node_event2str(int event) 13889b17223SAlexander Motin { 13989b17223SAlexander Motin 14089b17223SAlexander Motin switch (event) { 14189b17223SAlexander Motin case G_RAID_NODE_E_WAKE: 14289b17223SAlexander Motin return ("WAKE"); 14389b17223SAlexander Motin case G_RAID_NODE_E_START: 14489b17223SAlexander Motin return ("START"); 14589b17223SAlexander Motin default: 14689b17223SAlexander Motin return ("INVALID"); 14789b17223SAlexander Motin } 14889b17223SAlexander Motin } 14989b17223SAlexander Motin 15089b17223SAlexander Motin const char * 15189b17223SAlexander Motin g_raid_disk_state2str(int state) 15289b17223SAlexander Motin { 15389b17223SAlexander Motin 15489b17223SAlexander Motin switch (state) { 15589b17223SAlexander Motin case G_RAID_DISK_S_NONE: 15689b17223SAlexander Motin return ("NONE"); 15789b17223SAlexander Motin case G_RAID_DISK_S_OFFLINE: 15889b17223SAlexander Motin return ("OFFLINE"); 15926c538bcSAlexander Motin case G_RAID_DISK_S_DISABLED: 16026c538bcSAlexander Motin return ("DISABLED"); 16189b17223SAlexander Motin case G_RAID_DISK_S_FAILED: 16289b17223SAlexander Motin return ("FAILED"); 16389b17223SAlexander Motin case G_RAID_DISK_S_STALE_FAILED: 16489b17223SAlexander Motin return ("STALE_FAILED"); 16589b17223SAlexander Motin case G_RAID_DISK_S_SPARE: 16689b17223SAlexander Motin return ("SPARE"); 16789b17223SAlexander Motin case G_RAID_DISK_S_STALE: 16889b17223SAlexander Motin return ("STALE"); 16989b17223SAlexander Motin case G_RAID_DISK_S_ACTIVE: 17089b17223SAlexander Motin return ("ACTIVE"); 17189b17223SAlexander Motin default: 17289b17223SAlexander Motin return ("INVALID"); 17389b17223SAlexander Motin } 17489b17223SAlexander Motin } 17589b17223SAlexander Motin 17689b17223SAlexander Motin static const char * 17789b17223SAlexander Motin g_raid_disk_event2str(int event) 17889b17223SAlexander Motin { 17989b17223SAlexander Motin 18089b17223SAlexander Motin switch (event) { 18189b17223SAlexander Motin case G_RAID_DISK_E_DISCONNECTED: 18289b17223SAlexander Motin return ("DISCONNECTED"); 18389b17223SAlexander Motin default: 18489b17223SAlexander Motin return ("INVALID"); 18589b17223SAlexander Motin } 18689b17223SAlexander Motin } 18789b17223SAlexander Motin 18889b17223SAlexander Motin const char * 18989b17223SAlexander Motin g_raid_subdisk_state2str(int state) 19089b17223SAlexander Motin { 19189b17223SAlexander Motin 19289b17223SAlexander Motin switch (state) { 19389b17223SAlexander Motin case G_RAID_SUBDISK_S_NONE: 19489b17223SAlexander Motin return ("NONE"); 19589b17223SAlexander Motin case G_RAID_SUBDISK_S_FAILED: 19689b17223SAlexander Motin return ("FAILED"); 19789b17223SAlexander Motin case G_RAID_SUBDISK_S_NEW: 19889b17223SAlexander Motin return ("NEW"); 19989b17223SAlexander Motin case G_RAID_SUBDISK_S_REBUILD: 20089b17223SAlexander Motin return ("REBUILD"); 20189b17223SAlexander Motin case G_RAID_SUBDISK_S_UNINITIALIZED: 20289b17223SAlexander Motin return ("UNINITIALIZED"); 20389b17223SAlexander Motin case G_RAID_SUBDISK_S_STALE: 20489b17223SAlexander Motin return ("STALE"); 20589b17223SAlexander Motin case G_RAID_SUBDISK_S_RESYNC: 20689b17223SAlexander Motin return ("RESYNC"); 20789b17223SAlexander Motin case G_RAID_SUBDISK_S_ACTIVE: 20889b17223SAlexander Motin return ("ACTIVE"); 20989b17223SAlexander Motin default: 21089b17223SAlexander Motin return ("INVALID"); 21189b17223SAlexander Motin } 21289b17223SAlexander Motin } 21389b17223SAlexander Motin 21489b17223SAlexander Motin static const char * 21589b17223SAlexander Motin g_raid_subdisk_event2str(int event) 21689b17223SAlexander Motin { 21789b17223SAlexander Motin 21889b17223SAlexander Motin switch (event) { 21989b17223SAlexander Motin case G_RAID_SUBDISK_E_NEW: 22089b17223SAlexander Motin return ("NEW"); 221d9d68496SAlexander Motin case G_RAID_SUBDISK_E_FAILED: 222d9d68496SAlexander Motin return ("FAILED"); 22389b17223SAlexander Motin case G_RAID_SUBDISK_E_DISCONNECTED: 22489b17223SAlexander Motin return ("DISCONNECTED"); 22589b17223SAlexander Motin default: 22689b17223SAlexander Motin return ("INVALID"); 22789b17223SAlexander Motin } 22889b17223SAlexander Motin } 22989b17223SAlexander Motin 23089b17223SAlexander Motin const char * 23189b17223SAlexander Motin g_raid_volume_state2str(int state) 23289b17223SAlexander Motin { 23389b17223SAlexander Motin 23489b17223SAlexander Motin switch (state) { 23589b17223SAlexander Motin case G_RAID_VOLUME_S_STARTING: 23689b17223SAlexander Motin return ("STARTING"); 23789b17223SAlexander Motin case G_RAID_VOLUME_S_BROKEN: 23889b17223SAlexander Motin return ("BROKEN"); 23989b17223SAlexander Motin case G_RAID_VOLUME_S_DEGRADED: 24089b17223SAlexander Motin return ("DEGRADED"); 24189b17223SAlexander Motin case G_RAID_VOLUME_S_SUBOPTIMAL: 24289b17223SAlexander Motin return ("SUBOPTIMAL"); 24389b17223SAlexander Motin case G_RAID_VOLUME_S_OPTIMAL: 24489b17223SAlexander Motin return ("OPTIMAL"); 24589b17223SAlexander Motin case G_RAID_VOLUME_S_UNSUPPORTED: 24689b17223SAlexander Motin return ("UNSUPPORTED"); 24789b17223SAlexander Motin case G_RAID_VOLUME_S_STOPPED: 24889b17223SAlexander Motin return ("STOPPED"); 24989b17223SAlexander Motin default: 25089b17223SAlexander Motin return ("INVALID"); 25189b17223SAlexander Motin } 25289b17223SAlexander Motin } 25389b17223SAlexander Motin 25489b17223SAlexander Motin static const char * 25589b17223SAlexander Motin g_raid_volume_event2str(int event) 25689b17223SAlexander Motin { 25789b17223SAlexander Motin 25889b17223SAlexander Motin switch (event) { 25989b17223SAlexander Motin case G_RAID_VOLUME_E_UP: 26089b17223SAlexander Motin return ("UP"); 26189b17223SAlexander Motin case G_RAID_VOLUME_E_DOWN: 26289b17223SAlexander Motin return ("DOWN"); 26389b17223SAlexander Motin case G_RAID_VOLUME_E_START: 26489b17223SAlexander Motin return ("START"); 26589b17223SAlexander Motin case G_RAID_VOLUME_E_STARTMD: 26689b17223SAlexander Motin return ("STARTMD"); 26789b17223SAlexander Motin default: 26889b17223SAlexander Motin return ("INVALID"); 26989b17223SAlexander Motin } 27089b17223SAlexander Motin } 27189b17223SAlexander Motin 27289b17223SAlexander Motin const char * 27389b17223SAlexander Motin g_raid_volume_level2str(int level, int qual) 27489b17223SAlexander Motin { 27589b17223SAlexander Motin 27689b17223SAlexander Motin switch (level) { 27789b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID0: 27889b17223SAlexander Motin return ("RAID0"); 27989b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID1: 28089b17223SAlexander Motin return ("RAID1"); 28189b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID3: 282dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R3P0) 283dbb2e755SAlexander Motin return ("RAID3-P0"); 284dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R3PN) 285dbb2e755SAlexander Motin return ("RAID3-PN"); 28689b17223SAlexander Motin return ("RAID3"); 28789b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID4: 288dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R4P0) 289bafd0b5bSAlexander Motin return ("RAID4-P0"); 290dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R4PN) 291bafd0b5bSAlexander Motin return ("RAID4-PN"); 29289b17223SAlexander Motin return ("RAID4"); 29389b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID5: 294fc1de960SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5RA) 295dbb2e755SAlexander Motin return ("RAID5-RA"); 296fc1de960SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5RS) 297dbb2e755SAlexander Motin return ("RAID5-RS"); 298fc1de960SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5LA) 299dbb2e755SAlexander Motin return ("RAID5-LA"); 300fc1de960SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5LS) 301dbb2e755SAlexander Motin return ("RAID5-LS"); 30289b17223SAlexander Motin return ("RAID5"); 30389b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID6: 304dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R6RA) 305dbb2e755SAlexander Motin return ("RAID6-RA"); 306dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R6RS) 307dbb2e755SAlexander Motin return ("RAID6-RS"); 308dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R6LA) 309dbb2e755SAlexander Motin return ("RAID6-LA"); 310dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R6LS) 311dbb2e755SAlexander Motin return ("RAID6-LS"); 31289b17223SAlexander Motin return ("RAID6"); 313dbb2e755SAlexander Motin case G_RAID_VOLUME_RL_RAIDMDF: 314dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_RMDFRA) 315dbb2e755SAlexander Motin return ("RAIDMDF-RA"); 316dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_RMDFRS) 317dbb2e755SAlexander Motin return ("RAIDMDF-RS"); 318dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_RMDFLA) 319dbb2e755SAlexander Motin return ("RAIDMDF-LA"); 320dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_RMDFLS) 321dbb2e755SAlexander Motin return ("RAIDMDF-LS"); 322dbb2e755SAlexander Motin return ("RAIDMDF"); 32389b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID1E: 324dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R1EA) 325dbb2e755SAlexander Motin return ("RAID1E-A"); 326dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R1EO) 327dbb2e755SAlexander Motin return ("RAID1E-O"); 32889b17223SAlexander Motin return ("RAID1E"); 32989b17223SAlexander Motin case G_RAID_VOLUME_RL_SINGLE: 33089b17223SAlexander Motin return ("SINGLE"); 33189b17223SAlexander Motin case G_RAID_VOLUME_RL_CONCAT: 33289b17223SAlexander Motin return ("CONCAT"); 33389b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID5E: 334dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5ERA) 335dbb2e755SAlexander Motin return ("RAID5E-RA"); 336dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5ERS) 337dbb2e755SAlexander Motin return ("RAID5E-RS"); 338dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5ELA) 339dbb2e755SAlexander Motin return ("RAID5E-LA"); 340dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5ELS) 341dbb2e755SAlexander Motin return ("RAID5E-LS"); 34289b17223SAlexander Motin return ("RAID5E"); 34389b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID5EE: 344dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5EERA) 345dbb2e755SAlexander Motin return ("RAID5EE-RA"); 346dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5EERS) 347dbb2e755SAlexander Motin return ("RAID5EE-RS"); 348dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5EELA) 349dbb2e755SAlexander Motin return ("RAID5EE-LA"); 350dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5EELS) 351dbb2e755SAlexander Motin return ("RAID5EE-LS"); 35289b17223SAlexander Motin return ("RAID5EE"); 353dbb2e755SAlexander Motin case G_RAID_VOLUME_RL_RAID5R: 354dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5RRA) 355dbb2e755SAlexander Motin return ("RAID5R-RA"); 356dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5RRS) 357dbb2e755SAlexander Motin return ("RAID5R-RS"); 358dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5RLA) 359dbb2e755SAlexander Motin return ("RAID5R-LA"); 360dbb2e755SAlexander Motin if (qual == G_RAID_VOLUME_RLQ_R5RLS) 361dbb2e755SAlexander Motin return ("RAID5R-LS"); 362dbb2e755SAlexander Motin return ("RAID5E"); 36389b17223SAlexander Motin default: 36489b17223SAlexander Motin return ("UNKNOWN"); 36589b17223SAlexander Motin } 36689b17223SAlexander Motin } 36789b17223SAlexander Motin 36889b17223SAlexander Motin int 36989b17223SAlexander Motin g_raid_volume_str2level(const char *str, int *level, int *qual) 37089b17223SAlexander Motin { 37189b17223SAlexander Motin 37289b17223SAlexander Motin *level = G_RAID_VOLUME_RL_UNKNOWN; 37389b17223SAlexander Motin *qual = G_RAID_VOLUME_RLQ_NONE; 37489b17223SAlexander Motin if (strcasecmp(str, "RAID0") == 0) 37589b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID0; 37689b17223SAlexander Motin else if (strcasecmp(str, "RAID1") == 0) 37789b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID1; 378dbb2e755SAlexander Motin else if (strcasecmp(str, "RAID3-P0") == 0) { 37989b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID3; 380dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R3P0; 3814b97ff61SAlexander Motin } else if (strcasecmp(str, "RAID3-PN") == 0 || 382dbb2e755SAlexander Motin strcasecmp(str, "RAID3") == 0) { 383dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID3; 3844b97ff61SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R3PN; 385dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID4-P0") == 0) { 38689b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID4; 387dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R4P0; 3884b97ff61SAlexander Motin } else if (strcasecmp(str, "RAID4-PN") == 0 || 389dbb2e755SAlexander Motin strcasecmp(str, "RAID4") == 0) { 390dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID4; 3914b97ff61SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R4PN; 392dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5-RA") == 0) { 39389b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5; 394fc1de960SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5RA; 395dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5-RS") == 0) { 396fc1de960SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5; 397fc1de960SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5RS; 398fc1de960SAlexander Motin } else if (strcasecmp(str, "RAID5") == 0 || 399dbb2e755SAlexander Motin strcasecmp(str, "RAID5-LA") == 0) { 400fc1de960SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5; 401fc1de960SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5LA; 402dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5-LS") == 0) { 403fc1de960SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5; 404fc1de960SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5LS; 405dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID6-RA") == 0) { 40689b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID6; 407dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R6RA; 408dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID6-RS") == 0) { 409dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID6; 410dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R6RS; 411dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID6") == 0 || 412dbb2e755SAlexander Motin strcasecmp(str, "RAID6-LA") == 0) { 413dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID6; 414dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R6LA; 415dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID6-LS") == 0) { 416dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID6; 417dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R6LS; 418dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAIDMDF-RA") == 0) { 419dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAIDMDF; 420dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_RMDFRA; 421dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAIDMDF-RS") == 0) { 422dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAIDMDF; 423dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_RMDFRS; 424dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAIDMDF") == 0 || 425dbb2e755SAlexander Motin strcasecmp(str, "RAIDMDF-LA") == 0) { 426dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAIDMDF; 427dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_RMDFLA; 428dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAIDMDF-LS") == 0) { 429dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAIDMDF; 430dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_RMDFLS; 431dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID10") == 0 || 432dbb2e755SAlexander Motin strcasecmp(str, "RAID1E") == 0 || 433dbb2e755SAlexander Motin strcasecmp(str, "RAID1E-A") == 0) { 43489b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID1E; 435dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R1EA; 436dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID1E-O") == 0) { 437dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID1E; 438dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R1EO; 439dbb2e755SAlexander Motin } else if (strcasecmp(str, "SINGLE") == 0) 44089b17223SAlexander Motin *level = G_RAID_VOLUME_RL_SINGLE; 44189b17223SAlexander Motin else if (strcasecmp(str, "CONCAT") == 0) 44289b17223SAlexander Motin *level = G_RAID_VOLUME_RL_CONCAT; 443dbb2e755SAlexander Motin else if (strcasecmp(str, "RAID5E-RA") == 0) { 44489b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5E; 445dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5ERA; 446dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5E-RS") == 0) { 447dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5E; 448dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5ERS; 449dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5E") == 0 || 450dbb2e755SAlexander Motin strcasecmp(str, "RAID5E-LA") == 0) { 451dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5E; 452dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5ELA; 453dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5E-LS") == 0) { 454dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5E; 455dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5ELS; 456dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5EE-RA") == 0) { 45789b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5EE; 458dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5EERA; 459dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5EE-RS") == 0) { 460dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5EE; 461dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5EERS; 462dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5EE") == 0 || 463dbb2e755SAlexander Motin strcasecmp(str, "RAID5EE-LA") == 0) { 464dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5EE; 465dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5EELA; 466dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5EE-LS") == 0) { 467dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5EE; 468dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5EELS; 469dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5R-RA") == 0) { 470dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5R; 471dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5RRA; 472dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5R-RS") == 0) { 473dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5R; 474dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5RRS; 475dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5R") == 0 || 476dbb2e755SAlexander Motin strcasecmp(str, "RAID5R-LA") == 0) { 477dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5R; 478dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5RLA; 479dbb2e755SAlexander Motin } else if (strcasecmp(str, "RAID5R-LS") == 0) { 480dbb2e755SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5R; 481dbb2e755SAlexander Motin *qual = G_RAID_VOLUME_RLQ_R5RLS; 482dbb2e755SAlexander Motin } else 48389b17223SAlexander Motin return (-1); 48489b17223SAlexander Motin return (0); 48589b17223SAlexander Motin } 48689b17223SAlexander Motin 48789b17223SAlexander Motin const char * 48889b17223SAlexander Motin g_raid_get_diskname(struct g_raid_disk *disk) 48989b17223SAlexander Motin { 49089b17223SAlexander Motin 49189b17223SAlexander Motin if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 49289b17223SAlexander Motin return ("[unknown]"); 49389b17223SAlexander Motin return (disk->d_consumer->provider->name); 49489b17223SAlexander Motin } 49589b17223SAlexander Motin 49689b17223SAlexander Motin void 497609a7474SAlexander Motin g_raid_get_disk_info(struct g_raid_disk *disk) 498609a7474SAlexander Motin { 499609a7474SAlexander Motin struct g_consumer *cp = disk->d_consumer; 500609a7474SAlexander Motin int error, len; 501609a7474SAlexander Motin 502609a7474SAlexander Motin /* Read kernel dumping information. */ 503609a7474SAlexander Motin disk->d_kd.offset = 0; 504609a7474SAlexander Motin disk->d_kd.length = OFF_MAX; 505609a7474SAlexander Motin len = sizeof(disk->d_kd); 506609a7474SAlexander Motin error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); 507609a7474SAlexander Motin if (error) 508609a7474SAlexander Motin disk->d_kd.di.dumper = NULL; 509609a7474SAlexander Motin if (disk->d_kd.di.dumper == NULL) 510609a7474SAlexander Motin G_RAID_DEBUG1(2, disk->d_softc, 511609a7474SAlexander Motin "Dumping not supported by %s: %d.", 512609a7474SAlexander Motin cp->provider->name, error); 513609a7474SAlexander Motin 514609a7474SAlexander Motin /* Read BIO_DELETE support. */ 515609a7474SAlexander Motin error = g_getattr("GEOM::candelete", cp, &disk->d_candelete); 516609a7474SAlexander Motin if (error) 517609a7474SAlexander Motin disk->d_candelete = 0; 518609a7474SAlexander Motin if (!disk->d_candelete) 519609a7474SAlexander Motin G_RAID_DEBUG1(2, disk->d_softc, 520609a7474SAlexander Motin "BIO_DELETE not supported by %s: %d.", 521609a7474SAlexander Motin cp->provider->name, error); 522609a7474SAlexander Motin } 523609a7474SAlexander Motin 524609a7474SAlexander Motin void 52589b17223SAlexander Motin g_raid_report_disk_state(struct g_raid_disk *disk) 52689b17223SAlexander Motin { 52789b17223SAlexander Motin struct g_raid_subdisk *sd; 52889b17223SAlexander Motin int len, state; 52989b17223SAlexander Motin uint32_t s; 53089b17223SAlexander Motin 53189b17223SAlexander Motin if (disk->d_consumer == NULL) 53289b17223SAlexander Motin return; 53326c538bcSAlexander Motin if (disk->d_state == G_RAID_DISK_S_DISABLED) { 534b99586c2SAlexander Motin s = G_STATE_ACTIVE; /* XXX */ 53526c538bcSAlexander Motin } else if (disk->d_state == G_RAID_DISK_S_FAILED || 53689b17223SAlexander Motin disk->d_state == G_RAID_DISK_S_STALE_FAILED) { 53789b17223SAlexander Motin s = G_STATE_FAILED; 53889b17223SAlexander Motin } else { 53989b17223SAlexander Motin state = G_RAID_SUBDISK_S_ACTIVE; 54089b17223SAlexander Motin TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 54189b17223SAlexander Motin if (sd->sd_state < state) 54289b17223SAlexander Motin state = sd->sd_state; 54389b17223SAlexander Motin } 54489b17223SAlexander Motin if (state == G_RAID_SUBDISK_S_FAILED) 54589b17223SAlexander Motin s = G_STATE_FAILED; 54689b17223SAlexander Motin else if (state == G_RAID_SUBDISK_S_NEW || 54789b17223SAlexander Motin state == G_RAID_SUBDISK_S_REBUILD) 54889b17223SAlexander Motin s = G_STATE_REBUILD; 54989b17223SAlexander Motin else if (state == G_RAID_SUBDISK_S_STALE || 55089b17223SAlexander Motin state == G_RAID_SUBDISK_S_RESYNC) 55189b17223SAlexander Motin s = G_STATE_RESYNC; 55289b17223SAlexander Motin else 55389b17223SAlexander Motin s = G_STATE_ACTIVE; 55489b17223SAlexander Motin } 55589b17223SAlexander Motin len = sizeof(s); 55689b17223SAlexander Motin g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s); 55789b17223SAlexander Motin G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.", 55889b17223SAlexander Motin g_raid_get_diskname(disk), s); 55989b17223SAlexander Motin } 56089b17223SAlexander Motin 56189b17223SAlexander Motin void 56289b17223SAlexander Motin g_raid_change_disk_state(struct g_raid_disk *disk, int state) 56389b17223SAlexander Motin { 56489b17223SAlexander Motin 56589b17223SAlexander Motin G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.", 56689b17223SAlexander Motin g_raid_get_diskname(disk), 56789b17223SAlexander Motin g_raid_disk_state2str(disk->d_state), 56889b17223SAlexander Motin g_raid_disk_state2str(state)); 56989b17223SAlexander Motin disk->d_state = state; 57089b17223SAlexander Motin g_raid_report_disk_state(disk); 57189b17223SAlexander Motin } 57289b17223SAlexander Motin 57389b17223SAlexander Motin void 57489b17223SAlexander Motin g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state) 57589b17223SAlexander Motin { 57689b17223SAlexander Motin 57789b17223SAlexander Motin G_RAID_DEBUG1(0, sd->sd_softc, 57889b17223SAlexander Motin "Subdisk %s:%d-%s state changed from %s to %s.", 57989b17223SAlexander Motin sd->sd_volume->v_name, sd->sd_pos, 58089b17223SAlexander Motin sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", 58189b17223SAlexander Motin g_raid_subdisk_state2str(sd->sd_state), 58289b17223SAlexander Motin g_raid_subdisk_state2str(state)); 58389b17223SAlexander Motin sd->sd_state = state; 58489b17223SAlexander Motin if (sd->sd_disk) 58589b17223SAlexander Motin g_raid_report_disk_state(sd->sd_disk); 58689b17223SAlexander Motin } 58789b17223SAlexander Motin 58889b17223SAlexander Motin void 58989b17223SAlexander Motin g_raid_change_volume_state(struct g_raid_volume *vol, int state) 59089b17223SAlexander Motin { 59189b17223SAlexander Motin 59289b17223SAlexander Motin G_RAID_DEBUG1(0, vol->v_softc, 59389b17223SAlexander Motin "Volume %s state changed from %s to %s.", 59489b17223SAlexander Motin vol->v_name, 59589b17223SAlexander Motin g_raid_volume_state2str(vol->v_state), 59689b17223SAlexander Motin g_raid_volume_state2str(state)); 59789b17223SAlexander Motin vol->v_state = state; 59889b17223SAlexander Motin } 59989b17223SAlexander Motin 60089b17223SAlexander Motin /* 60189b17223SAlexander Motin * --- Events handling functions --- 60289b17223SAlexander Motin * Events in geom_raid are used to maintain subdisks and volumes status 60389b17223SAlexander Motin * from one thread to simplify locking. 60489b17223SAlexander Motin */ 60589b17223SAlexander Motin static void 60689b17223SAlexander Motin g_raid_event_free(struct g_raid_event *ep) 60789b17223SAlexander Motin { 60889b17223SAlexander Motin 60989b17223SAlexander Motin free(ep, M_RAID); 61089b17223SAlexander Motin } 61189b17223SAlexander Motin 61289b17223SAlexander Motin int 61389b17223SAlexander Motin g_raid_event_send(void *arg, int event, int flags) 61489b17223SAlexander Motin { 61589b17223SAlexander Motin struct g_raid_softc *sc; 61689b17223SAlexander Motin struct g_raid_event *ep; 61789b17223SAlexander Motin int error; 61889b17223SAlexander Motin 61989b17223SAlexander Motin if ((flags & G_RAID_EVENT_VOLUME) != 0) { 62089b17223SAlexander Motin sc = ((struct g_raid_volume *)arg)->v_softc; 62189b17223SAlexander Motin } else if ((flags & G_RAID_EVENT_DISK) != 0) { 62289b17223SAlexander Motin sc = ((struct g_raid_disk *)arg)->d_softc; 62389b17223SAlexander Motin } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) { 62489b17223SAlexander Motin sc = ((struct g_raid_subdisk *)arg)->sd_softc; 62589b17223SAlexander Motin } else { 62689b17223SAlexander Motin sc = arg; 62789b17223SAlexander Motin } 62889b17223SAlexander Motin ep = malloc(sizeof(*ep), M_RAID, 62989b17223SAlexander Motin sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT); 63089b17223SAlexander Motin if (ep == NULL) 63189b17223SAlexander Motin return (ENOMEM); 63289b17223SAlexander Motin ep->e_tgt = arg; 63389b17223SAlexander Motin ep->e_event = event; 63489b17223SAlexander Motin ep->e_flags = flags; 63589b17223SAlexander Motin ep->e_error = 0; 63689b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc); 63789b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 63889b17223SAlexander Motin TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 63989b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 64089b17223SAlexander Motin wakeup(sc); 64189b17223SAlexander Motin 64289b17223SAlexander Motin if ((flags & G_RAID_EVENT_WAIT) == 0) 64389b17223SAlexander Motin return (0); 64489b17223SAlexander Motin 64589b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 64689b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep); 64789b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 64889b17223SAlexander Motin while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) { 64989b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 65089b17223SAlexander Motin MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event", 65189b17223SAlexander Motin hz * 5); 65289b17223SAlexander Motin } 65389b17223SAlexander Motin error = ep->e_error; 65489b17223SAlexander Motin g_raid_event_free(ep); 65589b17223SAlexander Motin sx_xlock(&sc->sc_lock); 65689b17223SAlexander Motin return (error); 65789b17223SAlexander Motin } 65889b17223SAlexander Motin 65989b17223SAlexander Motin static void 66089b17223SAlexander Motin g_raid_event_cancel(struct g_raid_softc *sc, void *tgt) 66189b17223SAlexander Motin { 66289b17223SAlexander Motin struct g_raid_event *ep, *tmpep; 66389b17223SAlexander Motin 66489b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 66589b17223SAlexander Motin 66689b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 66789b17223SAlexander Motin TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 66889b17223SAlexander Motin if (ep->e_tgt != tgt) 66989b17223SAlexander Motin continue; 67089b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_events, ep, e_next); 67189b17223SAlexander Motin if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) 67289b17223SAlexander Motin g_raid_event_free(ep); 67389b17223SAlexander Motin else { 67489b17223SAlexander Motin ep->e_error = ECANCELED; 67589b17223SAlexander Motin wakeup(ep); 67689b17223SAlexander Motin } 67789b17223SAlexander Motin } 67889b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 67989b17223SAlexander Motin } 68089b17223SAlexander Motin 68189b17223SAlexander Motin static int 68289b17223SAlexander Motin g_raid_event_check(struct g_raid_softc *sc, void *tgt) 68389b17223SAlexander Motin { 68489b17223SAlexander Motin struct g_raid_event *ep; 68589b17223SAlexander Motin int res = 0; 68689b17223SAlexander Motin 68789b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 68889b17223SAlexander Motin 68989b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 69089b17223SAlexander Motin TAILQ_FOREACH(ep, &sc->sc_events, e_next) { 69189b17223SAlexander Motin if (ep->e_tgt != tgt) 69289b17223SAlexander Motin continue; 69389b17223SAlexander Motin res = 1; 69489b17223SAlexander Motin break; 69589b17223SAlexander Motin } 69689b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 69789b17223SAlexander Motin return (res); 69889b17223SAlexander Motin } 69989b17223SAlexander Motin 70089b17223SAlexander Motin /* 70189b17223SAlexander Motin * Return the number of disks in given state. 70289b17223SAlexander Motin * If state is equal to -1, count all connected disks. 70389b17223SAlexander Motin */ 70489b17223SAlexander Motin u_int 70589b17223SAlexander Motin g_raid_ndisks(struct g_raid_softc *sc, int state) 70689b17223SAlexander Motin { 70789b17223SAlexander Motin struct g_raid_disk *disk; 70889b17223SAlexander Motin u_int n; 70989b17223SAlexander Motin 71089b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 71189b17223SAlexander Motin 71289b17223SAlexander Motin n = 0; 71389b17223SAlexander Motin TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 71489b17223SAlexander Motin if (disk->d_state == state || state == -1) 71589b17223SAlexander Motin n++; 71689b17223SAlexander Motin } 71789b17223SAlexander Motin return (n); 71889b17223SAlexander Motin } 71989b17223SAlexander Motin 72089b17223SAlexander Motin /* 72189b17223SAlexander Motin * Return the number of subdisks in given state. 72289b17223SAlexander Motin * If state is equal to -1, count all connected disks. 72389b17223SAlexander Motin */ 72489b17223SAlexander Motin u_int 72589b17223SAlexander Motin g_raid_nsubdisks(struct g_raid_volume *vol, int state) 72689b17223SAlexander Motin { 72789b17223SAlexander Motin struct g_raid_subdisk *subdisk; 72889b17223SAlexander Motin struct g_raid_softc *sc; 72989b17223SAlexander Motin u_int i, n ; 73089b17223SAlexander Motin 73189b17223SAlexander Motin sc = vol->v_softc; 73289b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 73389b17223SAlexander Motin 73489b17223SAlexander Motin n = 0; 73589b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 73689b17223SAlexander Motin subdisk = &vol->v_subdisks[i]; 73789b17223SAlexander Motin if ((state == -1 && 73889b17223SAlexander Motin subdisk->sd_state != G_RAID_SUBDISK_S_NONE) || 73989b17223SAlexander Motin subdisk->sd_state == state) 74089b17223SAlexander Motin n++; 74189b17223SAlexander Motin } 74289b17223SAlexander Motin return (n); 74389b17223SAlexander Motin } 74489b17223SAlexander Motin 74589b17223SAlexander Motin /* 74689b17223SAlexander Motin * Return the first subdisk in given state. 74789b17223SAlexander Motin * If state is equal to -1, then the first connected disks. 74889b17223SAlexander Motin */ 74989b17223SAlexander Motin struct g_raid_subdisk * 75089b17223SAlexander Motin g_raid_get_subdisk(struct g_raid_volume *vol, int state) 75189b17223SAlexander Motin { 75289b17223SAlexander Motin struct g_raid_subdisk *sd; 75389b17223SAlexander Motin struct g_raid_softc *sc; 75489b17223SAlexander Motin u_int i; 75589b17223SAlexander Motin 75689b17223SAlexander Motin sc = vol->v_softc; 75789b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 75889b17223SAlexander Motin 75989b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 76089b17223SAlexander Motin sd = &vol->v_subdisks[i]; 76189b17223SAlexander Motin if ((state == -1 && 76289b17223SAlexander Motin sd->sd_state != G_RAID_SUBDISK_S_NONE) || 76389b17223SAlexander Motin sd->sd_state == state) 76489b17223SAlexander Motin return (sd); 76589b17223SAlexander Motin } 76689b17223SAlexander Motin return (NULL); 76789b17223SAlexander Motin } 76889b17223SAlexander Motin 76989b17223SAlexander Motin struct g_consumer * 77089b17223SAlexander Motin g_raid_open_consumer(struct g_raid_softc *sc, const char *name) 77189b17223SAlexander Motin { 77289b17223SAlexander Motin struct g_consumer *cp; 77389b17223SAlexander Motin struct g_provider *pp; 77489b17223SAlexander Motin 77589b17223SAlexander Motin g_topology_assert(); 77689b17223SAlexander Motin 77789b17223SAlexander Motin if (strncmp(name, "/dev/", 5) == 0) 77889b17223SAlexander Motin name += 5; 77989b17223SAlexander Motin pp = g_provider_by_name(name); 78089b17223SAlexander Motin if (pp == NULL) 78189b17223SAlexander Motin return (NULL); 78289b17223SAlexander Motin cp = g_new_consumer(sc->sc_geom); 78340ea77a0SAlexander Motin cp->flags |= G_CF_DIRECT_RECEIVE; 78489b17223SAlexander Motin if (g_attach(cp, pp) != 0) { 78589b17223SAlexander Motin g_destroy_consumer(cp); 78689b17223SAlexander Motin return (NULL); 78789b17223SAlexander Motin } 78889b17223SAlexander Motin if (g_access(cp, 1, 1, 1) != 0) { 78989b17223SAlexander Motin g_detach(cp); 79089b17223SAlexander Motin g_destroy_consumer(cp); 79189b17223SAlexander Motin return (NULL); 79289b17223SAlexander Motin } 79389b17223SAlexander Motin return (cp); 79489b17223SAlexander Motin } 79589b17223SAlexander Motin 79689b17223SAlexander Motin static u_int 79789b17223SAlexander Motin g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) 79889b17223SAlexander Motin { 79989b17223SAlexander Motin struct bio *bp; 80089b17223SAlexander Motin u_int nreqs = 0; 80189b17223SAlexander Motin 80289b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 80389b17223SAlexander Motin TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 80489b17223SAlexander Motin if (bp->bio_from == cp) 80589b17223SAlexander Motin nreqs++; 80689b17223SAlexander Motin } 80789b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 80889b17223SAlexander Motin return (nreqs); 80989b17223SAlexander Motin } 81089b17223SAlexander Motin 81189b17223SAlexander Motin u_int 81289b17223SAlexander Motin g_raid_nopens(struct g_raid_softc *sc) 81389b17223SAlexander Motin { 81489b17223SAlexander Motin struct g_raid_volume *vol; 81589b17223SAlexander Motin u_int opens; 81689b17223SAlexander Motin 81789b17223SAlexander Motin opens = 0; 81889b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 81989b17223SAlexander Motin if (vol->v_provider_open != 0) 82089b17223SAlexander Motin opens++; 82189b17223SAlexander Motin } 82289b17223SAlexander Motin return (opens); 82389b17223SAlexander Motin } 82489b17223SAlexander Motin 82589b17223SAlexander Motin static int 82689b17223SAlexander Motin g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) 82789b17223SAlexander Motin { 82889b17223SAlexander Motin 82989b17223SAlexander Motin if (cp->index > 0) { 83089b17223SAlexander Motin G_RAID_DEBUG1(2, sc, 83189b17223SAlexander Motin "I/O requests for %s exist, can't destroy it now.", 83289b17223SAlexander Motin cp->provider->name); 83389b17223SAlexander Motin return (1); 83489b17223SAlexander Motin } 83589b17223SAlexander Motin if (g_raid_nrequests(sc, cp) > 0) { 83689b17223SAlexander Motin G_RAID_DEBUG1(2, sc, 83789b17223SAlexander Motin "I/O requests for %s in queue, can't destroy it now.", 83889b17223SAlexander Motin cp->provider->name); 83989b17223SAlexander Motin return (1); 84089b17223SAlexander Motin } 84189b17223SAlexander Motin return (0); 84289b17223SAlexander Motin } 84389b17223SAlexander Motin 84489b17223SAlexander Motin static void 84589b17223SAlexander Motin g_raid_destroy_consumer(void *arg, int flags __unused) 84689b17223SAlexander Motin { 84789b17223SAlexander Motin struct g_consumer *cp; 84889b17223SAlexander Motin 84989b17223SAlexander Motin g_topology_assert(); 85089b17223SAlexander Motin 85189b17223SAlexander Motin cp = arg; 85289b17223SAlexander Motin G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 85389b17223SAlexander Motin g_detach(cp); 85489b17223SAlexander Motin g_destroy_consumer(cp); 85589b17223SAlexander Motin } 85689b17223SAlexander Motin 85789b17223SAlexander Motin void 85889b17223SAlexander Motin g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) 85989b17223SAlexander Motin { 86089b17223SAlexander Motin struct g_provider *pp; 86189b17223SAlexander Motin int retaste_wait; 86289b17223SAlexander Motin 86389b17223SAlexander Motin g_topology_assert_not(); 86489b17223SAlexander Motin 86589b17223SAlexander Motin g_topology_lock(); 86689b17223SAlexander Motin cp->private = NULL; 86789b17223SAlexander Motin if (g_raid_consumer_is_busy(sc, cp)) 86889b17223SAlexander Motin goto out; 86989b17223SAlexander Motin pp = cp->provider; 87089b17223SAlexander Motin retaste_wait = 0; 87189b17223SAlexander Motin if (cp->acw == 1) { 87289b17223SAlexander Motin if ((pp->geom->flags & G_GEOM_WITHER) == 0) 87389b17223SAlexander Motin retaste_wait = 1; 87489b17223SAlexander Motin } 87589b17223SAlexander Motin if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 87689b17223SAlexander Motin g_access(cp, -cp->acr, -cp->acw, -cp->ace); 87789b17223SAlexander Motin if (retaste_wait) { 87889b17223SAlexander Motin /* 87989b17223SAlexander Motin * After retaste event was send (inside g_access()), we can send 88089b17223SAlexander Motin * event to detach and destroy consumer. 88189b17223SAlexander Motin * A class, which has consumer to the given provider connected 88289b17223SAlexander Motin * will not receive retaste event for the provider. 88389b17223SAlexander Motin * This is the way how I ignore retaste events when I close 88489b17223SAlexander Motin * consumers opened for write: I detach and destroy consumer 88589b17223SAlexander Motin * after retaste event is sent. 88689b17223SAlexander Motin */ 88789b17223SAlexander Motin g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); 88889b17223SAlexander Motin goto out; 88989b17223SAlexander Motin } 89089b17223SAlexander Motin G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); 89189b17223SAlexander Motin g_detach(cp); 89289b17223SAlexander Motin g_destroy_consumer(cp); 89389b17223SAlexander Motin out: 89489b17223SAlexander Motin g_topology_unlock(); 89589b17223SAlexander Motin } 89689b17223SAlexander Motin 89789b17223SAlexander Motin static void 89889b17223SAlexander Motin g_raid_orphan(struct g_consumer *cp) 89989b17223SAlexander Motin { 90089b17223SAlexander Motin struct g_raid_disk *disk; 90189b17223SAlexander Motin 90289b17223SAlexander Motin g_topology_assert(); 90389b17223SAlexander Motin 90489b17223SAlexander Motin disk = cp->private; 90589b17223SAlexander Motin if (disk == NULL) 90689b17223SAlexander Motin return; 90789b17223SAlexander Motin g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, 90889b17223SAlexander Motin G_RAID_EVENT_DISK); 90989b17223SAlexander Motin } 91089b17223SAlexander Motin 911a479c51bSAlexander Motin static void 91289b17223SAlexander Motin g_raid_clean(struct g_raid_volume *vol, int acw) 91389b17223SAlexander Motin { 91489b17223SAlexander Motin struct g_raid_softc *sc; 91589b17223SAlexander Motin int timeout; 91689b17223SAlexander Motin 91789b17223SAlexander Motin sc = vol->v_softc; 91889b17223SAlexander Motin g_topology_assert_not(); 91989b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 92089b17223SAlexander Motin 92189b17223SAlexander Motin // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 922a479c51bSAlexander Motin // return; 92389b17223SAlexander Motin if (!vol->v_dirty) 924a479c51bSAlexander Motin return; 92589b17223SAlexander Motin if (vol->v_writes > 0) 926a479c51bSAlexander Motin return; 92789b17223SAlexander Motin if (acw > 0 || (acw == -1 && 92889b17223SAlexander Motin vol->v_provider != NULL && vol->v_provider->acw > 0)) { 92989b17223SAlexander Motin timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); 930a479c51bSAlexander Motin if (!g_raid_shutdown && timeout > 0) 931a479c51bSAlexander Motin return; 93289b17223SAlexander Motin } 93389b17223SAlexander Motin vol->v_dirty = 0; 93489b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", 93589b17223SAlexander Motin vol->v_name); 93689b17223SAlexander Motin g_raid_write_metadata(sc, vol, NULL, NULL); 93789b17223SAlexander Motin } 93889b17223SAlexander Motin 93989b17223SAlexander Motin static void 94089b17223SAlexander Motin g_raid_dirty(struct g_raid_volume *vol) 94189b17223SAlexander Motin { 94289b17223SAlexander Motin struct g_raid_softc *sc; 94389b17223SAlexander Motin 94489b17223SAlexander Motin sc = vol->v_softc; 94589b17223SAlexander Motin g_topology_assert_not(); 94689b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 94789b17223SAlexander Motin 94889b17223SAlexander Motin // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 94989b17223SAlexander Motin // return; 95089b17223SAlexander Motin vol->v_dirty = 1; 95189b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", 95289b17223SAlexander Motin vol->v_name); 95389b17223SAlexander Motin g_raid_write_metadata(sc, vol, NULL, NULL); 95489b17223SAlexander Motin } 95589b17223SAlexander Motin 95689b17223SAlexander Motin void 95789b17223SAlexander Motin g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) 95889b17223SAlexander Motin { 95989b17223SAlexander Motin struct g_raid_volume *vol; 96089b17223SAlexander Motin struct g_raid_subdisk *sd; 96189b17223SAlexander Motin struct bio_queue_head queue; 96289b17223SAlexander Motin struct bio *cbp; 96389b17223SAlexander Motin int i; 96489b17223SAlexander Motin 96589b17223SAlexander Motin vol = tr->tro_volume; 96689b17223SAlexander Motin 96789b17223SAlexander Motin /* 96889b17223SAlexander Motin * Allocate all bios before sending any request, so we can return 96989b17223SAlexander Motin * ENOMEM in nice and clean way. 97089b17223SAlexander Motin */ 97189b17223SAlexander Motin bioq_init(&queue); 97289b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 97389b17223SAlexander Motin sd = &vol->v_subdisks[i]; 97489b17223SAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_NONE || 97589b17223SAlexander Motin sd->sd_state == G_RAID_SUBDISK_S_FAILED) 97689b17223SAlexander Motin continue; 97789b17223SAlexander Motin cbp = g_clone_bio(bp); 97889b17223SAlexander Motin if (cbp == NULL) 97989b17223SAlexander Motin goto failure; 98089b17223SAlexander Motin cbp->bio_caller1 = sd; 98189b17223SAlexander Motin bioq_insert_tail(&queue, cbp); 98289b17223SAlexander Motin } 983b43560abSAlexander Motin while ((cbp = bioq_takefirst(&queue)) != NULL) { 98489b17223SAlexander Motin sd = cbp->bio_caller1; 98589b17223SAlexander Motin cbp->bio_caller1 = NULL; 98689b17223SAlexander Motin g_raid_subdisk_iostart(sd, cbp); 98789b17223SAlexander Motin } 98889b17223SAlexander Motin return; 98989b17223SAlexander Motin failure: 990b43560abSAlexander Motin while ((cbp = bioq_takefirst(&queue)) != NULL) 99189b17223SAlexander Motin g_destroy_bio(cbp); 99289b17223SAlexander Motin if (bp->bio_error == 0) 99389b17223SAlexander Motin bp->bio_error = ENOMEM; 99489b17223SAlexander Motin g_raid_iodone(bp, bp->bio_error); 99589b17223SAlexander Motin } 99689b17223SAlexander Motin 99789b17223SAlexander Motin static void 99889b17223SAlexander Motin g_raid_tr_kerneldump_common_done(struct bio *bp) 99989b17223SAlexander Motin { 100089b17223SAlexander Motin 100189b17223SAlexander Motin bp->bio_flags |= BIO_DONE; 100289b17223SAlexander Motin } 100389b17223SAlexander Motin 100489b17223SAlexander Motin int 100589b17223SAlexander Motin g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, 100689b17223SAlexander Motin void *virtual, vm_offset_t physical, off_t offset, size_t length) 100789b17223SAlexander Motin { 100889b17223SAlexander Motin struct g_raid_softc *sc; 100989b17223SAlexander Motin struct g_raid_volume *vol; 101089b17223SAlexander Motin struct bio bp; 101189b17223SAlexander Motin 101289b17223SAlexander Motin vol = tr->tro_volume; 101389b17223SAlexander Motin sc = vol->v_softc; 101489b17223SAlexander Motin 1015c55f5707SWarner Losh g_reset_bio(&bp); 101689b17223SAlexander Motin bp.bio_cmd = BIO_WRITE; 101789b17223SAlexander Motin bp.bio_done = g_raid_tr_kerneldump_common_done; 101889b17223SAlexander Motin bp.bio_attribute = NULL; 101989b17223SAlexander Motin bp.bio_offset = offset; 102089b17223SAlexander Motin bp.bio_length = length; 102189b17223SAlexander Motin bp.bio_data = virtual; 102289b17223SAlexander Motin bp.bio_to = vol->v_provider; 102389b17223SAlexander Motin 102489b17223SAlexander Motin g_raid_start(&bp); 102589b17223SAlexander Motin while (!(bp.bio_flags & BIO_DONE)) { 102689b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Poll..."); 102789b17223SAlexander Motin g_raid_poll(sc); 102889b17223SAlexander Motin DELAY(10); 102989b17223SAlexander Motin } 103089b17223SAlexander Motin 103189b17223SAlexander Motin return (bp.bio_error != 0 ? EIO : 0); 103289b17223SAlexander Motin } 103389b17223SAlexander Motin 103489b17223SAlexander Motin static int 103589b17223SAlexander Motin g_raid_dump(void *arg, 103689b17223SAlexander Motin void *virtual, vm_offset_t physical, off_t offset, size_t length) 103789b17223SAlexander Motin { 103889b17223SAlexander Motin struct g_raid_volume *vol; 103989b17223SAlexander Motin int error; 104089b17223SAlexander Motin 104189b17223SAlexander Motin vol = (struct g_raid_volume *)arg; 104289b17223SAlexander Motin G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", 104389b17223SAlexander Motin (long long unsigned)offset, (long long unsigned)length); 104489b17223SAlexander Motin 104589b17223SAlexander Motin error = G_RAID_TR_KERNELDUMP(vol->v_tr, 104689b17223SAlexander Motin virtual, physical, offset, length); 104789b17223SAlexander Motin return (error); 104889b17223SAlexander Motin } 104989b17223SAlexander Motin 105089b17223SAlexander Motin static void 105189b17223SAlexander Motin g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) 105289b17223SAlexander Motin { 105389b17223SAlexander Motin struct g_kerneldump *gkd; 105489b17223SAlexander Motin struct g_provider *pp; 105589b17223SAlexander Motin struct g_raid_volume *vol; 105689b17223SAlexander Motin 105789b17223SAlexander Motin gkd = (struct g_kerneldump*)bp->bio_data; 105889b17223SAlexander Motin pp = bp->bio_to; 105989b17223SAlexander Motin vol = pp->private; 106089b17223SAlexander Motin g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", 106189b17223SAlexander Motin pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 106289b17223SAlexander Motin gkd->di.dumper = g_raid_dump; 106389b17223SAlexander Motin gkd->di.priv = vol; 106489b17223SAlexander Motin gkd->di.blocksize = vol->v_sectorsize; 106589b17223SAlexander Motin gkd->di.maxiosize = DFLTPHYS; 106689b17223SAlexander Motin gkd->di.mediaoffset = gkd->offset; 106789b17223SAlexander Motin if ((gkd->offset + gkd->length) > vol->v_mediasize) 106889b17223SAlexander Motin gkd->length = vol->v_mediasize - gkd->offset; 106989b17223SAlexander Motin gkd->di.mediasize = gkd->length; 107089b17223SAlexander Motin g_io_deliver(bp, 0); 107189b17223SAlexander Motin } 107289b17223SAlexander Motin 107389b17223SAlexander Motin static void 1074609a7474SAlexander Motin g_raid_candelete(struct g_raid_softc *sc, struct bio *bp) 1075609a7474SAlexander Motin { 1076609a7474SAlexander Motin struct g_provider *pp; 1077609a7474SAlexander Motin struct g_raid_volume *vol; 1078609a7474SAlexander Motin struct g_raid_subdisk *sd; 1079438622afSMark Johnston int i, val; 1080609a7474SAlexander Motin 1081609a7474SAlexander Motin pp = bp->bio_to; 1082609a7474SAlexander Motin vol = pp->private; 1083609a7474SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 1084609a7474SAlexander Motin sd = &vol->v_subdisks[i]; 1085609a7474SAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_NONE) 1086609a7474SAlexander Motin continue; 1087438622afSMark Johnston if (sd->sd_disk->d_candelete) 1088609a7474SAlexander Motin break; 1089609a7474SAlexander Motin } 1090438622afSMark Johnston val = i < vol->v_disks_count; 1091438622afSMark Johnston g_handleattr(bp, "GEOM::candelete", &val, sizeof(val)); 1092609a7474SAlexander Motin } 1093609a7474SAlexander Motin 1094609a7474SAlexander Motin static void 109589b17223SAlexander Motin g_raid_start(struct bio *bp) 109689b17223SAlexander Motin { 109789b17223SAlexander Motin struct g_raid_softc *sc; 109889b17223SAlexander Motin 109989b17223SAlexander Motin sc = bp->bio_to->geom->softc; 110089b17223SAlexander Motin /* 110189b17223SAlexander Motin * If sc == NULL or there are no valid disks, provider's error 110289b17223SAlexander Motin * should be set and g_raid_start() should not be called at all. 110389b17223SAlexander Motin */ 110489b17223SAlexander Motin // KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, 110589b17223SAlexander Motin // ("Provider's error should be set (error=%d)(mirror=%s).", 110689b17223SAlexander Motin // bp->bio_to->error, bp->bio_to->name)); 110789b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Request received."); 110889b17223SAlexander Motin 110989b17223SAlexander Motin switch (bp->bio_cmd) { 111089b17223SAlexander Motin case BIO_READ: 111189b17223SAlexander Motin case BIO_WRITE: 111289b17223SAlexander Motin case BIO_DELETE: 111389b17223SAlexander Motin case BIO_FLUSH: 111489b17223SAlexander Motin break; 111589b17223SAlexander Motin case BIO_GETATTR: 1116609a7474SAlexander Motin if (!strcmp(bp->bio_attribute, "GEOM::candelete")) 1117609a7474SAlexander Motin g_raid_candelete(sc, bp); 1118609a7474SAlexander Motin else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 111989b17223SAlexander Motin g_raid_kerneldump(sc, bp); 112089b17223SAlexander Motin else 112189b17223SAlexander Motin g_io_deliver(bp, EOPNOTSUPP); 112289b17223SAlexander Motin return; 112389b17223SAlexander Motin default: 112489b17223SAlexander Motin g_io_deliver(bp, EOPNOTSUPP); 112589b17223SAlexander Motin return; 112689b17223SAlexander Motin } 112789b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 11283ab0187aSAlexander Motin bioq_insert_tail(&sc->sc_queue, bp); 112989b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 113089b17223SAlexander Motin if (!dumping) { 113189b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); 113289b17223SAlexander Motin wakeup(sc); 113389b17223SAlexander Motin } 113489b17223SAlexander Motin } 113589b17223SAlexander Motin 113689b17223SAlexander Motin static int 113789b17223SAlexander Motin g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) 113889b17223SAlexander Motin { 113989b17223SAlexander Motin /* 114089b17223SAlexander Motin * 5 cases: 114189b17223SAlexander Motin * (1) bp entirely below NO 114289b17223SAlexander Motin * (2) bp entirely above NO 114389b17223SAlexander Motin * (3) bp start below, but end in range YES 114489b17223SAlexander Motin * (4) bp entirely within YES 114589b17223SAlexander Motin * (5) bp starts within, ends above YES 114689b17223SAlexander Motin * 114789b17223SAlexander Motin * lock range 10-19 (offset 10 length 10) 114889b17223SAlexander Motin * (1) 1-5: first if kicks it out 114989b17223SAlexander Motin * (2) 30-35: second if kicks it out 115089b17223SAlexander Motin * (3) 5-15: passes both ifs 115189b17223SAlexander Motin * (4) 12-14: passes both ifs 115289b17223SAlexander Motin * (5) 19-20: passes both 115389b17223SAlexander Motin */ 115489b17223SAlexander Motin off_t lend = lstart + len - 1; 115589b17223SAlexander Motin off_t bstart = bp->bio_offset; 115689b17223SAlexander Motin off_t bend = bp->bio_offset + bp->bio_length - 1; 115789b17223SAlexander Motin 115889b17223SAlexander Motin if (bend < lstart) 115989b17223SAlexander Motin return (0); 116089b17223SAlexander Motin if (lend < bstart) 116189b17223SAlexander Motin return (0); 116289b17223SAlexander Motin return (1); 116389b17223SAlexander Motin } 116489b17223SAlexander Motin 116589b17223SAlexander Motin static int 116689b17223SAlexander Motin g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) 116789b17223SAlexander Motin { 116889b17223SAlexander Motin struct g_raid_lock *lp; 116989b17223SAlexander Motin 117089b17223SAlexander Motin sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); 117189b17223SAlexander Motin 117289b17223SAlexander Motin LIST_FOREACH(lp, &vol->v_locks, l_next) { 117389b17223SAlexander Motin if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) 117489b17223SAlexander Motin return (1); 117589b17223SAlexander Motin } 117689b17223SAlexander Motin return (0); 117789b17223SAlexander Motin } 117889b17223SAlexander Motin 117989b17223SAlexander Motin static void 118089b17223SAlexander Motin g_raid_start_request(struct bio *bp) 118189b17223SAlexander Motin { 118289b17223SAlexander Motin struct g_raid_softc *sc; 118389b17223SAlexander Motin struct g_raid_volume *vol; 118489b17223SAlexander Motin 118589b17223SAlexander Motin sc = bp->bio_to->geom->softc; 118689b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 118789b17223SAlexander Motin vol = bp->bio_to->private; 118889b17223SAlexander Motin 118989b17223SAlexander Motin /* 119089b17223SAlexander Motin * Check to see if this item is in a locked range. If so, 119189b17223SAlexander Motin * queue it to our locked queue and return. We'll requeue 119289b17223SAlexander Motin * it when the range is unlocked. Internal I/O for the 119389b17223SAlexander Motin * rebuild/rescan/recovery process is excluded from this 119489b17223SAlexander Motin * check so we can actually do the recovery. 119589b17223SAlexander Motin */ 119689b17223SAlexander Motin if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && 119789b17223SAlexander Motin g_raid_is_in_locked_range(vol, bp)) { 119889b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Defer request."); 119989b17223SAlexander Motin bioq_insert_tail(&vol->v_locked, bp); 120089b17223SAlexander Motin return; 120189b17223SAlexander Motin } 120289b17223SAlexander Motin 120389b17223SAlexander Motin /* 120489b17223SAlexander Motin * If we're actually going to do the write/delete, then 120589b17223SAlexander Motin * update the idle stats for the volume. 120689b17223SAlexander Motin */ 120789b17223SAlexander Motin if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 120889b17223SAlexander Motin if (!vol->v_dirty) 120989b17223SAlexander Motin g_raid_dirty(vol); 121089b17223SAlexander Motin vol->v_writes++; 121189b17223SAlexander Motin } 121289b17223SAlexander Motin 121389b17223SAlexander Motin /* 121489b17223SAlexander Motin * Put request onto inflight queue, so we can check if new 121589b17223SAlexander Motin * synchronization requests don't collide with it. Then tell 121689b17223SAlexander Motin * the transformation layer to start the I/O. 121789b17223SAlexander Motin */ 121889b17223SAlexander Motin bioq_insert_tail(&vol->v_inflight, bp); 121989b17223SAlexander Motin G_RAID_LOGREQ(4, bp, "Request started"); 122089b17223SAlexander Motin G_RAID_TR_IOSTART(vol->v_tr, bp); 122189b17223SAlexander Motin } 122289b17223SAlexander Motin 122389b17223SAlexander Motin static void 122489b17223SAlexander Motin g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) 122589b17223SAlexander Motin { 122689b17223SAlexander Motin off_t off, len; 122789b17223SAlexander Motin struct bio *nbp; 122889b17223SAlexander Motin struct g_raid_lock *lp; 122989b17223SAlexander Motin 123089b17223SAlexander Motin vol->v_pending_lock = 0; 123189b17223SAlexander Motin LIST_FOREACH(lp, &vol->v_locks, l_next) { 123289b17223SAlexander Motin if (lp->l_pending) { 123389b17223SAlexander Motin off = lp->l_offset; 123489b17223SAlexander Motin len = lp->l_length; 123589b17223SAlexander Motin lp->l_pending = 0; 123689b17223SAlexander Motin TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { 123789b17223SAlexander Motin if (g_raid_bio_overlaps(nbp, off, len)) 123889b17223SAlexander Motin lp->l_pending++; 123989b17223SAlexander Motin } 124089b17223SAlexander Motin if (lp->l_pending) { 124189b17223SAlexander Motin vol->v_pending_lock = 1; 124289b17223SAlexander Motin G_RAID_DEBUG1(4, vol->v_softc, 124389b17223SAlexander Motin "Deferred lock(%jd, %jd) has %d pending", 124489b17223SAlexander Motin (intmax_t)off, (intmax_t)(off + len), 124589b17223SAlexander Motin lp->l_pending); 124689b17223SAlexander Motin continue; 124789b17223SAlexander Motin } 124889b17223SAlexander Motin G_RAID_DEBUG1(4, vol->v_softc, 124989b17223SAlexander Motin "Deferred lock of %jd to %jd completed", 125089b17223SAlexander Motin (intmax_t)off, (intmax_t)(off + len)); 125189b17223SAlexander Motin G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 125289b17223SAlexander Motin } 125389b17223SAlexander Motin } 125489b17223SAlexander Motin } 125589b17223SAlexander Motin 125689b17223SAlexander Motin void 125789b17223SAlexander Motin g_raid_iodone(struct bio *bp, int error) 125889b17223SAlexander Motin { 125989b17223SAlexander Motin struct g_raid_softc *sc; 126089b17223SAlexander Motin struct g_raid_volume *vol; 126189b17223SAlexander Motin 126289b17223SAlexander Motin sc = bp->bio_to->geom->softc; 126389b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 126489b17223SAlexander Motin vol = bp->bio_to->private; 126589b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Request done: %d.", error); 126689b17223SAlexander Motin 126789b17223SAlexander Motin /* Update stats if we done write/delete. */ 126889b17223SAlexander Motin if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 126989b17223SAlexander Motin vol->v_writes--; 127089b17223SAlexander Motin vol->v_last_write = time_uptime; 127189b17223SAlexander Motin } 127289b17223SAlexander Motin 127389b17223SAlexander Motin bioq_remove(&vol->v_inflight, bp); 127489b17223SAlexander Motin if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) 127589b17223SAlexander Motin g_raid_finish_with_locked_ranges(vol, bp); 127689b17223SAlexander Motin getmicrouptime(&vol->v_last_done); 127789b17223SAlexander Motin g_io_deliver(bp, error); 127889b17223SAlexander Motin } 127989b17223SAlexander Motin 128089b17223SAlexander Motin int 128189b17223SAlexander Motin g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, 128289b17223SAlexander Motin struct bio *ignore, void *argp) 128389b17223SAlexander Motin { 128489b17223SAlexander Motin struct g_raid_softc *sc; 128589b17223SAlexander Motin struct g_raid_lock *lp; 128689b17223SAlexander Motin struct bio *bp; 128789b17223SAlexander Motin 128889b17223SAlexander Motin sc = vol->v_softc; 128989b17223SAlexander Motin lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); 129089b17223SAlexander Motin LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); 129189b17223SAlexander Motin lp->l_offset = off; 129289b17223SAlexander Motin lp->l_length = len; 129389b17223SAlexander Motin lp->l_callback_arg = argp; 129489b17223SAlexander Motin 129589b17223SAlexander Motin lp->l_pending = 0; 129689b17223SAlexander Motin TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { 129789b17223SAlexander Motin if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) 129889b17223SAlexander Motin lp->l_pending++; 129989b17223SAlexander Motin } 130089b17223SAlexander Motin 130189b17223SAlexander Motin /* 130289b17223SAlexander Motin * If there are any writes that are pending, we return EBUSY. All 130389b17223SAlexander Motin * callers will have to wait until all pending writes clear. 130489b17223SAlexander Motin */ 130589b17223SAlexander Motin if (lp->l_pending > 0) { 130689b17223SAlexander Motin vol->v_pending_lock = 1; 130789b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", 130889b17223SAlexander Motin (intmax_t)off, (intmax_t)(off+len), lp->l_pending); 130989b17223SAlexander Motin return (EBUSY); 131089b17223SAlexander Motin } 131189b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", 131289b17223SAlexander Motin (intmax_t)off, (intmax_t)(off+len)); 131389b17223SAlexander Motin G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 131489b17223SAlexander Motin return (0); 131589b17223SAlexander Motin } 131689b17223SAlexander Motin 131789b17223SAlexander Motin int 131889b17223SAlexander Motin g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) 131989b17223SAlexander Motin { 132089b17223SAlexander Motin struct g_raid_lock *lp; 132189b17223SAlexander Motin struct g_raid_softc *sc; 132289b17223SAlexander Motin struct bio *bp; 132389b17223SAlexander Motin 132489b17223SAlexander Motin sc = vol->v_softc; 132589b17223SAlexander Motin LIST_FOREACH(lp, &vol->v_locks, l_next) { 132689b17223SAlexander Motin if (lp->l_offset == off && lp->l_length == len) { 132789b17223SAlexander Motin LIST_REMOVE(lp, l_next); 132889b17223SAlexander Motin /* XXX 132989b17223SAlexander Motin * Right now we just put them all back on the queue 133089b17223SAlexander Motin * and hope for the best. We hope this because any 133189b17223SAlexander Motin * locked ranges will go right back on this list 133289b17223SAlexander Motin * when the worker thread runs. 133389b17223SAlexander Motin * XXX 133489b17223SAlexander Motin */ 133589b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", 133689b17223SAlexander Motin (intmax_t)lp->l_offset, 133789b17223SAlexander Motin (intmax_t)(lp->l_offset+lp->l_length)); 133889b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 133989b17223SAlexander Motin while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) 13403ab0187aSAlexander Motin bioq_insert_tail(&sc->sc_queue, bp); 134189b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 134289b17223SAlexander Motin free(lp, M_RAID); 134389b17223SAlexander Motin return (0); 134489b17223SAlexander Motin } 134589b17223SAlexander Motin } 134689b17223SAlexander Motin return (EINVAL); 134789b17223SAlexander Motin } 134889b17223SAlexander Motin 134989b17223SAlexander Motin void 135089b17223SAlexander Motin g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) 135189b17223SAlexander Motin { 135289b17223SAlexander Motin struct g_consumer *cp; 135389b17223SAlexander Motin struct g_raid_disk *disk, *tdisk; 135489b17223SAlexander Motin 135589b17223SAlexander Motin bp->bio_caller1 = sd; 135689b17223SAlexander Motin 135789b17223SAlexander Motin /* 135889b17223SAlexander Motin * Make sure that the disk is present. Generally it is a task of 135989b17223SAlexander Motin * transformation layers to not send requests to absent disks, but 136089b17223SAlexander Motin * it is better to be safe and report situation then sorry. 136189b17223SAlexander Motin */ 136289b17223SAlexander Motin if (sd->sd_disk == NULL) { 136389b17223SAlexander Motin G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); 136489b17223SAlexander Motin nodisk: 136589b17223SAlexander Motin bp->bio_from = NULL; 136689b17223SAlexander Motin bp->bio_to = NULL; 136789b17223SAlexander Motin bp->bio_error = ENXIO; 136889b17223SAlexander Motin g_raid_disk_done(bp); 136989b17223SAlexander Motin return; 137089b17223SAlexander Motin } 137189b17223SAlexander Motin disk = sd->sd_disk; 137289b17223SAlexander Motin if (disk->d_state != G_RAID_DISK_S_ACTIVE && 137389b17223SAlexander Motin disk->d_state != G_RAID_DISK_S_FAILED) { 137489b17223SAlexander Motin G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " 137589b17223SAlexander Motin "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 137689b17223SAlexander Motin goto nodisk; 137789b17223SAlexander Motin } 137889b17223SAlexander Motin 137989b17223SAlexander Motin cp = disk->d_consumer; 138089b17223SAlexander Motin bp->bio_from = cp; 138189b17223SAlexander Motin bp->bio_to = cp->provider; 138289b17223SAlexander Motin cp->index++; 138389b17223SAlexander Motin 138489b17223SAlexander Motin /* Update average disks load. */ 138589b17223SAlexander Motin TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { 138689b17223SAlexander Motin if (tdisk->d_consumer == NULL) 138789b17223SAlexander Motin tdisk->d_load = 0; 138889b17223SAlexander Motin else 138989b17223SAlexander Motin tdisk->d_load = (tdisk->d_consumer->index * 139089b17223SAlexander Motin G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; 139189b17223SAlexander Motin } 139289b17223SAlexander Motin 139389b17223SAlexander Motin disk->d_last_offset = bp->bio_offset + bp->bio_length; 139489b17223SAlexander Motin if (dumping) { 139589b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Sending dumping request."); 139689b17223SAlexander Motin if (bp->bio_cmd == BIO_WRITE) { 139789b17223SAlexander Motin bp->bio_error = g_raid_subdisk_kerneldump(sd, 139889b17223SAlexander Motin bp->bio_data, 0, bp->bio_offset, bp->bio_length); 139989b17223SAlexander Motin } else 140089b17223SAlexander Motin bp->bio_error = EOPNOTSUPP; 140189b17223SAlexander Motin g_raid_disk_done(bp); 140289b17223SAlexander Motin } else { 140389b17223SAlexander Motin bp->bio_done = g_raid_disk_done; 140489b17223SAlexander Motin bp->bio_offset += sd->sd_offset; 140589b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Sending request."); 140689b17223SAlexander Motin g_io_request(bp, cp); 140789b17223SAlexander Motin } 140889b17223SAlexander Motin } 140989b17223SAlexander Motin 141089b17223SAlexander Motin int 141189b17223SAlexander Motin g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, 141289b17223SAlexander Motin void *virtual, vm_offset_t physical, off_t offset, size_t length) 141389b17223SAlexander Motin { 141489b17223SAlexander Motin 141589b17223SAlexander Motin if (sd->sd_disk == NULL) 141689b17223SAlexander Motin return (ENXIO); 141789b17223SAlexander Motin if (sd->sd_disk->d_kd.di.dumper == NULL) 141889b17223SAlexander Motin return (EOPNOTSUPP); 141989b17223SAlexander Motin return (dump_write(&sd->sd_disk->d_kd.di, 142089b17223SAlexander Motin virtual, physical, 142189b17223SAlexander Motin sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, 142289b17223SAlexander Motin length)); 142389b17223SAlexander Motin } 142489b17223SAlexander Motin 142589b17223SAlexander Motin static void 142689b17223SAlexander Motin g_raid_disk_done(struct bio *bp) 142789b17223SAlexander Motin { 142889b17223SAlexander Motin struct g_raid_softc *sc; 142989b17223SAlexander Motin struct g_raid_subdisk *sd; 143089b17223SAlexander Motin 143189b17223SAlexander Motin sd = bp->bio_caller1; 143289b17223SAlexander Motin sc = sd->sd_softc; 143389b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 14343ab0187aSAlexander Motin bioq_insert_tail(&sc->sc_queue, bp); 143589b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 143689b17223SAlexander Motin if (!dumping) 143789b17223SAlexander Motin wakeup(sc); 143889b17223SAlexander Motin } 143989b17223SAlexander Motin 144089b17223SAlexander Motin static void 144189b17223SAlexander Motin g_raid_disk_done_request(struct bio *bp) 144289b17223SAlexander Motin { 144389b17223SAlexander Motin struct g_raid_softc *sc; 144489b17223SAlexander Motin struct g_raid_disk *disk; 144589b17223SAlexander Motin struct g_raid_subdisk *sd; 144689b17223SAlexander Motin struct g_raid_volume *vol; 144789b17223SAlexander Motin 144889b17223SAlexander Motin g_topology_assert_not(); 144989b17223SAlexander Motin 145089b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); 145189b17223SAlexander Motin sd = bp->bio_caller1; 145289b17223SAlexander Motin sc = sd->sd_softc; 145389b17223SAlexander Motin vol = sd->sd_volume; 145489b17223SAlexander Motin if (bp->bio_from != NULL) { 145589b17223SAlexander Motin bp->bio_from->index--; 145689b17223SAlexander Motin disk = bp->bio_from->private; 145789b17223SAlexander Motin if (disk == NULL) 145889b17223SAlexander Motin g_raid_kill_consumer(sc, bp->bio_from); 145989b17223SAlexander Motin } 146089b17223SAlexander Motin bp->bio_offset -= sd->sd_offset; 146189b17223SAlexander Motin 146289b17223SAlexander Motin G_RAID_TR_IODONE(vol->v_tr, sd, bp); 146389b17223SAlexander Motin } 146489b17223SAlexander Motin 146589b17223SAlexander Motin static void 146689b17223SAlexander Motin g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) 146789b17223SAlexander Motin { 146889b17223SAlexander Motin 146989b17223SAlexander Motin if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) 147089b17223SAlexander Motin ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); 147189b17223SAlexander Motin else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) 147289b17223SAlexander Motin ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); 147389b17223SAlexander Motin else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) 147489b17223SAlexander Motin ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); 147589b17223SAlexander Motin else 147689b17223SAlexander Motin ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); 147789b17223SAlexander Motin if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { 147889b17223SAlexander Motin KASSERT(ep->e_error == 0, 147989b17223SAlexander Motin ("Error cannot be handled.")); 148089b17223SAlexander Motin g_raid_event_free(ep); 148189b17223SAlexander Motin } else { 148289b17223SAlexander Motin ep->e_flags |= G_RAID_EVENT_DONE; 148389b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); 148489b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 148589b17223SAlexander Motin wakeup(ep); 148689b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 148789b17223SAlexander Motin } 148889b17223SAlexander Motin } 148989b17223SAlexander Motin 149089b17223SAlexander Motin /* 149189b17223SAlexander Motin * Worker thread. 149289b17223SAlexander Motin */ 149389b17223SAlexander Motin static void 149489b17223SAlexander Motin g_raid_worker(void *arg) 149589b17223SAlexander Motin { 149689b17223SAlexander Motin struct g_raid_softc *sc; 149789b17223SAlexander Motin struct g_raid_event *ep; 149889b17223SAlexander Motin struct g_raid_volume *vol; 149989b17223SAlexander Motin struct bio *bp; 150089b17223SAlexander Motin struct timeval now, t; 150189b17223SAlexander Motin int timeout, rv; 150289b17223SAlexander Motin 150389b17223SAlexander Motin sc = arg; 150489b17223SAlexander Motin thread_lock(curthread); 150589b17223SAlexander Motin sched_prio(curthread, PRIBIO); 150689b17223SAlexander Motin thread_unlock(curthread); 150789b17223SAlexander Motin 150889b17223SAlexander Motin sx_xlock(&sc->sc_lock); 150989b17223SAlexander Motin for (;;) { 151089b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 151189b17223SAlexander Motin /* 151289b17223SAlexander Motin * First take a look at events. 151389b17223SAlexander Motin * This is important to handle events before any I/O requests. 151489b17223SAlexander Motin */ 151589b17223SAlexander Motin bp = NULL; 151689b17223SAlexander Motin vol = NULL; 151789b17223SAlexander Motin rv = 0; 151889b17223SAlexander Motin ep = TAILQ_FIRST(&sc->sc_events); 151989b17223SAlexander Motin if (ep != NULL) 152089b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_events, ep, e_next); 152189b17223SAlexander Motin else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) 152289b17223SAlexander Motin ; 152389b17223SAlexander Motin else { 152489b17223SAlexander Motin getmicrouptime(&now); 152589b17223SAlexander Motin t = now; 152689b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 152789b17223SAlexander Motin if (bioq_first(&vol->v_inflight) == NULL && 152889b17223SAlexander Motin vol->v_tr && 152989b17223SAlexander Motin timevalcmp(&vol->v_last_done, &t, < )) 153089b17223SAlexander Motin t = vol->v_last_done; 153189b17223SAlexander Motin } 153289b17223SAlexander Motin timevalsub(&t, &now); 153389b17223SAlexander Motin timeout = g_raid_idle_threshold + 153489b17223SAlexander Motin t.tv_sec * 1000000 + t.tv_usec; 153589b17223SAlexander Motin if (timeout > 0) { 153689b17223SAlexander Motin /* 153789b17223SAlexander Motin * Two steps to avoid overflows at HZ=1000 153889b17223SAlexander Motin * and idle timeouts > 2.1s. Some rounding 153989b17223SAlexander Motin * errors can occur, but they are < 1tick, 154089b17223SAlexander Motin * which is deemed to be close enough for 154189b17223SAlexander Motin * this purpose. 154289b17223SAlexander Motin */ 154389b17223SAlexander Motin int micpertic = 1000000 / hz; 154489b17223SAlexander Motin timeout = (timeout + micpertic - 1) / micpertic; 154589b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 154689b17223SAlexander Motin MSLEEP(rv, sc, &sc->sc_queue_mtx, 154789b17223SAlexander Motin PRIBIO | PDROP, "-", timeout); 154889b17223SAlexander Motin sx_xlock(&sc->sc_lock); 154989b17223SAlexander Motin goto process; 155089b17223SAlexander Motin } else 155189b17223SAlexander Motin rv = EWOULDBLOCK; 155289b17223SAlexander Motin } 155389b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 155489b17223SAlexander Motin process: 155589b17223SAlexander Motin if (ep != NULL) { 155689b17223SAlexander Motin g_raid_handle_event(sc, ep); 155789b17223SAlexander Motin } else if (bp != NULL) { 155889b17223SAlexander Motin if (bp->bio_to != NULL && 155989b17223SAlexander Motin bp->bio_to->geom == sc->sc_geom) 156089b17223SAlexander Motin g_raid_start_request(bp); 156189b17223SAlexander Motin else 156289b17223SAlexander Motin g_raid_disk_done_request(bp); 156389b17223SAlexander Motin } else if (rv == EWOULDBLOCK) { 156489b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 156589b17223SAlexander Motin g_raid_clean(vol, -1); 156689b17223SAlexander Motin if (bioq_first(&vol->v_inflight) == NULL && 156789b17223SAlexander Motin vol->v_tr) { 156889b17223SAlexander Motin t.tv_sec = g_raid_idle_threshold / 1000000; 156989b17223SAlexander Motin t.tv_usec = g_raid_idle_threshold % 1000000; 157089b17223SAlexander Motin timevaladd(&t, &vol->v_last_done); 157189b17223SAlexander Motin getmicrouptime(&now); 157289b17223SAlexander Motin if (timevalcmp(&t, &now, <= )) { 157389b17223SAlexander Motin G_RAID_TR_IDLE(vol->v_tr); 157489b17223SAlexander Motin vol->v_last_done = now; 157589b17223SAlexander Motin } 157689b17223SAlexander Motin } 157789b17223SAlexander Motin } 157889b17223SAlexander Motin } 157989b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_HARD) 158089b17223SAlexander Motin g_raid_destroy_node(sc, 1); /* May not return. */ 158189b17223SAlexander Motin } 158289b17223SAlexander Motin } 158389b17223SAlexander Motin 158489b17223SAlexander Motin static void 158589b17223SAlexander Motin g_raid_poll(struct g_raid_softc *sc) 158689b17223SAlexander Motin { 158789b17223SAlexander Motin struct g_raid_event *ep; 158889b17223SAlexander Motin struct bio *bp; 158989b17223SAlexander Motin 159089b17223SAlexander Motin sx_xlock(&sc->sc_lock); 159189b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 159289b17223SAlexander Motin /* 159389b17223SAlexander Motin * First take a look at events. 159489b17223SAlexander Motin * This is important to handle events before any I/O requests. 159589b17223SAlexander Motin */ 159689b17223SAlexander Motin ep = TAILQ_FIRST(&sc->sc_events); 159789b17223SAlexander Motin if (ep != NULL) { 159889b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_events, ep, e_next); 159989b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 160089b17223SAlexander Motin g_raid_handle_event(sc, ep); 160189b17223SAlexander Motin goto out; 160289b17223SAlexander Motin } 160389b17223SAlexander Motin bp = bioq_takefirst(&sc->sc_queue); 160489b17223SAlexander Motin if (bp != NULL) { 160589b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 160689b17223SAlexander Motin if (bp->bio_from == NULL || 160789b17223SAlexander Motin bp->bio_from->geom != sc->sc_geom) 160889b17223SAlexander Motin g_raid_start_request(bp); 160989b17223SAlexander Motin else 161089b17223SAlexander Motin g_raid_disk_done_request(bp); 161189b17223SAlexander Motin } 161289b17223SAlexander Motin out: 161389b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 161489b17223SAlexander Motin } 161589b17223SAlexander Motin 161689b17223SAlexander Motin static void 161789b17223SAlexander Motin g_raid_launch_provider(struct g_raid_volume *vol) 161889b17223SAlexander Motin { 161989b17223SAlexander Motin struct g_raid_disk *disk; 1620b43560abSAlexander Motin struct g_raid_subdisk *sd; 162189b17223SAlexander Motin struct g_raid_softc *sc; 162289b17223SAlexander Motin struct g_provider *pp; 162389b17223SAlexander Motin char name[G_RAID_MAX_VOLUMENAME]; 162489b17223SAlexander Motin off_t off; 1625b43560abSAlexander Motin int i; 162689b17223SAlexander Motin 162789b17223SAlexander Motin sc = vol->v_softc; 162889b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 162989b17223SAlexander Motin 163089b17223SAlexander Motin g_topology_lock(); 163189b17223SAlexander Motin /* Try to name provider with volume name. */ 163289b17223SAlexander Motin snprintf(name, sizeof(name), "raid/%s", vol->v_name); 163389b17223SAlexander Motin if (g_raid_name_format == 0 || vol->v_name[0] == 0 || 163489b17223SAlexander Motin g_provider_by_name(name) != NULL) { 163589b17223SAlexander Motin /* Otherwise use sequential volume number. */ 163689b17223SAlexander Motin snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); 163789b17223SAlexander Motin } 1638bd9fba0cSSean Bruno 163989b17223SAlexander Motin pp = g_new_providerf(sc->sc_geom, "%s", name); 164040ea77a0SAlexander Motin pp->flags |= G_PF_DIRECT_RECEIVE; 1641b43560abSAlexander Motin if (vol->v_tr->tro_class->trc_accept_unmapped) { 1642b43560abSAlexander Motin pp->flags |= G_PF_ACCEPT_UNMAPPED; 1643b43560abSAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 1644b43560abSAlexander Motin sd = &vol->v_subdisks[i]; 1645b43560abSAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_NONE) 1646b43560abSAlexander Motin continue; 1647b43560abSAlexander Motin if ((sd->sd_disk->d_consumer->provider->flags & 1648b43560abSAlexander Motin G_PF_ACCEPT_UNMAPPED) == 0) 1649b43560abSAlexander Motin pp->flags &= ~G_PF_ACCEPT_UNMAPPED; 1650b43560abSAlexander Motin } 1651b43560abSAlexander Motin } 165289b17223SAlexander Motin pp->private = vol; 165389b17223SAlexander Motin pp->mediasize = vol->v_mediasize; 165489b17223SAlexander Motin pp->sectorsize = vol->v_sectorsize; 165589b17223SAlexander Motin pp->stripesize = 0; 165689b17223SAlexander Motin pp->stripeoffset = 0; 165789b17223SAlexander Motin if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 165889b17223SAlexander Motin vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || 165989b17223SAlexander Motin vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || 166089b17223SAlexander Motin vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { 166189b17223SAlexander Motin if ((disk = vol->v_subdisks[0].sd_disk) != NULL && 166289b17223SAlexander Motin disk->d_consumer != NULL && 166389b17223SAlexander Motin disk->d_consumer->provider != NULL) { 166489b17223SAlexander Motin pp->stripesize = disk->d_consumer->provider->stripesize; 166589b17223SAlexander Motin off = disk->d_consumer->provider->stripeoffset; 166689b17223SAlexander Motin pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; 166789b17223SAlexander Motin if (off > 0) 166889b17223SAlexander Motin pp->stripeoffset %= off; 166989b17223SAlexander Motin } 167089b17223SAlexander Motin if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { 167189b17223SAlexander Motin pp->stripesize *= (vol->v_disks_count - 1); 167289b17223SAlexander Motin pp->stripeoffset *= (vol->v_disks_count - 1); 167389b17223SAlexander Motin } 167489b17223SAlexander Motin } else 167589b17223SAlexander Motin pp->stripesize = vol->v_strip_size; 167689b17223SAlexander Motin vol->v_provider = pp; 167789b17223SAlexander Motin g_error_provider(pp, 0); 167889b17223SAlexander Motin g_topology_unlock(); 167989b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", 168089b17223SAlexander Motin pp->name, vol->v_name); 168189b17223SAlexander Motin } 168289b17223SAlexander Motin 168389b17223SAlexander Motin static void 168489b17223SAlexander Motin g_raid_destroy_provider(struct g_raid_volume *vol) 168589b17223SAlexander Motin { 168689b17223SAlexander Motin struct g_raid_softc *sc; 168789b17223SAlexander Motin struct g_provider *pp; 168889b17223SAlexander Motin struct bio *bp, *tmp; 168989b17223SAlexander Motin 169089b17223SAlexander Motin g_topology_assert_not(); 169189b17223SAlexander Motin sc = vol->v_softc; 169289b17223SAlexander Motin pp = vol->v_provider; 169389b17223SAlexander Motin KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); 169489b17223SAlexander Motin 169589b17223SAlexander Motin g_topology_lock(); 169689b17223SAlexander Motin g_error_provider(pp, ENXIO); 169789b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 169889b17223SAlexander Motin TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { 169989b17223SAlexander Motin if (bp->bio_to != pp) 170089b17223SAlexander Motin continue; 170189b17223SAlexander Motin bioq_remove(&sc->sc_queue, bp); 170289b17223SAlexander Motin g_io_deliver(bp, ENXIO); 170389b17223SAlexander Motin } 170489b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 170589b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", 170689b17223SAlexander Motin pp->name, vol->v_name); 170789b17223SAlexander Motin g_wither_provider(pp, ENXIO); 170889b17223SAlexander Motin g_topology_unlock(); 170989b17223SAlexander Motin vol->v_provider = NULL; 171089b17223SAlexander Motin } 171189b17223SAlexander Motin 171289b17223SAlexander Motin /* 171389b17223SAlexander Motin * Update device state. 171489b17223SAlexander Motin */ 171589b17223SAlexander Motin static int 171689b17223SAlexander Motin g_raid_update_volume(struct g_raid_volume *vol, u_int event) 171789b17223SAlexander Motin { 171889b17223SAlexander Motin struct g_raid_softc *sc; 171989b17223SAlexander Motin 172089b17223SAlexander Motin sc = vol->v_softc; 172189b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 172289b17223SAlexander Motin 172389b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", 172489b17223SAlexander Motin g_raid_volume_event2str(event), 172589b17223SAlexander Motin vol->v_name); 172689b17223SAlexander Motin switch (event) { 172789b17223SAlexander Motin case G_RAID_VOLUME_E_DOWN: 172889b17223SAlexander Motin if (vol->v_provider != NULL) 172989b17223SAlexander Motin g_raid_destroy_provider(vol); 173089b17223SAlexander Motin break; 173189b17223SAlexander Motin case G_RAID_VOLUME_E_UP: 173289b17223SAlexander Motin if (vol->v_provider == NULL) 173389b17223SAlexander Motin g_raid_launch_provider(vol); 173489b17223SAlexander Motin break; 173589b17223SAlexander Motin case G_RAID_VOLUME_E_START: 173689b17223SAlexander Motin if (vol->v_tr) 173789b17223SAlexander Motin G_RAID_TR_START(vol->v_tr); 173889b17223SAlexander Motin return (0); 173989b17223SAlexander Motin default: 174089b17223SAlexander Motin if (sc->sc_md) 174189b17223SAlexander Motin G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); 174289b17223SAlexander Motin return (0); 174389b17223SAlexander Motin } 174489b17223SAlexander Motin 174589b17223SAlexander Motin /* Manage root mount release. */ 174689b17223SAlexander Motin if (vol->v_starting) { 174789b17223SAlexander Motin vol->v_starting = 0; 174889b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); 174989b17223SAlexander Motin root_mount_rel(vol->v_rootmount); 175089b17223SAlexander Motin vol->v_rootmount = NULL; 175189b17223SAlexander Motin } 175289b17223SAlexander Motin if (vol->v_stopping && vol->v_provider_open == 0) 175389b17223SAlexander Motin g_raid_destroy_volume(vol); 175489b17223SAlexander Motin return (0); 175589b17223SAlexander Motin } 175689b17223SAlexander Motin 175789b17223SAlexander Motin /* 175889b17223SAlexander Motin * Update subdisk state. 175989b17223SAlexander Motin */ 176089b17223SAlexander Motin static int 176189b17223SAlexander Motin g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) 176289b17223SAlexander Motin { 176389b17223SAlexander Motin struct g_raid_softc *sc; 176489b17223SAlexander Motin struct g_raid_volume *vol; 176589b17223SAlexander Motin 176689b17223SAlexander Motin sc = sd->sd_softc; 176789b17223SAlexander Motin vol = sd->sd_volume; 176889b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 176989b17223SAlexander Motin 177089b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", 177189b17223SAlexander Motin g_raid_subdisk_event2str(event), 177289b17223SAlexander Motin vol->v_name, sd->sd_pos, 177389b17223SAlexander Motin sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); 177489b17223SAlexander Motin if (vol->v_tr) 177589b17223SAlexander Motin G_RAID_TR_EVENT(vol->v_tr, sd, event); 177689b17223SAlexander Motin 177789b17223SAlexander Motin return (0); 177889b17223SAlexander Motin } 177989b17223SAlexander Motin 178089b17223SAlexander Motin /* 178189b17223SAlexander Motin * Update disk state. 178289b17223SAlexander Motin */ 178389b17223SAlexander Motin static int 178489b17223SAlexander Motin g_raid_update_disk(struct g_raid_disk *disk, u_int event) 178589b17223SAlexander Motin { 178689b17223SAlexander Motin struct g_raid_softc *sc; 178789b17223SAlexander Motin 178889b17223SAlexander Motin sc = disk->d_softc; 178989b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 179089b17223SAlexander Motin 179189b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", 179289b17223SAlexander Motin g_raid_disk_event2str(event), 179389b17223SAlexander Motin g_raid_get_diskname(disk)); 179489b17223SAlexander Motin 179589b17223SAlexander Motin if (sc->sc_md) 179689b17223SAlexander Motin G_RAID_MD_EVENT(sc->sc_md, disk, event); 179789b17223SAlexander Motin return (0); 179889b17223SAlexander Motin } 179989b17223SAlexander Motin 180089b17223SAlexander Motin /* 180189b17223SAlexander Motin * Node event. 180289b17223SAlexander Motin */ 180389b17223SAlexander Motin static int 180489b17223SAlexander Motin g_raid_update_node(struct g_raid_softc *sc, u_int event) 180589b17223SAlexander Motin { 180689b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 180789b17223SAlexander Motin 180889b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for the array.", 180989b17223SAlexander Motin g_raid_node_event2str(event)); 181089b17223SAlexander Motin 181189b17223SAlexander Motin if (event == G_RAID_NODE_E_WAKE) 181289b17223SAlexander Motin return (0); 181389b17223SAlexander Motin if (sc->sc_md) 181489b17223SAlexander Motin G_RAID_MD_EVENT(sc->sc_md, NULL, event); 181589b17223SAlexander Motin return (0); 181689b17223SAlexander Motin } 181789b17223SAlexander Motin 181889b17223SAlexander Motin static int 181989b17223SAlexander Motin g_raid_access(struct g_provider *pp, int acr, int acw, int ace) 182089b17223SAlexander Motin { 182189b17223SAlexander Motin struct g_raid_volume *vol; 182289b17223SAlexander Motin struct g_raid_softc *sc; 182314e2cd0aSAlexander Motin int dcw, opens, error = 0; 182489b17223SAlexander Motin 182589b17223SAlexander Motin g_topology_assert(); 182689b17223SAlexander Motin sc = pp->geom->softc; 182789b17223SAlexander Motin vol = pp->private; 182889b17223SAlexander Motin KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 182989b17223SAlexander Motin KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); 183089b17223SAlexander Motin 183189b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, 183289b17223SAlexander Motin acr, acw, ace); 183389b17223SAlexander Motin dcw = pp->acw + acw; 183489b17223SAlexander Motin 183589b17223SAlexander Motin g_topology_unlock(); 183689b17223SAlexander Motin sx_xlock(&sc->sc_lock); 183789b17223SAlexander Motin /* Deny new opens while dying. */ 183889b17223SAlexander Motin if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { 183989b17223SAlexander Motin error = ENXIO; 184089b17223SAlexander Motin goto out; 184189b17223SAlexander Motin } 18420f0b2fd8SAlexander Motin /* Deny write opens for read-only volumes. */ 18430f0b2fd8SAlexander Motin if (vol->v_read_only && acw > 0) { 18440f0b2fd8SAlexander Motin error = EROFS; 18450f0b2fd8SAlexander Motin goto out; 18460f0b2fd8SAlexander Motin } 1847a479c51bSAlexander Motin if (dcw == 0) 184889b17223SAlexander Motin g_raid_clean(vol, dcw); 184989b17223SAlexander Motin vol->v_provider_open += acr + acw + ace; 185089b17223SAlexander Motin /* Handle delayed node destruction. */ 185189b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && 185289b17223SAlexander Motin vol->v_provider_open == 0) { 185389b17223SAlexander Motin /* Count open volumes. */ 185489b17223SAlexander Motin opens = g_raid_nopens(sc); 185589b17223SAlexander Motin if (opens == 0) { 185689b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_HARD; 185789b17223SAlexander Motin /* Wake up worker to make it selfdestruct. */ 185889b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 185989b17223SAlexander Motin } 186089b17223SAlexander Motin } 186189b17223SAlexander Motin /* Handle open volume destruction. */ 186289b17223SAlexander Motin if (vol->v_stopping && vol->v_provider_open == 0) 186389b17223SAlexander Motin g_raid_destroy_volume(vol); 186489b17223SAlexander Motin out: 186589b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 186689b17223SAlexander Motin g_topology_lock(); 186789b17223SAlexander Motin return (error); 186889b17223SAlexander Motin } 186989b17223SAlexander Motin 187089b17223SAlexander Motin struct g_raid_softc * 187189b17223SAlexander Motin g_raid_create_node(struct g_class *mp, 187289b17223SAlexander Motin const char *name, struct g_raid_md_object *md) 187389b17223SAlexander Motin { 187489b17223SAlexander Motin struct g_raid_softc *sc; 187589b17223SAlexander Motin struct g_geom *gp; 187689b17223SAlexander Motin int error; 187789b17223SAlexander Motin 187889b17223SAlexander Motin g_topology_assert(); 187989b17223SAlexander Motin G_RAID_DEBUG(1, "Creating array %s.", name); 188089b17223SAlexander Motin 188189b17223SAlexander Motin gp = g_new_geomf(mp, "%s", name); 188289b17223SAlexander Motin sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); 188389b17223SAlexander Motin gp->start = g_raid_start; 188489b17223SAlexander Motin gp->orphan = g_raid_orphan; 188589b17223SAlexander Motin gp->access = g_raid_access; 188689b17223SAlexander Motin gp->dumpconf = g_raid_dumpconf; 188789b17223SAlexander Motin 188889b17223SAlexander Motin sc->sc_md = md; 188989b17223SAlexander Motin sc->sc_geom = gp; 189089b17223SAlexander Motin sc->sc_flags = 0; 189189b17223SAlexander Motin TAILQ_INIT(&sc->sc_volumes); 189289b17223SAlexander Motin TAILQ_INIT(&sc->sc_disks); 1893c9f545e5SAlexander Motin sx_init(&sc->sc_lock, "graid:lock"); 1894c9f545e5SAlexander Motin mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF); 189589b17223SAlexander Motin TAILQ_INIT(&sc->sc_events); 189689b17223SAlexander Motin bioq_init(&sc->sc_queue); 189789b17223SAlexander Motin gp->softc = sc; 189889b17223SAlexander Motin error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, 189989b17223SAlexander Motin "g_raid %s", name); 190089b17223SAlexander Motin if (error != 0) { 190189b17223SAlexander Motin G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); 190289b17223SAlexander Motin mtx_destroy(&sc->sc_queue_mtx); 190389b17223SAlexander Motin sx_destroy(&sc->sc_lock); 190489b17223SAlexander Motin g_destroy_geom(sc->sc_geom); 190589b17223SAlexander Motin free(sc, M_RAID); 190689b17223SAlexander Motin return (NULL); 190789b17223SAlexander Motin } 190889b17223SAlexander Motin 190989b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Array %s created.", name); 191089b17223SAlexander Motin return (sc); 191189b17223SAlexander Motin } 191289b17223SAlexander Motin 191389b17223SAlexander Motin struct g_raid_volume * 191489b17223SAlexander Motin g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) 191589b17223SAlexander Motin { 191689b17223SAlexander Motin struct g_raid_volume *vol, *vol1; 191789b17223SAlexander Motin int i; 191889b17223SAlexander Motin 191989b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); 192089b17223SAlexander Motin vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); 192189b17223SAlexander Motin vol->v_softc = sc; 192289b17223SAlexander Motin strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); 192389b17223SAlexander Motin vol->v_state = G_RAID_VOLUME_S_STARTING; 192489b17223SAlexander Motin vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 192589b17223SAlexander Motin vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; 19268f12ca2eSAlexander Motin vol->v_rotate_parity = 1; 192789b17223SAlexander Motin bioq_init(&vol->v_inflight); 192889b17223SAlexander Motin bioq_init(&vol->v_locked); 192989b17223SAlexander Motin LIST_INIT(&vol->v_locks); 193089b17223SAlexander Motin for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 193189b17223SAlexander Motin vol->v_subdisks[i].sd_softc = sc; 193289b17223SAlexander Motin vol->v_subdisks[i].sd_volume = vol; 193389b17223SAlexander Motin vol->v_subdisks[i].sd_pos = i; 193489b17223SAlexander Motin vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; 193589b17223SAlexander Motin } 193689b17223SAlexander Motin 193789b17223SAlexander Motin /* Find free ID for this volume. */ 193889b17223SAlexander Motin g_topology_lock(); 193989b17223SAlexander Motin vol1 = vol; 194089b17223SAlexander Motin if (id >= 0) { 194189b17223SAlexander Motin LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 194289b17223SAlexander Motin if (vol1->v_global_id == id) 194389b17223SAlexander Motin break; 194489b17223SAlexander Motin } 194589b17223SAlexander Motin } 194689b17223SAlexander Motin if (vol1 != NULL) { 194789b17223SAlexander Motin for (id = 0; ; id++) { 194889b17223SAlexander Motin LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 194989b17223SAlexander Motin if (vol1->v_global_id == id) 195089b17223SAlexander Motin break; 195189b17223SAlexander Motin } 195289b17223SAlexander Motin if (vol1 == NULL) 195389b17223SAlexander Motin break; 195489b17223SAlexander Motin } 195589b17223SAlexander Motin } 195689b17223SAlexander Motin vol->v_global_id = id; 195789b17223SAlexander Motin LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); 195889b17223SAlexander Motin g_topology_unlock(); 195989b17223SAlexander Motin 196089b17223SAlexander Motin /* Delay root mounting. */ 196189b17223SAlexander Motin vol->v_rootmount = root_mount_hold("GRAID"); 196289b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); 196389b17223SAlexander Motin vol->v_starting = 1; 196489b17223SAlexander Motin TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); 196589b17223SAlexander Motin return (vol); 196689b17223SAlexander Motin } 196789b17223SAlexander Motin 196889b17223SAlexander Motin struct g_raid_disk * 196989b17223SAlexander Motin g_raid_create_disk(struct g_raid_softc *sc) 197089b17223SAlexander Motin { 197189b17223SAlexander Motin struct g_raid_disk *disk; 197289b17223SAlexander Motin 197389b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Creating disk."); 197489b17223SAlexander Motin disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); 197589b17223SAlexander Motin disk->d_softc = sc; 197689b17223SAlexander Motin disk->d_state = G_RAID_DISK_S_NONE; 197789b17223SAlexander Motin TAILQ_INIT(&disk->d_subdisks); 197889b17223SAlexander Motin TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); 197989b17223SAlexander Motin return (disk); 198089b17223SAlexander Motin } 198189b17223SAlexander Motin 198289b17223SAlexander Motin int g_raid_start_volume(struct g_raid_volume *vol) 198389b17223SAlexander Motin { 198489b17223SAlexander Motin struct g_raid_tr_class *class; 198589b17223SAlexander Motin struct g_raid_tr_object *obj; 198689b17223SAlexander Motin int status; 198789b17223SAlexander Motin 198889b17223SAlexander Motin G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); 198989b17223SAlexander Motin LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { 1990c89d2fbeSAlexander Motin if (!class->trc_enable) 1991c89d2fbeSAlexander Motin continue; 199289b17223SAlexander Motin G_RAID_DEBUG1(2, vol->v_softc, 199389b17223SAlexander Motin "Tasting volume %s for %s transformation.", 199489b17223SAlexander Motin vol->v_name, class->name); 199589b17223SAlexander Motin obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 199689b17223SAlexander Motin M_WAITOK); 199789b17223SAlexander Motin obj->tro_class = class; 199889b17223SAlexander Motin obj->tro_volume = vol; 199989b17223SAlexander Motin status = G_RAID_TR_TASTE(obj, vol); 200089b17223SAlexander Motin if (status != G_RAID_TR_TASTE_FAIL) 200189b17223SAlexander Motin break; 200289b17223SAlexander Motin kobj_delete((kobj_t)obj, M_RAID); 200389b17223SAlexander Motin } 200489b17223SAlexander Motin if (class == NULL) { 200589b17223SAlexander Motin G_RAID_DEBUG1(0, vol->v_softc, 200689b17223SAlexander Motin "No transformation module found for %s.", 200789b17223SAlexander Motin vol->v_name); 200889b17223SAlexander Motin vol->v_tr = NULL; 200989b17223SAlexander Motin g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); 201089b17223SAlexander Motin g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, 201189b17223SAlexander Motin G_RAID_EVENT_VOLUME); 201289b17223SAlexander Motin return (-1); 201389b17223SAlexander Motin } 201489b17223SAlexander Motin G_RAID_DEBUG1(2, vol->v_softc, 201589b17223SAlexander Motin "Transformation module %s chosen for %s.", 201689b17223SAlexander Motin class->name, vol->v_name); 201789b17223SAlexander Motin vol->v_tr = obj; 201889b17223SAlexander Motin return (0); 201989b17223SAlexander Motin } 202089b17223SAlexander Motin 202189b17223SAlexander Motin int 202289b17223SAlexander Motin g_raid_destroy_node(struct g_raid_softc *sc, int worker) 202389b17223SAlexander Motin { 202489b17223SAlexander Motin struct g_raid_volume *vol, *tmpv; 202589b17223SAlexander Motin struct g_raid_disk *disk, *tmpd; 202689b17223SAlexander Motin int error = 0; 202789b17223SAlexander Motin 202889b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_HARD; 202989b17223SAlexander Motin TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { 203089b17223SAlexander Motin if (g_raid_destroy_volume(vol)) 203189b17223SAlexander Motin error = EBUSY; 203289b17223SAlexander Motin } 203389b17223SAlexander Motin if (error) 203489b17223SAlexander Motin return (error); 203589b17223SAlexander Motin TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { 203689b17223SAlexander Motin if (g_raid_destroy_disk(disk)) 203789b17223SAlexander Motin error = EBUSY; 203889b17223SAlexander Motin } 203989b17223SAlexander Motin if (error) 204089b17223SAlexander Motin return (error); 204189b17223SAlexander Motin if (sc->sc_md) { 204289b17223SAlexander Motin G_RAID_MD_FREE(sc->sc_md); 204389b17223SAlexander Motin kobj_delete((kobj_t)sc->sc_md, M_RAID); 204489b17223SAlexander Motin sc->sc_md = NULL; 204589b17223SAlexander Motin } 204689b17223SAlexander Motin if (sc->sc_geom != NULL) { 204789b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); 204889b17223SAlexander Motin g_topology_lock(); 204989b17223SAlexander Motin sc->sc_geom->softc = NULL; 205089b17223SAlexander Motin g_wither_geom(sc->sc_geom, ENXIO); 205189b17223SAlexander Motin g_topology_unlock(); 205289b17223SAlexander Motin sc->sc_geom = NULL; 205389b17223SAlexander Motin } else 205489b17223SAlexander Motin G_RAID_DEBUG(1, "Array destroyed."); 205589b17223SAlexander Motin if (worker) { 205689b17223SAlexander Motin g_raid_event_cancel(sc, sc); 205789b17223SAlexander Motin mtx_destroy(&sc->sc_queue_mtx); 205889b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 205989b17223SAlexander Motin sx_destroy(&sc->sc_lock); 206089b17223SAlexander Motin wakeup(&sc->sc_stopping); 206189b17223SAlexander Motin free(sc, M_RAID); 206289b17223SAlexander Motin curthread->td_pflags &= ~TDP_GEOM; 206389b17223SAlexander Motin G_RAID_DEBUG(1, "Thread exiting."); 206489b17223SAlexander Motin kproc_exit(0); 206589b17223SAlexander Motin } else { 206689b17223SAlexander Motin /* Wake up worker to make it selfdestruct. */ 206789b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 206889b17223SAlexander Motin } 206989b17223SAlexander Motin return (0); 207089b17223SAlexander Motin } 207189b17223SAlexander Motin 207289b17223SAlexander Motin int 207389b17223SAlexander Motin g_raid_destroy_volume(struct g_raid_volume *vol) 207489b17223SAlexander Motin { 207589b17223SAlexander Motin struct g_raid_softc *sc; 207689b17223SAlexander Motin struct g_raid_disk *disk; 207789b17223SAlexander Motin int i; 207889b17223SAlexander Motin 207989b17223SAlexander Motin sc = vol->v_softc; 208089b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); 208189b17223SAlexander Motin vol->v_stopping = 1; 208289b17223SAlexander Motin if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { 208389b17223SAlexander Motin if (vol->v_tr) { 208489b17223SAlexander Motin G_RAID_TR_STOP(vol->v_tr); 208589b17223SAlexander Motin return (EBUSY); 208689b17223SAlexander Motin } else 208789b17223SAlexander Motin vol->v_state = G_RAID_VOLUME_S_STOPPED; 208889b17223SAlexander Motin } 208989b17223SAlexander Motin if (g_raid_event_check(sc, vol) != 0) 209089b17223SAlexander Motin return (EBUSY); 209189b17223SAlexander Motin if (vol->v_provider != NULL) 209289b17223SAlexander Motin return (EBUSY); 209389b17223SAlexander Motin if (vol->v_provider_open != 0) 209489b17223SAlexander Motin return (EBUSY); 209589b17223SAlexander Motin if (vol->v_tr) { 209689b17223SAlexander Motin G_RAID_TR_FREE(vol->v_tr); 209789b17223SAlexander Motin kobj_delete((kobj_t)vol->v_tr, M_RAID); 209889b17223SAlexander Motin vol->v_tr = NULL; 209989b17223SAlexander Motin } 210089b17223SAlexander Motin if (vol->v_rootmount) 210189b17223SAlexander Motin root_mount_rel(vol->v_rootmount); 210289b17223SAlexander Motin g_topology_lock(); 210389b17223SAlexander Motin LIST_REMOVE(vol, v_global_next); 210489b17223SAlexander Motin g_topology_unlock(); 210589b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); 210689b17223SAlexander Motin for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 210789b17223SAlexander Motin g_raid_event_cancel(sc, &vol->v_subdisks[i]); 210889b17223SAlexander Motin disk = vol->v_subdisks[i].sd_disk; 210989b17223SAlexander Motin if (disk == NULL) 211089b17223SAlexander Motin continue; 211189b17223SAlexander Motin TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); 211289b17223SAlexander Motin } 211389b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); 211489b17223SAlexander Motin if (sc->sc_md) 211589b17223SAlexander Motin G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); 211689b17223SAlexander Motin g_raid_event_cancel(sc, vol); 211789b17223SAlexander Motin free(vol, M_RAID); 211889b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_HARD) { 211989b17223SAlexander Motin /* Wake up worker to let it selfdestruct. */ 212089b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 212189b17223SAlexander Motin } 212289b17223SAlexander Motin return (0); 212389b17223SAlexander Motin } 212489b17223SAlexander Motin 212589b17223SAlexander Motin int 212689b17223SAlexander Motin g_raid_destroy_disk(struct g_raid_disk *disk) 212789b17223SAlexander Motin { 212889b17223SAlexander Motin struct g_raid_softc *sc; 212989b17223SAlexander Motin struct g_raid_subdisk *sd, *tmp; 213089b17223SAlexander Motin 213189b17223SAlexander Motin sc = disk->d_softc; 213289b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Destroying disk."); 213389b17223SAlexander Motin if (disk->d_consumer) { 213489b17223SAlexander Motin g_raid_kill_consumer(sc, disk->d_consumer); 213589b17223SAlexander Motin disk->d_consumer = NULL; 213689b17223SAlexander Motin } 213789b17223SAlexander Motin TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { 213889b17223SAlexander Motin g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); 213989b17223SAlexander Motin g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 214089b17223SAlexander Motin G_RAID_EVENT_SUBDISK); 214189b17223SAlexander Motin TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); 214289b17223SAlexander Motin sd->sd_disk = NULL; 214389b17223SAlexander Motin } 214489b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_disks, disk, d_next); 214589b17223SAlexander Motin if (sc->sc_md) 214689b17223SAlexander Motin G_RAID_MD_FREE_DISK(sc->sc_md, disk); 214789b17223SAlexander Motin g_raid_event_cancel(sc, disk); 214889b17223SAlexander Motin free(disk, M_RAID); 214989b17223SAlexander Motin return (0); 215089b17223SAlexander Motin } 215189b17223SAlexander Motin 215289b17223SAlexander Motin int 215389b17223SAlexander Motin g_raid_destroy(struct g_raid_softc *sc, int how) 215489b17223SAlexander Motin { 21558531bb3fSAlexander Motin int error, opens; 215689b17223SAlexander Motin 215789b17223SAlexander Motin g_topology_assert_not(); 215889b17223SAlexander Motin if (sc == NULL) 215989b17223SAlexander Motin return (ENXIO); 216089b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 216189b17223SAlexander Motin 216289b17223SAlexander Motin /* Count open volumes. */ 216389b17223SAlexander Motin opens = g_raid_nopens(sc); 216489b17223SAlexander Motin 216589b17223SAlexander Motin /* React on some opened volumes. */ 216689b17223SAlexander Motin if (opens > 0) { 216789b17223SAlexander Motin switch (how) { 216889b17223SAlexander Motin case G_RAID_DESTROY_SOFT: 216989b17223SAlexander Motin G_RAID_DEBUG1(1, sc, 217089b17223SAlexander Motin "%d volumes are still open.", 217189b17223SAlexander Motin opens); 21728531bb3fSAlexander Motin sx_xunlock(&sc->sc_lock); 217389b17223SAlexander Motin return (EBUSY); 217489b17223SAlexander Motin case G_RAID_DESTROY_DELAYED: 217589b17223SAlexander Motin G_RAID_DEBUG1(1, sc, 217689b17223SAlexander Motin "Array will be destroyed on last close."); 217789b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_DELAYED; 21788531bb3fSAlexander Motin sx_xunlock(&sc->sc_lock); 217989b17223SAlexander Motin return (EBUSY); 218089b17223SAlexander Motin case G_RAID_DESTROY_HARD: 218189b17223SAlexander Motin G_RAID_DEBUG1(1, sc, 218289b17223SAlexander Motin "%d volumes are still open.", 218389b17223SAlexander Motin opens); 218489b17223SAlexander Motin } 218589b17223SAlexander Motin } 218689b17223SAlexander Motin 218789b17223SAlexander Motin /* Mark node for destruction. */ 218889b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_HARD; 218989b17223SAlexander Motin /* Wake up worker to let it selfdestruct. */ 219089b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 219189b17223SAlexander Motin /* Sleep until node destroyed. */ 21928531bb3fSAlexander Motin error = sx_sleep(&sc->sc_stopping, &sc->sc_lock, 21938531bb3fSAlexander Motin PRIBIO | PDROP, "r:destroy", hz * 3); 21948531bb3fSAlexander Motin return (error == EWOULDBLOCK ? EBUSY : 0); 219589b17223SAlexander Motin } 219689b17223SAlexander Motin 219789b17223SAlexander Motin static void 219889b17223SAlexander Motin g_raid_taste_orphan(struct g_consumer *cp) 219989b17223SAlexander Motin { 220089b17223SAlexander Motin 220189b17223SAlexander Motin KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 220289b17223SAlexander Motin cp->provider->name)); 220389b17223SAlexander Motin } 220489b17223SAlexander Motin 220589b17223SAlexander Motin static struct g_geom * 220689b17223SAlexander Motin g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 220789b17223SAlexander Motin { 220889b17223SAlexander Motin struct g_consumer *cp; 220989b17223SAlexander Motin struct g_geom *gp, *geom; 221089b17223SAlexander Motin struct g_raid_md_class *class; 221189b17223SAlexander Motin struct g_raid_md_object *obj; 221289b17223SAlexander Motin int status; 221389b17223SAlexander Motin 221489b17223SAlexander Motin g_topology_assert(); 221589b17223SAlexander Motin g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 2216c89d2fbeSAlexander Motin if (!g_raid_enable) 2217c89d2fbeSAlexander Motin return (NULL); 221889b17223SAlexander Motin G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); 221989b17223SAlexander Motin 2220dea1e226SAlexander Motin geom = NULL; 2221dea1e226SAlexander Motin status = G_RAID_MD_TASTE_FAIL; 22228df8e26aSAlexander Motin gp = g_new_geomf(mp, "raid:taste"); 222389b17223SAlexander Motin /* 222489b17223SAlexander Motin * This orphan function should be never called. 222589b17223SAlexander Motin */ 222689b17223SAlexander Motin gp->orphan = g_raid_taste_orphan; 222789b17223SAlexander Motin cp = g_new_consumer(gp); 222840ea77a0SAlexander Motin cp->flags |= G_CF_DIRECT_RECEIVE; 222989b17223SAlexander Motin g_attach(cp, pp); 2230dea1e226SAlexander Motin if (g_access(cp, 1, 0, 0) != 0) 2231dea1e226SAlexander Motin goto ofail; 223289b17223SAlexander Motin 223389b17223SAlexander Motin LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 2234c89d2fbeSAlexander Motin if (!class->mdc_enable) 2235c89d2fbeSAlexander Motin continue; 223689b17223SAlexander Motin G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", 223789b17223SAlexander Motin pp->name, class->name); 223889b17223SAlexander Motin obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 223989b17223SAlexander Motin M_WAITOK); 224089b17223SAlexander Motin obj->mdo_class = class; 224189b17223SAlexander Motin status = G_RAID_MD_TASTE(obj, mp, cp, &geom); 224289b17223SAlexander Motin if (status != G_RAID_MD_TASTE_NEW) 224389b17223SAlexander Motin kobj_delete((kobj_t)obj, M_RAID); 224489b17223SAlexander Motin if (status != G_RAID_MD_TASTE_FAIL) 224589b17223SAlexander Motin break; 224689b17223SAlexander Motin } 224789b17223SAlexander Motin 2248dea1e226SAlexander Motin if (status == G_RAID_MD_TASTE_FAIL) 2249dea1e226SAlexander Motin (void)g_access(cp, -1, 0, 0); 2250dea1e226SAlexander Motin ofail: 225189b17223SAlexander Motin g_detach(cp); 225289b17223SAlexander Motin g_destroy_consumer(cp); 225389b17223SAlexander Motin g_destroy_geom(gp); 225489b17223SAlexander Motin G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); 225589b17223SAlexander Motin return (geom); 225689b17223SAlexander Motin } 225789b17223SAlexander Motin 225889b17223SAlexander Motin int 22598df8e26aSAlexander Motin g_raid_create_node_format(const char *format, struct gctl_req *req, 22608df8e26aSAlexander Motin struct g_geom **gp) 226189b17223SAlexander Motin { 226289b17223SAlexander Motin struct g_raid_md_class *class; 226389b17223SAlexander Motin struct g_raid_md_object *obj; 226489b17223SAlexander Motin int status; 226589b17223SAlexander Motin 226689b17223SAlexander Motin G_RAID_DEBUG(2, "Creating array for %s metadata.", format); 226789b17223SAlexander Motin LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 226889b17223SAlexander Motin if (strcasecmp(class->name, format) == 0) 226989b17223SAlexander Motin break; 227089b17223SAlexander Motin } 227189b17223SAlexander Motin if (class == NULL) { 227289b17223SAlexander Motin G_RAID_DEBUG(1, "No support for %s metadata.", format); 227389b17223SAlexander Motin return (G_RAID_MD_TASTE_FAIL); 227489b17223SAlexander Motin } 227589b17223SAlexander Motin obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 227689b17223SAlexander Motin M_WAITOK); 227789b17223SAlexander Motin obj->mdo_class = class; 22788df8e26aSAlexander Motin status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp); 227989b17223SAlexander Motin if (status != G_RAID_MD_TASTE_NEW) 228089b17223SAlexander Motin kobj_delete((kobj_t)obj, M_RAID); 228189b17223SAlexander Motin return (status); 228289b17223SAlexander Motin } 228389b17223SAlexander Motin 228489b17223SAlexander Motin static int 228589b17223SAlexander Motin g_raid_destroy_geom(struct gctl_req *req __unused, 228689b17223SAlexander Motin struct g_class *mp __unused, struct g_geom *gp) 228789b17223SAlexander Motin { 228889b17223SAlexander Motin struct g_raid_softc *sc; 228989b17223SAlexander Motin int error; 229089b17223SAlexander Motin 229189b17223SAlexander Motin g_topology_unlock(); 229289b17223SAlexander Motin sc = gp->softc; 229389b17223SAlexander Motin sx_xlock(&sc->sc_lock); 229489b17223SAlexander Motin g_cancel_event(sc); 229589b17223SAlexander Motin error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); 229689b17223SAlexander Motin g_topology_lock(); 229789b17223SAlexander Motin return (error); 229889b17223SAlexander Motin } 229989b17223SAlexander Motin 230089b17223SAlexander Motin void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, 230189b17223SAlexander Motin struct g_raid_subdisk *sd, struct g_raid_disk *disk) 230289b17223SAlexander Motin { 230389b17223SAlexander Motin 230489b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_HARD) 230589b17223SAlexander Motin return; 230689b17223SAlexander Motin if (sc->sc_md) 230789b17223SAlexander Motin G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); 230889b17223SAlexander Motin } 230989b17223SAlexander Motin 231089b17223SAlexander Motin void g_raid_fail_disk(struct g_raid_softc *sc, 231189b17223SAlexander Motin struct g_raid_subdisk *sd, struct g_raid_disk *disk) 231289b17223SAlexander Motin { 231389b17223SAlexander Motin 231489b17223SAlexander Motin if (disk == NULL) 231589b17223SAlexander Motin disk = sd->sd_disk; 231689b17223SAlexander Motin if (disk == NULL) { 231789b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); 231889b17223SAlexander Motin return; 231989b17223SAlexander Motin } 232089b17223SAlexander Motin if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 232189b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " 232289b17223SAlexander Motin "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 232389b17223SAlexander Motin return; 232489b17223SAlexander Motin } 232589b17223SAlexander Motin if (sc->sc_md) 232689b17223SAlexander Motin G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); 232789b17223SAlexander Motin } 232889b17223SAlexander Motin 232989b17223SAlexander Motin static void 233089b17223SAlexander Motin g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 233189b17223SAlexander Motin struct g_consumer *cp, struct g_provider *pp) 233289b17223SAlexander Motin { 233389b17223SAlexander Motin struct g_raid_softc *sc; 233489b17223SAlexander Motin struct g_raid_volume *vol; 233589b17223SAlexander Motin struct g_raid_subdisk *sd; 233689b17223SAlexander Motin struct g_raid_disk *disk; 233789b17223SAlexander Motin int i, s; 233889b17223SAlexander Motin 233989b17223SAlexander Motin g_topology_assert(); 234089b17223SAlexander Motin 234189b17223SAlexander Motin sc = gp->softc; 234289b17223SAlexander Motin if (sc == NULL) 234389b17223SAlexander Motin return; 234489b17223SAlexander Motin if (pp != NULL) { 234589b17223SAlexander Motin vol = pp->private; 234689b17223SAlexander Motin g_topology_unlock(); 234789b17223SAlexander Motin sx_xlock(&sc->sc_lock); 2348bcb6ad36SAlexander Motin sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent, 2349bcb6ad36SAlexander Motin sc->sc_md->mdo_class->name, 2350bcb6ad36SAlexander Motin g_raid_volume_level2str(vol->v_raid_level, 2351bcb6ad36SAlexander Motin vol->v_raid_level_qualifier)); 235289b17223SAlexander Motin sbuf_printf(sb, "%s<Label>%s</Label>\n", indent, 235389b17223SAlexander Motin vol->v_name); 235489b17223SAlexander Motin sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent, 235589b17223SAlexander Motin g_raid_volume_level2str(vol->v_raid_level, 235689b17223SAlexander Motin vol->v_raid_level_qualifier)); 235789b17223SAlexander Motin sbuf_printf(sb, 235889b17223SAlexander Motin "%s<Transformation>%s</Transformation>\n", indent, 235989b17223SAlexander Motin vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); 236089b17223SAlexander Motin sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 236189b17223SAlexander Motin vol->v_disks_count); 236289b17223SAlexander Motin sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent, 236389b17223SAlexander Motin vol->v_strip_size); 236489b17223SAlexander Motin sbuf_printf(sb, "%s<State>%s</State>\n", indent, 236589b17223SAlexander Motin g_raid_volume_state2str(vol->v_state)); 236689b17223SAlexander Motin sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent, 236789b17223SAlexander Motin vol->v_dirty ? "Yes" : "No"); 236889b17223SAlexander Motin sbuf_printf(sb, "%s<Subdisks>", indent); 236989b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 237089b17223SAlexander Motin sd = &vol->v_subdisks[i]; 237189b17223SAlexander Motin if (sd->sd_disk != NULL && 237289b17223SAlexander Motin sd->sd_disk->d_consumer != NULL) { 237389b17223SAlexander Motin sbuf_printf(sb, "%s ", 237489b17223SAlexander Motin g_raid_get_diskname(sd->sd_disk)); 237589b17223SAlexander Motin } else { 237649ee0fceSAlexander Motin sbuf_cat(sb, "NONE "); 237789b17223SAlexander Motin } 237889b17223SAlexander Motin sbuf_printf(sb, "(%s", 237989b17223SAlexander Motin g_raid_subdisk_state2str(sd->sd_state)); 238089b17223SAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 238189b17223SAlexander Motin sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 238289b17223SAlexander Motin sbuf_printf(sb, " %d%%", 238389b17223SAlexander Motin (int)(sd->sd_rebuild_pos * 100 / 238489b17223SAlexander Motin sd->sd_size)); 238589b17223SAlexander Motin } 238649ee0fceSAlexander Motin sbuf_cat(sb, ")"); 238789b17223SAlexander Motin if (i + 1 < vol->v_disks_count) 238849ee0fceSAlexander Motin sbuf_cat(sb, ", "); 238989b17223SAlexander Motin } 239049ee0fceSAlexander Motin sbuf_cat(sb, "</Subdisks>\n"); 239189b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 239289b17223SAlexander Motin g_topology_lock(); 239389b17223SAlexander Motin } else if (cp != NULL) { 239489b17223SAlexander Motin disk = cp->private; 239589b17223SAlexander Motin if (disk == NULL) 239689b17223SAlexander Motin return; 239789b17223SAlexander Motin g_topology_unlock(); 239889b17223SAlexander Motin sx_xlock(&sc->sc_lock); 239989b17223SAlexander Motin sbuf_printf(sb, "%s<State>%s", indent, 240089b17223SAlexander Motin g_raid_disk_state2str(disk->d_state)); 240189b17223SAlexander Motin if (!TAILQ_EMPTY(&disk->d_subdisks)) { 240249ee0fceSAlexander Motin sbuf_cat(sb, " ("); 240389b17223SAlexander Motin TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 240489b17223SAlexander Motin sbuf_printf(sb, "%s", 240589b17223SAlexander Motin g_raid_subdisk_state2str(sd->sd_state)); 240689b17223SAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 240789b17223SAlexander Motin sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 240889b17223SAlexander Motin sbuf_printf(sb, " %d%%", 240989b17223SAlexander Motin (int)(sd->sd_rebuild_pos * 100 / 241089b17223SAlexander Motin sd->sd_size)); 241189b17223SAlexander Motin } 241289b17223SAlexander Motin if (TAILQ_NEXT(sd, sd_next)) 241349ee0fceSAlexander Motin sbuf_cat(sb, ", "); 241489b17223SAlexander Motin } 241549ee0fceSAlexander Motin sbuf_cat(sb, ")"); 241689b17223SAlexander Motin } 241749ee0fceSAlexander Motin sbuf_cat(sb, "</State>\n"); 241889b17223SAlexander Motin sbuf_printf(sb, "%s<Subdisks>", indent); 241989b17223SAlexander Motin TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 242089b17223SAlexander Motin sbuf_printf(sb, "r%d(%s):%d@%ju", 242189b17223SAlexander Motin sd->sd_volume->v_global_id, 242289b17223SAlexander Motin sd->sd_volume->v_name, 24236d305ab0SEugene Grosbein sd->sd_pos, (uintmax_t)sd->sd_offset); 242489b17223SAlexander Motin if (TAILQ_NEXT(sd, sd_next)) 242549ee0fceSAlexander Motin sbuf_cat(sb, ", "); 242689b17223SAlexander Motin } 242749ee0fceSAlexander Motin sbuf_cat(sb, "</Subdisks>\n"); 242889b17223SAlexander Motin sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent, 242989b17223SAlexander Motin disk->d_read_errs); 243089b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 243189b17223SAlexander Motin g_topology_lock(); 243289b17223SAlexander Motin } else { 243389b17223SAlexander Motin g_topology_unlock(); 243489b17223SAlexander Motin sx_xlock(&sc->sc_lock); 243589b17223SAlexander Motin if (sc->sc_md) { 243689b17223SAlexander Motin sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent, 243789b17223SAlexander Motin sc->sc_md->mdo_class->name); 243889b17223SAlexander Motin } 243989b17223SAlexander Motin if (!TAILQ_EMPTY(&sc->sc_volumes)) { 244089b17223SAlexander Motin s = 0xff; 244189b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 244289b17223SAlexander Motin if (vol->v_state < s) 244389b17223SAlexander Motin s = vol->v_state; 244489b17223SAlexander Motin } 244589b17223SAlexander Motin sbuf_printf(sb, "%s<State>%s</State>\n", indent, 244689b17223SAlexander Motin g_raid_volume_state2str(s)); 244789b17223SAlexander Motin } 244889b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 244989b17223SAlexander Motin g_topology_lock(); 245089b17223SAlexander Motin } 245189b17223SAlexander Motin } 245289b17223SAlexander Motin 245389b17223SAlexander Motin static void 2454a479c51bSAlexander Motin g_raid_shutdown_post_sync(void *arg, int howto) 245589b17223SAlexander Motin { 245689b17223SAlexander Motin struct g_class *mp; 245789b17223SAlexander Motin struct g_geom *gp, *gp2; 245889b17223SAlexander Motin struct g_raid_softc *sc; 2459a479c51bSAlexander Motin struct g_raid_volume *vol; 246089b17223SAlexander Motin 246189b17223SAlexander Motin mp = arg; 246289b17223SAlexander Motin g_topology_lock(); 2463a479c51bSAlexander Motin g_raid_shutdown = 1; 246489b17223SAlexander Motin LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 246589b17223SAlexander Motin if ((sc = gp->softc) == NULL) 246689b17223SAlexander Motin continue; 246789b17223SAlexander Motin g_topology_unlock(); 246889b17223SAlexander Motin sx_xlock(&sc->sc_lock); 2469a479c51bSAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) 2470a479c51bSAlexander Motin g_raid_clean(vol, -1); 247189b17223SAlexander Motin g_cancel_event(sc); 24728531bb3fSAlexander Motin g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); 247389b17223SAlexander Motin g_topology_lock(); 247489b17223SAlexander Motin } 247589b17223SAlexander Motin g_topology_unlock(); 247689b17223SAlexander Motin } 247789b17223SAlexander Motin 247889b17223SAlexander Motin static void 247989b17223SAlexander Motin g_raid_init(struct g_class *mp) 248089b17223SAlexander Motin { 248189b17223SAlexander Motin 2482a479c51bSAlexander Motin g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync, 2483a479c51bSAlexander Motin g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST); 2484a479c51bSAlexander Motin if (g_raid_post_sync == NULL) 248589b17223SAlexander Motin G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); 248689b17223SAlexander Motin g_raid_started = 1; 248789b17223SAlexander Motin } 248889b17223SAlexander Motin 248989b17223SAlexander Motin static void 249089b17223SAlexander Motin g_raid_fini(struct g_class *mp) 249189b17223SAlexander Motin { 249289b17223SAlexander Motin 2493a479c51bSAlexander Motin if (g_raid_post_sync != NULL) 2494a479c51bSAlexander Motin EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync); 249589b17223SAlexander Motin g_raid_started = 0; 249689b17223SAlexander Motin } 249789b17223SAlexander Motin 249889b17223SAlexander Motin int 249989b17223SAlexander Motin g_raid_md_modevent(module_t mod, int type, void *arg) 250089b17223SAlexander Motin { 250189b17223SAlexander Motin struct g_raid_md_class *class, *c, *nc; 250289b17223SAlexander Motin int error; 250389b17223SAlexander Motin 250489b17223SAlexander Motin error = 0; 250589b17223SAlexander Motin class = arg; 250689b17223SAlexander Motin switch (type) { 250789b17223SAlexander Motin case MOD_LOAD: 250889b17223SAlexander Motin c = LIST_FIRST(&g_raid_md_classes); 250989b17223SAlexander Motin if (c == NULL || c->mdc_priority > class->mdc_priority) 251089b17223SAlexander Motin LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); 251189b17223SAlexander Motin else { 251289b17223SAlexander Motin while ((nc = LIST_NEXT(c, mdc_list)) != NULL && 251389b17223SAlexander Motin nc->mdc_priority < class->mdc_priority) 251489b17223SAlexander Motin c = nc; 251589b17223SAlexander Motin LIST_INSERT_AFTER(c, class, mdc_list); 251689b17223SAlexander Motin } 251789b17223SAlexander Motin if (g_raid_started) 251889b17223SAlexander Motin g_retaste(&g_raid_class); 251989b17223SAlexander Motin break; 252089b17223SAlexander Motin case MOD_UNLOAD: 252189b17223SAlexander Motin LIST_REMOVE(class, mdc_list); 252289b17223SAlexander Motin break; 252389b17223SAlexander Motin default: 252489b17223SAlexander Motin error = EOPNOTSUPP; 252589b17223SAlexander Motin break; 252689b17223SAlexander Motin } 252789b17223SAlexander Motin 252889b17223SAlexander Motin return (error); 252989b17223SAlexander Motin } 253089b17223SAlexander Motin 253189b17223SAlexander Motin int 253289b17223SAlexander Motin g_raid_tr_modevent(module_t mod, int type, void *arg) 253389b17223SAlexander Motin { 253489b17223SAlexander Motin struct g_raid_tr_class *class, *c, *nc; 253589b17223SAlexander Motin int error; 253689b17223SAlexander Motin 253789b17223SAlexander Motin error = 0; 253889b17223SAlexander Motin class = arg; 253989b17223SAlexander Motin switch (type) { 254089b17223SAlexander Motin case MOD_LOAD: 254189b17223SAlexander Motin c = LIST_FIRST(&g_raid_tr_classes); 254289b17223SAlexander Motin if (c == NULL || c->trc_priority > class->trc_priority) 254389b17223SAlexander Motin LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); 254489b17223SAlexander Motin else { 254589b17223SAlexander Motin while ((nc = LIST_NEXT(c, trc_list)) != NULL && 254689b17223SAlexander Motin nc->trc_priority < class->trc_priority) 254789b17223SAlexander Motin c = nc; 254889b17223SAlexander Motin LIST_INSERT_AFTER(c, class, trc_list); 254989b17223SAlexander Motin } 255089b17223SAlexander Motin break; 255189b17223SAlexander Motin case MOD_UNLOAD: 255289b17223SAlexander Motin LIST_REMOVE(class, trc_list); 255389b17223SAlexander Motin break; 255489b17223SAlexander Motin default: 255589b17223SAlexander Motin error = EOPNOTSUPP; 255689b17223SAlexander Motin break; 255789b17223SAlexander Motin } 255889b17223SAlexander Motin 255989b17223SAlexander Motin return (error); 256089b17223SAlexander Motin } 256189b17223SAlexander Motin 256289b17223SAlexander Motin /* 256389b17223SAlexander Motin * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) 256489b17223SAlexander Motin * to reduce module priority, allowing submodules to register them first. 256589b17223SAlexander Motin */ 256689b17223SAlexander Motin static moduledata_t g_raid_mod = { 256789b17223SAlexander Motin "g_raid", 256889b17223SAlexander Motin g_modevent, 256989b17223SAlexander Motin &g_raid_class 257089b17223SAlexander Motin }; 257189b17223SAlexander Motin DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD); 257289b17223SAlexander Motin MODULE_VERSION(geom_raid, 0); 2573