189b17223SAlexander Motin /*- 289b17223SAlexander Motin * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org> 389b17223SAlexander Motin * All rights reserved. 489b17223SAlexander Motin * 589b17223SAlexander Motin * Redistribution and use in source and binary forms, with or without 689b17223SAlexander Motin * modification, are permitted provided that the following conditions 789b17223SAlexander Motin * are met: 889b17223SAlexander Motin * 1. Redistributions of source code must retain the above copyright 989b17223SAlexander Motin * notice, this list of conditions and the following disclaimer. 1089b17223SAlexander Motin * 2. Redistributions in binary form must reproduce the above copyright 1189b17223SAlexander Motin * notice, this list of conditions and the following disclaimer in the 1289b17223SAlexander Motin * documentation and/or other materials provided with the distribution. 1389b17223SAlexander Motin * 1489b17223SAlexander Motin * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND 1589b17223SAlexander Motin * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1689b17223SAlexander Motin * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1789b17223SAlexander Motin * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE 1889b17223SAlexander Motin * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 1989b17223SAlexander Motin * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2089b17223SAlexander Motin * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2189b17223SAlexander Motin * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2289b17223SAlexander Motin * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2389b17223SAlexander Motin * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2489b17223SAlexander Motin * SUCH DAMAGE. 2589b17223SAlexander Motin */ 2689b17223SAlexander Motin 2789b17223SAlexander Motin #include <sys/cdefs.h> 2889b17223SAlexander Motin __FBSDID("$FreeBSD$"); 2989b17223SAlexander Motin 3089b17223SAlexander Motin #include <sys/param.h> 3189b17223SAlexander Motin #include <sys/systm.h> 3289b17223SAlexander Motin #include <sys/kernel.h> 3389b17223SAlexander Motin #include <sys/module.h> 3489b17223SAlexander Motin #include <sys/limits.h> 3589b17223SAlexander Motin #include <sys/lock.h> 3689b17223SAlexander Motin #include <sys/mutex.h> 3789b17223SAlexander Motin #include <sys/bio.h> 385d807a0eSAndrey V. Elsukov #include <sys/sbuf.h> 3989b17223SAlexander Motin #include <sys/sysctl.h> 4089b17223SAlexander Motin #include <sys/malloc.h> 4189b17223SAlexander Motin #include <sys/eventhandler.h> 4289b17223SAlexander Motin #include <vm/uma.h> 4389b17223SAlexander Motin #include <geom/geom.h> 4489b17223SAlexander Motin #include <sys/proc.h> 4589b17223SAlexander Motin #include <sys/kthread.h> 4689b17223SAlexander Motin #include <sys/sched.h> 4789b17223SAlexander Motin #include <geom/raid/g_raid.h> 4889b17223SAlexander Motin #include "g_raid_md_if.h" 4989b17223SAlexander Motin #include "g_raid_tr_if.h" 5089b17223SAlexander Motin 5189b17223SAlexander Motin static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data"); 5289b17223SAlexander Motin 5389b17223SAlexander Motin SYSCTL_DECL(_kern_geom); 5489b17223SAlexander Motin SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff"); 5589b17223SAlexander Motin u_int g_raid_aggressive_spare = 0; 5689b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare); 5789b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW, 5889b17223SAlexander Motin &g_raid_aggressive_spare, 0, "Use disks without metadata as spare"); 59fe51d6c1SAlexander Motin u_int g_raid_debug = 0; 6089b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.debug", &g_raid_debug); 6189b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0, 6289b17223SAlexander Motin "Debug level"); 6389b17223SAlexander Motin int g_raid_read_err_thresh = 10; 6489b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh); 6589b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW, 6689b17223SAlexander Motin &g_raid_read_err_thresh, 0, 6789b17223SAlexander Motin "Number of read errors equated to disk failure"); 6889b17223SAlexander Motin u_int g_raid_start_timeout = 30; 6989b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout); 7089b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW, 7189b17223SAlexander Motin &g_raid_start_timeout, 0, 7289b17223SAlexander Motin "Time to wait for all array components"); 7389b17223SAlexander Motin static u_int g_raid_clean_time = 5; 7489b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time); 7589b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW, 7689b17223SAlexander Motin &g_raid_clean_time, 0, "Mark volume as clean when idling"); 7789b17223SAlexander Motin static u_int g_raid_disconnect_on_failure = 1; 7889b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.disconnect_on_failure", 7989b17223SAlexander Motin &g_raid_disconnect_on_failure); 8089b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, 8189b17223SAlexander Motin &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 8289b17223SAlexander Motin static u_int g_raid_name_format = 0; 8389b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format); 8489b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW, 8589b17223SAlexander Motin &g_raid_name_format, 0, "Providers name format."); 8689b17223SAlexander Motin static u_int g_raid_idle_threshold = 1000000; 8789b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold); 8889b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW, 8989b17223SAlexander Motin &g_raid_idle_threshold, 1000000, 9089b17223SAlexander Motin "Time in microseconds to consider a volume idle."); 9189b17223SAlexander Motin 9289b17223SAlexander Motin #define MSLEEP(rv, ident, mtx, priority, wmesg, timeout) do { \ 9389b17223SAlexander Motin G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 9489b17223SAlexander Motin rv = msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 9589b17223SAlexander Motin G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 9689b17223SAlexander Motin } while (0) 9789b17223SAlexander Motin 9889b17223SAlexander Motin LIST_HEAD(, g_raid_md_class) g_raid_md_classes = 9989b17223SAlexander Motin LIST_HEAD_INITIALIZER(g_raid_md_classes); 10089b17223SAlexander Motin 10189b17223SAlexander Motin LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes = 10289b17223SAlexander Motin LIST_HEAD_INITIALIZER(g_raid_tr_classes); 10389b17223SAlexander Motin 10489b17223SAlexander Motin LIST_HEAD(, g_raid_volume) g_raid_volumes = 10589b17223SAlexander Motin LIST_HEAD_INITIALIZER(g_raid_volumes); 10689b17223SAlexander Motin 10789b17223SAlexander Motin static eventhandler_tag g_raid_pre_sync = NULL; 10889b17223SAlexander Motin static int g_raid_started = 0; 10989b17223SAlexander Motin 11089b17223SAlexander Motin static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp, 11189b17223SAlexander Motin struct g_geom *gp); 11289b17223SAlexander Motin static g_taste_t g_raid_taste; 11389b17223SAlexander Motin static void g_raid_init(struct g_class *mp); 11489b17223SAlexander Motin static void g_raid_fini(struct g_class *mp); 11589b17223SAlexander Motin 11689b17223SAlexander Motin struct g_class g_raid_class = { 11789b17223SAlexander Motin .name = G_RAID_CLASS_NAME, 11889b17223SAlexander Motin .version = G_VERSION, 11989b17223SAlexander Motin .ctlreq = g_raid_ctl, 12089b17223SAlexander Motin .taste = g_raid_taste, 12189b17223SAlexander Motin .destroy_geom = g_raid_destroy_geom, 12289b17223SAlexander Motin .init = g_raid_init, 12389b17223SAlexander Motin .fini = g_raid_fini 12489b17223SAlexander Motin }; 12589b17223SAlexander Motin 12689b17223SAlexander Motin static void g_raid_destroy_provider(struct g_raid_volume *vol); 12789b17223SAlexander Motin static int g_raid_update_disk(struct g_raid_disk *disk, u_int event); 12889b17223SAlexander Motin static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event); 12989b17223SAlexander Motin static int g_raid_update_volume(struct g_raid_volume *vol, u_int event); 13089b17223SAlexander Motin static int g_raid_update_node(struct g_raid_softc *sc, u_int event); 13189b17223SAlexander Motin static void g_raid_dumpconf(struct sbuf *sb, const char *indent, 13289b17223SAlexander Motin struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 13389b17223SAlexander Motin static void g_raid_start(struct bio *bp); 13489b17223SAlexander Motin static void g_raid_start_request(struct bio *bp); 13589b17223SAlexander Motin static void g_raid_disk_done(struct bio *bp); 13689b17223SAlexander Motin static void g_raid_poll(struct g_raid_softc *sc); 13789b17223SAlexander Motin 13889b17223SAlexander Motin static const char * 13989b17223SAlexander Motin g_raid_node_event2str(int event) 14089b17223SAlexander Motin { 14189b17223SAlexander Motin 14289b17223SAlexander Motin switch (event) { 14389b17223SAlexander Motin case G_RAID_NODE_E_WAKE: 14489b17223SAlexander Motin return ("WAKE"); 14589b17223SAlexander Motin case G_RAID_NODE_E_START: 14689b17223SAlexander Motin return ("START"); 14789b17223SAlexander Motin default: 14889b17223SAlexander Motin return ("INVALID"); 14989b17223SAlexander Motin } 15089b17223SAlexander Motin } 15189b17223SAlexander Motin 15289b17223SAlexander Motin const char * 15389b17223SAlexander Motin g_raid_disk_state2str(int state) 15489b17223SAlexander Motin { 15589b17223SAlexander Motin 15689b17223SAlexander Motin switch (state) { 15789b17223SAlexander Motin case G_RAID_DISK_S_NONE: 15889b17223SAlexander Motin return ("NONE"); 15989b17223SAlexander Motin case G_RAID_DISK_S_OFFLINE: 16089b17223SAlexander Motin return ("OFFLINE"); 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 49789b17223SAlexander Motin g_raid_report_disk_state(struct g_raid_disk *disk) 49889b17223SAlexander Motin { 49989b17223SAlexander Motin struct g_raid_subdisk *sd; 50089b17223SAlexander Motin int len, state; 50189b17223SAlexander Motin uint32_t s; 50289b17223SAlexander Motin 50389b17223SAlexander Motin if (disk->d_consumer == NULL) 50489b17223SAlexander Motin return; 50589b17223SAlexander Motin if (disk->d_state == G_RAID_DISK_S_FAILED || 50689b17223SAlexander Motin disk->d_state == G_RAID_DISK_S_STALE_FAILED) { 50789b17223SAlexander Motin s = G_STATE_FAILED; 50889b17223SAlexander Motin } else { 50989b17223SAlexander Motin state = G_RAID_SUBDISK_S_ACTIVE; 51089b17223SAlexander Motin TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 51189b17223SAlexander Motin if (sd->sd_state < state) 51289b17223SAlexander Motin state = sd->sd_state; 51389b17223SAlexander Motin } 51489b17223SAlexander Motin if (state == G_RAID_SUBDISK_S_FAILED) 51589b17223SAlexander Motin s = G_STATE_FAILED; 51689b17223SAlexander Motin else if (state == G_RAID_SUBDISK_S_NEW || 51789b17223SAlexander Motin state == G_RAID_SUBDISK_S_REBUILD) 51889b17223SAlexander Motin s = G_STATE_REBUILD; 51989b17223SAlexander Motin else if (state == G_RAID_SUBDISK_S_STALE || 52089b17223SAlexander Motin state == G_RAID_SUBDISK_S_RESYNC) 52189b17223SAlexander Motin s = G_STATE_RESYNC; 52289b17223SAlexander Motin else 52389b17223SAlexander Motin s = G_STATE_ACTIVE; 52489b17223SAlexander Motin } 52589b17223SAlexander Motin len = sizeof(s); 52689b17223SAlexander Motin g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s); 52789b17223SAlexander Motin G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.", 52889b17223SAlexander Motin g_raid_get_diskname(disk), s); 52989b17223SAlexander Motin } 53089b17223SAlexander Motin 53189b17223SAlexander Motin void 53289b17223SAlexander Motin g_raid_change_disk_state(struct g_raid_disk *disk, int state) 53389b17223SAlexander Motin { 53489b17223SAlexander Motin 53589b17223SAlexander Motin G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.", 53689b17223SAlexander Motin g_raid_get_diskname(disk), 53789b17223SAlexander Motin g_raid_disk_state2str(disk->d_state), 53889b17223SAlexander Motin g_raid_disk_state2str(state)); 53989b17223SAlexander Motin disk->d_state = state; 54089b17223SAlexander Motin g_raid_report_disk_state(disk); 54189b17223SAlexander Motin } 54289b17223SAlexander Motin 54389b17223SAlexander Motin void 54489b17223SAlexander Motin g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state) 54589b17223SAlexander Motin { 54689b17223SAlexander Motin 54789b17223SAlexander Motin G_RAID_DEBUG1(0, sd->sd_softc, 54889b17223SAlexander Motin "Subdisk %s:%d-%s state changed from %s to %s.", 54989b17223SAlexander Motin sd->sd_volume->v_name, sd->sd_pos, 55089b17223SAlexander Motin sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", 55189b17223SAlexander Motin g_raid_subdisk_state2str(sd->sd_state), 55289b17223SAlexander Motin g_raid_subdisk_state2str(state)); 55389b17223SAlexander Motin sd->sd_state = state; 55489b17223SAlexander Motin if (sd->sd_disk) 55589b17223SAlexander Motin g_raid_report_disk_state(sd->sd_disk); 55689b17223SAlexander Motin } 55789b17223SAlexander Motin 55889b17223SAlexander Motin void 55989b17223SAlexander Motin g_raid_change_volume_state(struct g_raid_volume *vol, int state) 56089b17223SAlexander Motin { 56189b17223SAlexander Motin 56289b17223SAlexander Motin G_RAID_DEBUG1(0, vol->v_softc, 56389b17223SAlexander Motin "Volume %s state changed from %s to %s.", 56489b17223SAlexander Motin vol->v_name, 56589b17223SAlexander Motin g_raid_volume_state2str(vol->v_state), 56689b17223SAlexander Motin g_raid_volume_state2str(state)); 56789b17223SAlexander Motin vol->v_state = state; 56889b17223SAlexander Motin } 56989b17223SAlexander Motin 57089b17223SAlexander Motin /* 57189b17223SAlexander Motin * --- Events handling functions --- 57289b17223SAlexander Motin * Events in geom_raid are used to maintain subdisks and volumes status 57389b17223SAlexander Motin * from one thread to simplify locking. 57489b17223SAlexander Motin */ 57589b17223SAlexander Motin static void 57689b17223SAlexander Motin g_raid_event_free(struct g_raid_event *ep) 57789b17223SAlexander Motin { 57889b17223SAlexander Motin 57989b17223SAlexander Motin free(ep, M_RAID); 58089b17223SAlexander Motin } 58189b17223SAlexander Motin 58289b17223SAlexander Motin int 58389b17223SAlexander Motin g_raid_event_send(void *arg, int event, int flags) 58489b17223SAlexander Motin { 58589b17223SAlexander Motin struct g_raid_softc *sc; 58689b17223SAlexander Motin struct g_raid_event *ep; 58789b17223SAlexander Motin int error; 58889b17223SAlexander Motin 58989b17223SAlexander Motin if ((flags & G_RAID_EVENT_VOLUME) != 0) { 59089b17223SAlexander Motin sc = ((struct g_raid_volume *)arg)->v_softc; 59189b17223SAlexander Motin } else if ((flags & G_RAID_EVENT_DISK) != 0) { 59289b17223SAlexander Motin sc = ((struct g_raid_disk *)arg)->d_softc; 59389b17223SAlexander Motin } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) { 59489b17223SAlexander Motin sc = ((struct g_raid_subdisk *)arg)->sd_softc; 59589b17223SAlexander Motin } else { 59689b17223SAlexander Motin sc = arg; 59789b17223SAlexander Motin } 59889b17223SAlexander Motin ep = malloc(sizeof(*ep), M_RAID, 59989b17223SAlexander Motin sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT); 60089b17223SAlexander Motin if (ep == NULL) 60189b17223SAlexander Motin return (ENOMEM); 60289b17223SAlexander Motin ep->e_tgt = arg; 60389b17223SAlexander Motin ep->e_event = event; 60489b17223SAlexander Motin ep->e_flags = flags; 60589b17223SAlexander Motin ep->e_error = 0; 60689b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc); 60789b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 60889b17223SAlexander Motin TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 60989b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 61089b17223SAlexander Motin wakeup(sc); 61189b17223SAlexander Motin 61289b17223SAlexander Motin if ((flags & G_RAID_EVENT_WAIT) == 0) 61389b17223SAlexander Motin return (0); 61489b17223SAlexander Motin 61589b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 61689b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep); 61789b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 61889b17223SAlexander Motin while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) { 61989b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 62089b17223SAlexander Motin MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event", 62189b17223SAlexander Motin hz * 5); 62289b17223SAlexander Motin } 62389b17223SAlexander Motin error = ep->e_error; 62489b17223SAlexander Motin g_raid_event_free(ep); 62589b17223SAlexander Motin sx_xlock(&sc->sc_lock); 62689b17223SAlexander Motin return (error); 62789b17223SAlexander Motin } 62889b17223SAlexander Motin 62989b17223SAlexander Motin static void 63089b17223SAlexander Motin g_raid_event_cancel(struct g_raid_softc *sc, void *tgt) 63189b17223SAlexander Motin { 63289b17223SAlexander Motin struct g_raid_event *ep, *tmpep; 63389b17223SAlexander Motin 63489b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 63589b17223SAlexander Motin 63689b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 63789b17223SAlexander Motin TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 63889b17223SAlexander Motin if (ep->e_tgt != tgt) 63989b17223SAlexander Motin continue; 64089b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_events, ep, e_next); 64189b17223SAlexander Motin if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) 64289b17223SAlexander Motin g_raid_event_free(ep); 64389b17223SAlexander Motin else { 64489b17223SAlexander Motin ep->e_error = ECANCELED; 64589b17223SAlexander Motin wakeup(ep); 64689b17223SAlexander Motin } 64789b17223SAlexander Motin } 64889b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 64989b17223SAlexander Motin } 65089b17223SAlexander Motin 65189b17223SAlexander Motin static int 65289b17223SAlexander Motin g_raid_event_check(struct g_raid_softc *sc, void *tgt) 65389b17223SAlexander Motin { 65489b17223SAlexander Motin struct g_raid_event *ep; 65589b17223SAlexander Motin int res = 0; 65689b17223SAlexander Motin 65789b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 65889b17223SAlexander Motin 65989b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 66089b17223SAlexander Motin TAILQ_FOREACH(ep, &sc->sc_events, e_next) { 66189b17223SAlexander Motin if (ep->e_tgt != tgt) 66289b17223SAlexander Motin continue; 66389b17223SAlexander Motin res = 1; 66489b17223SAlexander Motin break; 66589b17223SAlexander Motin } 66689b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 66789b17223SAlexander Motin return (res); 66889b17223SAlexander Motin } 66989b17223SAlexander Motin 67089b17223SAlexander Motin /* 67189b17223SAlexander Motin * Return the number of disks in given state. 67289b17223SAlexander Motin * If state is equal to -1, count all connected disks. 67389b17223SAlexander Motin */ 67489b17223SAlexander Motin u_int 67589b17223SAlexander Motin g_raid_ndisks(struct g_raid_softc *sc, int state) 67689b17223SAlexander Motin { 67789b17223SAlexander Motin struct g_raid_disk *disk; 67889b17223SAlexander Motin u_int n; 67989b17223SAlexander Motin 68089b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 68189b17223SAlexander Motin 68289b17223SAlexander Motin n = 0; 68389b17223SAlexander Motin TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 68489b17223SAlexander Motin if (disk->d_state == state || state == -1) 68589b17223SAlexander Motin n++; 68689b17223SAlexander Motin } 68789b17223SAlexander Motin return (n); 68889b17223SAlexander Motin } 68989b17223SAlexander Motin 69089b17223SAlexander Motin /* 69189b17223SAlexander Motin * Return the number of subdisks in given state. 69289b17223SAlexander Motin * If state is equal to -1, count all connected disks. 69389b17223SAlexander Motin */ 69489b17223SAlexander Motin u_int 69589b17223SAlexander Motin g_raid_nsubdisks(struct g_raid_volume *vol, int state) 69689b17223SAlexander Motin { 69789b17223SAlexander Motin struct g_raid_subdisk *subdisk; 69889b17223SAlexander Motin struct g_raid_softc *sc; 69989b17223SAlexander Motin u_int i, n ; 70089b17223SAlexander Motin 70189b17223SAlexander Motin sc = vol->v_softc; 70289b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 70389b17223SAlexander Motin 70489b17223SAlexander Motin n = 0; 70589b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 70689b17223SAlexander Motin subdisk = &vol->v_subdisks[i]; 70789b17223SAlexander Motin if ((state == -1 && 70889b17223SAlexander Motin subdisk->sd_state != G_RAID_SUBDISK_S_NONE) || 70989b17223SAlexander Motin subdisk->sd_state == state) 71089b17223SAlexander Motin n++; 71189b17223SAlexander Motin } 71289b17223SAlexander Motin return (n); 71389b17223SAlexander Motin } 71489b17223SAlexander Motin 71589b17223SAlexander Motin /* 71689b17223SAlexander Motin * Return the first subdisk in given state. 71789b17223SAlexander Motin * If state is equal to -1, then the first connected disks. 71889b17223SAlexander Motin */ 71989b17223SAlexander Motin struct g_raid_subdisk * 72089b17223SAlexander Motin g_raid_get_subdisk(struct g_raid_volume *vol, int state) 72189b17223SAlexander Motin { 72289b17223SAlexander Motin struct g_raid_subdisk *sd; 72389b17223SAlexander Motin struct g_raid_softc *sc; 72489b17223SAlexander Motin u_int i; 72589b17223SAlexander Motin 72689b17223SAlexander Motin sc = vol->v_softc; 72789b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 72889b17223SAlexander Motin 72989b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 73089b17223SAlexander Motin sd = &vol->v_subdisks[i]; 73189b17223SAlexander Motin if ((state == -1 && 73289b17223SAlexander Motin sd->sd_state != G_RAID_SUBDISK_S_NONE) || 73389b17223SAlexander Motin sd->sd_state == state) 73489b17223SAlexander Motin return (sd); 73589b17223SAlexander Motin } 73689b17223SAlexander Motin return (NULL); 73789b17223SAlexander Motin } 73889b17223SAlexander Motin 73989b17223SAlexander Motin struct g_consumer * 74089b17223SAlexander Motin g_raid_open_consumer(struct g_raid_softc *sc, const char *name) 74189b17223SAlexander Motin { 74289b17223SAlexander Motin struct g_consumer *cp; 74389b17223SAlexander Motin struct g_provider *pp; 74489b17223SAlexander Motin 74589b17223SAlexander Motin g_topology_assert(); 74689b17223SAlexander Motin 74789b17223SAlexander Motin if (strncmp(name, "/dev/", 5) == 0) 74889b17223SAlexander Motin name += 5; 74989b17223SAlexander Motin pp = g_provider_by_name(name); 75089b17223SAlexander Motin if (pp == NULL) 75189b17223SAlexander Motin return (NULL); 75289b17223SAlexander Motin cp = g_new_consumer(sc->sc_geom); 75389b17223SAlexander Motin if (g_attach(cp, pp) != 0) { 75489b17223SAlexander Motin g_destroy_consumer(cp); 75589b17223SAlexander Motin return (NULL); 75689b17223SAlexander Motin } 75789b17223SAlexander Motin if (g_access(cp, 1, 1, 1) != 0) { 75889b17223SAlexander Motin g_detach(cp); 75989b17223SAlexander Motin g_destroy_consumer(cp); 76089b17223SAlexander Motin return (NULL); 76189b17223SAlexander Motin } 76289b17223SAlexander Motin return (cp); 76389b17223SAlexander Motin } 76489b17223SAlexander Motin 76589b17223SAlexander Motin static u_int 76689b17223SAlexander Motin g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) 76789b17223SAlexander Motin { 76889b17223SAlexander Motin struct bio *bp; 76989b17223SAlexander Motin u_int nreqs = 0; 77089b17223SAlexander Motin 77189b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 77289b17223SAlexander Motin TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 77389b17223SAlexander Motin if (bp->bio_from == cp) 77489b17223SAlexander Motin nreqs++; 77589b17223SAlexander Motin } 77689b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 77789b17223SAlexander Motin return (nreqs); 77889b17223SAlexander Motin } 77989b17223SAlexander Motin 78089b17223SAlexander Motin u_int 78189b17223SAlexander Motin g_raid_nopens(struct g_raid_softc *sc) 78289b17223SAlexander Motin { 78389b17223SAlexander Motin struct g_raid_volume *vol; 78489b17223SAlexander Motin u_int opens; 78589b17223SAlexander Motin 78689b17223SAlexander Motin opens = 0; 78789b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 78889b17223SAlexander Motin if (vol->v_provider_open != 0) 78989b17223SAlexander Motin opens++; 79089b17223SAlexander Motin } 79189b17223SAlexander Motin return (opens); 79289b17223SAlexander Motin } 79389b17223SAlexander Motin 79489b17223SAlexander Motin static int 79589b17223SAlexander Motin g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) 79689b17223SAlexander Motin { 79789b17223SAlexander Motin 79889b17223SAlexander Motin if (cp->index > 0) { 79989b17223SAlexander Motin G_RAID_DEBUG1(2, sc, 80089b17223SAlexander Motin "I/O requests for %s exist, can't destroy it now.", 80189b17223SAlexander Motin cp->provider->name); 80289b17223SAlexander Motin return (1); 80389b17223SAlexander Motin } 80489b17223SAlexander Motin if (g_raid_nrequests(sc, cp) > 0) { 80589b17223SAlexander Motin G_RAID_DEBUG1(2, sc, 80689b17223SAlexander Motin "I/O requests for %s in queue, can't destroy it now.", 80789b17223SAlexander Motin cp->provider->name); 80889b17223SAlexander Motin return (1); 80989b17223SAlexander Motin } 81089b17223SAlexander Motin return (0); 81189b17223SAlexander Motin } 81289b17223SAlexander Motin 81389b17223SAlexander Motin static void 81489b17223SAlexander Motin g_raid_destroy_consumer(void *arg, int flags __unused) 81589b17223SAlexander Motin { 81689b17223SAlexander Motin struct g_consumer *cp; 81789b17223SAlexander Motin 81889b17223SAlexander Motin g_topology_assert(); 81989b17223SAlexander Motin 82089b17223SAlexander Motin cp = arg; 82189b17223SAlexander Motin G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 82289b17223SAlexander Motin g_detach(cp); 82389b17223SAlexander Motin g_destroy_consumer(cp); 82489b17223SAlexander Motin } 82589b17223SAlexander Motin 82689b17223SAlexander Motin void 82789b17223SAlexander Motin g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) 82889b17223SAlexander Motin { 82989b17223SAlexander Motin struct g_provider *pp; 83089b17223SAlexander Motin int retaste_wait; 83189b17223SAlexander Motin 83289b17223SAlexander Motin g_topology_assert_not(); 83389b17223SAlexander Motin 83489b17223SAlexander Motin g_topology_lock(); 83589b17223SAlexander Motin cp->private = NULL; 83689b17223SAlexander Motin if (g_raid_consumer_is_busy(sc, cp)) 83789b17223SAlexander Motin goto out; 83889b17223SAlexander Motin pp = cp->provider; 83989b17223SAlexander Motin retaste_wait = 0; 84089b17223SAlexander Motin if (cp->acw == 1) { 84189b17223SAlexander Motin if ((pp->geom->flags & G_GEOM_WITHER) == 0) 84289b17223SAlexander Motin retaste_wait = 1; 84389b17223SAlexander Motin } 84489b17223SAlexander Motin if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 84589b17223SAlexander Motin g_access(cp, -cp->acr, -cp->acw, -cp->ace); 84689b17223SAlexander Motin if (retaste_wait) { 84789b17223SAlexander Motin /* 84889b17223SAlexander Motin * After retaste event was send (inside g_access()), we can send 84989b17223SAlexander Motin * event to detach and destroy consumer. 85089b17223SAlexander Motin * A class, which has consumer to the given provider connected 85189b17223SAlexander Motin * will not receive retaste event for the provider. 85289b17223SAlexander Motin * This is the way how I ignore retaste events when I close 85389b17223SAlexander Motin * consumers opened for write: I detach and destroy consumer 85489b17223SAlexander Motin * after retaste event is sent. 85589b17223SAlexander Motin */ 85689b17223SAlexander Motin g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); 85789b17223SAlexander Motin goto out; 85889b17223SAlexander Motin } 85989b17223SAlexander Motin G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); 86089b17223SAlexander Motin g_detach(cp); 86189b17223SAlexander Motin g_destroy_consumer(cp); 86289b17223SAlexander Motin out: 86389b17223SAlexander Motin g_topology_unlock(); 86489b17223SAlexander Motin } 86589b17223SAlexander Motin 86689b17223SAlexander Motin static void 86789b17223SAlexander Motin g_raid_orphan(struct g_consumer *cp) 86889b17223SAlexander Motin { 86989b17223SAlexander Motin struct g_raid_disk *disk; 87089b17223SAlexander Motin 87189b17223SAlexander Motin g_topology_assert(); 87289b17223SAlexander Motin 87389b17223SAlexander Motin disk = cp->private; 87489b17223SAlexander Motin if (disk == NULL) 87589b17223SAlexander Motin return; 87689b17223SAlexander Motin g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, 87789b17223SAlexander Motin G_RAID_EVENT_DISK); 87889b17223SAlexander Motin } 87989b17223SAlexander Motin 88089b17223SAlexander Motin static int 88189b17223SAlexander Motin g_raid_clean(struct g_raid_volume *vol, int acw) 88289b17223SAlexander Motin { 88389b17223SAlexander Motin struct g_raid_softc *sc; 88489b17223SAlexander Motin int timeout; 88589b17223SAlexander Motin 88689b17223SAlexander Motin sc = vol->v_softc; 88789b17223SAlexander Motin g_topology_assert_not(); 88889b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 88989b17223SAlexander Motin 89089b17223SAlexander Motin // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 89189b17223SAlexander Motin // return (0); 89289b17223SAlexander Motin if (!vol->v_dirty) 89389b17223SAlexander Motin return (0); 89489b17223SAlexander Motin if (vol->v_writes > 0) 89589b17223SAlexander Motin return (0); 89689b17223SAlexander Motin if (acw > 0 || (acw == -1 && 89789b17223SAlexander Motin vol->v_provider != NULL && vol->v_provider->acw > 0)) { 89889b17223SAlexander Motin timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); 89989b17223SAlexander Motin if (timeout > 0) 90089b17223SAlexander Motin return (timeout); 90189b17223SAlexander Motin } 90289b17223SAlexander Motin vol->v_dirty = 0; 90389b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", 90489b17223SAlexander Motin vol->v_name); 90589b17223SAlexander Motin g_raid_write_metadata(sc, vol, NULL, NULL); 90689b17223SAlexander Motin return (0); 90789b17223SAlexander Motin } 90889b17223SAlexander Motin 90989b17223SAlexander Motin static void 91089b17223SAlexander Motin g_raid_dirty(struct g_raid_volume *vol) 91189b17223SAlexander Motin { 91289b17223SAlexander Motin struct g_raid_softc *sc; 91389b17223SAlexander Motin 91489b17223SAlexander Motin sc = vol->v_softc; 91589b17223SAlexander Motin g_topology_assert_not(); 91689b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 91789b17223SAlexander Motin 91889b17223SAlexander Motin // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 91989b17223SAlexander Motin // return; 92089b17223SAlexander Motin vol->v_dirty = 1; 92189b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", 92289b17223SAlexander Motin vol->v_name); 92389b17223SAlexander Motin g_raid_write_metadata(sc, vol, NULL, NULL); 92489b17223SAlexander Motin } 92589b17223SAlexander Motin 92689b17223SAlexander Motin void 92789b17223SAlexander Motin g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) 92889b17223SAlexander Motin { 92989b17223SAlexander Motin struct g_raid_softc *sc; 93089b17223SAlexander Motin struct g_raid_volume *vol; 93189b17223SAlexander Motin struct g_raid_subdisk *sd; 93289b17223SAlexander Motin struct bio_queue_head queue; 93389b17223SAlexander Motin struct bio *cbp; 93489b17223SAlexander Motin int i; 93589b17223SAlexander Motin 93689b17223SAlexander Motin vol = tr->tro_volume; 93789b17223SAlexander Motin sc = vol->v_softc; 93889b17223SAlexander Motin 93989b17223SAlexander Motin /* 94089b17223SAlexander Motin * Allocate all bios before sending any request, so we can return 94189b17223SAlexander Motin * ENOMEM in nice and clean way. 94289b17223SAlexander Motin */ 94389b17223SAlexander Motin bioq_init(&queue); 94489b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 94589b17223SAlexander Motin sd = &vol->v_subdisks[i]; 94689b17223SAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_NONE || 94789b17223SAlexander Motin sd->sd_state == G_RAID_SUBDISK_S_FAILED) 94889b17223SAlexander Motin continue; 94989b17223SAlexander Motin cbp = g_clone_bio(bp); 95089b17223SAlexander Motin if (cbp == NULL) 95189b17223SAlexander Motin goto failure; 95289b17223SAlexander Motin cbp->bio_caller1 = sd; 95389b17223SAlexander Motin bioq_insert_tail(&queue, cbp); 95489b17223SAlexander Motin } 95589b17223SAlexander Motin for (cbp = bioq_first(&queue); cbp != NULL; 95689b17223SAlexander Motin cbp = bioq_first(&queue)) { 95789b17223SAlexander Motin bioq_remove(&queue, cbp); 95889b17223SAlexander Motin sd = cbp->bio_caller1; 95989b17223SAlexander Motin cbp->bio_caller1 = NULL; 96089b17223SAlexander Motin g_raid_subdisk_iostart(sd, cbp); 96189b17223SAlexander Motin } 96289b17223SAlexander Motin return; 96389b17223SAlexander Motin failure: 96489b17223SAlexander Motin for (cbp = bioq_first(&queue); cbp != NULL; 96589b17223SAlexander Motin cbp = bioq_first(&queue)) { 96689b17223SAlexander Motin bioq_remove(&queue, cbp); 96789b17223SAlexander Motin g_destroy_bio(cbp); 96889b17223SAlexander Motin } 96989b17223SAlexander Motin if (bp->bio_error == 0) 97089b17223SAlexander Motin bp->bio_error = ENOMEM; 97189b17223SAlexander Motin g_raid_iodone(bp, bp->bio_error); 97289b17223SAlexander Motin } 97389b17223SAlexander Motin 97489b17223SAlexander Motin static void 97589b17223SAlexander Motin g_raid_tr_kerneldump_common_done(struct bio *bp) 97689b17223SAlexander Motin { 97789b17223SAlexander Motin 97889b17223SAlexander Motin bp->bio_flags |= BIO_DONE; 97989b17223SAlexander Motin } 98089b17223SAlexander Motin 98189b17223SAlexander Motin int 98289b17223SAlexander Motin g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, 98389b17223SAlexander Motin void *virtual, vm_offset_t physical, off_t offset, size_t length) 98489b17223SAlexander Motin { 98589b17223SAlexander Motin struct g_raid_softc *sc; 98689b17223SAlexander Motin struct g_raid_volume *vol; 98789b17223SAlexander Motin struct bio bp; 98889b17223SAlexander Motin 98989b17223SAlexander Motin vol = tr->tro_volume; 99089b17223SAlexander Motin sc = vol->v_softc; 99189b17223SAlexander Motin 99289b17223SAlexander Motin bzero(&bp, sizeof(bp)); 99389b17223SAlexander Motin bp.bio_cmd = BIO_WRITE; 99489b17223SAlexander Motin bp.bio_done = g_raid_tr_kerneldump_common_done; 99589b17223SAlexander Motin bp.bio_attribute = NULL; 99689b17223SAlexander Motin bp.bio_offset = offset; 99789b17223SAlexander Motin bp.bio_length = length; 99889b17223SAlexander Motin bp.bio_data = virtual; 99989b17223SAlexander Motin bp.bio_to = vol->v_provider; 100089b17223SAlexander Motin 100189b17223SAlexander Motin g_raid_start(&bp); 100289b17223SAlexander Motin while (!(bp.bio_flags & BIO_DONE)) { 100389b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Poll..."); 100489b17223SAlexander Motin g_raid_poll(sc); 100589b17223SAlexander Motin DELAY(10); 100689b17223SAlexander Motin } 100789b17223SAlexander Motin 100889b17223SAlexander Motin return (bp.bio_error != 0 ? EIO : 0); 100989b17223SAlexander Motin } 101089b17223SAlexander Motin 101189b17223SAlexander Motin static int 101289b17223SAlexander Motin g_raid_dump(void *arg, 101389b17223SAlexander Motin void *virtual, vm_offset_t physical, off_t offset, size_t length) 101489b17223SAlexander Motin { 101589b17223SAlexander Motin struct g_raid_volume *vol; 101689b17223SAlexander Motin int error; 101789b17223SAlexander Motin 101889b17223SAlexander Motin vol = (struct g_raid_volume *)arg; 101989b17223SAlexander Motin G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", 102089b17223SAlexander Motin (long long unsigned)offset, (long long unsigned)length); 102189b17223SAlexander Motin 102289b17223SAlexander Motin error = G_RAID_TR_KERNELDUMP(vol->v_tr, 102389b17223SAlexander Motin virtual, physical, offset, length); 102489b17223SAlexander Motin return (error); 102589b17223SAlexander Motin } 102689b17223SAlexander Motin 102789b17223SAlexander Motin static void 102889b17223SAlexander Motin g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) 102989b17223SAlexander Motin { 103089b17223SAlexander Motin struct g_kerneldump *gkd; 103189b17223SAlexander Motin struct g_provider *pp; 103289b17223SAlexander Motin struct g_raid_volume *vol; 103389b17223SAlexander Motin 103489b17223SAlexander Motin gkd = (struct g_kerneldump*)bp->bio_data; 103589b17223SAlexander Motin pp = bp->bio_to; 103689b17223SAlexander Motin vol = pp->private; 103789b17223SAlexander Motin g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", 103889b17223SAlexander Motin pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 103989b17223SAlexander Motin gkd->di.dumper = g_raid_dump; 104089b17223SAlexander Motin gkd->di.priv = vol; 104189b17223SAlexander Motin gkd->di.blocksize = vol->v_sectorsize; 104289b17223SAlexander Motin gkd->di.maxiosize = DFLTPHYS; 104389b17223SAlexander Motin gkd->di.mediaoffset = gkd->offset; 104489b17223SAlexander Motin if ((gkd->offset + gkd->length) > vol->v_mediasize) 104589b17223SAlexander Motin gkd->length = vol->v_mediasize - gkd->offset; 104689b17223SAlexander Motin gkd->di.mediasize = gkd->length; 104789b17223SAlexander Motin g_io_deliver(bp, 0); 104889b17223SAlexander Motin } 104989b17223SAlexander Motin 105089b17223SAlexander Motin static void 105189b17223SAlexander Motin g_raid_start(struct bio *bp) 105289b17223SAlexander Motin { 105389b17223SAlexander Motin struct g_raid_softc *sc; 105489b17223SAlexander Motin 105589b17223SAlexander Motin sc = bp->bio_to->geom->softc; 105689b17223SAlexander Motin /* 105789b17223SAlexander Motin * If sc == NULL or there are no valid disks, provider's error 105889b17223SAlexander Motin * should be set and g_raid_start() should not be called at all. 105989b17223SAlexander Motin */ 106089b17223SAlexander Motin // KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, 106189b17223SAlexander Motin // ("Provider's error should be set (error=%d)(mirror=%s).", 106289b17223SAlexander Motin // bp->bio_to->error, bp->bio_to->name)); 106389b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Request received."); 106489b17223SAlexander Motin 106589b17223SAlexander Motin switch (bp->bio_cmd) { 106689b17223SAlexander Motin case BIO_READ: 106789b17223SAlexander Motin case BIO_WRITE: 106889b17223SAlexander Motin case BIO_DELETE: 106989b17223SAlexander Motin case BIO_FLUSH: 107089b17223SAlexander Motin break; 107189b17223SAlexander Motin case BIO_GETATTR: 107289b17223SAlexander Motin if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 107389b17223SAlexander Motin g_raid_kerneldump(sc, bp); 107489b17223SAlexander Motin else 107589b17223SAlexander Motin g_io_deliver(bp, EOPNOTSUPP); 107689b17223SAlexander Motin return; 107789b17223SAlexander Motin default: 107889b17223SAlexander Motin g_io_deliver(bp, EOPNOTSUPP); 107989b17223SAlexander Motin return; 108089b17223SAlexander Motin } 108189b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 108289b17223SAlexander Motin bioq_disksort(&sc->sc_queue, bp); 108389b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 108489b17223SAlexander Motin if (!dumping) { 108589b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); 108689b17223SAlexander Motin wakeup(sc); 108789b17223SAlexander Motin } 108889b17223SAlexander Motin } 108989b17223SAlexander Motin 109089b17223SAlexander Motin static int 109189b17223SAlexander Motin g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) 109289b17223SAlexander Motin { 109389b17223SAlexander Motin /* 109489b17223SAlexander Motin * 5 cases: 109589b17223SAlexander Motin * (1) bp entirely below NO 109689b17223SAlexander Motin * (2) bp entirely above NO 109789b17223SAlexander Motin * (3) bp start below, but end in range YES 109889b17223SAlexander Motin * (4) bp entirely within YES 109989b17223SAlexander Motin * (5) bp starts within, ends above YES 110089b17223SAlexander Motin * 110189b17223SAlexander Motin * lock range 10-19 (offset 10 length 10) 110289b17223SAlexander Motin * (1) 1-5: first if kicks it out 110389b17223SAlexander Motin * (2) 30-35: second if kicks it out 110489b17223SAlexander Motin * (3) 5-15: passes both ifs 110589b17223SAlexander Motin * (4) 12-14: passes both ifs 110689b17223SAlexander Motin * (5) 19-20: passes both 110789b17223SAlexander Motin */ 110889b17223SAlexander Motin off_t lend = lstart + len - 1; 110989b17223SAlexander Motin off_t bstart = bp->bio_offset; 111089b17223SAlexander Motin off_t bend = bp->bio_offset + bp->bio_length - 1; 111189b17223SAlexander Motin 111289b17223SAlexander Motin if (bend < lstart) 111389b17223SAlexander Motin return (0); 111489b17223SAlexander Motin if (lend < bstart) 111589b17223SAlexander Motin return (0); 111689b17223SAlexander Motin return (1); 111789b17223SAlexander Motin } 111889b17223SAlexander Motin 111989b17223SAlexander Motin static int 112089b17223SAlexander Motin g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) 112189b17223SAlexander Motin { 112289b17223SAlexander Motin struct g_raid_lock *lp; 112389b17223SAlexander Motin 112489b17223SAlexander Motin sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); 112589b17223SAlexander Motin 112689b17223SAlexander Motin LIST_FOREACH(lp, &vol->v_locks, l_next) { 112789b17223SAlexander Motin if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) 112889b17223SAlexander Motin return (1); 112989b17223SAlexander Motin } 113089b17223SAlexander Motin return (0); 113189b17223SAlexander Motin } 113289b17223SAlexander Motin 113389b17223SAlexander Motin static void 113489b17223SAlexander Motin g_raid_start_request(struct bio *bp) 113589b17223SAlexander Motin { 113689b17223SAlexander Motin struct g_raid_softc *sc; 113789b17223SAlexander Motin struct g_raid_volume *vol; 113889b17223SAlexander Motin 113989b17223SAlexander Motin sc = bp->bio_to->geom->softc; 114089b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 114189b17223SAlexander Motin vol = bp->bio_to->private; 114289b17223SAlexander Motin 114389b17223SAlexander Motin /* 114489b17223SAlexander Motin * Check to see if this item is in a locked range. If so, 114589b17223SAlexander Motin * queue it to our locked queue and return. We'll requeue 114689b17223SAlexander Motin * it when the range is unlocked. Internal I/O for the 114789b17223SAlexander Motin * rebuild/rescan/recovery process is excluded from this 114889b17223SAlexander Motin * check so we can actually do the recovery. 114989b17223SAlexander Motin */ 115089b17223SAlexander Motin if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && 115189b17223SAlexander Motin g_raid_is_in_locked_range(vol, bp)) { 115289b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Defer request."); 115389b17223SAlexander Motin bioq_insert_tail(&vol->v_locked, bp); 115489b17223SAlexander Motin return; 115589b17223SAlexander Motin } 115689b17223SAlexander Motin 115789b17223SAlexander Motin /* 115889b17223SAlexander Motin * If we're actually going to do the write/delete, then 115989b17223SAlexander Motin * update the idle stats for the volume. 116089b17223SAlexander Motin */ 116189b17223SAlexander Motin if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 116289b17223SAlexander Motin if (!vol->v_dirty) 116389b17223SAlexander Motin g_raid_dirty(vol); 116489b17223SAlexander Motin vol->v_writes++; 116589b17223SAlexander Motin } 116689b17223SAlexander Motin 116789b17223SAlexander Motin /* 116889b17223SAlexander Motin * Put request onto inflight queue, so we can check if new 116989b17223SAlexander Motin * synchronization requests don't collide with it. Then tell 117089b17223SAlexander Motin * the transformation layer to start the I/O. 117189b17223SAlexander Motin */ 117289b17223SAlexander Motin bioq_insert_tail(&vol->v_inflight, bp); 117389b17223SAlexander Motin G_RAID_LOGREQ(4, bp, "Request started"); 117489b17223SAlexander Motin G_RAID_TR_IOSTART(vol->v_tr, bp); 117589b17223SAlexander Motin } 117689b17223SAlexander Motin 117789b17223SAlexander Motin static void 117889b17223SAlexander Motin g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) 117989b17223SAlexander Motin { 118089b17223SAlexander Motin off_t off, len; 118189b17223SAlexander Motin struct bio *nbp; 118289b17223SAlexander Motin struct g_raid_lock *lp; 118389b17223SAlexander Motin 118489b17223SAlexander Motin vol->v_pending_lock = 0; 118589b17223SAlexander Motin LIST_FOREACH(lp, &vol->v_locks, l_next) { 118689b17223SAlexander Motin if (lp->l_pending) { 118789b17223SAlexander Motin off = lp->l_offset; 118889b17223SAlexander Motin len = lp->l_length; 118989b17223SAlexander Motin lp->l_pending = 0; 119089b17223SAlexander Motin TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { 119189b17223SAlexander Motin if (g_raid_bio_overlaps(nbp, off, len)) 119289b17223SAlexander Motin lp->l_pending++; 119389b17223SAlexander Motin } 119489b17223SAlexander Motin if (lp->l_pending) { 119589b17223SAlexander Motin vol->v_pending_lock = 1; 119689b17223SAlexander Motin G_RAID_DEBUG1(4, vol->v_softc, 119789b17223SAlexander Motin "Deferred lock(%jd, %jd) has %d pending", 119889b17223SAlexander Motin (intmax_t)off, (intmax_t)(off + len), 119989b17223SAlexander Motin lp->l_pending); 120089b17223SAlexander Motin continue; 120189b17223SAlexander Motin } 120289b17223SAlexander Motin G_RAID_DEBUG1(4, vol->v_softc, 120389b17223SAlexander Motin "Deferred lock of %jd to %jd completed", 120489b17223SAlexander Motin (intmax_t)off, (intmax_t)(off + len)); 120589b17223SAlexander Motin G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 120689b17223SAlexander Motin } 120789b17223SAlexander Motin } 120889b17223SAlexander Motin } 120989b17223SAlexander Motin 121089b17223SAlexander Motin void 121189b17223SAlexander Motin g_raid_iodone(struct bio *bp, int error) 121289b17223SAlexander Motin { 121389b17223SAlexander Motin struct g_raid_softc *sc; 121489b17223SAlexander Motin struct g_raid_volume *vol; 121589b17223SAlexander Motin 121689b17223SAlexander Motin sc = bp->bio_to->geom->softc; 121789b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 121889b17223SAlexander Motin vol = bp->bio_to->private; 121989b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Request done: %d.", error); 122089b17223SAlexander Motin 122189b17223SAlexander Motin /* Update stats if we done write/delete. */ 122289b17223SAlexander Motin if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 122389b17223SAlexander Motin vol->v_writes--; 122489b17223SAlexander Motin vol->v_last_write = time_uptime; 122589b17223SAlexander Motin } 122689b17223SAlexander Motin 122789b17223SAlexander Motin bioq_remove(&vol->v_inflight, bp); 122889b17223SAlexander Motin if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) 122989b17223SAlexander Motin g_raid_finish_with_locked_ranges(vol, bp); 123089b17223SAlexander Motin getmicrouptime(&vol->v_last_done); 123189b17223SAlexander Motin g_io_deliver(bp, error); 123289b17223SAlexander Motin } 123389b17223SAlexander Motin 123489b17223SAlexander Motin int 123589b17223SAlexander Motin g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, 123689b17223SAlexander Motin struct bio *ignore, void *argp) 123789b17223SAlexander Motin { 123889b17223SAlexander Motin struct g_raid_softc *sc; 123989b17223SAlexander Motin struct g_raid_lock *lp; 124089b17223SAlexander Motin struct bio *bp; 124189b17223SAlexander Motin 124289b17223SAlexander Motin sc = vol->v_softc; 124389b17223SAlexander Motin lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); 124489b17223SAlexander Motin LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); 124589b17223SAlexander Motin lp->l_offset = off; 124689b17223SAlexander Motin lp->l_length = len; 124789b17223SAlexander Motin lp->l_callback_arg = argp; 124889b17223SAlexander Motin 124989b17223SAlexander Motin lp->l_pending = 0; 125089b17223SAlexander Motin TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { 125189b17223SAlexander Motin if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) 125289b17223SAlexander Motin lp->l_pending++; 125389b17223SAlexander Motin } 125489b17223SAlexander Motin 125589b17223SAlexander Motin /* 125689b17223SAlexander Motin * If there are any writes that are pending, we return EBUSY. All 125789b17223SAlexander Motin * callers will have to wait until all pending writes clear. 125889b17223SAlexander Motin */ 125989b17223SAlexander Motin if (lp->l_pending > 0) { 126089b17223SAlexander Motin vol->v_pending_lock = 1; 126189b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", 126289b17223SAlexander Motin (intmax_t)off, (intmax_t)(off+len), lp->l_pending); 126389b17223SAlexander Motin return (EBUSY); 126489b17223SAlexander Motin } 126589b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", 126689b17223SAlexander Motin (intmax_t)off, (intmax_t)(off+len)); 126789b17223SAlexander Motin G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 126889b17223SAlexander Motin return (0); 126989b17223SAlexander Motin } 127089b17223SAlexander Motin 127189b17223SAlexander Motin int 127289b17223SAlexander Motin g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) 127389b17223SAlexander Motin { 127489b17223SAlexander Motin struct g_raid_lock *lp; 127589b17223SAlexander Motin struct g_raid_softc *sc; 127689b17223SAlexander Motin struct bio *bp; 127789b17223SAlexander Motin 127889b17223SAlexander Motin sc = vol->v_softc; 127989b17223SAlexander Motin LIST_FOREACH(lp, &vol->v_locks, l_next) { 128089b17223SAlexander Motin if (lp->l_offset == off && lp->l_length == len) { 128189b17223SAlexander Motin LIST_REMOVE(lp, l_next); 128289b17223SAlexander Motin /* XXX 128389b17223SAlexander Motin * Right now we just put them all back on the queue 128489b17223SAlexander Motin * and hope for the best. We hope this because any 128589b17223SAlexander Motin * locked ranges will go right back on this list 128689b17223SAlexander Motin * when the worker thread runs. 128789b17223SAlexander Motin * XXX 128889b17223SAlexander Motin */ 128989b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", 129089b17223SAlexander Motin (intmax_t)lp->l_offset, 129189b17223SAlexander Motin (intmax_t)(lp->l_offset+lp->l_length)); 129289b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 129389b17223SAlexander Motin while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) 129489b17223SAlexander Motin bioq_disksort(&sc->sc_queue, bp); 129589b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 129689b17223SAlexander Motin free(lp, M_RAID); 129789b17223SAlexander Motin return (0); 129889b17223SAlexander Motin } 129989b17223SAlexander Motin } 130089b17223SAlexander Motin return (EINVAL); 130189b17223SAlexander Motin } 130289b17223SAlexander Motin 130389b17223SAlexander Motin void 130489b17223SAlexander Motin g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) 130589b17223SAlexander Motin { 130689b17223SAlexander Motin struct g_consumer *cp; 130789b17223SAlexander Motin struct g_raid_disk *disk, *tdisk; 130889b17223SAlexander Motin 130989b17223SAlexander Motin bp->bio_caller1 = sd; 131089b17223SAlexander Motin 131189b17223SAlexander Motin /* 131289b17223SAlexander Motin * Make sure that the disk is present. Generally it is a task of 131389b17223SAlexander Motin * transformation layers to not send requests to absent disks, but 131489b17223SAlexander Motin * it is better to be safe and report situation then sorry. 131589b17223SAlexander Motin */ 131689b17223SAlexander Motin if (sd->sd_disk == NULL) { 131789b17223SAlexander Motin G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); 131889b17223SAlexander Motin nodisk: 131989b17223SAlexander Motin bp->bio_from = NULL; 132089b17223SAlexander Motin bp->bio_to = NULL; 132189b17223SAlexander Motin bp->bio_error = ENXIO; 132289b17223SAlexander Motin g_raid_disk_done(bp); 132389b17223SAlexander Motin return; 132489b17223SAlexander Motin } 132589b17223SAlexander Motin disk = sd->sd_disk; 132689b17223SAlexander Motin if (disk->d_state != G_RAID_DISK_S_ACTIVE && 132789b17223SAlexander Motin disk->d_state != G_RAID_DISK_S_FAILED) { 132889b17223SAlexander Motin G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " 132989b17223SAlexander Motin "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 133089b17223SAlexander Motin goto nodisk; 133189b17223SAlexander Motin } 133289b17223SAlexander Motin 133389b17223SAlexander Motin cp = disk->d_consumer; 133489b17223SAlexander Motin bp->bio_from = cp; 133589b17223SAlexander Motin bp->bio_to = cp->provider; 133689b17223SAlexander Motin cp->index++; 133789b17223SAlexander Motin 133889b17223SAlexander Motin /* Update average disks load. */ 133989b17223SAlexander Motin TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { 134089b17223SAlexander Motin if (tdisk->d_consumer == NULL) 134189b17223SAlexander Motin tdisk->d_load = 0; 134289b17223SAlexander Motin else 134389b17223SAlexander Motin tdisk->d_load = (tdisk->d_consumer->index * 134489b17223SAlexander Motin G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; 134589b17223SAlexander Motin } 134689b17223SAlexander Motin 134789b17223SAlexander Motin disk->d_last_offset = bp->bio_offset + bp->bio_length; 134889b17223SAlexander Motin if (dumping) { 134989b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Sending dumping request."); 135089b17223SAlexander Motin if (bp->bio_cmd == BIO_WRITE) { 135189b17223SAlexander Motin bp->bio_error = g_raid_subdisk_kerneldump(sd, 135289b17223SAlexander Motin bp->bio_data, 0, bp->bio_offset, bp->bio_length); 135389b17223SAlexander Motin } else 135489b17223SAlexander Motin bp->bio_error = EOPNOTSUPP; 135589b17223SAlexander Motin g_raid_disk_done(bp); 135689b17223SAlexander Motin } else { 135789b17223SAlexander Motin bp->bio_done = g_raid_disk_done; 135889b17223SAlexander Motin bp->bio_offset += sd->sd_offset; 135989b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Sending request."); 136089b17223SAlexander Motin g_io_request(bp, cp); 136189b17223SAlexander Motin } 136289b17223SAlexander Motin } 136389b17223SAlexander Motin 136489b17223SAlexander Motin int 136589b17223SAlexander Motin g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, 136689b17223SAlexander Motin void *virtual, vm_offset_t physical, off_t offset, size_t length) 136789b17223SAlexander Motin { 136889b17223SAlexander Motin 136989b17223SAlexander Motin if (sd->sd_disk == NULL) 137089b17223SAlexander Motin return (ENXIO); 137189b17223SAlexander Motin if (sd->sd_disk->d_kd.di.dumper == NULL) 137289b17223SAlexander Motin return (EOPNOTSUPP); 137389b17223SAlexander Motin return (dump_write(&sd->sd_disk->d_kd.di, 137489b17223SAlexander Motin virtual, physical, 137589b17223SAlexander Motin sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, 137689b17223SAlexander Motin length)); 137789b17223SAlexander Motin } 137889b17223SAlexander Motin 137989b17223SAlexander Motin static void 138089b17223SAlexander Motin g_raid_disk_done(struct bio *bp) 138189b17223SAlexander Motin { 138289b17223SAlexander Motin struct g_raid_softc *sc; 138389b17223SAlexander Motin struct g_raid_subdisk *sd; 138489b17223SAlexander Motin 138589b17223SAlexander Motin sd = bp->bio_caller1; 138689b17223SAlexander Motin sc = sd->sd_softc; 138789b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 138889b17223SAlexander Motin bioq_disksort(&sc->sc_queue, bp); 138989b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 139089b17223SAlexander Motin if (!dumping) 139189b17223SAlexander Motin wakeup(sc); 139289b17223SAlexander Motin } 139389b17223SAlexander Motin 139489b17223SAlexander Motin static void 139589b17223SAlexander Motin g_raid_disk_done_request(struct bio *bp) 139689b17223SAlexander Motin { 139789b17223SAlexander Motin struct g_raid_softc *sc; 139889b17223SAlexander Motin struct g_raid_disk *disk; 139989b17223SAlexander Motin struct g_raid_subdisk *sd; 140089b17223SAlexander Motin struct g_raid_volume *vol; 140189b17223SAlexander Motin 140289b17223SAlexander Motin g_topology_assert_not(); 140389b17223SAlexander Motin 140489b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); 140589b17223SAlexander Motin sd = bp->bio_caller1; 140689b17223SAlexander Motin sc = sd->sd_softc; 140789b17223SAlexander Motin vol = sd->sd_volume; 140889b17223SAlexander Motin if (bp->bio_from != NULL) { 140989b17223SAlexander Motin bp->bio_from->index--; 141089b17223SAlexander Motin disk = bp->bio_from->private; 141189b17223SAlexander Motin if (disk == NULL) 141289b17223SAlexander Motin g_raid_kill_consumer(sc, bp->bio_from); 141389b17223SAlexander Motin } 141489b17223SAlexander Motin bp->bio_offset -= sd->sd_offset; 141589b17223SAlexander Motin 141689b17223SAlexander Motin G_RAID_TR_IODONE(vol->v_tr, sd, bp); 141789b17223SAlexander Motin } 141889b17223SAlexander Motin 141989b17223SAlexander Motin static void 142089b17223SAlexander Motin g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) 142189b17223SAlexander Motin { 142289b17223SAlexander Motin 142389b17223SAlexander Motin if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) 142489b17223SAlexander Motin ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); 142589b17223SAlexander Motin else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) 142689b17223SAlexander Motin ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); 142789b17223SAlexander Motin else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) 142889b17223SAlexander Motin ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); 142989b17223SAlexander Motin else 143089b17223SAlexander Motin ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); 143189b17223SAlexander Motin if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { 143289b17223SAlexander Motin KASSERT(ep->e_error == 0, 143389b17223SAlexander Motin ("Error cannot be handled.")); 143489b17223SAlexander Motin g_raid_event_free(ep); 143589b17223SAlexander Motin } else { 143689b17223SAlexander Motin ep->e_flags |= G_RAID_EVENT_DONE; 143789b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); 143889b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 143989b17223SAlexander Motin wakeup(ep); 144089b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 144189b17223SAlexander Motin } 144289b17223SAlexander Motin } 144389b17223SAlexander Motin 144489b17223SAlexander Motin /* 144589b17223SAlexander Motin * Worker thread. 144689b17223SAlexander Motin */ 144789b17223SAlexander Motin static void 144889b17223SAlexander Motin g_raid_worker(void *arg) 144989b17223SAlexander Motin { 145089b17223SAlexander Motin struct g_raid_softc *sc; 145189b17223SAlexander Motin struct g_raid_event *ep; 145289b17223SAlexander Motin struct g_raid_volume *vol; 145389b17223SAlexander Motin struct bio *bp; 145489b17223SAlexander Motin struct timeval now, t; 145589b17223SAlexander Motin int timeout, rv; 145689b17223SAlexander Motin 145789b17223SAlexander Motin sc = arg; 145889b17223SAlexander Motin thread_lock(curthread); 145989b17223SAlexander Motin sched_prio(curthread, PRIBIO); 146089b17223SAlexander Motin thread_unlock(curthread); 146189b17223SAlexander Motin 146289b17223SAlexander Motin sx_xlock(&sc->sc_lock); 146389b17223SAlexander Motin for (;;) { 146489b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 146589b17223SAlexander Motin /* 146689b17223SAlexander Motin * First take a look at events. 146789b17223SAlexander Motin * This is important to handle events before any I/O requests. 146889b17223SAlexander Motin */ 146989b17223SAlexander Motin bp = NULL; 147089b17223SAlexander Motin vol = NULL; 147189b17223SAlexander Motin rv = 0; 147289b17223SAlexander Motin ep = TAILQ_FIRST(&sc->sc_events); 147389b17223SAlexander Motin if (ep != NULL) 147489b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_events, ep, e_next); 147589b17223SAlexander Motin else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) 147689b17223SAlexander Motin ; 147789b17223SAlexander Motin else { 147889b17223SAlexander Motin getmicrouptime(&now); 147989b17223SAlexander Motin t = now; 148089b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 148189b17223SAlexander Motin if (bioq_first(&vol->v_inflight) == NULL && 148289b17223SAlexander Motin vol->v_tr && 148389b17223SAlexander Motin timevalcmp(&vol->v_last_done, &t, < )) 148489b17223SAlexander Motin t = vol->v_last_done; 148589b17223SAlexander Motin } 148689b17223SAlexander Motin timevalsub(&t, &now); 148789b17223SAlexander Motin timeout = g_raid_idle_threshold + 148889b17223SAlexander Motin t.tv_sec * 1000000 + t.tv_usec; 148989b17223SAlexander Motin if (timeout > 0) { 149089b17223SAlexander Motin /* 149189b17223SAlexander Motin * Two steps to avoid overflows at HZ=1000 149289b17223SAlexander Motin * and idle timeouts > 2.1s. Some rounding 149389b17223SAlexander Motin * errors can occur, but they are < 1tick, 149489b17223SAlexander Motin * which is deemed to be close enough for 149589b17223SAlexander Motin * this purpose. 149689b17223SAlexander Motin */ 149789b17223SAlexander Motin int micpertic = 1000000 / hz; 149889b17223SAlexander Motin timeout = (timeout + micpertic - 1) / micpertic; 149989b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 150089b17223SAlexander Motin MSLEEP(rv, sc, &sc->sc_queue_mtx, 150189b17223SAlexander Motin PRIBIO | PDROP, "-", timeout); 150289b17223SAlexander Motin sx_xlock(&sc->sc_lock); 150389b17223SAlexander Motin goto process; 150489b17223SAlexander Motin } else 150589b17223SAlexander Motin rv = EWOULDBLOCK; 150689b17223SAlexander Motin } 150789b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 150889b17223SAlexander Motin process: 150989b17223SAlexander Motin if (ep != NULL) { 151089b17223SAlexander Motin g_raid_handle_event(sc, ep); 151189b17223SAlexander Motin } else if (bp != NULL) { 151289b17223SAlexander Motin if (bp->bio_to != NULL && 151389b17223SAlexander Motin bp->bio_to->geom == sc->sc_geom) 151489b17223SAlexander Motin g_raid_start_request(bp); 151589b17223SAlexander Motin else 151689b17223SAlexander Motin g_raid_disk_done_request(bp); 151789b17223SAlexander Motin } else if (rv == EWOULDBLOCK) { 151889b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 151989b17223SAlexander Motin if (vol->v_writes == 0 && vol->v_dirty) 152089b17223SAlexander Motin g_raid_clean(vol, -1); 152189b17223SAlexander Motin if (bioq_first(&vol->v_inflight) == NULL && 152289b17223SAlexander Motin vol->v_tr) { 152389b17223SAlexander Motin t.tv_sec = g_raid_idle_threshold / 1000000; 152489b17223SAlexander Motin t.tv_usec = g_raid_idle_threshold % 1000000; 152589b17223SAlexander Motin timevaladd(&t, &vol->v_last_done); 152689b17223SAlexander Motin getmicrouptime(&now); 152789b17223SAlexander Motin if (timevalcmp(&t, &now, <= )) { 152889b17223SAlexander Motin G_RAID_TR_IDLE(vol->v_tr); 152989b17223SAlexander Motin vol->v_last_done = now; 153089b17223SAlexander Motin } 153189b17223SAlexander Motin } 153289b17223SAlexander Motin } 153389b17223SAlexander Motin } 153489b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_HARD) 153589b17223SAlexander Motin g_raid_destroy_node(sc, 1); /* May not return. */ 153689b17223SAlexander Motin } 153789b17223SAlexander Motin } 153889b17223SAlexander Motin 153989b17223SAlexander Motin static void 154089b17223SAlexander Motin g_raid_poll(struct g_raid_softc *sc) 154189b17223SAlexander Motin { 154289b17223SAlexander Motin struct g_raid_event *ep; 154389b17223SAlexander Motin struct bio *bp; 154489b17223SAlexander Motin 154589b17223SAlexander Motin sx_xlock(&sc->sc_lock); 154689b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 154789b17223SAlexander Motin /* 154889b17223SAlexander Motin * First take a look at events. 154989b17223SAlexander Motin * This is important to handle events before any I/O requests. 155089b17223SAlexander Motin */ 155189b17223SAlexander Motin ep = TAILQ_FIRST(&sc->sc_events); 155289b17223SAlexander Motin if (ep != NULL) { 155389b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_events, ep, e_next); 155489b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 155589b17223SAlexander Motin g_raid_handle_event(sc, ep); 155689b17223SAlexander Motin goto out; 155789b17223SAlexander Motin } 155889b17223SAlexander Motin bp = bioq_takefirst(&sc->sc_queue); 155989b17223SAlexander Motin if (bp != NULL) { 156089b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 156189b17223SAlexander Motin if (bp->bio_from == NULL || 156289b17223SAlexander Motin bp->bio_from->geom != sc->sc_geom) 156389b17223SAlexander Motin g_raid_start_request(bp); 156489b17223SAlexander Motin else 156589b17223SAlexander Motin g_raid_disk_done_request(bp); 156689b17223SAlexander Motin } 156789b17223SAlexander Motin out: 156889b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 156989b17223SAlexander Motin } 157089b17223SAlexander Motin 157189b17223SAlexander Motin static void 157289b17223SAlexander Motin g_raid_launch_provider(struct g_raid_volume *vol) 157389b17223SAlexander Motin { 157489b17223SAlexander Motin struct g_raid_disk *disk; 157589b17223SAlexander Motin struct g_raid_softc *sc; 157689b17223SAlexander Motin struct g_provider *pp; 157789b17223SAlexander Motin char name[G_RAID_MAX_VOLUMENAME]; 157889b17223SAlexander Motin off_t off; 157989b17223SAlexander Motin 158089b17223SAlexander Motin sc = vol->v_softc; 158189b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 158289b17223SAlexander Motin 158389b17223SAlexander Motin g_topology_lock(); 158489b17223SAlexander Motin /* Try to name provider with volume name. */ 158589b17223SAlexander Motin snprintf(name, sizeof(name), "raid/%s", vol->v_name); 158689b17223SAlexander Motin if (g_raid_name_format == 0 || vol->v_name[0] == 0 || 158789b17223SAlexander Motin g_provider_by_name(name) != NULL) { 158889b17223SAlexander Motin /* Otherwise use sequential volume number. */ 158989b17223SAlexander Motin snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); 159089b17223SAlexander Motin } 159189b17223SAlexander Motin pp = g_new_providerf(sc->sc_geom, "%s", name); 159289b17223SAlexander Motin pp->private = vol; 159389b17223SAlexander Motin pp->mediasize = vol->v_mediasize; 159489b17223SAlexander Motin pp->sectorsize = vol->v_sectorsize; 159589b17223SAlexander Motin pp->stripesize = 0; 159689b17223SAlexander Motin pp->stripeoffset = 0; 159789b17223SAlexander Motin if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 159889b17223SAlexander Motin vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || 159989b17223SAlexander Motin vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || 160089b17223SAlexander Motin vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { 160189b17223SAlexander Motin if ((disk = vol->v_subdisks[0].sd_disk) != NULL && 160289b17223SAlexander Motin disk->d_consumer != NULL && 160389b17223SAlexander Motin disk->d_consumer->provider != NULL) { 160489b17223SAlexander Motin pp->stripesize = disk->d_consumer->provider->stripesize; 160589b17223SAlexander Motin off = disk->d_consumer->provider->stripeoffset; 160689b17223SAlexander Motin pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; 160789b17223SAlexander Motin if (off > 0) 160889b17223SAlexander Motin pp->stripeoffset %= off; 160989b17223SAlexander Motin } 161089b17223SAlexander Motin if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { 161189b17223SAlexander Motin pp->stripesize *= (vol->v_disks_count - 1); 161289b17223SAlexander Motin pp->stripeoffset *= (vol->v_disks_count - 1); 161389b17223SAlexander Motin } 161489b17223SAlexander Motin } else 161589b17223SAlexander Motin pp->stripesize = vol->v_strip_size; 161689b17223SAlexander Motin vol->v_provider = pp; 161789b17223SAlexander Motin g_error_provider(pp, 0); 161889b17223SAlexander Motin g_topology_unlock(); 161989b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", 162089b17223SAlexander Motin pp->name, vol->v_name); 162189b17223SAlexander Motin } 162289b17223SAlexander Motin 162389b17223SAlexander Motin static void 162489b17223SAlexander Motin g_raid_destroy_provider(struct g_raid_volume *vol) 162589b17223SAlexander Motin { 162689b17223SAlexander Motin struct g_raid_softc *sc; 162789b17223SAlexander Motin struct g_provider *pp; 162889b17223SAlexander Motin struct bio *bp, *tmp; 162989b17223SAlexander Motin 163089b17223SAlexander Motin g_topology_assert_not(); 163189b17223SAlexander Motin sc = vol->v_softc; 163289b17223SAlexander Motin pp = vol->v_provider; 163389b17223SAlexander Motin KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); 163489b17223SAlexander Motin 163589b17223SAlexander Motin g_topology_lock(); 163689b17223SAlexander Motin g_error_provider(pp, ENXIO); 163789b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 163889b17223SAlexander Motin TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { 163989b17223SAlexander Motin if (bp->bio_to != pp) 164089b17223SAlexander Motin continue; 164189b17223SAlexander Motin bioq_remove(&sc->sc_queue, bp); 164289b17223SAlexander Motin g_io_deliver(bp, ENXIO); 164389b17223SAlexander Motin } 164489b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 164589b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", 164689b17223SAlexander Motin pp->name, vol->v_name); 164789b17223SAlexander Motin g_wither_provider(pp, ENXIO); 164889b17223SAlexander Motin g_topology_unlock(); 164989b17223SAlexander Motin vol->v_provider = NULL; 165089b17223SAlexander Motin } 165189b17223SAlexander Motin 165289b17223SAlexander Motin /* 165389b17223SAlexander Motin * Update device state. 165489b17223SAlexander Motin */ 165589b17223SAlexander Motin static int 165689b17223SAlexander Motin g_raid_update_volume(struct g_raid_volume *vol, u_int event) 165789b17223SAlexander Motin { 165889b17223SAlexander Motin struct g_raid_softc *sc; 165989b17223SAlexander Motin 166089b17223SAlexander Motin sc = vol->v_softc; 166189b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 166289b17223SAlexander Motin 166389b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", 166489b17223SAlexander Motin g_raid_volume_event2str(event), 166589b17223SAlexander Motin vol->v_name); 166689b17223SAlexander Motin switch (event) { 166789b17223SAlexander Motin case G_RAID_VOLUME_E_DOWN: 166889b17223SAlexander Motin if (vol->v_provider != NULL) 166989b17223SAlexander Motin g_raid_destroy_provider(vol); 167089b17223SAlexander Motin break; 167189b17223SAlexander Motin case G_RAID_VOLUME_E_UP: 167289b17223SAlexander Motin if (vol->v_provider == NULL) 167389b17223SAlexander Motin g_raid_launch_provider(vol); 167489b17223SAlexander Motin break; 167589b17223SAlexander Motin case G_RAID_VOLUME_E_START: 167689b17223SAlexander Motin if (vol->v_tr) 167789b17223SAlexander Motin G_RAID_TR_START(vol->v_tr); 167889b17223SAlexander Motin return (0); 167989b17223SAlexander Motin default: 168089b17223SAlexander Motin if (sc->sc_md) 168189b17223SAlexander Motin G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); 168289b17223SAlexander Motin return (0); 168389b17223SAlexander Motin } 168489b17223SAlexander Motin 168589b17223SAlexander Motin /* Manage root mount release. */ 168689b17223SAlexander Motin if (vol->v_starting) { 168789b17223SAlexander Motin vol->v_starting = 0; 168889b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); 168989b17223SAlexander Motin root_mount_rel(vol->v_rootmount); 169089b17223SAlexander Motin vol->v_rootmount = NULL; 169189b17223SAlexander Motin } 169289b17223SAlexander Motin if (vol->v_stopping && vol->v_provider_open == 0) 169389b17223SAlexander Motin g_raid_destroy_volume(vol); 169489b17223SAlexander Motin return (0); 169589b17223SAlexander Motin } 169689b17223SAlexander Motin 169789b17223SAlexander Motin /* 169889b17223SAlexander Motin * Update subdisk state. 169989b17223SAlexander Motin */ 170089b17223SAlexander Motin static int 170189b17223SAlexander Motin g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) 170289b17223SAlexander Motin { 170389b17223SAlexander Motin struct g_raid_softc *sc; 170489b17223SAlexander Motin struct g_raid_volume *vol; 170589b17223SAlexander Motin 170689b17223SAlexander Motin sc = sd->sd_softc; 170789b17223SAlexander Motin vol = sd->sd_volume; 170889b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 170989b17223SAlexander Motin 171089b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", 171189b17223SAlexander Motin g_raid_subdisk_event2str(event), 171289b17223SAlexander Motin vol->v_name, sd->sd_pos, 171389b17223SAlexander Motin sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); 171489b17223SAlexander Motin if (vol->v_tr) 171589b17223SAlexander Motin G_RAID_TR_EVENT(vol->v_tr, sd, event); 171689b17223SAlexander Motin 171789b17223SAlexander Motin return (0); 171889b17223SAlexander Motin } 171989b17223SAlexander Motin 172089b17223SAlexander Motin /* 172189b17223SAlexander Motin * Update disk state. 172289b17223SAlexander Motin */ 172389b17223SAlexander Motin static int 172489b17223SAlexander Motin g_raid_update_disk(struct g_raid_disk *disk, u_int event) 172589b17223SAlexander Motin { 172689b17223SAlexander Motin struct g_raid_softc *sc; 172789b17223SAlexander Motin 172889b17223SAlexander Motin sc = disk->d_softc; 172989b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 173089b17223SAlexander Motin 173189b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", 173289b17223SAlexander Motin g_raid_disk_event2str(event), 173389b17223SAlexander Motin g_raid_get_diskname(disk)); 173489b17223SAlexander Motin 173589b17223SAlexander Motin if (sc->sc_md) 173689b17223SAlexander Motin G_RAID_MD_EVENT(sc->sc_md, disk, event); 173789b17223SAlexander Motin return (0); 173889b17223SAlexander Motin } 173989b17223SAlexander Motin 174089b17223SAlexander Motin /* 174189b17223SAlexander Motin * Node event. 174289b17223SAlexander Motin */ 174389b17223SAlexander Motin static int 174489b17223SAlexander Motin g_raid_update_node(struct g_raid_softc *sc, u_int event) 174589b17223SAlexander Motin { 174689b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 174789b17223SAlexander Motin 174889b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for the array.", 174989b17223SAlexander Motin g_raid_node_event2str(event)); 175089b17223SAlexander Motin 175189b17223SAlexander Motin if (event == G_RAID_NODE_E_WAKE) 175289b17223SAlexander Motin return (0); 175389b17223SAlexander Motin if (sc->sc_md) 175489b17223SAlexander Motin G_RAID_MD_EVENT(sc->sc_md, NULL, event); 175589b17223SAlexander Motin return (0); 175689b17223SAlexander Motin } 175789b17223SAlexander Motin 175889b17223SAlexander Motin static int 175989b17223SAlexander Motin g_raid_access(struct g_provider *pp, int acr, int acw, int ace) 176089b17223SAlexander Motin { 176189b17223SAlexander Motin struct g_raid_volume *vol; 176289b17223SAlexander Motin struct g_raid_softc *sc; 176314e2cd0aSAlexander Motin int dcw, opens, error = 0; 176489b17223SAlexander Motin 176589b17223SAlexander Motin g_topology_assert(); 176689b17223SAlexander Motin sc = pp->geom->softc; 176789b17223SAlexander Motin vol = pp->private; 176889b17223SAlexander Motin KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 176989b17223SAlexander Motin KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); 177089b17223SAlexander Motin 177189b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, 177289b17223SAlexander Motin acr, acw, ace); 177389b17223SAlexander Motin dcw = pp->acw + acw; 177489b17223SAlexander Motin 177589b17223SAlexander Motin g_topology_unlock(); 177689b17223SAlexander Motin sx_xlock(&sc->sc_lock); 177789b17223SAlexander Motin /* Deny new opens while dying. */ 177889b17223SAlexander Motin if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { 177989b17223SAlexander Motin error = ENXIO; 178089b17223SAlexander Motin goto out; 178189b17223SAlexander Motin } 178289b17223SAlexander Motin if (dcw == 0 && vol->v_dirty) 178389b17223SAlexander Motin g_raid_clean(vol, dcw); 178489b17223SAlexander Motin vol->v_provider_open += acr + acw + ace; 178589b17223SAlexander Motin /* Handle delayed node destruction. */ 178689b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && 178789b17223SAlexander Motin vol->v_provider_open == 0) { 178889b17223SAlexander Motin /* Count open volumes. */ 178989b17223SAlexander Motin opens = g_raid_nopens(sc); 179089b17223SAlexander Motin if (opens == 0) { 179189b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_HARD; 179289b17223SAlexander Motin /* Wake up worker to make it selfdestruct. */ 179389b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 179489b17223SAlexander Motin } 179589b17223SAlexander Motin } 179689b17223SAlexander Motin /* Handle open volume destruction. */ 179789b17223SAlexander Motin if (vol->v_stopping && vol->v_provider_open == 0) 179889b17223SAlexander Motin g_raid_destroy_volume(vol); 179989b17223SAlexander Motin out: 180089b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 180189b17223SAlexander Motin g_topology_lock(); 180289b17223SAlexander Motin return (error); 180389b17223SAlexander Motin } 180489b17223SAlexander Motin 180589b17223SAlexander Motin struct g_raid_softc * 180689b17223SAlexander Motin g_raid_create_node(struct g_class *mp, 180789b17223SAlexander Motin const char *name, struct g_raid_md_object *md) 180889b17223SAlexander Motin { 180989b17223SAlexander Motin struct g_raid_softc *sc; 181089b17223SAlexander Motin struct g_geom *gp; 181189b17223SAlexander Motin int error; 181289b17223SAlexander Motin 181389b17223SAlexander Motin g_topology_assert(); 181489b17223SAlexander Motin G_RAID_DEBUG(1, "Creating array %s.", name); 181589b17223SAlexander Motin 181689b17223SAlexander Motin gp = g_new_geomf(mp, "%s", name); 181789b17223SAlexander Motin sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); 181889b17223SAlexander Motin gp->start = g_raid_start; 181989b17223SAlexander Motin gp->orphan = g_raid_orphan; 182089b17223SAlexander Motin gp->access = g_raid_access; 182189b17223SAlexander Motin gp->dumpconf = g_raid_dumpconf; 182289b17223SAlexander Motin 182389b17223SAlexander Motin sc->sc_md = md; 182489b17223SAlexander Motin sc->sc_geom = gp; 182589b17223SAlexander Motin sc->sc_flags = 0; 182689b17223SAlexander Motin TAILQ_INIT(&sc->sc_volumes); 182789b17223SAlexander Motin TAILQ_INIT(&sc->sc_disks); 1828c9f545e5SAlexander Motin sx_init(&sc->sc_lock, "graid:lock"); 1829c9f545e5SAlexander Motin mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF); 183089b17223SAlexander Motin TAILQ_INIT(&sc->sc_events); 183189b17223SAlexander Motin bioq_init(&sc->sc_queue); 183289b17223SAlexander Motin gp->softc = sc; 183389b17223SAlexander Motin error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, 183489b17223SAlexander Motin "g_raid %s", name); 183589b17223SAlexander Motin if (error != 0) { 183689b17223SAlexander Motin G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); 183789b17223SAlexander Motin mtx_destroy(&sc->sc_queue_mtx); 183889b17223SAlexander Motin sx_destroy(&sc->sc_lock); 183989b17223SAlexander Motin g_destroy_geom(sc->sc_geom); 184089b17223SAlexander Motin free(sc, M_RAID); 184189b17223SAlexander Motin return (NULL); 184289b17223SAlexander Motin } 184389b17223SAlexander Motin 184489b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Array %s created.", name); 184589b17223SAlexander Motin return (sc); 184689b17223SAlexander Motin } 184789b17223SAlexander Motin 184889b17223SAlexander Motin struct g_raid_volume * 184989b17223SAlexander Motin g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) 185089b17223SAlexander Motin { 185189b17223SAlexander Motin struct g_raid_volume *vol, *vol1; 185289b17223SAlexander Motin int i; 185389b17223SAlexander Motin 185489b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); 185589b17223SAlexander Motin vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); 185689b17223SAlexander Motin vol->v_softc = sc; 185789b17223SAlexander Motin strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); 185889b17223SAlexander Motin vol->v_state = G_RAID_VOLUME_S_STARTING; 185989b17223SAlexander Motin vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 186089b17223SAlexander Motin vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; 18618f12ca2eSAlexander Motin vol->v_rotate_parity = 1; 186289b17223SAlexander Motin bioq_init(&vol->v_inflight); 186389b17223SAlexander Motin bioq_init(&vol->v_locked); 186489b17223SAlexander Motin LIST_INIT(&vol->v_locks); 186589b17223SAlexander Motin for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 186689b17223SAlexander Motin vol->v_subdisks[i].sd_softc = sc; 186789b17223SAlexander Motin vol->v_subdisks[i].sd_volume = vol; 186889b17223SAlexander Motin vol->v_subdisks[i].sd_pos = i; 186989b17223SAlexander Motin vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; 187089b17223SAlexander Motin } 187189b17223SAlexander Motin 187289b17223SAlexander Motin /* Find free ID for this volume. */ 187389b17223SAlexander Motin g_topology_lock(); 187489b17223SAlexander Motin vol1 = vol; 187589b17223SAlexander Motin if (id >= 0) { 187689b17223SAlexander Motin LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 187789b17223SAlexander Motin if (vol1->v_global_id == id) 187889b17223SAlexander Motin break; 187989b17223SAlexander Motin } 188089b17223SAlexander Motin } 188189b17223SAlexander Motin if (vol1 != NULL) { 188289b17223SAlexander Motin for (id = 0; ; id++) { 188389b17223SAlexander Motin LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 188489b17223SAlexander Motin if (vol1->v_global_id == id) 188589b17223SAlexander Motin break; 188689b17223SAlexander Motin } 188789b17223SAlexander Motin if (vol1 == NULL) 188889b17223SAlexander Motin break; 188989b17223SAlexander Motin } 189089b17223SAlexander Motin } 189189b17223SAlexander Motin vol->v_global_id = id; 189289b17223SAlexander Motin LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); 189389b17223SAlexander Motin g_topology_unlock(); 189489b17223SAlexander Motin 189589b17223SAlexander Motin /* Delay root mounting. */ 189689b17223SAlexander Motin vol->v_rootmount = root_mount_hold("GRAID"); 189789b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); 189889b17223SAlexander Motin vol->v_starting = 1; 189989b17223SAlexander Motin TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); 190089b17223SAlexander Motin return (vol); 190189b17223SAlexander Motin } 190289b17223SAlexander Motin 190389b17223SAlexander Motin struct g_raid_disk * 190489b17223SAlexander Motin g_raid_create_disk(struct g_raid_softc *sc) 190589b17223SAlexander Motin { 190689b17223SAlexander Motin struct g_raid_disk *disk; 190789b17223SAlexander Motin 190889b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Creating disk."); 190989b17223SAlexander Motin disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); 191089b17223SAlexander Motin disk->d_softc = sc; 191189b17223SAlexander Motin disk->d_state = G_RAID_DISK_S_NONE; 191289b17223SAlexander Motin TAILQ_INIT(&disk->d_subdisks); 191389b17223SAlexander Motin TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); 191489b17223SAlexander Motin return (disk); 191589b17223SAlexander Motin } 191689b17223SAlexander Motin 191789b17223SAlexander Motin int g_raid_start_volume(struct g_raid_volume *vol) 191889b17223SAlexander Motin { 191989b17223SAlexander Motin struct g_raid_tr_class *class; 192089b17223SAlexander Motin struct g_raid_tr_object *obj; 192189b17223SAlexander Motin int status; 192289b17223SAlexander Motin 192389b17223SAlexander Motin G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); 192489b17223SAlexander Motin LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { 192589b17223SAlexander Motin G_RAID_DEBUG1(2, vol->v_softc, 192689b17223SAlexander Motin "Tasting volume %s for %s transformation.", 192789b17223SAlexander Motin vol->v_name, class->name); 192889b17223SAlexander Motin obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 192989b17223SAlexander Motin M_WAITOK); 193089b17223SAlexander Motin obj->tro_class = class; 193189b17223SAlexander Motin obj->tro_volume = vol; 193289b17223SAlexander Motin status = G_RAID_TR_TASTE(obj, vol); 193389b17223SAlexander Motin if (status != G_RAID_TR_TASTE_FAIL) 193489b17223SAlexander Motin break; 193589b17223SAlexander Motin kobj_delete((kobj_t)obj, M_RAID); 193689b17223SAlexander Motin } 193789b17223SAlexander Motin if (class == NULL) { 193889b17223SAlexander Motin G_RAID_DEBUG1(0, vol->v_softc, 193989b17223SAlexander Motin "No transformation module found for %s.", 194089b17223SAlexander Motin vol->v_name); 194189b17223SAlexander Motin vol->v_tr = NULL; 194289b17223SAlexander Motin g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); 194389b17223SAlexander Motin g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, 194489b17223SAlexander Motin G_RAID_EVENT_VOLUME); 194589b17223SAlexander Motin return (-1); 194689b17223SAlexander Motin } 194789b17223SAlexander Motin G_RAID_DEBUG1(2, vol->v_softc, 194889b17223SAlexander Motin "Transformation module %s chosen for %s.", 194989b17223SAlexander Motin class->name, vol->v_name); 195089b17223SAlexander Motin vol->v_tr = obj; 195189b17223SAlexander Motin return (0); 195289b17223SAlexander Motin } 195389b17223SAlexander Motin 195489b17223SAlexander Motin int 195589b17223SAlexander Motin g_raid_destroy_node(struct g_raid_softc *sc, int worker) 195689b17223SAlexander Motin { 195789b17223SAlexander Motin struct g_raid_volume *vol, *tmpv; 195889b17223SAlexander Motin struct g_raid_disk *disk, *tmpd; 195989b17223SAlexander Motin int error = 0; 196089b17223SAlexander Motin 196189b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_HARD; 196289b17223SAlexander Motin TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { 196389b17223SAlexander Motin if (g_raid_destroy_volume(vol)) 196489b17223SAlexander Motin error = EBUSY; 196589b17223SAlexander Motin } 196689b17223SAlexander Motin if (error) 196789b17223SAlexander Motin return (error); 196889b17223SAlexander Motin TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { 196989b17223SAlexander Motin if (g_raid_destroy_disk(disk)) 197089b17223SAlexander Motin error = EBUSY; 197189b17223SAlexander Motin } 197289b17223SAlexander Motin if (error) 197389b17223SAlexander Motin return (error); 197489b17223SAlexander Motin if (sc->sc_md) { 197589b17223SAlexander Motin G_RAID_MD_FREE(sc->sc_md); 197689b17223SAlexander Motin kobj_delete((kobj_t)sc->sc_md, M_RAID); 197789b17223SAlexander Motin sc->sc_md = NULL; 197889b17223SAlexander Motin } 197989b17223SAlexander Motin if (sc->sc_geom != NULL) { 198089b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); 198189b17223SAlexander Motin g_topology_lock(); 198289b17223SAlexander Motin sc->sc_geom->softc = NULL; 198389b17223SAlexander Motin g_wither_geom(sc->sc_geom, ENXIO); 198489b17223SAlexander Motin g_topology_unlock(); 198589b17223SAlexander Motin sc->sc_geom = NULL; 198689b17223SAlexander Motin } else 198789b17223SAlexander Motin G_RAID_DEBUG(1, "Array destroyed."); 198889b17223SAlexander Motin if (worker) { 198989b17223SAlexander Motin g_raid_event_cancel(sc, sc); 199089b17223SAlexander Motin mtx_destroy(&sc->sc_queue_mtx); 199189b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 199289b17223SAlexander Motin sx_destroy(&sc->sc_lock); 199389b17223SAlexander Motin wakeup(&sc->sc_stopping); 199489b17223SAlexander Motin free(sc, M_RAID); 199589b17223SAlexander Motin curthread->td_pflags &= ~TDP_GEOM; 199689b17223SAlexander Motin G_RAID_DEBUG(1, "Thread exiting."); 199789b17223SAlexander Motin kproc_exit(0); 199889b17223SAlexander Motin } else { 199989b17223SAlexander Motin /* Wake up worker to make it selfdestruct. */ 200089b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 200189b17223SAlexander Motin } 200289b17223SAlexander Motin return (0); 200389b17223SAlexander Motin } 200489b17223SAlexander Motin 200589b17223SAlexander Motin int 200689b17223SAlexander Motin g_raid_destroy_volume(struct g_raid_volume *vol) 200789b17223SAlexander Motin { 200889b17223SAlexander Motin struct g_raid_softc *sc; 200989b17223SAlexander Motin struct g_raid_disk *disk; 201089b17223SAlexander Motin int i; 201189b17223SAlexander Motin 201289b17223SAlexander Motin sc = vol->v_softc; 201389b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); 201489b17223SAlexander Motin vol->v_stopping = 1; 201589b17223SAlexander Motin if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { 201689b17223SAlexander Motin if (vol->v_tr) { 201789b17223SAlexander Motin G_RAID_TR_STOP(vol->v_tr); 201889b17223SAlexander Motin return (EBUSY); 201989b17223SAlexander Motin } else 202089b17223SAlexander Motin vol->v_state = G_RAID_VOLUME_S_STOPPED; 202189b17223SAlexander Motin } 202289b17223SAlexander Motin if (g_raid_event_check(sc, vol) != 0) 202389b17223SAlexander Motin return (EBUSY); 202489b17223SAlexander Motin if (vol->v_provider != NULL) 202589b17223SAlexander Motin return (EBUSY); 202689b17223SAlexander Motin if (vol->v_provider_open != 0) 202789b17223SAlexander Motin return (EBUSY); 202889b17223SAlexander Motin if (vol->v_tr) { 202989b17223SAlexander Motin G_RAID_TR_FREE(vol->v_tr); 203089b17223SAlexander Motin kobj_delete((kobj_t)vol->v_tr, M_RAID); 203189b17223SAlexander Motin vol->v_tr = NULL; 203289b17223SAlexander Motin } 203389b17223SAlexander Motin if (vol->v_rootmount) 203489b17223SAlexander Motin root_mount_rel(vol->v_rootmount); 203589b17223SAlexander Motin g_topology_lock(); 203689b17223SAlexander Motin LIST_REMOVE(vol, v_global_next); 203789b17223SAlexander Motin g_topology_unlock(); 203889b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); 203989b17223SAlexander Motin for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 204089b17223SAlexander Motin g_raid_event_cancel(sc, &vol->v_subdisks[i]); 204189b17223SAlexander Motin disk = vol->v_subdisks[i].sd_disk; 204289b17223SAlexander Motin if (disk == NULL) 204389b17223SAlexander Motin continue; 204489b17223SAlexander Motin TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); 204589b17223SAlexander Motin } 204689b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); 204789b17223SAlexander Motin if (sc->sc_md) 204889b17223SAlexander Motin G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); 204989b17223SAlexander Motin g_raid_event_cancel(sc, vol); 205089b17223SAlexander Motin free(vol, M_RAID); 205189b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_HARD) { 205289b17223SAlexander Motin /* Wake up worker to let it selfdestruct. */ 205389b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 205489b17223SAlexander Motin } 205589b17223SAlexander Motin return (0); 205689b17223SAlexander Motin } 205789b17223SAlexander Motin 205889b17223SAlexander Motin int 205989b17223SAlexander Motin g_raid_destroy_disk(struct g_raid_disk *disk) 206089b17223SAlexander Motin { 206189b17223SAlexander Motin struct g_raid_softc *sc; 206289b17223SAlexander Motin struct g_raid_subdisk *sd, *tmp; 206389b17223SAlexander Motin 206489b17223SAlexander Motin sc = disk->d_softc; 206589b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Destroying disk."); 206689b17223SAlexander Motin if (disk->d_consumer) { 206789b17223SAlexander Motin g_raid_kill_consumer(sc, disk->d_consumer); 206889b17223SAlexander Motin disk->d_consumer = NULL; 206989b17223SAlexander Motin } 207089b17223SAlexander Motin TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { 207189b17223SAlexander Motin g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); 207289b17223SAlexander Motin g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 207389b17223SAlexander Motin G_RAID_EVENT_SUBDISK); 207489b17223SAlexander Motin TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); 207589b17223SAlexander Motin sd->sd_disk = NULL; 207689b17223SAlexander Motin } 207789b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_disks, disk, d_next); 207889b17223SAlexander Motin if (sc->sc_md) 207989b17223SAlexander Motin G_RAID_MD_FREE_DISK(sc->sc_md, disk); 208089b17223SAlexander Motin g_raid_event_cancel(sc, disk); 208189b17223SAlexander Motin free(disk, M_RAID); 208289b17223SAlexander Motin return (0); 208389b17223SAlexander Motin } 208489b17223SAlexander Motin 208589b17223SAlexander Motin int 208689b17223SAlexander Motin g_raid_destroy(struct g_raid_softc *sc, int how) 208789b17223SAlexander Motin { 208889b17223SAlexander Motin int opens; 208989b17223SAlexander Motin 209089b17223SAlexander Motin g_topology_assert_not(); 209189b17223SAlexander Motin if (sc == NULL) 209289b17223SAlexander Motin return (ENXIO); 209389b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 209489b17223SAlexander Motin 209589b17223SAlexander Motin /* Count open volumes. */ 209689b17223SAlexander Motin opens = g_raid_nopens(sc); 209789b17223SAlexander Motin 209889b17223SAlexander Motin /* React on some opened volumes. */ 209989b17223SAlexander Motin if (opens > 0) { 210089b17223SAlexander Motin switch (how) { 210189b17223SAlexander Motin case G_RAID_DESTROY_SOFT: 210289b17223SAlexander Motin G_RAID_DEBUG1(1, sc, 210389b17223SAlexander Motin "%d volumes are still open.", 210489b17223SAlexander Motin opens); 210589b17223SAlexander Motin return (EBUSY); 210689b17223SAlexander Motin case G_RAID_DESTROY_DELAYED: 210789b17223SAlexander Motin G_RAID_DEBUG1(1, sc, 210889b17223SAlexander Motin "Array will be destroyed on last close."); 210989b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_DELAYED; 211089b17223SAlexander Motin return (EBUSY); 211189b17223SAlexander Motin case G_RAID_DESTROY_HARD: 211289b17223SAlexander Motin G_RAID_DEBUG1(1, sc, 211389b17223SAlexander Motin "%d volumes are still open.", 211489b17223SAlexander Motin opens); 211589b17223SAlexander Motin } 211689b17223SAlexander Motin } 211789b17223SAlexander Motin 211889b17223SAlexander Motin /* Mark node for destruction. */ 211989b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_HARD; 212089b17223SAlexander Motin /* Wake up worker to let it selfdestruct. */ 212189b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 212289b17223SAlexander Motin /* Sleep until node destroyed. */ 212389b17223SAlexander Motin sx_sleep(&sc->sc_stopping, &sc->sc_lock, 212489b17223SAlexander Motin PRIBIO | PDROP, "r:destroy", 0); 212589b17223SAlexander Motin return (0); 212689b17223SAlexander Motin } 212789b17223SAlexander Motin 212889b17223SAlexander Motin static void 212989b17223SAlexander Motin g_raid_taste_orphan(struct g_consumer *cp) 213089b17223SAlexander Motin { 213189b17223SAlexander Motin 213289b17223SAlexander Motin KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 213389b17223SAlexander Motin cp->provider->name)); 213489b17223SAlexander Motin } 213589b17223SAlexander Motin 213689b17223SAlexander Motin static struct g_geom * 213789b17223SAlexander Motin g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 213889b17223SAlexander Motin { 213989b17223SAlexander Motin struct g_consumer *cp; 214089b17223SAlexander Motin struct g_geom *gp, *geom; 214189b17223SAlexander Motin struct g_raid_md_class *class; 214289b17223SAlexander Motin struct g_raid_md_object *obj; 214389b17223SAlexander Motin int status; 214489b17223SAlexander Motin 214589b17223SAlexander Motin g_topology_assert(); 214689b17223SAlexander Motin g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 214789b17223SAlexander Motin G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); 214889b17223SAlexander Motin 21498df8e26aSAlexander Motin gp = g_new_geomf(mp, "raid:taste"); 215089b17223SAlexander Motin /* 215189b17223SAlexander Motin * This orphan function should be never called. 215289b17223SAlexander Motin */ 215389b17223SAlexander Motin gp->orphan = g_raid_taste_orphan; 215489b17223SAlexander Motin cp = g_new_consumer(gp); 215589b17223SAlexander Motin g_attach(cp, pp); 215689b17223SAlexander Motin 215789b17223SAlexander Motin geom = NULL; 215889b17223SAlexander Motin LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 215989b17223SAlexander Motin G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", 216089b17223SAlexander Motin pp->name, class->name); 216189b17223SAlexander Motin obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 216289b17223SAlexander Motin M_WAITOK); 216389b17223SAlexander Motin obj->mdo_class = class; 216489b17223SAlexander Motin status = G_RAID_MD_TASTE(obj, mp, cp, &geom); 216589b17223SAlexander Motin if (status != G_RAID_MD_TASTE_NEW) 216689b17223SAlexander Motin kobj_delete((kobj_t)obj, M_RAID); 216789b17223SAlexander Motin if (status != G_RAID_MD_TASTE_FAIL) 216889b17223SAlexander Motin break; 216989b17223SAlexander Motin } 217089b17223SAlexander Motin 217189b17223SAlexander Motin g_detach(cp); 217289b17223SAlexander Motin g_destroy_consumer(cp); 217389b17223SAlexander Motin g_destroy_geom(gp); 217489b17223SAlexander Motin G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); 217589b17223SAlexander Motin return (geom); 217689b17223SAlexander Motin } 217789b17223SAlexander Motin 217889b17223SAlexander Motin int 21798df8e26aSAlexander Motin g_raid_create_node_format(const char *format, struct gctl_req *req, 21808df8e26aSAlexander Motin struct g_geom **gp) 218189b17223SAlexander Motin { 218289b17223SAlexander Motin struct g_raid_md_class *class; 218389b17223SAlexander Motin struct g_raid_md_object *obj; 218489b17223SAlexander Motin int status; 218589b17223SAlexander Motin 218689b17223SAlexander Motin G_RAID_DEBUG(2, "Creating array for %s metadata.", format); 218789b17223SAlexander Motin LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 218889b17223SAlexander Motin if (strcasecmp(class->name, format) == 0) 218989b17223SAlexander Motin break; 219089b17223SAlexander Motin } 219189b17223SAlexander Motin if (class == NULL) { 219289b17223SAlexander Motin G_RAID_DEBUG(1, "No support for %s metadata.", format); 219389b17223SAlexander Motin return (G_RAID_MD_TASTE_FAIL); 219489b17223SAlexander Motin } 219589b17223SAlexander Motin obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 219689b17223SAlexander Motin M_WAITOK); 219789b17223SAlexander Motin obj->mdo_class = class; 21988df8e26aSAlexander Motin status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp); 219989b17223SAlexander Motin if (status != G_RAID_MD_TASTE_NEW) 220089b17223SAlexander Motin kobj_delete((kobj_t)obj, M_RAID); 220189b17223SAlexander Motin return (status); 220289b17223SAlexander Motin } 220389b17223SAlexander Motin 220489b17223SAlexander Motin static int 220589b17223SAlexander Motin g_raid_destroy_geom(struct gctl_req *req __unused, 220689b17223SAlexander Motin struct g_class *mp __unused, struct g_geom *gp) 220789b17223SAlexander Motin { 220889b17223SAlexander Motin struct g_raid_softc *sc; 220989b17223SAlexander Motin int error; 221089b17223SAlexander Motin 221189b17223SAlexander Motin g_topology_unlock(); 221289b17223SAlexander Motin sc = gp->softc; 221389b17223SAlexander Motin sx_xlock(&sc->sc_lock); 221489b17223SAlexander Motin g_cancel_event(sc); 221589b17223SAlexander Motin error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); 221689b17223SAlexander Motin if (error != 0) 221789b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 221889b17223SAlexander Motin g_topology_lock(); 221989b17223SAlexander Motin return (error); 222089b17223SAlexander Motin } 222189b17223SAlexander Motin 222289b17223SAlexander Motin void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, 222389b17223SAlexander Motin struct g_raid_subdisk *sd, struct g_raid_disk *disk) 222489b17223SAlexander Motin { 222589b17223SAlexander Motin 222689b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_HARD) 222789b17223SAlexander Motin return; 222889b17223SAlexander Motin if (sc->sc_md) 222989b17223SAlexander Motin G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); 223089b17223SAlexander Motin } 223189b17223SAlexander Motin 223289b17223SAlexander Motin void g_raid_fail_disk(struct g_raid_softc *sc, 223389b17223SAlexander Motin struct g_raid_subdisk *sd, struct g_raid_disk *disk) 223489b17223SAlexander Motin { 223589b17223SAlexander Motin 223689b17223SAlexander Motin if (disk == NULL) 223789b17223SAlexander Motin disk = sd->sd_disk; 223889b17223SAlexander Motin if (disk == NULL) { 223989b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); 224089b17223SAlexander Motin return; 224189b17223SAlexander Motin } 224289b17223SAlexander Motin if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 224389b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " 224489b17223SAlexander Motin "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 224589b17223SAlexander Motin return; 224689b17223SAlexander Motin } 224789b17223SAlexander Motin if (sc->sc_md) 224889b17223SAlexander Motin G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); 224989b17223SAlexander Motin } 225089b17223SAlexander Motin 225189b17223SAlexander Motin static void 225289b17223SAlexander Motin g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 225389b17223SAlexander Motin struct g_consumer *cp, struct g_provider *pp) 225489b17223SAlexander Motin { 225589b17223SAlexander Motin struct g_raid_softc *sc; 225689b17223SAlexander Motin struct g_raid_volume *vol; 225789b17223SAlexander Motin struct g_raid_subdisk *sd; 225889b17223SAlexander Motin struct g_raid_disk *disk; 225989b17223SAlexander Motin int i, s; 226089b17223SAlexander Motin 226189b17223SAlexander Motin g_topology_assert(); 226289b17223SAlexander Motin 226389b17223SAlexander Motin sc = gp->softc; 226489b17223SAlexander Motin if (sc == NULL) 226589b17223SAlexander Motin return; 226689b17223SAlexander Motin if (pp != NULL) { 226789b17223SAlexander Motin vol = pp->private; 226889b17223SAlexander Motin g_topology_unlock(); 226989b17223SAlexander Motin sx_xlock(&sc->sc_lock); 227089b17223SAlexander Motin sbuf_printf(sb, "%s<Label>%s</Label>\n", indent, 227189b17223SAlexander Motin vol->v_name); 227289b17223SAlexander Motin sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent, 227389b17223SAlexander Motin g_raid_volume_level2str(vol->v_raid_level, 227489b17223SAlexander Motin vol->v_raid_level_qualifier)); 227589b17223SAlexander Motin sbuf_printf(sb, 227689b17223SAlexander Motin "%s<Transformation>%s</Transformation>\n", indent, 227789b17223SAlexander Motin vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); 227889b17223SAlexander Motin sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 227989b17223SAlexander Motin vol->v_disks_count); 228089b17223SAlexander Motin sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent, 228189b17223SAlexander Motin vol->v_strip_size); 228289b17223SAlexander Motin sbuf_printf(sb, "%s<State>%s</State>\n", indent, 228389b17223SAlexander Motin g_raid_volume_state2str(vol->v_state)); 228489b17223SAlexander Motin sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent, 228589b17223SAlexander Motin vol->v_dirty ? "Yes" : "No"); 228689b17223SAlexander Motin sbuf_printf(sb, "%s<Subdisks>", indent); 228789b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 228889b17223SAlexander Motin sd = &vol->v_subdisks[i]; 228989b17223SAlexander Motin if (sd->sd_disk != NULL && 229089b17223SAlexander Motin sd->sd_disk->d_consumer != NULL) { 229189b17223SAlexander Motin sbuf_printf(sb, "%s ", 229289b17223SAlexander Motin g_raid_get_diskname(sd->sd_disk)); 229389b17223SAlexander Motin } else { 229489b17223SAlexander Motin sbuf_printf(sb, "NONE "); 229589b17223SAlexander Motin } 229689b17223SAlexander Motin sbuf_printf(sb, "(%s", 229789b17223SAlexander Motin g_raid_subdisk_state2str(sd->sd_state)); 229889b17223SAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 229989b17223SAlexander Motin sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 230089b17223SAlexander Motin sbuf_printf(sb, " %d%%", 230189b17223SAlexander Motin (int)(sd->sd_rebuild_pos * 100 / 230289b17223SAlexander Motin sd->sd_size)); 230389b17223SAlexander Motin } 230489b17223SAlexander Motin sbuf_printf(sb, ")"); 230589b17223SAlexander Motin if (i + 1 < vol->v_disks_count) 230689b17223SAlexander Motin sbuf_printf(sb, ", "); 230789b17223SAlexander Motin } 230889b17223SAlexander Motin sbuf_printf(sb, "</Subdisks>\n"); 230989b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 231089b17223SAlexander Motin g_topology_lock(); 231189b17223SAlexander Motin } else if (cp != NULL) { 231289b17223SAlexander Motin disk = cp->private; 231389b17223SAlexander Motin if (disk == NULL) 231489b17223SAlexander Motin return; 231589b17223SAlexander Motin g_topology_unlock(); 231689b17223SAlexander Motin sx_xlock(&sc->sc_lock); 231789b17223SAlexander Motin sbuf_printf(sb, "%s<State>%s", indent, 231889b17223SAlexander Motin g_raid_disk_state2str(disk->d_state)); 231989b17223SAlexander Motin if (!TAILQ_EMPTY(&disk->d_subdisks)) { 232089b17223SAlexander Motin sbuf_printf(sb, " ("); 232189b17223SAlexander Motin TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 232289b17223SAlexander Motin sbuf_printf(sb, "%s", 232389b17223SAlexander Motin g_raid_subdisk_state2str(sd->sd_state)); 232489b17223SAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 232589b17223SAlexander Motin sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 232689b17223SAlexander Motin sbuf_printf(sb, " %d%%", 232789b17223SAlexander Motin (int)(sd->sd_rebuild_pos * 100 / 232889b17223SAlexander Motin sd->sd_size)); 232989b17223SAlexander Motin } 233089b17223SAlexander Motin if (TAILQ_NEXT(sd, sd_next)) 233189b17223SAlexander Motin sbuf_printf(sb, ", "); 233289b17223SAlexander Motin } 233389b17223SAlexander Motin sbuf_printf(sb, ")"); 233489b17223SAlexander Motin } 233589b17223SAlexander Motin sbuf_printf(sb, "</State>\n"); 233689b17223SAlexander Motin sbuf_printf(sb, "%s<Subdisks>", indent); 233789b17223SAlexander Motin TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 233889b17223SAlexander Motin sbuf_printf(sb, "r%d(%s):%d@%ju", 233989b17223SAlexander Motin sd->sd_volume->v_global_id, 234089b17223SAlexander Motin sd->sd_volume->v_name, 234189b17223SAlexander Motin sd->sd_pos, sd->sd_offset); 234289b17223SAlexander Motin if (TAILQ_NEXT(sd, sd_next)) 234389b17223SAlexander Motin sbuf_printf(sb, ", "); 234489b17223SAlexander Motin } 234589b17223SAlexander Motin sbuf_printf(sb, "</Subdisks>\n"); 234689b17223SAlexander Motin sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent, 234789b17223SAlexander Motin disk->d_read_errs); 234889b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 234989b17223SAlexander Motin g_topology_lock(); 235089b17223SAlexander Motin } else { 235189b17223SAlexander Motin g_topology_unlock(); 235289b17223SAlexander Motin sx_xlock(&sc->sc_lock); 235389b17223SAlexander Motin if (sc->sc_md) { 235489b17223SAlexander Motin sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent, 235589b17223SAlexander Motin sc->sc_md->mdo_class->name); 235689b17223SAlexander Motin } 235789b17223SAlexander Motin if (!TAILQ_EMPTY(&sc->sc_volumes)) { 235889b17223SAlexander Motin s = 0xff; 235989b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 236089b17223SAlexander Motin if (vol->v_state < s) 236189b17223SAlexander Motin s = vol->v_state; 236289b17223SAlexander Motin } 236389b17223SAlexander Motin sbuf_printf(sb, "%s<State>%s</State>\n", indent, 236489b17223SAlexander Motin g_raid_volume_state2str(s)); 236589b17223SAlexander Motin } 236689b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 236789b17223SAlexander Motin g_topology_lock(); 236889b17223SAlexander Motin } 236989b17223SAlexander Motin } 237089b17223SAlexander Motin 237189b17223SAlexander Motin static void 237289b17223SAlexander Motin g_raid_shutdown_pre_sync(void *arg, int howto) 237389b17223SAlexander Motin { 237489b17223SAlexander Motin struct g_class *mp; 237589b17223SAlexander Motin struct g_geom *gp, *gp2; 237689b17223SAlexander Motin struct g_raid_softc *sc; 237789b17223SAlexander Motin int error; 237889b17223SAlexander Motin 237989b17223SAlexander Motin mp = arg; 238089b17223SAlexander Motin DROP_GIANT(); 238189b17223SAlexander Motin g_topology_lock(); 238289b17223SAlexander Motin LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 238389b17223SAlexander Motin if ((sc = gp->softc) == NULL) 238489b17223SAlexander Motin continue; 238589b17223SAlexander Motin g_topology_unlock(); 238689b17223SAlexander Motin sx_xlock(&sc->sc_lock); 238789b17223SAlexander Motin g_cancel_event(sc); 238889b17223SAlexander Motin error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); 238989b17223SAlexander Motin if (error != 0) 239089b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 239189b17223SAlexander Motin g_topology_lock(); 239289b17223SAlexander Motin } 239389b17223SAlexander Motin g_topology_unlock(); 239489b17223SAlexander Motin PICKUP_GIANT(); 239589b17223SAlexander Motin } 239689b17223SAlexander Motin 239789b17223SAlexander Motin static void 239889b17223SAlexander Motin g_raid_init(struct g_class *mp) 239989b17223SAlexander Motin { 240089b17223SAlexander Motin 240189b17223SAlexander Motin g_raid_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync, 240289b17223SAlexander Motin g_raid_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST); 240389b17223SAlexander Motin if (g_raid_pre_sync == NULL) 240489b17223SAlexander Motin G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); 240589b17223SAlexander Motin g_raid_started = 1; 240689b17223SAlexander Motin } 240789b17223SAlexander Motin 240889b17223SAlexander Motin static void 240989b17223SAlexander Motin g_raid_fini(struct g_class *mp) 241089b17223SAlexander Motin { 241189b17223SAlexander Motin 241289b17223SAlexander Motin if (g_raid_pre_sync != NULL) 241389b17223SAlexander Motin EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_raid_pre_sync); 241489b17223SAlexander Motin g_raid_started = 0; 241589b17223SAlexander Motin } 241689b17223SAlexander Motin 241789b17223SAlexander Motin int 241889b17223SAlexander Motin g_raid_md_modevent(module_t mod, int type, void *arg) 241989b17223SAlexander Motin { 242089b17223SAlexander Motin struct g_raid_md_class *class, *c, *nc; 242189b17223SAlexander Motin int error; 242289b17223SAlexander Motin 242389b17223SAlexander Motin error = 0; 242489b17223SAlexander Motin class = arg; 242589b17223SAlexander Motin switch (type) { 242689b17223SAlexander Motin case MOD_LOAD: 242789b17223SAlexander Motin c = LIST_FIRST(&g_raid_md_classes); 242889b17223SAlexander Motin if (c == NULL || c->mdc_priority > class->mdc_priority) 242989b17223SAlexander Motin LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); 243089b17223SAlexander Motin else { 243189b17223SAlexander Motin while ((nc = LIST_NEXT(c, mdc_list)) != NULL && 243289b17223SAlexander Motin nc->mdc_priority < class->mdc_priority) 243389b17223SAlexander Motin c = nc; 243489b17223SAlexander Motin LIST_INSERT_AFTER(c, class, mdc_list); 243589b17223SAlexander Motin } 243689b17223SAlexander Motin if (g_raid_started) 243789b17223SAlexander Motin g_retaste(&g_raid_class); 243889b17223SAlexander Motin break; 243989b17223SAlexander Motin case MOD_UNLOAD: 244089b17223SAlexander Motin LIST_REMOVE(class, mdc_list); 244189b17223SAlexander Motin break; 244289b17223SAlexander Motin default: 244389b17223SAlexander Motin error = EOPNOTSUPP; 244489b17223SAlexander Motin break; 244589b17223SAlexander Motin } 244689b17223SAlexander Motin 244789b17223SAlexander Motin return (error); 244889b17223SAlexander Motin } 244989b17223SAlexander Motin 245089b17223SAlexander Motin int 245189b17223SAlexander Motin g_raid_tr_modevent(module_t mod, int type, void *arg) 245289b17223SAlexander Motin { 245389b17223SAlexander Motin struct g_raid_tr_class *class, *c, *nc; 245489b17223SAlexander Motin int error; 245589b17223SAlexander Motin 245689b17223SAlexander Motin error = 0; 245789b17223SAlexander Motin class = arg; 245889b17223SAlexander Motin switch (type) { 245989b17223SAlexander Motin case MOD_LOAD: 246089b17223SAlexander Motin c = LIST_FIRST(&g_raid_tr_classes); 246189b17223SAlexander Motin if (c == NULL || c->trc_priority > class->trc_priority) 246289b17223SAlexander Motin LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); 246389b17223SAlexander Motin else { 246489b17223SAlexander Motin while ((nc = LIST_NEXT(c, trc_list)) != NULL && 246589b17223SAlexander Motin nc->trc_priority < class->trc_priority) 246689b17223SAlexander Motin c = nc; 246789b17223SAlexander Motin LIST_INSERT_AFTER(c, class, trc_list); 246889b17223SAlexander Motin } 246989b17223SAlexander Motin break; 247089b17223SAlexander Motin case MOD_UNLOAD: 247189b17223SAlexander Motin LIST_REMOVE(class, trc_list); 247289b17223SAlexander Motin break; 247389b17223SAlexander Motin default: 247489b17223SAlexander Motin error = EOPNOTSUPP; 247589b17223SAlexander Motin break; 247689b17223SAlexander Motin } 247789b17223SAlexander Motin 247889b17223SAlexander Motin return (error); 247989b17223SAlexander Motin } 248089b17223SAlexander Motin 248189b17223SAlexander Motin /* 248289b17223SAlexander Motin * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) 248389b17223SAlexander Motin * to reduce module priority, allowing submodules to register them first. 248489b17223SAlexander Motin */ 248589b17223SAlexander Motin static moduledata_t g_raid_mod = { 248689b17223SAlexander Motin "g_raid", 248789b17223SAlexander Motin g_modevent, 248889b17223SAlexander Motin &g_raid_class 248989b17223SAlexander Motin }; 249089b17223SAlexander Motin DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD); 249189b17223SAlexander Motin MODULE_VERSION(geom_raid, 0); 2492