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> 3889b17223SAlexander Motin #include <sys/sysctl.h> 3989b17223SAlexander Motin #include <sys/malloc.h> 4089b17223SAlexander Motin #include <sys/eventhandler.h> 4189b17223SAlexander Motin #include <vm/uma.h> 4289b17223SAlexander Motin #include <geom/geom.h> 4389b17223SAlexander Motin #include <sys/proc.h> 4489b17223SAlexander Motin #include <sys/kthread.h> 4589b17223SAlexander Motin #include <sys/sched.h> 4689b17223SAlexander Motin #include <geom/raid/g_raid.h> 4789b17223SAlexander Motin #include "g_raid_md_if.h" 4889b17223SAlexander Motin #include "g_raid_tr_if.h" 4989b17223SAlexander Motin 5089b17223SAlexander Motin static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data"); 5189b17223SAlexander Motin 5289b17223SAlexander Motin SYSCTL_DECL(_kern_geom); 5389b17223SAlexander Motin SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff"); 5489b17223SAlexander Motin u_int g_raid_aggressive_spare = 0; 5589b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare); 5689b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW, 5789b17223SAlexander Motin &g_raid_aggressive_spare, 0, "Use disks without metadata as spare"); 5889b17223SAlexander Motin u_int g_raid_debug = 2; 5989b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.debug", &g_raid_debug); 6089b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0, 6189b17223SAlexander Motin "Debug level"); 6289b17223SAlexander Motin int g_raid_read_err_thresh = 10; 6389b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh); 6489b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW, 6589b17223SAlexander Motin &g_raid_read_err_thresh, 0, 6689b17223SAlexander Motin "Number of read errors equated to disk failure"); 6789b17223SAlexander Motin u_int g_raid_start_timeout = 30; 6889b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout); 6989b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW, 7089b17223SAlexander Motin &g_raid_start_timeout, 0, 7189b17223SAlexander Motin "Time to wait for all array components"); 7289b17223SAlexander Motin static u_int g_raid_clean_time = 5; 7389b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time); 7489b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW, 7589b17223SAlexander Motin &g_raid_clean_time, 0, "Mark volume as clean when idling"); 7689b17223SAlexander Motin static u_int g_raid_disconnect_on_failure = 1; 7789b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.disconnect_on_failure", 7889b17223SAlexander Motin &g_raid_disconnect_on_failure); 7989b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, 8089b17223SAlexander Motin &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure."); 8189b17223SAlexander Motin static u_int g_raid_name_format = 0; 8289b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format); 8389b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW, 8489b17223SAlexander Motin &g_raid_name_format, 0, "Providers name format."); 8589b17223SAlexander Motin static u_int g_raid_idle_threshold = 1000000; 8689b17223SAlexander Motin TUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold); 8789b17223SAlexander Motin SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW, 8889b17223SAlexander Motin &g_raid_idle_threshold, 1000000, 8989b17223SAlexander Motin "Time in microseconds to consider a volume idle."); 9089b17223SAlexander Motin 9189b17223SAlexander Motin #define MSLEEP(rv, ident, mtx, priority, wmesg, timeout) do { \ 9289b17223SAlexander Motin G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ 9389b17223SAlexander Motin rv = msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ 9489b17223SAlexander Motin G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ 9589b17223SAlexander Motin } while (0) 9689b17223SAlexander Motin 9789b17223SAlexander Motin LIST_HEAD(, g_raid_md_class) g_raid_md_classes = 9889b17223SAlexander Motin LIST_HEAD_INITIALIZER(g_raid_md_classes); 9989b17223SAlexander Motin 10089b17223SAlexander Motin LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes = 10189b17223SAlexander Motin LIST_HEAD_INITIALIZER(g_raid_tr_classes); 10289b17223SAlexander Motin 10389b17223SAlexander Motin LIST_HEAD(, g_raid_volume) g_raid_volumes = 10489b17223SAlexander Motin LIST_HEAD_INITIALIZER(g_raid_volumes); 10589b17223SAlexander Motin 10689b17223SAlexander Motin static eventhandler_tag g_raid_pre_sync = NULL; 10789b17223SAlexander Motin static int g_raid_started = 0; 10889b17223SAlexander Motin 10989b17223SAlexander Motin static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp, 11089b17223SAlexander Motin struct g_geom *gp); 11189b17223SAlexander Motin static g_taste_t g_raid_taste; 11289b17223SAlexander Motin static void g_raid_init(struct g_class *mp); 11389b17223SAlexander Motin static void g_raid_fini(struct g_class *mp); 11489b17223SAlexander Motin 11589b17223SAlexander Motin struct g_class g_raid_class = { 11689b17223SAlexander Motin .name = G_RAID_CLASS_NAME, 11789b17223SAlexander Motin .version = G_VERSION, 11889b17223SAlexander Motin .ctlreq = g_raid_ctl, 11989b17223SAlexander Motin .taste = g_raid_taste, 12089b17223SAlexander Motin .destroy_geom = g_raid_destroy_geom, 12189b17223SAlexander Motin .init = g_raid_init, 12289b17223SAlexander Motin .fini = g_raid_fini 12389b17223SAlexander Motin }; 12489b17223SAlexander Motin 12589b17223SAlexander Motin static void g_raid_destroy_provider(struct g_raid_volume *vol); 12689b17223SAlexander Motin static int g_raid_update_disk(struct g_raid_disk *disk, u_int event); 12789b17223SAlexander Motin static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event); 12889b17223SAlexander Motin static int g_raid_update_volume(struct g_raid_volume *vol, u_int event); 12989b17223SAlexander Motin static int g_raid_update_node(struct g_raid_softc *sc, u_int event); 13089b17223SAlexander Motin static void g_raid_dumpconf(struct sbuf *sb, const char *indent, 13189b17223SAlexander Motin struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); 13289b17223SAlexander Motin static void g_raid_start(struct bio *bp); 13389b17223SAlexander Motin static void g_raid_start_request(struct bio *bp); 13489b17223SAlexander Motin static void g_raid_disk_done(struct bio *bp); 13589b17223SAlexander Motin static void g_raid_poll(struct g_raid_softc *sc); 13689b17223SAlexander Motin 13789b17223SAlexander Motin static const char * 13889b17223SAlexander Motin g_raid_node_event2str(int event) 13989b17223SAlexander Motin { 14089b17223SAlexander Motin 14189b17223SAlexander Motin switch (event) { 14289b17223SAlexander Motin case G_RAID_NODE_E_WAKE: 14389b17223SAlexander Motin return ("WAKE"); 14489b17223SAlexander Motin case G_RAID_NODE_E_START: 14589b17223SAlexander Motin return ("START"); 14689b17223SAlexander Motin default: 14789b17223SAlexander Motin return ("INVALID"); 14889b17223SAlexander Motin } 14989b17223SAlexander Motin } 15089b17223SAlexander Motin 15189b17223SAlexander Motin const char * 15289b17223SAlexander Motin g_raid_disk_state2str(int state) 15389b17223SAlexander Motin { 15489b17223SAlexander Motin 15589b17223SAlexander Motin switch (state) { 15689b17223SAlexander Motin case G_RAID_DISK_S_NONE: 15789b17223SAlexander Motin return ("NONE"); 15889b17223SAlexander Motin case G_RAID_DISK_S_OFFLINE: 15989b17223SAlexander Motin return ("OFFLINE"); 16089b17223SAlexander Motin case G_RAID_DISK_S_FAILED: 16189b17223SAlexander Motin return ("FAILED"); 16289b17223SAlexander Motin case G_RAID_DISK_S_STALE_FAILED: 16389b17223SAlexander Motin return ("STALE_FAILED"); 16489b17223SAlexander Motin case G_RAID_DISK_S_SPARE: 16589b17223SAlexander Motin return ("SPARE"); 16689b17223SAlexander Motin case G_RAID_DISK_S_STALE: 16789b17223SAlexander Motin return ("STALE"); 16889b17223SAlexander Motin case G_RAID_DISK_S_ACTIVE: 16989b17223SAlexander Motin return ("ACTIVE"); 17089b17223SAlexander Motin default: 17189b17223SAlexander Motin return ("INVALID"); 17289b17223SAlexander Motin } 17389b17223SAlexander Motin } 17489b17223SAlexander Motin 17589b17223SAlexander Motin static const char * 17689b17223SAlexander Motin g_raid_disk_event2str(int event) 17789b17223SAlexander Motin { 17889b17223SAlexander Motin 17989b17223SAlexander Motin switch (event) { 18089b17223SAlexander Motin case G_RAID_DISK_E_DISCONNECTED: 18189b17223SAlexander Motin return ("DISCONNECTED"); 18289b17223SAlexander Motin default: 18389b17223SAlexander Motin return ("INVALID"); 18489b17223SAlexander Motin } 18589b17223SAlexander Motin } 18689b17223SAlexander Motin 18789b17223SAlexander Motin const char * 18889b17223SAlexander Motin g_raid_subdisk_state2str(int state) 18989b17223SAlexander Motin { 19089b17223SAlexander Motin 19189b17223SAlexander Motin switch (state) { 19289b17223SAlexander Motin case G_RAID_SUBDISK_S_NONE: 19389b17223SAlexander Motin return ("NONE"); 19489b17223SAlexander Motin case G_RAID_SUBDISK_S_FAILED: 19589b17223SAlexander Motin return ("FAILED"); 19689b17223SAlexander Motin case G_RAID_SUBDISK_S_NEW: 19789b17223SAlexander Motin return ("NEW"); 19889b17223SAlexander Motin case G_RAID_SUBDISK_S_REBUILD: 19989b17223SAlexander Motin return ("REBUILD"); 20089b17223SAlexander Motin case G_RAID_SUBDISK_S_UNINITIALIZED: 20189b17223SAlexander Motin return ("UNINITIALIZED"); 20289b17223SAlexander Motin case G_RAID_SUBDISK_S_STALE: 20389b17223SAlexander Motin return ("STALE"); 20489b17223SAlexander Motin case G_RAID_SUBDISK_S_RESYNC: 20589b17223SAlexander Motin return ("RESYNC"); 20689b17223SAlexander Motin case G_RAID_SUBDISK_S_ACTIVE: 20789b17223SAlexander Motin return ("ACTIVE"); 20889b17223SAlexander Motin default: 20989b17223SAlexander Motin return ("INVALID"); 21089b17223SAlexander Motin } 21189b17223SAlexander Motin } 21289b17223SAlexander Motin 21389b17223SAlexander Motin static const char * 21489b17223SAlexander Motin g_raid_subdisk_event2str(int event) 21589b17223SAlexander Motin { 21689b17223SAlexander Motin 21789b17223SAlexander Motin switch (event) { 21889b17223SAlexander Motin case G_RAID_SUBDISK_E_NEW: 21989b17223SAlexander Motin return ("NEW"); 22089b17223SAlexander Motin case G_RAID_SUBDISK_E_DISCONNECTED: 22189b17223SAlexander Motin return ("DISCONNECTED"); 22289b17223SAlexander Motin default: 22389b17223SAlexander Motin return ("INVALID"); 22489b17223SAlexander Motin } 22589b17223SAlexander Motin } 22689b17223SAlexander Motin 22789b17223SAlexander Motin const char * 22889b17223SAlexander Motin g_raid_volume_state2str(int state) 22989b17223SAlexander Motin { 23089b17223SAlexander Motin 23189b17223SAlexander Motin switch (state) { 23289b17223SAlexander Motin case G_RAID_VOLUME_S_STARTING: 23389b17223SAlexander Motin return ("STARTING"); 23489b17223SAlexander Motin case G_RAID_VOLUME_S_BROKEN: 23589b17223SAlexander Motin return ("BROKEN"); 23689b17223SAlexander Motin case G_RAID_VOLUME_S_DEGRADED: 23789b17223SAlexander Motin return ("DEGRADED"); 23889b17223SAlexander Motin case G_RAID_VOLUME_S_SUBOPTIMAL: 23989b17223SAlexander Motin return ("SUBOPTIMAL"); 24089b17223SAlexander Motin case G_RAID_VOLUME_S_OPTIMAL: 24189b17223SAlexander Motin return ("OPTIMAL"); 24289b17223SAlexander Motin case G_RAID_VOLUME_S_UNSUPPORTED: 24389b17223SAlexander Motin return ("UNSUPPORTED"); 24489b17223SAlexander Motin case G_RAID_VOLUME_S_STOPPED: 24589b17223SAlexander Motin return ("STOPPED"); 24689b17223SAlexander Motin default: 24789b17223SAlexander Motin return ("INVALID"); 24889b17223SAlexander Motin } 24989b17223SAlexander Motin } 25089b17223SAlexander Motin 25189b17223SAlexander Motin static const char * 25289b17223SAlexander Motin g_raid_volume_event2str(int event) 25389b17223SAlexander Motin { 25489b17223SAlexander Motin 25589b17223SAlexander Motin switch (event) { 25689b17223SAlexander Motin case G_RAID_VOLUME_E_UP: 25789b17223SAlexander Motin return ("UP"); 25889b17223SAlexander Motin case G_RAID_VOLUME_E_DOWN: 25989b17223SAlexander Motin return ("DOWN"); 26089b17223SAlexander Motin case G_RAID_VOLUME_E_START: 26189b17223SAlexander Motin return ("START"); 26289b17223SAlexander Motin case G_RAID_VOLUME_E_STARTMD: 26389b17223SAlexander Motin return ("STARTMD"); 26489b17223SAlexander Motin default: 26589b17223SAlexander Motin return ("INVALID"); 26689b17223SAlexander Motin } 26789b17223SAlexander Motin } 26889b17223SAlexander Motin 26989b17223SAlexander Motin const char * 27089b17223SAlexander Motin g_raid_volume_level2str(int level, int qual) 27189b17223SAlexander Motin { 27289b17223SAlexander Motin 27389b17223SAlexander Motin switch (level) { 27489b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID0: 27589b17223SAlexander Motin return ("RAID0"); 27689b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID1: 27789b17223SAlexander Motin return ("RAID1"); 27889b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID3: 27989b17223SAlexander Motin return ("RAID3"); 28089b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID4: 28189b17223SAlexander Motin return ("RAID4"); 28289b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID5: 28389b17223SAlexander Motin return ("RAID5"); 28489b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID6: 28589b17223SAlexander Motin return ("RAID6"); 28689b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID1E: 28789b17223SAlexander Motin return ("RAID1E"); 28889b17223SAlexander Motin case G_RAID_VOLUME_RL_SINGLE: 28989b17223SAlexander Motin return ("SINGLE"); 29089b17223SAlexander Motin case G_RAID_VOLUME_RL_CONCAT: 29189b17223SAlexander Motin return ("CONCAT"); 29289b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID5E: 29389b17223SAlexander Motin return ("RAID5E"); 29489b17223SAlexander Motin case G_RAID_VOLUME_RL_RAID5EE: 29589b17223SAlexander Motin return ("RAID5EE"); 29689b17223SAlexander Motin default: 29789b17223SAlexander Motin return ("UNKNOWN"); 29889b17223SAlexander Motin } 29989b17223SAlexander Motin } 30089b17223SAlexander Motin 30189b17223SAlexander Motin int 30289b17223SAlexander Motin g_raid_volume_str2level(const char *str, int *level, int *qual) 30389b17223SAlexander Motin { 30489b17223SAlexander Motin 30589b17223SAlexander Motin *level = G_RAID_VOLUME_RL_UNKNOWN; 30689b17223SAlexander Motin *qual = G_RAID_VOLUME_RLQ_NONE; 30789b17223SAlexander Motin if (strcasecmp(str, "RAID0") == 0) 30889b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID0; 30989b17223SAlexander Motin else if (strcasecmp(str, "RAID1") == 0) 31089b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID1; 31189b17223SAlexander Motin else if (strcasecmp(str, "RAID3") == 0) 31289b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID3; 31389b17223SAlexander Motin else if (strcasecmp(str, "RAID4") == 0) 31489b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID4; 31589b17223SAlexander Motin else if (strcasecmp(str, "RAID5") == 0) 31689b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5; 31789b17223SAlexander Motin else if (strcasecmp(str, "RAID6") == 0) 31889b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID6; 31989b17223SAlexander Motin else if (strcasecmp(str, "RAID10") == 0 || 32089b17223SAlexander Motin strcasecmp(str, "RAID1E") == 0) 32189b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID1E; 32289b17223SAlexander Motin else if (strcasecmp(str, "SINGLE") == 0) 32389b17223SAlexander Motin *level = G_RAID_VOLUME_RL_SINGLE; 32489b17223SAlexander Motin else if (strcasecmp(str, "CONCAT") == 0) 32589b17223SAlexander Motin *level = G_RAID_VOLUME_RL_CONCAT; 32689b17223SAlexander Motin else if (strcasecmp(str, "RAID5E") == 0) 32789b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5E; 32889b17223SAlexander Motin else if (strcasecmp(str, "RAID5EE") == 0) 32989b17223SAlexander Motin *level = G_RAID_VOLUME_RL_RAID5EE; 33089b17223SAlexander Motin else 33189b17223SAlexander Motin return (-1); 33289b17223SAlexander Motin return (0); 33389b17223SAlexander Motin } 33489b17223SAlexander Motin 33589b17223SAlexander Motin const char * 33689b17223SAlexander Motin g_raid_get_diskname(struct g_raid_disk *disk) 33789b17223SAlexander Motin { 33889b17223SAlexander Motin 33989b17223SAlexander Motin if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) 34089b17223SAlexander Motin return ("[unknown]"); 34189b17223SAlexander Motin return (disk->d_consumer->provider->name); 34289b17223SAlexander Motin } 34389b17223SAlexander Motin 34489b17223SAlexander Motin void 34589b17223SAlexander Motin g_raid_report_disk_state(struct g_raid_disk *disk) 34689b17223SAlexander Motin { 34789b17223SAlexander Motin struct g_raid_subdisk *sd; 34889b17223SAlexander Motin int len, state; 34989b17223SAlexander Motin uint32_t s; 35089b17223SAlexander Motin 35189b17223SAlexander Motin if (disk->d_consumer == NULL) 35289b17223SAlexander Motin return; 35389b17223SAlexander Motin if (disk->d_state == G_RAID_DISK_S_FAILED || 35489b17223SAlexander Motin disk->d_state == G_RAID_DISK_S_STALE_FAILED) { 35589b17223SAlexander Motin s = G_STATE_FAILED; 35689b17223SAlexander Motin } else { 35789b17223SAlexander Motin state = G_RAID_SUBDISK_S_ACTIVE; 35889b17223SAlexander Motin TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 35989b17223SAlexander Motin if (sd->sd_state < state) 36089b17223SAlexander Motin state = sd->sd_state; 36189b17223SAlexander Motin } 36289b17223SAlexander Motin if (state == G_RAID_SUBDISK_S_FAILED) 36389b17223SAlexander Motin s = G_STATE_FAILED; 36489b17223SAlexander Motin else if (state == G_RAID_SUBDISK_S_NEW || 36589b17223SAlexander Motin state == G_RAID_SUBDISK_S_REBUILD) 36689b17223SAlexander Motin s = G_STATE_REBUILD; 36789b17223SAlexander Motin else if (state == G_RAID_SUBDISK_S_STALE || 36889b17223SAlexander Motin state == G_RAID_SUBDISK_S_RESYNC) 36989b17223SAlexander Motin s = G_STATE_RESYNC; 37089b17223SAlexander Motin else 37189b17223SAlexander Motin s = G_STATE_ACTIVE; 37289b17223SAlexander Motin } 37389b17223SAlexander Motin len = sizeof(s); 37489b17223SAlexander Motin g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s); 37589b17223SAlexander Motin G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.", 37689b17223SAlexander Motin g_raid_get_diskname(disk), s); 37789b17223SAlexander Motin } 37889b17223SAlexander Motin 37989b17223SAlexander Motin void 38089b17223SAlexander Motin g_raid_change_disk_state(struct g_raid_disk *disk, int state) 38189b17223SAlexander Motin { 38289b17223SAlexander Motin 38389b17223SAlexander Motin G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.", 38489b17223SAlexander Motin g_raid_get_diskname(disk), 38589b17223SAlexander Motin g_raid_disk_state2str(disk->d_state), 38689b17223SAlexander Motin g_raid_disk_state2str(state)); 38789b17223SAlexander Motin disk->d_state = state; 38889b17223SAlexander Motin g_raid_report_disk_state(disk); 38989b17223SAlexander Motin } 39089b17223SAlexander Motin 39189b17223SAlexander Motin void 39289b17223SAlexander Motin g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state) 39389b17223SAlexander Motin { 39489b17223SAlexander Motin 39589b17223SAlexander Motin G_RAID_DEBUG1(0, sd->sd_softc, 39689b17223SAlexander Motin "Subdisk %s:%d-%s state changed from %s to %s.", 39789b17223SAlexander Motin sd->sd_volume->v_name, sd->sd_pos, 39889b17223SAlexander Motin sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", 39989b17223SAlexander Motin g_raid_subdisk_state2str(sd->sd_state), 40089b17223SAlexander Motin g_raid_subdisk_state2str(state)); 40189b17223SAlexander Motin sd->sd_state = state; 40289b17223SAlexander Motin if (sd->sd_disk) 40389b17223SAlexander Motin g_raid_report_disk_state(sd->sd_disk); 40489b17223SAlexander Motin } 40589b17223SAlexander Motin 40689b17223SAlexander Motin void 40789b17223SAlexander Motin g_raid_change_volume_state(struct g_raid_volume *vol, int state) 40889b17223SAlexander Motin { 40989b17223SAlexander Motin 41089b17223SAlexander Motin G_RAID_DEBUG1(0, vol->v_softc, 41189b17223SAlexander Motin "Volume %s state changed from %s to %s.", 41289b17223SAlexander Motin vol->v_name, 41389b17223SAlexander Motin g_raid_volume_state2str(vol->v_state), 41489b17223SAlexander Motin g_raid_volume_state2str(state)); 41589b17223SAlexander Motin vol->v_state = state; 41689b17223SAlexander Motin } 41789b17223SAlexander Motin 41889b17223SAlexander Motin /* 41989b17223SAlexander Motin * --- Events handling functions --- 42089b17223SAlexander Motin * Events in geom_raid are used to maintain subdisks and volumes status 42189b17223SAlexander Motin * from one thread to simplify locking. 42289b17223SAlexander Motin */ 42389b17223SAlexander Motin static void 42489b17223SAlexander Motin g_raid_event_free(struct g_raid_event *ep) 42589b17223SAlexander Motin { 42689b17223SAlexander Motin 42789b17223SAlexander Motin free(ep, M_RAID); 42889b17223SAlexander Motin } 42989b17223SAlexander Motin 43089b17223SAlexander Motin int 43189b17223SAlexander Motin g_raid_event_send(void *arg, int event, int flags) 43289b17223SAlexander Motin { 43389b17223SAlexander Motin struct g_raid_softc *sc; 43489b17223SAlexander Motin struct g_raid_event *ep; 43589b17223SAlexander Motin int error; 43689b17223SAlexander Motin 43789b17223SAlexander Motin if ((flags & G_RAID_EVENT_VOLUME) != 0) { 43889b17223SAlexander Motin sc = ((struct g_raid_volume *)arg)->v_softc; 43989b17223SAlexander Motin } else if ((flags & G_RAID_EVENT_DISK) != 0) { 44089b17223SAlexander Motin sc = ((struct g_raid_disk *)arg)->d_softc; 44189b17223SAlexander Motin } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) { 44289b17223SAlexander Motin sc = ((struct g_raid_subdisk *)arg)->sd_softc; 44389b17223SAlexander Motin } else { 44489b17223SAlexander Motin sc = arg; 44589b17223SAlexander Motin } 44689b17223SAlexander Motin ep = malloc(sizeof(*ep), M_RAID, 44789b17223SAlexander Motin sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT); 44889b17223SAlexander Motin if (ep == NULL) 44989b17223SAlexander Motin return (ENOMEM); 45089b17223SAlexander Motin ep->e_tgt = arg; 45189b17223SAlexander Motin ep->e_event = event; 45289b17223SAlexander Motin ep->e_flags = flags; 45389b17223SAlexander Motin ep->e_error = 0; 45489b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc); 45589b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 45689b17223SAlexander Motin TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); 45789b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 45889b17223SAlexander Motin wakeup(sc); 45989b17223SAlexander Motin 46089b17223SAlexander Motin if ((flags & G_RAID_EVENT_WAIT) == 0) 46189b17223SAlexander Motin return (0); 46289b17223SAlexander Motin 46389b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 46489b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep); 46589b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 46689b17223SAlexander Motin while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) { 46789b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 46889b17223SAlexander Motin MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event", 46989b17223SAlexander Motin hz * 5); 47089b17223SAlexander Motin } 47189b17223SAlexander Motin error = ep->e_error; 47289b17223SAlexander Motin g_raid_event_free(ep); 47389b17223SAlexander Motin sx_xlock(&sc->sc_lock); 47489b17223SAlexander Motin return (error); 47589b17223SAlexander Motin } 47689b17223SAlexander Motin 47789b17223SAlexander Motin static void 47889b17223SAlexander Motin g_raid_event_cancel(struct g_raid_softc *sc, void *tgt) 47989b17223SAlexander Motin { 48089b17223SAlexander Motin struct g_raid_event *ep, *tmpep; 48189b17223SAlexander Motin 48289b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 48389b17223SAlexander Motin 48489b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 48589b17223SAlexander Motin TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { 48689b17223SAlexander Motin if (ep->e_tgt != tgt) 48789b17223SAlexander Motin continue; 48889b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_events, ep, e_next); 48989b17223SAlexander Motin if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) 49089b17223SAlexander Motin g_raid_event_free(ep); 49189b17223SAlexander Motin else { 49289b17223SAlexander Motin ep->e_error = ECANCELED; 49389b17223SAlexander Motin wakeup(ep); 49489b17223SAlexander Motin } 49589b17223SAlexander Motin } 49689b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 49789b17223SAlexander Motin } 49889b17223SAlexander Motin 49989b17223SAlexander Motin static int 50089b17223SAlexander Motin g_raid_event_check(struct g_raid_softc *sc, void *tgt) 50189b17223SAlexander Motin { 50289b17223SAlexander Motin struct g_raid_event *ep; 50389b17223SAlexander Motin int res = 0; 50489b17223SAlexander Motin 50589b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 50689b17223SAlexander Motin 50789b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 50889b17223SAlexander Motin TAILQ_FOREACH(ep, &sc->sc_events, e_next) { 50989b17223SAlexander Motin if (ep->e_tgt != tgt) 51089b17223SAlexander Motin continue; 51189b17223SAlexander Motin res = 1; 51289b17223SAlexander Motin break; 51389b17223SAlexander Motin } 51489b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 51589b17223SAlexander Motin return (res); 51689b17223SAlexander Motin } 51789b17223SAlexander Motin 51889b17223SAlexander Motin /* 51989b17223SAlexander Motin * Return the number of disks in given state. 52089b17223SAlexander Motin * If state is equal to -1, count all connected disks. 52189b17223SAlexander Motin */ 52289b17223SAlexander Motin u_int 52389b17223SAlexander Motin g_raid_ndisks(struct g_raid_softc *sc, int state) 52489b17223SAlexander Motin { 52589b17223SAlexander Motin struct g_raid_disk *disk; 52689b17223SAlexander Motin u_int n; 52789b17223SAlexander Motin 52889b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 52989b17223SAlexander Motin 53089b17223SAlexander Motin n = 0; 53189b17223SAlexander Motin TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { 53289b17223SAlexander Motin if (disk->d_state == state || state == -1) 53389b17223SAlexander Motin n++; 53489b17223SAlexander Motin } 53589b17223SAlexander Motin return (n); 53689b17223SAlexander Motin } 53789b17223SAlexander Motin 53889b17223SAlexander Motin /* 53989b17223SAlexander Motin * Return the number of subdisks in given state. 54089b17223SAlexander Motin * If state is equal to -1, count all connected disks. 54189b17223SAlexander Motin */ 54289b17223SAlexander Motin u_int 54389b17223SAlexander Motin g_raid_nsubdisks(struct g_raid_volume *vol, int state) 54489b17223SAlexander Motin { 54589b17223SAlexander Motin struct g_raid_subdisk *subdisk; 54689b17223SAlexander Motin struct g_raid_softc *sc; 54789b17223SAlexander Motin u_int i, n ; 54889b17223SAlexander Motin 54989b17223SAlexander Motin sc = vol->v_softc; 55089b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 55189b17223SAlexander Motin 55289b17223SAlexander Motin n = 0; 55389b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 55489b17223SAlexander Motin subdisk = &vol->v_subdisks[i]; 55589b17223SAlexander Motin if ((state == -1 && 55689b17223SAlexander Motin subdisk->sd_state != G_RAID_SUBDISK_S_NONE) || 55789b17223SAlexander Motin subdisk->sd_state == state) 55889b17223SAlexander Motin n++; 55989b17223SAlexander Motin } 56089b17223SAlexander Motin return (n); 56189b17223SAlexander Motin } 56289b17223SAlexander Motin 56389b17223SAlexander Motin /* 56489b17223SAlexander Motin * Return the first subdisk in given state. 56589b17223SAlexander Motin * If state is equal to -1, then the first connected disks. 56689b17223SAlexander Motin */ 56789b17223SAlexander Motin struct g_raid_subdisk * 56889b17223SAlexander Motin g_raid_get_subdisk(struct g_raid_volume *vol, int state) 56989b17223SAlexander Motin { 57089b17223SAlexander Motin struct g_raid_subdisk *sd; 57189b17223SAlexander Motin struct g_raid_softc *sc; 57289b17223SAlexander Motin u_int i; 57389b17223SAlexander Motin 57489b17223SAlexander Motin sc = vol->v_softc; 57589b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 57689b17223SAlexander Motin 57789b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 57889b17223SAlexander Motin sd = &vol->v_subdisks[i]; 57989b17223SAlexander Motin if ((state == -1 && 58089b17223SAlexander Motin sd->sd_state != G_RAID_SUBDISK_S_NONE) || 58189b17223SAlexander Motin sd->sd_state == state) 58289b17223SAlexander Motin return (sd); 58389b17223SAlexander Motin } 58489b17223SAlexander Motin return (NULL); 58589b17223SAlexander Motin } 58689b17223SAlexander Motin 58789b17223SAlexander Motin struct g_consumer * 58889b17223SAlexander Motin g_raid_open_consumer(struct g_raid_softc *sc, const char *name) 58989b17223SAlexander Motin { 59089b17223SAlexander Motin struct g_consumer *cp; 59189b17223SAlexander Motin struct g_provider *pp; 59289b17223SAlexander Motin 59389b17223SAlexander Motin g_topology_assert(); 59489b17223SAlexander Motin 59589b17223SAlexander Motin if (strncmp(name, "/dev/", 5) == 0) 59689b17223SAlexander Motin name += 5; 59789b17223SAlexander Motin pp = g_provider_by_name(name); 59889b17223SAlexander Motin if (pp == NULL) 59989b17223SAlexander Motin return (NULL); 60089b17223SAlexander Motin cp = g_new_consumer(sc->sc_geom); 60189b17223SAlexander Motin if (g_attach(cp, pp) != 0) { 60289b17223SAlexander Motin g_destroy_consumer(cp); 60389b17223SAlexander Motin return (NULL); 60489b17223SAlexander Motin } 60589b17223SAlexander Motin if (g_access(cp, 1, 1, 1) != 0) { 60689b17223SAlexander Motin g_detach(cp); 60789b17223SAlexander Motin g_destroy_consumer(cp); 60889b17223SAlexander Motin return (NULL); 60989b17223SAlexander Motin } 61089b17223SAlexander Motin return (cp); 61189b17223SAlexander Motin } 61289b17223SAlexander Motin 61389b17223SAlexander Motin static u_int 61489b17223SAlexander Motin g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) 61589b17223SAlexander Motin { 61689b17223SAlexander Motin struct bio *bp; 61789b17223SAlexander Motin u_int nreqs = 0; 61889b17223SAlexander Motin 61989b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 62089b17223SAlexander Motin TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { 62189b17223SAlexander Motin if (bp->bio_from == cp) 62289b17223SAlexander Motin nreqs++; 62389b17223SAlexander Motin } 62489b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 62589b17223SAlexander Motin return (nreqs); 62689b17223SAlexander Motin } 62789b17223SAlexander Motin 62889b17223SAlexander Motin u_int 62989b17223SAlexander Motin g_raid_nopens(struct g_raid_softc *sc) 63089b17223SAlexander Motin { 63189b17223SAlexander Motin struct g_raid_volume *vol; 63289b17223SAlexander Motin u_int opens; 63389b17223SAlexander Motin 63489b17223SAlexander Motin opens = 0; 63589b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 63689b17223SAlexander Motin if (vol->v_provider_open != 0) 63789b17223SAlexander Motin opens++; 63889b17223SAlexander Motin } 63989b17223SAlexander Motin return (opens); 64089b17223SAlexander Motin } 64189b17223SAlexander Motin 64289b17223SAlexander Motin static int 64389b17223SAlexander Motin g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) 64489b17223SAlexander Motin { 64589b17223SAlexander Motin 64689b17223SAlexander Motin if (cp->index > 0) { 64789b17223SAlexander Motin G_RAID_DEBUG1(2, sc, 64889b17223SAlexander Motin "I/O requests for %s exist, can't destroy it now.", 64989b17223SAlexander Motin cp->provider->name); 65089b17223SAlexander Motin return (1); 65189b17223SAlexander Motin } 65289b17223SAlexander Motin if (g_raid_nrequests(sc, cp) > 0) { 65389b17223SAlexander Motin G_RAID_DEBUG1(2, sc, 65489b17223SAlexander Motin "I/O requests for %s in queue, can't destroy it now.", 65589b17223SAlexander Motin cp->provider->name); 65689b17223SAlexander Motin return (1); 65789b17223SAlexander Motin } 65889b17223SAlexander Motin return (0); 65989b17223SAlexander Motin } 66089b17223SAlexander Motin 66189b17223SAlexander Motin static void 66289b17223SAlexander Motin g_raid_destroy_consumer(void *arg, int flags __unused) 66389b17223SAlexander Motin { 66489b17223SAlexander Motin struct g_consumer *cp; 66589b17223SAlexander Motin 66689b17223SAlexander Motin g_topology_assert(); 66789b17223SAlexander Motin 66889b17223SAlexander Motin cp = arg; 66989b17223SAlexander Motin G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); 67089b17223SAlexander Motin g_detach(cp); 67189b17223SAlexander Motin g_destroy_consumer(cp); 67289b17223SAlexander Motin } 67389b17223SAlexander Motin 67489b17223SAlexander Motin void 67589b17223SAlexander Motin g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) 67689b17223SAlexander Motin { 67789b17223SAlexander Motin struct g_provider *pp; 67889b17223SAlexander Motin int retaste_wait; 67989b17223SAlexander Motin 68089b17223SAlexander Motin g_topology_assert_not(); 68189b17223SAlexander Motin 68289b17223SAlexander Motin g_topology_lock(); 68389b17223SAlexander Motin cp->private = NULL; 68489b17223SAlexander Motin if (g_raid_consumer_is_busy(sc, cp)) 68589b17223SAlexander Motin goto out; 68689b17223SAlexander Motin pp = cp->provider; 68789b17223SAlexander Motin retaste_wait = 0; 68889b17223SAlexander Motin if (cp->acw == 1) { 68989b17223SAlexander Motin if ((pp->geom->flags & G_GEOM_WITHER) == 0) 69089b17223SAlexander Motin retaste_wait = 1; 69189b17223SAlexander Motin } 69289b17223SAlexander Motin if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) 69389b17223SAlexander Motin g_access(cp, -cp->acr, -cp->acw, -cp->ace); 69489b17223SAlexander Motin if (retaste_wait) { 69589b17223SAlexander Motin /* 69689b17223SAlexander Motin * After retaste event was send (inside g_access()), we can send 69789b17223SAlexander Motin * event to detach and destroy consumer. 69889b17223SAlexander Motin * A class, which has consumer to the given provider connected 69989b17223SAlexander Motin * will not receive retaste event for the provider. 70089b17223SAlexander Motin * This is the way how I ignore retaste events when I close 70189b17223SAlexander Motin * consumers opened for write: I detach and destroy consumer 70289b17223SAlexander Motin * after retaste event is sent. 70389b17223SAlexander Motin */ 70489b17223SAlexander Motin g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); 70589b17223SAlexander Motin goto out; 70689b17223SAlexander Motin } 70789b17223SAlexander Motin G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); 70889b17223SAlexander Motin g_detach(cp); 70989b17223SAlexander Motin g_destroy_consumer(cp); 71089b17223SAlexander Motin out: 71189b17223SAlexander Motin g_topology_unlock(); 71289b17223SAlexander Motin } 71389b17223SAlexander Motin 71489b17223SAlexander Motin static void 71589b17223SAlexander Motin g_raid_orphan(struct g_consumer *cp) 71689b17223SAlexander Motin { 71789b17223SAlexander Motin struct g_raid_disk *disk; 71889b17223SAlexander Motin 71989b17223SAlexander Motin g_topology_assert(); 72089b17223SAlexander Motin 72189b17223SAlexander Motin disk = cp->private; 72289b17223SAlexander Motin if (disk == NULL) 72389b17223SAlexander Motin return; 72489b17223SAlexander Motin g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, 72589b17223SAlexander Motin G_RAID_EVENT_DISK); 72689b17223SAlexander Motin } 72789b17223SAlexander Motin 72889b17223SAlexander Motin static int 72989b17223SAlexander Motin g_raid_clean(struct g_raid_volume *vol, int acw) 73089b17223SAlexander Motin { 73189b17223SAlexander Motin struct g_raid_softc *sc; 73289b17223SAlexander Motin int timeout; 73389b17223SAlexander Motin 73489b17223SAlexander Motin sc = vol->v_softc; 73589b17223SAlexander Motin g_topology_assert_not(); 73689b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 73789b17223SAlexander Motin 73889b17223SAlexander Motin // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 73989b17223SAlexander Motin // return (0); 74089b17223SAlexander Motin if (!vol->v_dirty) 74189b17223SAlexander Motin return (0); 74289b17223SAlexander Motin if (vol->v_writes > 0) 74389b17223SAlexander Motin return (0); 74489b17223SAlexander Motin if (acw > 0 || (acw == -1 && 74589b17223SAlexander Motin vol->v_provider != NULL && vol->v_provider->acw > 0)) { 74689b17223SAlexander Motin timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); 74789b17223SAlexander Motin if (timeout > 0) 74889b17223SAlexander Motin return (timeout); 74989b17223SAlexander Motin } 75089b17223SAlexander Motin vol->v_dirty = 0; 75189b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", 75289b17223SAlexander Motin vol->v_name); 75389b17223SAlexander Motin g_raid_write_metadata(sc, vol, NULL, NULL); 75489b17223SAlexander Motin return (0); 75589b17223SAlexander Motin } 75689b17223SAlexander Motin 75789b17223SAlexander Motin static void 75889b17223SAlexander Motin g_raid_dirty(struct g_raid_volume *vol) 75989b17223SAlexander Motin { 76089b17223SAlexander Motin struct g_raid_softc *sc; 76189b17223SAlexander Motin 76289b17223SAlexander Motin sc = vol->v_softc; 76389b17223SAlexander Motin g_topology_assert_not(); 76489b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 76589b17223SAlexander Motin 76689b17223SAlexander Motin // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) 76789b17223SAlexander Motin // return; 76889b17223SAlexander Motin vol->v_dirty = 1; 76989b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", 77089b17223SAlexander Motin vol->v_name); 77189b17223SAlexander Motin g_raid_write_metadata(sc, vol, NULL, NULL); 77289b17223SAlexander Motin } 77389b17223SAlexander Motin 77489b17223SAlexander Motin void 77589b17223SAlexander Motin g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) 77689b17223SAlexander Motin { 77789b17223SAlexander Motin struct g_raid_softc *sc; 77889b17223SAlexander Motin struct g_raid_volume *vol; 77989b17223SAlexander Motin struct g_raid_subdisk *sd; 78089b17223SAlexander Motin struct bio_queue_head queue; 78189b17223SAlexander Motin struct bio *cbp; 78289b17223SAlexander Motin int i; 78389b17223SAlexander Motin 78489b17223SAlexander Motin vol = tr->tro_volume; 78589b17223SAlexander Motin sc = vol->v_softc; 78689b17223SAlexander Motin 78789b17223SAlexander Motin /* 78889b17223SAlexander Motin * Allocate all bios before sending any request, so we can return 78989b17223SAlexander Motin * ENOMEM in nice and clean way. 79089b17223SAlexander Motin */ 79189b17223SAlexander Motin bioq_init(&queue); 79289b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 79389b17223SAlexander Motin sd = &vol->v_subdisks[i]; 79489b17223SAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_NONE || 79589b17223SAlexander Motin sd->sd_state == G_RAID_SUBDISK_S_FAILED) 79689b17223SAlexander Motin continue; 79789b17223SAlexander Motin cbp = g_clone_bio(bp); 79889b17223SAlexander Motin if (cbp == NULL) 79989b17223SAlexander Motin goto failure; 80089b17223SAlexander Motin cbp->bio_caller1 = sd; 80189b17223SAlexander Motin bioq_insert_tail(&queue, cbp); 80289b17223SAlexander Motin } 80389b17223SAlexander Motin for (cbp = bioq_first(&queue); cbp != NULL; 80489b17223SAlexander Motin cbp = bioq_first(&queue)) { 80589b17223SAlexander Motin bioq_remove(&queue, cbp); 80689b17223SAlexander Motin sd = cbp->bio_caller1; 80789b17223SAlexander Motin cbp->bio_caller1 = NULL; 80889b17223SAlexander Motin g_raid_subdisk_iostart(sd, cbp); 80989b17223SAlexander Motin } 81089b17223SAlexander Motin return; 81189b17223SAlexander Motin failure: 81289b17223SAlexander Motin for (cbp = bioq_first(&queue); cbp != NULL; 81389b17223SAlexander Motin cbp = bioq_first(&queue)) { 81489b17223SAlexander Motin bioq_remove(&queue, cbp); 81589b17223SAlexander Motin g_destroy_bio(cbp); 81689b17223SAlexander Motin } 81789b17223SAlexander Motin if (bp->bio_error == 0) 81889b17223SAlexander Motin bp->bio_error = ENOMEM; 81989b17223SAlexander Motin g_raid_iodone(bp, bp->bio_error); 82089b17223SAlexander Motin } 82189b17223SAlexander Motin 82289b17223SAlexander Motin static void 82389b17223SAlexander Motin g_raid_tr_kerneldump_common_done(struct bio *bp) 82489b17223SAlexander Motin { 82589b17223SAlexander Motin 82689b17223SAlexander Motin bp->bio_flags |= BIO_DONE; 82789b17223SAlexander Motin } 82889b17223SAlexander Motin 82989b17223SAlexander Motin int 83089b17223SAlexander Motin g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, 83189b17223SAlexander Motin void *virtual, vm_offset_t physical, off_t offset, size_t length) 83289b17223SAlexander Motin { 83389b17223SAlexander Motin struct g_raid_softc *sc; 83489b17223SAlexander Motin struct g_raid_volume *vol; 83589b17223SAlexander Motin struct bio bp; 83689b17223SAlexander Motin 83789b17223SAlexander Motin vol = tr->tro_volume; 83889b17223SAlexander Motin sc = vol->v_softc; 83989b17223SAlexander Motin 84089b17223SAlexander Motin bzero(&bp, sizeof(bp)); 84189b17223SAlexander Motin bp.bio_cmd = BIO_WRITE; 84289b17223SAlexander Motin bp.bio_done = g_raid_tr_kerneldump_common_done; 84389b17223SAlexander Motin bp.bio_attribute = NULL; 84489b17223SAlexander Motin bp.bio_offset = offset; 84589b17223SAlexander Motin bp.bio_length = length; 84689b17223SAlexander Motin bp.bio_data = virtual; 84789b17223SAlexander Motin bp.bio_to = vol->v_provider; 84889b17223SAlexander Motin 84989b17223SAlexander Motin g_raid_start(&bp); 85089b17223SAlexander Motin while (!(bp.bio_flags & BIO_DONE)) { 85189b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Poll..."); 85289b17223SAlexander Motin g_raid_poll(sc); 85389b17223SAlexander Motin DELAY(10); 85489b17223SAlexander Motin } 85589b17223SAlexander Motin 85689b17223SAlexander Motin return (bp.bio_error != 0 ? EIO : 0); 85789b17223SAlexander Motin } 85889b17223SAlexander Motin 85989b17223SAlexander Motin static int 86089b17223SAlexander Motin g_raid_dump(void *arg, 86189b17223SAlexander Motin void *virtual, vm_offset_t physical, off_t offset, size_t length) 86289b17223SAlexander Motin { 86389b17223SAlexander Motin struct g_raid_volume *vol; 86489b17223SAlexander Motin int error; 86589b17223SAlexander Motin 86689b17223SAlexander Motin vol = (struct g_raid_volume *)arg; 86789b17223SAlexander Motin G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", 86889b17223SAlexander Motin (long long unsigned)offset, (long long unsigned)length); 86989b17223SAlexander Motin 87089b17223SAlexander Motin error = G_RAID_TR_KERNELDUMP(vol->v_tr, 87189b17223SAlexander Motin virtual, physical, offset, length); 87289b17223SAlexander Motin return (error); 87389b17223SAlexander Motin } 87489b17223SAlexander Motin 87589b17223SAlexander Motin static void 87689b17223SAlexander Motin g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) 87789b17223SAlexander Motin { 87889b17223SAlexander Motin struct g_kerneldump *gkd; 87989b17223SAlexander Motin struct g_provider *pp; 88089b17223SAlexander Motin struct g_raid_volume *vol; 88189b17223SAlexander Motin 88289b17223SAlexander Motin gkd = (struct g_kerneldump*)bp->bio_data; 88389b17223SAlexander Motin pp = bp->bio_to; 88489b17223SAlexander Motin vol = pp->private; 88589b17223SAlexander Motin g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", 88689b17223SAlexander Motin pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); 88789b17223SAlexander Motin gkd->di.dumper = g_raid_dump; 88889b17223SAlexander Motin gkd->di.priv = vol; 88989b17223SAlexander Motin gkd->di.blocksize = vol->v_sectorsize; 89089b17223SAlexander Motin gkd->di.maxiosize = DFLTPHYS; 89189b17223SAlexander Motin gkd->di.mediaoffset = gkd->offset; 89289b17223SAlexander Motin if ((gkd->offset + gkd->length) > vol->v_mediasize) 89389b17223SAlexander Motin gkd->length = vol->v_mediasize - gkd->offset; 89489b17223SAlexander Motin gkd->di.mediasize = gkd->length; 89589b17223SAlexander Motin g_io_deliver(bp, 0); 89689b17223SAlexander Motin } 89789b17223SAlexander Motin 89889b17223SAlexander Motin static void 89989b17223SAlexander Motin g_raid_start(struct bio *bp) 90089b17223SAlexander Motin { 90189b17223SAlexander Motin struct g_raid_softc *sc; 90289b17223SAlexander Motin 90389b17223SAlexander Motin sc = bp->bio_to->geom->softc; 90489b17223SAlexander Motin /* 90589b17223SAlexander Motin * If sc == NULL or there are no valid disks, provider's error 90689b17223SAlexander Motin * should be set and g_raid_start() should not be called at all. 90789b17223SAlexander Motin */ 90889b17223SAlexander Motin // KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, 90989b17223SAlexander Motin // ("Provider's error should be set (error=%d)(mirror=%s).", 91089b17223SAlexander Motin // bp->bio_to->error, bp->bio_to->name)); 91189b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Request received."); 91289b17223SAlexander Motin 91389b17223SAlexander Motin switch (bp->bio_cmd) { 91489b17223SAlexander Motin case BIO_READ: 91589b17223SAlexander Motin case BIO_WRITE: 91689b17223SAlexander Motin case BIO_DELETE: 91789b17223SAlexander Motin case BIO_FLUSH: 91889b17223SAlexander Motin break; 91989b17223SAlexander Motin case BIO_GETATTR: 92089b17223SAlexander Motin if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) 92189b17223SAlexander Motin g_raid_kerneldump(sc, bp); 92289b17223SAlexander Motin else 92389b17223SAlexander Motin g_io_deliver(bp, EOPNOTSUPP); 92489b17223SAlexander Motin return; 92589b17223SAlexander Motin default: 92689b17223SAlexander Motin g_io_deliver(bp, EOPNOTSUPP); 92789b17223SAlexander Motin return; 92889b17223SAlexander Motin } 92989b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 93089b17223SAlexander Motin bioq_disksort(&sc->sc_queue, bp); 93189b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 93289b17223SAlexander Motin if (!dumping) { 93389b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); 93489b17223SAlexander Motin wakeup(sc); 93589b17223SAlexander Motin } 93689b17223SAlexander Motin } 93789b17223SAlexander Motin 93889b17223SAlexander Motin static int 93989b17223SAlexander Motin g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) 94089b17223SAlexander Motin { 94189b17223SAlexander Motin /* 94289b17223SAlexander Motin * 5 cases: 94389b17223SAlexander Motin * (1) bp entirely below NO 94489b17223SAlexander Motin * (2) bp entirely above NO 94589b17223SAlexander Motin * (3) bp start below, but end in range YES 94689b17223SAlexander Motin * (4) bp entirely within YES 94789b17223SAlexander Motin * (5) bp starts within, ends above YES 94889b17223SAlexander Motin * 94989b17223SAlexander Motin * lock range 10-19 (offset 10 length 10) 95089b17223SAlexander Motin * (1) 1-5: first if kicks it out 95189b17223SAlexander Motin * (2) 30-35: second if kicks it out 95289b17223SAlexander Motin * (3) 5-15: passes both ifs 95389b17223SAlexander Motin * (4) 12-14: passes both ifs 95489b17223SAlexander Motin * (5) 19-20: passes both 95589b17223SAlexander Motin */ 95689b17223SAlexander Motin off_t lend = lstart + len - 1; 95789b17223SAlexander Motin off_t bstart = bp->bio_offset; 95889b17223SAlexander Motin off_t bend = bp->bio_offset + bp->bio_length - 1; 95989b17223SAlexander Motin 96089b17223SAlexander Motin if (bend < lstart) 96189b17223SAlexander Motin return (0); 96289b17223SAlexander Motin if (lend < bstart) 96389b17223SAlexander Motin return (0); 96489b17223SAlexander Motin return (1); 96589b17223SAlexander Motin } 96689b17223SAlexander Motin 96789b17223SAlexander Motin static int 96889b17223SAlexander Motin g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) 96989b17223SAlexander Motin { 97089b17223SAlexander Motin struct g_raid_lock *lp; 97189b17223SAlexander Motin 97289b17223SAlexander Motin sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); 97389b17223SAlexander Motin 97489b17223SAlexander Motin LIST_FOREACH(lp, &vol->v_locks, l_next) { 97589b17223SAlexander Motin if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) 97689b17223SAlexander Motin return (1); 97789b17223SAlexander Motin } 97889b17223SAlexander Motin return (0); 97989b17223SAlexander Motin } 98089b17223SAlexander Motin 98189b17223SAlexander Motin static void 98289b17223SAlexander Motin g_raid_start_request(struct bio *bp) 98389b17223SAlexander Motin { 98489b17223SAlexander Motin struct g_raid_softc *sc; 98589b17223SAlexander Motin struct g_raid_volume *vol; 98689b17223SAlexander Motin 98789b17223SAlexander Motin sc = bp->bio_to->geom->softc; 98889b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 98989b17223SAlexander Motin vol = bp->bio_to->private; 99089b17223SAlexander Motin 99189b17223SAlexander Motin /* 99289b17223SAlexander Motin * Check to see if this item is in a locked range. If so, 99389b17223SAlexander Motin * queue it to our locked queue and return. We'll requeue 99489b17223SAlexander Motin * it when the range is unlocked. Internal I/O for the 99589b17223SAlexander Motin * rebuild/rescan/recovery process is excluded from this 99689b17223SAlexander Motin * check so we can actually do the recovery. 99789b17223SAlexander Motin */ 99889b17223SAlexander Motin if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && 99989b17223SAlexander Motin g_raid_is_in_locked_range(vol, bp)) { 100089b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Defer request."); 100189b17223SAlexander Motin bioq_insert_tail(&vol->v_locked, bp); 100289b17223SAlexander Motin return; 100389b17223SAlexander Motin } 100489b17223SAlexander Motin 100589b17223SAlexander Motin /* 100689b17223SAlexander Motin * If we're actually going to do the write/delete, then 100789b17223SAlexander Motin * update the idle stats for the volume. 100889b17223SAlexander Motin */ 100989b17223SAlexander Motin if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 101089b17223SAlexander Motin if (!vol->v_dirty) 101189b17223SAlexander Motin g_raid_dirty(vol); 101289b17223SAlexander Motin vol->v_writes++; 101389b17223SAlexander Motin } 101489b17223SAlexander Motin 101589b17223SAlexander Motin /* 101689b17223SAlexander Motin * Put request onto inflight queue, so we can check if new 101789b17223SAlexander Motin * synchronization requests don't collide with it. Then tell 101889b17223SAlexander Motin * the transformation layer to start the I/O. 101989b17223SAlexander Motin */ 102089b17223SAlexander Motin bioq_insert_tail(&vol->v_inflight, bp); 102189b17223SAlexander Motin G_RAID_LOGREQ(4, bp, "Request started"); 102289b17223SAlexander Motin G_RAID_TR_IOSTART(vol->v_tr, bp); 102389b17223SAlexander Motin } 102489b17223SAlexander Motin 102589b17223SAlexander Motin static void 102689b17223SAlexander Motin g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) 102789b17223SAlexander Motin { 102889b17223SAlexander Motin off_t off, len; 102989b17223SAlexander Motin struct bio *nbp; 103089b17223SAlexander Motin struct g_raid_lock *lp; 103189b17223SAlexander Motin 103289b17223SAlexander Motin vol->v_pending_lock = 0; 103389b17223SAlexander Motin LIST_FOREACH(lp, &vol->v_locks, l_next) { 103489b17223SAlexander Motin if (lp->l_pending) { 103589b17223SAlexander Motin off = lp->l_offset; 103689b17223SAlexander Motin len = lp->l_length; 103789b17223SAlexander Motin lp->l_pending = 0; 103889b17223SAlexander Motin TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { 103989b17223SAlexander Motin if (g_raid_bio_overlaps(nbp, off, len)) 104089b17223SAlexander Motin lp->l_pending++; 104189b17223SAlexander Motin } 104289b17223SAlexander Motin if (lp->l_pending) { 104389b17223SAlexander Motin vol->v_pending_lock = 1; 104489b17223SAlexander Motin G_RAID_DEBUG1(4, vol->v_softc, 104589b17223SAlexander Motin "Deferred lock(%jd, %jd) has %d pending", 104689b17223SAlexander Motin (intmax_t)off, (intmax_t)(off + len), 104789b17223SAlexander Motin lp->l_pending); 104889b17223SAlexander Motin continue; 104989b17223SAlexander Motin } 105089b17223SAlexander Motin G_RAID_DEBUG1(4, vol->v_softc, 105189b17223SAlexander Motin "Deferred lock of %jd to %jd completed", 105289b17223SAlexander Motin (intmax_t)off, (intmax_t)(off + len)); 105389b17223SAlexander Motin G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 105489b17223SAlexander Motin } 105589b17223SAlexander Motin } 105689b17223SAlexander Motin } 105789b17223SAlexander Motin 105889b17223SAlexander Motin void 105989b17223SAlexander Motin g_raid_iodone(struct bio *bp, int error) 106089b17223SAlexander Motin { 106189b17223SAlexander Motin struct g_raid_softc *sc; 106289b17223SAlexander Motin struct g_raid_volume *vol; 106389b17223SAlexander Motin 106489b17223SAlexander Motin sc = bp->bio_to->geom->softc; 106589b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 106689b17223SAlexander Motin vol = bp->bio_to->private; 106789b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Request done: %d.", error); 106889b17223SAlexander Motin 106989b17223SAlexander Motin /* Update stats if we done write/delete. */ 107089b17223SAlexander Motin if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { 107189b17223SAlexander Motin vol->v_writes--; 107289b17223SAlexander Motin vol->v_last_write = time_uptime; 107389b17223SAlexander Motin } 107489b17223SAlexander Motin 107589b17223SAlexander Motin bioq_remove(&vol->v_inflight, bp); 107689b17223SAlexander Motin if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) 107789b17223SAlexander Motin g_raid_finish_with_locked_ranges(vol, bp); 107889b17223SAlexander Motin getmicrouptime(&vol->v_last_done); 107989b17223SAlexander Motin g_io_deliver(bp, error); 108089b17223SAlexander Motin } 108189b17223SAlexander Motin 108289b17223SAlexander Motin int 108389b17223SAlexander Motin g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, 108489b17223SAlexander Motin struct bio *ignore, void *argp) 108589b17223SAlexander Motin { 108689b17223SAlexander Motin struct g_raid_softc *sc; 108789b17223SAlexander Motin struct g_raid_lock *lp; 108889b17223SAlexander Motin struct bio *bp; 108989b17223SAlexander Motin 109089b17223SAlexander Motin sc = vol->v_softc; 109189b17223SAlexander Motin lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); 109289b17223SAlexander Motin LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); 109389b17223SAlexander Motin lp->l_offset = off; 109489b17223SAlexander Motin lp->l_length = len; 109589b17223SAlexander Motin lp->l_callback_arg = argp; 109689b17223SAlexander Motin 109789b17223SAlexander Motin lp->l_pending = 0; 109889b17223SAlexander Motin TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { 109989b17223SAlexander Motin if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) 110089b17223SAlexander Motin lp->l_pending++; 110189b17223SAlexander Motin } 110289b17223SAlexander Motin 110389b17223SAlexander Motin /* 110489b17223SAlexander Motin * If there are any writes that are pending, we return EBUSY. All 110589b17223SAlexander Motin * callers will have to wait until all pending writes clear. 110689b17223SAlexander Motin */ 110789b17223SAlexander Motin if (lp->l_pending > 0) { 110889b17223SAlexander Motin vol->v_pending_lock = 1; 110989b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", 111089b17223SAlexander Motin (intmax_t)off, (intmax_t)(off+len), lp->l_pending); 111189b17223SAlexander Motin return (EBUSY); 111289b17223SAlexander Motin } 111389b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", 111489b17223SAlexander Motin (intmax_t)off, (intmax_t)(off+len)); 111589b17223SAlexander Motin G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); 111689b17223SAlexander Motin return (0); 111789b17223SAlexander Motin } 111889b17223SAlexander Motin 111989b17223SAlexander Motin int 112089b17223SAlexander Motin g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) 112189b17223SAlexander Motin { 112289b17223SAlexander Motin struct g_raid_lock *lp; 112389b17223SAlexander Motin struct g_raid_softc *sc; 112489b17223SAlexander Motin struct bio *bp; 112589b17223SAlexander Motin 112689b17223SAlexander Motin sc = vol->v_softc; 112789b17223SAlexander Motin LIST_FOREACH(lp, &vol->v_locks, l_next) { 112889b17223SAlexander Motin if (lp->l_offset == off && lp->l_length == len) { 112989b17223SAlexander Motin LIST_REMOVE(lp, l_next); 113089b17223SAlexander Motin /* XXX 113189b17223SAlexander Motin * Right now we just put them all back on the queue 113289b17223SAlexander Motin * and hope for the best. We hope this because any 113389b17223SAlexander Motin * locked ranges will go right back on this list 113489b17223SAlexander Motin * when the worker thread runs. 113589b17223SAlexander Motin * XXX 113689b17223SAlexander Motin */ 113789b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", 113889b17223SAlexander Motin (intmax_t)lp->l_offset, 113989b17223SAlexander Motin (intmax_t)(lp->l_offset+lp->l_length)); 114089b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 114189b17223SAlexander Motin while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) 114289b17223SAlexander Motin bioq_disksort(&sc->sc_queue, bp); 114389b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 114489b17223SAlexander Motin free(lp, M_RAID); 114589b17223SAlexander Motin return (0); 114689b17223SAlexander Motin } 114789b17223SAlexander Motin } 114889b17223SAlexander Motin return (EINVAL); 114989b17223SAlexander Motin } 115089b17223SAlexander Motin 115189b17223SAlexander Motin void 115289b17223SAlexander Motin g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) 115389b17223SAlexander Motin { 115489b17223SAlexander Motin struct g_consumer *cp; 115589b17223SAlexander Motin struct g_raid_disk *disk, *tdisk; 115689b17223SAlexander Motin 115789b17223SAlexander Motin bp->bio_caller1 = sd; 115889b17223SAlexander Motin 115989b17223SAlexander Motin /* 116089b17223SAlexander Motin * Make sure that the disk is present. Generally it is a task of 116189b17223SAlexander Motin * transformation layers to not send requests to absent disks, but 116289b17223SAlexander Motin * it is better to be safe and report situation then sorry. 116389b17223SAlexander Motin */ 116489b17223SAlexander Motin if (sd->sd_disk == NULL) { 116589b17223SAlexander Motin G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); 116689b17223SAlexander Motin nodisk: 116789b17223SAlexander Motin bp->bio_from = NULL; 116889b17223SAlexander Motin bp->bio_to = NULL; 116989b17223SAlexander Motin bp->bio_error = ENXIO; 117089b17223SAlexander Motin g_raid_disk_done(bp); 117189b17223SAlexander Motin return; 117289b17223SAlexander Motin } 117389b17223SAlexander Motin disk = sd->sd_disk; 117489b17223SAlexander Motin if (disk->d_state != G_RAID_DISK_S_ACTIVE && 117589b17223SAlexander Motin disk->d_state != G_RAID_DISK_S_FAILED) { 117689b17223SAlexander Motin G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " 117789b17223SAlexander Motin "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 117889b17223SAlexander Motin goto nodisk; 117989b17223SAlexander Motin } 118089b17223SAlexander Motin 118189b17223SAlexander Motin cp = disk->d_consumer; 118289b17223SAlexander Motin bp->bio_from = cp; 118389b17223SAlexander Motin bp->bio_to = cp->provider; 118489b17223SAlexander Motin cp->index++; 118589b17223SAlexander Motin 118689b17223SAlexander Motin /* Update average disks load. */ 118789b17223SAlexander Motin TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { 118889b17223SAlexander Motin if (tdisk->d_consumer == NULL) 118989b17223SAlexander Motin tdisk->d_load = 0; 119089b17223SAlexander Motin else 119189b17223SAlexander Motin tdisk->d_load = (tdisk->d_consumer->index * 119289b17223SAlexander Motin G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; 119389b17223SAlexander Motin } 119489b17223SAlexander Motin 119589b17223SAlexander Motin disk->d_last_offset = bp->bio_offset + bp->bio_length; 119689b17223SAlexander Motin if (dumping) { 119789b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Sending dumping request."); 119889b17223SAlexander Motin if (bp->bio_cmd == BIO_WRITE) { 119989b17223SAlexander Motin bp->bio_error = g_raid_subdisk_kerneldump(sd, 120089b17223SAlexander Motin bp->bio_data, 0, bp->bio_offset, bp->bio_length); 120189b17223SAlexander Motin } else 120289b17223SAlexander Motin bp->bio_error = EOPNOTSUPP; 120389b17223SAlexander Motin g_raid_disk_done(bp); 120489b17223SAlexander Motin } else { 120589b17223SAlexander Motin bp->bio_done = g_raid_disk_done; 120689b17223SAlexander Motin bp->bio_offset += sd->sd_offset; 120789b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Sending request."); 120889b17223SAlexander Motin g_io_request(bp, cp); 120989b17223SAlexander Motin } 121089b17223SAlexander Motin } 121189b17223SAlexander Motin 121289b17223SAlexander Motin int 121389b17223SAlexander Motin g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, 121489b17223SAlexander Motin void *virtual, vm_offset_t physical, off_t offset, size_t length) 121589b17223SAlexander Motin { 121689b17223SAlexander Motin 121789b17223SAlexander Motin if (sd->sd_disk == NULL) 121889b17223SAlexander Motin return (ENXIO); 121989b17223SAlexander Motin if (sd->sd_disk->d_kd.di.dumper == NULL) 122089b17223SAlexander Motin return (EOPNOTSUPP); 122189b17223SAlexander Motin return (dump_write(&sd->sd_disk->d_kd.di, 122289b17223SAlexander Motin virtual, physical, 122389b17223SAlexander Motin sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, 122489b17223SAlexander Motin length)); 122589b17223SAlexander Motin } 122689b17223SAlexander Motin 122789b17223SAlexander Motin static void 122889b17223SAlexander Motin g_raid_disk_done(struct bio *bp) 122989b17223SAlexander Motin { 123089b17223SAlexander Motin struct g_raid_softc *sc; 123189b17223SAlexander Motin struct g_raid_subdisk *sd; 123289b17223SAlexander Motin 123389b17223SAlexander Motin sd = bp->bio_caller1; 123489b17223SAlexander Motin sc = sd->sd_softc; 123589b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 123689b17223SAlexander Motin bioq_disksort(&sc->sc_queue, bp); 123789b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 123889b17223SAlexander Motin if (!dumping) 123989b17223SAlexander Motin wakeup(sc); 124089b17223SAlexander Motin } 124189b17223SAlexander Motin 124289b17223SAlexander Motin static void 124389b17223SAlexander Motin g_raid_disk_done_request(struct bio *bp) 124489b17223SAlexander Motin { 124589b17223SAlexander Motin struct g_raid_softc *sc; 124689b17223SAlexander Motin struct g_raid_disk *disk; 124789b17223SAlexander Motin struct g_raid_subdisk *sd; 124889b17223SAlexander Motin struct g_raid_volume *vol; 124989b17223SAlexander Motin 125089b17223SAlexander Motin g_topology_assert_not(); 125189b17223SAlexander Motin 125289b17223SAlexander Motin G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); 125389b17223SAlexander Motin sd = bp->bio_caller1; 125489b17223SAlexander Motin sc = sd->sd_softc; 125589b17223SAlexander Motin vol = sd->sd_volume; 125689b17223SAlexander Motin if (bp->bio_from != NULL) { 125789b17223SAlexander Motin bp->bio_from->index--; 125889b17223SAlexander Motin disk = bp->bio_from->private; 125989b17223SAlexander Motin if (disk == NULL) 126089b17223SAlexander Motin g_raid_kill_consumer(sc, bp->bio_from); 126189b17223SAlexander Motin } 126289b17223SAlexander Motin bp->bio_offset -= sd->sd_offset; 126389b17223SAlexander Motin 126489b17223SAlexander Motin G_RAID_TR_IODONE(vol->v_tr, sd, bp); 126589b17223SAlexander Motin } 126689b17223SAlexander Motin 126789b17223SAlexander Motin static void 126889b17223SAlexander Motin g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) 126989b17223SAlexander Motin { 127089b17223SAlexander Motin 127189b17223SAlexander Motin if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) 127289b17223SAlexander Motin ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); 127389b17223SAlexander Motin else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) 127489b17223SAlexander Motin ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); 127589b17223SAlexander Motin else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) 127689b17223SAlexander Motin ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); 127789b17223SAlexander Motin else 127889b17223SAlexander Motin ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); 127989b17223SAlexander Motin if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { 128089b17223SAlexander Motin KASSERT(ep->e_error == 0, 128189b17223SAlexander Motin ("Error cannot be handled.")); 128289b17223SAlexander Motin g_raid_event_free(ep); 128389b17223SAlexander Motin } else { 128489b17223SAlexander Motin ep->e_flags |= G_RAID_EVENT_DONE; 128589b17223SAlexander Motin G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); 128689b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 128789b17223SAlexander Motin wakeup(ep); 128889b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 128989b17223SAlexander Motin } 129089b17223SAlexander Motin } 129189b17223SAlexander Motin 129289b17223SAlexander Motin /* 129389b17223SAlexander Motin * Worker thread. 129489b17223SAlexander Motin */ 129589b17223SAlexander Motin static void 129689b17223SAlexander Motin g_raid_worker(void *arg) 129789b17223SAlexander Motin { 129889b17223SAlexander Motin struct g_raid_softc *sc; 129989b17223SAlexander Motin struct g_raid_event *ep; 130089b17223SAlexander Motin struct g_raid_volume *vol; 130189b17223SAlexander Motin struct bio *bp; 130289b17223SAlexander Motin struct timeval now, t; 130389b17223SAlexander Motin int timeout, rv; 130489b17223SAlexander Motin 130589b17223SAlexander Motin sc = arg; 130689b17223SAlexander Motin thread_lock(curthread); 130789b17223SAlexander Motin sched_prio(curthread, PRIBIO); 130889b17223SAlexander Motin thread_unlock(curthread); 130989b17223SAlexander Motin 131089b17223SAlexander Motin sx_xlock(&sc->sc_lock); 131189b17223SAlexander Motin for (;;) { 131289b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 131389b17223SAlexander Motin /* 131489b17223SAlexander Motin * First take a look at events. 131589b17223SAlexander Motin * This is important to handle events before any I/O requests. 131689b17223SAlexander Motin */ 131789b17223SAlexander Motin bp = NULL; 131889b17223SAlexander Motin vol = NULL; 131989b17223SAlexander Motin rv = 0; 132089b17223SAlexander Motin ep = TAILQ_FIRST(&sc->sc_events); 132189b17223SAlexander Motin if (ep != NULL) 132289b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_events, ep, e_next); 132389b17223SAlexander Motin else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) 132489b17223SAlexander Motin ; 132589b17223SAlexander Motin else { 132689b17223SAlexander Motin getmicrouptime(&now); 132789b17223SAlexander Motin t = now; 132889b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 132989b17223SAlexander Motin if (bioq_first(&vol->v_inflight) == NULL && 133089b17223SAlexander Motin vol->v_tr && 133189b17223SAlexander Motin timevalcmp(&vol->v_last_done, &t, < )) 133289b17223SAlexander Motin t = vol->v_last_done; 133389b17223SAlexander Motin } 133489b17223SAlexander Motin timevalsub(&t, &now); 133589b17223SAlexander Motin timeout = g_raid_idle_threshold + 133689b17223SAlexander Motin t.tv_sec * 1000000 + t.tv_usec; 133789b17223SAlexander Motin if (timeout > 0) { 133889b17223SAlexander Motin /* 133989b17223SAlexander Motin * Two steps to avoid overflows at HZ=1000 134089b17223SAlexander Motin * and idle timeouts > 2.1s. Some rounding 134189b17223SAlexander Motin * errors can occur, but they are < 1tick, 134289b17223SAlexander Motin * which is deemed to be close enough for 134389b17223SAlexander Motin * this purpose. 134489b17223SAlexander Motin */ 134589b17223SAlexander Motin int micpertic = 1000000 / hz; 134689b17223SAlexander Motin timeout = (timeout + micpertic - 1) / micpertic; 134789b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 134889b17223SAlexander Motin MSLEEP(rv, sc, &sc->sc_queue_mtx, 134989b17223SAlexander Motin PRIBIO | PDROP, "-", timeout); 135089b17223SAlexander Motin sx_xlock(&sc->sc_lock); 135189b17223SAlexander Motin goto process; 135289b17223SAlexander Motin } else 135389b17223SAlexander Motin rv = EWOULDBLOCK; 135489b17223SAlexander Motin } 135589b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 135689b17223SAlexander Motin process: 135789b17223SAlexander Motin if (ep != NULL) { 135889b17223SAlexander Motin g_raid_handle_event(sc, ep); 135989b17223SAlexander Motin } else if (bp != NULL) { 136089b17223SAlexander Motin if (bp->bio_to != NULL && 136189b17223SAlexander Motin bp->bio_to->geom == sc->sc_geom) 136289b17223SAlexander Motin g_raid_start_request(bp); 136389b17223SAlexander Motin else 136489b17223SAlexander Motin g_raid_disk_done_request(bp); 136589b17223SAlexander Motin } else if (rv == EWOULDBLOCK) { 136689b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 136789b17223SAlexander Motin if (vol->v_writes == 0 && vol->v_dirty) 136889b17223SAlexander Motin g_raid_clean(vol, -1); 136989b17223SAlexander Motin if (bioq_first(&vol->v_inflight) == NULL && 137089b17223SAlexander Motin vol->v_tr) { 137189b17223SAlexander Motin t.tv_sec = g_raid_idle_threshold / 1000000; 137289b17223SAlexander Motin t.tv_usec = g_raid_idle_threshold % 1000000; 137389b17223SAlexander Motin timevaladd(&t, &vol->v_last_done); 137489b17223SAlexander Motin getmicrouptime(&now); 137589b17223SAlexander Motin if (timevalcmp(&t, &now, <= )) { 137689b17223SAlexander Motin G_RAID_TR_IDLE(vol->v_tr); 137789b17223SAlexander Motin vol->v_last_done = now; 137889b17223SAlexander Motin } 137989b17223SAlexander Motin } 138089b17223SAlexander Motin } 138189b17223SAlexander Motin } 138289b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_HARD) 138389b17223SAlexander Motin g_raid_destroy_node(sc, 1); /* May not return. */ 138489b17223SAlexander Motin } 138589b17223SAlexander Motin } 138689b17223SAlexander Motin 138789b17223SAlexander Motin static void 138889b17223SAlexander Motin g_raid_poll(struct g_raid_softc *sc) 138989b17223SAlexander Motin { 139089b17223SAlexander Motin struct g_raid_event *ep; 139189b17223SAlexander Motin struct bio *bp; 139289b17223SAlexander Motin 139389b17223SAlexander Motin sx_xlock(&sc->sc_lock); 139489b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 139589b17223SAlexander Motin /* 139689b17223SAlexander Motin * First take a look at events. 139789b17223SAlexander Motin * This is important to handle events before any I/O requests. 139889b17223SAlexander Motin */ 139989b17223SAlexander Motin ep = TAILQ_FIRST(&sc->sc_events); 140089b17223SAlexander Motin if (ep != NULL) { 140189b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_events, ep, e_next); 140289b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 140389b17223SAlexander Motin g_raid_handle_event(sc, ep); 140489b17223SAlexander Motin goto out; 140589b17223SAlexander Motin } 140689b17223SAlexander Motin bp = bioq_takefirst(&sc->sc_queue); 140789b17223SAlexander Motin if (bp != NULL) { 140889b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 140989b17223SAlexander Motin if (bp->bio_from == NULL || 141089b17223SAlexander Motin bp->bio_from->geom != sc->sc_geom) 141189b17223SAlexander Motin g_raid_start_request(bp); 141289b17223SAlexander Motin else 141389b17223SAlexander Motin g_raid_disk_done_request(bp); 141489b17223SAlexander Motin } 141589b17223SAlexander Motin out: 141689b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 141789b17223SAlexander Motin } 141889b17223SAlexander Motin 141989b17223SAlexander Motin static void 142089b17223SAlexander Motin g_raid_launch_provider(struct g_raid_volume *vol) 142189b17223SAlexander Motin { 142289b17223SAlexander Motin struct g_raid_disk *disk; 142389b17223SAlexander Motin struct g_raid_softc *sc; 142489b17223SAlexander Motin struct g_provider *pp; 142589b17223SAlexander Motin char name[G_RAID_MAX_VOLUMENAME]; 142689b17223SAlexander Motin off_t off; 142789b17223SAlexander Motin 142889b17223SAlexander Motin sc = vol->v_softc; 142989b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_LOCKED); 143089b17223SAlexander Motin 143189b17223SAlexander Motin g_topology_lock(); 143289b17223SAlexander Motin /* Try to name provider with volume name. */ 143389b17223SAlexander Motin snprintf(name, sizeof(name), "raid/%s", vol->v_name); 143489b17223SAlexander Motin if (g_raid_name_format == 0 || vol->v_name[0] == 0 || 143589b17223SAlexander Motin g_provider_by_name(name) != NULL) { 143689b17223SAlexander Motin /* Otherwise use sequential volume number. */ 143789b17223SAlexander Motin snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); 143889b17223SAlexander Motin } 143989b17223SAlexander Motin pp = g_new_providerf(sc->sc_geom, "%s", name); 144089b17223SAlexander Motin pp->private = vol; 144189b17223SAlexander Motin pp->mediasize = vol->v_mediasize; 144289b17223SAlexander Motin pp->sectorsize = vol->v_sectorsize; 144389b17223SAlexander Motin pp->stripesize = 0; 144489b17223SAlexander Motin pp->stripeoffset = 0; 144589b17223SAlexander Motin if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || 144689b17223SAlexander Motin vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || 144789b17223SAlexander Motin vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || 144889b17223SAlexander Motin vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { 144989b17223SAlexander Motin if ((disk = vol->v_subdisks[0].sd_disk) != NULL && 145089b17223SAlexander Motin disk->d_consumer != NULL && 145189b17223SAlexander Motin disk->d_consumer->provider != NULL) { 145289b17223SAlexander Motin pp->stripesize = disk->d_consumer->provider->stripesize; 145389b17223SAlexander Motin off = disk->d_consumer->provider->stripeoffset; 145489b17223SAlexander Motin pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; 145589b17223SAlexander Motin if (off > 0) 145689b17223SAlexander Motin pp->stripeoffset %= off; 145789b17223SAlexander Motin } 145889b17223SAlexander Motin if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { 145989b17223SAlexander Motin pp->stripesize *= (vol->v_disks_count - 1); 146089b17223SAlexander Motin pp->stripeoffset *= (vol->v_disks_count - 1); 146189b17223SAlexander Motin } 146289b17223SAlexander Motin } else 146389b17223SAlexander Motin pp->stripesize = vol->v_strip_size; 146489b17223SAlexander Motin vol->v_provider = pp; 146589b17223SAlexander Motin g_error_provider(pp, 0); 146689b17223SAlexander Motin g_topology_unlock(); 146789b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", 146889b17223SAlexander Motin pp->name, vol->v_name); 146989b17223SAlexander Motin } 147089b17223SAlexander Motin 147189b17223SAlexander Motin static void 147289b17223SAlexander Motin g_raid_destroy_provider(struct g_raid_volume *vol) 147389b17223SAlexander Motin { 147489b17223SAlexander Motin struct g_raid_softc *sc; 147589b17223SAlexander Motin struct g_provider *pp; 147689b17223SAlexander Motin struct bio *bp, *tmp; 147789b17223SAlexander Motin 147889b17223SAlexander Motin g_topology_assert_not(); 147989b17223SAlexander Motin sc = vol->v_softc; 148089b17223SAlexander Motin pp = vol->v_provider; 148189b17223SAlexander Motin KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); 148289b17223SAlexander Motin 148389b17223SAlexander Motin g_topology_lock(); 148489b17223SAlexander Motin g_error_provider(pp, ENXIO); 148589b17223SAlexander Motin mtx_lock(&sc->sc_queue_mtx); 148689b17223SAlexander Motin TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { 148789b17223SAlexander Motin if (bp->bio_to != pp) 148889b17223SAlexander Motin continue; 148989b17223SAlexander Motin bioq_remove(&sc->sc_queue, bp); 149089b17223SAlexander Motin g_io_deliver(bp, ENXIO); 149189b17223SAlexander Motin } 149289b17223SAlexander Motin mtx_unlock(&sc->sc_queue_mtx); 149389b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", 149489b17223SAlexander Motin pp->name, vol->v_name); 149589b17223SAlexander Motin g_wither_provider(pp, ENXIO); 149689b17223SAlexander Motin g_topology_unlock(); 149789b17223SAlexander Motin vol->v_provider = NULL; 149889b17223SAlexander Motin } 149989b17223SAlexander Motin 150089b17223SAlexander Motin /* 150189b17223SAlexander Motin * Update device state. 150289b17223SAlexander Motin */ 150389b17223SAlexander Motin static int 150489b17223SAlexander Motin g_raid_update_volume(struct g_raid_volume *vol, u_int event) 150589b17223SAlexander Motin { 150689b17223SAlexander Motin struct g_raid_softc *sc; 150789b17223SAlexander Motin 150889b17223SAlexander Motin sc = vol->v_softc; 150989b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 151089b17223SAlexander Motin 151189b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", 151289b17223SAlexander Motin g_raid_volume_event2str(event), 151389b17223SAlexander Motin vol->v_name); 151489b17223SAlexander Motin switch (event) { 151589b17223SAlexander Motin case G_RAID_VOLUME_E_DOWN: 151689b17223SAlexander Motin if (vol->v_provider != NULL) 151789b17223SAlexander Motin g_raid_destroy_provider(vol); 151889b17223SAlexander Motin break; 151989b17223SAlexander Motin case G_RAID_VOLUME_E_UP: 152089b17223SAlexander Motin if (vol->v_provider == NULL) 152189b17223SAlexander Motin g_raid_launch_provider(vol); 152289b17223SAlexander Motin break; 152389b17223SAlexander Motin case G_RAID_VOLUME_E_START: 152489b17223SAlexander Motin if (vol->v_tr) 152589b17223SAlexander Motin G_RAID_TR_START(vol->v_tr); 152689b17223SAlexander Motin return (0); 152789b17223SAlexander Motin default: 152889b17223SAlexander Motin if (sc->sc_md) 152989b17223SAlexander Motin G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); 153089b17223SAlexander Motin return (0); 153189b17223SAlexander Motin } 153289b17223SAlexander Motin 153389b17223SAlexander Motin /* Manage root mount release. */ 153489b17223SAlexander Motin if (vol->v_starting) { 153589b17223SAlexander Motin vol->v_starting = 0; 153689b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); 153789b17223SAlexander Motin root_mount_rel(vol->v_rootmount); 153889b17223SAlexander Motin vol->v_rootmount = NULL; 153989b17223SAlexander Motin } 154089b17223SAlexander Motin if (vol->v_stopping && vol->v_provider_open == 0) 154189b17223SAlexander Motin g_raid_destroy_volume(vol); 154289b17223SAlexander Motin return (0); 154389b17223SAlexander Motin } 154489b17223SAlexander Motin 154589b17223SAlexander Motin /* 154689b17223SAlexander Motin * Update subdisk state. 154789b17223SAlexander Motin */ 154889b17223SAlexander Motin static int 154989b17223SAlexander Motin g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) 155089b17223SAlexander Motin { 155189b17223SAlexander Motin struct g_raid_softc *sc; 155289b17223SAlexander Motin struct g_raid_volume *vol; 155389b17223SAlexander Motin 155489b17223SAlexander Motin sc = sd->sd_softc; 155589b17223SAlexander Motin vol = sd->sd_volume; 155689b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 155789b17223SAlexander Motin 155889b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", 155989b17223SAlexander Motin g_raid_subdisk_event2str(event), 156089b17223SAlexander Motin vol->v_name, sd->sd_pos, 156189b17223SAlexander Motin sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); 156289b17223SAlexander Motin if (vol->v_tr) 156389b17223SAlexander Motin G_RAID_TR_EVENT(vol->v_tr, sd, event); 156489b17223SAlexander Motin 156589b17223SAlexander Motin return (0); 156689b17223SAlexander Motin } 156789b17223SAlexander Motin 156889b17223SAlexander Motin /* 156989b17223SAlexander Motin * Update disk state. 157089b17223SAlexander Motin */ 157189b17223SAlexander Motin static int 157289b17223SAlexander Motin g_raid_update_disk(struct g_raid_disk *disk, u_int event) 157389b17223SAlexander Motin { 157489b17223SAlexander Motin struct g_raid_softc *sc; 157589b17223SAlexander Motin 157689b17223SAlexander Motin sc = disk->d_softc; 157789b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 157889b17223SAlexander Motin 157989b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", 158089b17223SAlexander Motin g_raid_disk_event2str(event), 158189b17223SAlexander Motin g_raid_get_diskname(disk)); 158289b17223SAlexander Motin 158389b17223SAlexander Motin if (sc->sc_md) 158489b17223SAlexander Motin G_RAID_MD_EVENT(sc->sc_md, disk, event); 158589b17223SAlexander Motin return (0); 158689b17223SAlexander Motin } 158789b17223SAlexander Motin 158889b17223SAlexander Motin /* 158989b17223SAlexander Motin * Node event. 159089b17223SAlexander Motin */ 159189b17223SAlexander Motin static int 159289b17223SAlexander Motin g_raid_update_node(struct g_raid_softc *sc, u_int event) 159389b17223SAlexander Motin { 159489b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 159589b17223SAlexander Motin 159689b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Event %s for the array.", 159789b17223SAlexander Motin g_raid_node_event2str(event)); 159889b17223SAlexander Motin 159989b17223SAlexander Motin if (event == G_RAID_NODE_E_WAKE) 160089b17223SAlexander Motin return (0); 160189b17223SAlexander Motin if (sc->sc_md) 160289b17223SAlexander Motin G_RAID_MD_EVENT(sc->sc_md, NULL, event); 160389b17223SAlexander Motin return (0); 160489b17223SAlexander Motin } 160589b17223SAlexander Motin 160689b17223SAlexander Motin static int 160789b17223SAlexander Motin g_raid_access(struct g_provider *pp, int acr, int acw, int ace) 160889b17223SAlexander Motin { 160989b17223SAlexander Motin struct g_raid_volume *vol; 161089b17223SAlexander Motin struct g_raid_softc *sc; 161114e2cd0aSAlexander Motin int dcw, opens, error = 0; 161289b17223SAlexander Motin 161389b17223SAlexander Motin g_topology_assert(); 161489b17223SAlexander Motin sc = pp->geom->softc; 161589b17223SAlexander Motin vol = pp->private; 161689b17223SAlexander Motin KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); 161789b17223SAlexander Motin KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); 161889b17223SAlexander Motin 161989b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, 162089b17223SAlexander Motin acr, acw, ace); 162189b17223SAlexander Motin dcw = pp->acw + acw; 162289b17223SAlexander Motin 162389b17223SAlexander Motin g_topology_unlock(); 162489b17223SAlexander Motin sx_xlock(&sc->sc_lock); 162589b17223SAlexander Motin /* Deny new opens while dying. */ 162689b17223SAlexander Motin if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { 162789b17223SAlexander Motin error = ENXIO; 162889b17223SAlexander Motin goto out; 162989b17223SAlexander Motin } 163089b17223SAlexander Motin if (dcw == 0 && vol->v_dirty) 163189b17223SAlexander Motin g_raid_clean(vol, dcw); 163289b17223SAlexander Motin vol->v_provider_open += acr + acw + ace; 163389b17223SAlexander Motin /* Handle delayed node destruction. */ 163489b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && 163589b17223SAlexander Motin vol->v_provider_open == 0) { 163689b17223SAlexander Motin /* Count open volumes. */ 163789b17223SAlexander Motin opens = g_raid_nopens(sc); 163889b17223SAlexander Motin if (opens == 0) { 163989b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_HARD; 164089b17223SAlexander Motin /* Wake up worker to make it selfdestruct. */ 164189b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 164289b17223SAlexander Motin } 164389b17223SAlexander Motin } 164489b17223SAlexander Motin /* Handle open volume destruction. */ 164589b17223SAlexander Motin if (vol->v_stopping && vol->v_provider_open == 0) 164689b17223SAlexander Motin g_raid_destroy_volume(vol); 164789b17223SAlexander Motin out: 164889b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 164989b17223SAlexander Motin g_topology_lock(); 165089b17223SAlexander Motin return (error); 165189b17223SAlexander Motin } 165289b17223SAlexander Motin 165389b17223SAlexander Motin struct g_raid_softc * 165489b17223SAlexander Motin g_raid_create_node(struct g_class *mp, 165589b17223SAlexander Motin const char *name, struct g_raid_md_object *md) 165689b17223SAlexander Motin { 165789b17223SAlexander Motin struct g_raid_softc *sc; 165889b17223SAlexander Motin struct g_geom *gp; 165989b17223SAlexander Motin int error; 166089b17223SAlexander Motin 166189b17223SAlexander Motin g_topology_assert(); 166289b17223SAlexander Motin G_RAID_DEBUG(1, "Creating array %s.", name); 166389b17223SAlexander Motin 166489b17223SAlexander Motin gp = g_new_geomf(mp, "%s", name); 166589b17223SAlexander Motin sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); 166689b17223SAlexander Motin gp->start = g_raid_start; 166789b17223SAlexander Motin gp->orphan = g_raid_orphan; 166889b17223SAlexander Motin gp->access = g_raid_access; 166989b17223SAlexander Motin gp->dumpconf = g_raid_dumpconf; 167089b17223SAlexander Motin 167189b17223SAlexander Motin sc->sc_md = md; 167289b17223SAlexander Motin sc->sc_geom = gp; 167389b17223SAlexander Motin sc->sc_flags = 0; 167489b17223SAlexander Motin TAILQ_INIT(&sc->sc_volumes); 167589b17223SAlexander Motin TAILQ_INIT(&sc->sc_disks); 167689b17223SAlexander Motin sx_init(&sc->sc_lock, "gmirror:lock"); 167789b17223SAlexander Motin mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); 167889b17223SAlexander Motin TAILQ_INIT(&sc->sc_events); 167989b17223SAlexander Motin bioq_init(&sc->sc_queue); 168089b17223SAlexander Motin gp->softc = sc; 168189b17223SAlexander Motin error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, 168289b17223SAlexander Motin "g_raid %s", name); 168389b17223SAlexander Motin if (error != 0) { 168489b17223SAlexander Motin G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); 168589b17223SAlexander Motin mtx_destroy(&sc->sc_queue_mtx); 168689b17223SAlexander Motin sx_destroy(&sc->sc_lock); 168789b17223SAlexander Motin g_destroy_geom(sc->sc_geom); 168889b17223SAlexander Motin free(sc, M_RAID); 168989b17223SAlexander Motin return (NULL); 169089b17223SAlexander Motin } 169189b17223SAlexander Motin 169289b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Array %s created.", name); 169389b17223SAlexander Motin return (sc); 169489b17223SAlexander Motin } 169589b17223SAlexander Motin 169689b17223SAlexander Motin struct g_raid_volume * 169789b17223SAlexander Motin g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) 169889b17223SAlexander Motin { 169989b17223SAlexander Motin struct g_raid_volume *vol, *vol1; 170089b17223SAlexander Motin int i; 170189b17223SAlexander Motin 170289b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); 170389b17223SAlexander Motin vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); 170489b17223SAlexander Motin vol->v_softc = sc; 170589b17223SAlexander Motin strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); 170689b17223SAlexander Motin vol->v_state = G_RAID_VOLUME_S_STARTING; 170789b17223SAlexander Motin vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; 170889b17223SAlexander Motin vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; 170989b17223SAlexander Motin bioq_init(&vol->v_inflight); 171089b17223SAlexander Motin bioq_init(&vol->v_locked); 171189b17223SAlexander Motin LIST_INIT(&vol->v_locks); 171289b17223SAlexander Motin for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 171389b17223SAlexander Motin vol->v_subdisks[i].sd_softc = sc; 171489b17223SAlexander Motin vol->v_subdisks[i].sd_volume = vol; 171589b17223SAlexander Motin vol->v_subdisks[i].sd_pos = i; 171689b17223SAlexander Motin vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; 171789b17223SAlexander Motin } 171889b17223SAlexander Motin 171989b17223SAlexander Motin /* Find free ID for this volume. */ 172089b17223SAlexander Motin g_topology_lock(); 172189b17223SAlexander Motin vol1 = vol; 172289b17223SAlexander Motin if (id >= 0) { 172389b17223SAlexander Motin LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 172489b17223SAlexander Motin if (vol1->v_global_id == id) 172589b17223SAlexander Motin break; 172689b17223SAlexander Motin } 172789b17223SAlexander Motin } 172889b17223SAlexander Motin if (vol1 != NULL) { 172989b17223SAlexander Motin for (id = 0; ; id++) { 173089b17223SAlexander Motin LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { 173189b17223SAlexander Motin if (vol1->v_global_id == id) 173289b17223SAlexander Motin break; 173389b17223SAlexander Motin } 173489b17223SAlexander Motin if (vol1 == NULL) 173589b17223SAlexander Motin break; 173689b17223SAlexander Motin } 173789b17223SAlexander Motin } 173889b17223SAlexander Motin vol->v_global_id = id; 173989b17223SAlexander Motin LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); 174089b17223SAlexander Motin g_topology_unlock(); 174189b17223SAlexander Motin 174289b17223SAlexander Motin /* Delay root mounting. */ 174389b17223SAlexander Motin vol->v_rootmount = root_mount_hold("GRAID"); 174489b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); 174589b17223SAlexander Motin vol->v_starting = 1; 174689b17223SAlexander Motin TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); 174789b17223SAlexander Motin return (vol); 174889b17223SAlexander Motin } 174989b17223SAlexander Motin 175089b17223SAlexander Motin struct g_raid_disk * 175189b17223SAlexander Motin g_raid_create_disk(struct g_raid_softc *sc) 175289b17223SAlexander Motin { 175389b17223SAlexander Motin struct g_raid_disk *disk; 175489b17223SAlexander Motin 175589b17223SAlexander Motin G_RAID_DEBUG1(1, sc, "Creating disk."); 175689b17223SAlexander Motin disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); 175789b17223SAlexander Motin disk->d_softc = sc; 175889b17223SAlexander Motin disk->d_state = G_RAID_DISK_S_NONE; 175989b17223SAlexander Motin TAILQ_INIT(&disk->d_subdisks); 176089b17223SAlexander Motin TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); 176189b17223SAlexander Motin return (disk); 176289b17223SAlexander Motin } 176389b17223SAlexander Motin 176489b17223SAlexander Motin int g_raid_start_volume(struct g_raid_volume *vol) 176589b17223SAlexander Motin { 176689b17223SAlexander Motin struct g_raid_tr_class *class; 176789b17223SAlexander Motin struct g_raid_tr_object *obj; 176889b17223SAlexander Motin int status; 176989b17223SAlexander Motin 177089b17223SAlexander Motin G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); 177189b17223SAlexander Motin LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { 177289b17223SAlexander Motin G_RAID_DEBUG1(2, vol->v_softc, 177389b17223SAlexander Motin "Tasting volume %s for %s transformation.", 177489b17223SAlexander Motin vol->v_name, class->name); 177589b17223SAlexander Motin obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 177689b17223SAlexander Motin M_WAITOK); 177789b17223SAlexander Motin obj->tro_class = class; 177889b17223SAlexander Motin obj->tro_volume = vol; 177989b17223SAlexander Motin status = G_RAID_TR_TASTE(obj, vol); 178089b17223SAlexander Motin if (status != G_RAID_TR_TASTE_FAIL) 178189b17223SAlexander Motin break; 178289b17223SAlexander Motin kobj_delete((kobj_t)obj, M_RAID); 178389b17223SAlexander Motin } 178489b17223SAlexander Motin if (class == NULL) { 178589b17223SAlexander Motin G_RAID_DEBUG1(0, vol->v_softc, 178689b17223SAlexander Motin "No transformation module found for %s.", 178789b17223SAlexander Motin vol->v_name); 178889b17223SAlexander Motin vol->v_tr = NULL; 178989b17223SAlexander Motin g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); 179089b17223SAlexander Motin g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, 179189b17223SAlexander Motin G_RAID_EVENT_VOLUME); 179289b17223SAlexander Motin return (-1); 179389b17223SAlexander Motin } 179489b17223SAlexander Motin G_RAID_DEBUG1(2, vol->v_softc, 179589b17223SAlexander Motin "Transformation module %s chosen for %s.", 179689b17223SAlexander Motin class->name, vol->v_name); 179789b17223SAlexander Motin vol->v_tr = obj; 179889b17223SAlexander Motin return (0); 179989b17223SAlexander Motin } 180089b17223SAlexander Motin 180189b17223SAlexander Motin int 180289b17223SAlexander Motin g_raid_destroy_node(struct g_raid_softc *sc, int worker) 180389b17223SAlexander Motin { 180489b17223SAlexander Motin struct g_raid_volume *vol, *tmpv; 180589b17223SAlexander Motin struct g_raid_disk *disk, *tmpd; 180689b17223SAlexander Motin int error = 0; 180789b17223SAlexander Motin 180889b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_HARD; 180989b17223SAlexander Motin TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { 181089b17223SAlexander Motin if (g_raid_destroy_volume(vol)) 181189b17223SAlexander Motin error = EBUSY; 181289b17223SAlexander Motin } 181389b17223SAlexander Motin if (error) 181489b17223SAlexander Motin return (error); 181589b17223SAlexander Motin TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { 181689b17223SAlexander Motin if (g_raid_destroy_disk(disk)) 181789b17223SAlexander Motin error = EBUSY; 181889b17223SAlexander Motin } 181989b17223SAlexander Motin if (error) 182089b17223SAlexander Motin return (error); 182189b17223SAlexander Motin if (sc->sc_md) { 182289b17223SAlexander Motin G_RAID_MD_FREE(sc->sc_md); 182389b17223SAlexander Motin kobj_delete((kobj_t)sc->sc_md, M_RAID); 182489b17223SAlexander Motin sc->sc_md = NULL; 182589b17223SAlexander Motin } 182689b17223SAlexander Motin if (sc->sc_geom != NULL) { 182789b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); 182889b17223SAlexander Motin g_topology_lock(); 182989b17223SAlexander Motin sc->sc_geom->softc = NULL; 183089b17223SAlexander Motin g_wither_geom(sc->sc_geom, ENXIO); 183189b17223SAlexander Motin g_topology_unlock(); 183289b17223SAlexander Motin sc->sc_geom = NULL; 183389b17223SAlexander Motin } else 183489b17223SAlexander Motin G_RAID_DEBUG(1, "Array destroyed."); 183589b17223SAlexander Motin if (worker) { 183689b17223SAlexander Motin g_raid_event_cancel(sc, sc); 183789b17223SAlexander Motin mtx_destroy(&sc->sc_queue_mtx); 183889b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 183989b17223SAlexander Motin sx_destroy(&sc->sc_lock); 184089b17223SAlexander Motin wakeup(&sc->sc_stopping); 184189b17223SAlexander Motin free(sc, M_RAID); 184289b17223SAlexander Motin curthread->td_pflags &= ~TDP_GEOM; 184389b17223SAlexander Motin G_RAID_DEBUG(1, "Thread exiting."); 184489b17223SAlexander Motin kproc_exit(0); 184589b17223SAlexander Motin } else { 184689b17223SAlexander Motin /* Wake up worker to make it selfdestruct. */ 184789b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 184889b17223SAlexander Motin } 184989b17223SAlexander Motin return (0); 185089b17223SAlexander Motin } 185189b17223SAlexander Motin 185289b17223SAlexander Motin int 185389b17223SAlexander Motin g_raid_destroy_volume(struct g_raid_volume *vol) 185489b17223SAlexander Motin { 185589b17223SAlexander Motin struct g_raid_softc *sc; 185689b17223SAlexander Motin struct g_raid_disk *disk; 185789b17223SAlexander Motin int i; 185889b17223SAlexander Motin 185989b17223SAlexander Motin sc = vol->v_softc; 186089b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); 186189b17223SAlexander Motin vol->v_stopping = 1; 186289b17223SAlexander Motin if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { 186389b17223SAlexander Motin if (vol->v_tr) { 186489b17223SAlexander Motin G_RAID_TR_STOP(vol->v_tr); 186589b17223SAlexander Motin return (EBUSY); 186689b17223SAlexander Motin } else 186789b17223SAlexander Motin vol->v_state = G_RAID_VOLUME_S_STOPPED; 186889b17223SAlexander Motin } 186989b17223SAlexander Motin if (g_raid_event_check(sc, vol) != 0) 187089b17223SAlexander Motin return (EBUSY); 187189b17223SAlexander Motin if (vol->v_provider != NULL) 187289b17223SAlexander Motin return (EBUSY); 187389b17223SAlexander Motin if (vol->v_provider_open != 0) 187489b17223SAlexander Motin return (EBUSY); 187589b17223SAlexander Motin if (vol->v_tr) { 187689b17223SAlexander Motin G_RAID_TR_FREE(vol->v_tr); 187789b17223SAlexander Motin kobj_delete((kobj_t)vol->v_tr, M_RAID); 187889b17223SAlexander Motin vol->v_tr = NULL; 187989b17223SAlexander Motin } 188089b17223SAlexander Motin if (vol->v_rootmount) 188189b17223SAlexander Motin root_mount_rel(vol->v_rootmount); 188289b17223SAlexander Motin g_topology_lock(); 188389b17223SAlexander Motin LIST_REMOVE(vol, v_global_next); 188489b17223SAlexander Motin g_topology_unlock(); 188589b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); 188689b17223SAlexander Motin for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { 188789b17223SAlexander Motin g_raid_event_cancel(sc, &vol->v_subdisks[i]); 188889b17223SAlexander Motin disk = vol->v_subdisks[i].sd_disk; 188989b17223SAlexander Motin if (disk == NULL) 189089b17223SAlexander Motin continue; 189189b17223SAlexander Motin TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); 189289b17223SAlexander Motin } 189389b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); 189489b17223SAlexander Motin if (sc->sc_md) 189589b17223SAlexander Motin G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); 189689b17223SAlexander Motin g_raid_event_cancel(sc, vol); 189789b17223SAlexander Motin free(vol, M_RAID); 189889b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_HARD) { 189989b17223SAlexander Motin /* Wake up worker to let it selfdestruct. */ 190089b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 190189b17223SAlexander Motin } 190289b17223SAlexander Motin return (0); 190389b17223SAlexander Motin } 190489b17223SAlexander Motin 190589b17223SAlexander Motin int 190689b17223SAlexander Motin g_raid_destroy_disk(struct g_raid_disk *disk) 190789b17223SAlexander Motin { 190889b17223SAlexander Motin struct g_raid_softc *sc; 190989b17223SAlexander Motin struct g_raid_subdisk *sd, *tmp; 191089b17223SAlexander Motin 191189b17223SAlexander Motin sc = disk->d_softc; 191289b17223SAlexander Motin G_RAID_DEBUG1(2, sc, "Destroying disk."); 191389b17223SAlexander Motin if (disk->d_consumer) { 191489b17223SAlexander Motin g_raid_kill_consumer(sc, disk->d_consumer); 191589b17223SAlexander Motin disk->d_consumer = NULL; 191689b17223SAlexander Motin } 191789b17223SAlexander Motin TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { 191889b17223SAlexander Motin g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); 191989b17223SAlexander Motin g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, 192089b17223SAlexander Motin G_RAID_EVENT_SUBDISK); 192189b17223SAlexander Motin TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); 192289b17223SAlexander Motin sd->sd_disk = NULL; 192389b17223SAlexander Motin } 192489b17223SAlexander Motin TAILQ_REMOVE(&sc->sc_disks, disk, d_next); 192589b17223SAlexander Motin if (sc->sc_md) 192689b17223SAlexander Motin G_RAID_MD_FREE_DISK(sc->sc_md, disk); 192789b17223SAlexander Motin g_raid_event_cancel(sc, disk); 192889b17223SAlexander Motin free(disk, M_RAID); 192989b17223SAlexander Motin return (0); 193089b17223SAlexander Motin } 193189b17223SAlexander Motin 193289b17223SAlexander Motin int 193389b17223SAlexander Motin g_raid_destroy(struct g_raid_softc *sc, int how) 193489b17223SAlexander Motin { 193589b17223SAlexander Motin int opens; 193689b17223SAlexander Motin 193789b17223SAlexander Motin g_topology_assert_not(); 193889b17223SAlexander Motin if (sc == NULL) 193989b17223SAlexander Motin return (ENXIO); 194089b17223SAlexander Motin sx_assert(&sc->sc_lock, SX_XLOCKED); 194189b17223SAlexander Motin 194289b17223SAlexander Motin /* Count open volumes. */ 194389b17223SAlexander Motin opens = g_raid_nopens(sc); 194489b17223SAlexander Motin 194589b17223SAlexander Motin /* React on some opened volumes. */ 194689b17223SAlexander Motin if (opens > 0) { 194789b17223SAlexander Motin switch (how) { 194889b17223SAlexander Motin case G_RAID_DESTROY_SOFT: 194989b17223SAlexander Motin G_RAID_DEBUG1(1, sc, 195089b17223SAlexander Motin "%d volumes are still open.", 195189b17223SAlexander Motin opens); 195289b17223SAlexander Motin return (EBUSY); 195389b17223SAlexander Motin case G_RAID_DESTROY_DELAYED: 195489b17223SAlexander Motin G_RAID_DEBUG1(1, sc, 195589b17223SAlexander Motin "Array will be destroyed on last close."); 195689b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_DELAYED; 195789b17223SAlexander Motin return (EBUSY); 195889b17223SAlexander Motin case G_RAID_DESTROY_HARD: 195989b17223SAlexander Motin G_RAID_DEBUG1(1, sc, 196089b17223SAlexander Motin "%d volumes are still open.", 196189b17223SAlexander Motin opens); 196289b17223SAlexander Motin } 196389b17223SAlexander Motin } 196489b17223SAlexander Motin 196589b17223SAlexander Motin /* Mark node for destruction. */ 196689b17223SAlexander Motin sc->sc_stopping = G_RAID_DESTROY_HARD; 196789b17223SAlexander Motin /* Wake up worker to let it selfdestruct. */ 196889b17223SAlexander Motin g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); 196989b17223SAlexander Motin /* Sleep until node destroyed. */ 197089b17223SAlexander Motin sx_sleep(&sc->sc_stopping, &sc->sc_lock, 197189b17223SAlexander Motin PRIBIO | PDROP, "r:destroy", 0); 197289b17223SAlexander Motin return (0); 197389b17223SAlexander Motin } 197489b17223SAlexander Motin 197589b17223SAlexander Motin static void 197689b17223SAlexander Motin g_raid_taste_orphan(struct g_consumer *cp) 197789b17223SAlexander Motin { 197889b17223SAlexander Motin 197989b17223SAlexander Motin KASSERT(1 == 0, ("%s called while tasting %s.", __func__, 198089b17223SAlexander Motin cp->provider->name)); 198189b17223SAlexander Motin } 198289b17223SAlexander Motin 198389b17223SAlexander Motin static struct g_geom * 198489b17223SAlexander Motin g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) 198589b17223SAlexander Motin { 198689b17223SAlexander Motin struct g_consumer *cp; 198789b17223SAlexander Motin struct g_geom *gp, *geom; 198889b17223SAlexander Motin struct g_raid_md_class *class; 198989b17223SAlexander Motin struct g_raid_md_object *obj; 199089b17223SAlexander Motin int status; 199189b17223SAlexander Motin 199289b17223SAlexander Motin g_topology_assert(); 199389b17223SAlexander Motin g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); 199489b17223SAlexander Motin G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); 199589b17223SAlexander Motin 199689b17223SAlexander Motin gp = g_new_geomf(mp, "mirror:taste"); 199789b17223SAlexander Motin /* 199889b17223SAlexander Motin * This orphan function should be never called. 199989b17223SAlexander Motin */ 200089b17223SAlexander Motin gp->orphan = g_raid_taste_orphan; 200189b17223SAlexander Motin cp = g_new_consumer(gp); 200289b17223SAlexander Motin g_attach(cp, pp); 200389b17223SAlexander Motin 200489b17223SAlexander Motin geom = NULL; 200589b17223SAlexander Motin LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 200689b17223SAlexander Motin G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", 200789b17223SAlexander Motin pp->name, class->name); 200889b17223SAlexander Motin obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 200989b17223SAlexander Motin M_WAITOK); 201089b17223SAlexander Motin obj->mdo_class = class; 201189b17223SAlexander Motin status = G_RAID_MD_TASTE(obj, mp, cp, &geom); 201289b17223SAlexander Motin if (status != G_RAID_MD_TASTE_NEW) 201389b17223SAlexander Motin kobj_delete((kobj_t)obj, M_RAID); 201489b17223SAlexander Motin if (status != G_RAID_MD_TASTE_FAIL) 201589b17223SAlexander Motin break; 201689b17223SAlexander Motin } 201789b17223SAlexander Motin 201889b17223SAlexander Motin g_detach(cp); 201989b17223SAlexander Motin g_destroy_consumer(cp); 202089b17223SAlexander Motin g_destroy_geom(gp); 202189b17223SAlexander Motin G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); 202289b17223SAlexander Motin return (geom); 202389b17223SAlexander Motin } 202489b17223SAlexander Motin 202589b17223SAlexander Motin int 202689b17223SAlexander Motin g_raid_create_node_format(const char *format, struct g_geom **gp) 202789b17223SAlexander Motin { 202889b17223SAlexander Motin struct g_raid_md_class *class; 202989b17223SAlexander Motin struct g_raid_md_object *obj; 203089b17223SAlexander Motin int status; 203189b17223SAlexander Motin 203289b17223SAlexander Motin G_RAID_DEBUG(2, "Creating array for %s metadata.", format); 203389b17223SAlexander Motin LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { 203489b17223SAlexander Motin if (strcasecmp(class->name, format) == 0) 203589b17223SAlexander Motin break; 203689b17223SAlexander Motin } 203789b17223SAlexander Motin if (class == NULL) { 203889b17223SAlexander Motin G_RAID_DEBUG(1, "No support for %s metadata.", format); 203989b17223SAlexander Motin return (G_RAID_MD_TASTE_FAIL); 204089b17223SAlexander Motin } 204189b17223SAlexander Motin obj = (void *)kobj_create((kobj_class_t)class, M_RAID, 204289b17223SAlexander Motin M_WAITOK); 204389b17223SAlexander Motin obj->mdo_class = class; 204489b17223SAlexander Motin status = G_RAID_MD_CREATE(obj, &g_raid_class, gp); 204589b17223SAlexander Motin if (status != G_RAID_MD_TASTE_NEW) 204689b17223SAlexander Motin kobj_delete((kobj_t)obj, M_RAID); 204789b17223SAlexander Motin return (status); 204889b17223SAlexander Motin } 204989b17223SAlexander Motin 205089b17223SAlexander Motin static int 205189b17223SAlexander Motin g_raid_destroy_geom(struct gctl_req *req __unused, 205289b17223SAlexander Motin struct g_class *mp __unused, struct g_geom *gp) 205389b17223SAlexander Motin { 205489b17223SAlexander Motin struct g_raid_softc *sc; 205589b17223SAlexander Motin int error; 205689b17223SAlexander Motin 205789b17223SAlexander Motin g_topology_unlock(); 205889b17223SAlexander Motin sc = gp->softc; 205989b17223SAlexander Motin sx_xlock(&sc->sc_lock); 206089b17223SAlexander Motin g_cancel_event(sc); 206189b17223SAlexander Motin error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); 206289b17223SAlexander Motin if (error != 0) 206389b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 206489b17223SAlexander Motin g_topology_lock(); 206589b17223SAlexander Motin return (error); 206689b17223SAlexander Motin } 206789b17223SAlexander Motin 206889b17223SAlexander Motin void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, 206989b17223SAlexander Motin struct g_raid_subdisk *sd, struct g_raid_disk *disk) 207089b17223SAlexander Motin { 207189b17223SAlexander Motin 207289b17223SAlexander Motin if (sc->sc_stopping == G_RAID_DESTROY_HARD) 207389b17223SAlexander Motin return; 207489b17223SAlexander Motin if (sc->sc_md) 207589b17223SAlexander Motin G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); 207689b17223SAlexander Motin } 207789b17223SAlexander Motin 207889b17223SAlexander Motin void g_raid_fail_disk(struct g_raid_softc *sc, 207989b17223SAlexander Motin struct g_raid_subdisk *sd, struct g_raid_disk *disk) 208089b17223SAlexander Motin { 208189b17223SAlexander Motin 208289b17223SAlexander Motin if (disk == NULL) 208389b17223SAlexander Motin disk = sd->sd_disk; 208489b17223SAlexander Motin if (disk == NULL) { 208589b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); 208689b17223SAlexander Motin return; 208789b17223SAlexander Motin } 208889b17223SAlexander Motin if (disk->d_state != G_RAID_DISK_S_ACTIVE) { 208989b17223SAlexander Motin G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " 209089b17223SAlexander Motin "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); 209189b17223SAlexander Motin return; 209289b17223SAlexander Motin } 209389b17223SAlexander Motin if (sc->sc_md) 209489b17223SAlexander Motin G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); 209589b17223SAlexander Motin } 209689b17223SAlexander Motin 209789b17223SAlexander Motin static void 209889b17223SAlexander Motin g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, 209989b17223SAlexander Motin struct g_consumer *cp, struct g_provider *pp) 210089b17223SAlexander Motin { 210189b17223SAlexander Motin struct g_raid_softc *sc; 210289b17223SAlexander Motin struct g_raid_volume *vol; 210389b17223SAlexander Motin struct g_raid_subdisk *sd; 210489b17223SAlexander Motin struct g_raid_disk *disk; 210589b17223SAlexander Motin int i, s; 210689b17223SAlexander Motin 210789b17223SAlexander Motin g_topology_assert(); 210889b17223SAlexander Motin 210989b17223SAlexander Motin sc = gp->softc; 211089b17223SAlexander Motin if (sc == NULL) 211189b17223SAlexander Motin return; 211289b17223SAlexander Motin if (pp != NULL) { 211389b17223SAlexander Motin vol = pp->private; 211489b17223SAlexander Motin g_topology_unlock(); 211589b17223SAlexander Motin sx_xlock(&sc->sc_lock); 211689b17223SAlexander Motin sbuf_printf(sb, "%s<Label>%s</Label>\n", indent, 211789b17223SAlexander Motin vol->v_name); 211889b17223SAlexander Motin sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent, 211989b17223SAlexander Motin g_raid_volume_level2str(vol->v_raid_level, 212089b17223SAlexander Motin vol->v_raid_level_qualifier)); 212189b17223SAlexander Motin sbuf_printf(sb, 212289b17223SAlexander Motin "%s<Transformation>%s</Transformation>\n", indent, 212389b17223SAlexander Motin vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); 212489b17223SAlexander Motin sbuf_printf(sb, "%s<Components>%u</Components>\n", indent, 212589b17223SAlexander Motin vol->v_disks_count); 212689b17223SAlexander Motin sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent, 212789b17223SAlexander Motin vol->v_strip_size); 212889b17223SAlexander Motin sbuf_printf(sb, "%s<State>%s</State>\n", indent, 212989b17223SAlexander Motin g_raid_volume_state2str(vol->v_state)); 213089b17223SAlexander Motin sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent, 213189b17223SAlexander Motin vol->v_dirty ? "Yes" : "No"); 213289b17223SAlexander Motin sbuf_printf(sb, "%s<Subdisks>", indent); 213389b17223SAlexander Motin for (i = 0; i < vol->v_disks_count; i++) { 213489b17223SAlexander Motin sd = &vol->v_subdisks[i]; 213589b17223SAlexander Motin if (sd->sd_disk != NULL && 213689b17223SAlexander Motin sd->sd_disk->d_consumer != NULL) { 213789b17223SAlexander Motin sbuf_printf(sb, "%s ", 213889b17223SAlexander Motin g_raid_get_diskname(sd->sd_disk)); 213989b17223SAlexander Motin } else { 214089b17223SAlexander Motin sbuf_printf(sb, "NONE "); 214189b17223SAlexander Motin } 214289b17223SAlexander Motin sbuf_printf(sb, "(%s", 214389b17223SAlexander Motin g_raid_subdisk_state2str(sd->sd_state)); 214489b17223SAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 214589b17223SAlexander Motin sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 214689b17223SAlexander Motin sbuf_printf(sb, " %d%%", 214789b17223SAlexander Motin (int)(sd->sd_rebuild_pos * 100 / 214889b17223SAlexander Motin sd->sd_size)); 214989b17223SAlexander Motin } 215089b17223SAlexander Motin sbuf_printf(sb, ")"); 215189b17223SAlexander Motin if (i + 1 < vol->v_disks_count) 215289b17223SAlexander Motin sbuf_printf(sb, ", "); 215389b17223SAlexander Motin } 215489b17223SAlexander Motin sbuf_printf(sb, "</Subdisks>\n"); 215589b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 215689b17223SAlexander Motin g_topology_lock(); 215789b17223SAlexander Motin } else if (cp != NULL) { 215889b17223SAlexander Motin disk = cp->private; 215989b17223SAlexander Motin if (disk == NULL) 216089b17223SAlexander Motin return; 216189b17223SAlexander Motin g_topology_unlock(); 216289b17223SAlexander Motin sx_xlock(&sc->sc_lock); 216389b17223SAlexander Motin sbuf_printf(sb, "%s<State>%s", indent, 216489b17223SAlexander Motin g_raid_disk_state2str(disk->d_state)); 216589b17223SAlexander Motin if (!TAILQ_EMPTY(&disk->d_subdisks)) { 216689b17223SAlexander Motin sbuf_printf(sb, " ("); 216789b17223SAlexander Motin TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 216889b17223SAlexander Motin sbuf_printf(sb, "%s", 216989b17223SAlexander Motin g_raid_subdisk_state2str(sd->sd_state)); 217089b17223SAlexander Motin if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || 217189b17223SAlexander Motin sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { 217289b17223SAlexander Motin sbuf_printf(sb, " %d%%", 217389b17223SAlexander Motin (int)(sd->sd_rebuild_pos * 100 / 217489b17223SAlexander Motin sd->sd_size)); 217589b17223SAlexander Motin } 217689b17223SAlexander Motin if (TAILQ_NEXT(sd, sd_next)) 217789b17223SAlexander Motin sbuf_printf(sb, ", "); 217889b17223SAlexander Motin } 217989b17223SAlexander Motin sbuf_printf(sb, ")"); 218089b17223SAlexander Motin } 218189b17223SAlexander Motin sbuf_printf(sb, "</State>\n"); 218289b17223SAlexander Motin sbuf_printf(sb, "%s<Subdisks>", indent); 218389b17223SAlexander Motin TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { 218489b17223SAlexander Motin sbuf_printf(sb, "r%d(%s):%d@%ju", 218589b17223SAlexander Motin sd->sd_volume->v_global_id, 218689b17223SAlexander Motin sd->sd_volume->v_name, 218789b17223SAlexander Motin sd->sd_pos, sd->sd_offset); 218889b17223SAlexander Motin if (TAILQ_NEXT(sd, sd_next)) 218989b17223SAlexander Motin sbuf_printf(sb, ", "); 219089b17223SAlexander Motin } 219189b17223SAlexander Motin sbuf_printf(sb, "</Subdisks>\n"); 219289b17223SAlexander Motin sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent, 219389b17223SAlexander Motin disk->d_read_errs); 219489b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 219589b17223SAlexander Motin g_topology_lock(); 219689b17223SAlexander Motin } else { 219789b17223SAlexander Motin g_topology_unlock(); 219889b17223SAlexander Motin sx_xlock(&sc->sc_lock); 219989b17223SAlexander Motin if (sc->sc_md) { 220089b17223SAlexander Motin sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent, 220189b17223SAlexander Motin sc->sc_md->mdo_class->name); 220289b17223SAlexander Motin } 220389b17223SAlexander Motin if (!TAILQ_EMPTY(&sc->sc_volumes)) { 220489b17223SAlexander Motin s = 0xff; 220589b17223SAlexander Motin TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { 220689b17223SAlexander Motin if (vol->v_state < s) 220789b17223SAlexander Motin s = vol->v_state; 220889b17223SAlexander Motin } 220989b17223SAlexander Motin sbuf_printf(sb, "%s<State>%s</State>\n", indent, 221089b17223SAlexander Motin g_raid_volume_state2str(s)); 221189b17223SAlexander Motin } 221289b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 221389b17223SAlexander Motin g_topology_lock(); 221489b17223SAlexander Motin } 221589b17223SAlexander Motin } 221689b17223SAlexander Motin 221789b17223SAlexander Motin static void 221889b17223SAlexander Motin g_raid_shutdown_pre_sync(void *arg, int howto) 221989b17223SAlexander Motin { 222089b17223SAlexander Motin struct g_class *mp; 222189b17223SAlexander Motin struct g_geom *gp, *gp2; 222289b17223SAlexander Motin struct g_raid_softc *sc; 222389b17223SAlexander Motin int error; 222489b17223SAlexander Motin 222589b17223SAlexander Motin mp = arg; 222689b17223SAlexander Motin DROP_GIANT(); 222789b17223SAlexander Motin g_topology_lock(); 222889b17223SAlexander Motin LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { 222989b17223SAlexander Motin if ((sc = gp->softc) == NULL) 223089b17223SAlexander Motin continue; 223189b17223SAlexander Motin g_topology_unlock(); 223289b17223SAlexander Motin sx_xlock(&sc->sc_lock); 223389b17223SAlexander Motin g_cancel_event(sc); 223489b17223SAlexander Motin error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); 223589b17223SAlexander Motin if (error != 0) 223689b17223SAlexander Motin sx_xunlock(&sc->sc_lock); 223789b17223SAlexander Motin g_topology_lock(); 223889b17223SAlexander Motin } 223989b17223SAlexander Motin g_topology_unlock(); 224089b17223SAlexander Motin PICKUP_GIANT(); 224189b17223SAlexander Motin } 224289b17223SAlexander Motin 224389b17223SAlexander Motin static void 224489b17223SAlexander Motin g_raid_init(struct g_class *mp) 224589b17223SAlexander Motin { 224689b17223SAlexander Motin 224789b17223SAlexander Motin g_raid_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync, 224889b17223SAlexander Motin g_raid_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST); 224989b17223SAlexander Motin if (g_raid_pre_sync == NULL) 225089b17223SAlexander Motin G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); 225189b17223SAlexander Motin g_raid_started = 1; 225289b17223SAlexander Motin } 225389b17223SAlexander Motin 225489b17223SAlexander Motin static void 225589b17223SAlexander Motin g_raid_fini(struct g_class *mp) 225689b17223SAlexander Motin { 225789b17223SAlexander Motin 225889b17223SAlexander Motin if (g_raid_pre_sync != NULL) 225989b17223SAlexander Motin EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_raid_pre_sync); 226089b17223SAlexander Motin g_raid_started = 0; 226189b17223SAlexander Motin } 226289b17223SAlexander Motin 226389b17223SAlexander Motin int 226489b17223SAlexander Motin g_raid_md_modevent(module_t mod, int type, void *arg) 226589b17223SAlexander Motin { 226689b17223SAlexander Motin struct g_raid_md_class *class, *c, *nc; 226789b17223SAlexander Motin int error; 226889b17223SAlexander Motin 226989b17223SAlexander Motin error = 0; 227089b17223SAlexander Motin class = arg; 227189b17223SAlexander Motin switch (type) { 227289b17223SAlexander Motin case MOD_LOAD: 227389b17223SAlexander Motin c = LIST_FIRST(&g_raid_md_classes); 227489b17223SAlexander Motin if (c == NULL || c->mdc_priority > class->mdc_priority) 227589b17223SAlexander Motin LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); 227689b17223SAlexander Motin else { 227789b17223SAlexander Motin while ((nc = LIST_NEXT(c, mdc_list)) != NULL && 227889b17223SAlexander Motin nc->mdc_priority < class->mdc_priority) 227989b17223SAlexander Motin c = nc; 228089b17223SAlexander Motin LIST_INSERT_AFTER(c, class, mdc_list); 228189b17223SAlexander Motin } 228289b17223SAlexander Motin if (g_raid_started) 228389b17223SAlexander Motin g_retaste(&g_raid_class); 228489b17223SAlexander Motin break; 228589b17223SAlexander Motin case MOD_UNLOAD: 228689b17223SAlexander Motin LIST_REMOVE(class, mdc_list); 228789b17223SAlexander Motin break; 228889b17223SAlexander Motin default: 228989b17223SAlexander Motin error = EOPNOTSUPP; 229089b17223SAlexander Motin break; 229189b17223SAlexander Motin } 229289b17223SAlexander Motin 229389b17223SAlexander Motin return (error); 229489b17223SAlexander Motin } 229589b17223SAlexander Motin 229689b17223SAlexander Motin int 229789b17223SAlexander Motin g_raid_tr_modevent(module_t mod, int type, void *arg) 229889b17223SAlexander Motin { 229989b17223SAlexander Motin struct g_raid_tr_class *class, *c, *nc; 230089b17223SAlexander Motin int error; 230189b17223SAlexander Motin 230289b17223SAlexander Motin error = 0; 230389b17223SAlexander Motin class = arg; 230489b17223SAlexander Motin switch (type) { 230589b17223SAlexander Motin case MOD_LOAD: 230689b17223SAlexander Motin c = LIST_FIRST(&g_raid_tr_classes); 230789b17223SAlexander Motin if (c == NULL || c->trc_priority > class->trc_priority) 230889b17223SAlexander Motin LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); 230989b17223SAlexander Motin else { 231089b17223SAlexander Motin while ((nc = LIST_NEXT(c, trc_list)) != NULL && 231189b17223SAlexander Motin nc->trc_priority < class->trc_priority) 231289b17223SAlexander Motin c = nc; 231389b17223SAlexander Motin LIST_INSERT_AFTER(c, class, trc_list); 231489b17223SAlexander Motin } 231589b17223SAlexander Motin break; 231689b17223SAlexander Motin case MOD_UNLOAD: 231789b17223SAlexander Motin LIST_REMOVE(class, trc_list); 231889b17223SAlexander Motin break; 231989b17223SAlexander Motin default: 232089b17223SAlexander Motin error = EOPNOTSUPP; 232189b17223SAlexander Motin break; 232289b17223SAlexander Motin } 232389b17223SAlexander Motin 232489b17223SAlexander Motin return (error); 232589b17223SAlexander Motin } 232689b17223SAlexander Motin 232789b17223SAlexander Motin /* 232889b17223SAlexander Motin * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) 232989b17223SAlexander Motin * to reduce module priority, allowing submodules to register them first. 233089b17223SAlexander Motin */ 233189b17223SAlexander Motin static moduledata_t g_raid_mod = { 233289b17223SAlexander Motin "g_raid", 233389b17223SAlexander Motin g_modevent, 233489b17223SAlexander Motin &g_raid_class 233589b17223SAlexander Motin }; 233689b17223SAlexander Motin DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD); 233789b17223SAlexander Motin MODULE_VERSION(geom_raid, 0); 2338