diff --git a/etc/mtree/BSD.include.dist b/etc/mtree/BSD.include.dist index b227bdb30ca7..a19eddc1d4db 100644 --- a/etc/mtree/BSD.include.dist +++ b/etc/mtree/BSD.include.dist @@ -190,6 +190,8 @@ .. nop .. + raid + .. raid3 .. shsec diff --git a/include/Makefile b/include/Makefile index 249db95569bf..9bcced529890 100644 --- a/include/Makefile +++ b/include/Makefile @@ -47,7 +47,7 @@ LSUBDIRS= cam/ata cam/scsi \ ${_fs_nwfs} fs/portalfs fs/procfs fs/smbfs fs/udf fs/unionfs \ geom/cache geom/concat geom/eli geom/gate geom/journal geom/label \ geom/mirror geom/mountver geom/multipath geom/nop \ - geom/raid3 geom/shsec geom/stripe geom/virstor \ + geom/raid geom/raid3 geom/shsec geom/stripe geom/virstor \ netgraph/atm netgraph/netflow \ security/audit \ security/mac_biba security/mac_bsdextended security/mac_lomac \ diff --git a/sbin/geom/class/Makefile b/sbin/geom/class/Makefile index 0611cdd7f29e..912561fbc52f 100644 --- a/sbin/geom/class/Makefile +++ b/sbin/geom/class/Makefile @@ -14,6 +14,7 @@ SUBDIR+=mountver SUBDIR+=multipath SUBDIR+=nop SUBDIR+=part +SUBDIR+=raid SUBDIR+=raid3 SUBDIR+=sched SUBDIR+=shsec diff --git a/sbin/geom/class/raid/Makefile b/sbin/geom/class/raid/Makefile new file mode 100644 index 000000000000..743f690f0fd9 --- /dev/null +++ b/sbin/geom/class/raid/Makefile @@ -0,0 +1,10 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../misc + +GEOM_CLASS= raid + +DPADD= ${LIBMD} +LDADD= -lmd + +.include diff --git a/sbin/geom/class/raid/geom_raid.c b/sbin/geom/class/raid/geom_raid.c new file mode 100644 index 000000000000..2f1629532e70 --- /dev/null +++ b/sbin/geom/class/raid/geom_raid.c @@ -0,0 +1,91 @@ +/*- + * Copyright (c) 2010 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +uint32_t lib_version = G_LIB_VERSION; +uint32_t version = G_RAID_VERSION; + +struct g_command class_commands[] = { + { "label", G_FLAG_VERBOSE, NULL, + { + { 'f', "force", NULL, G_TYPE_BOOL }, + { 'S', "size", G_VAL_OPTIONAL, G_TYPE_NUMBER }, + { 's', "strip", G_VAL_OPTIONAL, G_TYPE_NUMBER }, + G_OPT_SENTINEL + }, + "[-fv] [-S size] [-s stripsize] format label level prov ..." + }, + { "add", G_FLAG_VERBOSE, NULL, + { + { 'f', "force", NULL, G_TYPE_BOOL }, + { 'S', "size", G_VAL_OPTIONAL, G_TYPE_NUMBER }, + { 's', "strip", G_VAL_OPTIONAL, G_TYPE_NUMBER }, + G_OPT_SENTINEL + }, + "[-fv] [-S size] [-s stripsize] name label level" + }, + { "delete", G_FLAG_VERBOSE, NULL, + { + { 'f', "force", NULL, G_TYPE_BOOL }, + G_OPT_SENTINEL + }, + "[-fv] name [label|num]" + }, + { "insert", G_FLAG_VERBOSE, NULL, G_NULL_OPTS, + "[-v] name prov ..." + }, + { "remove", G_FLAG_VERBOSE, NULL, G_NULL_OPTS, + "[-v] name prov ..." + }, + { "fail", G_FLAG_VERBOSE, NULL, G_NULL_OPTS, + "[-v] name prov ..." + }, + { "stop", G_FLAG_VERBOSE, NULL, + { + { 'f', "force", NULL, G_TYPE_BOOL }, + G_OPT_SENTINEL + }, + "[-fv] name" + }, + G_CMD_SENTINEL +}; + diff --git a/sbin/geom/class/raid/graid.8 b/sbin/geom/class/raid/graid.8 new file mode 100644 index 000000000000..d1c92a2148af --- /dev/null +++ b/sbin/geom/class/raid/graid.8 @@ -0,0 +1,266 @@ +.\" Copyright (c) 2010 Alexander Motin +.\" All rights reserved. +.\" +.\" Redistribution and use in source and binary forms, with or without +.\" modification, are permitted provided that the following conditions +.\" are met: +.\" 1. Redistributions of source code must retain the above copyright +.\" notice, this list of conditions and the following disclaimer. +.\" 2. Redistributions in binary form must reproduce the above copyright +.\" notice, this list of conditions and the following disclaimer in the +.\" documentation and/or other materials provided with the distribution. +.\" +.\" THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND +.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +.\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE +.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS +.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) +.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT +.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY +.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF +.\" SUCH DAMAGE. +.\" +.\" $FreeBSD$ +.\" +.Dd March 22, 2011 +.Dt GRAID 8 +.Os +.Sh NAME +.Nm graid +.Nd "control utility for software RAID devices" +.Sh SYNOPSIS +.Nm +.Cm label +.Op Fl f +.Op Fl S Ar size +.Op Fl s Ar strip +.Ar format +.Ar label +.Ar level +.Ar prov ... +.Nm +.Cm add +.Op Fl f +.Op Fl S Ar size +.Op Fl s Ar strip +.Ar name +.Ar label +.Ar level +.Nm +.Cm delete +.Op Fl f +.Ar name +.Op Ar label | Ar num +.Nm +.Cm insert +.Ar name +.Ar prov ... +.Nm +.Cm remove +.Ar name +.Ar prov ... +.Nm +.Cm fail +.Ar name +.Ar prov ... +.Nm +.Cm stop +.Op Fl fv +.Ar name ... +.Nm +.Cm list +.Nm +.Cm status +.Nm +.Cm load +.Nm +.Cm unload +.Sh DESCRIPTION +The +.Nm +utility is used to manage software RAID configurations, supported by the +GEOM RAID class. +GEOM RAID class uses on-disk metadata to provide access to software-RAID +volumes defined by different RAID BIOSes. +Depending on RAID BIOS type and it's metadata format, different subsets of +configurations and features are supported. +To allow booting from RAID volume, the metadata format should match the +RAID BIOS type and its capabilities. +To guarantee that these match, it is recommended to create volumes via the +RAID BIOS interface, while experienced users are free to do it using this +utility. +.Pp +The first argument to +.Nm +indicates an action to be performed: +.Bl -tag -width ".Cm destroy" +.It Cm label +Create an array with single volume. +The +.Ar format +argument specifies the on-disk metadata format to use for this array, +such as "Intel". +The +.Ar label +argument specifies the label of the created volume. +The +.Ar level +argument specifies the RAID level of the created volume, such as: +"RAID0", "RAID1", etc. +The subsequent list enumerates providers to use as array components. +The special name "NONE" can be used to reserve space for absent disks. +The order of components can be important, depending on specific RAID level +and metadata format. +.Pp +Additional options include: +.Bl -tag -width ".Fl s Ar strip" +.It Fl f +Enforce specified configuration creation if it is officially unsupported, +but technically can be created. +.It Fl S Ar size +Use +.Ar size +bytes on each component for this volume. +Should be used if several volumes per array are planned, or if smaller +components going to be inserted later. +Defaults to size of the smallest component. +.It Fl s Ar strip +Specifies strip size in bytes. +Defaults to 131072. +.El +.It Cm add +Create another volume on the existing array. +The +.Ar name +argument is the name of the existing array, reported by label command. +The rest of arguments are the same as for the label command. +.It Cm delete +Delete volume(s) from the existing array. +When the last volume is deleted, the array is also deleted and its metadata +erased. +The +.Ar name +argument is the name of existing array. +Optional +.Ar label +or +.Ar num +arguments allow specifying volume for deletion. +.Pp +Additional options include: +.Bl -tag -width ".Fl f" +.It Fl f +Delete volume(s) even if it is still open. +.El +.It Cm insert +Insert specified provider(s) into specified array instead of the first missing +or failed components. +If there are no such components, mark disk(s) as spare. +.It Cm remove +Remove the specified provider(s) from the specified array and erase metadata. +If there are spare disks present, the removed disk(s) will be replaced by +spares. +.It Cm fail +Mark the given disks(s) as failed, removing from active use unless absolutely +necessary due to exhausted redundancy. +If there are spare disks present - failed disk(s) will be replaced with one +of them. +.It Cm stop +Stop the given array. +The metadata will not be erased. +.Pp +Additional options include: +.Bl -tag -width ".Fl f" +.It Fl f +Stop the given array even if some of its volumes are opened. +.El +.It Cm list +See +.Xr geom 8 . +.It Cm status +See +.Xr geom 8 . +.It Cm load +See +.Xr geom 8 . +.It Cm unload +See +.Xr geom 8 . +.El +.Pp +Additional options include: +.Bl -tag -width ".Fl v" +.It Fl v +Be more verbose. +.El +.Sh SUPPORTED METADATA FORMATS +The GEOM RAID class follows a modular design, allowing different metadata +formats to be used. +Support is currently implemented for the following formats: +.Bl -tag -width "Intel" +.It Intel +The format used by Intel RAID BIOS. +Supports up to two volumes per array. +Supports configurations: RAID0 (2+ disks), RAID1 (2 disks), +RAID5 (3+ disks), RAID10 (4 disks). +Configurations not supported by Intel RAID BIOS, but enforceable on your own +risk: RAID1 (3+ disks), RAID1E (3+ disks), RAID10 (6+ disks). +.It JMicron +The format used by JMicron RAID BIOS. +Supports one volume per array. +Supports configurations: RAID0 (2+ disks), RAID1 (2 disks), +RAID10 (4 disks), CONCAT (2+ disks). +Configurations not supported by JMicron RAID BIOS, but enforceable on your own +risk: RAID1 (3+ disks), RAID1E (3+ disks), RAID10 (6+ disks), RAID5 (3+ disks). +.It NVIDIA +The format used by NVIDIA MediaShield RAID BIOS. +Supports one volume per array. +Supports configurations: RAID0 (2+ disks), RAID1 (2 disks), +RAID5 (3+ disks), RAID10 (4+ disks), SINGLE (1 disk), CONCAT (2+ disks). +Configurations not supported by NVIDIA MediaShield RAID BIOS, but enforceable +on your own risk: RAID1 (3+ disks). +.It Promise +The format used by Promise and AMD/ATI RAID BIOSes and FreeBSD ataraid(4) +driver. +Supports multiple volumes per array. +Each disk can be split to be used by up to two arbitrary volumes. +Supports configurations: RAID0 (2+ disks), RAID1 (2 disks), +RAID5 (3+ disks), RAID10 (4 disks), SINGLE (1 disk), CONCAT (2+ disks). +Configurations not supported by RAID BIOSes, but enforceable on your +own risk: RAID1 (3+ disks), RAID10 (6+ disks). +.It SiI +The format used by SiliconImage RAID BIOS. +Supports one volume per array. +Supports configurations: RAID0 (2+ disks), RAID1 (2 disks), +RAID5 (3+ disks), RAID10 (4 disks), SINGLE (1 disk), CONCAT (2+ disks). +Configurations not supported by SiliconImage RAID BIOS, but enforceable on your +own risk: RAID1 (3+ disks), RAID10 (6+ disks). +.El +.Sh SUPPORTED RAID LEVELS +The GEOM RAID class follows a modular design, allowing different RAID levels +to be used. +Support for the following RAID levels is currently implemented: RAID0, RAID1, +RAID1E, RAID10, SINGLE, CONCAT. +.Sh RAID LEVEL MIGRATION +The GEOM RAID class has no support for RAID level migration, allowed by some +metadata formats. +If you started migration using BIOS or in some other way, make sure to +complete it there. +Do not run GEOM RAID class on migrating volumes under pain of possible data +corruption! +.Sh EXIT STATUS +Exit status is 0 on success, and non-zero if the command fails. +.Sh SEE ALSO +.Xr geom 4 , +.Xr geom 8 , +.Xr vinum 8 +.Sh HISTORY +The +.Nm +utility appeared in +.Fx 9.0 . +.Sh AUTHORS +.An Alexander Motin Aq mav@FreeBSD.org +.An M. Warner Losh Aq imp@FreeBSD.org diff --git a/sys/conf/NOTES b/sys/conf/NOTES index cf8064f4b213..851b9b8f038c 100644 --- a/sys/conf/NOTES +++ b/sys/conf/NOTES @@ -163,6 +163,7 @@ options GEOM_PART_MBR # MBR partitioning options GEOM_PART_PC98 # PC-9800 disk partitioning options GEOM_PART_VTOC8 # SMI VTOC8 disk label options GEOM_PC98 # NEC PC9800 partitioning +options GEOM_RAID # Soft RAID functionality. options GEOM_RAID3 # RAID3 functionality. options GEOM_SHSEC # Shared secret. options GEOM_STRIPE # Disk striping. diff --git a/sys/conf/files b/sys/conf/files index 8af90a4c790b..bced83812b9d 100644 --- a/sys/conf/files +++ b/sys/conf/files @@ -2115,6 +2115,19 @@ geom/part/g_part_gpt.c optional geom_part_gpt geom/part/g_part_mbr.c optional geom_part_mbr geom/part/g_part_pc98.c optional geom_part_pc98 geom/part/g_part_vtoc8.c optional geom_part_vtoc8 +geom/raid/g_raid.c optional geom_raid +geom/raid/g_raid_ctl.c optional geom_raid +geom/raid/g_raid_md_if.m optional geom_raid +geom/raid/g_raid_tr_if.m optional geom_raid +geom/raid/md_intel.c optional geom_raid +geom/raid/md_jmicron.c optional geom_raid +geom/raid/md_nvidia.c optional geom_raid +geom/raid/md_promise.c optional geom_raid +geom/raid/md_sii.c optional geom_raid +geom/raid/tr_concat.c optional geom_raid +geom/raid/tr_raid0.c optional geom_raid +geom/raid/tr_raid1.c optional geom_raid +geom/raid/tr_raid1e.c optional geom_raid geom/raid3/g_raid3.c optional geom_raid3 geom/raid3/g_raid3_ctl.c optional geom_raid3 geom/shsec/g_shsec.c optional geom_shsec diff --git a/sys/conf/options b/sys/conf/options index b3642e9f1f1d..a507d6907983 100644 --- a/sys/conf/options +++ b/sys/conf/options @@ -102,6 +102,7 @@ GEOM_PART_MBR opt_geom.h GEOM_PART_PC98 opt_geom.h GEOM_PART_VTOC8 opt_geom.h GEOM_PC98 opt_geom.h +GEOM_RAID opt_geom.h GEOM_RAID3 opt_geom.h GEOM_SHSEC opt_geom.h GEOM_STRIPE opt_geom.h diff --git a/sys/geom/raid/g_raid.c b/sys/geom/raid/g_raid.c new file mode 100644 index 000000000000..eebb3605dba3 --- /dev/null +++ b/sys/geom/raid/g_raid.c @@ -0,0 +1,2340 @@ +/*- + * Copyright (c) 2010 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "g_raid_md_if.h" +#include "g_raid_tr_if.h" + +static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data"); + +SYSCTL_DECL(_kern_geom); +SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff"); +u_int g_raid_aggressive_spare = 0; +TUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare); +SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW, + &g_raid_aggressive_spare, 0, "Use disks without metadata as spare"); +u_int g_raid_debug = 2; +TUNABLE_INT("kern.geom.raid.debug", &g_raid_debug); +SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0, + "Debug level"); +int g_raid_read_err_thresh = 10; +TUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh); +SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW, + &g_raid_read_err_thresh, 0, + "Number of read errors equated to disk failure"); +u_int g_raid_start_timeout = 30; +TUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout); +SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW, + &g_raid_start_timeout, 0, + "Time to wait for all array components"); +static u_int g_raid_clean_time = 5; +TUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time); +SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW, + &g_raid_clean_time, 0, "Mark volume as clean when idling"); +static u_int g_raid_disconnect_on_failure = 1; +TUNABLE_INT("kern.geom.raid.disconnect_on_failure", + &g_raid_disconnect_on_failure); +SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW, + &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure."); +static u_int g_raid_name_format = 0; +TUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format); +SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW, + &g_raid_name_format, 0, "Providers name format."); +static u_int g_raid_idle_threshold = 1000000; +TUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold); +SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW, + &g_raid_idle_threshold, 1000000, + "Time in microseconds to consider a volume idle."); + +#define MSLEEP(rv, ident, mtx, priority, wmesg, timeout) do { \ + G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \ + rv = msleep((ident), (mtx), (priority), (wmesg), (timeout)); \ + G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \ +} while (0) + +LIST_HEAD(, g_raid_md_class) g_raid_md_classes = + LIST_HEAD_INITIALIZER(g_raid_md_classes); + +LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes = + LIST_HEAD_INITIALIZER(g_raid_tr_classes); + +LIST_HEAD(, g_raid_volume) g_raid_volumes = + LIST_HEAD_INITIALIZER(g_raid_volumes); + +static eventhandler_tag g_raid_pre_sync = NULL; +static int g_raid_started = 0; + +static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp, + struct g_geom *gp); +static g_taste_t g_raid_taste; +static void g_raid_init(struct g_class *mp); +static void g_raid_fini(struct g_class *mp); + +struct g_class g_raid_class = { + .name = G_RAID_CLASS_NAME, + .version = G_VERSION, + .ctlreq = g_raid_ctl, + .taste = g_raid_taste, + .destroy_geom = g_raid_destroy_geom, + .init = g_raid_init, + .fini = g_raid_fini +}; + +static void g_raid_destroy_provider(struct g_raid_volume *vol); +static int g_raid_update_disk(struct g_raid_disk *disk, u_int event); +static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event); +static int g_raid_update_volume(struct g_raid_volume *vol, u_int event); +static int g_raid_update_node(struct g_raid_softc *sc, u_int event); +static void g_raid_dumpconf(struct sbuf *sb, const char *indent, + struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp); +static void g_raid_start(struct bio *bp); +static void g_raid_start_request(struct bio *bp); +static void g_raid_disk_done(struct bio *bp); +static void g_raid_poll(struct g_raid_softc *sc); + +static const char * +g_raid_node_event2str(int event) +{ + + switch (event) { + case G_RAID_NODE_E_WAKE: + return ("WAKE"); + case G_RAID_NODE_E_START: + return ("START"); + default: + return ("INVALID"); + } +} + +const char * +g_raid_disk_state2str(int state) +{ + + switch (state) { + case G_RAID_DISK_S_NONE: + return ("NONE"); + case G_RAID_DISK_S_OFFLINE: + return ("OFFLINE"); + case G_RAID_DISK_S_FAILED: + return ("FAILED"); + case G_RAID_DISK_S_STALE_FAILED: + return ("STALE_FAILED"); + case G_RAID_DISK_S_SPARE: + return ("SPARE"); + case G_RAID_DISK_S_STALE: + return ("STALE"); + case G_RAID_DISK_S_ACTIVE: + return ("ACTIVE"); + default: + return ("INVALID"); + } +} + +static const char * +g_raid_disk_event2str(int event) +{ + + switch (event) { + case G_RAID_DISK_E_DISCONNECTED: + return ("DISCONNECTED"); + default: + return ("INVALID"); + } +} + +const char * +g_raid_subdisk_state2str(int state) +{ + + switch (state) { + case G_RAID_SUBDISK_S_NONE: + return ("NONE"); + case G_RAID_SUBDISK_S_FAILED: + return ("FAILED"); + case G_RAID_SUBDISK_S_NEW: + return ("NEW"); + case G_RAID_SUBDISK_S_REBUILD: + return ("REBUILD"); + case G_RAID_SUBDISK_S_UNINITIALIZED: + return ("UNINITIALIZED"); + case G_RAID_SUBDISK_S_STALE: + return ("STALE"); + case G_RAID_SUBDISK_S_RESYNC: + return ("RESYNC"); + case G_RAID_SUBDISK_S_ACTIVE: + return ("ACTIVE"); + default: + return ("INVALID"); + } +} + +static const char * +g_raid_subdisk_event2str(int event) +{ + + switch (event) { + case G_RAID_SUBDISK_E_NEW: + return ("NEW"); + case G_RAID_SUBDISK_E_DISCONNECTED: + return ("DISCONNECTED"); + default: + return ("INVALID"); + } +} + +const char * +g_raid_volume_state2str(int state) +{ + + switch (state) { + case G_RAID_VOLUME_S_STARTING: + return ("STARTING"); + case G_RAID_VOLUME_S_BROKEN: + return ("BROKEN"); + case G_RAID_VOLUME_S_DEGRADED: + return ("DEGRADED"); + case G_RAID_VOLUME_S_SUBOPTIMAL: + return ("SUBOPTIMAL"); + case G_RAID_VOLUME_S_OPTIMAL: + return ("OPTIMAL"); + case G_RAID_VOLUME_S_UNSUPPORTED: + return ("UNSUPPORTED"); + case G_RAID_VOLUME_S_STOPPED: + return ("STOPPED"); + default: + return ("INVALID"); + } +} + +static const char * +g_raid_volume_event2str(int event) +{ + + switch (event) { + case G_RAID_VOLUME_E_UP: + return ("UP"); + case G_RAID_VOLUME_E_DOWN: + return ("DOWN"); + case G_RAID_VOLUME_E_START: + return ("START"); + case G_RAID_VOLUME_E_STARTMD: + return ("STARTMD"); + default: + return ("INVALID"); + } +} + +const char * +g_raid_volume_level2str(int level, int qual) +{ + + switch (level) { + case G_RAID_VOLUME_RL_RAID0: + return ("RAID0"); + case G_RAID_VOLUME_RL_RAID1: + return ("RAID1"); + case G_RAID_VOLUME_RL_RAID3: + return ("RAID3"); + case G_RAID_VOLUME_RL_RAID4: + return ("RAID4"); + case G_RAID_VOLUME_RL_RAID5: + return ("RAID5"); + case G_RAID_VOLUME_RL_RAID6: + return ("RAID6"); + case G_RAID_VOLUME_RL_RAID1E: + return ("RAID1E"); + case G_RAID_VOLUME_RL_SINGLE: + return ("SINGLE"); + case G_RAID_VOLUME_RL_CONCAT: + return ("CONCAT"); + case G_RAID_VOLUME_RL_RAID5E: + return ("RAID5E"); + case G_RAID_VOLUME_RL_RAID5EE: + return ("RAID5EE"); + default: + return ("UNKNOWN"); + } +} + +int +g_raid_volume_str2level(const char *str, int *level, int *qual) +{ + + *level = G_RAID_VOLUME_RL_UNKNOWN; + *qual = G_RAID_VOLUME_RLQ_NONE; + if (strcasecmp(str, "RAID0") == 0) + *level = G_RAID_VOLUME_RL_RAID0; + else if (strcasecmp(str, "RAID1") == 0) + *level = G_RAID_VOLUME_RL_RAID1; + else if (strcasecmp(str, "RAID3") == 0) + *level = G_RAID_VOLUME_RL_RAID3; + else if (strcasecmp(str, "RAID4") == 0) + *level = G_RAID_VOLUME_RL_RAID4; + else if (strcasecmp(str, "RAID5") == 0) + *level = G_RAID_VOLUME_RL_RAID5; + else if (strcasecmp(str, "RAID6") == 0) + *level = G_RAID_VOLUME_RL_RAID6; + else if (strcasecmp(str, "RAID10") == 0 || + strcasecmp(str, "RAID1E") == 0) + *level = G_RAID_VOLUME_RL_RAID1E; + else if (strcasecmp(str, "SINGLE") == 0) + *level = G_RAID_VOLUME_RL_SINGLE; + else if (strcasecmp(str, "CONCAT") == 0) + *level = G_RAID_VOLUME_RL_CONCAT; + else if (strcasecmp(str, "RAID5E") == 0) + *level = G_RAID_VOLUME_RL_RAID5E; + else if (strcasecmp(str, "RAID5EE") == 0) + *level = G_RAID_VOLUME_RL_RAID5EE; + else + return (-1); + return (0); +} + +const char * +g_raid_get_diskname(struct g_raid_disk *disk) +{ + + if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL) + return ("[unknown]"); + return (disk->d_consumer->provider->name); +} + +void +g_raid_report_disk_state(struct g_raid_disk *disk) +{ + struct g_raid_subdisk *sd; + int len, state; + uint32_t s; + + if (disk->d_consumer == NULL) + return; + if (disk->d_state == G_RAID_DISK_S_FAILED || + disk->d_state == G_RAID_DISK_S_STALE_FAILED) { + s = G_STATE_FAILED; + } else { + state = G_RAID_SUBDISK_S_ACTIVE; + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + if (sd->sd_state < state) + state = sd->sd_state; + } + if (state == G_RAID_SUBDISK_S_FAILED) + s = G_STATE_FAILED; + else if (state == G_RAID_SUBDISK_S_NEW || + state == G_RAID_SUBDISK_S_REBUILD) + s = G_STATE_REBUILD; + else if (state == G_RAID_SUBDISK_S_STALE || + state == G_RAID_SUBDISK_S_RESYNC) + s = G_STATE_RESYNC; + else + s = G_STATE_ACTIVE; + } + len = sizeof(s); + g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s); + G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.", + g_raid_get_diskname(disk), s); +} + +void +g_raid_change_disk_state(struct g_raid_disk *disk, int state) +{ + + G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.", + g_raid_get_diskname(disk), + g_raid_disk_state2str(disk->d_state), + g_raid_disk_state2str(state)); + disk->d_state = state; + g_raid_report_disk_state(disk); +} + +void +g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state) +{ + + G_RAID_DEBUG1(0, sd->sd_softc, + "Subdisk %s:%d-%s state changed from %s to %s.", + sd->sd_volume->v_name, sd->sd_pos, + sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", + g_raid_subdisk_state2str(sd->sd_state), + g_raid_subdisk_state2str(state)); + sd->sd_state = state; + if (sd->sd_disk) + g_raid_report_disk_state(sd->sd_disk); +} + +void +g_raid_change_volume_state(struct g_raid_volume *vol, int state) +{ + + G_RAID_DEBUG1(0, vol->v_softc, + "Volume %s state changed from %s to %s.", + vol->v_name, + g_raid_volume_state2str(vol->v_state), + g_raid_volume_state2str(state)); + vol->v_state = state; +} + +/* + * --- Events handling functions --- + * Events in geom_raid are used to maintain subdisks and volumes status + * from one thread to simplify locking. + */ +static void +g_raid_event_free(struct g_raid_event *ep) +{ + + free(ep, M_RAID); +} + +int +g_raid_event_send(void *arg, int event, int flags) +{ + struct g_raid_softc *sc; + struct g_raid_event *ep; + int error; + + if ((flags & G_RAID_EVENT_VOLUME) != 0) { + sc = ((struct g_raid_volume *)arg)->v_softc; + } else if ((flags & G_RAID_EVENT_DISK) != 0) { + sc = ((struct g_raid_disk *)arg)->d_softc; + } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) { + sc = ((struct g_raid_subdisk *)arg)->sd_softc; + } else { + sc = arg; + } + ep = malloc(sizeof(*ep), M_RAID, + sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT); + if (ep == NULL) + return (ENOMEM); + ep->e_tgt = arg; + ep->e_event = event; + ep->e_flags = flags; + ep->e_error = 0; + G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc); + mtx_lock(&sc->sc_queue_mtx); + TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next); + mtx_unlock(&sc->sc_queue_mtx); + wakeup(sc); + + if ((flags & G_RAID_EVENT_WAIT) == 0) + return (0); + + sx_assert(&sc->sc_lock, SX_XLOCKED); + G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep); + sx_xunlock(&sc->sc_lock); + while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) { + mtx_lock(&sc->sc_queue_mtx); + MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event", + hz * 5); + } + error = ep->e_error; + g_raid_event_free(ep); + sx_xlock(&sc->sc_lock); + return (error); +} + +static void +g_raid_event_cancel(struct g_raid_softc *sc, void *tgt) +{ + struct g_raid_event *ep, *tmpep; + + sx_assert(&sc->sc_lock, SX_XLOCKED); + + mtx_lock(&sc->sc_queue_mtx); + TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) { + if (ep->e_tgt != tgt) + continue; + TAILQ_REMOVE(&sc->sc_events, ep, e_next); + if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) + g_raid_event_free(ep); + else { + ep->e_error = ECANCELED; + wakeup(ep); + } + } + mtx_unlock(&sc->sc_queue_mtx); +} + +static int +g_raid_event_check(struct g_raid_softc *sc, void *tgt) +{ + struct g_raid_event *ep; + int res = 0; + + sx_assert(&sc->sc_lock, SX_XLOCKED); + + mtx_lock(&sc->sc_queue_mtx); + TAILQ_FOREACH(ep, &sc->sc_events, e_next) { + if (ep->e_tgt != tgt) + continue; + res = 1; + break; + } + mtx_unlock(&sc->sc_queue_mtx); + return (res); +} + +/* + * Return the number of disks in given state. + * If state is equal to -1, count all connected disks. + */ +u_int +g_raid_ndisks(struct g_raid_softc *sc, int state) +{ + struct g_raid_disk *disk; + u_int n; + + sx_assert(&sc->sc_lock, SX_LOCKED); + + n = 0; + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == state || state == -1) + n++; + } + return (n); +} + +/* + * Return the number of subdisks in given state. + * If state is equal to -1, count all connected disks. + */ +u_int +g_raid_nsubdisks(struct g_raid_volume *vol, int state) +{ + struct g_raid_subdisk *subdisk; + struct g_raid_softc *sc; + u_int i, n ; + + sc = vol->v_softc; + sx_assert(&sc->sc_lock, SX_LOCKED); + + n = 0; + for (i = 0; i < vol->v_disks_count; i++) { + subdisk = &vol->v_subdisks[i]; + if ((state == -1 && + subdisk->sd_state != G_RAID_SUBDISK_S_NONE) || + subdisk->sd_state == state) + n++; + } + return (n); +} + +/* + * Return the first subdisk in given state. + * If state is equal to -1, then the first connected disks. + */ +struct g_raid_subdisk * +g_raid_get_subdisk(struct g_raid_volume *vol, int state) +{ + struct g_raid_subdisk *sd; + struct g_raid_softc *sc; + u_int i; + + sc = vol->v_softc; + sx_assert(&sc->sc_lock, SX_LOCKED); + + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if ((state == -1 && + sd->sd_state != G_RAID_SUBDISK_S_NONE) || + sd->sd_state == state) + return (sd); + } + return (NULL); +} + +struct g_consumer * +g_raid_open_consumer(struct g_raid_softc *sc, const char *name) +{ + struct g_consumer *cp; + struct g_provider *pp; + + g_topology_assert(); + + if (strncmp(name, "/dev/", 5) == 0) + name += 5; + pp = g_provider_by_name(name); + if (pp == NULL) + return (NULL); + cp = g_new_consumer(sc->sc_geom); + if (g_attach(cp, pp) != 0) { + g_destroy_consumer(cp); + return (NULL); + } + if (g_access(cp, 1, 1, 1) != 0) { + g_detach(cp); + g_destroy_consumer(cp); + return (NULL); + } + return (cp); +} + +static u_int +g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp) +{ + struct bio *bp; + u_int nreqs = 0; + + mtx_lock(&sc->sc_queue_mtx); + TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) { + if (bp->bio_from == cp) + nreqs++; + } + mtx_unlock(&sc->sc_queue_mtx); + return (nreqs); +} + +u_int +g_raid_nopens(struct g_raid_softc *sc) +{ + struct g_raid_volume *vol; + u_int opens; + + opens = 0; + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + if (vol->v_provider_open != 0) + opens++; + } + return (opens); +} + +static int +g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp) +{ + + if (cp->index > 0) { + G_RAID_DEBUG1(2, sc, + "I/O requests for %s exist, can't destroy it now.", + cp->provider->name); + return (1); + } + if (g_raid_nrequests(sc, cp) > 0) { + G_RAID_DEBUG1(2, sc, + "I/O requests for %s in queue, can't destroy it now.", + cp->provider->name); + return (1); + } + return (0); +} + +static void +g_raid_destroy_consumer(void *arg, int flags __unused) +{ + struct g_consumer *cp; + + g_topology_assert(); + + cp = arg; + G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name); + g_detach(cp); + g_destroy_consumer(cp); +} + +void +g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp) +{ + struct g_provider *pp; + int retaste_wait; + + g_topology_assert_not(); + + g_topology_lock(); + cp->private = NULL; + if (g_raid_consumer_is_busy(sc, cp)) + goto out; + pp = cp->provider; + retaste_wait = 0; + if (cp->acw == 1) { + if ((pp->geom->flags & G_GEOM_WITHER) == 0) + retaste_wait = 1; + } + if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) + g_access(cp, -cp->acr, -cp->acw, -cp->ace); + if (retaste_wait) { + /* + * After retaste event was send (inside g_access()), we can send + * event to detach and destroy consumer. + * A class, which has consumer to the given provider connected + * will not receive retaste event for the provider. + * This is the way how I ignore retaste events when I close + * consumers opened for write: I detach and destroy consumer + * after retaste event is sent. + */ + g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL); + goto out; + } + G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name); + g_detach(cp); + g_destroy_consumer(cp); +out: + g_topology_unlock(); +} + +static void +g_raid_orphan(struct g_consumer *cp) +{ + struct g_raid_disk *disk; + + g_topology_assert(); + + disk = cp->private; + if (disk == NULL) + return; + g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED, + G_RAID_EVENT_DISK); +} + +static int +g_raid_clean(struct g_raid_volume *vol, int acw) +{ + struct g_raid_softc *sc; + int timeout; + + sc = vol->v_softc; + g_topology_assert_not(); + sx_assert(&sc->sc_lock, SX_XLOCKED); + +// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) +// return (0); + if (!vol->v_dirty) + return (0); + if (vol->v_writes > 0) + return (0); + if (acw > 0 || (acw == -1 && + vol->v_provider != NULL && vol->v_provider->acw > 0)) { + timeout = g_raid_clean_time - (time_uptime - vol->v_last_write); + if (timeout > 0) + return (timeout); + } + vol->v_dirty = 0; + G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.", + vol->v_name); + g_raid_write_metadata(sc, vol, NULL, NULL); + return (0); +} + +static void +g_raid_dirty(struct g_raid_volume *vol) +{ + struct g_raid_softc *sc; + + sc = vol->v_softc; + g_topology_assert_not(); + sx_assert(&sc->sc_lock, SX_XLOCKED); + +// if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0) +// return; + vol->v_dirty = 1; + G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.", + vol->v_name); + g_raid_write_metadata(sc, vol, NULL, NULL); +} + +void +g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct bio_queue_head queue; + struct bio *cbp; + int i; + + vol = tr->tro_volume; + sc = vol->v_softc; + + /* + * Allocate all bios before sending any request, so we can return + * ENOMEM in nice and clean way. + */ + bioq_init(&queue); + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if (sd->sd_state == G_RAID_SUBDISK_S_NONE || + sd->sd_state == G_RAID_SUBDISK_S_FAILED) + continue; + cbp = g_clone_bio(bp); + if (cbp == NULL) + goto failure; + cbp->bio_caller1 = sd; + bioq_insert_tail(&queue, cbp); + } + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + sd = cbp->bio_caller1; + cbp->bio_caller1 = NULL; + g_raid_subdisk_iostart(sd, cbp); + } + return; +failure: + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + g_destroy_bio(cbp); + } + if (bp->bio_error == 0) + bp->bio_error = ENOMEM; + g_raid_iodone(bp, bp->bio_error); +} + +static void +g_raid_tr_kerneldump_common_done(struct bio *bp) +{ + + bp->bio_flags |= BIO_DONE; +} + +int +g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, + void *virtual, vm_offset_t physical, off_t offset, size_t length) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct bio bp; + + vol = tr->tro_volume; + sc = vol->v_softc; + + bzero(&bp, sizeof(bp)); + bp.bio_cmd = BIO_WRITE; + bp.bio_done = g_raid_tr_kerneldump_common_done; + bp.bio_attribute = NULL; + bp.bio_offset = offset; + bp.bio_length = length; + bp.bio_data = virtual; + bp.bio_to = vol->v_provider; + + g_raid_start(&bp); + while (!(bp.bio_flags & BIO_DONE)) { + G_RAID_DEBUG1(4, sc, "Poll..."); + g_raid_poll(sc); + DELAY(10); + } + + return (bp.bio_error != 0 ? EIO : 0); +} + +static int +g_raid_dump(void *arg, + void *virtual, vm_offset_t physical, off_t offset, size_t length) +{ + struct g_raid_volume *vol; + int error; + + vol = (struct g_raid_volume *)arg; + G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.", + (long long unsigned)offset, (long long unsigned)length); + + error = G_RAID_TR_KERNELDUMP(vol->v_tr, + virtual, physical, offset, length); + return (error); +} + +static void +g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp) +{ + struct g_kerneldump *gkd; + struct g_provider *pp; + struct g_raid_volume *vol; + + gkd = (struct g_kerneldump*)bp->bio_data; + pp = bp->bio_to; + vol = pp->private; + g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)", + pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length); + gkd->di.dumper = g_raid_dump; + gkd->di.priv = vol; + gkd->di.blocksize = vol->v_sectorsize; + gkd->di.maxiosize = DFLTPHYS; + gkd->di.mediaoffset = gkd->offset; + if ((gkd->offset + gkd->length) > vol->v_mediasize) + gkd->length = vol->v_mediasize - gkd->offset; + gkd->di.mediasize = gkd->length; + g_io_deliver(bp, 0); +} + +static void +g_raid_start(struct bio *bp) +{ + struct g_raid_softc *sc; + + sc = bp->bio_to->geom->softc; + /* + * If sc == NULL or there are no valid disks, provider's error + * should be set and g_raid_start() should not be called at all. + */ +// KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING, +// ("Provider's error should be set (error=%d)(mirror=%s).", +// bp->bio_to->error, bp->bio_to->name)); + G_RAID_LOGREQ(3, bp, "Request received."); + + switch (bp->bio_cmd) { + case BIO_READ: + case BIO_WRITE: + case BIO_DELETE: + case BIO_FLUSH: + break; + case BIO_GETATTR: + if (!strcmp(bp->bio_attribute, "GEOM::kerneldump")) + g_raid_kerneldump(sc, bp); + else + g_io_deliver(bp, EOPNOTSUPP); + return; + default: + g_io_deliver(bp, EOPNOTSUPP); + return; + } + mtx_lock(&sc->sc_queue_mtx); + bioq_disksort(&sc->sc_queue, bp); + mtx_unlock(&sc->sc_queue_mtx); + if (!dumping) { + G_RAID_DEBUG1(4, sc, "Waking up %p.", sc); + wakeup(sc); + } +} + +static int +g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len) +{ + /* + * 5 cases: + * (1) bp entirely below NO + * (2) bp entirely above NO + * (3) bp start below, but end in range YES + * (4) bp entirely within YES + * (5) bp starts within, ends above YES + * + * lock range 10-19 (offset 10 length 10) + * (1) 1-5: first if kicks it out + * (2) 30-35: second if kicks it out + * (3) 5-15: passes both ifs + * (4) 12-14: passes both ifs + * (5) 19-20: passes both + */ + off_t lend = lstart + len - 1; + off_t bstart = bp->bio_offset; + off_t bend = bp->bio_offset + bp->bio_length - 1; + + if (bend < lstart) + return (0); + if (lend < bstart) + return (0); + return (1); +} + +static int +g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp) +{ + struct g_raid_lock *lp; + + sx_assert(&vol->v_softc->sc_lock, SX_LOCKED); + + LIST_FOREACH(lp, &vol->v_locks, l_next) { + if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length)) + return (1); + } + return (0); +} + +static void +g_raid_start_request(struct bio *bp) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + + sc = bp->bio_to->geom->softc; + sx_assert(&sc->sc_lock, SX_LOCKED); + vol = bp->bio_to->private; + + /* + * Check to see if this item is in a locked range. If so, + * queue it to our locked queue and return. We'll requeue + * it when the range is unlocked. Internal I/O for the + * rebuild/rescan/recovery process is excluded from this + * check so we can actually do the recovery. + */ + if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) && + g_raid_is_in_locked_range(vol, bp)) { + G_RAID_LOGREQ(3, bp, "Defer request."); + bioq_insert_tail(&vol->v_locked, bp); + return; + } + + /* + * If we're actually going to do the write/delete, then + * update the idle stats for the volume. + */ + if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { + if (!vol->v_dirty) + g_raid_dirty(vol); + vol->v_writes++; + } + + /* + * Put request onto inflight queue, so we can check if new + * synchronization requests don't collide with it. Then tell + * the transformation layer to start the I/O. + */ + bioq_insert_tail(&vol->v_inflight, bp); + G_RAID_LOGREQ(4, bp, "Request started"); + G_RAID_TR_IOSTART(vol->v_tr, bp); +} + +static void +g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp) +{ + off_t off, len; + struct bio *nbp; + struct g_raid_lock *lp; + + vol->v_pending_lock = 0; + LIST_FOREACH(lp, &vol->v_locks, l_next) { + if (lp->l_pending) { + off = lp->l_offset; + len = lp->l_length; + lp->l_pending = 0; + TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) { + if (g_raid_bio_overlaps(nbp, off, len)) + lp->l_pending++; + } + if (lp->l_pending) { + vol->v_pending_lock = 1; + G_RAID_DEBUG1(4, vol->v_softc, + "Deferred lock(%jd, %jd) has %d pending", + (intmax_t)off, (intmax_t)(off + len), + lp->l_pending); + continue; + } + G_RAID_DEBUG1(4, vol->v_softc, + "Deferred lock of %jd to %jd completed", + (intmax_t)off, (intmax_t)(off + len)); + G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); + } + } +} + +void +g_raid_iodone(struct bio *bp, int error) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + + sc = bp->bio_to->geom->softc; + sx_assert(&sc->sc_lock, SX_LOCKED); + vol = bp->bio_to->private; + G_RAID_LOGREQ(3, bp, "Request done: %d.", error); + + /* Update stats if we done write/delete. */ + if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) { + vol->v_writes--; + vol->v_last_write = time_uptime; + } + + bioq_remove(&vol->v_inflight, bp); + if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp)) + g_raid_finish_with_locked_ranges(vol, bp); + getmicrouptime(&vol->v_last_done); + g_io_deliver(bp, error); +} + +int +g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, + struct bio *ignore, void *argp) +{ + struct g_raid_softc *sc; + struct g_raid_lock *lp; + struct bio *bp; + + sc = vol->v_softc; + lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO); + LIST_INSERT_HEAD(&vol->v_locks, lp, l_next); + lp->l_offset = off; + lp->l_length = len; + lp->l_callback_arg = argp; + + lp->l_pending = 0; + TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) { + if (bp != ignore && g_raid_bio_overlaps(bp, off, len)) + lp->l_pending++; + } + + /* + * If there are any writes that are pending, we return EBUSY. All + * callers will have to wait until all pending writes clear. + */ + if (lp->l_pending > 0) { + vol->v_pending_lock = 1; + G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend", + (intmax_t)off, (intmax_t)(off+len), lp->l_pending); + return (EBUSY); + } + G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd", + (intmax_t)off, (intmax_t)(off+len)); + G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg); + return (0); +} + +int +g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len) +{ + struct g_raid_lock *lp; + struct g_raid_softc *sc; + struct bio *bp; + + sc = vol->v_softc; + LIST_FOREACH(lp, &vol->v_locks, l_next) { + if (lp->l_offset == off && lp->l_length == len) { + LIST_REMOVE(lp, l_next); + /* XXX + * Right now we just put them all back on the queue + * and hope for the best. We hope this because any + * locked ranges will go right back on this list + * when the worker thread runs. + * XXX + */ + G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd", + (intmax_t)lp->l_offset, + (intmax_t)(lp->l_offset+lp->l_length)); + mtx_lock(&sc->sc_queue_mtx); + while ((bp = bioq_takefirst(&vol->v_locked)) != NULL) + bioq_disksort(&sc->sc_queue, bp); + mtx_unlock(&sc->sc_queue_mtx); + free(lp, M_RAID); + return (0); + } + } + return (EINVAL); +} + +void +g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp) +{ + struct g_consumer *cp; + struct g_raid_disk *disk, *tdisk; + + bp->bio_caller1 = sd; + + /* + * Make sure that the disk is present. Generally it is a task of + * transformation layers to not send requests to absent disks, but + * it is better to be safe and report situation then sorry. + */ + if (sd->sd_disk == NULL) { + G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!"); +nodisk: + bp->bio_from = NULL; + bp->bio_to = NULL; + bp->bio_error = ENXIO; + g_raid_disk_done(bp); + return; + } + disk = sd->sd_disk; + if (disk->d_state != G_RAID_DISK_S_ACTIVE && + disk->d_state != G_RAID_DISK_S_FAILED) { + G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a " + "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); + goto nodisk; + } + + cp = disk->d_consumer; + bp->bio_from = cp; + bp->bio_to = cp->provider; + cp->index++; + + /* Update average disks load. */ + TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) { + if (tdisk->d_consumer == NULL) + tdisk->d_load = 0; + else + tdisk->d_load = (tdisk->d_consumer->index * + G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8; + } + + disk->d_last_offset = bp->bio_offset + bp->bio_length; + if (dumping) { + G_RAID_LOGREQ(3, bp, "Sending dumping request."); + if (bp->bio_cmd == BIO_WRITE) { + bp->bio_error = g_raid_subdisk_kerneldump(sd, + bp->bio_data, 0, bp->bio_offset, bp->bio_length); + } else + bp->bio_error = EOPNOTSUPP; + g_raid_disk_done(bp); + } else { + bp->bio_done = g_raid_disk_done; + bp->bio_offset += sd->sd_offset; + G_RAID_LOGREQ(3, bp, "Sending request."); + g_io_request(bp, cp); + } +} + +int +g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, + void *virtual, vm_offset_t physical, off_t offset, size_t length) +{ + + if (sd->sd_disk == NULL) + return (ENXIO); + if (sd->sd_disk->d_kd.di.dumper == NULL) + return (EOPNOTSUPP); + return (dump_write(&sd->sd_disk->d_kd.di, + virtual, physical, + sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset, + length)); +} + +static void +g_raid_disk_done(struct bio *bp) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd; + + sd = bp->bio_caller1; + sc = sd->sd_softc; + mtx_lock(&sc->sc_queue_mtx); + bioq_disksort(&sc->sc_queue, bp); + mtx_unlock(&sc->sc_queue_mtx); + if (!dumping) + wakeup(sc); +} + +static void +g_raid_disk_done_request(struct bio *bp) +{ + struct g_raid_softc *sc; + struct g_raid_disk *disk; + struct g_raid_subdisk *sd; + struct g_raid_volume *vol; + + g_topology_assert_not(); + + G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error); + sd = bp->bio_caller1; + sc = sd->sd_softc; + vol = sd->sd_volume; + if (bp->bio_from != NULL) { + bp->bio_from->index--; + disk = bp->bio_from->private; + if (disk == NULL) + g_raid_kill_consumer(sc, bp->bio_from); + } + bp->bio_offset -= sd->sd_offset; + + G_RAID_TR_IODONE(vol->v_tr, sd, bp); +} + +static void +g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep) +{ + + if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0) + ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event); + else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0) + ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event); + else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0) + ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event); + else + ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event); + if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) { + KASSERT(ep->e_error == 0, + ("Error cannot be handled.")); + g_raid_event_free(ep); + } else { + ep->e_flags |= G_RAID_EVENT_DONE; + G_RAID_DEBUG1(4, sc, "Waking up %p.", ep); + mtx_lock(&sc->sc_queue_mtx); + wakeup(ep); + mtx_unlock(&sc->sc_queue_mtx); + } +} + +/* + * Worker thread. + */ +static void +g_raid_worker(void *arg) +{ + struct g_raid_softc *sc; + struct g_raid_event *ep; + struct g_raid_volume *vol; + struct bio *bp; + struct timeval now, t; + int timeout, rv; + + sc = arg; + thread_lock(curthread); + sched_prio(curthread, PRIBIO); + thread_unlock(curthread); + + sx_xlock(&sc->sc_lock); + for (;;) { + mtx_lock(&sc->sc_queue_mtx); + /* + * First take a look at events. + * This is important to handle events before any I/O requests. + */ + bp = NULL; + vol = NULL; + rv = 0; + ep = TAILQ_FIRST(&sc->sc_events); + if (ep != NULL) + TAILQ_REMOVE(&sc->sc_events, ep, e_next); + else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL) + ; + else { + getmicrouptime(&now); + t = now; + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + if (bioq_first(&vol->v_inflight) == NULL && + vol->v_tr && + timevalcmp(&vol->v_last_done, &t, < )) + t = vol->v_last_done; + } + timevalsub(&t, &now); + timeout = g_raid_idle_threshold + + t.tv_sec * 1000000 + t.tv_usec; + if (timeout > 0) { + /* + * Two steps to avoid overflows at HZ=1000 + * and idle timeouts > 2.1s. Some rounding + * errors can occur, but they are < 1tick, + * which is deemed to be close enough for + * this purpose. + */ + int micpertic = 1000000 / hz; + timeout = (timeout + micpertic - 1) / micpertic; + sx_xunlock(&sc->sc_lock); + MSLEEP(rv, sc, &sc->sc_queue_mtx, + PRIBIO | PDROP, "-", timeout); + sx_xlock(&sc->sc_lock); + goto process; + } else + rv = EWOULDBLOCK; + } + mtx_unlock(&sc->sc_queue_mtx); +process: + if (ep != NULL) { + g_raid_handle_event(sc, ep); + } else if (bp != NULL) { + if (bp->bio_to != NULL && + bp->bio_to->geom == sc->sc_geom) + g_raid_start_request(bp); + else + g_raid_disk_done_request(bp); + } else if (rv == EWOULDBLOCK) { + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + if (vol->v_writes == 0 && vol->v_dirty) + g_raid_clean(vol, -1); + if (bioq_first(&vol->v_inflight) == NULL && + vol->v_tr) { + t.tv_sec = g_raid_idle_threshold / 1000000; + t.tv_usec = g_raid_idle_threshold % 1000000; + timevaladd(&t, &vol->v_last_done); + getmicrouptime(&now); + if (timevalcmp(&t, &now, <= )) { + G_RAID_TR_IDLE(vol->v_tr); + vol->v_last_done = now; + } + } + } + } + if (sc->sc_stopping == G_RAID_DESTROY_HARD) + g_raid_destroy_node(sc, 1); /* May not return. */ + } +} + +static void +g_raid_poll(struct g_raid_softc *sc) +{ + struct g_raid_event *ep; + struct bio *bp; + + sx_xlock(&sc->sc_lock); + mtx_lock(&sc->sc_queue_mtx); + /* + * First take a look at events. + * This is important to handle events before any I/O requests. + */ + ep = TAILQ_FIRST(&sc->sc_events); + if (ep != NULL) { + TAILQ_REMOVE(&sc->sc_events, ep, e_next); + mtx_unlock(&sc->sc_queue_mtx); + g_raid_handle_event(sc, ep); + goto out; + } + bp = bioq_takefirst(&sc->sc_queue); + if (bp != NULL) { + mtx_unlock(&sc->sc_queue_mtx); + if (bp->bio_from == NULL || + bp->bio_from->geom != sc->sc_geom) + g_raid_start_request(bp); + else + g_raid_disk_done_request(bp); + } +out: + sx_xunlock(&sc->sc_lock); +} + +static void +g_raid_launch_provider(struct g_raid_volume *vol) +{ + struct g_raid_disk *disk; + struct g_raid_softc *sc; + struct g_provider *pp; + char name[G_RAID_MAX_VOLUMENAME]; + off_t off; + + sc = vol->v_softc; + sx_assert(&sc->sc_lock, SX_LOCKED); + + g_topology_lock(); + /* Try to name provider with volume name. */ + snprintf(name, sizeof(name), "raid/%s", vol->v_name); + if (g_raid_name_format == 0 || vol->v_name[0] == 0 || + g_provider_by_name(name) != NULL) { + /* Otherwise use sequential volume number. */ + snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id); + } + pp = g_new_providerf(sc->sc_geom, "%s", name); + pp->private = vol; + pp->mediasize = vol->v_mediasize; + pp->sectorsize = vol->v_sectorsize; + pp->stripesize = 0; + pp->stripeoffset = 0; + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || + vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || + vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE || + vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) { + if ((disk = vol->v_subdisks[0].sd_disk) != NULL && + disk->d_consumer != NULL && + disk->d_consumer->provider != NULL) { + pp->stripesize = disk->d_consumer->provider->stripesize; + off = disk->d_consumer->provider->stripeoffset; + pp->stripeoffset = off + vol->v_subdisks[0].sd_offset; + if (off > 0) + pp->stripeoffset %= off; + } + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) { + pp->stripesize *= (vol->v_disks_count - 1); + pp->stripeoffset *= (vol->v_disks_count - 1); + } + } else + pp->stripesize = vol->v_strip_size; + vol->v_provider = pp; + g_error_provider(pp, 0); + g_topology_unlock(); + G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.", + pp->name, vol->v_name); +} + +static void +g_raid_destroy_provider(struct g_raid_volume *vol) +{ + struct g_raid_softc *sc; + struct g_provider *pp; + struct bio *bp, *tmp; + + g_topology_assert_not(); + sc = vol->v_softc; + pp = vol->v_provider; + KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name)); + + g_topology_lock(); + g_error_provider(pp, ENXIO); + mtx_lock(&sc->sc_queue_mtx); + TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) { + if (bp->bio_to != pp) + continue; + bioq_remove(&sc->sc_queue, bp); + g_io_deliver(bp, ENXIO); + } + mtx_unlock(&sc->sc_queue_mtx); + G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.", + pp->name, vol->v_name); + g_wither_provider(pp, ENXIO); + g_topology_unlock(); + vol->v_provider = NULL; +} + +/* + * Update device state. + */ +static int +g_raid_update_volume(struct g_raid_volume *vol, u_int event) +{ + struct g_raid_softc *sc; + + sc = vol->v_softc; + sx_assert(&sc->sc_lock, SX_XLOCKED); + + G_RAID_DEBUG1(2, sc, "Event %s for volume %s.", + g_raid_volume_event2str(event), + vol->v_name); + switch (event) { + case G_RAID_VOLUME_E_DOWN: + if (vol->v_provider != NULL) + g_raid_destroy_provider(vol); + break; + case G_RAID_VOLUME_E_UP: + if (vol->v_provider == NULL) + g_raid_launch_provider(vol); + break; + case G_RAID_VOLUME_E_START: + if (vol->v_tr) + G_RAID_TR_START(vol->v_tr); + return (0); + default: + if (sc->sc_md) + G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event); + return (0); + } + + /* Manage root mount release. */ + if (vol->v_starting) { + vol->v_starting = 0; + G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount); + root_mount_rel(vol->v_rootmount); + vol->v_rootmount = NULL; + } + if (vol->v_stopping && vol->v_provider_open == 0) + g_raid_destroy_volume(vol); + return (0); +} + +/* + * Update subdisk state. + */ +static int +g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + + sc = sd->sd_softc; + vol = sd->sd_volume; + sx_assert(&sc->sc_lock, SX_XLOCKED); + + G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.", + g_raid_subdisk_event2str(event), + vol->v_name, sd->sd_pos, + sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); + if (vol->v_tr) + G_RAID_TR_EVENT(vol->v_tr, sd, event); + + return (0); +} + +/* + * Update disk state. + */ +static int +g_raid_update_disk(struct g_raid_disk *disk, u_int event) +{ + struct g_raid_softc *sc; + + sc = disk->d_softc; + sx_assert(&sc->sc_lock, SX_XLOCKED); + + G_RAID_DEBUG1(2, sc, "Event %s for disk %s.", + g_raid_disk_event2str(event), + g_raid_get_diskname(disk)); + + if (sc->sc_md) + G_RAID_MD_EVENT(sc->sc_md, disk, event); + return (0); +} + +/* + * Node event. + */ +static int +g_raid_update_node(struct g_raid_softc *sc, u_int event) +{ + sx_assert(&sc->sc_lock, SX_XLOCKED); + + G_RAID_DEBUG1(2, sc, "Event %s for the array.", + g_raid_node_event2str(event)); + + if (event == G_RAID_NODE_E_WAKE) + return (0); + if (sc->sc_md) + G_RAID_MD_EVENT(sc->sc_md, NULL, event); + return (0); +} + +static int +g_raid_access(struct g_provider *pp, int acr, int acw, int ace) +{ + struct g_raid_volume *vol; + struct g_raid_softc *sc; + int dcr, dcw, dce, opens, error = 0; + + g_topology_assert(); + sc = pp->geom->softc; + vol = pp->private; + KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name)); + KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name)); + + G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name, + acr, acw, ace); + + dcr = pp->acr + acr; + dcw = pp->acw + acw; + dce = pp->ace + ace; + + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + /* Deny new opens while dying. */ + if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) { + error = ENXIO; + goto out; + } + if (dcw == 0 && vol->v_dirty) + g_raid_clean(vol, dcw); + vol->v_provider_open += acr + acw + ace; + /* Handle delayed node destruction. */ + if (sc->sc_stopping == G_RAID_DESTROY_DELAYED && + vol->v_provider_open == 0) { + /* Count open volumes. */ + opens = g_raid_nopens(sc); + if (opens == 0) { + sc->sc_stopping = G_RAID_DESTROY_HARD; + /* Wake up worker to make it selfdestruct. */ + g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); + } + } + /* Handle open volume destruction. */ + if (vol->v_stopping && vol->v_provider_open == 0) + g_raid_destroy_volume(vol); +out: + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + return (error); +} + +struct g_raid_softc * +g_raid_create_node(struct g_class *mp, + const char *name, struct g_raid_md_object *md) +{ + struct g_raid_softc *sc; + struct g_geom *gp; + int error; + + g_topology_assert(); + G_RAID_DEBUG(1, "Creating array %s.", name); + + gp = g_new_geomf(mp, "%s", name); + sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO); + gp->start = g_raid_start; + gp->orphan = g_raid_orphan; + gp->access = g_raid_access; + gp->dumpconf = g_raid_dumpconf; + + sc->sc_md = md; + sc->sc_geom = gp; + sc->sc_flags = 0; + TAILQ_INIT(&sc->sc_volumes); + TAILQ_INIT(&sc->sc_disks); + sx_init(&sc->sc_lock, "gmirror:lock"); + mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF); + TAILQ_INIT(&sc->sc_events); + bioq_init(&sc->sc_queue); + gp->softc = sc; + error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0, + "g_raid %s", name); + if (error != 0) { + G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name); + mtx_destroy(&sc->sc_queue_mtx); + sx_destroy(&sc->sc_lock); + g_destroy_geom(sc->sc_geom); + free(sc, M_RAID); + return (NULL); + } + + G_RAID_DEBUG1(0, sc, "Array %s created.", name); + return (sc); +} + +struct g_raid_volume * +g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id) +{ + struct g_raid_volume *vol, *vol1; + int i; + + G_RAID_DEBUG1(1, sc, "Creating volume %s.", name); + vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO); + vol->v_softc = sc; + strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME); + vol->v_state = G_RAID_VOLUME_S_STARTING; + vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN; + bioq_init(&vol->v_inflight); + bioq_init(&vol->v_locked); + LIST_INIT(&vol->v_locks); + for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { + vol->v_subdisks[i].sd_softc = sc; + vol->v_subdisks[i].sd_volume = vol; + vol->v_subdisks[i].sd_pos = i; + vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE; + } + + /* Find free ID for this volume. */ + g_topology_lock(); + vol1 = vol; + if (id >= 0) { + LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { + if (vol1->v_global_id == id) + break; + } + } + if (vol1 != NULL) { + for (id = 0; ; id++) { + LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) { + if (vol1->v_global_id == id) + break; + } + if (vol1 == NULL) + break; + } + } + vol->v_global_id = id; + LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next); + g_topology_unlock(); + + /* Delay root mounting. */ + vol->v_rootmount = root_mount_hold("GRAID"); + G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount); + vol->v_starting = 1; + TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next); + return (vol); +} + +struct g_raid_disk * +g_raid_create_disk(struct g_raid_softc *sc) +{ + struct g_raid_disk *disk; + + G_RAID_DEBUG1(1, sc, "Creating disk."); + disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO); + disk->d_softc = sc; + disk->d_state = G_RAID_DISK_S_NONE; + TAILQ_INIT(&disk->d_subdisks); + TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next); + return (disk); +} + +int g_raid_start_volume(struct g_raid_volume *vol) +{ + struct g_raid_tr_class *class; + struct g_raid_tr_object *obj; + int status; + + G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name); + LIST_FOREACH(class, &g_raid_tr_classes, trc_list) { + G_RAID_DEBUG1(2, vol->v_softc, + "Tasting volume %s for %s transformation.", + vol->v_name, class->name); + obj = (void *)kobj_create((kobj_class_t)class, M_RAID, + M_WAITOK); + obj->tro_class = class; + obj->tro_volume = vol; + status = G_RAID_TR_TASTE(obj, vol); + if (status != G_RAID_TR_TASTE_FAIL) + break; + kobj_delete((kobj_t)obj, M_RAID); + } + if (class == NULL) { + G_RAID_DEBUG1(0, vol->v_softc, + "No transformation module found for %s.", + vol->v_name); + vol->v_tr = NULL; + g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED); + g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN, + G_RAID_EVENT_VOLUME); + return (-1); + } + G_RAID_DEBUG1(2, vol->v_softc, + "Transformation module %s chosen for %s.", + class->name, vol->v_name); + vol->v_tr = obj; + return (0); +} + +int +g_raid_destroy_node(struct g_raid_softc *sc, int worker) +{ + struct g_raid_volume *vol, *tmpv; + struct g_raid_disk *disk, *tmpd; + int error = 0; + + sc->sc_stopping = G_RAID_DESTROY_HARD; + TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) { + if (g_raid_destroy_volume(vol)) + error = EBUSY; + } + if (error) + return (error); + TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) { + if (g_raid_destroy_disk(disk)) + error = EBUSY; + } + if (error) + return (error); + if (sc->sc_md) { + G_RAID_MD_FREE(sc->sc_md); + kobj_delete((kobj_t)sc->sc_md, M_RAID); + sc->sc_md = NULL; + } + if (sc->sc_geom != NULL) { + G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name); + g_topology_lock(); + sc->sc_geom->softc = NULL; + g_wither_geom(sc->sc_geom, ENXIO); + g_topology_unlock(); + sc->sc_geom = NULL; + } else + G_RAID_DEBUG(1, "Array destroyed."); + if (worker) { + g_raid_event_cancel(sc, sc); + mtx_destroy(&sc->sc_queue_mtx); + sx_xunlock(&sc->sc_lock); + sx_destroy(&sc->sc_lock); + wakeup(&sc->sc_stopping); + free(sc, M_RAID); + curthread->td_pflags &= ~TDP_GEOM; + G_RAID_DEBUG(1, "Thread exiting."); + kproc_exit(0); + } else { + /* Wake up worker to make it selfdestruct. */ + g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); + } + return (0); +} + +int +g_raid_destroy_volume(struct g_raid_volume *vol) +{ + struct g_raid_softc *sc; + struct g_raid_disk *disk; + int i; + + sc = vol->v_softc; + G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name); + vol->v_stopping = 1; + if (vol->v_state != G_RAID_VOLUME_S_STOPPED) { + if (vol->v_tr) { + G_RAID_TR_STOP(vol->v_tr); + return (EBUSY); + } else + vol->v_state = G_RAID_VOLUME_S_STOPPED; + } + if (g_raid_event_check(sc, vol) != 0) + return (EBUSY); + if (vol->v_provider != NULL) + return (EBUSY); + if (vol->v_provider_open != 0) + return (EBUSY); + if (vol->v_tr) { + G_RAID_TR_FREE(vol->v_tr); + kobj_delete((kobj_t)vol->v_tr, M_RAID); + vol->v_tr = NULL; + } + if (vol->v_rootmount) + root_mount_rel(vol->v_rootmount); + g_topology_lock(); + LIST_REMOVE(vol, v_global_next); + g_topology_unlock(); + TAILQ_REMOVE(&sc->sc_volumes, vol, v_next); + for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) { + g_raid_event_cancel(sc, &vol->v_subdisks[i]); + disk = vol->v_subdisks[i].sd_disk; + if (disk == NULL) + continue; + TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next); + } + G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name); + if (sc->sc_md) + G_RAID_MD_FREE_VOLUME(sc->sc_md, vol); + g_raid_event_cancel(sc, vol); + free(vol, M_RAID); + if (sc->sc_stopping == G_RAID_DESTROY_HARD) { + /* Wake up worker to let it selfdestruct. */ + g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); + } + return (0); +} + +int +g_raid_destroy_disk(struct g_raid_disk *disk) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd, *tmp; + + sc = disk->d_softc; + G_RAID_DEBUG1(2, sc, "Destroying disk."); + if (disk->d_consumer) { + g_raid_kill_consumer(sc, disk->d_consumer); + disk->d_consumer = NULL; + } + TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) { + g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, + G_RAID_EVENT_SUBDISK); + TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next); + sd->sd_disk = NULL; + } + TAILQ_REMOVE(&sc->sc_disks, disk, d_next); + if (sc->sc_md) + G_RAID_MD_FREE_DISK(sc->sc_md, disk); + g_raid_event_cancel(sc, disk); + free(disk, M_RAID); + return (0); +} + +int +g_raid_destroy(struct g_raid_softc *sc, int how) +{ + int opens; + + g_topology_assert_not(); + if (sc == NULL) + return (ENXIO); + sx_assert(&sc->sc_lock, SX_XLOCKED); + + /* Count open volumes. */ + opens = g_raid_nopens(sc); + + /* React on some opened volumes. */ + if (opens > 0) { + switch (how) { + case G_RAID_DESTROY_SOFT: + G_RAID_DEBUG1(1, sc, + "%d volumes are still open.", + opens); + return (EBUSY); + case G_RAID_DESTROY_DELAYED: + G_RAID_DEBUG1(1, sc, + "Array will be destroyed on last close."); + sc->sc_stopping = G_RAID_DESTROY_DELAYED; + return (EBUSY); + case G_RAID_DESTROY_HARD: + G_RAID_DEBUG1(1, sc, + "%d volumes are still open.", + opens); + } + } + + /* Mark node for destruction. */ + sc->sc_stopping = G_RAID_DESTROY_HARD; + /* Wake up worker to let it selfdestruct. */ + g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0); + /* Sleep until node destroyed. */ + sx_sleep(&sc->sc_stopping, &sc->sc_lock, + PRIBIO | PDROP, "r:destroy", 0); + return (0); +} + +static void +g_raid_taste_orphan(struct g_consumer *cp) +{ + + KASSERT(1 == 0, ("%s called while tasting %s.", __func__, + cp->provider->name)); +} + +static struct g_geom * +g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) +{ + struct g_consumer *cp; + struct g_geom *gp, *geom; + struct g_raid_md_class *class; + struct g_raid_md_object *obj; + int status; + + g_topology_assert(); + g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); + G_RAID_DEBUG(2, "Tasting provider %s.", pp->name); + + gp = g_new_geomf(mp, "mirror:taste"); + /* + * This orphan function should be never called. + */ + gp->orphan = g_raid_taste_orphan; + cp = g_new_consumer(gp); + g_attach(cp, pp); + + geom = NULL; + LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { + G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.", + pp->name, class->name); + obj = (void *)kobj_create((kobj_class_t)class, M_RAID, + M_WAITOK); + obj->mdo_class = class; + status = G_RAID_MD_TASTE(obj, mp, cp, &geom); + if (status != G_RAID_MD_TASTE_NEW) + kobj_delete((kobj_t)obj, M_RAID); + if (status != G_RAID_MD_TASTE_FAIL) + break; + } + + g_detach(cp); + g_destroy_consumer(cp); + g_destroy_geom(gp); + G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name); + return (geom); +} + +int +g_raid_create_node_format(const char *format, struct g_geom **gp) +{ + struct g_raid_md_class *class; + struct g_raid_md_object *obj; + int status; + + G_RAID_DEBUG(2, "Creating array for %s metadata.", format); + LIST_FOREACH(class, &g_raid_md_classes, mdc_list) { + if (strcasecmp(class->name, format) == 0) + break; + } + if (class == NULL) { + G_RAID_DEBUG(1, "No support for %s metadata.", format); + return (G_RAID_MD_TASTE_FAIL); + } + obj = (void *)kobj_create((kobj_class_t)class, M_RAID, + M_WAITOK); + obj->mdo_class = class; + status = G_RAID_MD_CREATE(obj, &g_raid_class, gp); + if (status != G_RAID_MD_TASTE_NEW) + kobj_delete((kobj_t)obj, M_RAID); + return (status); +} + +static int +g_raid_destroy_geom(struct gctl_req *req __unused, + struct g_class *mp __unused, struct g_geom *gp) +{ + struct g_raid_softc *sc; + int error; + + g_topology_unlock(); + sc = gp->softc; + sx_xlock(&sc->sc_lock); + g_cancel_event(sc); + error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT); + if (error != 0) + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + return (error); +} + +void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, + struct g_raid_subdisk *sd, struct g_raid_disk *disk) +{ + + if (sc->sc_stopping == G_RAID_DESTROY_HARD) + return; + if (sc->sc_md) + G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk); +} + +void g_raid_fail_disk(struct g_raid_softc *sc, + struct g_raid_subdisk *sd, struct g_raid_disk *disk) +{ + + if (disk == NULL) + disk = sd->sd_disk; + if (disk == NULL) { + G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!"); + return; + } + if (disk->d_state != G_RAID_DISK_S_ACTIVE) { + G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a " + "wrong state (%s)!", g_raid_disk_state2str(disk->d_state)); + return; + } + if (sc->sc_md) + G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk); +} + +static void +g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, + struct g_consumer *cp, struct g_provider *pp) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + int i, s; + + g_topology_assert(); + + sc = gp->softc; + if (sc == NULL) + return; + if (pp != NULL) { + vol = pp->private; + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + sbuf_printf(sb, "%s\n", indent, + vol->v_name); + sbuf_printf(sb, "%s%s\n", indent, + g_raid_volume_level2str(vol->v_raid_level, + vol->v_raid_level_qualifier)); + sbuf_printf(sb, + "%s%s\n", indent, + vol->v_tr ? vol->v_tr->tro_class->name : "NONE"); + sbuf_printf(sb, "%s%u\n", indent, + vol->v_disks_count); + sbuf_printf(sb, "%s%u\n", indent, + vol->v_strip_size); + sbuf_printf(sb, "%s%s\n", indent, + g_raid_volume_state2str(vol->v_state)); + sbuf_printf(sb, "%s%s\n", indent, + vol->v_dirty ? "Yes" : "No"); + sbuf_printf(sb, "%s", indent); + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if (sd->sd_disk != NULL && + sd->sd_disk->d_consumer != NULL) { + sbuf_printf(sb, "%s ", + g_raid_get_diskname(sd->sd_disk)); + } else { + sbuf_printf(sb, "NONE "); + } + sbuf_printf(sb, "(%s", + g_raid_subdisk_state2str(sd->sd_state)); + if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || + sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { + sbuf_printf(sb, " %d%%", + (int)(sd->sd_rebuild_pos * 100 / + sd->sd_size)); + } + sbuf_printf(sb, ")"); + if (i + 1 < vol->v_disks_count) + sbuf_printf(sb, ", "); + } + sbuf_printf(sb, "\n"); + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + } else if (cp != NULL) { + disk = cp->private; + if (disk == NULL) + return; + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + sbuf_printf(sb, "%s%s", indent, + g_raid_disk_state2str(disk->d_state)); + if (!TAILQ_EMPTY(&disk->d_subdisks)) { + sbuf_printf(sb, " ("); + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + sbuf_printf(sb, "%s", + g_raid_subdisk_state2str(sd->sd_state)); + if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || + sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { + sbuf_printf(sb, " %d%%", + (int)(sd->sd_rebuild_pos * 100 / + sd->sd_size)); + } + if (TAILQ_NEXT(sd, sd_next)) + sbuf_printf(sb, ", "); + } + sbuf_printf(sb, ")"); + } + sbuf_printf(sb, "\n"); + sbuf_printf(sb, "%s", indent); + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + sbuf_printf(sb, "r%d(%s):%d@%ju", + sd->sd_volume->v_global_id, + sd->sd_volume->v_name, + sd->sd_pos, sd->sd_offset); + if (TAILQ_NEXT(sd, sd_next)) + sbuf_printf(sb, ", "); + } + sbuf_printf(sb, "\n"); + sbuf_printf(sb, "%s%d\n", indent, + disk->d_read_errs); + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + } else { + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + if (sc->sc_md) { + sbuf_printf(sb, "%s%s\n", indent, + sc->sc_md->mdo_class->name); + } + if (!TAILQ_EMPTY(&sc->sc_volumes)) { + s = 0xff; + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + if (vol->v_state < s) + s = vol->v_state; + } + sbuf_printf(sb, "%s%s\n", indent, + g_raid_volume_state2str(s)); + } + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + } +} + +static void +g_raid_shutdown_pre_sync(void *arg, int howto) +{ + struct g_class *mp; + struct g_geom *gp, *gp2; + struct g_raid_softc *sc; + int error; + + mp = arg; + DROP_GIANT(); + g_topology_lock(); + LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { + if ((sc = gp->softc) == NULL) + continue; + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + g_cancel_event(sc); + error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED); + if (error != 0) + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + } + g_topology_unlock(); + PICKUP_GIANT(); +} + +static void +g_raid_init(struct g_class *mp) +{ + + g_raid_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync, + g_raid_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST); + if (g_raid_pre_sync == NULL) + G_RAID_DEBUG(0, "Warning! Cannot register shutdown event."); + g_raid_started = 1; +} + +static void +g_raid_fini(struct g_class *mp) +{ + + if (g_raid_pre_sync != NULL) + EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_raid_pre_sync); + g_raid_started = 0; +} + +int +g_raid_md_modevent(module_t mod, int type, void *arg) +{ + struct g_raid_md_class *class, *c, *nc; + int error; + + error = 0; + class = arg; + switch (type) { + case MOD_LOAD: + c = LIST_FIRST(&g_raid_md_classes); + if (c == NULL || c->mdc_priority > class->mdc_priority) + LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list); + else { + while ((nc = LIST_NEXT(c, mdc_list)) != NULL && + nc->mdc_priority < class->mdc_priority) + c = nc; + LIST_INSERT_AFTER(c, class, mdc_list); + } + if (g_raid_started) + g_retaste(&g_raid_class); + break; + case MOD_UNLOAD: + LIST_REMOVE(class, mdc_list); + break; + default: + error = EOPNOTSUPP; + break; + } + + return (error); +} + +int +g_raid_tr_modevent(module_t mod, int type, void *arg) +{ + struct g_raid_tr_class *class, *c, *nc; + int error; + + error = 0; + class = arg; + switch (type) { + case MOD_LOAD: + c = LIST_FIRST(&g_raid_tr_classes); + if (c == NULL || c->trc_priority > class->trc_priority) + LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list); + else { + while ((nc = LIST_NEXT(c, trc_list)) != NULL && + nc->trc_priority < class->trc_priority) + c = nc; + LIST_INSERT_AFTER(c, class, trc_list); + } + break; + case MOD_UNLOAD: + LIST_REMOVE(class, trc_list); + break; + default: + error = EOPNOTSUPP; + break; + } + + return (error); +} + +/* + * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid) + * to reduce module priority, allowing submodules to register them first. + */ +static moduledata_t g_raid_mod = { + "g_raid", + g_modevent, + &g_raid_class +}; +DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD); +MODULE_VERSION(geom_raid, 0); diff --git a/sys/geom/raid/g_raid.h b/sys/geom/raid/g_raid.h new file mode 100644 index 000000000000..1c14ad6738bf --- /dev/null +++ b/sys/geom/raid/g_raid.h @@ -0,0 +1,403 @@ +/*- + * Copyright (c) 2010 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + */ + +#ifndef _G_RAID_H_ +#define _G_RAID_H_ + +#include +#include +#include +#include + +#define G_RAID_CLASS_NAME "RAID" + +#define G_RAID_MAGIC "GEOM::RAID" + +#define G_RAID_VERSION 0 + +struct g_raid_md_object; +struct g_raid_tr_object; + +#define G_RAID_DEVICE_FLAG_NOAUTOSYNC 0x0000000000000001ULL +#define G_RAID_DEVICE_FLAG_NOFAILSYNC 0x0000000000000002ULL +#define G_RAID_DEVICE_FLAG_MASK (G_RAID_DEVICE_FLAG_NOAUTOSYNC | \ + G_RAID_DEVICE_FLAG_NOFAILSYNC) + +#ifdef _KERNEL +extern u_int g_raid_aggressive_spare; +extern u_int g_raid_debug; +extern int g_raid_read_err_thresh; +extern u_int g_raid_start_timeout; +extern struct g_class g_raid_class; + +#define G_RAID_DEBUG(lvl, fmt, ...) do { \ + if (g_raid_debug >= (lvl)) { \ + if (g_raid_debug > 0) { \ + printf("GEOM_RAID[%u]: " fmt "\n", \ + lvl, ## __VA_ARGS__); \ + } else { \ + printf("GEOM_RAID: " fmt "\n", \ + ## __VA_ARGS__); \ + } \ + } \ +} while (0) +#define G_RAID_DEBUG1(lvl, sc, fmt, ...) do { \ + if (g_raid_debug >= (lvl)) { \ + if (g_raid_debug > 0) { \ + printf("GEOM_RAID[%u]: %s: " fmt "\n", \ + lvl, (sc)->sc_name, ## __VA_ARGS__); \ + } else { \ + printf("GEOM_RAID: %s: " fmt "\n", \ + (sc)->sc_name, ## __VA_ARGS__); \ + } \ + } \ +} while (0) +#define G_RAID_LOGREQ(lvl, bp, fmt, ...) do { \ + if (g_raid_debug >= (lvl)) { \ + if (g_raid_debug > 0) { \ + printf("GEOM_RAID[%u]: " fmt " ", \ + lvl, ## __VA_ARGS__); \ + } else \ + printf("GEOM_RAID: " fmt " ", ## __VA_ARGS__); \ + g_print_bio(bp); \ + printf("\n"); \ + } \ +} while (0) + +/* + * Flags we use to distinguish I/O initiated by the TR layer to maintain + * the volume's characteristics, fix subdisks, extra copies of data, etc. + * + * G_RAID_BIO_FLAG_SYNC I/O to update an extra copy of the data + * for RAID volumes that maintain extra data + * and need to rebuild that data. + * G_RAID_BIO_FLAG_REMAP I/O done to try to provoke a subdisk into + * doing some desirable action such as bad + * block remapping after we detect a bad part + * of the disk. + * G_RAID_BIO_FLAG_LOCKED I/O holds range lock that should re released. + * + * and the following meta item: + * G_RAID_BIO_FLAG_SPECIAL And of the I/O flags that need to make it + * through the range locking which would + * otherwise defer the I/O until after that + * range is unlocked. + */ +#define G_RAID_BIO_FLAG_SYNC 0x01 +#define G_RAID_BIO_FLAG_REMAP 0x02 +#define G_RAID_BIO_FLAG_SPECIAL \ + (G_RAID_BIO_FLAG_SYNC|G_RAID_BIO_FLAG_REMAP) +#define G_RAID_BIO_FLAG_LOCKED 0x80 + +struct g_raid_lock { + off_t l_offset; + off_t l_length; + void *l_callback_arg; + int l_pending; + LIST_ENTRY(g_raid_lock) l_next; +}; + +#define G_RAID_EVENT_WAIT 0x01 +#define G_RAID_EVENT_VOLUME 0x02 +#define G_RAID_EVENT_SUBDISK 0x04 +#define G_RAID_EVENT_DISK 0x08 +#define G_RAID_EVENT_DONE 0x10 +struct g_raid_event { + void *e_tgt; + int e_event; + int e_flags; + int e_error; + TAILQ_ENTRY(g_raid_event) e_next; +}; +#define G_RAID_DISK_S_NONE 0x00 /* State is unknown. */ +#define G_RAID_DISK_S_OFFLINE 0x01 /* Missing disk placeholder. */ +#define G_RAID_DISK_S_FAILED 0x02 /* Failed. */ +#define G_RAID_DISK_S_STALE_FAILED 0x03 /* Old failed. */ +#define G_RAID_DISK_S_SPARE 0x04 /* Hot-spare. */ +#define G_RAID_DISK_S_STALE 0x05 /* Old disk, unused now. */ +#define G_RAID_DISK_S_ACTIVE 0x06 /* Operational. */ + +#define G_RAID_DISK_E_DISCONNECTED 0x01 + +struct g_raid_disk { + struct g_raid_softc *d_softc; /* Back-pointer to softc. */ + struct g_consumer *d_consumer; /* GEOM disk consumer. */ + void *d_md_data; /* Disk's metadata storage. */ + struct g_kerneldump d_kd; /* Kernel dumping method/args. */ + uint64_t d_flags; /* Additional flags. */ + u_int d_state; /* Disk state. */ + u_int d_load; /* Disk average load. */ + off_t d_last_offset; /* Last head offset. */ + int d_read_errs; /* Count of the read errors */ + TAILQ_HEAD(, g_raid_subdisk) d_subdisks; /* List of subdisks. */ + TAILQ_ENTRY(g_raid_disk) d_next; /* Next disk in the node. */ +}; + +#define G_RAID_SUBDISK_S_NONE 0x00 /* Absent. */ +#define G_RAID_SUBDISK_S_FAILED 0x01 /* Failed. */ +#define G_RAID_SUBDISK_S_NEW 0x02 /* Blank. */ +#define G_RAID_SUBDISK_S_REBUILD 0x03 /* Blank + rebuild. */ +#define G_RAID_SUBDISK_S_UNINITIALIZED 0x04 /* Disk of the new volume. */ +#define G_RAID_SUBDISK_S_STALE 0x05 /* Dirty. */ +#define G_RAID_SUBDISK_S_RESYNC 0x06 /* Dirty + check/repair. */ +#define G_RAID_SUBDISK_S_ACTIVE 0x07 /* Usable. */ + +#define G_RAID_SUBDISK_E_NEW 0x01 /* A new subdisk has arrived */ +#define G_RAID_SUBDISK_E_FAILED 0x02 /* A subdisk failed, but remains in volume */ +#define G_RAID_SUBDISK_E_DISCONNECTED 0x03 /* A subdisk removed from volume. */ +#define G_RAID_SUBDISK_E_FIRST_TR_PRIVATE 0x80 /* translation private events */ + +#define G_RAID_SUBDISK_POS(sd) \ + ((sd)->sd_disk ? ((sd)->sd_disk->d_last_offset - (sd)->sd_offset) : 0) +#define G_RAID_SUBDISK_TRACK_SIZE (1 * 1024 * 1024) +#define G_RAID_SUBDISK_LOAD(sd) \ + ((sd)->sd_disk ? ((sd)->sd_disk->d_load) : 0) +#define G_RAID_SUBDISK_LOAD_SCALE 256 + +struct g_raid_subdisk { + struct g_raid_softc *sd_softc; /* Back-pointer to softc. */ + struct g_raid_disk *sd_disk; /* Where this subdisk lives. */ + struct g_raid_volume *sd_volume; /* Volume, sd is a part of. */ + off_t sd_offset; /* Offset on the disk. */ + off_t sd_size; /* Size on the disk. */ + u_int sd_pos; /* Position in volume. */ + u_int sd_state; /* Subdisk state. */ + off_t sd_rebuild_pos; /* Rebuild position. */ + int sd_recovery; /* Count of recovery reqs. */ + TAILQ_ENTRY(g_raid_subdisk) sd_next; /* Next subdisk on disk. */ +}; + +#define G_RAID_MAX_SUBDISKS 16 +#define G_RAID_MAX_VOLUMENAME 32 + +#define G_RAID_VOLUME_S_STARTING 0x00 +#define G_RAID_VOLUME_S_BROKEN 0x01 +#define G_RAID_VOLUME_S_DEGRADED 0x02 +#define G_RAID_VOLUME_S_SUBOPTIMAL 0x03 +#define G_RAID_VOLUME_S_OPTIMAL 0x04 +#define G_RAID_VOLUME_S_UNSUPPORTED 0x05 +#define G_RAID_VOLUME_S_STOPPED 0x06 + +#define G_RAID_VOLUME_S_ALIVE(s) \ + ((s) == G_RAID_VOLUME_S_DEGRADED || \ + (s) == G_RAID_VOLUME_S_SUBOPTIMAL || \ + (s) == G_RAID_VOLUME_S_OPTIMAL) + +#define G_RAID_VOLUME_E_DOWN 0x00 +#define G_RAID_VOLUME_E_UP 0x01 +#define G_RAID_VOLUME_E_START 0x10 +#define G_RAID_VOLUME_E_STARTMD 0x11 + +#define G_RAID_VOLUME_RL_RAID0 0x00 +#define G_RAID_VOLUME_RL_RAID1 0x01 +#define G_RAID_VOLUME_RL_RAID3 0x03 +#define G_RAID_VOLUME_RL_RAID4 0x04 +#define G_RAID_VOLUME_RL_RAID5 0x05 +#define G_RAID_VOLUME_RL_RAID6 0x06 +#define G_RAID_VOLUME_RL_RAID1E 0x11 +#define G_RAID_VOLUME_RL_SINGLE 0x0f +#define G_RAID_VOLUME_RL_CONCAT 0x1f +#define G_RAID_VOLUME_RL_RAID5E 0x15 +#define G_RAID_VOLUME_RL_RAID5EE 0x25 +#define G_RAID_VOLUME_RL_UNKNOWN 0xff + +#define G_RAID_VOLUME_RLQ_NONE 0x00 +#define G_RAID_VOLUME_RLQ_UNKNOWN 0xff + +struct g_raid_volume; + +struct g_raid_volume { + struct g_raid_softc *v_softc; /* Back-pointer to softc. */ + struct g_provider *v_provider; /* GEOM provider. */ + struct g_raid_subdisk v_subdisks[G_RAID_MAX_SUBDISKS]; + /* Subdisks of this volume. */ + void *v_md_data; /* Volume's metadata storage. */ + struct g_raid_tr_object *v_tr; /* Transformation object. */ + char v_name[G_RAID_MAX_VOLUMENAME]; + /* Volume name. */ + u_int v_state; /* Volume state. */ + u_int v_raid_level; /* Array RAID level. */ + u_int v_raid_level_qualifier; /* RAID level det. */ + u_int v_disks_count; /* Number of disks in array. */ + u_int v_strip_size; /* Array strip size. */ + u_int v_sectorsize; /* Volume sector size. */ + off_t v_mediasize; /* Volume media size. */ + struct bio_queue_head v_inflight; /* In-flight write requests. */ + struct bio_queue_head v_locked; /* Blocked I/O requests. */ + LIST_HEAD(, g_raid_lock) v_locks; /* List of locked regions. */ + int v_pending_lock; /* writes to locked region */ + int v_dirty; /* Volume is DIRTY. */ + struct timeval v_last_done; /* Time of the last I/O. */ + time_t v_last_write; /* Time of the last write. */ + u_int v_writes; /* Number of active writes. */ + struct root_hold_token *v_rootmount; /* Root mount delay token. */ + int v_starting; /* Volume is starting */ + int v_stopping; /* Volume is stopping */ + int v_provider_open; /* Number of opens. */ + int v_global_id; /* Global volume ID (rX). */ + TAILQ_ENTRY(g_raid_volume) v_next; /* List of volumes entry. */ + LIST_ENTRY(g_raid_volume) v_global_next; /* Global list entry. */ +}; + +#define G_RAID_NODE_E_WAKE 0x00 +#define G_RAID_NODE_E_START 0x01 + +struct g_raid_softc { + struct g_raid_md_object *sc_md; /* Metadata object. */ + struct g_geom *sc_geom; /* GEOM class instance. */ + uint64_t sc_flags; /* Additional flags. */ + TAILQ_HEAD(, g_raid_volume) sc_volumes; /* List of volumes. */ + TAILQ_HEAD(, g_raid_disk) sc_disks; /* List of disks. */ + struct sx sc_lock; /* Main node lock. */ + struct proc *sc_worker; /* Worker process. */ + struct mtx sc_queue_mtx; /* Worker queues lock. */ + TAILQ_HEAD(, g_raid_event) sc_events; /* Worker events queue. */ + struct bio_queue_head sc_queue; /* Worker I/O queue. */ + int sc_stopping; /* Node is stopping */ +}; +#define sc_name sc_geom->name + +/* + * KOBJ parent class of metadata processing modules. + */ +struct g_raid_md_class { + KOBJ_CLASS_FIELDS; + int mdc_priority; + LIST_ENTRY(g_raid_md_class) mdc_list; +}; + +/* + * KOBJ instance of metadata processing module. + */ +struct g_raid_md_object { + KOBJ_FIELDS; + struct g_raid_md_class *mdo_class; + struct g_raid_softc *mdo_softc; /* Back-pointer to softc. */ +}; + +int g_raid_md_modevent(module_t, int, void *); + +#define G_RAID_MD_DECLARE(name) \ + static moduledata_t name##_mod = { \ + #name, \ + g_raid_md_modevent, \ + &name##_class \ + }; \ + DECLARE_MODULE(name, name##_mod, SI_SUB_DRIVERS, SI_ORDER_SECOND); \ + MODULE_DEPEND(name, geom_raid, 0, 0, 0) + +/* + * KOBJ parent class of data transformation modules. + */ +struct g_raid_tr_class { + KOBJ_CLASS_FIELDS; + int trc_priority; + LIST_ENTRY(g_raid_tr_class) trc_list; +}; + +/* + * KOBJ instance of data transformation module. + */ +struct g_raid_tr_object { + KOBJ_FIELDS; + struct g_raid_tr_class *tro_class; + struct g_raid_volume *tro_volume; /* Back-pointer to volume. */ +}; + +int g_raid_tr_modevent(module_t, int, void *); + +#define G_RAID_TR_DECLARE(name) \ + static moduledata_t name##_mod = { \ + #name, \ + g_raid_tr_modevent, \ + &name##_class \ + }; \ + DECLARE_MODULE(name, name##_mod, SI_SUB_DRIVERS, SI_ORDER_FIRST); \ + MODULE_DEPEND(name, geom_raid, 0, 0, 0) + +const char * g_raid_volume_level2str(int level, int qual); +int g_raid_volume_str2level(const char *str, int *level, int *qual); +const char * g_raid_volume_state2str(int state); +const char * g_raid_subdisk_state2str(int state); +const char * g_raid_disk_state2str(int state); + +struct g_raid_softc * g_raid_create_node(struct g_class *mp, + const char *name, struct g_raid_md_object *md); +int g_raid_create_node_format(const char *format, struct g_geom **gp); +struct g_raid_volume * g_raid_create_volume(struct g_raid_softc *sc, + const char *name, int id); +struct g_raid_disk * g_raid_create_disk(struct g_raid_softc *sc); +const char * g_raid_get_diskname(struct g_raid_disk *disk); + +int g_raid_start_volume(struct g_raid_volume *vol); + +int g_raid_destroy_node(struct g_raid_softc *sc, int worker); +int g_raid_destroy_volume(struct g_raid_volume *vol); +int g_raid_destroy_disk(struct g_raid_disk *disk); + +void g_raid_iodone(struct bio *bp, int error); +void g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp); +int g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd, + void *virtual, vm_offset_t physical, off_t offset, size_t length); + +struct g_consumer *g_raid_open_consumer(struct g_raid_softc *sc, + const char *name); +void g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp); + +void g_raid_report_disk_state(struct g_raid_disk *disk); +void g_raid_change_disk_state(struct g_raid_disk *disk, int state); +void g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state); +void g_raid_change_volume_state(struct g_raid_volume *vol, int state); + +void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol, + struct g_raid_subdisk *sd, struct g_raid_disk *disk); +void g_raid_fail_disk(struct g_raid_softc *sc, + struct g_raid_subdisk *sd, struct g_raid_disk *disk); + +void g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp); +int g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr, + void *virtual, vm_offset_t physical, off_t offset, size_t length); + +u_int g_raid_ndisks(struct g_raid_softc *sc, int state); +u_int g_raid_nsubdisks(struct g_raid_volume *vol, int state); +u_int g_raid_nopens(struct g_raid_softc *sc); +struct g_raid_subdisk * g_raid_get_subdisk(struct g_raid_volume *vol, + int state); +#define G_RAID_DESTROY_SOFT 0 +#define G_RAID_DESTROY_DELAYED 1 +#define G_RAID_DESTROY_HARD 2 +int g_raid_destroy(struct g_raid_softc *sc, int how); +int g_raid_event_send(void *arg, int event, int flags); +int g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len, + struct bio *ignore, void *argp); +int g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len); + +g_ctl_req_t g_raid_ctl; +#endif /* _KERNEL */ + +#endif /* !_G_RAID_H_ */ diff --git a/sys/geom/raid/g_raid_ctl.c b/sys/geom/raid/g_raid_ctl.c new file mode 100644 index 000000000000..028aa94c308e --- /dev/null +++ b/sys/geom/raid/g_raid_ctl.c @@ -0,0 +1,217 @@ +/*- + * Copyright (c) 2010 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "g_raid_md_if.h" + + +static struct g_raid_softc * +g_raid_find_node(struct g_class *mp, const char *name) +{ + struct g_raid_softc *sc; + struct g_geom *gp; + + LIST_FOREACH(gp, &mp->geom, geom) { + sc = gp->softc; + if (sc == NULL) + continue; + if (sc->sc_stopping != 0) + continue; + if (strcasecmp(sc->sc_name, name) == 0) + return (sc); + } + return (NULL); +} + +static void +g_raid_ctl_label(struct gctl_req *req, struct g_class *mp) +{ + struct g_geom *geom; + struct g_raid_softc *sc; + const char *format; + int *nargs; + int crstatus, ctlstatus; + char buf[64]; + + nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); + if (nargs == NULL) { + gctl_error(req, "No '%s' argument.", "nargs"); + return; + } + if (*nargs < 4) { + gctl_error(req, "Invalid number of arguments."); + return; + } + format = gctl_get_asciiparam(req, "arg0"); + if (format == NULL) { + gctl_error(req, "No format recieved."); + return; + } + crstatus = g_raid_create_node_format(format, &geom); + if (crstatus == G_RAID_MD_TASTE_FAIL) { + gctl_error(req, "Failed to create array with format '%s'.", + format); + return; + } + sc = (struct g_raid_softc *)geom->softc; + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + ctlstatus = G_RAID_MD_CTL(sc->sc_md, req); + if (ctlstatus < 0) { + gctl_error(req, "Command failed: %d.", ctlstatus); + if (crstatus == G_RAID_MD_TASTE_NEW) + g_raid_destroy_node(sc, 0); + } else { + if (crstatus == G_RAID_MD_TASTE_NEW) + snprintf(buf, sizeof(buf), "%s created\n", sc->sc_name); + else + snprintf(buf, sizeof(buf), "%s reused\n", sc->sc_name); + gctl_set_param_err(req, "output", buf, strlen(buf) + 1); + } + sx_xunlock(&sc->sc_lock); + g_topology_lock(); +} + +static void +g_raid_ctl_stop(struct gctl_req *req, struct g_class *mp) +{ + struct g_raid_softc *sc; + const char *nodename; + int *nargs, *force; + int error, how; + + nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); + if (nargs == NULL) { + gctl_error(req, "No '%s' argument.", "nargs"); + return; + } + if (*nargs != 1) { + gctl_error(req, "Invalid number of arguments."); + return; + } + nodename = gctl_get_asciiparam(req, "arg0"); + if (nodename == NULL) { + gctl_error(req, "No array name recieved."); + return; + } + sc = g_raid_find_node(mp, nodename); + if (sc == NULL) { + gctl_error(req, "Array '%s' not found.", nodename); + return; + } + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (force != NULL && *force) + how = G_RAID_DESTROY_HARD; + else + how = G_RAID_DESTROY_SOFT; + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + error = g_raid_destroy(sc, how); + if (error != 0) + sx_xunlock(&sc->sc_lock); + g_topology_lock(); +} + +static void +g_raid_ctl_other(struct gctl_req *req, struct g_class *mp) +{ + struct g_raid_softc *sc; + const char *nodename; + int *nargs; + int ctlstatus; + + nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); + if (nargs == NULL) { + gctl_error(req, "No '%s' argument.", "nargs"); + return; + } + if (*nargs < 1) { + gctl_error(req, "Invalid number of arguments."); + return; + } + nodename = gctl_get_asciiparam(req, "arg0"); + if (nodename == NULL) { + gctl_error(req, "No array name recieved."); + return; + } + sc = g_raid_find_node(mp, nodename); + if (sc == NULL) { + gctl_error(req, "Array '%s' not found.", nodename); + return; + } + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + if (sc->sc_md != NULL) { + ctlstatus = G_RAID_MD_CTL(sc->sc_md, req); + if (ctlstatus < 0) + gctl_error(req, "Command failed: %d.", ctlstatus); + } + sx_xunlock(&sc->sc_lock); + g_topology_lock(); +} + +void +g_raid_ctl(struct gctl_req *req, struct g_class *mp, const char *verb) +{ + uint32_t *version; + + g_topology_assert(); + + version = gctl_get_paraml(req, "version", sizeof(*version)); + if (version == NULL) { + gctl_error(req, "No '%s' argument.", "version"); + return; + } + if (*version != G_RAID_VERSION) { + gctl_error(req, "Userland and kernel parts are out of sync."); + return; + } + + if (strcmp(verb, "label") == 0) + g_raid_ctl_label(req, mp); + else if (strcmp(verb, "stop") == 0) + g_raid_ctl_stop(req, mp); + else + g_raid_ctl_other(req, mp); +} diff --git a/sys/geom/raid/g_raid_md_if.m b/sys/geom/raid/g_raid_md_if.m new file mode 100644 index 000000000000..05e9f66957cc --- /dev/null +++ b/sys/geom/raid/g_raid_md_if.m @@ -0,0 +1,156 @@ +#- +# Copyright (c) 2010 Alexander Motin +# All rights reserved. +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions +# are met: +# +# 1. Redistributions of source code must retain the above copyright +# notice, this list of conditions and the following disclaimer. +# 2. Redistributions in binary form must reproduce the above copyright +# notice, this list of conditions and the following disclaimer in the +# documentation and/or other materials provided with the distribution. +# +# THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR +# IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES +# OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. +# IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, +# INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT +# NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF +# THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +# +# $FreeBSD$ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +# The G_RAID metadata class interface. + +INTERFACE g_raid_md; + +HEADER { +#define G_RAID_MD_TASTE_FAIL -1 +#define G_RAID_MD_TASTE_EXISTING 0 +#define G_RAID_MD_TASTE_NEW 1 +}; + +# Default implementations of methods. +CODE { + static int + g_raid_md_create_default(struct g_raid_md_object *md) + { + + return (G_RAID_MD_TASTE_FAIL); + } + + static int + g_raid_md_ctl_default(struct g_raid_md_object *md, + struct gctl_req *req) + { + + return (-1); + } + + static int + g_raid_md_volume_event_default(struct g_raid_md_object *md, + struct g_raid_volume *vol, u_int event) + { + + return (-1); + } + + static int + g_raid_md_free_disk_default(struct g_raid_md_object *md, + struct g_raid_volume *vol) + { + + return (0); + } + + static int + g_raid_md_free_volume_default(struct g_raid_md_object *md, + struct g_raid_volume *vol) + { + + return (0); + } +}; + +# create() - create new node from scratch. +METHOD int create { + struct g_raid_md_object *md; + struct g_class *mp; + struct g_geom **gp; +} DEFAULT g_raid_md_create_default; + +# taste() - taste disk and, if needed, create new node. +METHOD int taste { + struct g_raid_md_object *md; + struct g_class *mp; + struct g_consumer *cp; + struct g_geom **gp; +}; + +# ctl() - user-level control commands handling method. +METHOD int ctl { + struct g_raid_md_object *md; + struct gctl_req *req; +} DEFAULT g_raid_md_ctl_default; + +# event() - events handling method. +METHOD int event { + struct g_raid_md_object *md; + struct g_raid_disk *disk; + u_int event; +}; + +# volume_event() - events handling method. +METHOD int volume_event { + struct g_raid_md_object *md; + struct g_raid_volume *vol; + u_int event; +} DEFAULT g_raid_md_volume_event_default; + +# write() - metadata write method. +METHOD int write { + struct g_raid_md_object *md; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; +}; + +# fail_disk() - mark disk as failed and remove it from use. +METHOD int fail_disk { + struct g_raid_md_object *md; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; +}; + +# free_disk() - disk destructor. +METHOD int free_disk { + struct g_raid_md_object *md; + struct g_raid_disk *disk; +} DEFAULT g_raid_md_free_disk_default; + +# free_volume() - volume destructor. +METHOD int free_volume { + struct g_raid_md_object *md; + struct g_raid_volume *vol; +} DEFAULT g_raid_md_free_volume_default; + +# free() - destructor. +METHOD int free { + struct g_raid_md_object *md; +}; diff --git a/sys/geom/raid/g_raid_tr_if.m b/sys/geom/raid/g_raid_tr_if.m new file mode 100644 index 000000000000..193b4298691a --- /dev/null +++ b/sys/geom/raid/g_raid_tr_if.m @@ -0,0 +1,118 @@ +#- +# Copyright (c) 2010 Alexander Motin +# All rights reserved. +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions +# are met: +# +# 1. Redistributions of source code must retain the above copyright +# notice, this list of conditions and the following disclaimer. +# 2. Redistributions in binary form must reproduce the above copyright +# notice, this list of conditions and the following disclaimer in the +# documentation and/or other materials provided with the distribution. +# +# THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR +# IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES +# OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. +# IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, +# INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT +# NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF +# THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +# +# $FreeBSD$ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +# The G_RAID transformation class interface. + +INTERFACE g_raid_tr; + +# Default implementations of methods. +CODE { + static int + g_raid_tr_locked_default(struct g_raid_tr_object *tr, void *argp) + { + + return (0); + } +}; + +HEADER { +#define G_RAID_TR_TASTE_FAIL -1 +#define G_RAID_TR_TASTE_SUCCEED 0 +}; + +# taste() - volume taste method. +METHOD int taste { + struct g_raid_tr_object *tr; + struct g_raid_volume *volume; +}; + +# event() - events handling method. +METHOD int event { + struct g_raid_tr_object *tr; + struct g_raid_subdisk *sd; + u_int event; +}; + +# start() - begin operation. +METHOD int start { + struct g_raid_tr_object *tr; +}; + +# stop() - stop operation. +METHOD int stop { + struct g_raid_tr_object *tr; +}; + +# iorequest() - manage forward transformation and generates requests to disks. +METHOD void iostart { + struct g_raid_tr_object *tr; + struct bio *bp; +}; + +# iodone() - manages backward transformation and reports completion status. +METHOD void iodone { + struct g_raid_tr_object *tr; + struct g_raid_subdisk *sd; + struct bio *bp; +}; + +# kerneldump() - optimized for rebustness (simplified) kernel dumping routine. +METHOD int kerneldump { + struct g_raid_tr_object *tr; + void *virtual; + vm_offset_t physical; + off_t offset; + size_t length; +} DEFAULT g_raid_tr_kerneldump_common; + +# locked() - callback method for lock(). +METHOD int locked { + struct g_raid_tr_object *tr; + void *argp; +} DEFAULT g_raid_tr_locked_default; + +# free() - destructor. +METHOD int free { + struct g_raid_tr_object *tr; +}; + +# idle() - callback when the volume is idle for a while and the TR wants +# to schedule some work for that idle period. +METHOD int idle { + struct g_raid_tr_object *tr; +}; diff --git a/sys/geom/raid/md_intel.c b/sys/geom/raid/md_intel.c new file mode 100644 index 000000000000..32dc8f0eafd6 --- /dev/null +++ b/sys/geom/raid/md_intel.c @@ -0,0 +1,2323 @@ +/*- + * Copyright (c) 2010 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "geom/raid/g_raid.h" +#include "g_raid_md_if.h" + +static MALLOC_DEFINE(M_MD_INTEL, "md_intel_data", "GEOM_RAID Intel metadata"); + +struct intel_raid_map { + uint32_t offset; + uint32_t disk_sectors; + uint32_t stripe_count; + uint16_t strip_sectors; + uint8_t status; +#define INTEL_S_READY 0x00 +#define INTEL_S_UNINITIALIZED 0x01 +#define INTEL_S_DEGRADED 0x02 +#define INTEL_S_FAILURE 0x03 + + uint8_t type; +#define INTEL_T_RAID0 0x00 +#define INTEL_T_RAID1 0x01 +#define INTEL_T_RAID5 0x05 + + uint8_t total_disks; + uint8_t total_domains; + uint8_t failed_disk_num; + uint8_t ddf; + uint32_t filler_2[7]; + uint32_t disk_idx[1]; /* total_disks entries. */ +#define INTEL_DI_IDX 0x00ffffff +#define INTEL_DI_RBLD 0x01000000 +} __packed; + +struct intel_raid_vol { + uint8_t name[16]; + u_int64_t total_sectors __packed; + uint32_t state; +#define INTEL_ST_BOOTABLE 0x00000001 +#define INTEL_ST_BOOT_DEVICE 0x00000002 +#define INTEL_ST_READ_COALESCING 0x00000004 +#define INTEL_ST_WRITE_COALESCING 0x00000008 +#define INTEL_ST_LAST_SHUTDOWN_DIRTY 0x00000010 +#define INTEL_ST_HIDDEN_AT_BOOT 0x00000020 +#define INTEL_ST_CURRENTLY_HIDDEN 0x00000040 +#define INTEL_ST_VERIFY_AND_FIX 0x00000080 +#define INTEL_ST_MAP_STATE_UNINIT 0x00000100 +#define INTEL_ST_NO_AUTO_RECOVERY 0x00000200 +#define INTEL_ST_CLONE_N_GO 0x00000400 +#define INTEL_ST_CLONE_MAN_SYNC 0x00000800 +#define INTEL_ST_CNG_MASTER_DISK_NUM 0x00001000 + uint32_t reserved; + uint8_t migr_priority; + uint8_t num_sub_vols; + uint8_t tid; + uint8_t cng_master_disk; + uint16_t cache_policy; + uint8_t cng_state; + uint8_t cng_sub_state; + uint32_t filler_0[10]; + + uint32_t curr_migr_unit; + uint32_t checkpoint_id; + uint8_t migr_state; + uint8_t migr_type; +#define INTEL_MT_INIT 0 +#define INTEL_MT_REBUILD 1 +#define INTEL_MT_VERIFY 2 +#define INTEL_MT_GEN_MIGR 3 +#define INTEL_MT_STATE_CHANGE 4 +#define INTEL_MT_REPAIR 5 + uint8_t dirty; + uint8_t fs_state; + uint16_t verify_errors; + uint16_t bad_blocks; + uint32_t filler_1[4]; + struct intel_raid_map map[1]; /* 2 entries if migr_state != 0. */ +} __packed; + +struct intel_raid_disk { +#define INTEL_SERIAL_LEN 16 + uint8_t serial[INTEL_SERIAL_LEN]; + uint32_t sectors; + uint32_t id; + uint32_t flags; +#define INTEL_F_SPARE 0x01 +#define INTEL_F_ASSIGNED 0x02 +#define INTEL_F_FAILED 0x04 +#define INTEL_F_ONLINE 0x08 + + uint32_t filler[5]; +} __packed; + +struct intel_raid_conf { + uint8_t intel_id[24]; +#define INTEL_MAGIC "Intel Raid ISM Cfg Sig. " + + uint8_t version[6]; +#define INTEL_VERSION_1000 "1.0.00" /* RAID0 */ +#define INTEL_VERSION_1100 "1.1.00" /* RAID1 */ +#define INTEL_VERSION_1200 "1.2.00" /* Many volumes */ +#define INTEL_VERSION_1201 "1.2.01" /* 3 or 4 disks */ +#define INTEL_VERSION_1202 "1.2.02" /* RAID5 */ +#define INTEL_VERSION_1204 "1.2.04" /* 5 or 6 disks */ +#define INTEL_VERSION_1206 "1.2.06" /* CNG */ +#define INTEL_VERSION_1300 "1.3.00" /* Attributes */ + + uint8_t dummy_0[2]; + uint32_t checksum; + uint32_t config_size; + uint32_t config_id; + uint32_t generation; + uint32_t error_log_size; + uint32_t attributes; +#define INTEL_ATTR_RAID0 0x00000001 +#define INTEL_ATTR_RAID1 0x00000002 +#define INTEL_ATTR_RAID10 0x00000004 +#define INTEL_ATTR_RAID1E 0x00000008 +#define INTEL_ATTR_RAID5 0x00000010 +#define INTEL_ATTR_RAIDCNG 0x00000020 +#define INTEL_ATTR_2TB 0x20000000 +#define INTEL_ATTR_PM 0x40000000 +#define INTEL_ATTR_CHECKSUM 0x80000000 + + uint8_t total_disks; + uint8_t total_volumes; + uint8_t dummy_2[2]; + uint32_t filler_0[39]; + struct intel_raid_disk disk[1]; /* total_disks entries. */ + /* Here goes total_volumes of struct intel_raid_vol. */ +} __packed; + +#define INTEL_MAX_MD_SIZE(ndisks) \ + (sizeof(struct intel_raid_conf) + \ + sizeof(struct intel_raid_disk) * (ndisks - 1) + \ + sizeof(struct intel_raid_vol) * 2 + \ + sizeof(struct intel_raid_map) * 2 + \ + sizeof(uint32_t) * (ndisks - 1) * 4) + +struct g_raid_md_intel_perdisk { + struct intel_raid_conf *pd_meta; + int pd_disk_pos; + struct intel_raid_disk pd_disk_meta; +}; + +struct g_raid_md_intel_object { + struct g_raid_md_object mdio_base; + uint32_t mdio_config_id; + uint32_t mdio_generation; + struct intel_raid_conf *mdio_meta; + struct callout mdio_start_co; /* STARTING state timer. */ + int mdio_disks_present; + int mdio_started; + int mdio_incomplete; + struct root_hold_token *mdio_rootmount; /* Root mount delay token. */ +}; + +static g_raid_md_create_t g_raid_md_create_intel; +static g_raid_md_taste_t g_raid_md_taste_intel; +static g_raid_md_event_t g_raid_md_event_intel; +static g_raid_md_ctl_t g_raid_md_ctl_intel; +static g_raid_md_write_t g_raid_md_write_intel; +static g_raid_md_fail_disk_t g_raid_md_fail_disk_intel; +static g_raid_md_free_disk_t g_raid_md_free_disk_intel; +static g_raid_md_free_t g_raid_md_free_intel; + +static kobj_method_t g_raid_md_intel_methods[] = { + KOBJMETHOD(g_raid_md_create, g_raid_md_create_intel), + KOBJMETHOD(g_raid_md_taste, g_raid_md_taste_intel), + KOBJMETHOD(g_raid_md_event, g_raid_md_event_intel), + KOBJMETHOD(g_raid_md_ctl, g_raid_md_ctl_intel), + KOBJMETHOD(g_raid_md_write, g_raid_md_write_intel), + KOBJMETHOD(g_raid_md_fail_disk, g_raid_md_fail_disk_intel), + KOBJMETHOD(g_raid_md_free_disk, g_raid_md_free_disk_intel), + KOBJMETHOD(g_raid_md_free, g_raid_md_free_intel), + { 0, 0 } +}; + +static struct g_raid_md_class g_raid_md_intel_class = { + "Intel", + g_raid_md_intel_methods, + sizeof(struct g_raid_md_intel_object), + .mdc_priority = 100 +}; + + +static struct intel_raid_map * +intel_get_map(struct intel_raid_vol *mvol, int i) +{ + struct intel_raid_map *mmap; + + if (i > (mvol->migr_state ? 1 : 0)) + return (NULL); + mmap = &mvol->map[0]; + for (; i > 0; i--) { + mmap = (struct intel_raid_map *) + &mmap->disk_idx[mmap->total_disks]; + } + return ((struct intel_raid_map *)mmap); +} + +static struct intel_raid_vol * +intel_get_volume(struct intel_raid_conf *meta, int i) +{ + struct intel_raid_vol *mvol; + struct intel_raid_map *mmap; + + if (i > 1) + return (NULL); + mvol = (struct intel_raid_vol *)&meta->disk[meta->total_disks]; + for (; i > 0; i--) { + mmap = intel_get_map(mvol, mvol->migr_state ? 1 : 0); + mvol = (struct intel_raid_vol *) + &mmap->disk_idx[mmap->total_disks]; + } + return (mvol); +} + +static void +g_raid_md_intel_print(struct intel_raid_conf *meta) +{ + struct intel_raid_vol *mvol; + struct intel_raid_map *mmap; + int i, j, k; + + if (g_raid_debug < 1) + return; + + printf("********* ATA Intel MatrixRAID Metadata *********\n"); + printf("intel_id <%.24s>\n", meta->intel_id); + printf("version <%.6s>\n", meta->version); + printf("checksum 0x%08x\n", meta->checksum); + printf("config_size 0x%08x\n", meta->config_size); + printf("config_id 0x%08x\n", meta->config_id); + printf("generation 0x%08x\n", meta->generation); + printf("attributes 0x%08x\n", meta->attributes); + printf("total_disks %u\n", meta->total_disks); + printf("total_volumes %u\n", meta->total_volumes); + printf("DISK# serial disk_sectors disk_id flags\n"); + for (i = 0; i < meta->total_disks; i++ ) { + printf(" %d <%.16s> %u 0x%08x 0x%08x\n", i, + meta->disk[i].serial, meta->disk[i].sectors, + meta->disk[i].id, meta->disk[i].flags); + } + for (i = 0; i < meta->total_volumes; i++) { + mvol = intel_get_volume(meta, i); + printf(" ****** Volume %d ******\n", i); + printf(" name %.16s\n", mvol->name); + printf(" total_sectors %ju\n", mvol->total_sectors); + printf(" state %u\n", mvol->state); + printf(" reserved %u\n", mvol->reserved); + printf(" curr_migr_unit %u\n", mvol->curr_migr_unit); + printf(" checkpoint_id %u\n", mvol->checkpoint_id); + printf(" migr_state %u\n", mvol->migr_state); + printf(" migr_type %u\n", mvol->migr_type); + printf(" dirty %u\n", mvol->dirty); + + for (j = 0; j < (mvol->migr_state ? 2 : 1); j++) { + printf(" *** Map %d ***\n", j); + mmap = intel_get_map(mvol, j); + printf(" offset %u\n", mmap->offset); + printf(" disk_sectors %u\n", mmap->disk_sectors); + printf(" stripe_count %u\n", mmap->stripe_count); + printf(" strip_sectors %u\n", mmap->strip_sectors); + printf(" status %u\n", mmap->status); + printf(" type %u\n", mmap->type); + printf(" total_disks %u\n", mmap->total_disks); + printf(" total_domains %u\n", mmap->total_domains); + printf(" failed_disk_num %u\n", mmap->failed_disk_num); + printf(" ddf %u\n", mmap->ddf); + printf(" disk_idx "); + for (k = 0; k < mmap->total_disks; k++) + printf(" 0x%08x", mmap->disk_idx[k]); + printf("\n"); + } + } + printf("=================================================\n"); +} + +static struct intel_raid_conf * +intel_meta_copy(struct intel_raid_conf *meta) +{ + struct intel_raid_conf *nmeta; + + nmeta = malloc(meta->config_size, M_MD_INTEL, M_WAITOK); + memcpy(nmeta, meta, meta->config_size); + return (nmeta); +} + +static int +intel_meta_find_disk(struct intel_raid_conf *meta, char *serial) +{ + int pos; + + for (pos = 0; pos < meta->total_disks; pos++) { + if (strncmp(meta->disk[pos].serial, + serial, INTEL_SERIAL_LEN) == 0) + return (pos); + } + return (-1); +} + +static struct intel_raid_conf * +intel_meta_read(struct g_consumer *cp) +{ + struct g_provider *pp; + struct intel_raid_conf *meta; + struct intel_raid_vol *mvol; + struct intel_raid_map *mmap; + char *buf; + int error, i, j, k, left, size; + uint32_t checksum, *ptr; + + pp = cp->provider; + + /* Read the anchor sector. */ + buf = g_read_data(cp, + pp->mediasize - pp->sectorsize * 2, pp->sectorsize, &error); + if (buf == NULL) { + G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).", + pp->name, error); + return (NULL); + } + meta = (struct intel_raid_conf *)buf; + + /* Check if this is an Intel RAID struct */ + if (strncmp(meta->intel_id, INTEL_MAGIC, strlen(INTEL_MAGIC))) { + G_RAID_DEBUG(1, "Intel signature check failed on %s", pp->name); + g_free(buf); + return (NULL); + } + if (meta->config_size > 65536 || + meta->config_size < sizeof(struct intel_raid_conf)) { + G_RAID_DEBUG(1, "Intel metadata size looks wrong: %d", + meta->config_size); + g_free(buf); + return (NULL); + } + size = meta->config_size; + meta = malloc(size, M_MD_INTEL, M_WAITOK); + memcpy(meta, buf, min(size, pp->sectorsize)); + g_free(buf); + + /* Read all the rest, if needed. */ + if (meta->config_size > pp->sectorsize) { + left = (meta->config_size - 1) / pp->sectorsize; + buf = g_read_data(cp, + pp->mediasize - pp->sectorsize * (2 + left), + pp->sectorsize * left, &error); + if (buf == NULL) { + G_RAID_DEBUG(1, "Cannot read remaining metadata" + " part from %s (error=%d).", + pp->name, error); + free(meta, M_MD_INTEL); + return (NULL); + } + memcpy(((char *)meta) + pp->sectorsize, buf, + pp->sectorsize * left); + g_free(buf); + } + + /* Check metadata checksum. */ + for (checksum = 0, ptr = (uint32_t *)meta, i = 0; + i < (meta->config_size / sizeof(uint32_t)); i++) { + checksum += *ptr++; + } + checksum -= meta->checksum; + if (checksum != meta->checksum) { + G_RAID_DEBUG(1, "Intel checksum check failed on %s", pp->name); + free(meta, M_MD_INTEL); + return (NULL); + } + + /* Validate metadata size. */ + size = sizeof(struct intel_raid_conf) + + sizeof(struct intel_raid_disk) * (meta->total_disks - 1) + + sizeof(struct intel_raid_vol) * meta->total_volumes; + if (size > meta->config_size) { +badsize: + G_RAID_DEBUG(1, "Intel metadata size incorrect %d < %d", + meta->config_size, size); + free(meta, M_MD_INTEL); + return (NULL); + } + for (i = 0; i < meta->total_volumes; i++) { + mvol = intel_get_volume(meta, i); + mmap = intel_get_map(mvol, 0); + size += 4 * (mmap->total_disks - 1); + if (size > meta->config_size) + goto badsize; + if (mvol->migr_state) { + size += sizeof(struct intel_raid_map); + if (size > meta->config_size) + goto badsize; + mmap = intel_get_map(mvol, 1); + size += 4 * (mmap->total_disks - 1); + if (size > meta->config_size) + goto badsize; + } + } + + /* Validate disk indexes. */ + for (i = 0; i < meta->total_volumes; i++) { + mvol = intel_get_volume(meta, i); + for (j = 0; j < (mvol->migr_state ? 2 : 1); j++) { + mmap = intel_get_map(mvol, j); + for (k = 0; k < mmap->total_disks; k++) { + if ((mmap->disk_idx[k] & INTEL_DI_IDX) > + meta->total_disks) { + G_RAID_DEBUG(1, "Intel metadata disk" + " index %d too big (>%d)", + mmap->disk_idx[k] & INTEL_DI_IDX, + meta->total_disks); + free(meta, M_MD_INTEL); + return (NULL); + } + } + } + } + + /* Validate migration types. */ + for (i = 0; i < meta->total_volumes; i++) { + mvol = intel_get_volume(meta, i); + if (mvol->migr_state && + mvol->migr_type != INTEL_MT_INIT && + mvol->migr_type != INTEL_MT_REBUILD && + mvol->migr_type != INTEL_MT_VERIFY && + mvol->migr_type != INTEL_MT_REPAIR) { + G_RAID_DEBUG(1, "Intel metadata has unsupported" + " migration type %d", mvol->migr_type); + free(meta, M_MD_INTEL); + return (NULL); + } + } + + return (meta); +} + +static int +intel_meta_write(struct g_consumer *cp, struct intel_raid_conf *meta) +{ + struct g_provider *pp; + char *buf; + int error, i, sectors; + uint32_t checksum, *ptr; + + pp = cp->provider; + + /* Recalculate checksum for case if metadata were changed. */ + meta->checksum = 0; + for (checksum = 0, ptr = (uint32_t *)meta, i = 0; + i < (meta->config_size / sizeof(uint32_t)); i++) { + checksum += *ptr++; + } + meta->checksum = checksum; + + /* Create and fill buffer. */ + sectors = (meta->config_size + pp->sectorsize - 1) / pp->sectorsize; + buf = malloc(sectors * pp->sectorsize, M_MD_INTEL, M_WAITOK | M_ZERO); + if (sectors > 1) { + memcpy(buf, ((char *)meta) + pp->sectorsize, + (sectors - 1) * pp->sectorsize); + } + memcpy(buf + (sectors - 1) * pp->sectorsize, meta, pp->sectorsize); + + error = g_write_data(cp, + pp->mediasize - pp->sectorsize * (1 + sectors), + buf, pp->sectorsize * sectors); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).", + pp->name, error); + } + + free(buf, M_MD_INTEL); + return (error); +} + +static int +intel_meta_erase(struct g_consumer *cp) +{ + struct g_provider *pp; + char *buf; + int error; + + pp = cp->provider; + buf = malloc(pp->sectorsize, M_MD_INTEL, M_WAITOK | M_ZERO); + error = g_write_data(cp, + pp->mediasize - 2 * pp->sectorsize, + buf, pp->sectorsize); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).", + pp->name, error); + } + free(buf, M_MD_INTEL); + return (error); +} + +static int +intel_meta_write_spare(struct g_consumer *cp, struct intel_raid_disk *d) +{ + struct intel_raid_conf *meta; + int error; + + /* Fill anchor and single disk. */ + meta = malloc(INTEL_MAX_MD_SIZE(1), M_MD_INTEL, M_WAITOK | M_ZERO); + memcpy(&meta->intel_id[0], INTEL_MAGIC, sizeof(INTEL_MAGIC)); + memcpy(&meta->version[0], INTEL_VERSION_1000, + sizeof(INTEL_VERSION_1000)); + meta->config_size = INTEL_MAX_MD_SIZE(1); + meta->config_id = arc4random(); + meta->generation = 1; + meta->total_disks = 1; + meta->disk[0] = *d; + error = intel_meta_write(cp, meta); + free(meta, M_MD_INTEL); + return (error); +} + +static struct g_raid_disk * +g_raid_md_intel_get_disk(struct g_raid_softc *sc, int id) +{ + struct g_raid_disk *disk; + struct g_raid_md_intel_perdisk *pd; + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; + if (pd->pd_disk_pos == id) + break; + } + return (disk); +} + +static int +g_raid_md_intel_supported(int level, int qual, int disks, int force) +{ + + switch (level) { + case G_RAID_VOLUME_RL_RAID0: + if (disks < 1) + return (0); + if (!force && (disks < 2 || disks > 6)) + return (0); + break; + case G_RAID_VOLUME_RL_RAID1: + if (disks < 1) + return (0); + if (!force && (disks != 2)) + return (0); + break; + case G_RAID_VOLUME_RL_RAID1E: + if (disks < 2) + return (0); + if (!force && (disks != 4)) + return (0); + break; + case G_RAID_VOLUME_RL_RAID5: + if (disks < 3) + return (0); + if (!force && disks > 6) + return (0); + break; + default: + return (0); + } + if (qual != G_RAID_VOLUME_RLQ_NONE) + return (0); + return (1); +} + +static struct g_raid_volume * +g_raid_md_intel_get_volume(struct g_raid_softc *sc, int id) +{ + struct g_raid_volume *mvol; + + TAILQ_FOREACH(mvol, &sc->sc_volumes, v_next) { + if ((intptr_t)(mvol->v_md_data) == id) + break; + } + return (mvol); +} + +static int +g_raid_md_intel_start_disk(struct g_raid_disk *disk) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd, *tmpsd; + struct g_raid_disk *olddisk, *tmpdisk; + struct g_raid_md_object *md; + struct g_raid_md_intel_object *mdi; + struct g_raid_md_intel_perdisk *pd, *oldpd; + struct intel_raid_conf *meta; + struct intel_raid_vol *mvol; + struct intel_raid_map *mmap0, *mmap1; + int disk_pos, resurrection = 0; + + sc = disk->d_softc; + md = sc->sc_md; + mdi = (struct g_raid_md_intel_object *)md; + meta = mdi->mdio_meta; + pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; + olddisk = NULL; + + /* Find disk position in metadata by it's serial. */ + disk_pos = intel_meta_find_disk(meta, pd->pd_disk_meta.serial); + if (disk_pos < 0) { + G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk"); + /* Failed stale disk is useless for us. */ + if (pd->pd_disk_meta.flags & INTEL_F_FAILED) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED); + return (0); + } + /* If we are in the start process, that's all for now. */ + if (!mdi->mdio_started) + goto nofit; + /* + * If we have already started - try to get use of the disk. + * Try to replace OFFLINE disks first, then FAILED. + */ + TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) { + if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE && + tmpdisk->d_state != G_RAID_DISK_S_FAILED) + continue; + /* Make sure this disk is big enough. */ + TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) { + if (sd->sd_offset + sd->sd_size + 4096 > + (off_t)pd->pd_disk_meta.sectors * 512) { + G_RAID_DEBUG1(1, sc, + "Disk too small (%llu < %llu)", + ((unsigned long long) + pd->pd_disk_meta.sectors) * 512, + (unsigned long long) + sd->sd_offset + sd->sd_size + 4096); + break; + } + } + if (sd != NULL) + continue; + if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) { + olddisk = tmpdisk; + break; + } else if (olddisk == NULL) + olddisk = tmpdisk; + } + if (olddisk == NULL) { +nofit: + if (pd->pd_disk_meta.flags & INTEL_F_SPARE) { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_SPARE); + return (1); + } else { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_STALE); + return (0); + } + } + oldpd = (struct g_raid_md_intel_perdisk *)olddisk->d_md_data; + disk_pos = oldpd->pd_disk_pos; + resurrection = 1; + } + + if (olddisk == NULL) { + /* Find placeholder by position. */ + olddisk = g_raid_md_intel_get_disk(sc, disk_pos); + if (olddisk == NULL) + panic("No disk at position %d!", disk_pos); + if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) { + G_RAID_DEBUG1(1, sc, "More then one disk for pos %d", + disk_pos); + g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE); + return (0); + } + oldpd = (struct g_raid_md_intel_perdisk *)olddisk->d_md_data; + } + + /* Replace failed disk or placeholder with new disk. */ + TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) { + TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next); + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + sd->sd_disk = disk; + } + oldpd->pd_disk_pos = -2; + pd->pd_disk_pos = disk_pos; + + /* If it was placeholder -- destroy it. */ + if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) { + g_raid_destroy_disk(olddisk); + } else { + /* Otherwise, make it STALE_FAILED. */ + g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED); + /* Update global metadata just in case. */ + memcpy(&meta->disk[disk_pos], &pd->pd_disk_meta, + sizeof(struct intel_raid_disk)); + } + + /* Welcome the new disk. */ + if (resurrection) + g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); + else if (meta->disk[disk_pos].flags & INTEL_F_FAILED) + g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED); + else if (meta->disk[disk_pos].flags & INTEL_F_SPARE) + g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE); + else + g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + mvol = intel_get_volume(meta, + (uintptr_t)(sd->sd_volume->v_md_data)); + mmap0 = intel_get_map(mvol, 0); + if (mvol->migr_state) + mmap1 = intel_get_map(mvol, 1); + else + mmap1 = mmap0; + + if (resurrection) { + /* Stale disk, almost same as new. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NEW); + } else if (meta->disk[disk_pos].flags & INTEL_F_FAILED) { + /* Failed disk, almost useless. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_FAILED); + } else if (mvol->migr_state == 0) { + if (mmap0->status == INTEL_S_UNINITIALIZED) { + /* Freshly created uninitialized volume. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_UNINITIALIZED); + } else if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { + /* Freshly inserted disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NEW); + } else if (mvol->dirty) { + /* Dirty volume (unclean shutdown). */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_STALE); + } else { + /* Up to date disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + } + } else if (mvol->migr_type == INTEL_MT_INIT || + mvol->migr_type == INTEL_MT_REBUILD) { + if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { + /* Freshly inserted disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NEW); + } else if (mmap1->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { + /* Rebuilding disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_REBUILD); + if (mvol->dirty) { + sd->sd_rebuild_pos = 0; + } else { + sd->sd_rebuild_pos = + (off_t)mvol->curr_migr_unit * + sd->sd_volume->v_strip_size * + mmap0->total_domains; + } + } else if (mvol->dirty) { + /* Dirty volume (unclean shutdown). */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_STALE); + } else { + /* Up to date disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + } + } else if (mvol->migr_type == INTEL_MT_VERIFY || + mvol->migr_type == INTEL_MT_REPAIR) { + if (mmap0->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { + /* Freshly inserted disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NEW); + } else if (mmap1->disk_idx[sd->sd_pos] & INTEL_DI_RBLD) { + /* Resyncing disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_RESYNC); + if (mvol->dirty) { + sd->sd_rebuild_pos = 0; + } else { + sd->sd_rebuild_pos = + (off_t)mvol->curr_migr_unit * + sd->sd_volume->v_strip_size * + mmap0->total_domains; + } + } else if (mvol->dirty) { + /* Dirty volume (unclean shutdown). */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_STALE); + } else { + /* Up to date disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + } + } + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + } + + /* Update status of our need for spare. */ + if (mdi->mdio_started) { + mdi->mdio_incomplete = + (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < + meta->total_disks); + } + + return (resurrection); +} + +static void +g_disk_md_intel_retaste(void *arg, int pending) +{ + + G_RAID_DEBUG(1, "Array is not complete, trying to retaste."); + g_retaste(&g_raid_class); + free(arg, M_MD_INTEL); +} + +static void +g_raid_md_intel_refill(struct g_raid_softc *sc) +{ + struct g_raid_md_object *md; + struct g_raid_md_intel_object *mdi; + struct intel_raid_conf *meta; + struct g_raid_disk *disk; + struct task *task; + int update, na; + + md = sc->sc_md; + mdi = (struct g_raid_md_intel_object *)md; + meta = mdi->mdio_meta; + update = 0; + do { + /* Make sure we miss anything. */ + na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE); + if (na == meta->total_disks) + break; + + G_RAID_DEBUG1(1, md->mdo_softc, + "Array is not complete (%d of %d), " + "trying to refill.", na, meta->total_disks); + + /* Try to get use some of STALE disks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_STALE) { + update += g_raid_md_intel_start_disk(disk); + if (disk->d_state == G_RAID_DISK_S_ACTIVE) + break; + } + } + if (disk != NULL) + continue; + + /* Try to get use some of SPARE disks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_SPARE) { + update += g_raid_md_intel_start_disk(disk); + if (disk->d_state == G_RAID_DISK_S_ACTIVE) + break; + } + } + } while (disk != NULL); + + /* Write new metadata if we changed something. */ + if (update) { + g_raid_md_write_intel(md, NULL, NULL, NULL); + meta = mdi->mdio_meta; + } + + /* Update status of our need for spare. */ + mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < + meta->total_disks); + + /* Request retaste hoping to find spare. */ + if (mdi->mdio_incomplete) { + task = malloc(sizeof(struct task), + M_MD_INTEL, M_WAITOK | M_ZERO); + TASK_INIT(task, 0, g_disk_md_intel_retaste, task); + taskqueue_enqueue(taskqueue_swi, task); + } +} + +static void +g_raid_md_intel_start(struct g_raid_softc *sc) +{ + struct g_raid_md_object *md; + struct g_raid_md_intel_object *mdi; + struct g_raid_md_intel_perdisk *pd; + struct intel_raid_conf *meta; + struct intel_raid_vol *mvol; + struct intel_raid_map *mmap; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + int i, j, disk_pos; + + md = sc->sc_md; + mdi = (struct g_raid_md_intel_object *)md; + meta = mdi->mdio_meta; + + /* Create volumes and subdisks. */ + for (i = 0; i < meta->total_volumes; i++) { + mvol = intel_get_volume(meta, i); + mmap = intel_get_map(mvol, 0); + vol = g_raid_create_volume(sc, mvol->name, -1); + vol->v_md_data = (void *)(intptr_t)i; + if (mmap->type == INTEL_T_RAID0) + vol->v_raid_level = G_RAID_VOLUME_RL_RAID0; + else if (mmap->type == INTEL_T_RAID1 && + mmap->total_domains >= 2 && + mmap->total_domains <= mmap->total_disks) { + /* Assume total_domains is correct. */ + if (mmap->total_domains == mmap->total_disks) + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1; + else + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E; + } else if (mmap->type == INTEL_T_RAID1) { + /* total_domains looks wrong. */ + if (mmap->total_disks <= 2) + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1; + else + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E; + } else if (mmap->type == INTEL_T_RAID5) + vol->v_raid_level = G_RAID_VOLUME_RL_RAID5; + else + vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + vol->v_strip_size = (u_int)mmap->strip_sectors * 512; //ZZZ + vol->v_disks_count = mmap->total_disks; + vol->v_mediasize = (off_t)mvol->total_sectors * 512; //ZZZ + vol->v_sectorsize = 512; //ZZZ + for (j = 0; j < vol->v_disks_count; j++) { + sd = &vol->v_subdisks[j]; + sd->sd_offset = (off_t)mmap->offset * 512; //ZZZ + sd->sd_size = (off_t)mmap->disk_sectors * 512; //ZZZ + } + g_raid_start_volume(vol); + } + + /* Create disk placeholders to store data for later writing. */ + for (disk_pos = 0; disk_pos < meta->total_disks; disk_pos++) { + pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); + pd->pd_disk_pos = disk_pos; + pd->pd_disk_meta = meta->disk[disk_pos]; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_state = G_RAID_DISK_S_OFFLINE; + for (i = 0; i < meta->total_volumes; i++) { + mvol = intel_get_volume(meta, i); + mmap = intel_get_map(mvol, 0); + for (j = 0; j < mmap->total_disks; j++) { + if ((mmap->disk_idx[j] & INTEL_DI_IDX) == disk_pos) + break; + } + if (j == mmap->total_disks) + continue; + vol = g_raid_md_intel_get_volume(sc, i); + sd = &vol->v_subdisks[j]; + sd->sd_disk = disk; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + } + } + + /* Make all disks found till the moment take their places. */ + do { + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_NONE) { + g_raid_md_intel_start_disk(disk); + break; + } + } + } while (disk != NULL); + + mdi->mdio_started = 1; + G_RAID_DEBUG1(0, sc, "Array started."); + g_raid_md_write_intel(md, NULL, NULL, NULL); + + /* Pickup any STALE/SPARE disks to refill array if needed. */ + g_raid_md_intel_refill(sc); + + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + g_raid_event_send(vol, G_RAID_VOLUME_E_START, + G_RAID_EVENT_VOLUME); + } + + callout_stop(&mdi->mdio_start_co); + G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount); + root_mount_rel(mdi->mdio_rootmount); + mdi->mdio_rootmount = NULL; +} + +static void +g_raid_md_intel_new_disk(struct g_raid_disk *disk) +{ + struct g_raid_softc *sc; + struct g_raid_md_object *md; + struct g_raid_md_intel_object *mdi; + struct intel_raid_conf *pdmeta; + struct g_raid_md_intel_perdisk *pd; + + sc = disk->d_softc; + md = sc->sc_md; + mdi = (struct g_raid_md_intel_object *)md; + pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; + pdmeta = pd->pd_meta; + + if (mdi->mdio_started) { + if (g_raid_md_intel_start_disk(disk)) + g_raid_md_write_intel(md, NULL, NULL, NULL); + } else { + /* If we haven't started yet - check metadata freshness. */ + if (mdi->mdio_meta == NULL || + ((int32_t)(pdmeta->generation - mdi->mdio_generation)) > 0) { + G_RAID_DEBUG1(1, sc, "Newer disk"); + if (mdi->mdio_meta != NULL) + free(mdi->mdio_meta, M_MD_INTEL); + mdi->mdio_meta = intel_meta_copy(pdmeta); + mdi->mdio_generation = mdi->mdio_meta->generation; + mdi->mdio_disks_present = 1; + } else if (pdmeta->generation == mdi->mdio_generation) { + mdi->mdio_disks_present++; + G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)", + mdi->mdio_disks_present, + mdi->mdio_meta->total_disks); + } else { + G_RAID_DEBUG1(1, sc, "Older disk"); + } + /* If we collected all needed disks - start array. */ + if (mdi->mdio_disks_present == mdi->mdio_meta->total_disks) + g_raid_md_intel_start(sc); + } +} + +static void +g_raid_intel_go(void *arg) +{ + struct g_raid_softc *sc; + struct g_raid_md_object *md; + struct g_raid_md_intel_object *mdi; + + sc = arg; + md = sc->sc_md; + mdi = (struct g_raid_md_intel_object *)md; + if (!mdi->mdio_started) { + G_RAID_DEBUG1(0, sc, "Force array start due to timeout."); + g_raid_event_send(sc, G_RAID_NODE_E_START, 0); + } +} + +static int +g_raid_md_create_intel(struct g_raid_md_object *md, struct g_class *mp, + struct g_geom **gp) +{ + struct g_raid_softc *sc; + struct g_raid_md_intel_object *mdi; + char name[16]; + + mdi = (struct g_raid_md_intel_object *)md; + mdi->mdio_config_id = arc4random(); + mdi->mdio_generation = 0; + snprintf(name, sizeof(name), "Intel-%08x", mdi->mdio_config_id); + sc = g_raid_create_node(mp, name, md); + if (sc == NULL) + return (G_RAID_MD_TASTE_FAIL); + md->mdo_softc = sc; + *gp = sc->sc_geom; + return (G_RAID_MD_TASTE_NEW); +} + +/* + * Return the last N characters of the serial label. The Linux and + * ataraid(7) code always uses the last 16 characters of the label to + * store into the Intel meta format. Generalize this to N characters + * since that's easy. Labels can be up to 20 characters for SATA drives + * and up 251 characters for SAS drives. Since intel controllers don't + * support SAS drives, just stick with the SATA limits for stack friendliness. + */ +static int +g_raid_md_get_label(struct g_consumer *cp, char *serial, int serlen) +{ + char serial_buffer[24]; + int len, error; + + len = sizeof(serial_buffer); + error = g_io_getattr("GEOM::ident", cp, &len, serial_buffer); + if (error != 0) + return (error); + len = strlen(serial_buffer); + if (len > serlen) + len -= serlen; + else + len = 0; + strncpy(serial, serial_buffer + len, serlen); + return (0); +} + +static int +g_raid_md_taste_intel(struct g_raid_md_object *md, struct g_class *mp, + struct g_consumer *cp, struct g_geom **gp) +{ + struct g_consumer *rcp; + struct g_provider *pp; + struct g_raid_md_intel_object *mdi, *mdi1; + struct g_raid_softc *sc; + struct g_raid_disk *disk; + struct intel_raid_conf *meta; + struct g_raid_md_intel_perdisk *pd; + struct g_geom *geom; + int error, disk_pos, result, spare, len; + char serial[INTEL_SERIAL_LEN]; + char name[16]; + uint16_t vendor; + + G_RAID_DEBUG(1, "Tasting Intel on %s", cp->provider->name); + mdi = (struct g_raid_md_intel_object *)md; + pp = cp->provider; + + /* Read metadata from device. */ + meta = NULL; + spare = 0; + vendor = 0xffff; + disk_pos = 0; + if (g_access(cp, 1, 0, 0) != 0) + return (G_RAID_MD_TASTE_FAIL); + g_topology_unlock(); + error = g_raid_md_get_label(cp, serial, sizeof(serial)); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot get serial number from %s (error=%d).", + pp->name, error); + goto fail2; + } + len = 2; + if (pp->geom->rank == 1) + g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor); + meta = intel_meta_read(cp); + g_topology_lock(); + g_access(cp, -1, 0, 0); + if (meta == NULL) { + if (g_raid_aggressive_spare) { + if (vendor == 0x8086) { + G_RAID_DEBUG(1, + "No Intel metadata, forcing spare."); + spare = 2; + goto search; + } else { + G_RAID_DEBUG(1, + "Intel vendor mismatch 0x%04x != 0x8086", + vendor); + } + } + return (G_RAID_MD_TASTE_FAIL); + } + + /* Check this disk position in obtained metadata. */ + disk_pos = intel_meta_find_disk(meta, serial); + if (disk_pos < 0) { + G_RAID_DEBUG(1, "Intel serial '%s' not found", serial); + goto fail1; + } + if (meta->disk[disk_pos].sectors != + (pp->mediasize / pp->sectorsize)) { + G_RAID_DEBUG(1, "Intel size mismatch %u != %u", + meta->disk[disk_pos].sectors, + (u_int)(pp->mediasize / pp->sectorsize)); + goto fail1; + } + + /* Metadata valid. Print it. */ + g_raid_md_intel_print(meta); + G_RAID_DEBUG(1, "Intel disk position %d", disk_pos); + spare = meta->disk[disk_pos].flags & INTEL_F_SPARE; + +search: + /* Search for matching node. */ + sc = NULL; + mdi1 = NULL; + LIST_FOREACH(geom, &mp->geom, geom) { + sc = geom->softc; + if (sc == NULL) + continue; + if (sc->sc_stopping != 0) + continue; + if (sc->sc_md->mdo_class != md->mdo_class) + continue; + mdi1 = (struct g_raid_md_intel_object *)sc->sc_md; + if (spare) { + if (mdi1->mdio_incomplete) + break; + } else { + if (mdi1->mdio_config_id == meta->config_id) + break; + } + } + + /* Found matching node. */ + if (geom != NULL) { + G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name); + result = G_RAID_MD_TASTE_EXISTING; + + } else if (spare) { /* Not found needy node -- left for later. */ + G_RAID_DEBUG(1, "Spare is not needed at this time"); + goto fail1; + + } else { /* Not found matching node -- create one. */ + result = G_RAID_MD_TASTE_NEW; + mdi->mdio_config_id = meta->config_id; + snprintf(name, sizeof(name), "Intel-%08x", meta->config_id); + sc = g_raid_create_node(mp, name, md); + md->mdo_softc = sc; + geom = sc->sc_geom; + callout_init(&mdi->mdio_start_co, 1); + callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz, + g_raid_intel_go, sc); + mdi->mdio_rootmount = root_mount_hold("GRAID-Intel"); + G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount); + } + + rcp = g_new_consumer(geom); + g_attach(rcp, pp); + if (g_access(rcp, 1, 1, 1) != 0) + ; //goto fail1; + + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + + pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); + pd->pd_meta = meta; + pd->pd_disk_pos = -1; + if (spare == 2) { + memcpy(&pd->pd_disk_meta.serial[0], serial, INTEL_SERIAL_LEN); + pd->pd_disk_meta.sectors = pp->mediasize / pp->sectorsize; + pd->pd_disk_meta.id = 0; + pd->pd_disk_meta.flags = INTEL_F_SPARE; + } else { + pd->pd_disk_meta = meta->disk[disk_pos]; + } + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_consumer = rcp; + rcp->private = disk; + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.", + rcp->provider->name, error); + + g_raid_md_intel_new_disk(disk); + + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + *gp = geom; + return (result); +fail2: + g_topology_lock(); + g_access(cp, -1, 0, 0); +fail1: + free(meta, M_MD_INTEL); + return (G_RAID_MD_TASTE_FAIL); +} + +static int +g_raid_md_event_intel(struct g_raid_md_object *md, + struct g_raid_disk *disk, u_int event) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd; + struct g_raid_md_intel_object *mdi; + struct g_raid_md_intel_perdisk *pd; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_intel_object *)md; + if (disk == NULL) { + switch (event) { + case G_RAID_NODE_E_START: + if (!mdi->mdio_started) + g_raid_md_intel_start(sc); + return (0); + } + return (-1); + } + pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; + switch (event) { + case G_RAID_DISK_E_DISCONNECTED: + /* If disk was assigned, just update statuses. */ + if (pd->pd_disk_pos >= 0) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + if (disk->d_consumer) { + g_raid_kill_consumer(sc, disk->d_consumer); + disk->d_consumer = NULL; + } + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NONE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, + G_RAID_EVENT_SUBDISK); + } + } else { + /* Otherwise -- delete. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); + g_raid_destroy_disk(disk); + } + + /* Write updated metadata to all disks. */ + g_raid_md_write_intel(md, NULL, NULL, NULL); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_intel_refill(sc); + return (0); + } + return (-2); +} + +static int +g_raid_md_ctl_intel(struct g_raid_md_object *md, + struct gctl_req *req) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol, *vol1; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_intel_object *mdi; + struct g_raid_md_intel_perdisk *pd; + struct g_consumer *cp; + struct g_provider *pp; + char arg[16], serial[INTEL_SERIAL_LEN]; + const char *verb, *volname, *levelname, *diskname; + char *tmp; + int *nargs, *force; + off_t off, size, sectorsize, strip; + intmax_t *sizearg, *striparg; + int numdisks, i, len, level, qual, update; + int error; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_intel_object *)md; + verb = gctl_get_param(req, "verb", NULL); + nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); + error = 0; + if (strcmp(verb, "label") == 0) { + + if (*nargs < 4) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + volname = gctl_get_asciiparam(req, "arg1"); + if (volname == NULL) { + gctl_error(req, "No volume name."); + return (-2); + } + levelname = gctl_get_asciiparam(req, "arg2"); + if (levelname == NULL) { + gctl_error(req, "No RAID level."); + return (-3); + } + if (g_raid_volume_str2level(levelname, &level, &qual)) { + gctl_error(req, "Unknown RAID level '%s'.", levelname); + return (-4); + } + numdisks = *nargs - 3; + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (!g_raid_md_intel_supported(level, qual, numdisks, + force ? *force : 0)) { + gctl_error(req, "Unsupported RAID level " + "(0x%02x/0x%02x), or number of disks (%d).", + level, qual, numdisks); + return (-5); + } + + /* Search for disks, connect them and probe. */ + size = 0x7fffffffffffffffllu; + sectorsize = 0; + for (i = 0; i < numdisks; i++) { + snprintf(arg, sizeof(arg), "arg%d", i + 3); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -6; + break; + } + if (strcmp(diskname, "NONE") == 0) { + cp = NULL; + pp = NULL; + } else { + g_topology_lock(); + cp = g_raid_open_consumer(sc, diskname); + if (cp == NULL) { + gctl_error(req, "Can't open disk '%s'.", + diskname); + g_topology_unlock(); + error = -4; + break; + } + pp = cp->provider; + } + pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); + pd->pd_disk_pos = i; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_consumer = cp; + if (cp == NULL) { + strcpy(&pd->pd_disk_meta.serial[0], "NONE"); + pd->pd_disk_meta.id = 0xffffffff; + pd->pd_disk_meta.flags = INTEL_F_ASSIGNED; + continue; + } + cp->private = disk; + g_topology_unlock(); + + error = g_raid_md_get_label(cp, + &pd->pd_disk_meta.serial[0], INTEL_SERIAL_LEN); + if (error != 0) { + gctl_error(req, + "Can't get serial for provider '%s'.", + diskname); + error = -8; + break; + } + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, + "Dumping not supported by %s.", + cp->provider->name); + + pd->pd_disk_meta.sectors = pp->mediasize / pp->sectorsize; + if (size > pp->mediasize) + size = pp->mediasize; + if (sectorsize < pp->sectorsize) + sectorsize = pp->sectorsize; + pd->pd_disk_meta.id = 0; + pd->pd_disk_meta.flags = INTEL_F_ASSIGNED | INTEL_F_ONLINE; + } + if (error != 0) + return (error); + + /* Reserve some space for metadata. */ + size -= ((4096 + sectorsize - 1) / sectorsize) * sectorsize; + + /* Handle size argument. */ + len = sizeof(*sizearg); + sizearg = gctl_get_param(req, "size", &len); + if (sizearg != NULL && len == sizeof(*sizearg) && + *sizearg > 0) { + if (*sizearg > size) { + gctl_error(req, "Size too big %lld > %lld.", + (long long)*sizearg, (long long)size); + return (-9); + } + size = *sizearg; + } + + /* Handle strip argument. */ + strip = 131072; + len = sizeof(*striparg); + striparg = gctl_get_param(req, "strip", &len); + if (striparg != NULL && len == sizeof(*striparg) && + *striparg > 0) { + if (*striparg < sectorsize) { + gctl_error(req, "Strip size too small."); + return (-10); + } + if (*striparg % sectorsize != 0) { + gctl_error(req, "Incorrect strip size."); + return (-11); + } + if (strip > 65535 * sectorsize) { + gctl_error(req, "Strip size too big."); + return (-12); + } + strip = *striparg; + } + + /* Round size down to strip or sector. */ + if (level == G_RAID_VOLUME_RL_RAID1) + size -= (size % sectorsize); + else if (level == G_RAID_VOLUME_RL_RAID1E && + (numdisks & 1) != 0) + size -= (size % (2 * strip)); + else + size -= (size % strip); + if (size <= 0) { + gctl_error(req, "Size too small."); + return (-13); + } + if (size > 0xffffffffllu * sectorsize) { + gctl_error(req, "Size too big."); + return (-14); + } + + /* We have all we need, create things: volume, ... */ + mdi->mdio_started = 1; + vol = g_raid_create_volume(sc, volname, -1); + vol->v_md_data = (void *)(intptr_t)0; + vol->v_raid_level = level; + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + vol->v_strip_size = strip; + vol->v_disks_count = numdisks; + if (level == G_RAID_VOLUME_RL_RAID0) + vol->v_mediasize = size * numdisks; + else if (level == G_RAID_VOLUME_RL_RAID1) + vol->v_mediasize = size; + else if (level == G_RAID_VOLUME_RL_RAID5) + vol->v_mediasize = size * (numdisks - 1); + else { /* RAID1E */ + vol->v_mediasize = ((size * numdisks) / strip / 2) * + strip; + } + vol->v_sectorsize = sectorsize; + g_raid_start_volume(vol); + + /* , and subdisks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; + sd = &vol->v_subdisks[pd->pd_disk_pos]; + sd->sd_disk = disk; + sd->sd_offset = 0; + sd->sd_size = size; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + if (sd->sd_disk->d_consumer != NULL) { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_ACTIVE); + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + } else { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + } + } + + /* Write metadata based on created entities. */ + G_RAID_DEBUG1(0, sc, "Array started."); + g_raid_md_write_intel(md, NULL, NULL, NULL); + + /* Pickup any STALE/SPARE disks to refill array if needed. */ + g_raid_md_intel_refill(sc); + + g_raid_event_send(vol, G_RAID_VOLUME_E_START, + G_RAID_EVENT_VOLUME); + return (0); + } + if (strcmp(verb, "add") == 0) { + + if (*nargs != 3) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + volname = gctl_get_asciiparam(req, "arg1"); + if (volname == NULL) { + gctl_error(req, "No volume name."); + return (-2); + } + levelname = gctl_get_asciiparam(req, "arg2"); + if (levelname == NULL) { + gctl_error(req, "No RAID level."); + return (-3); + } + if (g_raid_volume_str2level(levelname, &level, &qual)) { + gctl_error(req, "Unknown RAID level '%s'.", levelname); + return (-4); + } + + /* Look for existing volumes. */ + i = 0; + vol1 = NULL; + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + vol1 = vol; + i++; + } + if (i > 1) { + gctl_error(req, "Maximum two volumes supported."); + return (-6); + } + if (vol1 == NULL) { + gctl_error(req, "At least one volume must exist."); + return (-7); + } + + numdisks = vol1->v_disks_count; + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (!g_raid_md_intel_supported(level, qual, numdisks, + force ? *force : 0)) { + gctl_error(req, "Unsupported RAID level " + "(0x%02x/0x%02x), or number of disks (%d).", + level, qual, numdisks); + return (-5); + } + + /* Collect info about present disks. */ + size = 0x7fffffffffffffffllu; + sectorsize = 512; + for (i = 0; i < numdisks; i++) { + disk = vol1->v_subdisks[i].sd_disk; + pd = (struct g_raid_md_intel_perdisk *) + disk->d_md_data; + if ((off_t)pd->pd_disk_meta.sectors * 512 < size) + size = (off_t)pd->pd_disk_meta.sectors * 512; + if (disk->d_consumer != NULL && + disk->d_consumer->provider != NULL && + disk->d_consumer->provider->sectorsize > + sectorsize) { + sectorsize = + disk->d_consumer->provider->sectorsize; + } + } + + /* Reserve some space for metadata. */ + size -= ((4096 + sectorsize - 1) / sectorsize) * sectorsize; + + /* Decide insert before or after. */ + sd = &vol1->v_subdisks[0]; + if (sd->sd_offset > + size - (sd->sd_offset + sd->sd_size)) { + off = 0; + size = sd->sd_offset; + } else { + off = sd->sd_offset + sd->sd_size; + size = size - (sd->sd_offset + sd->sd_size); + } + + /* Handle strip argument. */ + strip = 131072; + len = sizeof(*striparg); + striparg = gctl_get_param(req, "strip", &len); + if (striparg != NULL && len == sizeof(*striparg) && + *striparg > 0) { + if (*striparg < sectorsize) { + gctl_error(req, "Strip size too small."); + return (-10); + } + if (*striparg % sectorsize != 0) { + gctl_error(req, "Incorrect strip size."); + return (-11); + } + if (strip > 65535 * sectorsize) { + gctl_error(req, "Strip size too big."); + return (-12); + } + strip = *striparg; + } + + /* Round offset up to strip. */ + if (off % strip != 0) { + size -= strip - off % strip; + off += strip - off % strip; + } + + /* Handle size argument. */ + len = sizeof(*sizearg); + sizearg = gctl_get_param(req, "size", &len); + if (sizearg != NULL && len == sizeof(*sizearg) && + *sizearg > 0) { + if (*sizearg > size) { + gctl_error(req, "Size too big %lld > %lld.", + (long long)*sizearg, (long long)size); + return (-9); + } + size = *sizearg; + } + + /* Round size down to strip or sector. */ + if (level == G_RAID_VOLUME_RL_RAID1) + size -= (size % sectorsize); + else + size -= (size % strip); + if (size <= 0) { + gctl_error(req, "Size too small."); + return (-13); + } + if (size > 0xffffffffllu * sectorsize) { + gctl_error(req, "Size too big."); + return (-14); + } + + /* We have all we need, create things: volume, ... */ + vol = g_raid_create_volume(sc, volname, -1); + vol->v_md_data = (void *)(intptr_t)i; + vol->v_raid_level = level; + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + vol->v_strip_size = strip; + vol->v_disks_count = numdisks; + if (level == G_RAID_VOLUME_RL_RAID0) + vol->v_mediasize = size * numdisks; + else if (level == G_RAID_VOLUME_RL_RAID1) + vol->v_mediasize = size; + else if (level == G_RAID_VOLUME_RL_RAID5) + vol->v_mediasize = size * (numdisks - 1); + else { /* RAID1E */ + vol->v_mediasize = ((size * numdisks) / strip / 2) * + strip; + } + vol->v_sectorsize = sectorsize; + g_raid_start_volume(vol); + + /* , and subdisks. */ + for (i = 0; i < numdisks; i++) { + disk = vol1->v_subdisks[i].sd_disk; + sd = &vol->v_subdisks[i]; + sd->sd_disk = disk; + sd->sd_offset = off; + sd->sd_size = size; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + if (disk->d_state == G_RAID_DISK_S_ACTIVE) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + } + } + + /* Write metadata based on created entities. */ + g_raid_md_write_intel(md, NULL, NULL, NULL); + + g_raid_event_send(vol, G_RAID_VOLUME_E_START, + G_RAID_EVENT_VOLUME); + return (0); + } + if (strcmp(verb, "delete") == 0) { + + /* Full node destruction. */ + if (*nargs == 1) { + /* Check if some volume is still open. */ + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (force != NULL && *force == 0 && + g_raid_nopens(sc) != 0) { + gctl_error(req, "Some volume is still open."); + return (-4); + } + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer) + intel_meta_erase(disk->d_consumer); + } + g_raid_destroy_node(sc, 0); + return (0); + } + + /* Destroy specified volume. If it was last - all node. */ + if (*nargs != 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + volname = gctl_get_asciiparam(req, "arg1"); + if (volname == NULL) { + gctl_error(req, "No volume name."); + return (-2); + } + + /* Search for volume. */ + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + if (strcmp(vol->v_name, volname) == 0) + break; + } + if (vol == NULL) { + i = strtol(volname, &tmp, 10); + if (verb != volname && tmp[0] == 0) { + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + if (vol->v_global_id == i) + break; + } + } + } + if (vol == NULL) { + gctl_error(req, "Volume '%s' not found.", volname); + return (-3); + } + + /* Check if volume is still open. */ + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (force != NULL && *force == 0 && + vol->v_provider_open != 0) { + gctl_error(req, "Volume is still open."); + return (-4); + } + + /* Destroy volume and potentially node. */ + i = 0; + TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next) + i++; + if (i >= 2) { + g_raid_destroy_volume(vol); + g_raid_md_write_intel(md, NULL, NULL, NULL); + } else { + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer) + intel_meta_erase(disk->d_consumer); + } + g_raid_destroy_node(sc, 0); + } + return (0); + } + if (strcmp(verb, "remove") == 0 || + strcmp(verb, "fail") == 0) { + if (*nargs < 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + for (i = 1; i < *nargs; i++) { + snprintf(arg, sizeof(arg), "arg%d", i); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -2; + break; + } + if (strncmp(diskname, "/dev/", 5) == 0) + diskname += 5; + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer != NULL && + disk->d_consumer->provider != NULL && + strcmp(disk->d_consumer->provider->name, + diskname) == 0) + break; + } + if (disk == NULL) { + gctl_error(req, "Disk '%s' not found.", + diskname); + error = -3; + break; + } + + if (strcmp(verb, "fail") == 0) { + g_raid_md_fail_disk_intel(md, NULL, disk); + continue; + } + + pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; + + /* Erase metadata on deleting disk. */ + intel_meta_erase(disk->d_consumer); + + /* If disk was assigned, just update statuses. */ + if (pd->pd_disk_pos >= 0) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + if (disk->d_consumer) { + g_raid_kill_consumer(sc, disk->d_consumer); + disk->d_consumer = NULL; + } + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NONE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, + G_RAID_EVENT_SUBDISK); + } + } else { + /* Otherwise -- delete. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); + g_raid_destroy_disk(disk); + } + } + + /* Write updated metadata to remaining disks. */ + g_raid_md_write_intel(md, NULL, NULL, NULL); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_intel_refill(sc); + return (error); + } + if (strcmp(verb, "insert") == 0) { + if (*nargs < 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + update = 0; + for (i = 1; i < *nargs; i++) { + /* Get disk name. */ + snprintf(arg, sizeof(arg), "arg%d", i); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -3; + break; + } + + /* Try to find provider with specified name. */ + g_topology_lock(); + cp = g_raid_open_consumer(sc, diskname); + if (cp == NULL) { + gctl_error(req, "Can't open disk '%s'.", + diskname); + g_topology_unlock(); + error = -4; + break; + } + pp = cp->provider; + g_topology_unlock(); + + /* Read disk serial. */ + error = g_raid_md_get_label(cp, + &serial[0], INTEL_SERIAL_LEN); + if (error != 0) { + gctl_error(req, + "Can't get serial for provider '%s'.", + diskname); + g_raid_kill_consumer(sc, cp); + error = -7; + break; + } + + pd = malloc(sizeof(*pd), M_MD_INTEL, M_WAITOK | M_ZERO); + pd->pd_disk_pos = -1; + + disk = g_raid_create_disk(sc); + disk->d_consumer = cp; + disk->d_consumer->private = disk; + disk->d_md_data = (void *)pd; + cp->private = disk; + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, + "Dumping not supported by %s.", + cp->provider->name); + + memcpy(&pd->pd_disk_meta.serial[0], &serial[0], + INTEL_SERIAL_LEN); + pd->pd_disk_meta.sectors = pp->mediasize / pp->sectorsize; + pd->pd_disk_meta.id = 0; + pd->pd_disk_meta.flags = INTEL_F_SPARE; + + /* Welcome the "new" disk. */ + update += g_raid_md_intel_start_disk(disk); + if (disk->d_state == G_RAID_DISK_S_SPARE) { + intel_meta_write_spare(cp, &pd->pd_disk_meta); + g_raid_destroy_disk(disk); + } else if (disk->d_state != G_RAID_DISK_S_ACTIVE) { + gctl_error(req, "Disk '%s' doesn't fit.", + diskname); + g_raid_destroy_disk(disk); + error = -8; + break; + } + } + + /* Write new metadata if we changed something. */ + if (update) + g_raid_md_write_intel(md, NULL, NULL, NULL); + return (error); + } + return (-100); +} + +static int +g_raid_md_write_intel(struct g_raid_md_object *md, struct g_raid_volume *tvol, + struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_intel_object *mdi; + struct g_raid_md_intel_perdisk *pd; + struct intel_raid_conf *meta; + struct intel_raid_vol *mvol; + struct intel_raid_map *mmap0, *mmap1; + off_t sectorsize = 512, pos; + const char *version, *cv; + int vi, sdi, numdisks, len, state, stale; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_intel_object *)md; + + if (sc->sc_stopping == G_RAID_DESTROY_HARD) + return (0); + + /* Bump generation. Newly written metadata may differ from previous. */ + mdi->mdio_generation++; + + /* Count number of disks. */ + numdisks = 0; + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; + if (pd->pd_disk_pos < 0) + continue; + numdisks++; + if (disk->d_state == G_RAID_DISK_S_ACTIVE) { + pd->pd_disk_meta.flags = + INTEL_F_ONLINE | INTEL_F_ASSIGNED; + } else if (disk->d_state == G_RAID_DISK_S_FAILED) { + pd->pd_disk_meta.flags = INTEL_F_FAILED | INTEL_F_ASSIGNED; + } else { + pd->pd_disk_meta.flags = INTEL_F_ASSIGNED; + if (pd->pd_disk_meta.id != 0xffffffff) { + pd->pd_disk_meta.id = 0xffffffff; + len = strlen(pd->pd_disk_meta.serial); + len = min(len, INTEL_SERIAL_LEN - 3); + strcpy(pd->pd_disk_meta.serial + len, ":0"); + } + } + } + + /* Fill anchor and disks. */ + meta = malloc(INTEL_MAX_MD_SIZE(numdisks), + M_MD_INTEL, M_WAITOK | M_ZERO); + memcpy(&meta->intel_id[0], INTEL_MAGIC, sizeof(INTEL_MAGIC)); + meta->config_size = INTEL_MAX_MD_SIZE(numdisks); + meta->config_id = mdi->mdio_config_id; + meta->generation = mdi->mdio_generation; + meta->attributes = INTEL_ATTR_CHECKSUM; + meta->total_disks = numdisks; + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; + if (pd->pd_disk_pos < 0) + continue; + meta->disk[pd->pd_disk_pos] = pd->pd_disk_meta; + } + + /* Fill volumes and maps. */ + vi = 0; + version = INTEL_VERSION_1000; + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + if (vol->v_stopping) + continue; + mvol = intel_get_volume(meta, vi); + + /* New metadata may have different volumes order. */ + vol->v_md_data = (void *)(intptr_t)vi; + + for (sdi = 0; sdi < vol->v_disks_count; sdi++) { + sd = &vol->v_subdisks[sdi]; + if (sd->sd_disk != NULL) + break; + } + if (sdi >= vol->v_disks_count) + panic("No any filled subdisk in volume"); + if (vol->v_mediasize >= 0x20000000000llu) + meta->attributes |= INTEL_ATTR_2TB; + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0) + meta->attributes |= INTEL_ATTR_RAID0; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) + meta->attributes |= INTEL_ATTR_RAID1; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) + meta->attributes |= INTEL_ATTR_RAID5; + else + meta->attributes |= INTEL_ATTR_RAID10; + + if (meta->attributes & INTEL_ATTR_2TB) + cv = INTEL_VERSION_1300; +// else if (dev->status == DEV_CLONE_N_GO) +// cv = INTEL_VERSION_1206; + else if (vol->v_disks_count > 4) + cv = INTEL_VERSION_1204; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) + cv = INTEL_VERSION_1202; + else if (vol->v_disks_count > 2) + cv = INTEL_VERSION_1201; + else if (vi > 0) + cv = INTEL_VERSION_1200; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) + cv = INTEL_VERSION_1100; + else + cv = INTEL_VERSION_1000; + if (strcmp(cv, version) > 0) + version = cv; + + strlcpy(&mvol->name[0], vol->v_name, sizeof(mvol->name)); + mvol->total_sectors = vol->v_mediasize / sectorsize; + + /* Check for any recovery in progress. */ + state = G_RAID_SUBDISK_S_ACTIVE; + pos = 0x7fffffffffffffffllu; + stale = 0; + for (sdi = 0; sdi < vol->v_disks_count; sdi++) { + sd = &vol->v_subdisks[sdi]; + if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) + state = G_RAID_SUBDISK_S_REBUILD; + else if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC && + state != G_RAID_SUBDISK_S_REBUILD) + state = G_RAID_SUBDISK_S_RESYNC; + else if (sd->sd_state == G_RAID_SUBDISK_S_STALE) + stale = 1; + if ((sd->sd_state == G_RAID_SUBDISK_S_REBUILD || + sd->sd_state == G_RAID_SUBDISK_S_RESYNC) && + sd->sd_rebuild_pos < pos) + pos = sd->sd_rebuild_pos; + } + if (state == G_RAID_SUBDISK_S_REBUILD) { + mvol->migr_state = 1; + mvol->migr_type = INTEL_MT_REBUILD; + } else if (state == G_RAID_SUBDISK_S_RESYNC) { + mvol->migr_state = 1; + /* mvol->migr_type = INTEL_MT_REPAIR; */ + mvol->migr_type = INTEL_MT_VERIFY; + mvol->state |= INTEL_ST_VERIFY_AND_FIX; + } else + mvol->migr_state = 0; + mvol->dirty = (vol->v_dirty || stale); + + mmap0 = intel_get_map(mvol, 0); + + /* Write map / common part of two maps. */ + mmap0->offset = sd->sd_offset / sectorsize; + mmap0->disk_sectors = sd->sd_size / sectorsize; + mmap0->strip_sectors = vol->v_strip_size / sectorsize; + if (vol->v_state == G_RAID_VOLUME_S_BROKEN) + mmap0->status = INTEL_S_FAILURE; + else if (vol->v_state == G_RAID_VOLUME_S_DEGRADED) + mmap0->status = INTEL_S_DEGRADED; + else + mmap0->status = INTEL_S_READY; + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0) + mmap0->type = INTEL_T_RAID0; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || + vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) + mmap0->type = INTEL_T_RAID1; + else + mmap0->type = INTEL_T_RAID5; + mmap0->total_disks = vol->v_disks_count; + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) + mmap0->total_domains = vol->v_disks_count; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) + mmap0->total_domains = 2; + else + mmap0->total_domains = 1; + mmap0->stripe_count = sd->sd_size / vol->v_strip_size / + mmap0->total_domains; + mmap0->failed_disk_num = 0xff; + mmap0->ddf = 1; + + /* If there are two maps - copy common and update. */ + if (mvol->migr_state) { + mvol->curr_migr_unit = pos / + vol->v_strip_size / mmap0->total_domains; + mmap1 = intel_get_map(mvol, 1); + memcpy(mmap1, mmap0, sizeof(struct intel_raid_map)); + mmap0->status = INTEL_S_READY; + } else + mmap1 = NULL; + + /* Write disk indexes and put rebuild flags. */ + for (sdi = 0; sdi < vol->v_disks_count; sdi++) { + sd = &vol->v_subdisks[sdi]; + pd = (struct g_raid_md_intel_perdisk *) + sd->sd_disk->d_md_data; + mmap0->disk_idx[sdi] = pd->pd_disk_pos; + if (mvol->migr_state) + mmap1->disk_idx[sdi] = pd->pd_disk_pos; + if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || + sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { + mmap1->disk_idx[sdi] |= INTEL_DI_RBLD; + } else if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE && + sd->sd_state != G_RAID_SUBDISK_S_STALE) { + mmap0->disk_idx[sdi] |= INTEL_DI_RBLD; + if (mvol->migr_state) + mmap1->disk_idx[sdi] |= INTEL_DI_RBLD; + } + if ((sd->sd_state == G_RAID_SUBDISK_S_NONE || + sd->sd_state == G_RAID_SUBDISK_S_FAILED) && + mmap0->failed_disk_num == 0xff) { + mmap0->failed_disk_num = sdi; + if (mvol->migr_state) + mmap1->failed_disk_num = sdi; + } + } + vi++; + } + meta->total_volumes = vi; + if (strcmp(version, INTEL_VERSION_1300) != 0) + meta->attributes &= INTEL_ATTR_CHECKSUM; + memcpy(&meta->version[0], version, sizeof(INTEL_VERSION_1000)); + + /* We are done. Print meta data and store them to disks. */ + g_raid_md_intel_print(meta); + if (mdi->mdio_meta != NULL) + free(mdi->mdio_meta, M_MD_INTEL); + mdi->mdio_meta = meta; + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; + if (disk->d_state != G_RAID_DISK_S_ACTIVE) + continue; + if (pd->pd_meta != NULL) { + free(pd->pd_meta, M_MD_INTEL); + pd->pd_meta = NULL; + } + pd->pd_meta = intel_meta_copy(meta); + intel_meta_write(disk->d_consumer, meta); + } + return (0); +} + +static int +g_raid_md_fail_disk_intel(struct g_raid_md_object *md, + struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) +{ + struct g_raid_softc *sc; + struct g_raid_md_intel_object *mdi; + struct g_raid_md_intel_perdisk *pd; + struct g_raid_subdisk *sd; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_intel_object *)md; + pd = (struct g_raid_md_intel_perdisk *)tdisk->d_md_data; + + /* We can't fail disk that is not a part of array now. */ + if (pd->pd_disk_pos < 0) + return (-1); + + /* + * Mark disk as failed in metadata and try to write that metadata + * to the disk itself to prevent it's later resurrection as STALE. + */ + mdi->mdio_meta->disk[pd->pd_disk_pos].flags = INTEL_F_FAILED; + pd->pd_disk_meta.flags = INTEL_F_FAILED; + g_raid_md_intel_print(mdi->mdio_meta); + if (tdisk->d_consumer) + intel_meta_write(tdisk->d_consumer, mdi->mdio_meta); + + /* Change states. */ + g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED); + TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_FAILED); + g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED, + G_RAID_EVENT_SUBDISK); + } + + /* Write updated metadata to remaining disks. */ + g_raid_md_write_intel(md, NULL, NULL, tdisk); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_intel_refill(sc); + return (0); +} + +static int +g_raid_md_free_disk_intel(struct g_raid_md_object *md, + struct g_raid_disk *disk) +{ + struct g_raid_md_intel_perdisk *pd; + + pd = (struct g_raid_md_intel_perdisk *)disk->d_md_data; + if (pd->pd_meta != NULL) { + free(pd->pd_meta, M_MD_INTEL); + pd->pd_meta = NULL; + } + free(pd, M_MD_INTEL); + disk->d_md_data = NULL; + return (0); +} + +static int +g_raid_md_free_intel(struct g_raid_md_object *md) +{ + struct g_raid_md_intel_object *mdi; + + mdi = (struct g_raid_md_intel_object *)md; + if (!mdi->mdio_started) { + mdi->mdio_started = 0; + callout_stop(&mdi->mdio_start_co); + G_RAID_DEBUG1(1, md->mdo_softc, + "root_mount_rel %p", mdi->mdio_rootmount); + root_mount_rel(mdi->mdio_rootmount); + mdi->mdio_rootmount = NULL; + } + if (mdi->mdio_meta != NULL) { + free(mdi->mdio_meta, M_MD_INTEL); + mdi->mdio_meta = NULL; + } + return (0); +} + +G_RAID_MD_DECLARE(g_raid_md_intel); diff --git a/sys/geom/raid/md_jmicron.c b/sys/geom/raid/md_jmicron.c new file mode 100644 index 000000000000..a56c543267e8 --- /dev/null +++ b/sys/geom/raid/md_jmicron.c @@ -0,0 +1,1582 @@ +/*- + * Copyright (c) 2010 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "geom/raid/g_raid.h" +#include "g_raid_md_if.h" + +static MALLOC_DEFINE(M_MD_JMICRON, "md_jmicron_data", "GEOM_RAID JMicron metadata"); + +#define JMICRON_MAX_DISKS 8 +#define JMICRON_MAX_SPARE 2 + +struct jmicron_raid_conf { + u_int8_t signature[2]; +#define JMICRON_MAGIC "JM" + + u_int16_t version; +#define JMICRON_VERSION 0x0001 + + u_int16_t checksum; + u_int8_t filler_1[10]; + u_int32_t disk_id; + u_int32_t offset; + u_int32_t disk_sectors_high; + u_int16_t disk_sectors_low; + u_int8_t filler_2[2]; + u_int8_t name[16]; + u_int8_t type; +#define JMICRON_T_RAID0 0 +#define JMICRON_T_RAID1 1 +#define JMICRON_T_RAID01 2 +#define JMICRON_T_CONCAT 3 +#define JMICRON_T_RAID5 5 + + u_int8_t stripe_shift; + u_int16_t flags; +#define JMICRON_F_READY 0x0001 +#define JMICRON_F_BOOTABLE 0x0002 +#define JMICRON_F_BADSEC 0x0004 +#define JMICRON_F_ACTIVE 0x0010 +#define JMICRON_F_UNSYNC 0x0020 +#define JMICRON_F_NEWEST 0x0040 + + u_int8_t filler_3[4]; + u_int32_t spare[JMICRON_MAX_SPARE]; + u_int32_t disks[JMICRON_MAX_DISKS]; +#define JMICRON_DISK_MASK 0xFFFFFFF0 +#define JMICRON_SEG_MASK 0x0000000F + u_int8_t filler_4[32]; + u_int8_t filler_5[384]; +}; + +struct g_raid_md_jmicron_perdisk { + struct jmicron_raid_conf *pd_meta; + int pd_disk_pos; + int pd_disk_id; + off_t pd_disk_size; +}; + +struct g_raid_md_jmicron_object { + struct g_raid_md_object mdio_base; + uint32_t mdio_config_id; + struct jmicron_raid_conf *mdio_meta; + struct callout mdio_start_co; /* STARTING state timer. */ + int mdio_total_disks; + int mdio_disks_present; + int mdio_started; + int mdio_incomplete; + struct root_hold_token *mdio_rootmount; /* Root mount delay token. */ +}; + +static g_raid_md_create_t g_raid_md_create_jmicron; +static g_raid_md_taste_t g_raid_md_taste_jmicron; +static g_raid_md_event_t g_raid_md_event_jmicron; +static g_raid_md_ctl_t g_raid_md_ctl_jmicron; +static g_raid_md_write_t g_raid_md_write_jmicron; +static g_raid_md_fail_disk_t g_raid_md_fail_disk_jmicron; +static g_raid_md_free_disk_t g_raid_md_free_disk_jmicron; +static g_raid_md_free_t g_raid_md_free_jmicron; + +static kobj_method_t g_raid_md_jmicron_methods[] = { + KOBJMETHOD(g_raid_md_create, g_raid_md_create_jmicron), + KOBJMETHOD(g_raid_md_taste, g_raid_md_taste_jmicron), + KOBJMETHOD(g_raid_md_event, g_raid_md_event_jmicron), + KOBJMETHOD(g_raid_md_ctl, g_raid_md_ctl_jmicron), + KOBJMETHOD(g_raid_md_write, g_raid_md_write_jmicron), + KOBJMETHOD(g_raid_md_fail_disk, g_raid_md_fail_disk_jmicron), + KOBJMETHOD(g_raid_md_free_disk, g_raid_md_free_disk_jmicron), + KOBJMETHOD(g_raid_md_free, g_raid_md_free_jmicron), + { 0, 0 } +}; + +static struct g_raid_md_class g_raid_md_jmicron_class = { + "JMicron", + g_raid_md_jmicron_methods, + sizeof(struct g_raid_md_jmicron_object), + .mdc_priority = 100 +}; + +static void +g_raid_md_jmicron_print(struct jmicron_raid_conf *meta) +{ + int k; + + if (g_raid_debug < 1) + return; + + printf("********* ATA JMicron RAID Metadata *********\n"); + printf("signature <%c%c>\n", meta->signature[0], meta->signature[1]); + printf("version %04x\n", meta->version); + printf("checksum 0x%04x\n", meta->checksum); + printf("disk_id 0x%08x\n", meta->disk_id); + printf("offset 0x%08x\n", meta->offset); + printf("disk_sectors_high 0x%08x\n", meta->disk_sectors_high); + printf("disk_sectors_low 0x%04x\n", meta->disk_sectors_low); + printf("name <%.16s>\n", meta->name); + printf("type %d\n", meta->type); + printf("stripe_shift %d\n", meta->stripe_shift); + printf("flags %04x\n", meta->flags); + printf("spare "); + for (k = 0; k < JMICRON_MAX_SPARE; k++) + printf(" 0x%08x", meta->spare[k]); + printf("\n"); + printf("disks "); + for (k = 0; k < JMICRON_MAX_DISKS; k++) + printf(" 0x%08x", meta->disks[k]); + printf("\n"); + printf("=================================================\n"); +} + +static struct jmicron_raid_conf * +jmicron_meta_copy(struct jmicron_raid_conf *meta) +{ + struct jmicron_raid_conf *nmeta; + + nmeta = malloc(sizeof(*meta), M_MD_JMICRON, M_WAITOK); + memcpy(nmeta, meta, sizeof(*meta)); + return (nmeta); +} + +static int +jmicron_meta_total_disks(struct jmicron_raid_conf *meta) +{ + int pos; + + for (pos = 0; pos < JMICRON_MAX_DISKS; pos++) { + if (meta->disks[pos] == 0) + break; + } + return (pos); +} + +static int +jmicron_meta_total_spare(struct jmicron_raid_conf *meta) +{ + int pos, n; + + n = 0; + for (pos = 0; pos < JMICRON_MAX_SPARE; pos++) { + if (meta->spare[pos] != 0) + n++; + } + return (n); +} + +/* + * Generate fake Configuration ID based on disk IDs. + * Note: it will change after each disk set change. + */ +static uint32_t +jmicron_meta_config_id(struct jmicron_raid_conf *meta) +{ + int pos; + uint32_t config_id; + + config_id = 0; + for (pos = 0; pos < JMICRON_MAX_DISKS; pos++) + config_id += meta->disks[pos] << pos; + return (config_id); +} + +static void +jmicron_meta_get_name(struct jmicron_raid_conf *meta, char *buf) +{ + int i; + + strncpy(buf, meta->name, 16); + buf[16] = 0; + for (i = 15; i >= 0; i--) { + if (buf[i] > 0x20) + break; + buf[i] = 0; + } +} + +static void +jmicron_meta_put_name(struct jmicron_raid_conf *meta, char *buf) +{ + + memset(meta->name, 0x20, 16); + memcpy(meta->name, buf, MIN(strlen(buf), 16)); +} + +static int +jmicron_meta_find_disk(struct jmicron_raid_conf *meta, uint32_t id) +{ + int pos; + + id &= JMICRON_DISK_MASK; + for (pos = 0; pos < JMICRON_MAX_DISKS; pos++) { + if ((meta->disks[pos] & JMICRON_DISK_MASK) == id) + return (pos); + } + for (pos = 0; pos < JMICRON_MAX_SPARE; pos++) { + if ((meta->spare[pos] & JMICRON_DISK_MASK) == id) + return (-3); + } + return (-1); +} + +static struct jmicron_raid_conf * +jmicron_meta_read(struct g_consumer *cp) +{ + struct g_provider *pp; + struct jmicron_raid_conf *meta; + char *buf; + int error, i; + uint16_t checksum, *ptr; + + pp = cp->provider; + + /* Read the anchor sector. */ + buf = g_read_data(cp, + pp->mediasize - pp->sectorsize, pp->sectorsize, &error); + if (buf == NULL) { + G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).", + pp->name, error); + return (NULL); + } + meta = (struct jmicron_raid_conf *)buf; + + /* Check if this is an JMicron RAID struct */ + if (strncmp(meta->signature, JMICRON_MAGIC, strlen(JMICRON_MAGIC))) { + G_RAID_DEBUG(1, "JMicron signature check failed on %s", pp->name); + g_free(buf); + return (NULL); + } + meta = malloc(sizeof(*meta), M_MD_JMICRON, M_WAITOK); + memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize)); + g_free(buf); + + /* Check metadata checksum. */ + for (checksum = 0, ptr = (uint16_t *)meta, i = 0; i < 64; i++) + checksum += *ptr++; + if (checksum != 0) { + G_RAID_DEBUG(1, "JMicron checksum check failed on %s", pp->name); + free(meta, M_MD_JMICRON); + return (NULL); + } + + return (meta); +} + +static int +jmicron_meta_write(struct g_consumer *cp, struct jmicron_raid_conf *meta) +{ + struct g_provider *pp; + char *buf; + int error, i; + uint16_t checksum, *ptr; + + pp = cp->provider; + + /* Recalculate checksum for case if metadata were changed. */ + meta->checksum = 0; + for (checksum = 0, ptr = (uint16_t *)meta, i = 0; i < 64; i++) + checksum += *ptr++; + meta->checksum -= checksum; + + /* Create and fill buffer. */ + buf = malloc(pp->sectorsize, M_MD_JMICRON, M_WAITOK | M_ZERO); + memcpy(buf, meta, sizeof(*meta)); + + error = g_write_data(cp, + pp->mediasize - pp->sectorsize, buf, pp->sectorsize); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).", + pp->name, error); + } + + free(buf, M_MD_JMICRON); + return (error); +} + +static int +jmicron_meta_erase(struct g_consumer *cp) +{ + struct g_provider *pp; + char *buf; + int error; + + pp = cp->provider; + buf = malloc(pp->sectorsize, M_MD_JMICRON, M_WAITOK | M_ZERO); + error = g_write_data(cp, + pp->mediasize - pp->sectorsize, buf, pp->sectorsize); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).", + pp->name, error); + } + free(buf, M_MD_JMICRON); + return (error); +} + +static struct g_raid_disk * +g_raid_md_jmicron_get_disk(struct g_raid_softc *sc, int id) +{ + struct g_raid_disk *disk; + struct g_raid_md_jmicron_perdisk *pd; + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data; + if (pd->pd_disk_pos == id) + break; + } + return (disk); +} + +static int +g_raid_md_jmicron_supported(int level, int qual, int disks, int force) +{ + + if (disks > 8) + return (0); + switch (level) { + case G_RAID_VOLUME_RL_RAID0: + if (disks < 1) + return (0); + if (!force && (disks < 2 || disks > 6)) + return (0); + break; + case G_RAID_VOLUME_RL_RAID1: + if (disks < 1) + return (0); + if (!force && (disks != 2)) + return (0); + break; + case G_RAID_VOLUME_RL_RAID1E: + if (disks < 2) + return (0); + if (!force && (disks != 4)) + return (0); + break; + case G_RAID_VOLUME_RL_SINGLE: + if (disks != 1) + return (0); + if (!force) + return (0); + break; + case G_RAID_VOLUME_RL_CONCAT: + if (disks < 2) + return (0); + break; + case G_RAID_VOLUME_RL_RAID5: + if (disks < 3) + return (0); + if (!force) + return (0); + break; + default: + return (0); + } + if (qual != G_RAID_VOLUME_RLQ_NONE) + return (0); + return (1); +} + +static int +g_raid_md_jmicron_start_disk(struct g_raid_disk *disk) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd, *tmpsd; + struct g_raid_disk *olddisk, *tmpdisk; + struct g_raid_md_object *md; + struct g_raid_md_jmicron_object *mdi; + struct g_raid_md_jmicron_perdisk *pd, *oldpd; + struct jmicron_raid_conf *meta; + int disk_pos, resurrection = 0; + + sc = disk->d_softc; + md = sc->sc_md; + mdi = (struct g_raid_md_jmicron_object *)md; + meta = mdi->mdio_meta; + pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data; + olddisk = NULL; + + /* Find disk position in metadata by it's serial. */ + disk_pos = jmicron_meta_find_disk(meta, pd->pd_disk_id); + if (disk_pos < 0) { + G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk"); + /* If we are in the start process, that's all for now. */ + if (!mdi->mdio_started) + goto nofit; + /* + * If we have already started - try to get use of the disk. + * Try to replace OFFLINE disks first, then FAILED. + */ + TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) { + if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE && + tmpdisk->d_state != G_RAID_DISK_S_FAILED) + continue; + /* Make sure this disk is big enough. */ + TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) { + if (sd->sd_offset + sd->sd_size + 512 > + pd->pd_disk_size) { + G_RAID_DEBUG1(1, sc, + "Disk too small (%ju < %ju)", + pd->pd_disk_size, + sd->sd_offset + sd->sd_size + 512); + break; + } + } + if (sd != NULL) + continue; + if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) { + olddisk = tmpdisk; + break; + } else if (olddisk == NULL) + olddisk = tmpdisk; + } + if (olddisk == NULL) { +nofit: + if (disk_pos == -3 || pd->pd_disk_pos == -3) { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_SPARE); + return (1); + } else { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_STALE); + return (0); + } + } + oldpd = (struct g_raid_md_jmicron_perdisk *)olddisk->d_md_data; + disk_pos = oldpd->pd_disk_pos; + resurrection = 1; + } + + if (olddisk == NULL) { + /* Find placeholder by position. */ + olddisk = g_raid_md_jmicron_get_disk(sc, disk_pos); + if (olddisk == NULL) + panic("No disk at position %d!", disk_pos); + if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) { + G_RAID_DEBUG1(1, sc, "More then one disk for pos %d", + disk_pos); + g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE); + return (0); + } + oldpd = (struct g_raid_md_jmicron_perdisk *)olddisk->d_md_data; + } + + /* Replace failed disk or placeholder with new disk. */ + TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) { + TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next); + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + sd->sd_disk = disk; + } + oldpd->pd_disk_pos = -2; + pd->pd_disk_pos = disk_pos; + /* Update global metadata just in case. */ + meta->disks[disk_pos] = pd->pd_disk_id; + + /* If it was placeholder -- destroy it. */ + if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) { + g_raid_destroy_disk(olddisk); + } else { + /* Otherwise, make it STALE_FAILED. */ + g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED); + } + + /* Welcome the new disk. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + + /* + * Different disks may have different sizes/offsets, + * especially in concat mode. Update. + */ + if (pd->pd_meta != NULL && !resurrection) { + sd->sd_offset = + (off_t)pd->pd_meta->offset * 16 * 512; //ZZZ + sd->sd_size = + (((off_t)pd->pd_meta->disk_sectors_high << 16) + + pd->pd_meta->disk_sectors_low) * 512; + } + + if (resurrection) { + /* Stale disk, almost same as new. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NEW); + } else if ((meta->flags & JMICRON_F_BADSEC) != 0 && + (pd->pd_meta->flags & JMICRON_F_BADSEC) == 0) { + /* Cold-inserted or rebuilding disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NEW); + } else if (pd->pd_meta->flags & JMICRON_F_UNSYNC) { + /* Dirty or resyncing disk.. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_STALE); + } else { + /* Up to date disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + } + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + } + + /* Update status of our need for spare. */ + if (mdi->mdio_started) { + mdi->mdio_incomplete = + (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < + mdi->mdio_total_disks); + } + + return (resurrection); +} + +static void +g_disk_md_jmicron_retaste(void *arg, int pending) +{ + + G_RAID_DEBUG(1, "Array is not complete, trying to retaste."); + g_retaste(&g_raid_class); + free(arg, M_MD_JMICRON); +} + +static void +g_raid_md_jmicron_refill(struct g_raid_softc *sc) +{ + struct g_raid_md_object *md; + struct g_raid_md_jmicron_object *mdi; + struct jmicron_raid_conf *meta; + struct g_raid_disk *disk; + struct task *task; + int update, na; + + md = sc->sc_md; + mdi = (struct g_raid_md_jmicron_object *)md; + meta = mdi->mdio_meta; + update = 0; + do { + /* Make sure we miss anything. */ + na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE); + if (na == mdi->mdio_total_disks) + break; + + G_RAID_DEBUG1(1, md->mdo_softc, + "Array is not complete (%d of %d), " + "trying to refill.", na, mdi->mdio_total_disks); + + /* Try to get use some of STALE disks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_STALE) { + update += g_raid_md_jmicron_start_disk(disk); + if (disk->d_state == G_RAID_DISK_S_ACTIVE) + break; + } + } + if (disk != NULL) + continue; + + /* Try to get use some of SPARE disks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_SPARE) { + update += g_raid_md_jmicron_start_disk(disk); + if (disk->d_state == G_RAID_DISK_S_ACTIVE) + break; + } + } + } while (disk != NULL); + + /* Write new metadata if we changed something. */ + if (update) { + g_raid_md_write_jmicron(md, NULL, NULL, NULL); + meta = mdi->mdio_meta; + } + + /* Update status of our need for spare. */ + mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < + mdi->mdio_total_disks); + + /* Request retaste hoping to find spare. */ + if (mdi->mdio_incomplete) { + task = malloc(sizeof(struct task), + M_MD_JMICRON, M_WAITOK | M_ZERO); + TASK_INIT(task, 0, g_disk_md_jmicron_retaste, task); + taskqueue_enqueue(taskqueue_swi, task); + } +} + +static void +g_raid_md_jmicron_start(struct g_raid_softc *sc) +{ + struct g_raid_md_object *md; + struct g_raid_md_jmicron_object *mdi; + struct g_raid_md_jmicron_perdisk *pd; + struct jmicron_raid_conf *meta; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + off_t size; + int j, disk_pos; + char buf[17]; + + md = sc->sc_md; + mdi = (struct g_raid_md_jmicron_object *)md; + meta = mdi->mdio_meta; + + /* Create volumes and subdisks. */ + jmicron_meta_get_name(meta, buf); + vol = g_raid_create_volume(sc, buf, -1); + size = ((off_t)meta->disk_sectors_high << 16) + meta->disk_sectors_low; + size *= 512; //ZZZ + if (meta->type == JMICRON_T_RAID0) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID0; + vol->v_mediasize = size * mdi->mdio_total_disks; + } else if (meta->type == JMICRON_T_RAID1) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1; + vol->v_mediasize = size; + } else if (meta->type == JMICRON_T_RAID01) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E; + vol->v_mediasize = size * mdi->mdio_total_disks / 2; + } else if (meta->type == JMICRON_T_CONCAT) { + if (mdi->mdio_total_disks == 1) + vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE; + else + vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT; + vol->v_mediasize = 0; + } else if (meta->type == JMICRON_T_RAID5) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID5; + vol->v_mediasize = size * (mdi->mdio_total_disks - 1); + } else { + vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; + vol->v_mediasize = 0; + } + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + vol->v_strip_size = 1024 << meta->stripe_shift; //ZZZ + vol->v_disks_count = mdi->mdio_total_disks; + vol->v_sectorsize = 512; //ZZZ + for (j = 0; j < vol->v_disks_count; j++) { + sd = &vol->v_subdisks[j]; + sd->sd_offset = (off_t)meta->offset * 16 * 512; //ZZZ + sd->sd_size = size; + } + g_raid_start_volume(vol); + + /* Create disk placeholders to store data for later writing. */ + for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) { + pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO); + pd->pd_disk_pos = disk_pos; + pd->pd_disk_id = meta->disks[disk_pos]; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_state = G_RAID_DISK_S_OFFLINE; + sd = &vol->v_subdisks[disk_pos]; + sd->sd_disk = disk; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + } + + /* Make all disks found till the moment take their places. */ + do { + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_NONE) { + g_raid_md_jmicron_start_disk(disk); + break; + } + } + } while (disk != NULL); + + mdi->mdio_started = 1; + G_RAID_DEBUG1(0, sc, "Array started."); + g_raid_md_write_jmicron(md, NULL, NULL, NULL); + + /* Pickup any STALE/SPARE disks to refill array if needed. */ + g_raid_md_jmicron_refill(sc); + + g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME); + + callout_stop(&mdi->mdio_start_co); + G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount); + root_mount_rel(mdi->mdio_rootmount); + mdi->mdio_rootmount = NULL; +} + +static void +g_raid_md_jmicron_new_disk(struct g_raid_disk *disk) +{ + struct g_raid_softc *sc; + struct g_raid_md_object *md; + struct g_raid_md_jmicron_object *mdi; + struct jmicron_raid_conf *pdmeta; + struct g_raid_md_jmicron_perdisk *pd; + + sc = disk->d_softc; + md = sc->sc_md; + mdi = (struct g_raid_md_jmicron_object *)md; + pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data; + pdmeta = pd->pd_meta; + + if (mdi->mdio_started) { + if (g_raid_md_jmicron_start_disk(disk)) + g_raid_md_write_jmicron(md, NULL, NULL, NULL); + } else { + /* + * If we haven't started yet - update common metadata + * to get subdisks details, avoiding data from spare disks. + */ + if (mdi->mdio_meta == NULL || + jmicron_meta_find_disk(mdi->mdio_meta, + mdi->mdio_meta->disk_id) == -3) { + if (mdi->mdio_meta != NULL) + free(mdi->mdio_meta, M_MD_JMICRON); + mdi->mdio_meta = jmicron_meta_copy(pdmeta); + mdi->mdio_total_disks = jmicron_meta_total_disks(pdmeta); + } + mdi->mdio_meta->flags |= pdmeta->flags & JMICRON_F_BADSEC; + + mdi->mdio_disks_present++; + G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d+%d up)", + mdi->mdio_disks_present, + mdi->mdio_total_disks, + jmicron_meta_total_spare(mdi->mdio_meta)); + + /* If we collected all needed disks - start array. */ + if (mdi->mdio_disks_present == mdi->mdio_total_disks + + jmicron_meta_total_spare(mdi->mdio_meta)) + g_raid_md_jmicron_start(sc); + } +} + +static void +g_raid_jmicron_go(void *arg) +{ + struct g_raid_softc *sc; + struct g_raid_md_object *md; + struct g_raid_md_jmicron_object *mdi; + + sc = arg; + md = sc->sc_md; + mdi = (struct g_raid_md_jmicron_object *)md; + if (!mdi->mdio_started) { + G_RAID_DEBUG1(0, sc, "Force array start due to timeout."); + g_raid_event_send(sc, G_RAID_NODE_E_START, 0); + } +} + +static int +g_raid_md_create_jmicron(struct g_raid_md_object *md, struct g_class *mp, + struct g_geom **gp) +{ + struct g_raid_softc *sc; + struct g_raid_md_jmicron_object *mdi; + char name[16]; + + mdi = (struct g_raid_md_jmicron_object *)md; + mdi->mdio_config_id = arc4random(); + snprintf(name, sizeof(name), "JMicron-%08x", mdi->mdio_config_id); + sc = g_raid_create_node(mp, name, md); + if (sc == NULL) + return (G_RAID_MD_TASTE_FAIL); + md->mdo_softc = sc; + *gp = sc->sc_geom; + return (G_RAID_MD_TASTE_NEW); +} + +static int +g_raid_md_taste_jmicron(struct g_raid_md_object *md, struct g_class *mp, + struct g_consumer *cp, struct g_geom **gp) +{ + struct g_consumer *rcp; + struct g_provider *pp; + struct g_raid_md_jmicron_object *mdi, *mdi1; + struct g_raid_softc *sc; + struct g_raid_disk *disk; + struct jmicron_raid_conf *meta; + struct g_raid_md_jmicron_perdisk *pd; + struct g_geom *geom; + int error, disk_pos, result, spare, len; + char name[16]; + uint16_t vendor; + + G_RAID_DEBUG(1, "Tasting JMicron on %s", cp->provider->name); + mdi = (struct g_raid_md_jmicron_object *)md; + pp = cp->provider; + + /* Read metadata from device. */ + meta = NULL; + spare = 0; + vendor = 0xffff; + disk_pos = 0; + if (g_access(cp, 1, 0, 0) != 0) + return (G_RAID_MD_TASTE_FAIL); + g_topology_unlock(); + len = 2; + if (pp->geom->rank == 1) + g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor); + meta = jmicron_meta_read(cp); + g_topology_lock(); + g_access(cp, -1, 0, 0); + if (meta == NULL) { + if (g_raid_aggressive_spare) { + if (vendor == 0x197b) { + G_RAID_DEBUG(1, + "No JMicron metadata, forcing spare."); + spare = 2; + goto search; + } else { + G_RAID_DEBUG(1, + "JMicron vendor mismatch 0x%04x != 0x197b", + vendor); + } + } + return (G_RAID_MD_TASTE_FAIL); + } + + /* Check this disk position in obtained metadata. */ + disk_pos = jmicron_meta_find_disk(meta, meta->disk_id); + if (disk_pos == -1) { + G_RAID_DEBUG(1, "JMicron disk_id %08x not found", + meta->disk_id); + goto fail1; + } + + /* Metadata valid. Print it. */ + g_raid_md_jmicron_print(meta); + G_RAID_DEBUG(1, "JMicron disk position %d", disk_pos); + spare = (disk_pos == -2) ? 1 : 0; + +search: + /* Search for matching node. */ + sc = NULL; + mdi1 = NULL; + LIST_FOREACH(geom, &mp->geom, geom) { + sc = geom->softc; + if (sc == NULL) + continue; + if (sc->sc_stopping != 0) + continue; + if (sc->sc_md->mdo_class != md->mdo_class) + continue; + mdi1 = (struct g_raid_md_jmicron_object *)sc->sc_md; + if (spare == 2) { + if (mdi1->mdio_incomplete) + break; + } else { + if (mdi1->mdio_config_id == + jmicron_meta_config_id(meta)) + break; + } + } + + /* Found matching node. */ + if (geom != NULL) { + G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name); + result = G_RAID_MD_TASTE_EXISTING; + + } else if (spare) { /* Not found needy node -- left for later. */ + G_RAID_DEBUG(1, "Spare is not needed at this time"); + goto fail1; + + } else { /* Not found matching node -- create one. */ + result = G_RAID_MD_TASTE_NEW; + mdi->mdio_config_id = jmicron_meta_config_id(meta); + snprintf(name, sizeof(name), "JMicron-%08x", + mdi->mdio_config_id); + sc = g_raid_create_node(mp, name, md); + md->mdo_softc = sc; + geom = sc->sc_geom; + callout_init(&mdi->mdio_start_co, 1); + callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz, + g_raid_jmicron_go, sc); + mdi->mdio_rootmount = root_mount_hold("GRAID-JMicron"); + G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount); + } + + rcp = g_new_consumer(geom); + g_attach(rcp, pp); + if (g_access(rcp, 1, 1, 1) != 0) + ; //goto fail1; + + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + + pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO); + pd->pd_meta = meta; + if (spare == 2) { + pd->pd_disk_pos = -3; + pd->pd_disk_id = arc4random() & JMICRON_DISK_MASK; + } else { + pd->pd_disk_pos = -1; + pd->pd_disk_id = meta->disk_id; + } + pd->pd_disk_size = pp->mediasize; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_consumer = rcp; + rcp->private = disk; + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.", + rcp->provider->name, error); + + g_raid_md_jmicron_new_disk(disk); + + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + *gp = geom; + return (result); +fail1: + free(meta, M_MD_JMICRON); + return (G_RAID_MD_TASTE_FAIL); +} + +static int +g_raid_md_event_jmicron(struct g_raid_md_object *md, + struct g_raid_disk *disk, u_int event) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd; + struct g_raid_md_jmicron_object *mdi; + struct g_raid_md_jmicron_perdisk *pd; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_jmicron_object *)md; + if (disk == NULL) { + switch (event) { + case G_RAID_NODE_E_START: + if (!mdi->mdio_started) + g_raid_md_jmicron_start(sc); + return (0); + } + return (-1); + } + pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data; + switch (event) { + case G_RAID_DISK_E_DISCONNECTED: + /* If disk was assigned, just update statuses. */ + if (pd->pd_disk_pos >= 0) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + if (disk->d_consumer) { + g_raid_kill_consumer(sc, disk->d_consumer); + disk->d_consumer = NULL; + } + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NONE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, + G_RAID_EVENT_SUBDISK); + } + } else { + /* Otherwise -- delete. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); + g_raid_destroy_disk(disk); + } + + /* Write updated metadata to all disks. */ + g_raid_md_write_jmicron(md, NULL, NULL, NULL); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_jmicron_refill(sc); + return (0); + } + return (-2); +} + +static int +g_raid_md_ctl_jmicron(struct g_raid_md_object *md, + struct gctl_req *req) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_jmicron_object *mdi; + struct g_raid_md_jmicron_perdisk *pd; + struct g_consumer *cp; + struct g_provider *pp; + char arg[16]; + const char *verb, *volname, *levelname, *diskname; + int *nargs, *force; + off_t size, sectorsize, strip; + intmax_t *sizearg, *striparg; + int numdisks, i, len, level, qual, update; + int error; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_jmicron_object *)md; + verb = gctl_get_param(req, "verb", NULL); + nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); + error = 0; + if (strcmp(verb, "label") == 0) { + + if (*nargs < 4) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + volname = gctl_get_asciiparam(req, "arg1"); + if (volname == NULL) { + gctl_error(req, "No volume name."); + return (-2); + } + levelname = gctl_get_asciiparam(req, "arg2"); + if (levelname == NULL) { + gctl_error(req, "No RAID level."); + return (-3); + } + if (g_raid_volume_str2level(levelname, &level, &qual)) { + gctl_error(req, "Unknown RAID level '%s'.", levelname); + return (-4); + } + numdisks = *nargs - 3; + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (!g_raid_md_jmicron_supported(level, qual, numdisks, + force ? *force : 0)) { + gctl_error(req, "Unsupported RAID level " + "(0x%02x/0x%02x), or number of disks (%d).", + level, qual, numdisks); + return (-5); + } + + /* Search for disks, connect them and probe. */ + size = 0x7fffffffffffffffllu; + sectorsize = 0; + for (i = 0; i < numdisks; i++) { + snprintf(arg, sizeof(arg), "arg%d", i + 3); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -6; + break; + } + if (strcmp(diskname, "NONE") == 0) { + cp = NULL; + pp = NULL; + } else { + g_topology_lock(); + cp = g_raid_open_consumer(sc, diskname); + if (cp == NULL) { + gctl_error(req, "Can't open '%s'.", + diskname); + g_topology_unlock(); + error = -7; + break; + } + pp = cp->provider; + } + pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO); + pd->pd_disk_pos = i; + pd->pd_disk_id = arc4random() & JMICRON_DISK_MASK; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_consumer = cp; + if (cp == NULL) + continue; + cp->private = disk; + g_topology_unlock(); + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, + "Dumping not supported by %s.", + cp->provider->name); + + pd->pd_disk_size = pp->mediasize; + if (size > pp->mediasize) + size = pp->mediasize; + if (sectorsize < pp->sectorsize) + sectorsize = pp->sectorsize; + } + if (error != 0) + return (error); + + /* Reserve space for metadata. */ + size -= sectorsize; + + /* Handle size argument. */ + len = sizeof(*sizearg); + sizearg = gctl_get_param(req, "size", &len); + if (sizearg != NULL && len == sizeof(*sizearg) && + *sizearg > 0) { + if (*sizearg > size) { + gctl_error(req, "Size too big %lld > %lld.", + (long long)*sizearg, (long long)size); + return (-9); + } + size = *sizearg; + } + + /* Handle strip argument. */ + strip = 131072; + len = sizeof(*striparg); + striparg = gctl_get_param(req, "strip", &len); + if (striparg != NULL && len == sizeof(*striparg) && + *striparg > 0) { + if (*striparg < sectorsize) { + gctl_error(req, "Strip size too small."); + return (-10); + } + if (*striparg % sectorsize != 0) { + gctl_error(req, "Incorrect strip size."); + return (-11); + } + if (strip > 65535 * sectorsize) { + gctl_error(req, "Strip size too big."); + return (-12); + } + strip = *striparg; + } + + /* Round size down to strip or sector. */ + if (level == G_RAID_VOLUME_RL_RAID1) + size -= (size % sectorsize); + else if (level == G_RAID_VOLUME_RL_RAID1E && + (numdisks & 1) != 0) + size -= (size % (2 * strip)); + else + size -= (size % strip); + if (size <= 0) { + gctl_error(req, "Size too small."); + return (-13); + } + if (size > 0xffffffffffffllu * sectorsize) { + gctl_error(req, "Size too big."); + return (-14); + } + + /* We have all we need, create things: volume, ... */ + mdi->mdio_total_disks = numdisks; + mdi->mdio_started = 1; + vol = g_raid_create_volume(sc, volname, -1); + vol->v_md_data = (void *)(intptr_t)0; + vol->v_raid_level = level; + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + vol->v_strip_size = strip; + vol->v_disks_count = numdisks; + if (level == G_RAID_VOLUME_RL_RAID0 || + level == G_RAID_VOLUME_RL_CONCAT || + level == G_RAID_VOLUME_RL_SINGLE) + vol->v_mediasize = size * numdisks; + else if (level == G_RAID_VOLUME_RL_RAID1) + vol->v_mediasize = size; + else if (level == G_RAID_VOLUME_RL_RAID5) + vol->v_mediasize = size * (numdisks - 1); + else { /* RAID1E */ + vol->v_mediasize = ((size * numdisks) / strip / 2) * + strip; + } + vol->v_sectorsize = sectorsize; + g_raid_start_volume(vol); + + /* , and subdisks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data; + sd = &vol->v_subdisks[pd->pd_disk_pos]; + sd->sd_disk = disk; + sd->sd_offset = 0; + sd->sd_size = size; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + if (sd->sd_disk->d_consumer != NULL) { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_ACTIVE); + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + } else { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + } + } + + /* Write metadata based on created entities. */ + G_RAID_DEBUG1(0, sc, "Array started."); + g_raid_md_write_jmicron(md, NULL, NULL, NULL); + + /* Pickup any STALE/SPARE disks to refill array if needed. */ + g_raid_md_jmicron_refill(sc); + + g_raid_event_send(vol, G_RAID_VOLUME_E_START, + G_RAID_EVENT_VOLUME); + return (0); + } + if (strcmp(verb, "delete") == 0) { + + /* Check if some volume is still open. */ + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (force != NULL && *force == 0 && + g_raid_nopens(sc) != 0) { + gctl_error(req, "Some volume is still open."); + return (-4); + } + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer) + jmicron_meta_erase(disk->d_consumer); + } + g_raid_destroy_node(sc, 0); + return (0); + } + if (strcmp(verb, "remove") == 0 || + strcmp(verb, "fail") == 0) { + if (*nargs < 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + for (i = 1; i < *nargs; i++) { + snprintf(arg, sizeof(arg), "arg%d", i); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -2; + break; + } + if (strncmp(diskname, "/dev/", 5) == 0) + diskname += 5; + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer != NULL && + disk->d_consumer->provider != NULL && + strcmp(disk->d_consumer->provider->name, + diskname) == 0) + break; + } + if (disk == NULL) { + gctl_error(req, "Disk '%s' not found.", + diskname); + error = -3; + break; + } + + if (strcmp(verb, "fail") == 0) { + g_raid_md_fail_disk_jmicron(md, NULL, disk); + continue; + } + + pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data; + + /* Erase metadata on deleting disk. */ + jmicron_meta_erase(disk->d_consumer); + + /* If disk was assigned, just update statuses. */ + if (pd->pd_disk_pos >= 0) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + if (disk->d_consumer) { + g_raid_kill_consumer(sc, disk->d_consumer); + disk->d_consumer = NULL; + } + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NONE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, + G_RAID_EVENT_SUBDISK); + } + } else { + /* Otherwise -- delete. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); + g_raid_destroy_disk(disk); + } + } + + /* Write updated metadata to remaining disks. */ + g_raid_md_write_jmicron(md, NULL, NULL, NULL); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_jmicron_refill(sc); + return (error); + } + if (strcmp(verb, "insert") == 0) { + if (*nargs < 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + update = 0; + for (i = 1; i < *nargs; i++) { + /* Get disk name. */ + snprintf(arg, sizeof(arg), "arg%d", i); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -3; + break; + } + + /* Try to find provider with specified name. */ + g_topology_lock(); + cp = g_raid_open_consumer(sc, diskname); + if (cp == NULL) { + gctl_error(req, "Can't open disk '%s'.", + diskname); + g_topology_unlock(); + error = -4; + break; + } + pp = cp->provider; + + pd = malloc(sizeof(*pd), M_MD_JMICRON, M_WAITOK | M_ZERO); + pd->pd_disk_pos = -3; + pd->pd_disk_id = arc4random() & JMICRON_DISK_MASK; + pd->pd_disk_size = pp->mediasize; + + disk = g_raid_create_disk(sc); + disk->d_consumer = cp; + disk->d_consumer->private = disk; + disk->d_md_data = (void *)pd; + cp->private = disk; + g_topology_unlock(); + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, + "Dumping not supported by %s.", + cp->provider->name); + + /* Welcome the "new" disk. */ + update += g_raid_md_jmicron_start_disk(disk); + if (disk->d_state != G_RAID_DISK_S_ACTIVE && + disk->d_state != G_RAID_DISK_S_SPARE) { + gctl_error(req, "Disk '%s' doesn't fit.", + diskname); + g_raid_destroy_disk(disk); + error = -8; + break; + } + } + + /* Write new metadata if we changed something. */ + if (update) + g_raid_md_write_jmicron(md, NULL, NULL, NULL); + return (error); + } + gctl_error(req, "Command '%s' is not supported.", verb); + return (-100); +} + +static int +g_raid_md_write_jmicron(struct g_raid_md_object *md, struct g_raid_volume *tvol, + struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_jmicron_object *mdi; + struct g_raid_md_jmicron_perdisk *pd; + struct jmicron_raid_conf *meta; + int i, spares; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_jmicron_object *)md; + + if (sc->sc_stopping == G_RAID_DESTROY_HARD) + return (0); + + /* There is only one volume. */ + vol = TAILQ_FIRST(&sc->sc_volumes); + + /* Fill global fields. */ + meta = malloc(sizeof(*meta), M_MD_JMICRON, M_WAITOK | M_ZERO); + strncpy(meta->signature, JMICRON_MAGIC, 2); + meta->version = JMICRON_VERSION; + jmicron_meta_put_name(meta, vol->v_name); + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0) + meta->type = JMICRON_T_RAID0; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) + meta->type = JMICRON_T_RAID1; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) + meta->type = JMICRON_T_RAID01; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT || + vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE) + meta->type = JMICRON_T_CONCAT; + else + meta->type = JMICRON_T_RAID5; + meta->stripe_shift = fls(vol->v_strip_size / 2048); + meta->flags = JMICRON_F_READY | JMICRON_F_BOOTABLE; + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if (sd->sd_disk == NULL || sd->sd_disk->d_md_data == NULL) + meta->disks[i] = 0xffffffff; + else { + pd = (struct g_raid_md_jmicron_perdisk *) + sd->sd_disk->d_md_data; + meta->disks[i] = pd->pd_disk_id; + } + if (sd->sd_state < G_RAID_SUBDISK_S_STALE) + meta->flags |= JMICRON_F_BADSEC; + if (vol->v_dirty) + meta->flags |= JMICRON_F_UNSYNC; + } + + /* Put spares to their slots. */ + spares = 0; + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data; + if (disk->d_state != G_RAID_DISK_S_SPARE) + continue; + meta->spare[spares] = pd->pd_disk_id; + if (++spares >= 2) + break; + } + + /* We are done. Print meta data and store them to disks. */ + if (mdi->mdio_meta != NULL) + free(mdi->mdio_meta, M_MD_JMICRON); + mdi->mdio_meta = meta; + i = 0; + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data; + if (disk->d_state != G_RAID_DISK_S_ACTIVE && + disk->d_state != G_RAID_DISK_S_SPARE) + continue; + if (pd->pd_meta != NULL) { + free(pd->pd_meta, M_MD_JMICRON); + pd->pd_meta = NULL; + } + pd->pd_meta = jmicron_meta_copy(meta); + pd->pd_meta->disk_id = pd->pd_disk_id; + if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) { + pd->pd_meta->offset = + (sd->sd_offset / 512) / 16; + pd->pd_meta->disk_sectors_high = + (sd->sd_size / 512) >> 16; + pd->pd_meta->disk_sectors_low = + (sd->sd_size / 512) & 0xffff; + if (sd->sd_state < G_RAID_SUBDISK_S_STALE) + pd->pd_meta->flags &= ~JMICRON_F_BADSEC; + else if (sd->sd_state < G_RAID_SUBDISK_S_ACTIVE) + pd->pd_meta->flags |= JMICRON_F_UNSYNC; + } + G_RAID_DEBUG(1, "Writing JMicron metadata to %s", + g_raid_get_diskname(disk)); + g_raid_md_jmicron_print(pd->pd_meta); + jmicron_meta_write(disk->d_consumer, pd->pd_meta); + } + return (0); +} + +static int +g_raid_md_fail_disk_jmicron(struct g_raid_md_object *md, + struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) +{ + struct g_raid_softc *sc; + struct g_raid_md_jmicron_object *mdi; + struct g_raid_md_jmicron_perdisk *pd; + struct g_raid_subdisk *sd; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_jmicron_object *)md; + pd = (struct g_raid_md_jmicron_perdisk *)tdisk->d_md_data; + + /* We can't fail disk that is not a part of array now. */ + if (pd->pd_disk_pos < 0) + return (-1); + + if (tdisk->d_consumer) + jmicron_meta_erase(tdisk->d_consumer); + + /* Change states. */ + g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED); + TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_FAILED); + g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED, + G_RAID_EVENT_SUBDISK); + } + + /* Write updated metadata to remaining disks. */ + g_raid_md_write_jmicron(md, NULL, NULL, tdisk); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_jmicron_refill(sc); + return (0); +} + +static int +g_raid_md_free_disk_jmicron(struct g_raid_md_object *md, + struct g_raid_disk *disk) +{ + struct g_raid_md_jmicron_perdisk *pd; + + pd = (struct g_raid_md_jmicron_perdisk *)disk->d_md_data; + if (pd->pd_meta != NULL) { + free(pd->pd_meta, M_MD_JMICRON); + pd->pd_meta = NULL; + } + free(pd, M_MD_JMICRON); + disk->d_md_data = NULL; + return (0); +} + +static int +g_raid_md_free_jmicron(struct g_raid_md_object *md) +{ + struct g_raid_md_jmicron_object *mdi; + + mdi = (struct g_raid_md_jmicron_object *)md; + if (!mdi->mdio_started) { + mdi->mdio_started = 0; + callout_stop(&mdi->mdio_start_co); + G_RAID_DEBUG1(1, md->mdo_softc, + "root_mount_rel %p", mdi->mdio_rootmount); + root_mount_rel(mdi->mdio_rootmount); + mdi->mdio_rootmount = NULL; + } + if (mdi->mdio_meta != NULL) { + free(mdi->mdio_meta, M_MD_JMICRON); + mdi->mdio_meta = NULL; + } + return (0); +} + +G_RAID_MD_DECLARE(g_raid_md_jmicron); diff --git a/sys/geom/raid/md_nvidia.c b/sys/geom/raid/md_nvidia.c new file mode 100644 index 000000000000..dbaee0a1593c --- /dev/null +++ b/sys/geom/raid/md_nvidia.c @@ -0,0 +1,1607 @@ +/*- + * Copyright (c) 2011 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "geom/raid/g_raid.h" +#include "g_raid_md_if.h" + +static MALLOC_DEFINE(M_MD_NVIDIA, "md_nvidia_data", "GEOM_RAID NVIDIA metadata"); + +struct nvidia_raid_conf { + uint8_t nvidia_id[8]; +#define NVIDIA_MAGIC "NVIDIA " + + uint32_t config_size; + uint32_t checksum; + uint16_t version; + uint8_t disk_number; + uint8_t dummy_0; + uint32_t total_sectors; + uint32_t sector_size; + uint8_t name[16]; + uint8_t revision[4]; + uint32_t disk_status; + + uint32_t magic_0; +#define NVIDIA_MAGIC0 0x00640044 + + uint64_t volume_id[2]; + uint8_t state; +#define NVIDIA_S_IDLE 0 +#define NVIDIA_S_INIT 2 +#define NVIDIA_S_REBUILD 3 +#define NVIDIA_S_UPGRADE 4 +#define NVIDIA_S_SYNC 5 + uint8_t array_width; + uint8_t total_disks; + uint8_t orig_array_width; + uint16_t type; +#define NVIDIA_T_RAID0 0x0080 +#define NVIDIA_T_RAID1 0x0081 +#define NVIDIA_T_RAID3 0x0083 +#define NVIDIA_T_RAID5 0x0085 /* RLQ = 00/02? */ +#define NVIDIA_T_RAID5_SYM 0x0095 /* RLQ = 03 */ +#define NVIDIA_T_RAID10 0x008a +#define NVIDIA_T_RAID01 0x8180 +#define NVIDIA_T_CONCAT 0x00ff + + uint16_t dummy_3; + uint32_t strip_sectors; + uint32_t strip_bytes; + uint32_t strip_shift; + uint32_t strip_mask; + uint32_t stripe_sectors; + uint32_t stripe_bytes; + uint32_t rebuild_lba; + uint32_t orig_type; + uint32_t orig_total_sectors; + uint32_t status; +#define NVIDIA_S_BOOTABLE 0x00000001 +#define NVIDIA_S_DEGRADED 0x00000002 + + uint32_t filler[98]; +} __packed; + +struct g_raid_md_nvidia_perdisk { + struct nvidia_raid_conf *pd_meta; + int pd_disk_pos; + off_t pd_disk_size; +}; + +struct g_raid_md_nvidia_object { + struct g_raid_md_object mdio_base; + uint64_t mdio_volume_id[2]; + struct nvidia_raid_conf *mdio_meta; + struct callout mdio_start_co; /* STARTING state timer. */ + int mdio_total_disks; + int mdio_disks_present; + int mdio_started; + int mdio_incomplete; + struct root_hold_token *mdio_rootmount; /* Root mount delay token. */ +}; + +static g_raid_md_create_t g_raid_md_create_nvidia; +static g_raid_md_taste_t g_raid_md_taste_nvidia; +static g_raid_md_event_t g_raid_md_event_nvidia; +static g_raid_md_ctl_t g_raid_md_ctl_nvidia; +static g_raid_md_write_t g_raid_md_write_nvidia; +static g_raid_md_fail_disk_t g_raid_md_fail_disk_nvidia; +static g_raid_md_free_disk_t g_raid_md_free_disk_nvidia; +static g_raid_md_free_t g_raid_md_free_nvidia; + +static kobj_method_t g_raid_md_nvidia_methods[] = { + KOBJMETHOD(g_raid_md_create, g_raid_md_create_nvidia), + KOBJMETHOD(g_raid_md_taste, g_raid_md_taste_nvidia), + KOBJMETHOD(g_raid_md_event, g_raid_md_event_nvidia), + KOBJMETHOD(g_raid_md_ctl, g_raid_md_ctl_nvidia), + KOBJMETHOD(g_raid_md_write, g_raid_md_write_nvidia), + KOBJMETHOD(g_raid_md_fail_disk, g_raid_md_fail_disk_nvidia), + KOBJMETHOD(g_raid_md_free_disk, g_raid_md_free_disk_nvidia), + KOBJMETHOD(g_raid_md_free, g_raid_md_free_nvidia), + { 0, 0 } +}; + +static struct g_raid_md_class g_raid_md_nvidia_class = { + "NVIDIA", + g_raid_md_nvidia_methods, + sizeof(struct g_raid_md_nvidia_object), + .mdc_priority = 100 +}; + +static int NVIDIANodeID = 1; + +static void +g_raid_md_nvidia_print(struct nvidia_raid_conf *meta) +{ + + if (g_raid_debug < 1) + return; + + printf("********* ATA NVIDIA RAID Metadata *********\n"); + printf("nvidia_id <%.8s>\n", meta->nvidia_id); + printf("config_size %u\n", meta->config_size); + printf("checksum 0x%08x\n", meta->checksum); + printf("version 0x%04x\n", meta->version); + printf("disk_number %d\n", meta->disk_number); + printf("dummy_0 0x%02x\n", meta->dummy_0); + printf("total_sectors %u\n", meta->total_sectors); + printf("sector_size %u\n", meta->sector_size); + printf("name <%.16s>\n", meta->name); + printf("revision 0x%02x%02x%02x%02x\n", + meta->revision[0], meta->revision[1], + meta->revision[2], meta->revision[3]); + printf("disk_status 0x%08x\n", meta->disk_status); + printf("magic_0 0x%08x\n", meta->magic_0); + printf("volume_id 0x%016jx%016jx\n", + meta->volume_id[1], meta->volume_id[0]); + printf("state 0x%02x\n", meta->state); + printf("array_width %u\n", meta->array_width); + printf("total_disks %u\n", meta->total_disks); + printf("orig_array_width %u\n", meta->orig_array_width); + printf("type 0x%04x\n", meta->type); + printf("dummy_3 0x%04x\n", meta->dummy_3); + printf("strip_sectors %u\n", meta->strip_sectors); + printf("strip_bytes %u\n", meta->strip_bytes); + printf("strip_shift %u\n", meta->strip_shift); + printf("strip_mask 0x%08x\n", meta->strip_mask); + printf("stripe_sectors %u\n", meta->stripe_sectors); + printf("stripe_bytes %u\n", meta->stripe_bytes); + printf("rebuild_lba %u\n", meta->rebuild_lba); + printf("orig_type 0x%04x\n", meta->orig_type); + printf("orig_total_sectors %u\n", meta->orig_total_sectors); + printf("status 0x%08x\n", meta->status); + printf("=================================================\n"); +} + +static struct nvidia_raid_conf * +nvidia_meta_copy(struct nvidia_raid_conf *meta) +{ + struct nvidia_raid_conf *nmeta; + + nmeta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK); + memcpy(nmeta, meta, sizeof(*meta)); + return (nmeta); +} + +static int +nvidia_meta_translate_disk(struct nvidia_raid_conf *meta, int md_disk_pos) +{ + int disk_pos; + + if (md_disk_pos >= 0 && meta->type == NVIDIA_T_RAID01) { + disk_pos = (md_disk_pos / meta->array_width) + + (md_disk_pos % meta->array_width) * meta->array_width; + } else + disk_pos = md_disk_pos; + return (disk_pos); +} + +static void +nvidia_meta_get_name(struct nvidia_raid_conf *meta, char *buf) +{ + int i; + + strncpy(buf, meta->name, 16); + buf[16] = 0; + for (i = 15; i >= 0; i--) { + if (buf[i] > 0x20) + break; + buf[i] = 0; + } +} + +static void +nvidia_meta_put_name(struct nvidia_raid_conf *meta, char *buf) +{ + + memset(meta->name, 0x20, 16); + memcpy(meta->name, buf, MIN(strlen(buf), 16)); +} + +static struct nvidia_raid_conf * +nvidia_meta_read(struct g_consumer *cp) +{ + struct g_provider *pp; + struct nvidia_raid_conf *meta; + char *buf; + int error, i; + uint32_t checksum, *ptr; + + pp = cp->provider; + + /* Read the anchor sector. */ + buf = g_read_data(cp, + pp->mediasize - 2 * pp->sectorsize, pp->sectorsize, &error); + if (buf == NULL) { + G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).", + pp->name, error); + return (NULL); + } + meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK); + memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize)); + g_free(buf); + + /* Check if this is an NVIDIA RAID struct */ + if (strncmp(meta->nvidia_id, NVIDIA_MAGIC, strlen(NVIDIA_MAGIC))) { + G_RAID_DEBUG(1, "NVIDIA signature check failed on %s", pp->name); + free(meta, M_MD_NVIDIA); + return (NULL); + } + if (meta->config_size > 128 || + meta->config_size < 30) { + G_RAID_DEBUG(1, "NVIDIA metadata size looks wrong: %d", + meta->config_size); + free(meta, M_MD_NVIDIA); + return (NULL); + } + + /* Check metadata checksum. */ + for (checksum = 0, ptr = (uint32_t *)meta, + i = 0; i < meta->config_size; i++) + checksum += *ptr++; + if (checksum != 0) { + G_RAID_DEBUG(1, "NVIDIA checksum check failed on %s", pp->name); + free(meta, M_MD_NVIDIA); + return (NULL); + } + + /* Check volume state. */ + if (meta->state != NVIDIA_S_IDLE && meta->state != NVIDIA_S_INIT && + meta->state != NVIDIA_S_REBUILD && meta->state != NVIDIA_S_SYNC) { + G_RAID_DEBUG(1, "NVIDIA unknown state on %s (0x%02x)", + pp->name, meta->state); + free(meta, M_MD_NVIDIA); + return (NULL); + } + + /* Check raid type. */ + if (meta->type != NVIDIA_T_RAID0 && meta->type != NVIDIA_T_RAID1 && + meta->type != NVIDIA_T_RAID3 && meta->type != NVIDIA_T_RAID5 && + meta->type != NVIDIA_T_RAID5_SYM && + meta->type != NVIDIA_T_RAID01 && meta->type != NVIDIA_T_CONCAT) { + G_RAID_DEBUG(1, "NVIDIA unknown RAID level on %s (0x%02x)", + pp->name, meta->type); + free(meta, M_MD_NVIDIA); + return (NULL); + } + + return (meta); +} + +static int +nvidia_meta_write(struct g_consumer *cp, struct nvidia_raid_conf *meta) +{ + struct g_provider *pp; + char *buf; + int error, i; + uint32_t checksum, *ptr; + + pp = cp->provider; + + /* Recalculate checksum for case if metadata were changed. */ + meta->checksum = 0; + for (checksum = 0, ptr = (uint32_t *)meta, + i = 0; i < meta->config_size; i++) + checksum += *ptr++; + meta->checksum -= checksum; + + /* Create and fill buffer. */ + buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO); + memcpy(buf, meta, sizeof(*meta)); + + /* Write metadata. */ + error = g_write_data(cp, + pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).", + pp->name, error); + } + + free(buf, M_MD_NVIDIA); + return (error); +} + +static int +nvidia_meta_erase(struct g_consumer *cp) +{ + struct g_provider *pp; + char *buf; + int error; + + pp = cp->provider; + buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO); + error = g_write_data(cp, + pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).", + pp->name, error); + } + free(buf, M_MD_NVIDIA); + return (error); +} + +static struct g_raid_disk * +g_raid_md_nvidia_get_disk(struct g_raid_softc *sc, int id) +{ + struct g_raid_disk *disk; + struct g_raid_md_nvidia_perdisk *pd; + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data; + if (pd->pd_disk_pos == id) + break; + } + return (disk); +} + +static int +g_raid_md_nvidia_supported(int level, int qual, int disks, int force) +{ + + switch (level) { + case G_RAID_VOLUME_RL_RAID0: + if (disks < 1) + return (0); + if (!force && (disks < 2 || disks > 6)) + return (0); + break; + case G_RAID_VOLUME_RL_RAID1: + if (disks < 1) + return (0); + if (!force && (disks != 2)) + return (0); + break; + case G_RAID_VOLUME_RL_RAID1E: + if (disks < 2) + return (0); + if (disks % 2 != 0) + return (0); + if (!force && (disks < 4)) + return (0); + break; + case G_RAID_VOLUME_RL_SINGLE: + if (disks != 1) + return (0); + break; + case G_RAID_VOLUME_RL_CONCAT: + if (disks < 2) + return (0); + break; + case G_RAID_VOLUME_RL_RAID5: + if (disks < 3) + return (0); + break; + default: + return (0); + } + if (qual != G_RAID_VOLUME_RLQ_NONE) + return (0); + return (1); +} + +static int +g_raid_md_nvidia_start_disk(struct g_raid_disk *disk) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd, *tmpsd; + struct g_raid_disk *olddisk, *tmpdisk; + struct g_raid_md_object *md; + struct g_raid_md_nvidia_object *mdi; + struct g_raid_md_nvidia_perdisk *pd, *oldpd; + struct nvidia_raid_conf *meta; + int disk_pos, resurrection = 0; + + sc = disk->d_softc; + md = sc->sc_md; + mdi = (struct g_raid_md_nvidia_object *)md; + meta = mdi->mdio_meta; + pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data; + olddisk = NULL; + + /* Find disk position in metadata by it's serial. */ + if (pd->pd_meta != NULL) { + disk_pos = pd->pd_meta->disk_number; + if (disk_pos >= meta->total_disks || mdi->mdio_started) + disk_pos = -3; + } else + disk_pos = -3; + /* For RAID0+1 we need to translate order. */ + disk_pos = nvidia_meta_translate_disk(meta, disk_pos); + if (disk_pos < 0) { + G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk"); + /* If we are in the start process, that's all for now. */ + if (!mdi->mdio_started) + goto nofit; + /* + * If we have already started - try to get use of the disk. + * Try to replace OFFLINE disks first, then FAILED. + */ + TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) { + if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE && + tmpdisk->d_state != G_RAID_DISK_S_FAILED) + continue; + /* Make sure this disk is big enough. */ + TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) { + if (sd->sd_offset + sd->sd_size + 2 * 512 > + pd->pd_disk_size) { + G_RAID_DEBUG1(1, sc, + "Disk too small (%ju < %ju)", + pd->pd_disk_size, + sd->sd_offset + sd->sd_size + 512); + break; + } + } + if (sd != NULL) + continue; + if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) { + olddisk = tmpdisk; + break; + } else if (olddisk == NULL) + olddisk = tmpdisk; + } + if (olddisk == NULL) { +nofit: + g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE); + return (1); + } + oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data; + disk_pos = oldpd->pd_disk_pos; + resurrection = 1; + } + + if (olddisk == NULL) { + /* Find placeholder by position. */ + olddisk = g_raid_md_nvidia_get_disk(sc, disk_pos); + if (olddisk == NULL) + panic("No disk at position %d!", disk_pos); + if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) { + G_RAID_DEBUG1(1, sc, "More then one disk for pos %d", + disk_pos); + g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE); + return (0); + } + oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data; + } + + /* Replace failed disk or placeholder with new disk. */ + TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) { + TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next); + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + sd->sd_disk = disk; + } + oldpd->pd_disk_pos = -2; + pd->pd_disk_pos = disk_pos; + + /* If it was placeholder -- destroy it. */ + if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) { + g_raid_destroy_disk(olddisk); + } else { + /* Otherwise, make it STALE_FAILED. */ + g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED); + } + + /* Welcome the new disk. */ + if (resurrection) + g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); + else// if (pd->pd_meta->disk_status == NVIDIA_S_CURRENT || + //pd->pd_meta->disk_status == NVIDIA_S_REBUILD) + g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); +// else +// g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED); + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + + /* + * Different disks may have different sizes, + * in concat mode. Update from real disk size. + */ + if (meta->type == NVIDIA_T_CONCAT) + sd->sd_size = pd->pd_disk_size - 0x800 * 512; + + if (resurrection) { + /* New or ex-spare disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NEW); + } else if (meta->state == NVIDIA_S_REBUILD && + (pd->pd_meta->disk_status & 0x100)) { + /* Rebuilding disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_REBUILD); + sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba / + meta->array_width * pd->pd_meta->sector_size; + } else if (meta->state == NVIDIA_S_SYNC) { + /* Resyncing/dirty disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_RESYNC); + sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba / + meta->array_width * pd->pd_meta->sector_size; + } else { + /* Up to date disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + } + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + } + + /* Update status of our need for spare. */ + if (mdi->mdio_started) { + mdi->mdio_incomplete = + (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < + mdi->mdio_total_disks); + } + + return (resurrection); +} + +static void +g_disk_md_nvidia_retaste(void *arg, int pending) +{ + + G_RAID_DEBUG(1, "Array is not complete, trying to retaste."); + g_retaste(&g_raid_class); + free(arg, M_MD_NVIDIA); +} + +static void +g_raid_md_nvidia_refill(struct g_raid_softc *sc) +{ + struct g_raid_md_object *md; + struct g_raid_md_nvidia_object *mdi; + struct nvidia_raid_conf *meta; + struct g_raid_disk *disk; + struct task *task; + int update, na; + + md = sc->sc_md; + mdi = (struct g_raid_md_nvidia_object *)md; + meta = mdi->mdio_meta; + update = 0; + do { + /* Make sure we miss anything. */ + na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE); + if (na == mdi->mdio_total_disks) + break; + + G_RAID_DEBUG1(1, md->mdo_softc, + "Array is not complete (%d of %d), " + "trying to refill.", na, mdi->mdio_total_disks); + + /* Try to get use some of STALE disks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_STALE) { + update += g_raid_md_nvidia_start_disk(disk); + if (disk->d_state == G_RAID_DISK_S_ACTIVE) + break; + } + } + if (disk != NULL) + continue; + + /* Try to get use some of SPARE disks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_SPARE) { + update += g_raid_md_nvidia_start_disk(disk); + if (disk->d_state == G_RAID_DISK_S_ACTIVE) + break; + } + } + } while (disk != NULL); + + /* Write new metadata if we changed something. */ + if (update) { + g_raid_md_write_nvidia(md, NULL, NULL, NULL); + meta = mdi->mdio_meta; + } + + /* Update status of our need for spare. */ + mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < + mdi->mdio_total_disks); + + /* Request retaste hoping to find spare. */ + if (mdi->mdio_incomplete) { + task = malloc(sizeof(struct task), + M_MD_NVIDIA, M_WAITOK | M_ZERO); + TASK_INIT(task, 0, g_disk_md_nvidia_retaste, task); + taskqueue_enqueue(taskqueue_swi, task); + } +} + +static void +g_raid_md_nvidia_start(struct g_raid_softc *sc) +{ + struct g_raid_md_object *md; + struct g_raid_md_nvidia_object *mdi; + struct g_raid_md_nvidia_perdisk *pd; + struct nvidia_raid_conf *meta; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + off_t size; + int j, disk_pos; + char buf[17]; + + md = sc->sc_md; + mdi = (struct g_raid_md_nvidia_object *)md; + meta = mdi->mdio_meta; + + /* Create volumes and subdisks. */ + nvidia_meta_get_name(meta, buf); + vol = g_raid_create_volume(sc, buf, -1); + vol->v_mediasize = (off_t)meta->total_sectors * 512; + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + if (meta->type == NVIDIA_T_RAID0) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID0; + size = vol->v_mediasize / mdi->mdio_total_disks; + } else if (meta->type == NVIDIA_T_RAID1) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1; + size = vol->v_mediasize; + } else if (meta->type == NVIDIA_T_RAID01) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E; + size = vol->v_mediasize / (mdi->mdio_total_disks / 2); + } else if (meta->type == NVIDIA_T_CONCAT) { + if (mdi->mdio_total_disks == 1) + vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE; + else + vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT; + size = 0; + } else if (meta->type == NVIDIA_T_RAID5) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID5; + size = vol->v_mediasize / (mdi->mdio_total_disks - 1); + } else if (meta->type == NVIDIA_T_RAID5_SYM) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID5; +// vol->v_raid_level_qualifier = 0x03; + size = vol->v_mediasize / (mdi->mdio_total_disks - 1); + } else { + vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; + size = 0; + } + vol->v_strip_size = meta->strip_sectors * 512; //ZZZ + vol->v_disks_count = mdi->mdio_total_disks; + vol->v_sectorsize = 512; //ZZZ + for (j = 0; j < vol->v_disks_count; j++) { + sd = &vol->v_subdisks[j]; + sd->sd_offset = 0; + sd->sd_size = size; + } + g_raid_start_volume(vol); + + /* Create disk placeholders to store data for later writing. */ + for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) { + pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO); + pd->pd_disk_pos = disk_pos; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_state = G_RAID_DISK_S_OFFLINE; + sd = &vol->v_subdisks[disk_pos]; + sd->sd_disk = disk; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + } + + /* Make all disks found till the moment take their places. */ + do { + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_NONE) { + g_raid_md_nvidia_start_disk(disk); + break; + } + } + } while (disk != NULL); + + mdi->mdio_started = 1; + G_RAID_DEBUG1(0, sc, "Array started."); + g_raid_md_write_nvidia(md, NULL, NULL, NULL); + + /* Pickup any STALE/SPARE disks to refill array if needed. */ + g_raid_md_nvidia_refill(sc); + + g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME); + + callout_stop(&mdi->mdio_start_co); + G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount); + root_mount_rel(mdi->mdio_rootmount); + mdi->mdio_rootmount = NULL; +} + +static void +g_raid_md_nvidia_new_disk(struct g_raid_disk *disk) +{ + struct g_raid_softc *sc; + struct g_raid_md_object *md; + struct g_raid_md_nvidia_object *mdi; + struct nvidia_raid_conf *pdmeta; + struct g_raid_md_nvidia_perdisk *pd; + + sc = disk->d_softc; + md = sc->sc_md; + mdi = (struct g_raid_md_nvidia_object *)md; + pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data; + pdmeta = pd->pd_meta; + + if (mdi->mdio_started) { + if (g_raid_md_nvidia_start_disk(disk)) + g_raid_md_write_nvidia(md, NULL, NULL, NULL); + } else { + if (mdi->mdio_meta == NULL || + mdi->mdio_meta->disk_number >= mdi->mdio_meta->total_disks) { + G_RAID_DEBUG1(1, sc, "Newer disk"); + if (mdi->mdio_meta != NULL) + free(mdi->mdio_meta, M_MD_NVIDIA); + mdi->mdio_meta = nvidia_meta_copy(pdmeta); + mdi->mdio_total_disks = pdmeta->total_disks; + mdi->mdio_disks_present = 1; + } else if (pdmeta->disk_number < mdi->mdio_meta->total_disks) { + mdi->mdio_disks_present++; + G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)", + mdi->mdio_disks_present, + mdi->mdio_total_disks); + } else + G_RAID_DEBUG1(1, sc, "Spare disk"); + + /* If we collected all needed disks - start array. */ + if (mdi->mdio_disks_present == mdi->mdio_total_disks) + g_raid_md_nvidia_start(sc); + } +} + +static void +g_raid_nvidia_go(void *arg) +{ + struct g_raid_softc *sc; + struct g_raid_md_object *md; + struct g_raid_md_nvidia_object *mdi; + + sc = arg; + md = sc->sc_md; + mdi = (struct g_raid_md_nvidia_object *)md; + if (!mdi->mdio_started) { + G_RAID_DEBUG1(0, sc, "Force array start due to timeout."); + g_raid_event_send(sc, G_RAID_NODE_E_START, 0); + } +} + +static int +g_raid_md_create_nvidia(struct g_raid_md_object *md, struct g_class *mp, + struct g_geom **gp) +{ + struct g_raid_softc *sc; + struct g_raid_md_nvidia_object *mdi; + char name[32]; + + mdi = (struct g_raid_md_nvidia_object *)md; + arc4rand(&mdi->mdio_volume_id, 16, 0); + snprintf(name, sizeof(name), "NVIDIA-%d", + atomic_fetchadd_int(&NVIDIANodeID, 1)); + sc = g_raid_create_node(mp, name, md); + if (sc == NULL) + return (G_RAID_MD_TASTE_FAIL); + md->mdo_softc = sc; + *gp = sc->sc_geom; + return (G_RAID_MD_TASTE_NEW); +} + +static int +g_raid_md_taste_nvidia(struct g_raid_md_object *md, struct g_class *mp, + struct g_consumer *cp, struct g_geom **gp) +{ + struct g_consumer *rcp; + struct g_provider *pp; + struct g_raid_md_nvidia_object *mdi, *mdi1; + struct g_raid_softc *sc; + struct g_raid_disk *disk; + struct nvidia_raid_conf *meta; + struct g_raid_md_nvidia_perdisk *pd; + struct g_geom *geom; + int error, disk_pos, result, spare, len; + char name[32]; + uint16_t vendor; + + G_RAID_DEBUG(1, "Tasting NVIDIA on %s", cp->provider->name); + mdi = (struct g_raid_md_nvidia_object *)md; + pp = cp->provider; + + /* Read metadata from device. */ + meta = NULL; + spare = 0; + vendor = 0xffff; + disk_pos = 0; + if (g_access(cp, 1, 0, 0) != 0) + return (G_RAID_MD_TASTE_FAIL); + g_topology_unlock(); + len = 2; + if (pp->geom->rank == 1) + g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor); + meta = nvidia_meta_read(cp); + g_topology_lock(); + g_access(cp, -1, 0, 0); + if (meta == NULL) { + if (g_raid_aggressive_spare) { + if (vendor == 0x10de) { + G_RAID_DEBUG(1, + "No NVIDIA metadata, forcing spare."); + spare = 2; + goto search; + } else { + G_RAID_DEBUG(1, + "NVIDIA vendor mismatch 0x%04x != 0x10de", + vendor); + } + } + return (G_RAID_MD_TASTE_FAIL); + } + + /* Check this disk position in obtained metadata. */ + disk_pos = meta->disk_number; + if (disk_pos == -1) { + G_RAID_DEBUG(1, "NVIDIA disk position not found"); + goto fail1; + } + + /* Metadata valid. Print it. */ + g_raid_md_nvidia_print(meta); + G_RAID_DEBUG(1, "NVIDIA disk position %d", disk_pos); + spare = 0;//(meta->type == NVIDIA_T_SPARE) ? 1 : 0; + +search: + /* Search for matching node. */ + sc = NULL; + mdi1 = NULL; + LIST_FOREACH(geom, &mp->geom, geom) { + sc = geom->softc; + if (sc == NULL) + continue; + if (sc->sc_stopping != 0) + continue; + if (sc->sc_md->mdo_class != md->mdo_class) + continue; + mdi1 = (struct g_raid_md_nvidia_object *)sc->sc_md; + if (spare) { + if (mdi1->mdio_incomplete) + break; + } else { + if (memcmp(&mdi1->mdio_volume_id, + &meta->volume_id, 16) == 0) + break; + } + } + + /* Found matching node. */ + if (geom != NULL) { + G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name); + result = G_RAID_MD_TASTE_EXISTING; + + } else if (spare) { /* Not found needy node -- left for later. */ + G_RAID_DEBUG(1, "Spare is not needed at this time"); + goto fail1; + + } else { /* Not found matching node -- create one. */ + result = G_RAID_MD_TASTE_NEW; + memcpy(&mdi->mdio_volume_id, &meta->volume_id, 16); + snprintf(name, sizeof(name), "NVIDIA-%d", + atomic_fetchadd_int(&NVIDIANodeID, 1)); + sc = g_raid_create_node(mp, name, md); + md->mdo_softc = sc; + geom = sc->sc_geom; + callout_init(&mdi->mdio_start_co, 1); + callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz, + g_raid_nvidia_go, sc); + mdi->mdio_rootmount = root_mount_hold("GRAID-NVIDIA"); + G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount); + } + + rcp = g_new_consumer(geom); + g_attach(rcp, pp); + if (g_access(rcp, 1, 1, 1) != 0) + ; //goto fail1; + + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + + pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO); + pd->pd_meta = meta; + if (spare == 2) { + pd->pd_disk_pos = -3; + } else { + pd->pd_disk_pos = -1; + } + pd->pd_disk_size = pp->mediasize; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_consumer = rcp; + rcp->private = disk; + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.", + rcp->provider->name, error); + + g_raid_md_nvidia_new_disk(disk); + + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + *gp = geom; + return (result); +fail1: + free(meta, M_MD_NVIDIA); + return (G_RAID_MD_TASTE_FAIL); +} + +static int +g_raid_md_event_nvidia(struct g_raid_md_object *md, + struct g_raid_disk *disk, u_int event) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd; + struct g_raid_md_nvidia_object *mdi; + struct g_raid_md_nvidia_perdisk *pd; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_nvidia_object *)md; + if (disk == NULL) { + switch (event) { + case G_RAID_NODE_E_START: + if (!mdi->mdio_started) { + /* Bump volume ID to drop missing disks. */ + arc4rand(&mdi->mdio_volume_id, 16, 0); + g_raid_md_nvidia_start(sc); + } + return (0); + } + return (-1); + } + pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data; + switch (event) { + case G_RAID_DISK_E_DISCONNECTED: + /* If disk was assigned, just update statuses. */ + if (pd->pd_disk_pos >= 0) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + if (disk->d_consumer) { + g_raid_kill_consumer(sc, disk->d_consumer); + disk->d_consumer = NULL; + } + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NONE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, + G_RAID_EVENT_SUBDISK); + } + } else { + /* Otherwise -- delete. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); + g_raid_destroy_disk(disk); + } + + if (mdi->mdio_started) { + /* Bump volume ID to prevent disk resurrection. */ + if (pd->pd_disk_pos >= 0) + arc4rand(&mdi->mdio_volume_id, 16, 0); + + /* Write updated metadata to all disks. */ + g_raid_md_write_nvidia(md, NULL, NULL, NULL); + } + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_nvidia_refill(sc); + return (0); + } + return (-2); +} + +static int +g_raid_md_ctl_nvidia(struct g_raid_md_object *md, + struct gctl_req *req) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_nvidia_object *mdi; + struct g_raid_md_nvidia_perdisk *pd; + struct g_consumer *cp; + struct g_provider *pp; + char arg[16]; + const char *verb, *volname, *levelname, *diskname; + int *nargs, *force; + off_t size, sectorsize, strip; + intmax_t *sizearg, *striparg; + int numdisks, i, len, level, qual, update; + int error; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_nvidia_object *)md; + verb = gctl_get_param(req, "verb", NULL); + nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); + error = 0; + if (strcmp(verb, "label") == 0) { + + if (*nargs < 4) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + volname = gctl_get_asciiparam(req, "arg1"); + if (volname == NULL) { + gctl_error(req, "No volume name."); + return (-2); + } + levelname = gctl_get_asciiparam(req, "arg2"); + if (levelname == NULL) { + gctl_error(req, "No RAID level."); + return (-3); + } + if (g_raid_volume_str2level(levelname, &level, &qual)) { + gctl_error(req, "Unknown RAID level '%s'.", levelname); + return (-4); + } + numdisks = *nargs - 3; + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (!g_raid_md_nvidia_supported(level, qual, numdisks, + force ? *force : 0)) { + gctl_error(req, "Unsupported RAID level " + "(0x%02x/0x%02x), or number of disks (%d).", + level, qual, numdisks); + return (-5); + } + + /* Search for disks, connect them and probe. */ + size = 0x7fffffffffffffffllu; + sectorsize = 0; + for (i = 0; i < numdisks; i++) { + snprintf(arg, sizeof(arg), "arg%d", i + 3); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -6; + break; + } + if (strcmp(diskname, "NONE") == 0) { + cp = NULL; + pp = NULL; + } else { + g_topology_lock(); + cp = g_raid_open_consumer(sc, diskname); + if (cp == NULL) { + gctl_error(req, "Can't open '%s'.", + diskname); + g_topology_unlock(); + error = -7; + break; + } + pp = cp->provider; + } + pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO); + pd->pd_disk_pos = i; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_consumer = cp; + if (cp == NULL) + continue; + cp->private = disk; + g_topology_unlock(); + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, + "Dumping not supported by %s.", + cp->provider->name); + + pd->pd_disk_size = pp->mediasize; + if (size > pp->mediasize) + size = pp->mediasize; + if (sectorsize < pp->sectorsize) + sectorsize = pp->sectorsize; + } + if (error != 0) + return (error); + + /* Reserve space for metadata. */ + size -= 2 * sectorsize; + + /* Handle size argument. */ + len = sizeof(*sizearg); + sizearg = gctl_get_param(req, "size", &len); + if (sizearg != NULL && len == sizeof(*sizearg) && + *sizearg > 0) { + if (*sizearg > size) { + gctl_error(req, "Size too big %lld > %lld.", + (long long)*sizearg, (long long)size); + return (-9); + } + size = *sizearg; + } + + /* Handle strip argument. */ + strip = 131072; + len = sizeof(*striparg); + striparg = gctl_get_param(req, "strip", &len); + if (striparg != NULL && len == sizeof(*striparg) && + *striparg > 0) { + if (*striparg < sectorsize) { + gctl_error(req, "Strip size too small."); + return (-10); + } + if (*striparg % sectorsize != 0) { + gctl_error(req, "Incorrect strip size."); + return (-11); + } + if (strip > 65535 * sectorsize) { + gctl_error(req, "Strip size too big."); + return (-12); + } + strip = *striparg; + } + + /* Round size down to strip or sector. */ + if (level == G_RAID_VOLUME_RL_RAID1) + size -= (size % sectorsize); + else if (level == G_RAID_VOLUME_RL_RAID1E && + (numdisks & 1) != 0) + size -= (size % (2 * strip)); + else + size -= (size % strip); + if (size <= 0) { + gctl_error(req, "Size too small."); + return (-13); + } + if (size > 0xffffffffffffllu * sectorsize) { + gctl_error(req, "Size too big."); + return (-14); + } + + /* We have all we need, create things: volume, ... */ + mdi->mdio_total_disks = numdisks; + mdi->mdio_started = 1; + vol = g_raid_create_volume(sc, volname, -1); + vol->v_md_data = (void *)(intptr_t)0; + vol->v_raid_level = level; + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + vol->v_strip_size = strip; + vol->v_disks_count = numdisks; + if (level == G_RAID_VOLUME_RL_RAID0 || + level == G_RAID_VOLUME_RL_CONCAT || + level == G_RAID_VOLUME_RL_SINGLE) + vol->v_mediasize = size * numdisks; + else if (level == G_RAID_VOLUME_RL_RAID1) + vol->v_mediasize = size; + else if (level == G_RAID_VOLUME_RL_RAID5) + vol->v_mediasize = size * (numdisks - 1); + else { /* RAID1E */ + vol->v_mediasize = ((size * numdisks) / strip / 2) * + strip; + } + vol->v_sectorsize = sectorsize; + g_raid_start_volume(vol); + + /* , and subdisks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data; + sd = &vol->v_subdisks[pd->pd_disk_pos]; + sd->sd_disk = disk; + sd->sd_offset = 0; + sd->sd_size = size; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + if (sd->sd_disk->d_consumer != NULL) { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_ACTIVE); + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + } else { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + } + } + + /* Write metadata based on created entities. */ + G_RAID_DEBUG1(0, sc, "Array started."); + g_raid_md_write_nvidia(md, NULL, NULL, NULL); + + /* Pickup any STALE/SPARE disks to refill array if needed. */ + g_raid_md_nvidia_refill(sc); + + g_raid_event_send(vol, G_RAID_VOLUME_E_START, + G_RAID_EVENT_VOLUME); + return (0); + } + if (strcmp(verb, "delete") == 0) { + + /* Check if some volume is still open. */ + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (force != NULL && *force == 0 && + g_raid_nopens(sc) != 0) { + gctl_error(req, "Some volume is still open."); + return (-4); + } + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer) + nvidia_meta_erase(disk->d_consumer); + } + g_raid_destroy_node(sc, 0); + return (0); + } + if (strcmp(verb, "remove") == 0 || + strcmp(verb, "fail") == 0) { + if (*nargs < 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + for (i = 1; i < *nargs; i++) { + snprintf(arg, sizeof(arg), "arg%d", i); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -2; + break; + } + if (strncmp(diskname, "/dev/", 5) == 0) + diskname += 5; + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer != NULL && + disk->d_consumer->provider != NULL && + strcmp(disk->d_consumer->provider->name, + diskname) == 0) + break; + } + if (disk == NULL) { + gctl_error(req, "Disk '%s' not found.", + diskname); + error = -3; + break; + } + + if (strcmp(verb, "fail") == 0) { + g_raid_md_fail_disk_nvidia(md, NULL, disk); + continue; + } + + pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data; + + /* Erase metadata on deleting disk. */ + nvidia_meta_erase(disk->d_consumer); + + /* If disk was assigned, just update statuses. */ + if (pd->pd_disk_pos >= 0) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + if (disk->d_consumer) { + g_raid_kill_consumer(sc, disk->d_consumer); + disk->d_consumer = NULL; + } + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NONE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, + G_RAID_EVENT_SUBDISK); + } + } else { + /* Otherwise -- delete. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); + g_raid_destroy_disk(disk); + } + } + + /* Write updated metadata to remaining disks. */ + g_raid_md_write_nvidia(md, NULL, NULL, NULL); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_nvidia_refill(sc); + return (error); + } + if (strcmp(verb, "insert") == 0) { + if (*nargs < 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + update = 0; + for (i = 1; i < *nargs; i++) { + /* Get disk name. */ + snprintf(arg, sizeof(arg), "arg%d", i); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -3; + break; + } + + /* Try to find provider with specified name. */ + g_topology_lock(); + cp = g_raid_open_consumer(sc, diskname); + if (cp == NULL) { + gctl_error(req, "Can't open disk '%s'.", + diskname); + g_topology_unlock(); + error = -4; + break; + } + pp = cp->provider; + + pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO); + pd->pd_disk_pos = -3; + pd->pd_disk_size = pp->mediasize; + + disk = g_raid_create_disk(sc); + disk->d_consumer = cp; + disk->d_consumer->private = disk; + disk->d_md_data = (void *)pd; + cp->private = disk; + g_topology_unlock(); + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, + "Dumping not supported by %s.", + cp->provider->name); + + /* Welcome the "new" disk. */ + update += g_raid_md_nvidia_start_disk(disk); + if (disk->d_state != G_RAID_DISK_S_SPARE && + disk->d_state != G_RAID_DISK_S_ACTIVE) { + gctl_error(req, "Disk '%s' doesn't fit.", + diskname); + g_raid_destroy_disk(disk); + error = -8; + break; + } + } + + /* Write new metadata if we changed something. */ + if (update) + g_raid_md_write_nvidia(md, NULL, NULL, NULL); + return (error); + } + gctl_error(req, "Command '%s' is not supported.", verb); + return (-100); +} + +static int +g_raid_md_write_nvidia(struct g_raid_md_object *md, struct g_raid_volume *tvol, + struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_nvidia_object *mdi; + struct g_raid_md_nvidia_perdisk *pd; + struct nvidia_raid_conf *meta; + int i, spares; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_nvidia_object *)md; + + if (sc->sc_stopping == G_RAID_DESTROY_HARD) + return (0); + + /* There is only one volume. */ + vol = TAILQ_FIRST(&sc->sc_volumes); + + /* Fill global fields. */ + meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK | M_ZERO); + if (mdi->mdio_meta) + memcpy(meta, mdi->mdio_meta, sizeof(*meta)); + memcpy(meta->nvidia_id, NVIDIA_MAGIC, sizeof(NVIDIA_MAGIC)); + meta->config_size = 30; + meta->version = 0x0064; + meta->total_sectors = vol->v_mediasize / vol->v_sectorsize; + meta->sector_size = vol->v_sectorsize; + nvidia_meta_put_name(meta, vol->v_name); + meta->magic_0 = NVIDIA_MAGIC0; + memcpy(&meta->volume_id, &mdi->mdio_volume_id, 16); + meta->state = NVIDIA_S_IDLE; + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) + meta->array_width = 1; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) + meta->array_width = vol->v_disks_count / 2; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) + meta->array_width = vol->v_disks_count - 1; + else + meta->array_width = vol->v_disks_count; + meta->total_disks = vol->v_disks_count; + meta->orig_array_width = meta->array_width; + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0) + meta->type = NVIDIA_T_RAID0; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) + meta->type = NVIDIA_T_RAID1; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) + meta->type = NVIDIA_T_RAID01; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT || + vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE) + meta->type = NVIDIA_T_CONCAT; +// else if (vol->v_raid_level_qualifier == 0) +// meta->type = NVIDIA_T_RAID5; + else + meta->type = NVIDIA_T_RAID5_SYM; + meta->strip_sectors = vol->v_strip_size / vol->v_sectorsize; + meta->strip_bytes = vol->v_strip_size; + meta->strip_shift = ffs(meta->strip_sectors) - 1; + meta->strip_mask = meta->strip_sectors - 1; + meta->stripe_sectors = meta->strip_sectors * meta->orig_array_width; + meta->stripe_bytes = meta->stripe_sectors * vol->v_sectorsize; + meta->rebuild_lba = 0; + meta->orig_type = meta->type; + meta->orig_total_sectors = meta->total_sectors; + meta->status = 0; + + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if ((sd->sd_state == G_RAID_SUBDISK_S_STALE || + sd->sd_state == G_RAID_SUBDISK_S_RESYNC || + vol->v_dirty) && + meta->state != NVIDIA_S_REBUILD) + meta->state = NVIDIA_S_SYNC; + else if (sd->sd_state == G_RAID_SUBDISK_S_NEW || + sd->sd_state == G_RAID_SUBDISK_S_REBUILD) + meta->state = NVIDIA_S_REBUILD; + } + + /* We are done. Print meta data and store them to disks. */ + if (mdi->mdio_meta != NULL) + free(mdi->mdio_meta, M_MD_NVIDIA); + mdi->mdio_meta = meta; + spares = 0; + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data; + if (disk->d_state != G_RAID_DISK_S_ACTIVE && + disk->d_state != G_RAID_DISK_S_SPARE) + continue; + if (pd->pd_meta != NULL) { + free(pd->pd_meta, M_MD_NVIDIA); + pd->pd_meta = NULL; + } + pd->pd_meta = nvidia_meta_copy(meta); + if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) { + /* For RAID0+1 we need to translate order. */ + pd->pd_meta->disk_number = + nvidia_meta_translate_disk(meta, sd->sd_pos); + if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE) { + pd->pd_meta->disk_status = 0x100; + pd->pd_meta->rebuild_lba = + sd->sd_rebuild_pos / vol->v_sectorsize * + meta->array_width; + } + } else + pd->pd_meta->disk_number = meta->total_disks + spares++; + G_RAID_DEBUG(1, "Writing NVIDIA metadata to %s", + g_raid_get_diskname(disk)); + g_raid_md_nvidia_print(pd->pd_meta); + nvidia_meta_write(disk->d_consumer, pd->pd_meta); + } + return (0); +} + +static int +g_raid_md_fail_disk_nvidia(struct g_raid_md_object *md, + struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) +{ + struct g_raid_softc *sc; + struct g_raid_md_nvidia_object *mdi; + struct g_raid_md_nvidia_perdisk *pd; + struct g_raid_subdisk *sd; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_nvidia_object *)md; + pd = (struct g_raid_md_nvidia_perdisk *)tdisk->d_md_data; + + /* We can't fail disk that is not a part of array now. */ + if (pd->pd_disk_pos < 0) + return (-1); + + /* Erase metadata to prevent disks's later resurrection. */ + if (tdisk->d_consumer) + nvidia_meta_erase(tdisk->d_consumer); + + /* Change states. */ + g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED); + TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_FAILED); + g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED, + G_RAID_EVENT_SUBDISK); + } + + /* Write updated metadata to remaining disks. */ + g_raid_md_write_nvidia(md, NULL, NULL, tdisk); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_nvidia_refill(sc); + return (0); +} + +static int +g_raid_md_free_disk_nvidia(struct g_raid_md_object *md, + struct g_raid_disk *disk) +{ + struct g_raid_md_nvidia_perdisk *pd; + + pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data; + if (pd->pd_meta != NULL) { + free(pd->pd_meta, M_MD_NVIDIA); + pd->pd_meta = NULL; + } + free(pd, M_MD_NVIDIA); + disk->d_md_data = NULL; + return (0); +} + +static int +g_raid_md_free_nvidia(struct g_raid_md_object *md) +{ + struct g_raid_md_nvidia_object *mdi; + + mdi = (struct g_raid_md_nvidia_object *)md; + if (!mdi->mdio_started) { + mdi->mdio_started = 0; + callout_stop(&mdi->mdio_start_co); + G_RAID_DEBUG1(1, md->mdo_softc, + "root_mount_rel %p", mdi->mdio_rootmount); + root_mount_rel(mdi->mdio_rootmount); + mdi->mdio_rootmount = NULL; + } + if (mdi->mdio_meta != NULL) { + free(mdi->mdio_meta, M_MD_NVIDIA); + mdi->mdio_meta = NULL; + } + return (0); +} + +G_RAID_MD_DECLARE(g_raid_md_nvidia); diff --git a/sys/geom/raid/md_promise.c b/sys/geom/raid/md_promise.c new file mode 100644 index 000000000000..b7bf070f9543 --- /dev/null +++ b/sys/geom/raid/md_promise.c @@ -0,0 +1,1940 @@ +/*- + * Copyright (c) 2011 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "geom/raid/g_raid.h" +#include "g_raid_md_if.h" + +static MALLOC_DEFINE(M_MD_PROMISE, "md_promise_data", "GEOM_RAID Promise metadata"); + +#define PROMISE_MAX_DISKS 8 +#define PROMISE_MAX_SUBDISKS 2 +#define PROMISE_META_OFFSET 14 + +struct promise_raid_disk { + uint8_t flags; /* Subdisk status. */ +#define PROMISE_F_VALID 0x01 +#define PROMISE_F_ONLINE 0x02 +#define PROMISE_F_ASSIGNED 0x04 +#define PROMISE_F_SPARE 0x08 +#define PROMISE_F_DUPLICATE 0x10 +#define PROMISE_F_REDIR 0x20 +#define PROMISE_F_DOWN 0x40 +#define PROMISE_F_READY 0x80 + + uint8_t number; /* Position in a volume. */ + uint8_t channel; /* ATA channel number. */ + uint8_t device; /* ATA device number. */ + uint64_t id __packed; /* Subdisk ID. */ +} __packed; + +struct promise_raid_conf { + char promise_id[24]; +#define PROMISE_MAGIC "Promise Technology, Inc." +#define FREEBSD_MAGIC "FreeBSD ATA driver RAID " + + uint32_t dummy_0; + uint64_t magic_0; +#define PROMISE_MAGIC0(x) (((uint64_t)(x.channel) << 48) | \ + ((uint64_t)(x.device != 0) << 56)) + uint16_t magic_1; + uint32_t magic_2; + uint8_t filler1[470]; + + uint32_t integrity; +#define PROMISE_I_VALID 0x00000080 + + struct promise_raid_disk disk; /* This subdisk info. */ + uint32_t disk_offset; /* Subdisk offset. */ + uint32_t disk_sectors; /* Subdisk size */ + uint32_t rebuild_lba; /* Rebuild position. */ + uint16_t generation; /* Generation number. */ + uint8_t status; /* Volume status. */ +#define PROMISE_S_VALID 0x01 +#define PROMISE_S_ONLINE 0x02 +#define PROMISE_S_INITED 0x04 +#define PROMISE_S_READY 0x08 +#define PROMISE_S_DEGRADED 0x10 +#define PROMISE_S_MARKED 0x20 +#define PROMISE_S_MIGRATING 0x40 +#define PROMISE_S_FUNCTIONAL 0x80 + + uint8_t type; /* Voluem type. */ +#define PROMISE_T_RAID0 0x00 +#define PROMISE_T_RAID1 0x01 +#define PROMISE_T_RAID3 0x02 +#define PROMISE_T_RAID5 0x04 +#define PROMISE_T_SPAN 0x08 +#define PROMISE_T_JBOD 0x10 + + uint8_t total_disks; /* Disks in this volume. */ + uint8_t stripe_shift; /* Strip size. */ + uint8_t array_width; /* Number of RAID0 stripes. */ + uint8_t array_number; /* Global volume number. */ + uint32_t total_sectors; /* Volume size. */ + uint16_t cylinders; /* Volume geometry: C. */ + uint8_t heads; /* Volume geometry: H. */ + uint8_t sectors; /* Volume geometry: S. */ + uint64_t volume_id __packed; /* Volume ID, */ + struct promise_raid_disk disks[PROMISE_MAX_DISKS]; + /* Subdisks in this volume. */ + char name[32]; /* Volume label. */ + + uint32_t filler2[8]; + uint32_t magic_3; /* Something related to rebuild. */ + uint64_t rebuild_lba64; /* Per-volume rebuild position. */ + uint32_t magic_4; + uint32_t magic_5; + uint32_t filler3[325]; + uint32_t checksum; +} __packed; + +struct g_raid_md_promise_perdisk { + int pd_updated; + int pd_subdisks; + struct promise_raid_conf *pd_meta[PROMISE_MAX_SUBDISKS]; +}; + +struct g_raid_md_promise_pervolume { + struct promise_raid_conf *pv_meta; + uint64_t pv_id; + uint16_t pv_generation; + int pv_disks_present; + int pv_started; + struct callout pv_start_co; /* STARTING state timer. */ +}; + +static g_raid_md_create_t g_raid_md_create_promise; +static g_raid_md_taste_t g_raid_md_taste_promise; +static g_raid_md_event_t g_raid_md_event_promise; +static g_raid_md_volume_event_t g_raid_md_volume_event_promise; +static g_raid_md_ctl_t g_raid_md_ctl_promise; +static g_raid_md_write_t g_raid_md_write_promise; +static g_raid_md_fail_disk_t g_raid_md_fail_disk_promise; +static g_raid_md_free_disk_t g_raid_md_free_disk_promise; +static g_raid_md_free_volume_t g_raid_md_free_volume_promise; +static g_raid_md_free_t g_raid_md_free_promise; + +static kobj_method_t g_raid_md_promise_methods[] = { + KOBJMETHOD(g_raid_md_create, g_raid_md_create_promise), + KOBJMETHOD(g_raid_md_taste, g_raid_md_taste_promise), + KOBJMETHOD(g_raid_md_event, g_raid_md_event_promise), + KOBJMETHOD(g_raid_md_volume_event, g_raid_md_volume_event_promise), + KOBJMETHOD(g_raid_md_ctl, g_raid_md_ctl_promise), + KOBJMETHOD(g_raid_md_write, g_raid_md_write_promise), + KOBJMETHOD(g_raid_md_fail_disk, g_raid_md_fail_disk_promise), + KOBJMETHOD(g_raid_md_free_disk, g_raid_md_free_disk_promise), + KOBJMETHOD(g_raid_md_free_volume, g_raid_md_free_volume_promise), + KOBJMETHOD(g_raid_md_free, g_raid_md_free_promise), + { 0, 0 } +}; + +static struct g_raid_md_class g_raid_md_promise_class = { + "Promise", + g_raid_md_promise_methods, + sizeof(struct g_raid_md_object), + .mdc_priority = 100 +}; + + +static void +g_raid_md_promise_print(struct promise_raid_conf *meta) +{ + int i; + + if (g_raid_debug < 1) + return; + + printf("********* ATA Promise Metadata *********\n"); + printf("promise_id <%.24s>\n", meta->promise_id); + printf("disk %02x %02x %02x %02x %016jx\n", + meta->disk.flags, meta->disk.number, meta->disk.channel, + meta->disk.device, meta->disk.id); + printf("disk_offset %u\n", meta->disk_offset); + printf("disk_sectors %u\n", meta->disk_sectors); + printf("rebuild_lba %u\n", meta->rebuild_lba); + printf("generation %u\n", meta->generation); + printf("status 0x%02x\n", meta->status); + printf("type %u\n", meta->type); + printf("total_disks %u\n", meta->total_disks); + printf("stripe_shift %u\n", meta->stripe_shift); + printf("array_width %u\n", meta->array_width); + printf("array_number %u\n", meta->array_number); + printf("total_sectors %u\n", meta->total_sectors); + printf("cylinders %u\n", meta->cylinders); + printf("heads %u\n", meta->heads); + printf("sectors %u\n", meta->sectors); + printf("volume_id 0x%016jx\n", meta->volume_id); + printf("disks:\n"); + for (i = 0; i < PROMISE_MAX_DISKS; i++ ) { + printf(" %02x %02x %02x %02x %016jx\n", + meta->disks[i].flags, meta->disks[i].number, + meta->disks[i].channel, meta->disks[i].device, + meta->disks[i].id); + } + printf("name <%.32s>\n", meta->name); + printf("magic_3 0x%08x\n", meta->magic_3); + printf("rebuild_lba64 %ju\n", meta->rebuild_lba64); + printf("magic_4 0x%08x\n", meta->magic_4); + printf("magic_5 0x%08x\n", meta->magic_5); + printf("=================================================\n"); +} + +static struct promise_raid_conf * +promise_meta_copy(struct promise_raid_conf *meta) +{ + struct promise_raid_conf *nmeta; + + nmeta = malloc(sizeof(*nmeta), M_MD_PROMISE, M_WAITOK); + memcpy(nmeta, meta, sizeof(*nmeta)); + return (nmeta); +} + +static int +promise_meta_find_disk(struct promise_raid_conf *meta, uint64_t id) +{ + int pos; + + for (pos = 0; pos < meta->total_disks; pos++) { + if (meta->disks[pos].id == id) + return (pos); + } + return (-1); +} + +static int +promise_meta_unused_range(struct promise_raid_conf **metaarr, int nsd, + uint32_t sectors, uint32_t *off, uint32_t *size) +{ + uint32_t coff, csize; + int i, j; + + sectors -= 131072; + *off = 0; + *size = 0; + coff = 0; + csize = sectors; + i = 0; + while (1) { + for (j = 0; j < nsd; j++) { + if (metaarr[j]->disk_offset >= coff) { + csize = MIN(csize, + metaarr[j]->disk_offset - coff); + } + } + if (csize > *size) { + *off = coff; + *size = csize; + } + if (i >= nsd) + break; + coff = metaarr[i]->disk_offset + metaarr[i]->disk_sectors; + csize = sectors - coff; + i++; + }; + return ((*size > 0) ? 1 : 0); +} + +static int +promise_meta_translate_disk(struct g_raid_volume *vol, int md_disk_pos) +{ + int disk_pos, width; + + if (md_disk_pos >= 0 && vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) { + width = vol->v_disks_count / 2; + disk_pos = (md_disk_pos / width) + + (md_disk_pos % width) * width; + } else + disk_pos = md_disk_pos; + return (disk_pos); +} + +static void +promise_meta_get_name(struct promise_raid_conf *meta, char *buf) +{ + int i; + + strncpy(buf, meta->name, 32); + buf[32] = 0; + for (i = 31; i >= 0; i--) { + if (buf[i] > 0x20) + break; + buf[i] = 0; + } +} + +static void +promise_meta_put_name(struct promise_raid_conf *meta, char *buf) +{ + + memset(meta->name, 0x20, 32); + memcpy(meta->name, buf, MIN(strlen(buf), 32)); +} + +static int +promise_meta_read(struct g_consumer *cp, struct promise_raid_conf **metaarr) +{ + struct g_provider *pp; + struct promise_raid_conf *meta; + char *buf; + int error, i, subdisks; + uint32_t checksum, *ptr; + + pp = cp->provider; + subdisks = 0; +next: + /* Read metadata block. */ + buf = g_read_data(cp, pp->mediasize - pp->sectorsize * + (63 - subdisks * PROMISE_META_OFFSET), + pp->sectorsize * 4, &error); + if (buf == NULL) { + G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).", + pp->name, error); + return (subdisks); + } + meta = (struct promise_raid_conf *)buf; + + /* Check if this is an Promise RAID struct */ + if (strncmp(meta->promise_id, PROMISE_MAGIC, strlen(PROMISE_MAGIC)) && + strncmp(meta->promise_id, FREEBSD_MAGIC, strlen(FREEBSD_MAGIC))) { + if (subdisks == 0) + G_RAID_DEBUG(1, + "Promise signature check failed on %s", pp->name); + g_free(buf); + return (subdisks); + } + meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK); + memcpy(meta, buf, MIN(sizeof(*meta), pp->sectorsize * 4)); + g_free(buf); + + /* Check metadata checksum. */ + for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++) + checksum += *ptr++; + if (checksum != meta->checksum) { + G_RAID_DEBUG(1, "Promise checksum check failed on %s", pp->name); + free(meta, M_MD_PROMISE); + return (subdisks); + } + + if ((meta->integrity & PROMISE_I_VALID) == 0) { + G_RAID_DEBUG(1, "Promise metadata is invalid on %s", pp->name); + free(meta, M_MD_PROMISE); + return (subdisks); + } + + if (meta->total_disks > PROMISE_MAX_DISKS) { + G_RAID_DEBUG(1, "Wrong number of disks on %s (%d)", + pp->name, meta->total_disks); + free(meta, M_MD_PROMISE); + return (subdisks); + } + + /* Save this part and look for next. */ + *metaarr = meta; + metaarr++; + subdisks++; + if (subdisks < PROMISE_MAX_SUBDISKS) + goto next; + + return (subdisks); +} + +static int +promise_meta_write(struct g_consumer *cp, + struct promise_raid_conf **metaarr, int nsd) +{ + struct g_provider *pp; + struct promise_raid_conf *meta; + char *buf; + int error, i, subdisk, fake; + uint32_t checksum, *ptr, off, size; + + pp = cp->provider; + subdisk = 0; + fake = 0; +next: + buf = malloc(pp->sectorsize * 4, M_MD_PROMISE, M_WAITOK | M_ZERO); + meta = NULL; + if (subdisk < nsd) { + meta = metaarr[subdisk]; + } else if (!fake && promise_meta_unused_range(metaarr, nsd, + cp->provider->mediasize / cp->provider->sectorsize, + &off, &size)) { + /* Optionally add record for unused space. */ + meta = (struct promise_raid_conf *)buf; + memcpy(&meta->promise_id[0], PROMISE_MAGIC, sizeof(PROMISE_MAGIC)); + meta->dummy_0 = 0x00020000; + meta->integrity = PROMISE_I_VALID; + meta->disk.flags = PROMISE_F_ONLINE | PROMISE_F_VALID; + meta->disk.number = 0xff; + arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0); + meta->disk_offset = off; + meta->disk_sectors = size; + meta->rebuild_lba = UINT32_MAX; + fake = 1; + } + if (meta != NULL) { + /* Recalculate checksum for case if metadata were changed. */ + meta->checksum = 0; + for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++) + checksum += *ptr++; + meta->checksum = checksum; + memcpy(buf, meta, MIN(pp->sectorsize * 4, sizeof(*meta))); + } + error = g_write_data(cp, pp->mediasize - pp->sectorsize * + (63 - subdisk * PROMISE_META_OFFSET), + buf, pp->sectorsize * 4); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).", + pp->name, error); + } + free(buf, M_MD_PROMISE); + + subdisk++; + if (subdisk < PROMISE_MAX_SUBDISKS) + goto next; + + return (error); +} + +static int +promise_meta_erase(struct g_consumer *cp) +{ + struct g_provider *pp; + char *buf; + int error, subdisk; + + pp = cp->provider; + buf = malloc(4 * pp->sectorsize, M_MD_PROMISE, M_WAITOK | M_ZERO); + for (subdisk = 0; subdisk < PROMISE_MAX_SUBDISKS; subdisk++) { + error = g_write_data(cp, pp->mediasize - pp->sectorsize * + (63 - subdisk * PROMISE_META_OFFSET), + buf, 4 * pp->sectorsize); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).", + pp->name, error); + } + } + free(buf, M_MD_PROMISE); + return (error); +} + +static int +promise_meta_write_spare(struct g_consumer *cp) +{ + struct promise_raid_conf *meta; + int error; + + meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO); + memcpy(&meta->promise_id[0], PROMISE_MAGIC, sizeof(PROMISE_MAGIC)); + meta->dummy_0 = 0x00020000; + meta->integrity = PROMISE_I_VALID; + meta->disk.flags = PROMISE_F_SPARE | PROMISE_F_ONLINE | PROMISE_F_VALID; + meta->disk.number = 0xff; + arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0); + meta->disk_sectors = cp->provider->mediasize / cp->provider->sectorsize; + meta->disk_sectors -= 131072; + meta->rebuild_lba = UINT32_MAX; + error = promise_meta_write(cp, &meta, 1); + free(meta, M_MD_PROMISE); + return (error); +} + +static struct g_raid_volume * +g_raid_md_promise_get_volume(struct g_raid_softc *sc, uint64_t id) +{ + struct g_raid_volume *vol; + struct g_raid_md_promise_pervolume *pv; + + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + pv = vol->v_md_data; + if (pv->pv_id == id) + break; + } + return (vol); +} + +static int +g_raid_md_promise_purge_volumes(struct g_raid_softc *sc) +{ + struct g_raid_volume *vol, *tvol; + struct g_raid_md_promise_pervolume *pv; + int i, res; + + res = 0; + TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tvol) { + pv = vol->v_md_data; + if (!pv->pv_started || vol->v_stopping) + continue; + for (i = 0; i < vol->v_disks_count; i++) { + if (vol->v_subdisks[i].sd_state != G_RAID_SUBDISK_S_NONE) + break; + } + if (i >= vol->v_disks_count) { + g_raid_destroy_volume(vol); + res = 1; + } + } + return (res); +} + +static int +g_raid_md_promise_purge_disks(struct g_raid_softc *sc) +{ + struct g_raid_disk *disk, *tdisk; + struct g_raid_volume *vol; + struct g_raid_md_promise_perdisk *pd; + int i, j, res; + + res = 0; + TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) { + if (disk->d_state == G_RAID_DISK_S_SPARE) + continue; + pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data; + + /* Scan for deleted volumes. */ + for (i = 0; i < pd->pd_subdisks; ) { + vol = g_raid_md_promise_get_volume(sc, + pd->pd_meta[i]->volume_id); + if (vol != NULL && !vol->v_stopping) { + i++; + continue; + } + free(pd->pd_meta[i], M_MD_PROMISE); + for (j = i; j < pd->pd_subdisks - 1; j++) + pd->pd_meta[j] = pd->pd_meta[j + 1]; + pd->pd_meta[PROMISE_MAX_SUBDISKS - 1] = NULL; + pd->pd_subdisks--; + pd->pd_updated = 1; + } + + /* If there is no metadata left - erase and delete disk. */ + if (pd->pd_subdisks == 0) { + promise_meta_erase(disk->d_consumer); + g_raid_destroy_disk(disk); + res = 1; + } + } + return (res); +} + +static int +g_raid_md_promise_supported(int level, int qual, int disks, int force) +{ + + if (disks > PROMISE_MAX_DISKS) + return (0); + switch (level) { + case G_RAID_VOLUME_RL_RAID0: + if (disks < 1) + return (0); + if (!force && disks < 2) + return (0); + break; + case G_RAID_VOLUME_RL_RAID1: + if (disks < 1) + return (0); + if (!force && (disks != 2)) + return (0); + break; + case G_RAID_VOLUME_RL_RAID1E: + if (disks < 2) + return (0); + if (disks % 2 != 0) + return (0); + if (!force && (disks != 4)) + return (0); + break; + case G_RAID_VOLUME_RL_SINGLE: + if (disks != 1) + return (0); + break; + case G_RAID_VOLUME_RL_CONCAT: + if (disks < 2) + return (0); + break; + case G_RAID_VOLUME_RL_RAID5: + if (disks < 3) + return (0); + break; + default: + return (0); + } + if (qual != G_RAID_VOLUME_RLQ_NONE) + return (0); + return (1); +} + +static int +g_raid_md_promise_start_disk(struct g_raid_disk *disk, int sdn, + struct g_raid_volume *vol) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd; + struct g_raid_md_promise_perdisk *pd; + struct g_raid_md_promise_pervolume *pv; + struct promise_raid_conf *meta; + off_t size; + int disk_pos, md_disk_pos, i, resurrection = 0; + uint32_t eoff, esize; + + sc = disk->d_softc; + pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data; + + pv = vol->v_md_data; + meta = pv->pv_meta; + + if (sdn >= 0) { + /* Find disk position in metadata by it's serial. */ + md_disk_pos = promise_meta_find_disk(meta, pd->pd_meta[sdn]->disk.id); + /* For RAID0+1 we need to translate order. */ + disk_pos = promise_meta_translate_disk(vol, md_disk_pos); + } else { + md_disk_pos = -1; + disk_pos = -1; + } + if (disk_pos < 0) { + G_RAID_DEBUG1(1, sc, "Disk %s is not part of the volume %s", + g_raid_get_diskname(disk), vol->v_name); + /* Failed stale disk is useless for us. */ + if (sdn >= 0 && + pd->pd_meta[sdn]->disk.flags & PROMISE_F_DOWN) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED); + return (0); + } + /* If we were given specific metadata subdisk - erase it. */ + if (sdn >= 0) { + free(pd->pd_meta[sdn], M_MD_PROMISE); + for (i = sdn; i < pd->pd_subdisks - 1; i++) + pd->pd_meta[i] = pd->pd_meta[i + 1]; + pd->pd_meta[PROMISE_MAX_SUBDISKS - 1] = NULL; + pd->pd_subdisks--; + } + /* If we are in the start process, that's all for now. */ + if (!pv->pv_started) + goto nofit; + /* + * If we have already started - try to get use of the disk. + * Try to replace OFFLINE disks first, then FAILED. + */ + promise_meta_unused_range(pd->pd_meta, pd->pd_subdisks, + disk->d_consumer->provider->mediasize / + disk->d_consumer->provider->sectorsize, + &eoff, &esize); + if (esize == 0) { + G_RAID_DEBUG1(1, sc, "No free space on disk %s", + g_raid_get_diskname(disk)); + goto nofit; + } + size = INT64_MAX; + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if (sd->sd_state != G_RAID_SUBDISK_S_NONE) + size = sd->sd_size; + if (sd->sd_state <= G_RAID_SUBDISK_S_FAILED && + (disk_pos < 0 || + vol->v_subdisks[i].sd_state < sd->sd_state)) + disk_pos = i; + } + if (disk_pos >= 0 && + vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT && + (off_t)esize * 512 < size) { + G_RAID_DEBUG1(1, sc, "Disk %s free space " + "is too small (%ju < %ju)", + g_raid_get_diskname(disk), + (off_t)esize * 512, size); + disk_pos = -1; + } + if (disk_pos >= 0) { + if (vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT) + esize = size / 512; + /* For RAID0+1 we need to translate order. */ + md_disk_pos = promise_meta_translate_disk(vol, disk_pos); + } else { +nofit: + if (pd->pd_subdisks == 0) { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_SPARE); + } + return (0); + } + G_RAID_DEBUG1(1, sc, "Disk %s takes pos %d in the volume %s", + g_raid_get_diskname(disk), disk_pos, vol->v_name); + resurrection = 1; + } + + sd = &vol->v_subdisks[disk_pos]; + + if (resurrection && sd->sd_disk != NULL) { + g_raid_change_disk_state(sd->sd_disk, + G_RAID_DISK_S_STALE_FAILED); + TAILQ_REMOVE(&sd->sd_disk->d_subdisks, + sd, sd_next); + } + vol->v_subdisks[disk_pos].sd_disk = disk; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + + /* Welcome the new disk. */ + if (resurrection) + g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); + else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN) + g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED); + else + g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); + + if (resurrection) { + sd->sd_offset = (off_t)eoff * 512; + sd->sd_size = (off_t)esize * 512; + } else { + sd->sd_offset = (off_t)pd->pd_meta[sdn]->disk_offset * 512; + sd->sd_size = (off_t)pd->pd_meta[sdn]->disk_sectors * 512; + } + + if (resurrection) { + /* Stale disk, almost same as new. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NEW); + } else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN) { + /* Failed disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_FAILED); + } else if (meta->disks[md_disk_pos].flags & PROMISE_F_REDIR) { + /* Rebuilding disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_REBUILD); + if (pd->pd_meta[sdn]->generation != meta->generation) + sd->sd_rebuild_pos = 0; + else { + sd->sd_rebuild_pos = + (off_t)pd->pd_meta[sdn]->rebuild_lba * 512; + } + } else if (!(meta->disks[md_disk_pos].flags & PROMISE_F_ONLINE)) { + /* Rebuilding disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NEW); + } else if (pd->pd_meta[sdn]->generation != meta->generation || + (meta->status & PROMISE_S_MARKED)) { + /* Stale disk or dirty volume (unclean shutdown). */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_STALE); + } else { + /* Up to date disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + } + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + + return (resurrection); +} + +static void +g_raid_md_promise_refill(struct g_raid_softc *sc) +{ + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_object *md; + struct g_raid_md_promise_perdisk *pd; + struct g_raid_md_promise_pervolume *pv; + int update, updated, i, bad; + + md = sc->sc_md; +restart: + updated = 0; + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + pv = vol->v_md_data; + if (!pv->pv_started || vol->v_stopping) + continue; + + /* Search for subdisk that needs replacement. */ + bad = 0; + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if (sd->sd_state == G_RAID_SUBDISK_S_NONE || + sd->sd_state == G_RAID_SUBDISK_S_FAILED) + bad = 1; + } + if (!bad) + continue; + + G_RAID_DEBUG1(1, sc, "Volume %s is not complete, " + "trying to refill.", vol->v_name); + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + /* Skip failed. */ + if (disk->d_state < G_RAID_DISK_S_SPARE) + continue; + /* Skip already used by this volume. */ + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if (sd->sd_disk == disk) + break; + } + if (i < vol->v_disks_count) + continue; + + /* Try to use disk if it has empty extents. */ + pd = disk->d_md_data; + if (pd->pd_subdisks < PROMISE_MAX_SUBDISKS) { + update = + g_raid_md_promise_start_disk(disk, -1, vol); + } else + update = 0; + if (update) { + g_raid_md_write_promise(md, vol, NULL, disk); + break; + } + updated += update; + } + } + if (updated) + goto restart; +} + +static void +g_raid_md_promise_start(struct g_raid_volume *vol) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_object *md; + struct g_raid_md_promise_perdisk *pd; + struct g_raid_md_promise_pervolume *pv; + struct promise_raid_conf *meta; + int i; + + sc = vol->v_softc; + md = sc->sc_md; + pv = vol->v_md_data; + meta = pv->pv_meta; + + if (meta->type == PROMISE_T_RAID0) + vol->v_raid_level = G_RAID_VOLUME_RL_RAID0; + else if (meta->type == PROMISE_T_RAID1) { + if (meta->array_width == 1) + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1; + else + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E; + } else if (meta->type == PROMISE_T_RAID3) + vol->v_raid_level = G_RAID_VOLUME_RL_RAID3; + else if (meta->type == PROMISE_T_RAID5) + vol->v_raid_level = G_RAID_VOLUME_RL_RAID5; + else if (meta->type == PROMISE_T_SPAN) + vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT; + else if (meta->type == PROMISE_T_JBOD) + vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE; + else + vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + vol->v_strip_size = 512 << meta->stripe_shift; //ZZZ + vol->v_disks_count = meta->total_disks; + vol->v_mediasize = (off_t)meta->total_sectors * 512; //ZZZ + vol->v_sectorsize = 512; //ZZZ + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + sd->sd_offset = (off_t)meta->disk_offset * 512; //ZZZ + sd->sd_size = (off_t)meta->disk_sectors * 512; //ZZZ + } + g_raid_start_volume(vol); + + /* Make all disks found till the moment take their places. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = disk->d_md_data; + for (i = 0; i < pd->pd_subdisks; i++) { + if (pd->pd_meta[i]->volume_id == meta->volume_id) + g_raid_md_promise_start_disk(disk, i, vol); + } + } + + pv->pv_started = 1; + callout_stop(&pv->pv_start_co); + G_RAID_DEBUG1(0, sc, "Volume started."); + g_raid_md_write_promise(md, vol, NULL, NULL); + + /* Pickup any STALE/SPARE disks to refill array if needed. */ + g_raid_md_promise_refill(sc); + + g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME); +} + +static void +g_raid_promise_go(void *arg) +{ + struct g_raid_volume *vol; + struct g_raid_softc *sc; + struct g_raid_md_promise_pervolume *pv; + + vol = arg; + pv = vol->v_md_data; + sc = vol->v_softc; + if (!pv->pv_started) { + G_RAID_DEBUG1(0, sc, "Force volume start due to timeout."); + g_raid_event_send(vol, G_RAID_VOLUME_E_STARTMD, + G_RAID_EVENT_VOLUME); + } +} + +static void +g_raid_md_promise_new_disk(struct g_raid_disk *disk) +{ + struct g_raid_softc *sc; + struct g_raid_md_object *md; + struct promise_raid_conf *pdmeta; + struct g_raid_md_promise_perdisk *pd; + struct g_raid_md_promise_pervolume *pv; + struct g_raid_volume *vol; + int i; + char buf[33]; + + sc = disk->d_softc; + md = sc->sc_md; + pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data; + + if (pd->pd_subdisks == 0) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE); + g_raid_md_promise_refill(sc); + return; + } + + for (i = 0; i < pd->pd_subdisks; i++) { + pdmeta = pd->pd_meta[i]; + + /* Look for volume with matching ID. */ + vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id); + if (vol == NULL) { + promise_meta_get_name(pdmeta, buf); + vol = g_raid_create_volume(sc, buf, pdmeta->array_number); + pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO); + pv->pv_id = pdmeta->volume_id; + vol->v_md_data = pv; + callout_init(&pv->pv_start_co, 1); + callout_reset(&pv->pv_start_co, + g_raid_start_timeout * hz, + g_raid_promise_go, vol); + } else + pv = vol->v_md_data; + + /* If we haven't started yet - check metadata freshness. */ + if (pv->pv_meta == NULL || !pv->pv_started) { + if (pv->pv_meta == NULL || + ((int16_t)(pdmeta->generation - pv->pv_generation)) > 0) { + G_RAID_DEBUG1(1, sc, "Newer disk"); + if (pv->pv_meta != NULL) + free(pv->pv_meta, M_MD_PROMISE); + pv->pv_meta = promise_meta_copy(pdmeta); + pv->pv_generation = pv->pv_meta->generation; + pv->pv_disks_present = 1; + } else if (pdmeta->generation == pv->pv_generation) { + pv->pv_disks_present++; + G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)", + pv->pv_disks_present, + pv->pv_meta->total_disks); + } else { + G_RAID_DEBUG1(1, sc, "Older disk"); + } + } + } + + for (i = 0; i < pd->pd_subdisks; i++) { + pdmeta = pd->pd_meta[i]; + + /* Look for volume with matching ID. */ + vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id); + if (vol == NULL) + continue; + pv = vol->v_md_data; + + if (pv->pv_started) { + if (g_raid_md_promise_start_disk(disk, i, vol)) + g_raid_md_write_promise(md, vol, NULL, NULL); + } else { + /* If we collected all needed disks - start array. */ + if (pv->pv_disks_present == pv->pv_meta->total_disks) + g_raid_md_promise_start(vol); + } + } +} + +static int +g_raid_md_create_promise(struct g_raid_md_object *md, struct g_class *mp, + struct g_geom **gp) +{ + struct g_geom *geom; + struct g_raid_softc *sc; + + /* Search for existing node. */ + LIST_FOREACH(geom, &mp->geom, geom) { + sc = geom->softc; + if (sc == NULL) + continue; + if (sc->sc_stopping != 0) + continue; + if (sc->sc_md->mdo_class != md->mdo_class) + continue; + break; + } + if (geom != NULL) { + *gp = geom; + return (G_RAID_MD_TASTE_EXISTING); + } + + /* Create new one if not found. */ + sc = g_raid_create_node(mp, "Promise", md); + if (sc == NULL) + return (G_RAID_MD_TASTE_FAIL); + md->mdo_softc = sc; + *gp = sc->sc_geom; + return (G_RAID_MD_TASTE_NEW); +} + +static int +g_raid_md_taste_promise(struct g_raid_md_object *md, struct g_class *mp, + struct g_consumer *cp, struct g_geom **gp) +{ + struct g_consumer *rcp; + struct g_provider *pp; + struct g_raid_softc *sc; + struct g_raid_disk *disk; + struct promise_raid_conf *meta, *metaarr[4]; + struct g_raid_md_promise_perdisk *pd; + struct g_geom *geom; + int error, i, j, result, len, subdisks; + char name[16]; + uint16_t vendor; + + G_RAID_DEBUG(1, "Tasting Promise on %s", cp->provider->name); + pp = cp->provider; + + /* Read metadata from device. */ + meta = NULL; + vendor = 0xffff; + if (g_access(cp, 1, 0, 0) != 0) + return (G_RAID_MD_TASTE_FAIL); + g_topology_unlock(); + len = 2; + if (pp->geom->rank == 1) + g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor); + subdisks = promise_meta_read(cp, metaarr); + g_topology_lock(); + g_access(cp, -1, 0, 0); + if (subdisks == 0) { + if (g_raid_aggressive_spare) { + if (vendor == 0x105a || vendor == 0x1002) { + G_RAID_DEBUG(1, + "No Promise metadata, forcing spare."); + goto search; + } else { + G_RAID_DEBUG(1, + "Promise/ATI vendor mismatch " + "0x%04x != 0x105a/0x1002", + vendor); + } + } + return (G_RAID_MD_TASTE_FAIL); + } + + /* Metadata valid. Print it. */ + for (i = 0; i < subdisks; i++) + g_raid_md_promise_print(metaarr[i]); + + /* Purge meaningless (empty/spare) records. */ + for (i = 0; i < subdisks; ) { + if (metaarr[i]->disk.flags & PROMISE_F_ASSIGNED) { + i++; + continue; + } + free(metaarr[i], M_MD_PROMISE); + for (j = i; j < subdisks - 1; j++) + metaarr[i] = metaarr[j + 1]; + metaarr[PROMISE_MAX_SUBDISKS - 1] = NULL; + subdisks--; + } + +search: + /* Search for matching node. */ + sc = NULL; + LIST_FOREACH(geom, &mp->geom, geom) { + sc = geom->softc; + if (sc == NULL) + continue; + if (sc->sc_stopping != 0) + continue; + if (sc->sc_md->mdo_class != md->mdo_class) + continue; + break; + } + + /* Found matching node. */ + if (geom != NULL) { + G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name); + result = G_RAID_MD_TASTE_EXISTING; + + } else { /* Not found matching node -- create one. */ + result = G_RAID_MD_TASTE_NEW; + snprintf(name, sizeof(name), "Promise"); + sc = g_raid_create_node(mp, name, md); + md->mdo_softc = sc; + geom = sc->sc_geom; + } + + rcp = g_new_consumer(geom); + g_attach(rcp, pp); + if (g_access(rcp, 1, 1, 1) != 0) + ; //goto fail1; + + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + + pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO); + pd->pd_subdisks = subdisks; + for (i = 0; i < subdisks; i++) + pd->pd_meta[i] = metaarr[i]; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_consumer = rcp; + rcp->private = disk; + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.", + rcp->provider->name, error); + + g_raid_md_promise_new_disk(disk); + + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + *gp = geom; + return (result); +} + +static int +g_raid_md_event_promise(struct g_raid_md_object *md, + struct g_raid_disk *disk, u_int event) +{ + struct g_raid_softc *sc; + struct g_raid_md_promise_perdisk *pd; + + sc = md->mdo_softc; + if (disk == NULL) + return (-1); + pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data; + switch (event) { + case G_RAID_DISK_E_DISCONNECTED: + /* Delete disk. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); + g_raid_destroy_disk(disk); + g_raid_md_promise_purge_volumes(sc); + + /* Write updated metadata to all disks. */ + g_raid_md_write_promise(md, NULL, NULL, NULL); + + /* Check if anything left. */ + if (g_raid_ndisks(sc, -1) == 0) + g_raid_destroy_node(sc, 0); + else + g_raid_md_promise_refill(sc); + return (0); + } + return (-2); +} + +static int +g_raid_md_volume_event_promise(struct g_raid_md_object *md, + struct g_raid_volume *vol, u_int event) +{ + struct g_raid_softc *sc; + struct g_raid_md_promise_pervolume *pv; + + sc = md->mdo_softc; + pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data; + switch (event) { + case G_RAID_VOLUME_E_STARTMD: + if (!pv->pv_started) + g_raid_md_promise_start(vol); + return (0); + } + return (-2); +} + +static int +g_raid_md_ctl_promise(struct g_raid_md_object *md, + struct gctl_req *req) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol, *vol1; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk, *disks[PROMISE_MAX_DISKS]; + struct g_raid_md_promise_perdisk *pd; + struct g_raid_md_promise_pervolume *pv; + struct g_consumer *cp; + struct g_provider *pp; + char arg[16]; + const char *verb, *volname, *levelname, *diskname; + char *tmp; + int *nargs, *force; + off_t size, sectorsize, strip; + intmax_t *sizearg, *striparg; + uint32_t offs[PROMISE_MAX_DISKS], esize; + int numdisks, i, len, level, qual; + int error; + + sc = md->mdo_softc; + verb = gctl_get_param(req, "verb", NULL); + nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); + error = 0; + if (strcmp(verb, "label") == 0) { + + if (*nargs < 4) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + volname = gctl_get_asciiparam(req, "arg1"); + if (volname == NULL) { + gctl_error(req, "No volume name."); + return (-2); + } + levelname = gctl_get_asciiparam(req, "arg2"); + if (levelname == NULL) { + gctl_error(req, "No RAID level."); + return (-3); + } + if (g_raid_volume_str2level(levelname, &level, &qual)) { + gctl_error(req, "Unknown RAID level '%s'.", levelname); + return (-4); + } + numdisks = *nargs - 3; + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (!g_raid_md_promise_supported(level, qual, numdisks, + force ? *force : 0)) { + gctl_error(req, "Unsupported RAID level " + "(0x%02x/0x%02x), or number of disks (%d).", + level, qual, numdisks); + return (-5); + } + + /* Search for disks, connect them and probe. */ + size = INT64_MAX; + sectorsize = 0; + bzero(disks, sizeof(disks)); + bzero(offs, sizeof(offs)); + for (i = 0; i < numdisks; i++) { + snprintf(arg, sizeof(arg), "arg%d", i + 3); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -6; + break; + } + if (strcmp(diskname, "NONE") == 0) + continue; + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer != NULL && + disk->d_consumer->provider != NULL && + strcmp(disk->d_consumer->provider->name, + diskname) == 0) + break; + } + if (disk != NULL) { + if (disk->d_state != G_RAID_DISK_S_ACTIVE) { + gctl_error(req, "Disk '%s' is in a " + "wrong state (%s).", diskname, + g_raid_disk_state2str(disk->d_state)); + error = -7; + break; + } + pd = disk->d_md_data; + if (pd->pd_subdisks >= PROMISE_MAX_SUBDISKS) { + gctl_error(req, "Disk '%s' already " + "used by %d volumes.", + diskname, pd->pd_subdisks); + error = -7; + break; + } + pp = disk->d_consumer->provider; + disks[i] = disk; + promise_meta_unused_range(pd->pd_meta, + pd->pd_subdisks, + pp->mediasize / pp->sectorsize, + &offs[i], &esize); + size = MIN(size, (off_t)esize * pp->sectorsize); + sectorsize = MAX(sectorsize, pp->sectorsize); + continue; + } + + g_topology_lock(); + cp = g_raid_open_consumer(sc, diskname); + if (cp == NULL) { + gctl_error(req, "Can't open disk '%s'.", + diskname); + g_topology_unlock(); + error = -8; + break; + } + pp = cp->provider; + pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO); + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_consumer = cp; + disks[i] = disk; + cp->private = disk; + g_topology_unlock(); + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, + "Dumping not supported by %s.", + cp->provider->name); + + /* Reserve some space for metadata. */ + size = MIN(size, pp->mediasize - 131072llu * pp->sectorsize); + sectorsize = MAX(sectorsize, pp->sectorsize); + } + if (error != 0) { + for (i = 0; i < numdisks; i++) { + if (disks[i] != NULL && + disks[i]->d_state == G_RAID_DISK_S_NONE) + g_raid_destroy_disk(disks[i]); + } + return (error); + } + + /* Handle size argument. */ + len = sizeof(*sizearg); + sizearg = gctl_get_param(req, "size", &len); + if (sizearg != NULL && len == sizeof(*sizearg) && + *sizearg > 0) { + if (*sizearg > size) { + gctl_error(req, "Size too big %lld > %lld.", + (long long)*sizearg, (long long)size); + return (-9); + } + size = *sizearg; + } + + /* Handle strip argument. */ + strip = 131072; + len = sizeof(*striparg); + striparg = gctl_get_param(req, "strip", &len); + if (striparg != NULL && len == sizeof(*striparg) && + *striparg > 0) { + if (*striparg < sectorsize) { + gctl_error(req, "Strip size too small."); + return (-10); + } + if (*striparg % sectorsize != 0) { + gctl_error(req, "Incorrect strip size."); + return (-11); + } + strip = *striparg; + } + + /* Round size down to strip or sector. */ + if (level == G_RAID_VOLUME_RL_RAID1 || + level == G_RAID_VOLUME_RL_SINGLE || + level == G_RAID_VOLUME_RL_CONCAT) + size -= (size % sectorsize); + else if (level == G_RAID_VOLUME_RL_RAID1E && + (numdisks & 1) != 0) + size -= (size % (2 * strip)); + else + size -= (size % strip); + if (size <= 0) { + gctl_error(req, "Size too small."); + return (-13); + } + if (size > 0xffffffffllu * sectorsize) { + gctl_error(req, "Size too big."); + return (-14); + } + + /* We have all we need, create things: volume, ... */ + pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO); + arc4rand(&pv->pv_id, sizeof(pv->pv_id), 0); + pv->pv_generation = 0; + pv->pv_started = 1; + vol = g_raid_create_volume(sc, volname, -1); + vol->v_md_data = pv; + vol->v_raid_level = level; + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + vol->v_strip_size = strip; + vol->v_disks_count = numdisks; + if (level == G_RAID_VOLUME_RL_RAID0 || + level == G_RAID_VOLUME_RL_CONCAT || + level == G_RAID_VOLUME_RL_SINGLE) + vol->v_mediasize = size * numdisks; + else if (level == G_RAID_VOLUME_RL_RAID1) + vol->v_mediasize = size; + else if (level == G_RAID_VOLUME_RL_RAID3 || + level == G_RAID_VOLUME_RL_RAID5) + vol->v_mediasize = size * (numdisks - 1); + else { /* RAID1E */ + vol->v_mediasize = ((size * numdisks) / strip / 2) * + strip; + } + vol->v_sectorsize = sectorsize; + g_raid_start_volume(vol); + + /* , and subdisks. */ + for (i = 0; i < numdisks; i++) { + disk = disks[i]; + sd = &vol->v_subdisks[i]; + sd->sd_disk = disk; + sd->sd_offset = (off_t)offs[i] * 512; + sd->sd_size = size; + if (disk == NULL) + continue; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + g_raid_change_disk_state(disk, + G_RAID_DISK_S_ACTIVE); + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + } + + /* Write metadata based on created entities. */ + G_RAID_DEBUG1(0, sc, "Array started."); + g_raid_md_write_promise(md, vol, NULL, NULL); + + /* Pickup any STALE/SPARE disks to refill array if needed. */ + g_raid_md_promise_refill(sc); + + g_raid_event_send(vol, G_RAID_VOLUME_E_START, + G_RAID_EVENT_VOLUME); + return (0); + } + if (strcmp(verb, "add") == 0) { + + gctl_error(req, "`add` command is not applicable, " + "use `label` instead."); + return (-99); + } + if (strcmp(verb, "delete") == 0) { + + /* Full node destruction. */ + if (*nargs == 1) { + /* Check if some volume is still open. */ + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (force != NULL && *force == 0 && + g_raid_nopens(sc) != 0) { + gctl_error(req, "Some volume is still open."); + return (-4); + } + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer) + promise_meta_erase(disk->d_consumer); + } + g_raid_destroy_node(sc, 0); + return (0); + } + + /* Destroy specified volume. If it was last - all node. */ + if (*nargs != 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + volname = gctl_get_asciiparam(req, "arg1"); + if (volname == NULL) { + gctl_error(req, "No volume name."); + return (-2); + } + + /* Search for volume. */ + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + if (strcmp(vol->v_name, volname) == 0) + break; + } + if (vol == NULL) { + i = strtol(volname, &tmp, 10); + if (verb != volname && tmp[0] == 0) { + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + if (vol->v_global_id == i) + break; + } + } + } + if (vol == NULL) { + gctl_error(req, "Volume '%s' not found.", volname); + return (-3); + } + + /* Check if volume is still open. */ + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (force != NULL && *force == 0 && + vol->v_provider_open != 0) { + gctl_error(req, "Volume is still open."); + return (-4); + } + + /* Destroy volume and potentially node. */ + i = 0; + TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next) + i++; + if (i >= 2) { + g_raid_destroy_volume(vol); + g_raid_md_promise_purge_disks(sc); + g_raid_md_write_promise(md, NULL, NULL, NULL); + } else { + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer) + promise_meta_erase(disk->d_consumer); + } + g_raid_destroy_node(sc, 0); + } + return (0); + } + if (strcmp(verb, "remove") == 0 || + strcmp(verb, "fail") == 0) { + if (*nargs < 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + for (i = 1; i < *nargs; i++) { + snprintf(arg, sizeof(arg), "arg%d", i); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -2; + break; + } + if (strncmp(diskname, "/dev/", 5) == 0) + diskname += 5; + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer != NULL && + disk->d_consumer->provider != NULL && + strcmp(disk->d_consumer->provider->name, + diskname) == 0) + break; + } + if (disk == NULL) { + gctl_error(req, "Disk '%s' not found.", + diskname); + error = -3; + break; + } + + if (strcmp(verb, "fail") == 0) { + g_raid_md_fail_disk_promise(md, NULL, disk); + continue; + } + + pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data; + + /* Erase metadata on deleting disk and destroy it. */ + promise_meta_erase(disk->d_consumer); + g_raid_destroy_disk(disk); + } + g_raid_md_promise_purge_volumes(sc); + + /* Write updated metadata to remaining disks. */ + g_raid_md_write_promise(md, NULL, NULL, NULL); + + /* Check if anything left. */ + if (g_raid_ndisks(sc, -1) == 0) + g_raid_destroy_node(sc, 0); + else + g_raid_md_promise_refill(sc); + return (error); + } + if (strcmp(verb, "insert") == 0) { + if (*nargs < 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + for (i = 1; i < *nargs; i++) { + /* Get disk name. */ + snprintf(arg, sizeof(arg), "arg%d", i); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -3; + break; + } + + /* Try to find provider with specified name. */ + g_topology_lock(); + cp = g_raid_open_consumer(sc, diskname); + if (cp == NULL) { + gctl_error(req, "Can't open disk '%s'.", + diskname); + g_topology_unlock(); + error = -4; + break; + } + pp = cp->provider; + g_topology_unlock(); + + pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO); + + disk = g_raid_create_disk(sc); + disk->d_consumer = cp; + disk->d_consumer->private = disk; + disk->d_md_data = (void *)pd; + cp->private = disk; + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, + "Dumping not supported by %s.", + cp->provider->name); + + /* Welcome the "new" disk. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE); + promise_meta_write_spare(cp); + g_raid_md_promise_refill(sc); + } + return (error); + } + return (-100); +} + +static int +g_raid_md_write_promise(struct g_raid_md_object *md, struct g_raid_volume *tvol, + struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_promise_perdisk *pd; + struct g_raid_md_promise_pervolume *pv; + struct promise_raid_conf *meta; + off_t rebuild_lba64; + int i, j, pos, rebuild; + + sc = md->mdo_softc; + + if (sc->sc_stopping == G_RAID_DESTROY_HARD) + return (0); + + /* Generate new per-volume metadata for affected volumes. */ + TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) { + if (vol->v_stopping) + continue; + + /* Skip volumes not related to specified targets. */ + if (tvol != NULL && vol != tvol) + continue; + if (tsd != NULL && vol != tsd->sd_volume) + continue; + if (tdisk != NULL) { + for (i = 0; i < vol->v_disks_count; i++) { + if (vol->v_subdisks[i].sd_disk == tdisk) + break; + } + if (i >= vol->v_disks_count) + continue; + } + + pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data; + pv->pv_generation++; + + meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO); + if (pv->pv_meta != NULL) + memcpy(meta, pv->pv_meta, sizeof(*meta)); + memcpy(meta->promise_id, PROMISE_MAGIC, sizeof(PROMISE_MAGIC)); + meta->dummy_0 = 0x00020000; + meta->integrity = PROMISE_I_VALID; + + meta->generation = pv->pv_generation; + meta->status = PROMISE_S_VALID | PROMISE_S_ONLINE | + PROMISE_S_INITED | PROMISE_S_READY; + if (vol->v_state <= G_RAID_VOLUME_S_DEGRADED) + meta->status |= PROMISE_S_DEGRADED; + if (vol->v_dirty) + meta->status |= PROMISE_S_MARKED; /* XXX: INVENTED! */ + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0 || + vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE) + meta->type = PROMISE_T_RAID0; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || + vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) + meta->type = PROMISE_T_RAID1; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) + meta->type = PROMISE_T_RAID3; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) + meta->type = PROMISE_T_RAID5; + else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) + meta->type = PROMISE_T_SPAN; + else + meta->type = PROMISE_T_JBOD; + meta->total_disks = vol->v_disks_count; + meta->stripe_shift = ffs(vol->v_strip_size / 1024); + meta->array_width = vol->v_disks_count; + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || + vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) + meta->array_width /= 2; + meta->array_number = vol->v_global_id; + meta->total_sectors = vol->v_mediasize / vol->v_sectorsize; + meta->cylinders = meta->total_sectors / (255 * 63) - 1; + meta->heads = 254; + meta->sectors = 63; + meta->volume_id = pv->pv_id; + rebuild_lba64 = UINT64_MAX; + rebuild = 0; + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + /* For RAID0+1 we need to translate order. */ + pos = promise_meta_translate_disk(vol, i); + meta->disks[pos].flags = PROMISE_F_VALID | + PROMISE_F_ASSIGNED; + if (sd->sd_state == G_RAID_SUBDISK_S_NONE) { + meta->disks[pos].flags |= 0; + } else if (sd->sd_state == G_RAID_SUBDISK_S_FAILED) { + meta->disks[pos].flags |= + PROMISE_F_DOWN | PROMISE_F_REDIR; + } else if (sd->sd_state <= G_RAID_SUBDISK_S_REBUILD) { + meta->disks[pos].flags |= + PROMISE_F_ONLINE | PROMISE_F_REDIR; + if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) { + rebuild_lba64 = MIN(rebuild_lba64, + sd->sd_rebuild_pos / 512); + } else + rebuild_lba64 = 0; + rebuild = 1; + } else { + meta->disks[pos].flags |= PROMISE_F_ONLINE; + if (sd->sd_state < G_RAID_SUBDISK_S_ACTIVE) { + meta->status |= PROMISE_S_MARKED; + if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC) { + rebuild_lba64 = MIN(rebuild_lba64, + sd->sd_rebuild_pos / 512); + } else + rebuild_lba64 = 0; + } + } + if (pv->pv_meta != NULL) { + meta->disks[pos].id = pv->pv_meta->disks[pos].id; + } else { + meta->disks[pos].number = i * 2; + arc4rand(&meta->disks[pos].id, + sizeof(meta->disks[pos].id), 0); + } + } + promise_meta_put_name(meta, vol->v_name); + + /* Try to mimic AMD BIOS rebuild/resync behavior. */ + if (rebuild_lba64 != UINT64_MAX) { + if (rebuild) + meta->magic_3 = 0x03040010UL; /* Rebuild? */ + else + meta->magic_3 = 0x03040008UL; /* Resync? */ + /* Translate from per-disk to per-volume LBA. */ + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 || + vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) { + rebuild_lba64 *= meta->array_width; + } else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 || + vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) { + rebuild_lba64 *= meta->array_width - 1; + } else + rebuild_lba64 = 0; + } else + meta->magic_3 = 0x03000000UL; + meta->rebuild_lba64 = rebuild_lba64; + meta->magic_4 = 0x04010101UL; + + /* Replace per-volume metadata with new. */ + if (pv->pv_meta != NULL) + free(pv->pv_meta, M_MD_PROMISE); + pv->pv_meta = meta; + + /* Copy new metadata to the disks, adding or replacing old. */ + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + disk = sd->sd_disk; + if (disk == NULL) + continue; + /* For RAID0+1 we need to translate order. */ + pos = promise_meta_translate_disk(vol, i); + pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data; + for (j = 0; j < pd->pd_subdisks; j++) { + if (pd->pd_meta[j]->volume_id == meta->volume_id) + break; + } + if (j == pd->pd_subdisks) + pd->pd_subdisks++; + if (pd->pd_meta[j] != NULL) + free(pd->pd_meta[j], M_MD_PROMISE); + pd->pd_meta[j] = promise_meta_copy(meta); + pd->pd_meta[j]->disk = meta->disks[pos]; + pd->pd_meta[j]->disk.number = pos; + pd->pd_meta[j]->disk_offset = sd->sd_offset / 512; + pd->pd_meta[j]->disk_sectors = sd->sd_size / 512; + if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) { + pd->pd_meta[j]->rebuild_lba = + sd->sd_rebuild_pos / 512; + } else if (sd->sd_state < G_RAID_SUBDISK_S_REBUILD) + pd->pd_meta[j]->rebuild_lba = 0; + else + pd->pd_meta[j]->rebuild_lba = UINT32_MAX; + pd->pd_updated = 1; + } + } + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data; + if (disk->d_state != G_RAID_DISK_S_ACTIVE) + continue; + if (!pd->pd_updated) + continue; + G_RAID_DEBUG(1, "Writing Promise metadata to %s", + g_raid_get_diskname(disk)); + for (i = 0; i < pd->pd_subdisks; i++) + g_raid_md_promise_print(pd->pd_meta[i]); + promise_meta_write(disk->d_consumer, + pd->pd_meta, pd->pd_subdisks); + pd->pd_updated = 0; + } + + return (0); +} + +static int +g_raid_md_fail_disk_promise(struct g_raid_md_object *md, + struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) +{ + struct g_raid_softc *sc; + struct g_raid_md_promise_perdisk *pd; + struct g_raid_subdisk *sd; + int i, pos; + + sc = md->mdo_softc; + pd = (struct g_raid_md_promise_perdisk *)tdisk->d_md_data; + + /* We can't fail disk that is not a part of array now. */ + if (tdisk->d_state != G_RAID_DISK_S_ACTIVE) + return (-1); + + /* + * Mark disk as failed in metadata and try to write that metadata + * to the disk itself to prevent it's later resurrection as STALE. + */ + if (pd->pd_subdisks > 0 && tdisk->d_consumer != NULL) + G_RAID_DEBUG(1, "Writing Promise metadata to %s", + g_raid_get_diskname(tdisk)); + for (i = 0; i < pd->pd_subdisks; i++) { + pd->pd_meta[i]->disk.flags |= + PROMISE_F_DOWN | PROMISE_F_REDIR; + pos = pd->pd_meta[i]->disk.number; + if (pos >= 0 && pos < PROMISE_MAX_DISKS) { + pd->pd_meta[i]->disks[pos].flags |= + PROMISE_F_DOWN | PROMISE_F_REDIR; + } + g_raid_md_promise_print(pd->pd_meta[i]); + } + if (tdisk->d_consumer != NULL) + promise_meta_write(tdisk->d_consumer, + pd->pd_meta, pd->pd_subdisks); + + /* Change states. */ + g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED); + TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_FAILED); + g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED, + G_RAID_EVENT_SUBDISK); + } + + /* Write updated metadata to remaining disks. */ + g_raid_md_write_promise(md, NULL, NULL, tdisk); + + g_raid_md_promise_refill(sc); + return (0); +} + +static int +g_raid_md_free_disk_promise(struct g_raid_md_object *md, + struct g_raid_disk *disk) +{ + struct g_raid_md_promise_perdisk *pd; + int i; + + pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data; + for (i = 0; i < pd->pd_subdisks; i++) { + if (pd->pd_meta[i] != NULL) { + free(pd->pd_meta[i], M_MD_PROMISE); + pd->pd_meta[i] = NULL; + } + } + free(pd, M_MD_PROMISE); + disk->d_md_data = NULL; + return (0); +} + +static int +g_raid_md_free_volume_promise(struct g_raid_md_object *md, + struct g_raid_volume *vol) +{ + struct g_raid_md_promise_pervolume *pv; + + pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data; + if (pv && pv->pv_meta != NULL) { + free(pv->pv_meta, M_MD_PROMISE); + pv->pv_meta = NULL; + } + if (pv && !pv->pv_started) { + pv->pv_started = 1; + callout_stop(&pv->pv_start_co); + } + return (0); +} + +static int +g_raid_md_free_promise(struct g_raid_md_object *md) +{ + + return (0); +} + +G_RAID_MD_DECLARE(g_raid_md_promise); diff --git a/sys/geom/raid/md_sii.c b/sys/geom/raid/md_sii.c new file mode 100644 index 000000000000..305accdfcf54 --- /dev/null +++ b/sys/geom/raid/md_sii.c @@ -0,0 +1,1692 @@ +/*- + * Copyright (c) 2011 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "geom/raid/g_raid.h" +#include "g_raid_md_if.h" + +static MALLOC_DEFINE(M_MD_SII, "md_sii_data", "GEOM_RAID SiI metadata"); + +struct sii_raid_conf { + uint16_t ata_params_00_53[54]; + uint64_t total_sectors; /* 54 - 57 */ + uint16_t ata_params_58_81[72]; + uint16_t product_id; /* 130 */ + uint16_t vendor_id; /* 131 */ + uint16_t version_minor; /* 132 */ + uint16_t version_major; /* 133 */ + uint8_t timestamp[6]; /* 134 - 136 */ + uint16_t strip_sectors; /* 137 */ + uint16_t dummy_2; + uint8_t disk_number; /* 139 */ + uint8_t type; +#define SII_T_RAID0 0x00 +#define SII_T_RAID1 0x01 +#define SII_T_RAID01 0x02 +#define SII_T_SPARE 0x03 +#define SII_T_CONCAT 0x04 +#define SII_T_RAID5 0x10 +#define SII_T_RESERVED 0xfd +#define SII_T_JBOD 0xff + + uint8_t raid0_disks; /* 140 */ + uint8_t raid0_ident; + uint8_t raid1_disks; /* 141 */ + uint8_t raid1_ident; + uint64_t rebuild_lba; /* 142 - 145 */ + uint32_t generation; /* 146 - 147 */ + uint8_t disk_status; /* 148 */ +#define SII_S_CURRENT 0x01 +#define SII_S_REBUILD 0x02 +#define SII_S_DROPPED 0x03 +#define SII_S_REMOVED 0x04 + + uint8_t raid_status; +#define SII_S_ONLINE 0x01 +#define SII_S_AVAILABLE 0x02 + + uint8_t raid_location; /* 149 */ + uint8_t disk_location; + uint8_t auto_rebuild; /* 150 */ +#define SII_R_REBUILD 0x00 +#define SII_R_NOREBUILD 0xff + + uint8_t dummy_3; + uint8_t name[16]; /* 151 - 158 */ + uint16_t checksum; /* 159 */ + uint16_t ata_params_160_255[96]; +} __packed; + +struct g_raid_md_sii_perdisk { + struct sii_raid_conf *pd_meta; + int pd_disk_pos; + off_t pd_disk_size; +}; + +struct g_raid_md_sii_object { + struct g_raid_md_object mdio_base; + uint8_t mdio_timestamp[6]; + uint8_t mdio_location; + uint32_t mdio_generation; + struct sii_raid_conf *mdio_meta; + struct callout mdio_start_co; /* STARTING state timer. */ + int mdio_total_disks; + int mdio_disks_present; + int mdio_started; + int mdio_incomplete; + struct root_hold_token *mdio_rootmount; /* Root mount delay token. */ +}; + +static g_raid_md_create_t g_raid_md_create_sii; +static g_raid_md_taste_t g_raid_md_taste_sii; +static g_raid_md_event_t g_raid_md_event_sii; +static g_raid_md_ctl_t g_raid_md_ctl_sii; +static g_raid_md_write_t g_raid_md_write_sii; +static g_raid_md_fail_disk_t g_raid_md_fail_disk_sii; +static g_raid_md_free_disk_t g_raid_md_free_disk_sii; +static g_raid_md_free_t g_raid_md_free_sii; + +static kobj_method_t g_raid_md_sii_methods[] = { + KOBJMETHOD(g_raid_md_create, g_raid_md_create_sii), + KOBJMETHOD(g_raid_md_taste, g_raid_md_taste_sii), + KOBJMETHOD(g_raid_md_event, g_raid_md_event_sii), + KOBJMETHOD(g_raid_md_ctl, g_raid_md_ctl_sii), + KOBJMETHOD(g_raid_md_write, g_raid_md_write_sii), + KOBJMETHOD(g_raid_md_fail_disk, g_raid_md_fail_disk_sii), + KOBJMETHOD(g_raid_md_free_disk, g_raid_md_free_disk_sii), + KOBJMETHOD(g_raid_md_free, g_raid_md_free_sii), + { 0, 0 } +}; + +static struct g_raid_md_class g_raid_md_sii_class = { + "SiI", + g_raid_md_sii_methods, + sizeof(struct g_raid_md_sii_object), + .mdc_priority = 100 +}; + +static void +g_raid_md_sii_print(struct sii_raid_conf *meta) +{ + + if (g_raid_debug < 1) + return; + + printf("********* ATA SiI RAID Metadata *********\n"); + printf("total_sectors %llu\n", + (long long unsigned)meta->total_sectors); + printf("product_id 0x%04x\n", meta->product_id); + printf("vendor_id 0x%04x\n", meta->vendor_id); + printf("version_minor 0x%04x\n", meta->version_minor); + printf("version_major 0x%04x\n", meta->version_major); + printf("timestamp 0x%02x%02x%02x%02x%02x%02x\n", + meta->timestamp[5], meta->timestamp[4], meta->timestamp[3], + meta->timestamp[2], meta->timestamp[1], meta->timestamp[0]); + printf("strip_sectors %d\n", meta->strip_sectors); + printf("disk_number %d\n", meta->disk_number); + printf("type 0x%02x\n", meta->type); + printf("raid0_disks %d\n", meta->raid0_disks); + printf("raid0_ident %d\n", meta->raid0_ident); + printf("raid1_disks %d\n", meta->raid1_disks); + printf("raid1_ident %d\n", meta->raid1_ident); + printf("rebuild_lba %llu\n", + (long long unsigned)meta->rebuild_lba); + printf("generation %d\n", meta->generation); + printf("disk_status %d\n", meta->disk_status); + printf("raid_status %d\n", meta->raid_status); + printf("raid_location %d\n", meta->raid_location); + printf("disk_location %d\n", meta->disk_location); + printf("auto_rebuild %d\n", meta->auto_rebuild); + printf("name <%.16s>\n", meta->name); + printf("checksum 0x%04x\n", meta->checksum); + printf("=================================================\n"); +} + +static struct sii_raid_conf * +sii_meta_copy(struct sii_raid_conf *meta) +{ + struct sii_raid_conf *nmeta; + + nmeta = malloc(sizeof(*meta), M_MD_SII, M_WAITOK); + memcpy(nmeta, meta, sizeof(*meta)); + return (nmeta); +} + +static int +sii_meta_total_disks(struct sii_raid_conf *meta) +{ + + switch (meta->type) { + case SII_T_RAID0: + case SII_T_RAID5: + case SII_T_CONCAT: + return (meta->raid0_disks); + case SII_T_RAID1: + return (meta->raid1_disks); + case SII_T_RAID01: + return (meta->raid0_disks * meta->raid1_disks); + case SII_T_SPARE: + case SII_T_JBOD: + return (1); + } + return (0); +} + +static int +sii_meta_disk_pos(struct sii_raid_conf *meta, struct sii_raid_conf *pdmeta) +{ + + if (pdmeta->type == SII_T_SPARE) + return (-3); + + if (memcmp(&meta->timestamp, &pdmeta->timestamp, 6) != 0) + return (-1); + + switch (pdmeta->type) { + case SII_T_RAID0: + case SII_T_RAID1: + case SII_T_RAID5: + case SII_T_CONCAT: + return (pdmeta->disk_number); + case SII_T_RAID01: + return (pdmeta->raid1_ident * pdmeta->raid1_disks + + pdmeta->raid0_ident); + case SII_T_JBOD: + return (0); + } + return (-1); +} + +static void +sii_meta_get_name(struct sii_raid_conf *meta, char *buf) +{ + int i; + + strncpy(buf, meta->name, 16); + buf[16] = 0; + for (i = 15; i >= 0; i--) { + if (buf[i] > 0x20) + break; + buf[i] = 0; + } +} + +static void +sii_meta_put_name(struct sii_raid_conf *meta, char *buf) +{ + + memset(meta->name, 0x20, 16); + memcpy(meta->name, buf, MIN(strlen(buf), 16)); +} + +static struct sii_raid_conf * +sii_meta_read(struct g_consumer *cp) +{ + struct g_provider *pp; + struct sii_raid_conf *meta; + char *buf; + int error, i; + uint16_t checksum, *ptr; + + pp = cp->provider; + + /* Read the anchor sector. */ + buf = g_read_data(cp, + pp->mediasize - pp->sectorsize, pp->sectorsize, &error); + if (buf == NULL) { + G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).", + pp->name, error); + return (NULL); + } + meta = malloc(sizeof(*meta), M_MD_SII, M_WAITOK); + memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize)); + g_free(buf); + + /* Check vendor ID. */ + if (meta->vendor_id != 0x1095) { + G_RAID_DEBUG(1, "SiI vendor ID check failed on %s (0x%04x)", + pp->name, meta->vendor_id); + free(meta, M_MD_SII); + return (NULL); + } + + /* Check metadata major version. */ + if (meta->version_major != 2) { + G_RAID_DEBUG(1, "SiI version check failed on %s (%d.%d)", + pp->name, meta->version_major, meta->version_minor); + free(meta, M_MD_SII); + return (NULL); + } + + /* Check metadata checksum. */ + for (checksum = 0, ptr = (uint16_t *)meta, i = 0; i <= 159; i++) + checksum += *ptr++; + if (checksum != 0) { + G_RAID_DEBUG(1, "SiI checksum check failed on %s", pp->name); + free(meta, M_MD_SII); + return (NULL); + } + + /* Check raid type. */ + if (meta->type != SII_T_RAID0 && meta->type != SII_T_RAID1 && + meta->type != SII_T_RAID01 && meta->type != SII_T_SPARE && + meta->type != SII_T_RAID5 && meta->type != SII_T_CONCAT && + meta->type != SII_T_JBOD) { + G_RAID_DEBUG(1, "SiI unknown RAID level on %s (0x%02x)", + pp->name, meta->type); + free(meta, M_MD_SII); + return (NULL); + } + + return (meta); +} + +static int +sii_meta_write(struct g_consumer *cp, struct sii_raid_conf *meta) +{ + struct g_provider *pp; + char *buf; + int error, i; + uint16_t checksum, *ptr; + + pp = cp->provider; + + /* Recalculate checksum for case if metadata were changed. */ + meta->checksum = 0; + for (checksum = 0, ptr = (uint16_t *)meta, i = 0; i < 159; i++) + checksum += *ptr++; + meta->checksum -= checksum; + + /* Create and fill buffer. */ + buf = malloc(pp->sectorsize, M_MD_SII, M_WAITOK | M_ZERO); + memcpy(buf, meta, sizeof(*meta)); + + /* Write 4 copies of metadata. */ + for (i = 0; i < 4; i++) { + error = g_write_data(cp, + pp->mediasize - (pp->sectorsize * (1 + 0x200 * i)), + buf, pp->sectorsize); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).", + pp->name, error); + break; + } + } + + free(buf, M_MD_SII); + return (error); +} + +static int +sii_meta_erase(struct g_consumer *cp) +{ + struct g_provider *pp; + char *buf; + int error, i; + + pp = cp->provider; + buf = malloc(pp->sectorsize, M_MD_SII, M_WAITOK | M_ZERO); + /* Write 4 copies of metadata. */ + for (i = 0; i < 4; i++) { + error = g_write_data(cp, + pp->mediasize - (pp->sectorsize * (1 + 0x200 * i)), + buf, pp->sectorsize); + if (error != 0) { + G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).", + pp->name, error); + } + } + free(buf, M_MD_SII); + return (error); +} + +static int +sii_meta_write_spare(struct g_consumer *cp) +{ + struct sii_raid_conf *meta; + int error; + + meta = malloc(sizeof(*meta), M_MD_SII, M_WAITOK | M_ZERO); + meta->total_sectors = cp->provider->mediasize / + cp->provider->sectorsize - 0x800; + meta->vendor_id = 0x1095; + meta->version_minor = 0; + meta->version_major = 2; + meta->timestamp[0] = arc4random(); + meta->timestamp[1] = arc4random(); + meta->timestamp[2] = arc4random(); + meta->timestamp[3] = arc4random(); + meta->timestamp[4] = arc4random(); + meta->timestamp[5] = arc4random(); + meta->type = SII_T_SPARE; + meta->generation = 1; + meta->raid1_ident = 0xff; + meta->raid_location = arc4random(); + error = sii_meta_write(cp, meta); + free(meta, M_MD_SII); + return (error); +} + +static struct g_raid_disk * +g_raid_md_sii_get_disk(struct g_raid_softc *sc, int id) +{ + struct g_raid_disk *disk; + struct g_raid_md_sii_perdisk *pd; + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data; + if (pd->pd_disk_pos == id) + break; + } + return (disk); +} + +static int +g_raid_md_sii_supported(int level, int qual, int disks, int force) +{ + + if (disks > 8) + return (0); + switch (level) { + case G_RAID_VOLUME_RL_RAID0: + if (disks < 1) + return (0); + if (!force && (disks < 2 || disks > 6)) + return (0); + break; + case G_RAID_VOLUME_RL_RAID1: + if (disks < 1) + return (0); + if (!force && (disks != 2)) + return (0); + break; + case G_RAID_VOLUME_RL_RAID1E: + if (disks < 2) + return (0); + if (disks % 2 != 0) + return (0); + if (!force && (disks < 4)) + return (0); + break; + case G_RAID_VOLUME_RL_SINGLE: + if (disks != 1) + return (0); + break; + case G_RAID_VOLUME_RL_CONCAT: + if (disks < 2) + return (0); + break; + case G_RAID_VOLUME_RL_RAID5: + if (disks < 3) + return (0); + break; + default: + return (0); + } + if (qual != G_RAID_VOLUME_RLQ_NONE) + return (0); + return (1); +} + +static int +g_raid_md_sii_start_disk(struct g_raid_disk *disk) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd, *tmpsd; + struct g_raid_disk *olddisk, *tmpdisk; + struct g_raid_md_object *md; + struct g_raid_md_sii_object *mdi; + struct g_raid_md_sii_perdisk *pd, *oldpd; + struct sii_raid_conf *meta; + int disk_pos, resurrection = 0; + + sc = disk->d_softc; + md = sc->sc_md; + mdi = (struct g_raid_md_sii_object *)md; + meta = mdi->mdio_meta; + pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data; + olddisk = NULL; + + /* Find disk position in metadata by it's serial. */ + if (pd->pd_meta != NULL) + disk_pos = sii_meta_disk_pos(meta, pd->pd_meta); + else + disk_pos = -3; + if (disk_pos < 0) { + G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk"); + /* If we are in the start process, that's all for now. */ + if (!mdi->mdio_started) + goto nofit; + /* + * If we have already started - try to get use of the disk. + * Try to replace OFFLINE disks first, then FAILED. + */ + TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) { + if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE && + tmpdisk->d_state != G_RAID_DISK_S_FAILED) + continue; + /* Make sure this disk is big enough. */ + TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) { + if (sd->sd_offset + sd->sd_size + 512 > + pd->pd_disk_size) { + G_RAID_DEBUG1(1, sc, + "Disk too small (%ju < %ju)", + pd->pd_disk_size, + sd->sd_offset + sd->sd_size + 512); + break; + } + } + if (sd != NULL) + continue; + if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) { + olddisk = tmpdisk; + break; + } else if (olddisk == NULL) + olddisk = tmpdisk; + } + if (olddisk == NULL) { +nofit: + if (disk_pos == -3 || pd->pd_disk_pos == -3) { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_SPARE); + return (1); + } else { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_STALE); + return (0); + } + } + oldpd = (struct g_raid_md_sii_perdisk *)olddisk->d_md_data; + disk_pos = oldpd->pd_disk_pos; + resurrection = 1; + } + + if (olddisk == NULL) { + /* Find placeholder by position. */ + olddisk = g_raid_md_sii_get_disk(sc, disk_pos); + if (olddisk == NULL) + panic("No disk at position %d!", disk_pos); + if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) { + G_RAID_DEBUG1(1, sc, "More then one disk for pos %d", + disk_pos); + g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE); + return (0); + } + oldpd = (struct g_raid_md_sii_perdisk *)olddisk->d_md_data; + } + + /* Replace failed disk or placeholder with new disk. */ + TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) { + TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next); + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + sd->sd_disk = disk; + } + oldpd->pd_disk_pos = -2; + pd->pd_disk_pos = disk_pos; + + /* If it was placeholder -- destroy it. */ + if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) { + g_raid_destroy_disk(olddisk); + } else { + /* Otherwise, make it STALE_FAILED. */ + g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED); + } + + /* Welcome the new disk. */ + if (resurrection) + g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); + else if (pd->pd_meta->disk_status == SII_S_CURRENT || + pd->pd_meta->disk_status == SII_S_REBUILD) + g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE); + else + g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED); + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + + /* + * Different disks may have different sizes, + * in concat mode. Update from real disk size. + */ + if (meta->type == SII_T_CONCAT || meta->type == SII_T_JBOD) + sd->sd_size = pd->pd_disk_size - 0x800 * 512; + + if (resurrection) { + /* New or ex-spare disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NEW); + } else if (pd->pd_meta->disk_status == SII_S_REBUILD) { + /* Rebuilding disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_REBUILD); + if (pd->pd_meta->generation == meta->generation) + sd->sd_rebuild_pos = pd->pd_meta->rebuild_lba * 512; + else + sd->sd_rebuild_pos = 0; + } else if (pd->pd_meta->disk_status == SII_S_CURRENT) { + if (pd->pd_meta->raid_status == SII_S_ONLINE || + pd->pd_meta->generation != meta->generation) { + /* Dirty or resyncing disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_STALE); + } else { + /* Up to date disk. */ + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + } + } else { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_FAILED); + } + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + } + + /* Update status of our need for spare. */ + if (mdi->mdio_started) { + mdi->mdio_incomplete = + (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < + mdi->mdio_total_disks); + } + + return (resurrection); +} + +static void +g_disk_md_sii_retaste(void *arg, int pending) +{ + + G_RAID_DEBUG(1, "Array is not complete, trying to retaste."); + g_retaste(&g_raid_class); + free(arg, M_MD_SII); +} + +static void +g_raid_md_sii_refill(struct g_raid_softc *sc) +{ + struct g_raid_md_object *md; + struct g_raid_md_sii_object *mdi; + struct sii_raid_conf *meta; + struct g_raid_disk *disk; + struct task *task; + int update, na; + + md = sc->sc_md; + mdi = (struct g_raid_md_sii_object *)md; + meta = mdi->mdio_meta; + update = 0; + do { + /* Make sure we miss anything. */ + na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE); + if (na == mdi->mdio_total_disks) + break; + + G_RAID_DEBUG1(1, md->mdo_softc, + "Array is not complete (%d of %d), " + "trying to refill.", na, mdi->mdio_total_disks); + + /* Try to get use some of STALE disks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_STALE) { + update += g_raid_md_sii_start_disk(disk); + if (disk->d_state == G_RAID_DISK_S_ACTIVE) + break; + } + } + if (disk != NULL) + continue; + + /* Try to get use some of SPARE disks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state == G_RAID_DISK_S_SPARE) { + update += g_raid_md_sii_start_disk(disk); + if (disk->d_state == G_RAID_DISK_S_ACTIVE) + break; + } + } + } while (disk != NULL); + + /* Write new metadata if we changed something. */ + if (update) { + g_raid_md_write_sii(md, NULL, NULL, NULL); + meta = mdi->mdio_meta; + } + + /* Update status of our need for spare. */ + mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) < + mdi->mdio_total_disks); + + /* Request retaste hoping to find spare. */ + if (mdi->mdio_incomplete) { + task = malloc(sizeof(struct task), + M_MD_SII, M_WAITOK | M_ZERO); + TASK_INIT(task, 0, g_disk_md_sii_retaste, task); + taskqueue_enqueue(taskqueue_swi, task); + } +} + +static void +g_raid_md_sii_start(struct g_raid_softc *sc) +{ + struct g_raid_md_object *md; + struct g_raid_md_sii_object *mdi; + struct g_raid_md_sii_perdisk *pd; + struct sii_raid_conf *meta; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk, *best; + off_t size; + int j, disk_pos; + uint32_t gendiff, bestgendiff; + char buf[17]; + + md = sc->sc_md; + mdi = (struct g_raid_md_sii_object *)md; + meta = mdi->mdio_meta; + + /* Create volumes and subdisks. */ + sii_meta_get_name(meta, buf); + vol = g_raid_create_volume(sc, buf, -1); + vol->v_mediasize = (off_t)meta->total_sectors * 512; + if (meta->type == SII_T_RAID0) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID0; + size = vol->v_mediasize / mdi->mdio_total_disks; + } else if (meta->type == SII_T_RAID1) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1; + size = vol->v_mediasize; + } else if (meta->type == SII_T_RAID01) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E; + size = vol->v_mediasize / (mdi->mdio_total_disks / 2); + } else if (meta->type == SII_T_CONCAT) { + if (mdi->mdio_total_disks == 1) + vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE; + else + vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT; + size = 0; + } else if (meta->type == SII_T_RAID5) { + vol->v_raid_level = G_RAID_VOLUME_RL_RAID5; + size = vol->v_mediasize / (mdi->mdio_total_disks - 1); + } else if (meta->type == SII_T_JBOD) { + vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE; + size = 0; + } else { + vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN; + size = 0; + } + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + vol->v_strip_size = meta->strip_sectors * 512; //ZZZ + vol->v_disks_count = mdi->mdio_total_disks; + vol->v_sectorsize = 512; //ZZZ + for (j = 0; j < vol->v_disks_count; j++) { + sd = &vol->v_subdisks[j]; + sd->sd_offset = 0; + sd->sd_size = size; + } + g_raid_start_volume(vol); + + /* Create disk placeholders to store data for later writing. */ + for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) { + pd = malloc(sizeof(*pd), M_MD_SII, M_WAITOK | M_ZERO); + pd->pd_disk_pos = disk_pos; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_state = G_RAID_DISK_S_OFFLINE; + sd = &vol->v_subdisks[disk_pos]; + sd->sd_disk = disk; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + } + + /* + * Make all disks found till the moment take their places + * in order of their generation numbers. + */ + do { + best = NULL; + bestgendiff = 0xffffffff; + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_state != G_RAID_DISK_S_NONE) + continue; + pd = disk->d_md_data; + if (pd->pd_meta == NULL) + gendiff = 0xfffffffe; + else + gendiff = meta->generation - + pd->pd_meta->generation; + if (gendiff < bestgendiff) { + best = disk; + bestgendiff = gendiff; + } + } + if (best != NULL) + g_raid_md_sii_start_disk(best); + } while (best != NULL); + + mdi->mdio_started = 1; + G_RAID_DEBUG1(0, sc, "Array started."); + g_raid_md_write_sii(md, NULL, NULL, NULL); + + /* Pickup any STALE/SPARE disks to refill array if needed. */ + g_raid_md_sii_refill(sc); + + g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME); + + callout_stop(&mdi->mdio_start_co); + G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount); + root_mount_rel(mdi->mdio_rootmount); + mdi->mdio_rootmount = NULL; +} + +static void +g_raid_md_sii_new_disk(struct g_raid_disk *disk) +{ + struct g_raid_softc *sc; + struct g_raid_md_object *md; + struct g_raid_md_sii_object *mdi; + struct sii_raid_conf *pdmeta; + struct g_raid_md_sii_perdisk *pd; + + sc = disk->d_softc; + md = sc->sc_md; + mdi = (struct g_raid_md_sii_object *)md; + pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data; + pdmeta = pd->pd_meta; + + if (mdi->mdio_started) { + if (g_raid_md_sii_start_disk(disk)) + g_raid_md_write_sii(md, NULL, NULL, NULL); + } else { + if (mdi->mdio_meta == NULL || + ((int32_t)(pdmeta->generation - mdi->mdio_generation)) > 0) { + G_RAID_DEBUG1(1, sc, "Newer disk"); + if (mdi->mdio_meta != NULL) + free(mdi->mdio_meta, M_MD_SII); + mdi->mdio_meta = sii_meta_copy(pdmeta); + mdi->mdio_generation = mdi->mdio_meta->generation; + mdi->mdio_total_disks = sii_meta_total_disks(pdmeta); + mdi->mdio_disks_present = 1; + } else if (pdmeta->generation == mdi->mdio_generation) { + mdi->mdio_disks_present++; + G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)", + mdi->mdio_disks_present, + mdi->mdio_total_disks); + } else { + G_RAID_DEBUG1(1, sc, "Older disk"); + } + + /* If we collected all needed disks - start array. */ + if (mdi->mdio_disks_present == mdi->mdio_total_disks) + g_raid_md_sii_start(sc); + } +} + +static void +g_raid_sii_go(void *arg) +{ + struct g_raid_softc *sc; + struct g_raid_md_object *md; + struct g_raid_md_sii_object *mdi; + + sc = arg; + md = sc->sc_md; + mdi = (struct g_raid_md_sii_object *)md; + if (!mdi->mdio_started) { + G_RAID_DEBUG1(0, sc, "Force array start due to timeout."); + g_raid_event_send(sc, G_RAID_NODE_E_START, 0); + } +} + +static int +g_raid_md_create_sii(struct g_raid_md_object *md, struct g_class *mp, + struct g_geom **gp) +{ + struct g_raid_softc *sc; + struct g_raid_md_sii_object *mdi; + char name[32]; + + mdi = (struct g_raid_md_sii_object *)md; + mdi->mdio_timestamp[5] = arc4random(); + mdi->mdio_timestamp[4] = arc4random(); + mdi->mdio_timestamp[3] = arc4random(); + mdi->mdio_timestamp[2] = arc4random(); + mdi->mdio_timestamp[1] = arc4random(); + mdi->mdio_timestamp[0] = arc4random(); + mdi->mdio_location = arc4random(); + mdi->mdio_generation = 0; + snprintf(name, sizeof(name), "SiI-%02x%02x%02x%02x%02x%02x", + mdi->mdio_timestamp[5], mdi->mdio_timestamp[4], + mdi->mdio_timestamp[3], mdi->mdio_timestamp[2], + mdi->mdio_timestamp[1], mdi->mdio_timestamp[0]); + sc = g_raid_create_node(mp, name, md); + if (sc == NULL) + return (G_RAID_MD_TASTE_FAIL); + md->mdo_softc = sc; + *gp = sc->sc_geom; + return (G_RAID_MD_TASTE_NEW); +} + +static int +g_raid_md_taste_sii(struct g_raid_md_object *md, struct g_class *mp, + struct g_consumer *cp, struct g_geom **gp) +{ + struct g_consumer *rcp; + struct g_provider *pp; + struct g_raid_md_sii_object *mdi, *mdi1; + struct g_raid_softc *sc; + struct g_raid_disk *disk; + struct sii_raid_conf *meta; + struct g_raid_md_sii_perdisk *pd; + struct g_geom *geom; + int error, disk_pos, result, spare, len; + char name[32]; + uint16_t vendor; + + G_RAID_DEBUG(1, "Tasting SiI on %s", cp->provider->name); + mdi = (struct g_raid_md_sii_object *)md; + pp = cp->provider; + + /* Read metadata from device. */ + meta = NULL; + spare = 0; + vendor = 0xffff; + disk_pos = 0; + if (g_access(cp, 1, 0, 0) != 0) + return (G_RAID_MD_TASTE_FAIL); + g_topology_unlock(); + len = 2; + if (pp->geom->rank == 1) + g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor); + meta = sii_meta_read(cp); + g_topology_lock(); + g_access(cp, -1, 0, 0); + if (meta == NULL) { + if (g_raid_aggressive_spare) { + if (vendor == 0x1095) { + G_RAID_DEBUG(1, + "No SiI metadata, forcing spare."); + spare = 2; + goto search; + } else { + G_RAID_DEBUG(1, + "SiI vendor mismatch 0x%04x != 0x1095", + vendor); + } + } + return (G_RAID_MD_TASTE_FAIL); + } + + /* Check this disk position in obtained metadata. */ + disk_pos = sii_meta_disk_pos(meta, meta); + if (disk_pos == -1) { + G_RAID_DEBUG(1, "SiI disk position not found"); + goto fail1; + } + + /* Metadata valid. Print it. */ + g_raid_md_sii_print(meta); + G_RAID_DEBUG(1, "SiI disk position %d", disk_pos); + spare = (meta->type == SII_T_SPARE) ? 1 : 0; + +search: + /* Search for matching node. */ + sc = NULL; + mdi1 = NULL; + LIST_FOREACH(geom, &mp->geom, geom) { + sc = geom->softc; + if (sc == NULL) + continue; + if (sc->sc_stopping != 0) + continue; + if (sc->sc_md->mdo_class != md->mdo_class) + continue; + mdi1 = (struct g_raid_md_sii_object *)sc->sc_md; + if (spare) { + if (mdi1->mdio_incomplete) + break; + } else { + if (mdi1->mdio_location == meta->raid_location && + memcmp(&mdi1->mdio_timestamp, + &meta->timestamp, 6) == 0) + break; + } + } + + /* Found matching node. */ + if (geom != NULL) { + G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name); + result = G_RAID_MD_TASTE_EXISTING; + + } else if (spare) { /* Not found needy node -- left for later. */ + G_RAID_DEBUG(1, "Spare is not needed at this time"); + goto fail1; + + } else { /* Not found matching node -- create one. */ + result = G_RAID_MD_TASTE_NEW; + memcpy(&mdi->mdio_timestamp, &meta->timestamp, 6); + mdi->mdio_location = meta->raid_location; + snprintf(name, sizeof(name), "SiI-%02x%02x%02x%02x%02x%02x", + mdi->mdio_timestamp[5], mdi->mdio_timestamp[4], + mdi->mdio_timestamp[3], mdi->mdio_timestamp[2], + mdi->mdio_timestamp[1], mdi->mdio_timestamp[0]); + sc = g_raid_create_node(mp, name, md); + md->mdo_softc = sc; + geom = sc->sc_geom; + callout_init(&mdi->mdio_start_co, 1); + callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz, + g_raid_sii_go, sc); + mdi->mdio_rootmount = root_mount_hold("GRAID-SiI"); + G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount); + } + + rcp = g_new_consumer(geom); + g_attach(rcp, pp); + if (g_access(rcp, 1, 1, 1) != 0) + ; //goto fail1; + + g_topology_unlock(); + sx_xlock(&sc->sc_lock); + + pd = malloc(sizeof(*pd), M_MD_SII, M_WAITOK | M_ZERO); + pd->pd_meta = meta; + if (spare == 2) { + pd->pd_disk_pos = -3; + } else { + pd->pd_disk_pos = -1; + } + pd->pd_disk_size = pp->mediasize; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_consumer = rcp; + rcp->private = disk; + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + error = g_io_getattr("GEOM::kerneldump", rcp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, "Dumping not supported by %s: %d.", + rcp->provider->name, error); + + g_raid_md_sii_new_disk(disk); + + sx_xunlock(&sc->sc_lock); + g_topology_lock(); + *gp = geom; + return (result); +fail1: + free(meta, M_MD_SII); + return (G_RAID_MD_TASTE_FAIL); +} + +static int +g_raid_md_event_sii(struct g_raid_md_object *md, + struct g_raid_disk *disk, u_int event) +{ + struct g_raid_softc *sc; + struct g_raid_subdisk *sd; + struct g_raid_md_sii_object *mdi; + struct g_raid_md_sii_perdisk *pd; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_sii_object *)md; + if (disk == NULL) { + switch (event) { + case G_RAID_NODE_E_START: + if (!mdi->mdio_started) + g_raid_md_sii_start(sc); + return (0); + } + return (-1); + } + pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data; + switch (event) { + case G_RAID_DISK_E_DISCONNECTED: + /* If disk was assigned, just update statuses. */ + if (pd->pd_disk_pos >= 0) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + if (disk->d_consumer) { + g_raid_kill_consumer(sc, disk->d_consumer); + disk->d_consumer = NULL; + } + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NONE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, + G_RAID_EVENT_SUBDISK); + } + } else { + /* Otherwise -- delete. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); + g_raid_destroy_disk(disk); + } + + /* Write updated metadata to all disks. */ + g_raid_md_write_sii(md, NULL, NULL, NULL); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_sii_refill(sc); + return (0); + } + return (-2); +} + +static int +g_raid_md_ctl_sii(struct g_raid_md_object *md, + struct gctl_req *req) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_sii_object *mdi; + struct g_raid_md_sii_perdisk *pd; + struct g_consumer *cp; + struct g_provider *pp; + char arg[16]; + const char *verb, *volname, *levelname, *diskname; + int *nargs, *force; + off_t size, sectorsize, strip; + intmax_t *sizearg, *striparg; + int numdisks, i, len, level, qual, update; + int error; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_sii_object *)md; + verb = gctl_get_param(req, "verb", NULL); + nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); + error = 0; + if (strcmp(verb, "label") == 0) { + + if (*nargs < 4) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + volname = gctl_get_asciiparam(req, "arg1"); + if (volname == NULL) { + gctl_error(req, "No volume name."); + return (-2); + } + levelname = gctl_get_asciiparam(req, "arg2"); + if (levelname == NULL) { + gctl_error(req, "No RAID level."); + return (-3); + } + if (g_raid_volume_str2level(levelname, &level, &qual)) { + gctl_error(req, "Unknown RAID level '%s'.", levelname); + return (-4); + } + numdisks = *nargs - 3; + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (!g_raid_md_sii_supported(level, qual, numdisks, + force ? *force : 0)) { + gctl_error(req, "Unsupported RAID level " + "(0x%02x/0x%02x), or number of disks (%d).", + level, qual, numdisks); + return (-5); + } + + /* Search for disks, connect them and probe. */ + size = 0x7fffffffffffffffllu; + sectorsize = 0; + for (i = 0; i < numdisks; i++) { + snprintf(arg, sizeof(arg), "arg%d", i + 3); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -6; + break; + } + if (strcmp(diskname, "NONE") == 0) { + cp = NULL; + pp = NULL; + } else { + g_topology_lock(); + cp = g_raid_open_consumer(sc, diskname); + if (cp == NULL) { + gctl_error(req, "Can't open '%s'.", + diskname); + g_topology_unlock(); + error = -7; + break; + } + pp = cp->provider; + } + pd = malloc(sizeof(*pd), M_MD_SII, M_WAITOK | M_ZERO); + pd->pd_disk_pos = i; + disk = g_raid_create_disk(sc); + disk->d_md_data = (void *)pd; + disk->d_consumer = cp; + if (cp == NULL) + continue; + cp->private = disk; + g_topology_unlock(); + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, + "Dumping not supported by %s.", + cp->provider->name); + + pd->pd_disk_size = pp->mediasize; + if (size > pp->mediasize) + size = pp->mediasize; + if (sectorsize < pp->sectorsize) + sectorsize = pp->sectorsize; + } + if (error != 0) + return (error); + + /* Reserve space for metadata. */ + size -= 0x800 * sectorsize; + + /* Handle size argument. */ + len = sizeof(*sizearg); + sizearg = gctl_get_param(req, "size", &len); + if (sizearg != NULL && len == sizeof(*sizearg) && + *sizearg > 0) { + if (*sizearg > size) { + gctl_error(req, "Size too big %lld > %lld.", + (long long)*sizearg, (long long)size); + return (-9); + } + size = *sizearg; + } + + /* Handle strip argument. */ + strip = 131072; + len = sizeof(*striparg); + striparg = gctl_get_param(req, "strip", &len); + if (striparg != NULL && len == sizeof(*striparg) && + *striparg > 0) { + if (*striparg < sectorsize) { + gctl_error(req, "Strip size too small."); + return (-10); + } + if (*striparg % sectorsize != 0) { + gctl_error(req, "Incorrect strip size."); + return (-11); + } + if (strip > 65535 * sectorsize) { + gctl_error(req, "Strip size too big."); + return (-12); + } + strip = *striparg; + } + + /* Round size down to strip or sector. */ + if (level == G_RAID_VOLUME_RL_RAID1) + size -= (size % sectorsize); + else if (level == G_RAID_VOLUME_RL_RAID1E && + (numdisks & 1) != 0) + size -= (size % (2 * strip)); + else + size -= (size % strip); + if (size <= 0) { + gctl_error(req, "Size too small."); + return (-13); + } + if (size > 0xffffffffffffllu * sectorsize) { + gctl_error(req, "Size too big."); + return (-14); + } + + /* We have all we need, create things: volume, ... */ + mdi->mdio_total_disks = numdisks; + mdi->mdio_started = 1; + vol = g_raid_create_volume(sc, volname, -1); + vol->v_md_data = (void *)(intptr_t)0; + vol->v_raid_level = level; + vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE; + vol->v_strip_size = strip; + vol->v_disks_count = numdisks; + if (level == G_RAID_VOLUME_RL_RAID0 || + level == G_RAID_VOLUME_RL_CONCAT || + level == G_RAID_VOLUME_RL_SINGLE) + vol->v_mediasize = size * numdisks; + else if (level == G_RAID_VOLUME_RL_RAID1) + vol->v_mediasize = size; + else if (level == G_RAID_VOLUME_RL_RAID5) + vol->v_mediasize = size * (numdisks - 1); + else { /* RAID1E */ + vol->v_mediasize = ((size * numdisks) / strip / 2) * + strip; + } + vol->v_sectorsize = sectorsize; + g_raid_start_volume(vol); + + /* , and subdisks. */ + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data; + sd = &vol->v_subdisks[pd->pd_disk_pos]; + sd->sd_disk = disk; + sd->sd_offset = 0; + sd->sd_size = size; + TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next); + if (sd->sd_disk->d_consumer != NULL) { + g_raid_change_disk_state(disk, + G_RAID_DISK_S_ACTIVE); + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_ACTIVE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW, + G_RAID_EVENT_SUBDISK); + } else { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + } + } + + /* Write metadata based on created entities. */ + G_RAID_DEBUG1(0, sc, "Array started."); + g_raid_md_write_sii(md, NULL, NULL, NULL); + + /* Pickup any STALE/SPARE disks to refill array if needed. */ + g_raid_md_sii_refill(sc); + + g_raid_event_send(vol, G_RAID_VOLUME_E_START, + G_RAID_EVENT_VOLUME); + return (0); + } + if (strcmp(verb, "delete") == 0) { + + /* Check if some volume is still open. */ + force = gctl_get_paraml(req, "force", sizeof(*force)); + if (force != NULL && *force == 0 && + g_raid_nopens(sc) != 0) { + gctl_error(req, "Some volume is still open."); + return (-4); + } + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer) + sii_meta_erase(disk->d_consumer); + } + g_raid_destroy_node(sc, 0); + return (0); + } + if (strcmp(verb, "remove") == 0 || + strcmp(verb, "fail") == 0) { + if (*nargs < 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + for (i = 1; i < *nargs; i++) { + snprintf(arg, sizeof(arg), "arg%d", i); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -2; + break; + } + if (strncmp(diskname, "/dev/", 5) == 0) + diskname += 5; + + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + if (disk->d_consumer != NULL && + disk->d_consumer->provider != NULL && + strcmp(disk->d_consumer->provider->name, + diskname) == 0) + break; + } + if (disk == NULL) { + gctl_error(req, "Disk '%s' not found.", + diskname); + error = -3; + break; + } + + if (strcmp(verb, "fail") == 0) { + g_raid_md_fail_disk_sii(md, NULL, disk); + continue; + } + + pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data; + + /* Erase metadata on deleting disk. */ + sii_meta_erase(disk->d_consumer); + + /* If disk was assigned, just update statuses. */ + if (pd->pd_disk_pos >= 0) { + g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE); + if (disk->d_consumer) { + g_raid_kill_consumer(sc, disk->d_consumer); + disk->d_consumer = NULL; + } + TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_NONE); + g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED, + G_RAID_EVENT_SUBDISK); + } + } else { + /* Otherwise -- delete. */ + g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE); + g_raid_destroy_disk(disk); + } + } + + /* Write updated metadata to remaining disks. */ + g_raid_md_write_sii(md, NULL, NULL, NULL); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_sii_refill(sc); + return (error); + } + if (strcmp(verb, "insert") == 0) { + if (*nargs < 2) { + gctl_error(req, "Invalid number of arguments."); + return (-1); + } + update = 0; + for (i = 1; i < *nargs; i++) { + /* Get disk name. */ + snprintf(arg, sizeof(arg), "arg%d", i); + diskname = gctl_get_asciiparam(req, arg); + if (diskname == NULL) { + gctl_error(req, "No disk name (%s).", arg); + error = -3; + break; + } + + /* Try to find provider with specified name. */ + g_topology_lock(); + cp = g_raid_open_consumer(sc, diskname); + if (cp == NULL) { + gctl_error(req, "Can't open disk '%s'.", + diskname); + g_topology_unlock(); + error = -4; + break; + } + pp = cp->provider; + + pd = malloc(sizeof(*pd), M_MD_SII, M_WAITOK | M_ZERO); + pd->pd_disk_pos = -3; + pd->pd_disk_size = pp->mediasize; + + disk = g_raid_create_disk(sc); + disk->d_consumer = cp; + disk->d_consumer->private = disk; + disk->d_md_data = (void *)pd; + cp->private = disk; + g_topology_unlock(); + + /* Read kernel dumping information. */ + disk->d_kd.offset = 0; + disk->d_kd.length = OFF_MAX; + len = sizeof(disk->d_kd); + g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd); + if (disk->d_kd.di.dumper == NULL) + G_RAID_DEBUG1(2, sc, + "Dumping not supported by %s.", + cp->provider->name); + + /* Welcome the "new" disk. */ + update += g_raid_md_sii_start_disk(disk); + if (disk->d_state == G_RAID_DISK_S_SPARE) { + sii_meta_write_spare(cp); + g_raid_destroy_disk(disk); + } else if (disk->d_state != G_RAID_DISK_S_ACTIVE) { + gctl_error(req, "Disk '%s' doesn't fit.", + diskname); + g_raid_destroy_disk(disk); + error = -8; + break; + } + } + + /* Write new metadata if we changed something. */ + if (update) + g_raid_md_write_sii(md, NULL, NULL, NULL); + return (error); + } + gctl_error(req, "Command '%s' is not supported.", verb); + return (-100); +} + +static int +g_raid_md_write_sii(struct g_raid_md_object *md, struct g_raid_volume *tvol, + struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct g_raid_disk *disk; + struct g_raid_md_sii_object *mdi; + struct g_raid_md_sii_perdisk *pd; + struct sii_raid_conf *meta; + int i; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_sii_object *)md; + + if (sc->sc_stopping == G_RAID_DESTROY_HARD) + return (0); + + /* Bump generation. Newly written metadata may differ from previous. */ + mdi->mdio_generation++; + + /* There is only one volume. */ + vol = TAILQ_FIRST(&sc->sc_volumes); + + /* Fill global fields. */ + meta = malloc(sizeof(*meta), M_MD_SII, M_WAITOK | M_ZERO); + if (mdi->mdio_meta) + memcpy(meta, mdi->mdio_meta, sizeof(*meta)); + meta->total_sectors = vol->v_mediasize / vol->v_sectorsize; + meta->vendor_id = 0x1095; + meta->version_minor = 0; + meta->version_major = 2; + memcpy(&meta->timestamp, &mdi->mdio_timestamp, 6); + meta->strip_sectors = vol->v_strip_size / vol->v_sectorsize; + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0) { + meta->type = SII_T_RAID0; + meta->raid0_disks = vol->v_disks_count; + meta->raid1_disks = 0xff; + } else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) { + meta->type = SII_T_RAID1; + meta->raid0_disks = 0xff; + meta->raid1_disks = vol->v_disks_count; + } else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) { + meta->type = SII_T_RAID01; + meta->raid0_disks = vol->v_disks_count / 2; + meta->raid1_disks = 2; + } else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT || + vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE) { + meta->type = SII_T_JBOD; + meta->raid0_disks = vol->v_disks_count; + meta->raid1_disks = 0xff; + } else { + meta->type = SII_T_RAID5; + meta->raid0_disks = vol->v_disks_count; + meta->raid1_disks = 0xff; + } + meta->generation = mdi->mdio_generation; + meta->raid_status = vol->v_dirty ? SII_S_ONLINE : SII_S_AVAILABLE; + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if (sd->sd_state == G_RAID_SUBDISK_S_STALE || + sd->sd_state == G_RAID_SUBDISK_S_RESYNC) + meta->raid_status = SII_S_ONLINE; + } + meta->raid_location = mdi->mdio_location; + sii_meta_put_name(meta, vol->v_name); + + /* We are done. Print meta data and store them to disks. */ + if (mdi->mdio_meta != NULL) + free(mdi->mdio_meta, M_MD_SII); + mdi->mdio_meta = meta; + i = 0; + TAILQ_FOREACH(disk, &sc->sc_disks, d_next) { + pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data; + if (disk->d_state != G_RAID_DISK_S_ACTIVE) + continue; + if (pd->pd_meta != NULL) { + free(pd->pd_meta, M_MD_SII); + pd->pd_meta = NULL; + } + pd->pd_meta = sii_meta_copy(meta); + if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) { + if (sd->sd_state < G_RAID_SUBDISK_S_NEW) + pd->pd_meta->disk_status = SII_S_DROPPED; + else if (sd->sd_state < G_RAID_SUBDISK_S_STALE) { + pd->pd_meta->disk_status = SII_S_REBUILD; + pd->pd_meta->rebuild_lba = + sd->sd_rebuild_pos / vol->v_sectorsize; + } else + pd->pd_meta->disk_status = SII_S_CURRENT; + if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1) { + pd->pd_meta->disk_number = sd->sd_pos; + pd->pd_meta->raid0_ident = 0xff; + pd->pd_meta->raid1_ident = 0; + } else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) { + pd->pd_meta->disk_number = sd->sd_pos / meta->raid1_disks; + pd->pd_meta->raid0_ident = sd->sd_pos % meta->raid1_disks; + pd->pd_meta->raid1_ident = sd->sd_pos / meta->raid1_disks; + } else { + pd->pd_meta->disk_number = sd->sd_pos; + pd->pd_meta->raid0_ident = 0; + pd->pd_meta->raid1_ident = 0xff; + } + } + G_RAID_DEBUG(1, "Writing SiI metadata to %s", + g_raid_get_diskname(disk)); + g_raid_md_sii_print(pd->pd_meta); + sii_meta_write(disk->d_consumer, pd->pd_meta); + } + return (0); +} + +static int +g_raid_md_fail_disk_sii(struct g_raid_md_object *md, + struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk) +{ + struct g_raid_softc *sc; + struct g_raid_md_sii_object *mdi; + struct g_raid_md_sii_perdisk *pd; + struct g_raid_subdisk *sd; + + sc = md->mdo_softc; + mdi = (struct g_raid_md_sii_object *)md; + pd = (struct g_raid_md_sii_perdisk *)tdisk->d_md_data; + + /* We can't fail disk that is not a part of array now. */ + if (pd->pd_disk_pos < 0) + return (-1); + + /* + * Mark disk as failed in metadata and try to write that metadata + * to the disk itself to prevent it's later resurrection as STALE. + */ + if (tdisk->d_consumer) { + if (pd->pd_meta) { + pd->pd_meta->disk_status = SII_S_REMOVED; + sii_meta_write(tdisk->d_consumer, pd->pd_meta); + } else + sii_meta_erase(tdisk->d_consumer); + } + + /* Change states. */ + g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED); + TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) { + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_FAILED); + g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED, + G_RAID_EVENT_SUBDISK); + } + + /* Write updated metadata to remaining disks. */ + g_raid_md_write_sii(md, NULL, NULL, tdisk); + + /* Check if anything left except placeholders. */ + if (g_raid_ndisks(sc, -1) == + g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE)) + g_raid_destroy_node(sc, 0); + else + g_raid_md_sii_refill(sc); + return (0); +} + +static int +g_raid_md_free_disk_sii(struct g_raid_md_object *md, + struct g_raid_disk *disk) +{ + struct g_raid_md_sii_perdisk *pd; + + pd = (struct g_raid_md_sii_perdisk *)disk->d_md_data; + if (pd->pd_meta != NULL) { + free(pd->pd_meta, M_MD_SII); + pd->pd_meta = NULL; + } + free(pd, M_MD_SII); + disk->d_md_data = NULL; + return (0); +} + +static int +g_raid_md_free_sii(struct g_raid_md_object *md) +{ + struct g_raid_md_sii_object *mdi; + + mdi = (struct g_raid_md_sii_object *)md; + if (!mdi->mdio_started) { + mdi->mdio_started = 0; + callout_stop(&mdi->mdio_start_co); + G_RAID_DEBUG1(1, md->mdo_softc, + "root_mount_rel %p", mdi->mdio_rootmount); + root_mount_rel(mdi->mdio_rootmount); + mdi->mdio_rootmount = NULL; + } + if (mdi->mdio_meta != NULL) { + free(mdi->mdio_meta, M_MD_SII); + mdi->mdio_meta = NULL; + } + return (0); +} + +G_RAID_MD_DECLARE(g_raid_md_sii); diff --git a/sys/geom/raid/tr_concat.c b/sys/geom/raid/tr_concat.c new file mode 100644 index 000000000000..c5f2913656c4 --- /dev/null +++ b/sys/geom/raid/tr_concat.c @@ -0,0 +1,343 @@ +/*- + * Copyright (c) 2010 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "geom/raid/g_raid.h" +#include "g_raid_tr_if.h" + +static MALLOC_DEFINE(M_TR_CONCAT, "tr_concat_data", "GEOM_RAID CONCAT data"); + +struct g_raid_tr_concat_object { + struct g_raid_tr_object trso_base; + int trso_starting; + int trso_stopped; +}; + +static g_raid_tr_taste_t g_raid_tr_taste_concat; +static g_raid_tr_event_t g_raid_tr_event_concat; +static g_raid_tr_start_t g_raid_tr_start_concat; +static g_raid_tr_stop_t g_raid_tr_stop_concat; +static g_raid_tr_iostart_t g_raid_tr_iostart_concat; +static g_raid_tr_iodone_t g_raid_tr_iodone_concat; +static g_raid_tr_kerneldump_t g_raid_tr_kerneldump_concat; +static g_raid_tr_free_t g_raid_tr_free_concat; + +static kobj_method_t g_raid_tr_concat_methods[] = { + KOBJMETHOD(g_raid_tr_taste, g_raid_tr_taste_concat), + KOBJMETHOD(g_raid_tr_event, g_raid_tr_event_concat), + KOBJMETHOD(g_raid_tr_start, g_raid_tr_start_concat), + KOBJMETHOD(g_raid_tr_stop, g_raid_tr_stop_concat), + KOBJMETHOD(g_raid_tr_iostart, g_raid_tr_iostart_concat), + KOBJMETHOD(g_raid_tr_iodone, g_raid_tr_iodone_concat), + KOBJMETHOD(g_raid_tr_kerneldump, g_raid_tr_kerneldump_concat), + KOBJMETHOD(g_raid_tr_free, g_raid_tr_free_concat), + { 0, 0 } +}; + +static struct g_raid_tr_class g_raid_tr_concat_class = { + "CONCAT", + g_raid_tr_concat_methods, + sizeof(struct g_raid_tr_concat_object), + .trc_priority = 50 +}; + +static int +g_raid_tr_taste_concat(struct g_raid_tr_object *tr, struct g_raid_volume *volume) +{ + struct g_raid_tr_concat_object *trs; + + trs = (struct g_raid_tr_concat_object *)tr; + if (tr->tro_volume->v_raid_level != G_RAID_VOLUME_RL_SINGLE && + tr->tro_volume->v_raid_level != G_RAID_VOLUME_RL_CONCAT && + !(tr->tro_volume->v_disks_count == 1 && + tr->tro_volume->v_raid_level != G_RAID_VOLUME_RL_UNKNOWN)) + return (G_RAID_TR_TASTE_FAIL); + trs->trso_starting = 1; + return (G_RAID_TR_TASTE_SUCCEED); +} + +static int +g_raid_tr_update_state_concat(struct g_raid_volume *vol) +{ + struct g_raid_tr_concat_object *trs; + struct g_raid_softc *sc; + off_t size; + u_int s; + int i, n, f; + + sc = vol->v_softc; + trs = (struct g_raid_tr_concat_object *)vol->v_tr; + if (trs->trso_stopped) + s = G_RAID_VOLUME_S_STOPPED; + else if (trs->trso_starting) + s = G_RAID_VOLUME_S_STARTING; + else { + n = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE); + f = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_FAILED); + if (n + f == vol->v_disks_count) { + if (f == 0) + s = G_RAID_VOLUME_S_OPTIMAL; + else + s = G_RAID_VOLUME_S_SUBOPTIMAL; + } else + s = G_RAID_VOLUME_S_BROKEN; + } + if (s != vol->v_state) { + + /* + * Some metadata modules may not know CONCAT volume + * mediasize until all disks connected. Recalculate. + */ + if (G_RAID_VOLUME_S_ALIVE(s) && + !G_RAID_VOLUME_S_ALIVE(vol->v_state)) { + size = 0; + for (i = 0; i < vol->v_disks_count; i++) { + if (vol->v_subdisks[i].sd_state != + G_RAID_SUBDISK_S_NONE) + size += vol->v_subdisks[i].sd_size; + } + vol->v_mediasize = size; + } + + g_raid_event_send(vol, G_RAID_VOLUME_S_ALIVE(s) ? + G_RAID_VOLUME_E_UP : G_RAID_VOLUME_E_DOWN, + G_RAID_EVENT_VOLUME); + g_raid_change_volume_state(vol, s); + if (!trs->trso_starting && !trs->trso_stopped) + g_raid_write_metadata(sc, vol, NULL, NULL); + } + return (0); +} + +static int +g_raid_tr_event_concat(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd, u_int event) +{ + struct g_raid_tr_concat_object *trs; + struct g_raid_softc *sc; + struct g_raid_volume *vol; + int state; + + trs = (struct g_raid_tr_concat_object *)tr; + vol = tr->tro_volume; + sc = vol->v_softc; + + state = sd->sd_state; + if (state != G_RAID_SUBDISK_S_NONE && + state != G_RAID_SUBDISK_S_FAILED && + state != G_RAID_SUBDISK_S_ACTIVE) { + G_RAID_DEBUG1(1, sc, + "Promote subdisk %s:%d from %s to ACTIVE.", + vol->v_name, sd->sd_pos, + g_raid_subdisk_state2str(sd->sd_state)); + g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_ACTIVE); + } + if (state != sd->sd_state && + !trs->trso_starting && !trs->trso_stopped) + g_raid_write_metadata(sc, vol, sd, NULL); + g_raid_tr_update_state_concat(vol); + return (0); +} + +static int +g_raid_tr_start_concat(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_concat_object *trs; + struct g_raid_volume *vol; + + trs = (struct g_raid_tr_concat_object *)tr; + vol = tr->tro_volume; + trs->trso_starting = 0; + g_raid_tr_update_state_concat(vol); + return (0); +} + +static int +g_raid_tr_stop_concat(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_concat_object *trs; + struct g_raid_volume *vol; + + trs = (struct g_raid_tr_concat_object *)tr; + vol = tr->tro_volume; + trs->trso_starting = 0; + trs->trso_stopped = 1; + g_raid_tr_update_state_concat(vol); + return (0); +} + +static void +g_raid_tr_iostart_concat(struct g_raid_tr_object *tr, struct bio *bp) +{ + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct bio_queue_head queue; + struct bio *cbp; + char *addr; + off_t offset, length, remain; + u_int no; + + vol = tr->tro_volume; + if (vol->v_state != G_RAID_VOLUME_S_OPTIMAL && + vol->v_state != G_RAID_VOLUME_S_SUBOPTIMAL) { + g_raid_iodone(bp, EIO); + return; + } + if (bp->bio_cmd == BIO_FLUSH) { + g_raid_tr_flush_common(tr, bp); + return; + } + + offset = bp->bio_offset; + remain = bp->bio_length; + addr = bp->bio_data; + no = 0; + while (no < vol->v_disks_count && + offset >= vol->v_subdisks[no].sd_size) { + offset -= vol->v_subdisks[no].sd_size; + no++; + } + KASSERT(no < vol->v_disks_count, + ("Request starts after volume end (%ju)", bp->bio_offset)); + bioq_init(&queue); + do { + sd = &vol->v_subdisks[no]; + length = MIN(sd->sd_size - offset, remain); + cbp = g_clone_bio(bp); + if (cbp == NULL) + goto failure; + cbp->bio_offset = offset; + cbp->bio_data = addr; + cbp->bio_length = length; + cbp->bio_caller1 = sd; + bioq_insert_tail(&queue, cbp); + remain -= length; + addr += length; + offset = 0; + no++; + KASSERT(no < vol->v_disks_count || remain == 0, + ("Request ends after volume end (%ju, %ju)", + bp->bio_offset, bp->bio_length)); + } while (remain > 0); + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + sd = cbp->bio_caller1; + cbp->bio_caller1 = NULL; + g_raid_subdisk_iostart(sd, cbp); + } + return; +failure: + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + g_destroy_bio(cbp); + } + if (bp->bio_error == 0) + bp->bio_error = ENOMEM; + g_raid_iodone(bp, bp->bio_error); +} + +static int +g_raid_tr_kerneldump_concat(struct g_raid_tr_object *tr, + void *virtual, vm_offset_t physical, off_t boffset, size_t blength) +{ + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + char *addr; + off_t offset, length, remain; + int error, no; + + vol = tr->tro_volume; + if (vol->v_state != G_RAID_VOLUME_S_OPTIMAL) + return (ENXIO); + + offset = boffset; + remain = blength; + addr = virtual; + no = 0; + while (no < vol->v_disks_count && + offset >= vol->v_subdisks[no].sd_size) { + offset -= vol->v_subdisks[no].sd_size; + no++; + } + KASSERT(no < vol->v_disks_count, + ("Request starts after volume end (%ju)", boffset)); + do { + sd = &vol->v_subdisks[no]; + length = MIN(sd->sd_size - offset, remain); + error = g_raid_subdisk_kerneldump(&vol->v_subdisks[no], + addr, 0, offset, length); + if (error != 0) + return (error); + remain -= length; + addr += length; + offset = 0; + no++; + KASSERT(no < vol->v_disks_count || remain == 0, + ("Request ends after volume end (%ju, %zu)", + boffset, blength)); + } while (remain > 0); + return (0); +} + +static void +g_raid_tr_iodone_concat(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd,struct bio *bp) +{ + struct bio *pbp; + + pbp = bp->bio_parent; + if (pbp->bio_error == 0) + pbp->bio_error = bp->bio_error; + g_destroy_bio(bp); + pbp->bio_inbed++; + if (pbp->bio_children == pbp->bio_inbed) { + pbp->bio_completed = pbp->bio_length; + g_raid_iodone(pbp, bp->bio_error); + } +} + +static int +g_raid_tr_free_concat(struct g_raid_tr_object *tr) +{ + + return (0); +} + +G_RAID_TR_DECLARE(g_raid_tr_concat); diff --git a/sys/geom/raid/tr_raid0.c b/sys/geom/raid/tr_raid0.c new file mode 100644 index 000000000000..0fb45a6da4d3 --- /dev/null +++ b/sys/geom/raid/tr_raid0.c @@ -0,0 +1,326 @@ +/*- + * Copyright (c) 2010 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "geom/raid/g_raid.h" +#include "g_raid_tr_if.h" + +static MALLOC_DEFINE(M_TR_RAID0, "tr_raid0_data", "GEOM_RAID RAID0 data"); + +struct g_raid_tr_raid0_object { + struct g_raid_tr_object trso_base; + int trso_starting; + int trso_stopped; +}; + +static g_raid_tr_taste_t g_raid_tr_taste_raid0; +static g_raid_tr_event_t g_raid_tr_event_raid0; +static g_raid_tr_start_t g_raid_tr_start_raid0; +static g_raid_tr_stop_t g_raid_tr_stop_raid0; +static g_raid_tr_iostart_t g_raid_tr_iostart_raid0; +static g_raid_tr_iodone_t g_raid_tr_iodone_raid0; +static g_raid_tr_kerneldump_t g_raid_tr_kerneldump_raid0; +static g_raid_tr_free_t g_raid_tr_free_raid0; + +static kobj_method_t g_raid_tr_raid0_methods[] = { + KOBJMETHOD(g_raid_tr_taste, g_raid_tr_taste_raid0), + KOBJMETHOD(g_raid_tr_event, g_raid_tr_event_raid0), + KOBJMETHOD(g_raid_tr_start, g_raid_tr_start_raid0), + KOBJMETHOD(g_raid_tr_stop, g_raid_tr_stop_raid0), + KOBJMETHOD(g_raid_tr_iostart, g_raid_tr_iostart_raid0), + KOBJMETHOD(g_raid_tr_iodone, g_raid_tr_iodone_raid0), + KOBJMETHOD(g_raid_tr_kerneldump, g_raid_tr_kerneldump_raid0), + KOBJMETHOD(g_raid_tr_free, g_raid_tr_free_raid0), + { 0, 0 } +}; + +static struct g_raid_tr_class g_raid_tr_raid0_class = { + "RAID0", + g_raid_tr_raid0_methods, + sizeof(struct g_raid_tr_raid0_object), + .trc_priority = 100 +}; + +static int +g_raid_tr_taste_raid0(struct g_raid_tr_object *tr, struct g_raid_volume *volume) +{ + struct g_raid_tr_raid0_object *trs; + + trs = (struct g_raid_tr_raid0_object *)tr; + if (tr->tro_volume->v_raid_level != G_RAID_VOLUME_RL_RAID0 || + tr->tro_volume->v_raid_level_qualifier != G_RAID_VOLUME_RLQ_NONE) + return (G_RAID_TR_TASTE_FAIL); + trs->trso_starting = 1; + return (G_RAID_TR_TASTE_SUCCEED); +} + +static int +g_raid_tr_update_state_raid0(struct g_raid_volume *vol) +{ + struct g_raid_tr_raid0_object *trs; + struct g_raid_softc *sc; + u_int s; + int n, f; + + sc = vol->v_softc; + trs = (struct g_raid_tr_raid0_object *)vol->v_tr; + if (trs->trso_stopped) + s = G_RAID_VOLUME_S_STOPPED; + else if (trs->trso_starting) + s = G_RAID_VOLUME_S_STARTING; + else { + n = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE); + f = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_FAILED); + if (n + f == vol->v_disks_count) { + if (f == 0) + s = G_RAID_VOLUME_S_OPTIMAL; + else + s = G_RAID_VOLUME_S_SUBOPTIMAL; + } else + s = G_RAID_VOLUME_S_BROKEN; + } + if (s != vol->v_state) { + g_raid_event_send(vol, G_RAID_VOLUME_S_ALIVE(s) ? + G_RAID_VOLUME_E_UP : G_RAID_VOLUME_E_DOWN, + G_RAID_EVENT_VOLUME); + g_raid_change_volume_state(vol, s); + if (!trs->trso_starting && !trs->trso_stopped) + g_raid_write_metadata(sc, vol, NULL, NULL); + } + return (0); +} + +static int +g_raid_tr_event_raid0(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd, u_int event) +{ + struct g_raid_tr_raid0_object *trs; + struct g_raid_softc *sc; + struct g_raid_volume *vol; + int state; + + trs = (struct g_raid_tr_raid0_object *)tr; + vol = tr->tro_volume; + sc = vol->v_softc; + + state = sd->sd_state; + if (state != G_RAID_SUBDISK_S_NONE && + state != G_RAID_SUBDISK_S_FAILED && + state != G_RAID_SUBDISK_S_ACTIVE) { + G_RAID_DEBUG1(1, sc, + "Promote subdisk %s:%d from %s to ACTIVE.", + vol->v_name, sd->sd_pos, + g_raid_subdisk_state2str(sd->sd_state)); + g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_ACTIVE); + } + if (state != sd->sd_state && + !trs->trso_starting && !trs->trso_stopped) + g_raid_write_metadata(sc, vol, sd, NULL); + g_raid_tr_update_state_raid0(vol); + return (0); +} + +static int +g_raid_tr_start_raid0(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid0_object *trs; + struct g_raid_volume *vol; + + trs = (struct g_raid_tr_raid0_object *)tr; + vol = tr->tro_volume; + trs->trso_starting = 0; + g_raid_tr_update_state_raid0(vol); + return (0); +} + +static int +g_raid_tr_stop_raid0(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid0_object *trs; + struct g_raid_volume *vol; + + trs = (struct g_raid_tr_raid0_object *)tr; + vol = tr->tro_volume; + trs->trso_starting = 0; + trs->trso_stopped = 1; + g_raid_tr_update_state_raid0(vol); + return (0); +} + +static void +g_raid_tr_iostart_raid0(struct g_raid_tr_object *tr, struct bio *bp) +{ + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct bio_queue_head queue; + struct bio *cbp; + char *addr; + off_t offset, start, length, nstripe, remain; + u_int no, strip_size; + + vol = tr->tro_volume; + if (vol->v_state != G_RAID_VOLUME_S_OPTIMAL && + vol->v_state != G_RAID_VOLUME_S_SUBOPTIMAL) { + g_raid_iodone(bp, EIO); + return; + } + if (bp->bio_cmd == BIO_FLUSH) { + g_raid_tr_flush_common(tr, bp); + return; + } + addr = bp->bio_data; + strip_size = vol->v_strip_size; + + /* Stripe number. */ + nstripe = bp->bio_offset / strip_size; + /* Start position in stripe. */ + start = bp->bio_offset % strip_size; + /* Disk number. */ + no = nstripe % vol->v_disks_count; + /* Stripe start position in disk. */ + offset = (nstripe / vol->v_disks_count) * strip_size; + /* Length of data to operate. */ + remain = bp->bio_length; + + bioq_init(&queue); + do { + length = MIN(strip_size - start, remain); + cbp = g_clone_bio(bp); + if (cbp == NULL) + goto failure; + cbp->bio_offset = offset + start; + cbp->bio_data = addr; + cbp->bio_length = length; + cbp->bio_caller1 = &vol->v_subdisks[no]; + bioq_insert_tail(&queue, cbp); + if (++no >= vol->v_disks_count) { + no = 0; + offset += strip_size; + } + remain -= length; + addr += length; + start = 0; + } while (remain > 0); + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + sd = cbp->bio_caller1; + cbp->bio_caller1 = NULL; + g_raid_subdisk_iostart(sd, cbp); + } + return; +failure: + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + g_destroy_bio(cbp); + } + if (bp->bio_error == 0) + bp->bio_error = ENOMEM; + g_raid_iodone(bp, bp->bio_error); +} + +static int +g_raid_tr_kerneldump_raid0(struct g_raid_tr_object *tr, + void *virtual, vm_offset_t physical, off_t boffset, size_t blength) +{ + struct g_raid_volume *vol; + char *addr; + off_t offset, start, length, nstripe, remain; + u_int no, strip_size; + int error; + + vol = tr->tro_volume; + if (vol->v_state != G_RAID_VOLUME_S_OPTIMAL) + return (ENXIO); + addr = virtual; + strip_size = vol->v_strip_size; + + /* Stripe number. */ + nstripe = boffset / strip_size; + /* Start position in stripe. */ + start = boffset % strip_size; + /* Disk number. */ + no = nstripe % vol->v_disks_count; + /* Stripe tart position in disk. */ + offset = (nstripe / vol->v_disks_count) * strip_size; + /* Length of data to operate. */ + remain = blength; + + do { + length = MIN(strip_size - start, remain); + error = g_raid_subdisk_kerneldump(&vol->v_subdisks[no], + addr, 0, offset + start, length); + if (error != 0) + return (error); + if (++no >= vol->v_disks_count) { + no = 0; + offset += strip_size; + } + remain -= length; + addr += length; + start = 0; + } while (remain > 0); + return (0); +} + +static void +g_raid_tr_iodone_raid0(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd,struct bio *bp) +{ + struct bio *pbp; + + pbp = bp->bio_parent; + if (pbp->bio_error == 0) + pbp->bio_error = bp->bio_error; + g_destroy_bio(bp); + pbp->bio_inbed++; + if (pbp->bio_children == pbp->bio_inbed) { + pbp->bio_completed = pbp->bio_length; + g_raid_iodone(pbp, bp->bio_error); + } +} + +static int +g_raid_tr_free_raid0(struct g_raid_tr_object *tr) +{ + + return (0); +} + +G_RAID_TR_DECLARE(g_raid_tr_raid0); diff --git a/sys/geom/raid/tr_raid1.c b/sys/geom/raid/tr_raid1.c new file mode 100644 index 000000000000..b5e495398c0a --- /dev/null +++ b/sys/geom/raid/tr_raid1.c @@ -0,0 +1,993 @@ +/*- + * Copyright (c) 2010 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "geom/raid/g_raid.h" +#include "g_raid_tr_if.h" + +SYSCTL_DECL(_kern_geom_raid); +SYSCTL_NODE(_kern_geom_raid, OID_AUTO, raid1, CTLFLAG_RW, 0, + "RAID1 parameters"); + +#define RAID1_REBUILD_SLAB (1 << 20) /* One transation in a rebuild */ +static int g_raid1_rebuild_slab = RAID1_REBUILD_SLAB; +TUNABLE_INT("kern.geom.raid.raid1.rebuild_slab_size", + &g_raid1_rebuild_slab); +SYSCTL_UINT(_kern_geom_raid_raid1, OID_AUTO, rebuild_slab_size, CTLFLAG_RW, + &g_raid1_rebuild_slab, 0, + "Amount of the disk to rebuild each read/write cycle of the rebuild."); + +#define RAID1_REBUILD_FAIR_IO 20 /* use 1/x of the available I/O */ +static int g_raid1_rebuild_fair_io = RAID1_REBUILD_FAIR_IO; +TUNABLE_INT("kern.geom.raid.raid1.rebuild_fair_io", + &g_raid1_rebuild_fair_io); +SYSCTL_UINT(_kern_geom_raid_raid1, OID_AUTO, rebuild_fair_io, CTLFLAG_RW, + &g_raid1_rebuild_fair_io, 0, + "Fraction of the I/O bandwidth to use when disk busy for rebuild."); + +#define RAID1_REBUILD_CLUSTER_IDLE 100 +static int g_raid1_rebuild_cluster_idle = RAID1_REBUILD_CLUSTER_IDLE; +TUNABLE_INT("kern.geom.raid.raid1.rebuild_cluster_idle", + &g_raid1_rebuild_cluster_idle); +SYSCTL_UINT(_kern_geom_raid_raid1, OID_AUTO, rebuild_cluster_idle, CTLFLAG_RW, + &g_raid1_rebuild_cluster_idle, 0, + "Number of slabs to do each time we trigger a rebuild cycle"); + +#define RAID1_REBUILD_META_UPDATE 1024 /* update meta data every 1GB or so */ +static int g_raid1_rebuild_meta_update = RAID1_REBUILD_META_UPDATE; +TUNABLE_INT("kern.geom.raid.raid1.rebuild_meta_update", + &g_raid1_rebuild_meta_update); +SYSCTL_UINT(_kern_geom_raid_raid1, OID_AUTO, rebuild_meta_update, CTLFLAG_RW, + &g_raid1_rebuild_meta_update, 0, + "When to update the meta data."); + +static MALLOC_DEFINE(M_TR_RAID1, "tr_raid1_data", "GEOM_RAID RAID1 data"); + +#define TR_RAID1_NONE 0 +#define TR_RAID1_REBUILD 1 +#define TR_RAID1_RESYNC 2 + +#define TR_RAID1_F_DOING_SOME 0x1 +#define TR_RAID1_F_LOCKED 0x2 +#define TR_RAID1_F_ABORT 0x4 + +struct g_raid_tr_raid1_object { + struct g_raid_tr_object trso_base; + int trso_starting; + int trso_stopping; + int trso_type; + int trso_recover_slabs; /* slabs before rest */ + int trso_fair_io; + int trso_meta_update; + int trso_flags; + struct g_raid_subdisk *trso_failed_sd; /* like per volume */ + void *trso_buffer; /* Buffer space */ + struct bio trso_bio; +}; + +static g_raid_tr_taste_t g_raid_tr_taste_raid1; +static g_raid_tr_event_t g_raid_tr_event_raid1; +static g_raid_tr_start_t g_raid_tr_start_raid1; +static g_raid_tr_stop_t g_raid_tr_stop_raid1; +static g_raid_tr_iostart_t g_raid_tr_iostart_raid1; +static g_raid_tr_iodone_t g_raid_tr_iodone_raid1; +static g_raid_tr_kerneldump_t g_raid_tr_kerneldump_raid1; +static g_raid_tr_locked_t g_raid_tr_locked_raid1; +static g_raid_tr_idle_t g_raid_tr_idle_raid1; +static g_raid_tr_free_t g_raid_tr_free_raid1; + +static kobj_method_t g_raid_tr_raid1_methods[] = { + KOBJMETHOD(g_raid_tr_taste, g_raid_tr_taste_raid1), + KOBJMETHOD(g_raid_tr_event, g_raid_tr_event_raid1), + KOBJMETHOD(g_raid_tr_start, g_raid_tr_start_raid1), + KOBJMETHOD(g_raid_tr_stop, g_raid_tr_stop_raid1), + KOBJMETHOD(g_raid_tr_iostart, g_raid_tr_iostart_raid1), + KOBJMETHOD(g_raid_tr_iodone, g_raid_tr_iodone_raid1), + KOBJMETHOD(g_raid_tr_kerneldump, g_raid_tr_kerneldump_raid1), + KOBJMETHOD(g_raid_tr_locked, g_raid_tr_locked_raid1), + KOBJMETHOD(g_raid_tr_idle, g_raid_tr_idle_raid1), + KOBJMETHOD(g_raid_tr_free, g_raid_tr_free_raid1), + { 0, 0 } +}; + +static struct g_raid_tr_class g_raid_tr_raid1_class = { + "RAID1", + g_raid_tr_raid1_methods, + sizeof(struct g_raid_tr_raid1_object), + .trc_priority = 100 +}; + +static void g_raid_tr_raid1_rebuild_abort(struct g_raid_tr_object *tr); +static void g_raid_tr_raid1_maybe_rebuild(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd); + +static int +g_raid_tr_taste_raid1(struct g_raid_tr_object *tr, struct g_raid_volume *vol) +{ + struct g_raid_tr_raid1_object *trs; + + trs = (struct g_raid_tr_raid1_object *)tr; + if (tr->tro_volume->v_raid_level != G_RAID_VOLUME_RL_RAID1 || + tr->tro_volume->v_raid_level_qualifier != G_RAID_VOLUME_RLQ_NONE) + return (G_RAID_TR_TASTE_FAIL); + trs->trso_starting = 1; + return (G_RAID_TR_TASTE_SUCCEED); +} + +static int +g_raid_tr_update_state_raid1(struct g_raid_volume *vol, + struct g_raid_subdisk *sd) +{ + struct g_raid_tr_raid1_object *trs; + struct g_raid_softc *sc; + struct g_raid_subdisk *tsd, *bestsd; + u_int s; + int i, na, ns; + + sc = vol->v_softc; + trs = (struct g_raid_tr_raid1_object *)vol->v_tr; + if (trs->trso_stopping && + (trs->trso_flags & TR_RAID1_F_DOING_SOME) == 0) + s = G_RAID_VOLUME_S_STOPPED; + else if (trs->trso_starting) + s = G_RAID_VOLUME_S_STARTING; + else { + /* Make sure we have at least one ACTIVE disk. */ + na = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE); + if (na == 0) { + /* + * Critical situation! We have no any active disk! + * Choose the best disk we have to make it active. + */ + bestsd = &vol->v_subdisks[0]; + for (i = 1; i < vol->v_disks_count; i++) { + tsd = &vol->v_subdisks[i]; + if (tsd->sd_state > bestsd->sd_state) + bestsd = tsd; + else if (tsd->sd_state == bestsd->sd_state && + (tsd->sd_state == G_RAID_SUBDISK_S_REBUILD || + tsd->sd_state == G_RAID_SUBDISK_S_RESYNC) && + tsd->sd_rebuild_pos > bestsd->sd_rebuild_pos) + bestsd = tsd; + } + if (bestsd->sd_state >= G_RAID_SUBDISK_S_UNINITIALIZED) { + /* We found reasonable candidate. */ + G_RAID_DEBUG1(1, sc, + "Promote subdisk %s:%d from %s to ACTIVE.", + vol->v_name, bestsd->sd_pos, + g_raid_subdisk_state2str(bestsd->sd_state)); + g_raid_change_subdisk_state(bestsd, + G_RAID_SUBDISK_S_ACTIVE); + g_raid_write_metadata(sc, + vol, bestsd, bestsd->sd_disk); + } + } + na = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE); + ns = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_STALE) + + g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_RESYNC); + if (na == vol->v_disks_count) + s = G_RAID_VOLUME_S_OPTIMAL; + else if (na + ns == vol->v_disks_count) + s = G_RAID_VOLUME_S_SUBOPTIMAL; + else if (na > 0) + s = G_RAID_VOLUME_S_DEGRADED; + else + s = G_RAID_VOLUME_S_BROKEN; + g_raid_tr_raid1_maybe_rebuild(vol->v_tr, sd); + } + if (s != vol->v_state) { + g_raid_event_send(vol, G_RAID_VOLUME_S_ALIVE(s) ? + G_RAID_VOLUME_E_UP : G_RAID_VOLUME_E_DOWN, + G_RAID_EVENT_VOLUME); + g_raid_change_volume_state(vol, s); + if (!trs->trso_starting && !trs->trso_stopping) + g_raid_write_metadata(sc, vol, NULL, NULL); + } + return (0); +} + +static void +g_raid_tr_raid1_fail_disk(struct g_raid_softc *sc, struct g_raid_subdisk *sd, + struct g_raid_disk *disk) +{ + /* + * We don't fail the last disk in the pack, since it still has decent + * data on it and that's better than failing the disk if it is the root + * file system. + * + * XXX should this be controlled via a tunable? It makes sense for + * the volume that has / on it. I can't think of a case where we'd + * want the volume to go away on this kind of event. + */ + if (g_raid_nsubdisks(sd->sd_volume, G_RAID_SUBDISK_S_ACTIVE) == 1 && + g_raid_get_subdisk(sd->sd_volume, G_RAID_SUBDISK_S_ACTIVE) == sd) + return; + g_raid_fail_disk(sc, sd, disk); +} + +static void +g_raid_tr_raid1_rebuild_some(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1_object *trs; + struct g_raid_subdisk *sd, *good_sd; + struct bio *bp; + + trs = (struct g_raid_tr_raid1_object *)tr; + if (trs->trso_flags & TR_RAID1_F_DOING_SOME) + return; + sd = trs->trso_failed_sd; + good_sd = g_raid_get_subdisk(sd->sd_volume, G_RAID_SUBDISK_S_ACTIVE); + if (good_sd == NULL) { + g_raid_tr_raid1_rebuild_abort(tr); + return; + } + bp = &trs->trso_bio; + memset(bp, 0, sizeof(*bp)); + bp->bio_offset = sd->sd_rebuild_pos; + bp->bio_length = MIN(g_raid1_rebuild_slab, + sd->sd_size - sd->sd_rebuild_pos); + bp->bio_data = trs->trso_buffer; + bp->bio_cmd = BIO_READ; + bp->bio_cflags = G_RAID_BIO_FLAG_SYNC; + bp->bio_caller1 = good_sd; + trs->trso_flags |= TR_RAID1_F_DOING_SOME; + trs->trso_flags |= TR_RAID1_F_LOCKED; + g_raid_lock_range(sd->sd_volume, /* Lock callback starts I/O */ + bp->bio_offset, bp->bio_length, NULL, bp); +} + +static void +g_raid_tr_raid1_rebuild_done(struct g_raid_tr_raid1_object *trs) +{ + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + + vol = trs->trso_base.tro_volume; + sd = trs->trso_failed_sd; + g_raid_write_metadata(vol->v_softc, vol, sd, sd->sd_disk); + free(trs->trso_buffer, M_TR_RAID1); + trs->trso_buffer = NULL; + trs->trso_flags &= ~TR_RAID1_F_DOING_SOME; + trs->trso_type = TR_RAID1_NONE; + trs->trso_recover_slabs = 0; + trs->trso_failed_sd = NULL; + g_raid_tr_update_state_raid1(vol, NULL); +} + +static void +g_raid_tr_raid1_rebuild_finish(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1_object *trs; + struct g_raid_subdisk *sd; + + trs = (struct g_raid_tr_raid1_object *)tr; + sd = trs->trso_failed_sd; + G_RAID_DEBUG1(0, tr->tro_volume->v_softc, + "Subdisk %s:%d-%s rebuild completed.", + sd->sd_volume->v_name, sd->sd_pos, + sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); + g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_ACTIVE); + sd->sd_rebuild_pos = 0; + g_raid_tr_raid1_rebuild_done(trs); +} + +static void +g_raid_tr_raid1_rebuild_abort(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1_object *trs; + struct g_raid_subdisk *sd; + struct g_raid_volume *vol; + off_t len; + + vol = tr->tro_volume; + trs = (struct g_raid_tr_raid1_object *)tr; + sd = trs->trso_failed_sd; + if (trs->trso_flags & TR_RAID1_F_DOING_SOME) { + G_RAID_DEBUG1(1, vol->v_softc, + "Subdisk %s:%d-%s rebuild is aborting.", + sd->sd_volume->v_name, sd->sd_pos, + sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); + trs->trso_flags |= TR_RAID1_F_ABORT; + } else { + G_RAID_DEBUG1(0, vol->v_softc, + "Subdisk %s:%d-%s rebuild aborted.", + sd->sd_volume->v_name, sd->sd_pos, + sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); + trs->trso_flags &= ~TR_RAID1_F_ABORT; + if (trs->trso_flags & TR_RAID1_F_LOCKED) { + trs->trso_flags &= ~TR_RAID1_F_LOCKED; + len = MIN(g_raid1_rebuild_slab, + sd->sd_size - sd->sd_rebuild_pos); + g_raid_unlock_range(tr->tro_volume, + sd->sd_rebuild_pos, len); + } + g_raid_tr_raid1_rebuild_done(trs); + } +} + +static void +g_raid_tr_raid1_rebuild_start(struct g_raid_tr_object *tr) +{ + struct g_raid_volume *vol; + struct g_raid_tr_raid1_object *trs; + struct g_raid_subdisk *sd, *fsd; + + vol = tr->tro_volume; + trs = (struct g_raid_tr_raid1_object *)tr; + if (trs->trso_failed_sd) { + G_RAID_DEBUG1(1, vol->v_softc, + "Already rebuild in start rebuild. pos %jd\n", + (intmax_t)trs->trso_failed_sd->sd_rebuild_pos); + return; + } + sd = g_raid_get_subdisk(vol, G_RAID_SUBDISK_S_ACTIVE); + if (sd == NULL) { + G_RAID_DEBUG1(1, vol->v_softc, + "No active disk to rebuild. night night."); + return; + } + fsd = g_raid_get_subdisk(vol, G_RAID_SUBDISK_S_RESYNC); + if (fsd == NULL) + fsd = g_raid_get_subdisk(vol, G_RAID_SUBDISK_S_REBUILD); + if (fsd == NULL) { + fsd = g_raid_get_subdisk(vol, G_RAID_SUBDISK_S_STALE); + if (fsd != NULL) { + fsd->sd_rebuild_pos = 0; + g_raid_change_subdisk_state(fsd, + G_RAID_SUBDISK_S_RESYNC); + g_raid_write_metadata(vol->v_softc, vol, fsd, NULL); + } else { + fsd = g_raid_get_subdisk(vol, + G_RAID_SUBDISK_S_UNINITIALIZED); + if (fsd == NULL) + fsd = g_raid_get_subdisk(vol, + G_RAID_SUBDISK_S_NEW); + if (fsd != NULL) { + fsd->sd_rebuild_pos = 0; + g_raid_change_subdisk_state(fsd, + G_RAID_SUBDISK_S_REBUILD); + g_raid_write_metadata(vol->v_softc, + vol, fsd, NULL); + } + } + } + if (fsd == NULL) { + G_RAID_DEBUG1(1, vol->v_softc, + "No failed disk to rebuild. night night."); + return; + } + trs->trso_failed_sd = fsd; + G_RAID_DEBUG1(0, vol->v_softc, + "Subdisk %s:%d-%s rebuild start at %jd.", + fsd->sd_volume->v_name, fsd->sd_pos, + fsd->sd_disk ? g_raid_get_diskname(fsd->sd_disk) : "[none]", + trs->trso_failed_sd->sd_rebuild_pos); + trs->trso_type = TR_RAID1_REBUILD; + trs->trso_buffer = malloc(g_raid1_rebuild_slab, M_TR_RAID1, M_WAITOK); + trs->trso_meta_update = g_raid1_rebuild_meta_update; + g_raid_tr_raid1_rebuild_some(tr); +} + + +static void +g_raid_tr_raid1_maybe_rebuild(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd) +{ + struct g_raid_volume *vol; + struct g_raid_tr_raid1_object *trs; + int na, nr; + + /* + * If we're stopping, don't do anything. If we don't have at least one + * good disk and one bad disk, we don't do anything. And if there's a + * 'good disk' stored in the trs, then we're in progress and we punt. + * If we make it past all these checks, we need to rebuild. + */ + vol = tr->tro_volume; + trs = (struct g_raid_tr_raid1_object *)tr; + if (trs->trso_stopping) + return; + na = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE); + nr = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_REBUILD) + + g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_RESYNC); + switch(trs->trso_type) { + case TR_RAID1_NONE: + if (na == 0) + return; + if (nr == 0) { + nr = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_NEW) + + g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_STALE) + + g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_UNINITIALIZED); + if (nr == 0) + return; + } + g_raid_tr_raid1_rebuild_start(tr); + break; + case TR_RAID1_REBUILD: + if (na == 0 || nr == 0 || trs->trso_failed_sd == sd) + g_raid_tr_raid1_rebuild_abort(tr); + break; + case TR_RAID1_RESYNC: + break; + } +} + +static int +g_raid_tr_event_raid1(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd, u_int event) +{ + + g_raid_tr_update_state_raid1(tr->tro_volume, sd); + return (0); +} + +static int +g_raid_tr_start_raid1(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1_object *trs; + struct g_raid_volume *vol; + + trs = (struct g_raid_tr_raid1_object *)tr; + vol = tr->tro_volume; + trs->trso_starting = 0; + g_raid_tr_update_state_raid1(vol, NULL); + return (0); +} + +static int +g_raid_tr_stop_raid1(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1_object *trs; + struct g_raid_volume *vol; + + trs = (struct g_raid_tr_raid1_object *)tr; + vol = tr->tro_volume; + trs->trso_starting = 0; + trs->trso_stopping = 1; + g_raid_tr_update_state_raid1(vol, NULL); + return (0); +} + +/* + * Select the disk to read from. Take into account: subdisk state, running + * error recovery, average disk load, head position and possible cache hits. + */ +#define ABS(x) (((x) >= 0) ? (x) : (-(x))) +static struct g_raid_subdisk * +g_raid_tr_raid1_select_read_disk(struct g_raid_volume *vol, struct bio *bp, + u_int mask) +{ + struct g_raid_subdisk *sd, *best; + int i, prio, bestprio; + + best = NULL; + bestprio = INT_MAX; + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE && + ((sd->sd_state != G_RAID_SUBDISK_S_REBUILD && + sd->sd_state != G_RAID_SUBDISK_S_RESYNC) || + bp->bio_offset + bp->bio_length > sd->sd_rebuild_pos)) + continue; + if ((mask & (1 << i)) != 0) + continue; + prio = G_RAID_SUBDISK_LOAD(sd); + prio += min(sd->sd_recovery, 255) << 22; + prio += (G_RAID_SUBDISK_S_ACTIVE - sd->sd_state) << 16; + /* If disk head is precisely in position - highly prefer it. */ + if (G_RAID_SUBDISK_POS(sd) == bp->bio_offset) + prio -= 2 * G_RAID_SUBDISK_LOAD_SCALE; + else + /* If disk head is close to position - prefer it. */ + if (ABS(G_RAID_SUBDISK_POS(sd) - bp->bio_offset) < + G_RAID_SUBDISK_TRACK_SIZE) + prio -= 1 * G_RAID_SUBDISK_LOAD_SCALE; + if (prio < bestprio) { + best = sd; + bestprio = prio; + } + } + return (best); +} + +static void +g_raid_tr_iostart_raid1_read(struct g_raid_tr_object *tr, struct bio *bp) +{ + struct g_raid_subdisk *sd; + struct bio *cbp; + + sd = g_raid_tr_raid1_select_read_disk(tr->tro_volume, bp, 0); + KASSERT(sd != NULL, ("No active disks in volume %s.", + tr->tro_volume->v_name)); + + cbp = g_clone_bio(bp); + if (cbp == NULL) { + g_raid_iodone(bp, ENOMEM); + return; + } + + g_raid_subdisk_iostart(sd, cbp); +} + +static void +g_raid_tr_iostart_raid1_write(struct g_raid_tr_object *tr, struct bio *bp) +{ + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct bio_queue_head queue; + struct bio *cbp; + int i; + + vol = tr->tro_volume; + sc = vol->v_softc; + + /* + * Allocate all bios before sending any request, so we can return + * ENOMEM in nice and clean way. + */ + bioq_init(&queue); + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + switch (sd->sd_state) { + case G_RAID_SUBDISK_S_ACTIVE: + break; + case G_RAID_SUBDISK_S_REBUILD: + /* + * When rebuilding, only part of this subdisk is + * writable, the rest will be written as part of the + * that process. + */ + if (bp->bio_offset >= sd->sd_rebuild_pos) + continue; + break; + case G_RAID_SUBDISK_S_STALE: + case G_RAID_SUBDISK_S_RESYNC: + /* + * Resyncing still writes on the theory that the + * resync'd disk is very close and writing it will + * keep it that way better if we keep up while + * resyncing. + */ + break; + default: + continue; + } + cbp = g_clone_bio(bp); + if (cbp == NULL) + goto failure; + cbp->bio_caller1 = sd; + bioq_insert_tail(&queue, cbp); + } + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + sd = cbp->bio_caller1; + cbp->bio_caller1 = NULL; + g_raid_subdisk_iostart(sd, cbp); + } + return; +failure: + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + g_destroy_bio(cbp); + } + if (bp->bio_error == 0) + bp->bio_error = ENOMEM; + g_raid_iodone(bp, bp->bio_error); +} + +static void +g_raid_tr_iostart_raid1(struct g_raid_tr_object *tr, struct bio *bp) +{ + struct g_raid_volume *vol; + struct g_raid_tr_raid1_object *trs; + + vol = tr->tro_volume; + trs = (struct g_raid_tr_raid1_object *)tr; + if (vol->v_state != G_RAID_VOLUME_S_OPTIMAL && + vol->v_state != G_RAID_VOLUME_S_SUBOPTIMAL && + vol->v_state != G_RAID_VOLUME_S_DEGRADED) { + g_raid_iodone(bp, EIO); + return; + } + /* + * If we're rebuilding, squeeze in rebuild activity every so often, + * even when the disk is busy. Be sure to only count real I/O + * to the disk. All 'SPECIAL' I/O is traffic generated to the disk + * by this module. + */ + if (trs->trso_failed_sd != NULL && + !(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL)) { + /* Make this new or running now round short. */ + trs->trso_recover_slabs = 0; + if (--trs->trso_fair_io <= 0) { + trs->trso_fair_io = g_raid1_rebuild_fair_io; + g_raid_tr_raid1_rebuild_some(tr); + } + } + switch (bp->bio_cmd) { + case BIO_READ: + g_raid_tr_iostart_raid1_read(tr, bp); + break; + case BIO_WRITE: + g_raid_tr_iostart_raid1_write(tr, bp); + break; + case BIO_DELETE: + g_raid_iodone(bp, EIO); + break; + case BIO_FLUSH: + g_raid_tr_flush_common(tr, bp); + break; + default: + KASSERT(1 == 0, ("Invalid command here: %u (volume=%s)", + bp->bio_cmd, vol->v_name)); + break; + } +} + +static void +g_raid_tr_iodone_raid1(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd, struct bio *bp) +{ + struct bio *cbp; + struct g_raid_subdisk *nsd; + struct g_raid_volume *vol; + struct bio *pbp; + struct g_raid_tr_raid1_object *trs; + uintptr_t *mask; + int error, do_write; + + trs = (struct g_raid_tr_raid1_object *)tr; + vol = tr->tro_volume; + if (bp->bio_cflags & G_RAID_BIO_FLAG_SYNC) { + /* + * This operation is part of a rebuild or resync operation. + * See what work just got done, then schedule the next bit of + * work, if any. Rebuild/resync is done a little bit at a + * time. Either when a timeout happens, or after we get a + * bunch of I/Os to the disk (to make sure an active system + * will complete in a sane amount of time). + * + * We are setup to do differing amounts of work for each of + * these cases. so long as the slabs is smallish (less than + * 50 or so, I'd guess, but that's just a WAG), we shouldn't + * have any bio starvation issues. For active disks, we do + * 5MB of data, for inactive ones, we do 50MB. + */ + if (trs->trso_type == TR_RAID1_REBUILD) { + if (bp->bio_cmd == BIO_READ) { + + /* Immediately abort rebuild, if requested. */ + if (trs->trso_flags & TR_RAID1_F_ABORT) { + trs->trso_flags &= ~TR_RAID1_F_DOING_SOME; + g_raid_tr_raid1_rebuild_abort(tr); + return; + } + + /* On read error, skip and cross fingers. */ + if (bp->bio_error != 0) { + G_RAID_LOGREQ(0, bp, + "Read error during rebuild (%d), " + "possible data loss!", + bp->bio_error); + goto rebuild_round_done; + } + + /* + * The read operation finished, queue the + * write and get out. + */ + G_RAID_LOGREQ(4, bp, "rebuild read done. %d", + bp->bio_error); + bp->bio_cmd = BIO_WRITE; + bp->bio_cflags = G_RAID_BIO_FLAG_SYNC; + bp->bio_offset = bp->bio_offset; + bp->bio_length = bp->bio_length; + G_RAID_LOGREQ(4, bp, "Queueing rebuild write."); + g_raid_subdisk_iostart(trs->trso_failed_sd, bp); + } else { + /* + * The write operation just finished. Do + * another. We keep cloning the master bio + * since it has the right buffers allocated to + * it. + */ + G_RAID_LOGREQ(4, bp, + "rebuild write done. Error %d", + bp->bio_error); + nsd = trs->trso_failed_sd; + if (bp->bio_error != 0 || + trs->trso_flags & TR_RAID1_F_ABORT) { + if ((trs->trso_flags & + TR_RAID1_F_ABORT) == 0) { + g_raid_tr_raid1_fail_disk(sd->sd_softc, + nsd, nsd->sd_disk); + } + trs->trso_flags &= ~TR_RAID1_F_DOING_SOME; + g_raid_tr_raid1_rebuild_abort(tr); + return; + } +rebuild_round_done: + nsd = trs->trso_failed_sd; + trs->trso_flags &= ~TR_RAID1_F_LOCKED; + g_raid_unlock_range(sd->sd_volume, + bp->bio_offset, bp->bio_length); + nsd->sd_rebuild_pos += bp->bio_length; + if (nsd->sd_rebuild_pos >= nsd->sd_size) { + g_raid_tr_raid1_rebuild_finish(tr); + return; + } + + /* Abort rebuild if we are stopping */ + if (trs->trso_stopping) { + trs->trso_flags &= ~TR_RAID1_F_DOING_SOME; + g_raid_tr_raid1_rebuild_abort(tr); + return; + } + + if (--trs->trso_meta_update <= 0) { + g_raid_write_metadata(vol->v_softc, + vol, nsd, nsd->sd_disk); + trs->trso_meta_update = + g_raid1_rebuild_meta_update; + } + trs->trso_flags &= ~TR_RAID1_F_DOING_SOME; + if (--trs->trso_recover_slabs <= 0) + return; + g_raid_tr_raid1_rebuild_some(tr); + } + } else if (trs->trso_type == TR_RAID1_RESYNC) { + /* + * read good sd, read bad sd in parallel. when both + * done, compare the buffers. write good to the bad + * if different. do the next bit of work. + */ + panic("Somehow, we think we're doing a resync"); + } + return; + } + pbp = bp->bio_parent; + pbp->bio_inbed++; + if (bp->bio_cmd == BIO_READ && bp->bio_error != 0) { + /* + * Read failed on first drive. Retry the read error on + * another disk drive, if available, before erroring out the + * read. + */ + sd->sd_disk->d_read_errs++; + G_RAID_LOGREQ(0, bp, + "Read error (%d), %d read errors total", + bp->bio_error, sd->sd_disk->d_read_errs); + + /* + * If there are too many read errors, we move to degraded. + * XXX Do we want to FAIL the drive (eg, make the user redo + * everything to get it back in sync), or just degrade the + * drive, which kicks off a resync? + */ + do_write = 1; + if (sd->sd_disk->d_read_errs > g_raid_read_err_thresh) { + g_raid_tr_raid1_fail_disk(sd->sd_softc, sd, sd->sd_disk); + if (pbp->bio_children == 1) + do_write = 0; + } + + /* + * Find the other disk, and try to do the I/O to it. + */ + mask = (uintptr_t *)(&pbp->bio_driver2); + if (pbp->bio_children == 1) { + /* Save original subdisk. */ + pbp->bio_driver1 = do_write ? sd : NULL; + *mask = 0; + } + *mask |= 1 << sd->sd_pos; + nsd = g_raid_tr_raid1_select_read_disk(vol, pbp, *mask); + if (nsd != NULL && (cbp = g_clone_bio(pbp)) != NULL) { + g_destroy_bio(bp); + G_RAID_LOGREQ(2, cbp, "Retrying read from %d", + nsd->sd_pos); + if (pbp->bio_children == 2 && do_write) { + sd->sd_recovery++; + cbp->bio_caller1 = nsd; + pbp->bio_pflags = G_RAID_BIO_FLAG_LOCKED; + /* Lock callback starts I/O */ + g_raid_lock_range(sd->sd_volume, + cbp->bio_offset, cbp->bio_length, pbp, cbp); + } else { + g_raid_subdisk_iostart(nsd, cbp); + } + return; + } + /* + * We can't retry. Return the original error by falling + * through. This will happen when there's only one good disk. + * We don't need to fail the raid, since its actual state is + * based on the state of the subdisks. + */ + G_RAID_LOGREQ(2, bp, "Couldn't retry read, failing it"); + } + if (bp->bio_cmd == BIO_READ && + bp->bio_error == 0 && + pbp->bio_children > 1 && + pbp->bio_driver1 != NULL) { + /* + * If it was a read, and bio_children is >1, then we just + * recovered the data from the second drive. We should try to + * write that data to the first drive if sector remapping is + * enabled. A write should put the data in a new place on the + * disk, remapping the bad sector. Do we need to do that by + * queueing a request to the main worker thread? It doesn't + * affect the return code of this current read, and can be + * done at our liesure. However, to make the code simpler, it + * is done syncrhonously. + */ + G_RAID_LOGREQ(3, bp, "Recovered data from other drive"); + cbp = g_clone_bio(pbp); + if (cbp != NULL) { + g_destroy_bio(bp); + cbp->bio_cmd = BIO_WRITE; + cbp->bio_cflags = G_RAID_BIO_FLAG_REMAP; + G_RAID_LOGREQ(2, cbp, + "Attempting bad sector remap on failing drive."); + g_raid_subdisk_iostart(pbp->bio_driver1, cbp); + return; + } + } + if (pbp->bio_pflags & G_RAID_BIO_FLAG_LOCKED) { + /* + * We're done with a recovery, mark the range as unlocked. + * For any write errors, we agressively fail the disk since + * there was both a READ and a WRITE error at this location. + * Both types of errors generally indicates the drive is on + * the verge of total failure anyway. Better to stop trusting + * it now. However, we need to reset error to 0 in that case + * because we're not failing the original I/O which succeeded. + */ + if (bp->bio_cmd == BIO_WRITE && bp->bio_error) { + G_RAID_LOGREQ(0, bp, "Remap write failed: " + "failing subdisk."); + g_raid_tr_raid1_fail_disk(sd->sd_softc, sd, sd->sd_disk); + bp->bio_error = 0; + } + if (pbp->bio_driver1 != NULL) { + ((struct g_raid_subdisk *)pbp->bio_driver1) + ->sd_recovery--; + } + G_RAID_LOGREQ(2, bp, "REMAP done %d.", bp->bio_error); + g_raid_unlock_range(sd->sd_volume, bp->bio_offset, + bp->bio_length); + } + error = bp->bio_error; + g_destroy_bio(bp); + if (pbp->bio_children == pbp->bio_inbed) { + pbp->bio_completed = pbp->bio_length; + g_raid_iodone(pbp, error); + } +} + +static int +g_raid_tr_kerneldump_raid1(struct g_raid_tr_object *tr, + void *virtual, vm_offset_t physical, off_t offset, size_t length) +{ + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + int error, i, ok; + + vol = tr->tro_volume; + error = 0; + ok = 0; + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + switch (sd->sd_state) { + case G_RAID_SUBDISK_S_ACTIVE: + break; + case G_RAID_SUBDISK_S_REBUILD: + /* + * When rebuilding, only part of this subdisk is + * writable, the rest will be written as part of the + * that process. + */ + if (offset >= sd->sd_rebuild_pos) + continue; + break; + case G_RAID_SUBDISK_S_STALE: + case G_RAID_SUBDISK_S_RESYNC: + /* + * Resyncing still writes on the theory that the + * resync'd disk is very close and writing it will + * keep it that way better if we keep up while + * resyncing. + */ + break; + default: + continue; + } + error = g_raid_subdisk_kerneldump(sd, + virtual, physical, offset, length); + if (error == 0) + ok++; + } + return (ok > 0 ? 0 : error); +} + +static int +g_raid_tr_locked_raid1(struct g_raid_tr_object *tr, void *argp) +{ + struct bio *bp; + struct g_raid_subdisk *sd; + + bp = (struct bio *)argp; + sd = (struct g_raid_subdisk *)bp->bio_caller1; + g_raid_subdisk_iostart(sd, bp); + + return (0); +} + +static int +g_raid_tr_idle_raid1(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1_object *trs; + + trs = (struct g_raid_tr_raid1_object *)tr; + trs->trso_fair_io = g_raid1_rebuild_fair_io; + trs->trso_recover_slabs = g_raid1_rebuild_cluster_idle; + if (trs->trso_type == TR_RAID1_REBUILD) + g_raid_tr_raid1_rebuild_some(tr); + return (0); +} + +static int +g_raid_tr_free_raid1(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1_object *trs; + + trs = (struct g_raid_tr_raid1_object *)tr; + + if (trs->trso_buffer != NULL) { + free(trs->trso_buffer, M_TR_RAID1); + trs->trso_buffer = NULL; + } + return (0); +} + +G_RAID_TR_DECLARE(g_raid_tr_raid1); diff --git a/sys/geom/raid/tr_raid1e.c b/sys/geom/raid/tr_raid1e.c new file mode 100644 index 000000000000..9ebe21864fe1 --- /dev/null +++ b/sys/geom/raid/tr_raid1e.c @@ -0,0 +1,1227 @@ +/*- + * Copyright (c) 2010 Alexander Motin + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "geom/raid/g_raid.h" +#include "g_raid_tr_if.h" + +#define N 2 + +SYSCTL_DECL(_kern_geom_raid); +SYSCTL_NODE(_kern_geom_raid, OID_AUTO, raid1e, CTLFLAG_RW, 0, + "RAID1E parameters"); + +#define RAID1E_REBUILD_SLAB (1 << 20) /* One transation in a rebuild */ +static int g_raid1e_rebuild_slab = RAID1E_REBUILD_SLAB; +TUNABLE_INT("kern.geom.raid.raid1e.rebuild_slab_size", + &g_raid1e_rebuild_slab); +SYSCTL_UINT(_kern_geom_raid_raid1e, OID_AUTO, rebuild_slab_size, CTLFLAG_RW, + &g_raid1e_rebuild_slab, 0, + "Amount of the disk to rebuild each read/write cycle of the rebuild."); + +#define RAID1E_REBUILD_FAIR_IO 20 /* use 1/x of the available I/O */ +static int g_raid1e_rebuild_fair_io = RAID1E_REBUILD_FAIR_IO; +TUNABLE_INT("kern.geom.raid.raid1e.rebuild_fair_io", + &g_raid1e_rebuild_fair_io); +SYSCTL_UINT(_kern_geom_raid_raid1e, OID_AUTO, rebuild_fair_io, CTLFLAG_RW, + &g_raid1e_rebuild_fair_io, 0, + "Fraction of the I/O bandwidth to use when disk busy for rebuild."); + +#define RAID1E_REBUILD_CLUSTER_IDLE 100 +static int g_raid1e_rebuild_cluster_idle = RAID1E_REBUILD_CLUSTER_IDLE; +TUNABLE_INT("kern.geom.raid.raid1e.rebuild_cluster_idle", + &g_raid1e_rebuild_cluster_idle); +SYSCTL_UINT(_kern_geom_raid_raid1e, OID_AUTO, rebuild_cluster_idle, CTLFLAG_RW, + &g_raid1e_rebuild_cluster_idle, 0, + "Number of slabs to do each time we trigger a rebuild cycle"); + +#define RAID1E_REBUILD_META_UPDATE 1024 /* update meta data every 1GB or so */ +static int g_raid1e_rebuild_meta_update = RAID1E_REBUILD_META_UPDATE; +TUNABLE_INT("kern.geom.raid.raid1e.rebuild_meta_update", + &g_raid1e_rebuild_meta_update); +SYSCTL_UINT(_kern_geom_raid_raid1e, OID_AUTO, rebuild_meta_update, CTLFLAG_RW, + &g_raid1e_rebuild_meta_update, 0, + "When to update the meta data."); + +static MALLOC_DEFINE(M_TR_RAID1E, "tr_raid1e_data", "GEOM_RAID RAID1E data"); + +#define TR_RAID1E_NONE 0 +#define TR_RAID1E_REBUILD 1 +#define TR_RAID1E_RESYNC 2 + +#define TR_RAID1E_F_DOING_SOME 0x1 +#define TR_RAID1E_F_LOCKED 0x2 +#define TR_RAID1E_F_ABORT 0x4 + +struct g_raid_tr_raid1e_object { + struct g_raid_tr_object trso_base; + int trso_starting; + int trso_stopping; + int trso_type; + int trso_recover_slabs; /* slabs before rest */ + int trso_fair_io; + int trso_meta_update; + int trso_flags; + struct g_raid_subdisk *trso_failed_sd; /* like per volume */ + void *trso_buffer; /* Buffer space */ + off_t trso_lock_pos; /* Locked range start. */ + off_t trso_lock_len; /* Locked range length. */ + struct bio trso_bio; +}; + +static g_raid_tr_taste_t g_raid_tr_taste_raid1e; +static g_raid_tr_event_t g_raid_tr_event_raid1e; +static g_raid_tr_start_t g_raid_tr_start_raid1e; +static g_raid_tr_stop_t g_raid_tr_stop_raid1e; +static g_raid_tr_iostart_t g_raid_tr_iostart_raid1e; +static g_raid_tr_iodone_t g_raid_tr_iodone_raid1e; +static g_raid_tr_kerneldump_t g_raid_tr_kerneldump_raid1e; +static g_raid_tr_locked_t g_raid_tr_locked_raid1e; +static g_raid_tr_idle_t g_raid_tr_idle_raid1e; +static g_raid_tr_free_t g_raid_tr_free_raid1e; + +static kobj_method_t g_raid_tr_raid1e_methods[] = { + KOBJMETHOD(g_raid_tr_taste, g_raid_tr_taste_raid1e), + KOBJMETHOD(g_raid_tr_event, g_raid_tr_event_raid1e), + KOBJMETHOD(g_raid_tr_start, g_raid_tr_start_raid1e), + KOBJMETHOD(g_raid_tr_stop, g_raid_tr_stop_raid1e), + KOBJMETHOD(g_raid_tr_iostart, g_raid_tr_iostart_raid1e), + KOBJMETHOD(g_raid_tr_iodone, g_raid_tr_iodone_raid1e), + KOBJMETHOD(g_raid_tr_kerneldump, g_raid_tr_kerneldump_raid1e), + KOBJMETHOD(g_raid_tr_locked, g_raid_tr_locked_raid1e), + KOBJMETHOD(g_raid_tr_idle, g_raid_tr_idle_raid1e), + KOBJMETHOD(g_raid_tr_free, g_raid_tr_free_raid1e), + { 0, 0 } +}; + +static struct g_raid_tr_class g_raid_tr_raid1e_class = { + "RAID1E", + g_raid_tr_raid1e_methods, + sizeof(struct g_raid_tr_raid1e_object), + .trc_priority = 200 +}; + +static void g_raid_tr_raid1e_rebuild_abort(struct g_raid_tr_object *tr); +static void g_raid_tr_raid1e_maybe_rebuild(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd); +static int g_raid_tr_raid1e_select_read_disk(struct g_raid_volume *vol, + int no, off_t off, off_t len, u_int mask); + +static inline void +V2P(struct g_raid_volume *vol, off_t virt, + int *disk, off_t *offset, off_t *start) +{ + off_t nstrip; + u_int strip_size; + + strip_size = vol->v_strip_size; + /* Strip number. */ + nstrip = virt / strip_size; + /* Start position in strip. */ + *start = virt % strip_size; + /* Disk number. */ + *disk = (nstrip * N) % vol->v_disks_count; + /* Strip start position in disk. */ + *offset = ((nstrip * N) / vol->v_disks_count) * strip_size; +} + +static inline void +P2V(struct g_raid_volume *vol, int disk, off_t offset, + off_t *virt, int *copy) +{ + off_t nstrip, start; + u_int strip_size; + + strip_size = vol->v_strip_size; + /* Start position in strip. */ + start = offset % strip_size; + /* Physical strip number. */ + nstrip = (offset / strip_size) * vol->v_disks_count + disk; + /* Number of physical strip (copy) inside virtual strip. */ + *copy = nstrip % N; + /* Offset in virtual space. */ + *virt = (nstrip / N) * strip_size + start; +} + +static int +g_raid_tr_taste_raid1e(struct g_raid_tr_object *tr, struct g_raid_volume *vol) +{ + struct g_raid_tr_raid1e_object *trs; + + trs = (struct g_raid_tr_raid1e_object *)tr; + if (tr->tro_volume->v_raid_level != G_RAID_VOLUME_RL_RAID1E || + tr->tro_volume->v_raid_level_qualifier != G_RAID_VOLUME_RLQ_NONE) + return (G_RAID_TR_TASTE_FAIL); + trs->trso_starting = 1; + return (G_RAID_TR_TASTE_SUCCEED); +} + +static int +g_raid_tr_update_state_raid1e_even(struct g_raid_volume *vol) +{ + struct g_raid_tr_raid1e_object *trs; + struct g_raid_softc *sc; + struct g_raid_subdisk *sd, *bestsd, *worstsd; + int i, j, state, sstate; + + sc = vol->v_softc; + trs = (struct g_raid_tr_raid1e_object *)vol->v_tr; + state = G_RAID_VOLUME_S_OPTIMAL; + for (i = 0; i < vol->v_disks_count / N; i++) { + bestsd = &vol->v_subdisks[i * N]; + for (j = 1; j < N; j++) { + sd = &vol->v_subdisks[i * N + j]; + if (sd->sd_state > bestsd->sd_state) + bestsd = sd; + else if (sd->sd_state == bestsd->sd_state && + (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || + sd->sd_state == G_RAID_SUBDISK_S_RESYNC) && + sd->sd_rebuild_pos > bestsd->sd_rebuild_pos) + bestsd = sd; + } + if (bestsd->sd_state >= G_RAID_SUBDISK_S_UNINITIALIZED && + bestsd->sd_state != G_RAID_SUBDISK_S_ACTIVE) { + /* We found reasonable candidate. */ + G_RAID_DEBUG1(1, sc, + "Promote subdisk %s:%d from %s to ACTIVE.", + vol->v_name, bestsd->sd_pos, + g_raid_subdisk_state2str(bestsd->sd_state)); + g_raid_change_subdisk_state(bestsd, + G_RAID_SUBDISK_S_ACTIVE); + g_raid_write_metadata(sc, + vol, bestsd, bestsd->sd_disk); + } + worstsd = &vol->v_subdisks[i * N]; + for (j = 1; j < N; j++) { + sd = &vol->v_subdisks[i * N + j]; + if (sd->sd_state < worstsd->sd_state) + worstsd = sd; + } + if (worstsd->sd_state == G_RAID_SUBDISK_S_ACTIVE) + sstate = G_RAID_VOLUME_S_OPTIMAL; + else if (worstsd->sd_state >= G_RAID_SUBDISK_S_STALE) + sstate = G_RAID_VOLUME_S_SUBOPTIMAL; + else if (bestsd->sd_state == G_RAID_SUBDISK_S_ACTIVE) + sstate = G_RAID_VOLUME_S_DEGRADED; + else + sstate = G_RAID_VOLUME_S_BROKEN; + if (sstate < state) + state = sstate; + } + return (state); +} + +static int +g_raid_tr_update_state_raid1e_odd(struct g_raid_volume *vol) +{ + struct g_raid_tr_raid1e_object *trs; + struct g_raid_softc *sc; + struct g_raid_subdisk *sd, *bestsd, *worstsd; + int i, j, state, sstate; + + sc = vol->v_softc; + trs = (struct g_raid_tr_raid1e_object *)vol->v_tr; + if (g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_ACTIVE) == + vol->v_disks_count) + return (G_RAID_VOLUME_S_OPTIMAL); + for (i = 0; i < vol->v_disks_count; i++) { + sd = &vol->v_subdisks[i]; + if (sd->sd_state == G_RAID_SUBDISK_S_UNINITIALIZED) { + /* We found reasonable candidate. */ + G_RAID_DEBUG1(1, sc, + "Promote subdisk %s:%d from %s to STALE.", + vol->v_name, sd->sd_pos, + g_raid_subdisk_state2str(sd->sd_state)); + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_STALE); + g_raid_write_metadata(sc, vol, sd, sd->sd_disk); + } + } + state = G_RAID_VOLUME_S_OPTIMAL; + for (i = 0; i < vol->v_disks_count; i++) { + bestsd = &vol->v_subdisks[i]; + worstsd = &vol->v_subdisks[i]; + for (j = 1; j < N; j++) { + sd = &vol->v_subdisks[(i + j) % vol->v_disks_count]; + if (sd->sd_state > bestsd->sd_state) + bestsd = sd; + else if (sd->sd_state == bestsd->sd_state && + (sd->sd_state == G_RAID_SUBDISK_S_REBUILD || + sd->sd_state == G_RAID_SUBDISK_S_RESYNC) && + sd->sd_rebuild_pos > bestsd->sd_rebuild_pos) + bestsd = sd; + if (sd->sd_state < worstsd->sd_state) + worstsd = sd; + } + if (worstsd->sd_state == G_RAID_SUBDISK_S_ACTIVE) + sstate = G_RAID_VOLUME_S_OPTIMAL; + else if (worstsd->sd_state >= G_RAID_SUBDISK_S_STALE) + sstate = G_RAID_VOLUME_S_SUBOPTIMAL; + else if (bestsd->sd_state >= G_RAID_SUBDISK_S_STALE) + sstate = G_RAID_VOLUME_S_DEGRADED; + else + sstate = G_RAID_VOLUME_S_BROKEN; + if (sstate < state) + state = sstate; + } + return (state); +} + +static int +g_raid_tr_update_state_raid1e(struct g_raid_volume *vol, + struct g_raid_subdisk *sd) +{ + struct g_raid_tr_raid1e_object *trs; + struct g_raid_softc *sc; + u_int s; + + sc = vol->v_softc; + trs = (struct g_raid_tr_raid1e_object *)vol->v_tr; + if (trs->trso_stopping && + (trs->trso_flags & TR_RAID1E_F_DOING_SOME) == 0) + s = G_RAID_VOLUME_S_STOPPED; + else if (trs->trso_starting) + s = G_RAID_VOLUME_S_STARTING; + else { + if ((vol->v_disks_count % N) == 0) + s = g_raid_tr_update_state_raid1e_even(vol); + else + s = g_raid_tr_update_state_raid1e_odd(vol); + } + if (s != vol->v_state) { + g_raid_event_send(vol, G_RAID_VOLUME_S_ALIVE(s) ? + G_RAID_VOLUME_E_UP : G_RAID_VOLUME_E_DOWN, + G_RAID_EVENT_VOLUME); + g_raid_change_volume_state(vol, s); + if (!trs->trso_starting && !trs->trso_stopping) + g_raid_write_metadata(sc, vol, NULL, NULL); + } + if (!trs->trso_starting && !trs->trso_stopping) + g_raid_tr_raid1e_maybe_rebuild(vol->v_tr, sd); + return (0); +} + +static void +g_raid_tr_raid1e_fail_disk(struct g_raid_softc *sc, struct g_raid_subdisk *sd, + struct g_raid_disk *disk) +{ + /* + * We don't fail the last disk in the pack, since it still has decent + * data on it and that's better than failing the disk if it is the root + * file system. + * + * XXX should this be controlled via a tunable? It makes sense for + * the volume that has / on it. I can't think of a case where we'd + * want the volume to go away on this kind of event. + */ + if (g_raid_nsubdisks(sd->sd_volume, G_RAID_SUBDISK_S_ACTIVE) == 1 && + g_raid_get_subdisk(sd->sd_volume, G_RAID_SUBDISK_S_ACTIVE) == sd) + return; + g_raid_fail_disk(sc, sd, disk); +} + +static void +g_raid_tr_raid1e_rebuild_done(struct g_raid_tr_raid1e_object *trs) +{ + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + + vol = trs->trso_base.tro_volume; + sd = trs->trso_failed_sd; + g_raid_write_metadata(vol->v_softc, vol, sd, sd->sd_disk); + free(trs->trso_buffer, M_TR_RAID1E); + trs->trso_buffer = NULL; + trs->trso_flags &= ~TR_RAID1E_F_DOING_SOME; + trs->trso_type = TR_RAID1E_NONE; + trs->trso_recover_slabs = 0; + trs->trso_failed_sd = NULL; + g_raid_tr_update_state_raid1e(vol, NULL); +} + +static void +g_raid_tr_raid1e_rebuild_finish(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1e_object *trs; + struct g_raid_subdisk *sd; + + trs = (struct g_raid_tr_raid1e_object *)tr; + sd = trs->trso_failed_sd; + G_RAID_DEBUG1(0, tr->tro_volume->v_softc, + "Subdisk %s:%d-%s rebuild completed.", + sd->sd_volume->v_name, sd->sd_pos, + sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); + g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_ACTIVE); + sd->sd_rebuild_pos = 0; + g_raid_tr_raid1e_rebuild_done(trs); +} + +static void +g_raid_tr_raid1e_rebuild_abort(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1e_object *trs; + struct g_raid_subdisk *sd; + struct g_raid_volume *vol; + + vol = tr->tro_volume; + trs = (struct g_raid_tr_raid1e_object *)tr; + sd = trs->trso_failed_sd; + if (trs->trso_flags & TR_RAID1E_F_DOING_SOME) { + G_RAID_DEBUG1(1, vol->v_softc, + "Subdisk %s:%d-%s rebuild is aborting.", + sd->sd_volume->v_name, sd->sd_pos, + sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); + trs->trso_flags |= TR_RAID1E_F_ABORT; + } else { + G_RAID_DEBUG1(0, vol->v_softc, + "Subdisk %s:%d-%s rebuild aborted.", + sd->sd_volume->v_name, sd->sd_pos, + sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]"); + trs->trso_flags &= ~TR_RAID1E_F_ABORT; + if (trs->trso_flags & TR_RAID1E_F_LOCKED) { + trs->trso_flags &= ~TR_RAID1E_F_LOCKED; + g_raid_unlock_range(tr->tro_volume, + trs->trso_lock_pos, trs->trso_lock_len); + } + g_raid_tr_raid1e_rebuild_done(trs); + } +} + +static void +g_raid_tr_raid1e_rebuild_some(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1e_object *trs; + struct g_raid_softc *sc; + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct bio *bp; + off_t len, virtual, vend, offset, start; + int disk, copy, best; + + trs = (struct g_raid_tr_raid1e_object *)tr; + if (trs->trso_flags & TR_RAID1E_F_DOING_SOME) + return; + vol = tr->tro_volume; + sc = vol->v_softc; + sd = trs->trso_failed_sd; + + while (1) { + if (sd->sd_rebuild_pos >= sd->sd_size) { + g_raid_tr_raid1e_rebuild_finish(tr); + return; + } + /* Get virtual offset from physical rebuild position. */ + P2V(vol, sd->sd_pos, sd->sd_rebuild_pos, &virtual, ©); + /* Get physical offset back to get first stripe position. */ + V2P(vol, virtual, &disk, &offset, &start); + /* Calculate contignous data length. */ + len = MIN(g_raid1e_rebuild_slab, + sd->sd_size - sd->sd_rebuild_pos); + if ((vol->v_disks_count % N) != 0) + len = MIN(len, vol->v_strip_size - start); + /* Find disk with most accurate data. */ + best = g_raid_tr_raid1e_select_read_disk(vol, disk, + offset + start, len, 0); + if (best < 0) { + /* There is no any valid disk. */ + g_raid_tr_raid1e_rebuild_abort(tr); + return; + } else if (best != copy) { + /* Some other disk has better data. */ + break; + } + /* We have the most accurate data. Skip the range. */ + G_RAID_DEBUG1(3, sc, "Skipping rebuild for range %ju - %ju", + sd->sd_rebuild_pos, sd->sd_rebuild_pos + len); + sd->sd_rebuild_pos += len; + } + + bp = &trs->trso_bio; + memset(bp, 0, sizeof(*bp)); + bp->bio_offset = offset + start + + ((disk + best >= vol->v_disks_count) ? vol->v_strip_size : 0); + bp->bio_length = len; + bp->bio_data = trs->trso_buffer; + bp->bio_cmd = BIO_READ; + bp->bio_cflags = G_RAID_BIO_FLAG_SYNC; + bp->bio_caller1 = &vol->v_subdisks[(disk + best) % vol->v_disks_count]; + G_RAID_LOGREQ(3, bp, "Queueing rebuild read"); + /* + * If we are crossing stripe boundary, correct affected virtual + * range we should lock. + */ + if (start + len > vol->v_strip_size) { + P2V(vol, sd->sd_pos, sd->sd_rebuild_pos + len, &vend, ©); + len = vend - virtual; + } + trs->trso_flags |= TR_RAID1E_F_DOING_SOME; + trs->trso_flags |= TR_RAID1E_F_LOCKED; + trs->trso_lock_pos = virtual; + trs->trso_lock_len = len; + /* Lock callback starts I/O */ + g_raid_lock_range(sd->sd_volume, virtual, len, NULL, bp); +} + +static void +g_raid_tr_raid1e_rebuild_start(struct g_raid_tr_object *tr) +{ + struct g_raid_volume *vol; + struct g_raid_tr_raid1e_object *trs; + struct g_raid_subdisk *sd; + + vol = tr->tro_volume; + trs = (struct g_raid_tr_raid1e_object *)tr; + if (trs->trso_failed_sd) { + G_RAID_DEBUG1(1, vol->v_softc, + "Already rebuild in start rebuild. pos %jd\n", + (intmax_t)trs->trso_failed_sd->sd_rebuild_pos); + return; + } + sd = g_raid_get_subdisk(vol, G_RAID_SUBDISK_S_RESYNC); + if (sd == NULL) + sd = g_raid_get_subdisk(vol, G_RAID_SUBDISK_S_REBUILD); + if (sd == NULL) { + sd = g_raid_get_subdisk(vol, G_RAID_SUBDISK_S_STALE); + if (sd != NULL) { + sd->sd_rebuild_pos = 0; + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_RESYNC); + g_raid_write_metadata(vol->v_softc, vol, sd, NULL); + } else { + sd = g_raid_get_subdisk(vol, + G_RAID_SUBDISK_S_UNINITIALIZED); + if (sd == NULL) + sd = g_raid_get_subdisk(vol, + G_RAID_SUBDISK_S_NEW); + if (sd != NULL) { + sd->sd_rebuild_pos = 0; + g_raid_change_subdisk_state(sd, + G_RAID_SUBDISK_S_REBUILD); + g_raid_write_metadata(vol->v_softc, + vol, sd, NULL); + } + } + } + if (sd == NULL) { + G_RAID_DEBUG1(1, vol->v_softc, + "No failed disk to rebuild. night night."); + return; + } + trs->trso_failed_sd = sd; + G_RAID_DEBUG1(0, vol->v_softc, + "Subdisk %s:%d-%s rebuild start at %jd.", + sd->sd_volume->v_name, sd->sd_pos, + sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]", + trs->trso_failed_sd->sd_rebuild_pos); + trs->trso_type = TR_RAID1E_REBUILD; + trs->trso_buffer = malloc(g_raid1e_rebuild_slab, M_TR_RAID1E, M_WAITOK); + trs->trso_meta_update = g_raid1e_rebuild_meta_update; + g_raid_tr_raid1e_rebuild_some(tr); +} + +static void +g_raid_tr_raid1e_maybe_rebuild(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd) +{ + struct g_raid_volume *vol; + struct g_raid_tr_raid1e_object *trs; + int nr; + + vol = tr->tro_volume; + trs = (struct g_raid_tr_raid1e_object *)tr; + if (trs->trso_stopping) + return; + nr = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_REBUILD) + + g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_RESYNC); + switch(trs->trso_type) { + case TR_RAID1E_NONE: + if (vol->v_state < G_RAID_VOLUME_S_DEGRADED) + return; + if (nr == 0) { + nr = g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_NEW) + + g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_STALE) + + g_raid_nsubdisks(vol, G_RAID_SUBDISK_S_UNINITIALIZED); + if (nr == 0) + return; + } + g_raid_tr_raid1e_rebuild_start(tr); + break; + case TR_RAID1E_REBUILD: + if (vol->v_state < G_RAID_VOLUME_S_DEGRADED || nr == 0 || + trs->trso_failed_sd == sd) + g_raid_tr_raid1e_rebuild_abort(tr); + break; + case TR_RAID1E_RESYNC: + break; + } +} + +static int +g_raid_tr_event_raid1e(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd, u_int event) +{ + + g_raid_tr_update_state_raid1e(tr->tro_volume, sd); + return (0); +} + +static int +g_raid_tr_start_raid1e(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1e_object *trs; + struct g_raid_volume *vol; + + trs = (struct g_raid_tr_raid1e_object *)tr; + vol = tr->tro_volume; + trs->trso_starting = 0; + g_raid_tr_update_state_raid1e(vol, NULL); + return (0); +} + +static int +g_raid_tr_stop_raid1e(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1e_object *trs; + struct g_raid_volume *vol; + + trs = (struct g_raid_tr_raid1e_object *)tr; + vol = tr->tro_volume; + trs->trso_starting = 0; + trs->trso_stopping = 1; + g_raid_tr_update_state_raid1e(vol, NULL); + return (0); +} + +/* + * Select the disk to read from. Take into account: subdisk state, running + * error recovery, average disk load, head position and possible cache hits. + */ +#define ABS(x) (((x) >= 0) ? (x) : (-(x))) +static int +g_raid_tr_raid1e_select_read_disk(struct g_raid_volume *vol, + int no, off_t off, off_t len, u_int mask) +{ + struct g_raid_subdisk *sd; + off_t offset; + int i, best, prio, bestprio; + + best = -1; + bestprio = INT_MAX; + for (i = 0; i < N; i++) { + sd = &vol->v_subdisks[(no + i) % vol->v_disks_count]; + offset = off; + if (no + i >= vol->v_disks_count) + offset += vol->v_strip_size; + + prio = G_RAID_SUBDISK_LOAD(sd); + if ((mask & (1 << sd->sd_pos)) != 0) + continue; + switch (sd->sd_state) { + case G_RAID_SUBDISK_S_ACTIVE: + break; + case G_RAID_SUBDISK_S_RESYNC: + if (offset + off < sd->sd_rebuild_pos) + break; + /* FALLTHROUGH */ + case G_RAID_SUBDISK_S_STALE: + prio += i << 24; + break; + case G_RAID_SUBDISK_S_REBUILD: + if (offset + off < sd->sd_rebuild_pos) + break; + /* FALLTHROUGH */ + default: + continue; + } + prio += min(sd->sd_recovery, 255) << 16; + /* If disk head is precisely in position - highly prefer it. */ + if (G_RAID_SUBDISK_POS(sd) == offset) + prio -= 2 * G_RAID_SUBDISK_LOAD_SCALE; + else + /* If disk head is close to position - prefer it. */ + if (ABS(G_RAID_SUBDISK_POS(sd) - offset) < + G_RAID_SUBDISK_TRACK_SIZE) + prio -= 1 * G_RAID_SUBDISK_LOAD_SCALE; + if (prio < bestprio) { + bestprio = prio; + best = i; + } + } + return (best); +} + +static void +g_raid_tr_iostart_raid1e_read(struct g_raid_tr_object *tr, struct bio *bp) +{ + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct bio_queue_head queue; + struct bio *cbp; + char *addr; + off_t offset, start, length, remain; + u_int no, strip_size; + int best; + + vol = tr->tro_volume; + addr = bp->bio_data; + strip_size = vol->v_strip_size; + V2P(vol, bp->bio_offset, &no, &offset, &start); + remain = bp->bio_length; + bioq_init(&queue); + while (remain > 0) { + length = MIN(strip_size - start, remain); + best = g_raid_tr_raid1e_select_read_disk(vol, + no, offset, length, 0); + KASSERT(best >= 0, ("No readable disk in volume %s!", + vol->v_name)); + no += best; + if (no >= vol->v_disks_count) { + no -= vol->v_disks_count; + offset += strip_size; + } + cbp = g_clone_bio(bp); + if (cbp == NULL) + goto failure; + cbp->bio_offset = offset + start; + cbp->bio_data = addr; + cbp->bio_length = length; + cbp->bio_caller1 = &vol->v_subdisks[no]; + bioq_insert_tail(&queue, cbp); + no += N - best; + if (no >= vol->v_disks_count) { + no -= vol->v_disks_count; + offset += strip_size; + } + remain -= length; + addr += length; + start = 0; + } + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + sd = cbp->bio_caller1; + cbp->bio_caller1 = NULL; + g_raid_subdisk_iostart(sd, cbp); + } + return; +failure: + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + g_destroy_bio(cbp); + } + if (bp->bio_error == 0) + bp->bio_error = ENOMEM; + g_raid_iodone(bp, bp->bio_error); +} + +static void +g_raid_tr_iostart_raid1e_write(struct g_raid_tr_object *tr, struct bio *bp) +{ + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct bio_queue_head queue; + struct bio *cbp; + char *addr; + off_t offset, start, length, remain; + u_int no, strip_size; + int i; + + vol = tr->tro_volume; + addr = bp->bio_data; + strip_size = vol->v_strip_size; + V2P(vol, bp->bio_offset, &no, &offset, &start); + remain = bp->bio_length; + bioq_init(&queue); + while (remain > 0) { + length = MIN(strip_size - start, remain); + for (i = 0; i < N; i++) { + sd = &vol->v_subdisks[no]; + switch (sd->sd_state) { + case G_RAID_SUBDISK_S_ACTIVE: + case G_RAID_SUBDISK_S_STALE: + case G_RAID_SUBDISK_S_RESYNC: + break; + case G_RAID_SUBDISK_S_REBUILD: + if (offset + start >= sd->sd_rebuild_pos) + goto nextdisk; + break; + default: + goto nextdisk; + } + cbp = g_clone_bio(bp); + if (cbp == NULL) + goto failure; + cbp->bio_offset = offset + start; + cbp->bio_data = addr; + cbp->bio_length = length; + cbp->bio_caller1 = sd; + bioq_insert_tail(&queue, cbp); +nextdisk: + if (++no >= vol->v_disks_count) { + no = 0; + offset += strip_size; + } + } + remain -= length; + addr += length; + start = 0; + } + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + sd = cbp->bio_caller1; + cbp->bio_caller1 = NULL; + g_raid_subdisk_iostart(sd, cbp); + } + return; +failure: + for (cbp = bioq_first(&queue); cbp != NULL; + cbp = bioq_first(&queue)) { + bioq_remove(&queue, cbp); + g_destroy_bio(cbp); + } + if (bp->bio_error == 0) + bp->bio_error = ENOMEM; + g_raid_iodone(bp, bp->bio_error); +} + +static void +g_raid_tr_iostart_raid1e(struct g_raid_tr_object *tr, struct bio *bp) +{ + struct g_raid_volume *vol; + struct g_raid_tr_raid1e_object *trs; + + vol = tr->tro_volume; + trs = (struct g_raid_tr_raid1e_object *)tr; + if (vol->v_state != G_RAID_VOLUME_S_OPTIMAL && + vol->v_state != G_RAID_VOLUME_S_SUBOPTIMAL && + vol->v_state != G_RAID_VOLUME_S_DEGRADED) { + g_raid_iodone(bp, EIO); + return; + } + /* + * If we're rebuilding, squeeze in rebuild activity every so often, + * even when the disk is busy. Be sure to only count real I/O + * to the disk. All 'SPECIAL' I/O is traffic generated to the disk + * by this module. + */ + if (trs->trso_failed_sd != NULL && + !(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL)) { + /* Make this new or running now round short. */ + trs->trso_recover_slabs = 0; + if (--trs->trso_fair_io <= 0) { + trs->trso_fair_io = g_raid1e_rebuild_fair_io; + g_raid_tr_raid1e_rebuild_some(tr); + } + } + switch (bp->bio_cmd) { + case BIO_READ: + g_raid_tr_iostart_raid1e_read(tr, bp); + break; + case BIO_WRITE: + g_raid_tr_iostart_raid1e_write(tr, bp); + break; + case BIO_DELETE: + g_raid_iodone(bp, EIO); + break; + case BIO_FLUSH: + g_raid_tr_flush_common(tr, bp); + break; + default: + KASSERT(1 == 0, ("Invalid command here: %u (volume=%s)", + bp->bio_cmd, vol->v_name)); + break; + } +} + +static void +g_raid_tr_iodone_raid1e(struct g_raid_tr_object *tr, + struct g_raid_subdisk *sd, struct bio *bp) +{ + struct bio *cbp; + struct g_raid_subdisk *nsd; + struct g_raid_volume *vol; + struct bio *pbp; + struct g_raid_tr_raid1e_object *trs; + off_t virtual, offset, start; + uintptr_t mask; + int error, do_write, copy, disk, best; + + trs = (struct g_raid_tr_raid1e_object *)tr; + vol = tr->tro_volume; + if (bp->bio_cflags & G_RAID_BIO_FLAG_SYNC) { + if (trs->trso_type == TR_RAID1E_REBUILD) { + nsd = trs->trso_failed_sd; + if (bp->bio_cmd == BIO_READ) { + + /* Immediately abort rebuild, if requested. */ + if (trs->trso_flags & TR_RAID1E_F_ABORT) { + trs->trso_flags &= ~TR_RAID1E_F_DOING_SOME; + g_raid_tr_raid1e_rebuild_abort(tr); + return; + } + + /* On read error, skip and cross fingers. */ + if (bp->bio_error != 0) { + G_RAID_LOGREQ(0, bp, + "Read error during rebuild (%d), " + "possible data loss!", + bp->bio_error); + goto rebuild_round_done; + } + + /* + * The read operation finished, queue the + * write and get out. + */ + G_RAID_LOGREQ(3, bp, "Rebuild read done: %d", + bp->bio_error); + bp->bio_cmd = BIO_WRITE; + bp->bio_cflags = G_RAID_BIO_FLAG_SYNC; + bp->bio_offset = nsd->sd_rebuild_pos; + G_RAID_LOGREQ(3, bp, "Queueing rebuild write."); + g_raid_subdisk_iostart(nsd, bp); + } else { + /* + * The write operation just finished. Do + * another. We keep cloning the master bio + * since it has the right buffers allocated to + * it. + */ + G_RAID_LOGREQ(3, bp, "Rebuild write done: %d", + bp->bio_error); + if (bp->bio_error != 0 || + trs->trso_flags & TR_RAID1E_F_ABORT) { + if ((trs->trso_flags & + TR_RAID1E_F_ABORT) == 0) { + g_raid_tr_raid1e_fail_disk(sd->sd_softc, + nsd, nsd->sd_disk); + } + trs->trso_flags &= ~TR_RAID1E_F_DOING_SOME; + g_raid_tr_raid1e_rebuild_abort(tr); + return; + } +rebuild_round_done: + trs->trso_flags &= ~TR_RAID1E_F_LOCKED; + g_raid_unlock_range(tr->tro_volume, + trs->trso_lock_pos, trs->trso_lock_len); + nsd->sd_rebuild_pos += bp->bio_length; + if (nsd->sd_rebuild_pos >= nsd->sd_size) { + g_raid_tr_raid1e_rebuild_finish(tr); + return; + } + + /* Abort rebuild if we are stopping */ + if (trs->trso_stopping) { + trs->trso_flags &= ~TR_RAID1E_F_DOING_SOME; + g_raid_tr_raid1e_rebuild_abort(tr); + return; + } + + if (--trs->trso_meta_update <= 0) { + g_raid_write_metadata(vol->v_softc, + vol, nsd, nsd->sd_disk); + trs->trso_meta_update = + g_raid1e_rebuild_meta_update; + /* Compensate short rebuild I/Os. */ + if ((vol->v_disks_count % N) != 0 && + vol->v_strip_size < + g_raid1e_rebuild_slab) { + trs->trso_meta_update *= + g_raid1e_rebuild_slab; + trs->trso_meta_update /= + vol->v_strip_size; + } + } + trs->trso_flags &= ~TR_RAID1E_F_DOING_SOME; + if (--trs->trso_recover_slabs <= 0) + return; + /* Run next rebuild iteration. */ + g_raid_tr_raid1e_rebuild_some(tr); + } + } else if (trs->trso_type == TR_RAID1E_RESYNC) { + /* + * read good sd, read bad sd in parallel. when both + * done, compare the buffers. write good to the bad + * if different. do the next bit of work. + */ + panic("Somehow, we think we're doing a resync"); + } + return; + } + pbp = bp->bio_parent; + pbp->bio_inbed++; + mask = (intptr_t)bp->bio_caller2; + if (bp->bio_cmd == BIO_READ && bp->bio_error != 0) { + /* + * Read failed on first drive. Retry the read error on + * another disk drive, if available, before erroring out the + * read. + */ + sd->sd_disk->d_read_errs++; + G_RAID_LOGREQ(0, bp, + "Read error (%d), %d read errors total", + bp->bio_error, sd->sd_disk->d_read_errs); + + /* + * If there are too many read errors, we move to degraded. + * XXX Do we want to FAIL the drive (eg, make the user redo + * everything to get it back in sync), or just degrade the + * drive, which kicks off a resync? + */ + do_write = 0; + if (sd->sd_disk->d_read_errs > g_raid_read_err_thresh) + g_raid_tr_raid1e_fail_disk(sd->sd_softc, sd, sd->sd_disk); + else if (mask == 0) + do_write = 1; + + /* Restore what we were doing. */ + P2V(vol, sd->sd_pos, bp->bio_offset, &virtual, ©); + V2P(vol, virtual, &disk, &offset, &start); + + /* Find the other disk, and try to do the I/O to it. */ + mask |= 1 << copy; + best = g_raid_tr_raid1e_select_read_disk(vol, + disk, offset, start, mask); + if (best >= 0 && (cbp = g_clone_bio(pbp)) != NULL) { + disk += best; + if (disk >= vol->v_disks_count) { + disk -= vol->v_disks_count; + offset += vol->v_strip_size; + } + cbp->bio_offset = offset + start; + cbp->bio_length = bp->bio_length; + cbp->bio_data = bp->bio_data; + g_destroy_bio(bp); + nsd = &vol->v_subdisks[disk]; + G_RAID_LOGREQ(2, cbp, "Retrying read from %d", + nsd->sd_pos); + if (do_write) + mask |= 1 << 31; + if ((mask & (1 << 31)) != 0) + sd->sd_recovery++; + cbp->bio_caller2 = (void *)mask; + if (do_write) { + cbp->bio_caller1 = nsd; + /* Lock callback starts I/O */ + g_raid_lock_range(sd->sd_volume, + virtual, cbp->bio_length, pbp, cbp); + } else { + g_raid_subdisk_iostart(nsd, cbp); + } + return; + } + /* + * We can't retry. Return the original error by falling + * through. This will happen when there's only one good disk. + * We don't need to fail the raid, since its actual state is + * based on the state of the subdisks. + */ + G_RAID_LOGREQ(2, bp, "Couldn't retry read, failing it"); + } + if (bp->bio_cmd == BIO_READ && + bp->bio_error == 0 && + (mask & (1 << 31)) != 0) { + G_RAID_LOGREQ(3, bp, "Recovered data from other drive"); + + /* Restore what we were doing. */ + P2V(vol, sd->sd_pos, bp->bio_offset, &virtual, ©); + V2P(vol, virtual, &disk, &offset, &start); + + /* Find best disk to write. */ + best = g_raid_tr_raid1e_select_read_disk(vol, + disk, offset, start, ~mask); + if (best >= 0 && (cbp = g_clone_bio(pbp)) != NULL) { + disk += best; + if (disk >= vol->v_disks_count) { + disk -= vol->v_disks_count; + offset += vol->v_strip_size; + } + cbp->bio_offset = offset + start; + cbp->bio_length = bp->bio_length; + cbp->bio_data = bp->bio_data; + cbp->bio_cmd = BIO_WRITE; + cbp->bio_cflags = G_RAID_BIO_FLAG_REMAP; + cbp->bio_caller2 = (void *)mask; + g_destroy_bio(bp); + G_RAID_LOGREQ(2, cbp, + "Attempting bad sector remap on failing drive."); + g_raid_subdisk_iostart(&vol->v_subdisks[disk], cbp); + return; + } + } + if ((mask & (1 << 31)) != 0) { + /* + * We're done with a recovery, mark the range as unlocked. + * For any write errors, we agressively fail the disk since + * there was both a READ and a WRITE error at this location. + * Both types of errors generally indicates the drive is on + * the verge of total failure anyway. Better to stop trusting + * it now. However, we need to reset error to 0 in that case + * because we're not failing the original I/O which succeeded. + */ + + /* Restore what we were doing. */ + P2V(vol, sd->sd_pos, bp->bio_offset, &virtual, ©); + V2P(vol, virtual, &disk, &offset, &start); + + for (copy = 0; copy < N; copy++) { + if ((mask & (1 << copy) ) != 0) + vol->v_subdisks[(disk + copy) % + vol->v_disks_count].sd_recovery--; + } + + if (bp->bio_cmd == BIO_WRITE && bp->bio_error) { + G_RAID_LOGREQ(0, bp, "Remap write failed: " + "failing subdisk."); + g_raid_tr_raid1e_fail_disk(sd->sd_softc, sd, sd->sd_disk); + bp->bio_error = 0; + } + G_RAID_LOGREQ(2, bp, "REMAP done %d.", bp->bio_error); + g_raid_unlock_range(sd->sd_volume, virtual, bp->bio_length); + } + error = bp->bio_error; + g_destroy_bio(bp); + if (pbp->bio_children == pbp->bio_inbed) { + pbp->bio_completed = pbp->bio_length; + g_raid_iodone(pbp, error); + } +} + +static int +g_raid_tr_kerneldump_raid1e(struct g_raid_tr_object *tr, + void *virtual, vm_offset_t physical, off_t boffset, size_t blength) +{ + struct g_raid_volume *vol; + struct g_raid_subdisk *sd; + struct bio_queue_head queue; + char *addr; + off_t offset, start, length, remain; + u_int no, strip_size; + int i, error; + + vol = tr->tro_volume; + addr = virtual; + strip_size = vol->v_strip_size; + V2P(vol, boffset, &no, &offset, &start); + remain = blength; + bioq_init(&queue); + while (remain > 0) { + length = MIN(strip_size - start, remain); + for (i = 0; i < N; i++) { + sd = &vol->v_subdisks[no]; + switch (sd->sd_state) { + case G_RAID_SUBDISK_S_ACTIVE: + case G_RAID_SUBDISK_S_STALE: + case G_RAID_SUBDISK_S_RESYNC: + break; + case G_RAID_SUBDISK_S_REBUILD: + if (offset + start >= sd->sd_rebuild_pos) + goto nextdisk; + break; + default: + goto nextdisk; + } + error = g_raid_subdisk_kerneldump(sd, + addr, 0, offset + start, length); + if (error != 0) + return (error); +nextdisk: + if (++no >= vol->v_disks_count) { + no = 0; + offset += strip_size; + } + } + remain -= length; + addr += length; + start = 0; + } + return (0); +} + +static int +g_raid_tr_locked_raid1e(struct g_raid_tr_object *tr, void *argp) +{ + struct bio *bp; + struct g_raid_subdisk *sd; + + bp = (struct bio *)argp; + sd = (struct g_raid_subdisk *)bp->bio_caller1; + g_raid_subdisk_iostart(sd, bp); + + return (0); +} + +static int +g_raid_tr_idle_raid1e(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1e_object *trs; + struct g_raid_volume *vol; + + vol = tr->tro_volume; + trs = (struct g_raid_tr_raid1e_object *)tr; + trs->trso_fair_io = g_raid1e_rebuild_fair_io; + trs->trso_recover_slabs = g_raid1e_rebuild_cluster_idle; + /* Compensate short rebuild I/Os. */ + if ((vol->v_disks_count % N) != 0 && + vol->v_strip_size < g_raid1e_rebuild_slab) { + trs->trso_recover_slabs *= g_raid1e_rebuild_slab; + trs->trso_recover_slabs /= vol->v_strip_size; + } + if (trs->trso_type == TR_RAID1E_REBUILD) + g_raid_tr_raid1e_rebuild_some(tr); + return (0); +} + +static int +g_raid_tr_free_raid1e(struct g_raid_tr_object *tr) +{ + struct g_raid_tr_raid1e_object *trs; + + trs = (struct g_raid_tr_raid1e_object *)tr; + + if (trs->trso_buffer != NULL) { + free(trs->trso_buffer, M_TR_RAID1E); + trs->trso_buffer = NULL; + } + return (0); +} + +G_RAID_TR_DECLARE(g_raid_tr_raid1e); diff --git a/sys/modules/geom/Makefile b/sys/modules/geom/Makefile index 0b2e3e8b6220..ca7d7e639558 100644 --- a/sys/modules/geom/Makefile +++ b/sys/modules/geom/Makefile @@ -18,6 +18,7 @@ SUBDIR= geom_bde \ geom_nop \ geom_part \ geom_pc98 \ + geom_raid \ geom_raid3 \ geom_sched \ geom_shsec \ diff --git a/sys/modules/geom/geom_raid/Makefile b/sys/modules/geom/geom_raid/Makefile new file mode 100644 index 000000000000..4487807161ac --- /dev/null +++ b/sys/modules/geom/geom_raid/Makefile @@ -0,0 +1,19 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../geom/raid + +KMOD= geom_raid +SRCS= g_raid.c +SRCS+= g_raid_ctl.c +SRCS+= bus_if.h device_if.h +SRCS+= g_raid_md_if.h g_raid_md_if.c +SRCS+= g_raid_tr_if.h g_raid_tr_if.c + +SRCS+= md_intel.c md_jmicron.c md_nvidia.c md_promise.c md_sii.c + +SRCS+= tr_concat.c tr_raid0.c tr_raid1.c tr_raid1e.c + +MFILES= kern/bus_if.m kern/device_if.m +MFILES+= geom/raid/g_raid_md_if.m geom/raid/g_raid_tr_if.m + +.include