#!/bin/sh
#
# License:      GNU General Public License (GPL)
# Support:      users@clusterlabs.org
#
# mdraid (inspired by the Raid1 upstream resource agent)
#
#      Description: Manages a Linux software RAID device on a (shared) storage medium.
#           Author: Heinz Mauelshagen (heinzm@redhat.com)
#          Release: Mar 2020
#
# usage: $0 {start|stop|monitor|validate-all|usage|meta-data}
#
# EXAMPLE config file /etc/mdadm.conf (for more info: mdadm.conf(5))
#
#  AUTO -all
#  ARRAY /dev/md0 UUID=4a865b55:ba27ef8d:29cd5701:6fb42799
#
#######################################################################
# Initialization:

: ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/lib/heartbeat}
. ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs

# Parameter defaults

OCF_RESKEY_mdadm_conf_default=""
OCF_RESKEY_md_dev_default=""
OCF_RESKEY_force_stop_default="false"
OCF_RESKEY_wait_for_udev_default="true"
OCF_RESKEY_force_clones_default="false"

: ${OCF_RESKEY_mdadm_conf=${OCF_RESKEY_mdadm_conf_default}}
: ${OCF_RESKEY_md_dev=${OCF_RESKEY_md_dev_default}}
: ${OCF_RESKEY_force_stop=${OCF_RESKEY_force_stop_default}}
: ${OCF_RESKEY_wait_for_udev=${OCF_RESKEY_wait_for_udev_default}}
: ${OCF_RESKEY_force_clones=${OCF_RESKEY_force_clones_default}}

#######################################################################

usage() {
	cat <<-EOT
	usage: $0 {start|stop|monitor|validate-all|usage|meta-data}
	EOT
}

#
# Action: provide meta-data (parameter specifications and descriptive text)
#
meta_data() {
	cat <<END
<?xml version="1.0"?>
<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
<resource-agent name="mdraid" version="1.0">
<version>1.0</version>

<longdesc lang="en">
This resource agent manages Linux software RAID (MD) devices on
a shared storage medium ensuring that non-clustered MD arrays
are prohibited from starting cloned (which would cause data corruption
(e.g., on raid6 arrays) unless forced (see force_clones parameter).
Clustered MD RAID layouts (see below) will be discovered and allowed
cloning by default; no need to set force_clones.

It uses mdadm(8) to start, stop, and monitor the MD devices.

Supported clustered (i.e., clonable active-active) arrays are linear,
raid0, and clustered raid1/raid10 (i.e. mdadm(8) created with
--bitmap=clustered).

Option: OCF_CHECK_LEVEL

When OCF_CHECK_LEVEL is set to any number greater than 0, the standard
monitor operation (including probe) will check the array and attempt
recovery sequence to re-add devices if any failed device exists. By
default, OCF_CHECK_LEVEL is unset, and this is disabled.

</longdesc>
<shortdesc lang="en">Manages Linux software RAID (MD) devices on shared
storage</shortdesc>

<parameters>
<parameter name="mdadm_conf" unique="0" required="1">
<longdesc lang="en">
The MD RAID configuration file (e.g., /etc/mdadm.conf).
</longdesc>
<shortdesc lang="en">MD config file</shortdesc>
<content type="string" default="${OCF_RESKEY_mdadm_conf_default}" />
</parameter>

<parameter name="md_dev" unique="0" required="1">
<longdesc lang="en">
MD array block device to use (e.g., /dev/md0 or /dev/md/3).
With shared access to the array's storage, this should
preferably be a clustered raid1 or raid10 array created
with --bitmap=clustered, assuming its resource will
be cloned (i.e., active-active access).

Be sure to disable auto-assembly for the resource-managed arrays!
</longdesc>
<shortdesc lang="en">MD block device</shortdesc>
<content type="string" default="${OCF_RESKEY_md_dev_default}" />
</parameter>

<parameter name="force_stop" unique="0" required="0">
<longdesc lang="en">
If processes or kernel threads are using the array, it cannot be
stopped. We will try to stop processes, first by sending TERM and
then, if that doesn't help in $PROC_CLEANUP_TIME seconds, using KILL.
The lsof(8) program is required to get the list of array users.
Of course, the kernel threads cannot be stopped this way.
If the processes are critical for data integrity, then set this
parameter to false. Note that in that case the stop operation
will fail and the node will be fenced.
</longdesc>
<shortdesc lang="en">force stop processes using the array</shortdesc>
<content type="boolean" default="${OCF_RESKEY_force_stop_default}" />
</parameter>

<parameter name="wait_for_udev" unique="0" required="0">
<longdesc lang="en">
Wait until udevd creates a device in the start operation. On a
normally loaded host this should happen quickly, but you may be
unlucky. If you are not using udev set this to "no".
</longdesc>
<shortdesc lang="en">wait_for_udev</shortdesc>
<content type="boolean" default="${OCF_RESKEY_wait_for_udev_default}" />
</parameter>

<parameter name="force_clones" unique="0" required="0">
<longdesc lang="en">
Activating the same, non-clustered MD RAID array (i.e. single-host
raid1/4/5/6/10) on multiple nodes at the same time will result in
data corruption and thus is forbidden by default.

A safe example could be an (exotic) array that is only named identically
across all nodes, but is in fact based on distinct (non-shared) storage.

Only set this to "true" if you know what you are doing!
</longdesc>
<shortdesc lang="en">force ability to run as a clone</shortdesc>
<content type="boolean" default="${OCF_RESKEY_force_clones_default}" />
</parameter>

</parameters>

<actions>
<action name="start" timeout="20s" />
<action name="stop" timeout="20s" />
<action name="monitor" depth="0" timeout="20s" interval="10s" />
<action name="validate-all" timeout="5s" />
<action name="meta-data" timeout="5s" />
</actions>
</resource-agent>
END
}

#######################################################################

# ocf-shellfunc ocf_is_clone() fails with meta attribute clone-max < 2.
# Checking for defined meta_clone_max reskey is sufficient until fixed.
resource_is_cloned() {
	[ -z "$OCF_RESKEY_CRM_meta_clone_max" ] && return 1 || return 0;
}

raid_validate_all() {
	if [ -z "$mdadm_conf" ] ; then
		ocf_exit_reason "Please set OCF_RESKEY_mdadm_conf"
		return $OCF_ERR_CONFIGURED
	fi
	if [ ! -r "$mdadm_conf" ] ; then
		ocf_exit_reason "Configuration file [$mdadm_conf] does not exist, or cannot be opened"
		return $OCF_ERR_ARGS
	fi
	if [ -z "$md_dev" ] ; then
		ocf_exit_reason "Please set OCF_RESKEY_md_dev to the MD RAID array block device you want to control"
		return $OCF_ERR_CONFIGURED
	fi
	case "$md_dev" in
	/dev/*) ;;
	*)	ocf_exit_reason "Bogus MD RAID array block device name (\"$md_dev\")"
		return $OCF_ERR_ARGS;;
	esac
	if ocf_is_true $wait_for_udev && ! have_binary udevadm && [ "$__OCF_ACTION" = "start" ]; then
		ocf_exit_reason "either install udevadm or set udev to false"
		return $OCF_ERR_INSTALLED
	fi
	if ocf_is_true $force_stop && ! have_binary lsof; then
		ocf_exit_reason "Please install lsof(8) or set force_stop to false."
		return $OCF_ERR_INSTALLED
	fi
	if ! have_binary $MDADM; then
		ocf_exit_reason "Please install mdadm(8)!"
		return $OCF_ERR_INSTALLED
	fi
	if ! have_binary blkid; then
		ocf_exit_reason "Please install blkid(8). We need it to list MD array UUIDs!"
		return $OCF_ERR_INSTALLED
	fi
	if [ `echo $md_dev|wc -w` -gt 1 ]; then
		ocf_exit_reason "Only one MD array supported"
		return $OCF_ERR_CONFIGURED
	fi

	return $OCF_SUCCESS
}

# Remove ':' or '-' from uuid string to be able to compare between MD and blkid format.
uuid_flat() {
	echo $1|sed 's/[-:]//g'
}

# Global variable for devices by MD uuid.
devs=""

# Get array uuid from mdadm_conf based on $md_dev.
get_array_uuid_by_mddev() {
	local array_uuid

	array_uuid="`grep $md_dev $mdadm_conf`"
	if [ -z "$array_uuid" ]
	then
		ocf_exit_reason "Entry for $MMDEV does not exist in $mdadm_conf!"
		return $OCF_ERR_CONFIGURED
	fi

	array_uuid=$(echo $array_uuid | sed 's/^.*UUID=//;s/ .*$//')
	if [ -z "$array_uuid" ]
	then
		ocf_exit_reason "Bogus entry for $MMDEV in $mdadm_conf!"
		return $OCF_ERR_CONFIGURED
	fi

	echo `uuid_flat $array_uuid`

	return $OCF_SUCCESS
}

# Use blkid to get to the subset of raid members by array uuid.
list_devices_for_mddev() {
	local array_uuid blkid_array_uuid dev line rc

	array_uuid=`get_array_uuid_by_mddev`
	rc=$?
	if [ $rc -ne $OCF_SUCCESS ]; then
		ocf_exit_reason "Failed to get UUID of $md_dev from $mdadm_conf"
		return $rc
	fi

	blkid | grep linux_raid_member 2>/dev/null | while read line
	do
		dev=`echo $line | sed 's/: .*//'`
		blkid_array_uuid=$(echo $line | sed 's/^.* UUID="//;s/" .*$//')
		[ "`uuid_flat $blkid_array_uuid`" = "$array_uuid" ] && echo $dev
	done
}

# Check for linear or raid0 array; presumes defined global devs() array.
array_is_linear_or_raid0() {
	local c=0 d

	for d in $devs
	do
		$MDADM -E $d 2>&1 | $EGREP -i "raid level : (raid0|linear)" >/dev/null 2>&1
		[ $? -eq 0 ] && c=$((c+1))
	done

	[ $c -eq `echo $devs|wc -w` ] && return 0 || return 1
}

# Return true for clustered RAID relying on all component devices reporting clustered type;
# presumes defined global devs variable with the component devices of the array.
array_is_clustered_raid() {
	local c=0 d dev_count=`echo $devs|wc -w` s
	# Check based on specific "intent bitmap : clustered" output once
	# available in mdadm output or fall back to "Cluster Name" defined
	# presuming bitmap is clustered if so.
	local strs="clustered cluster.name"

	for d in $devs
	do
		for s in $strs
		do
			$MDADM -E $d 2>&1 | grep -i "$s" >/dev/null 2>&1
			if [ $? -eq 0 ]; then
				c=$((c+1))
				break
			fi
		done
	done

	[ $c -eq $dev_count ] && return 0 || return 1
}

# Check for all clustered types (linear, raid0, and clustered raid1/raid10).
is_clustered_raid() {
	array_is_clustered_raid || array_is_linear_or_raid0
}

md_assemble() {
	local rc

	$MDADM --assemble $md_dev --config="$mdadm_conf"
	rc=$?
	[ $rc -eq 0 ] && ocf_is_true $wait_for_udev && udevadm settle --exit-if-exists=$md_dev

	return $rc
}

# Try setting an MD array to readonly.
mark_readonly() {
	local rc

	$MDADM --readonly $md_dev --config="$mdadm_conf"
	rc=$?
	[ $rc -ne 0 ] && ocf_exit_reason "Failed to set $md_dev readonly (rc=$rc)"

	return $rc
}

# Try stopping an MD array in case its block device is nonexistent for some reason.
mknod_raid_stop() {
	local rc n tmp_block_file

	# first create a block device file, then try to stop the array
	n=`echo $1 | sed 's/[^0-9]*//'`
	if ! ocf_is_decimal "$n"; then
		ocf_log warn "could not get the minor device number from $1"
		return 1
	fi
	tmp_block_file="$HA_RSCTMP/${OCF_RESOURCE_INSTANCE}-`basename $1`"
	rm -f $tmp_block_file
	ocf_log info "block device file $1 missing, creating one in order to stop the array"
	mknod $tmp_block_file b 9 $n
	$MDADM --stop $tmp_block_file --config="$mdadm_conf"
	rc=$?
	rm -f $tmp_block_file
	return $rc
}

# Stop an MD array.
raid_stop_one() {
	if [ -b "$1" ]; then
		$MDADM --stop $1 --config="$mdadm_conf" && return
	else
		# newer mdadm releases can stop arrays when given the
		# basename; try that first
		$MDADM --stop `basename $1` --config="$mdadm_conf" && return
		# otherwise create a block device file
		mknod_raid_stop $1
	fi
}

# Functions show/stop any resource holding processes.
get_users_pids() {
	ocf_log debug "running lsof to list $md_dev users..."
	ocf_run -warn 'lsof $md_dev | tail -n +2 | $AWK "{print $2}" | sort -u'
}

stop_raid_users() {
	local pids=`get_users_pids $md_dev`

	if [ -z "$pids" ]; then
		ocf_log warn "lsof reported no users holding arrays"
		return 2
	else
		ocf_stop_processes TERM $PROC_CLEANUP_TIME $pids
	fi
}

showusers() {
	local disk=`basename $md_dev`

	ocf_log info "running lsof to list $disk users..."
	ocf_run -warn lsof $md_dev

	if [ -d /sys/block/$disk/holders ]; then
		ocf_log info "ls -l /sys/block/$disk/holders"
		ocf_run -warn ls -l /sys/block/$disk/holders
	fi
}

#######################################################################

#
# Action: START up the MD RAID array.
#
raid_start() {
	local rc

	if resource_is_cloned && ! is_clustered_raid; then
		if ocf_is_true "$OCF_RESKEY_force_clones"; then
			ocf_log warn "Forced cloned starting non-clustered $md_dev which may lead to data corruption!"
		else
			ocf_exit_reason "Rejecting start: non-clustered MD RAID array $md_dev is NOT safe to run cloned"
			exit $OCF_ERR_CONFIGURED
		fi
	fi

	raid_monitor
	rc=$?
	# md array already online, nothing to do.
	[ $rc -eq $OCF_SUCCESS ] && return $rc

	if [ $rc -ne $OCF_NOT_RUNNING ]
	then
		# If the array is in a broken state, this agent doesn't know how to repair that.
		ocf_exit_reason "MD RAID array $md_dev in a broken state; cannot start (rc=$rc)"
		return $OCF_ERR_GENERIC
	fi

	md_assemble
	rc=$?
	if [ $rc -ne 0 ]; then
		ocf_exit_reason "Failed to assemble MD RAID array $md_dev (rc=$rc, is $mdadm_conf up-to-date?)"
		return $OCF_ERR_GENERIC
	fi

	raid_monitor
	[ $? -eq $OCF_SUCCESS ] && return $OCF_SUCCESS

	ocf_exit_reason "Couldn't start MD RAID array $md_dev (rc=$rc)"

	return $OCF_ERR_GENERIC
}

#
# Action: STOP the MD RAID array
#
raid_stop() {
	local rc

	# See if the MD device is already cleanly stopped:
	raid_monitor
	[ $? -eq $OCF_NOT_RUNNING ] && return $OCF_SUCCESS

	# Turn off raid
	if ! raid_stop_one $md_dev; then
		if ocf_is_true $force_stop; then
			stop_raid_users
			case $? in
			2) false;;
			*) raid_stop_one $md_dev;;
			esac
		else
			false
		fi
	fi
	rc=$?

	if [ $rc -ne 0 ]; then
		ocf_log warn "Couldn't stop MD RAID array $md_dev (rc=$rc)"
		showusers $md_dev
		mark_readonly $md_dev
		return $OCF_ERR_GENERIC
	fi

	raid_monitor
	rc=$?
	[ $rc -eq $OCF_NOT_RUNNING ] && return $OCF_SUCCESS

	ocf_exit_reason "MD RAID array $md_dev still active after stop command (rc=$rc)"
	return $OCF_ERR_GENERIC
}

#
# Action: monitor the MD RAID array.
#
raid_monitor() {
	local TRY_READD=0 md rc pbsize

	# check if the md device exists first
	# but not if we are in the stop operation
	# device existence is important only for the running arrays
	if [ "$__OCF_ACTION" != "stop" ]; then
		if [ -h "$md_dev" ]; then
			md=$(ls $md_dev -l | $AWK -F'/' '{print $NF}')
		elif [ -b "$md_dev" ]; then
			md=${md_dev#/dev/}
		else
			ocf_log info "$md_dev is not a block device"
			return $OCF_NOT_RUNNING
		fi
	fi

	if ! grep -e "^$md[ \t:]" /proc/mdstat >/dev/null ; then
		ocf_log info "$md not found in /proc/mdstat"
		return $OCF_NOT_RUNNING
	fi

	$MDADM --detail --test $md_dev >/dev/null 2>&1
	rc=$?
	case $rc in
	0)	;;
	1)	ocf_log warn "$md_dev has at least one failed device."
		TRY_READD=1;;
	2)	ocf_exit_reason "$md_dev has failed."
		return $OCF_ERR_GENERIC;;
	4)
		if [ "$__OCF_ACTION" = "stop" ] ; then
			# There may be a transient invalid device after
			# we stop MD due to uevent processing, the
			# original device is stopped though.
			return $OCF_NOT_RUNNING
		else
			ocf_exit_reason "mdadm failed on $md_dev."
			return $OCF_ERR_GENERIC
		fi;;
	*)	ocf_exit_reason "mdadm returned an unknown result ($rc)."
		return $OCF_ERR_GENERIC;;
	esac

	if ! array_is_linear_or_raid0; then
		if [ "$__OCF_ACTION" = "monitor" ] && [ "$OCF_RESKEY_CRM_meta_interval" != 0 \
			] && [ $TRY_READD -eq 1 ] && [ $OCF_CHECK_LEVEL -gt 0 ]; then
			ocf_log info "Attempting recovery sequence to re-add devices on MD RAID array $md_dev:"
			$MDADM $md_dev --fail detached
			$MDADM $md_dev --remove failed
			$MDADM $md_dev --re-add missing
			# TODO: At this stage, there's nothing to actually do
			# here. Either this worked or it did not.
		fi
	fi

	pbsize=`(blockdev --getpbsz $md_dev || stat -c "%o" $md_dev) 2>/dev/null`
	if [ -z "$pbsize" ]; then
		ocf_log warn "both blockdev and stat were unable to get the block size (will use 4k)"
		pbsize=4096 # try with 4k
	fi
	if ! dd if=$md_dev count=1 bs=$pbsize of=/dev/null iflag=direct >/dev/null 2>&1; then
		ocf_exit_reason "$md_dev: I/O error on read"
		return $OCF_ERR_GENERIC
	fi

	[ "$__OCF_ACTION" = "monitor" ] && ocf_log info "monitoring...($md_dev)"

	return $OCF_SUCCESS
}

if [ $# -ne 1 ]; then
	usage
	exit $OCF_ERR_ARGS
fi

# Process actions which are independant from validation
case "$1" in
meta-data)	meta_data
		exit $OCF_SUCCESS;;
usage)		usage
		exit $OCF_SUCCESS;;
*)		;;
esac

# Define global variables used in ^ functions.
mdadm_conf="${OCF_RESKEY_mdadm_conf}"
md_dev="${OCF_RESKEY_md_dev}"
force_stop="${OCF_RESKEY_force_stop}"
wait_for_udev="${OCF_RESKEY_wait_for_udev}"

# Validate all parameters and check for mandatory binaries present
raid_validate_all
rc=$?
[ $rc -ne $OCF_SUCCESS ] && exit $rc
# raid_validate_all already processed and result checked.
[ "$1" = "validate-all" ] && return ${OCF_SUCCESS}

# Required by start|stop|monitor processed below
devs="`list_devices_for_mddev`"
if [ `echo $devs|wc -l` -eq 0 ]; then
	ocf_exit_reason "No component device(s) found for MD RAID array $md_dev"
	exit $OCF_ERR_GENERIC
fi

case "$1" in
start)		raid_start;;
stop)		raid_stop;;
monitor)	raid_monitor;;
*)		usage
		exit $OCF_ERR_UNIMPLEMENTED;;
esac
rc=$?

ocf_log debug "${OCF_RESOURCE_INSTANCE} $__OCF_ACTION : $rc"
exit $rc