#!/bin/bash
#
#
#		OCF Resource Agent compliant drbd resource script.
#
# Copyright (c) 2009 LINBIT HA-Solutions GmbH,
# Copyright (c) 2009 Florian Haas, Lars Ellenberg
# Based on the Heartbeat drbd OCF Resource Agent by Lars Marowsky-Bree
# (though it turned out to be an almost complete rewrite)
#
#					 All Rights Reserved.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of version 2 of the GNU General Public License as
# published by the Free Software Foundation.
#
# This program is distributed in the hope that it would be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
#
# Further, this software is distributed without any warranty that it is
# free of the rightful claim of any third person regarding infringement
# or the like.	Any license provided herein, whether implied or
# otherwise, applies only to this software file.  Patent licenses, if
# any, provided herein do not apply to combinations of this program with
# other software, or any other product whatsoever.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write the Free Software Foundation,
# Inc., 59 Temple Place - Suite 330, Boston MA 02111-1307, USA.
#
#

# OCF instance parameters
#
# OCF_RESKEY_drbd_resource
# OCF_RESKEY_drbdconf
# OCF_RESKEY_stop_outdates_secondary
# OCF_RESKEY_adjust_master_score
# OCF_RESKEY_ignore_missing_notifications
# OCF_RESKEY_unfence_if_all_uptodate
# OCF_RESKEY_unfence_extra_args
# OCF_RESKEY_require_drbd_module_version_ge
# OCF_RESKEY_require_drbd_module_version_lt
# OCF_RESKEY_connect_only_after_promote
# OCF_RESKEY_wfc_timeout
# OCF_RESKEY_remove_master_score_if_peer_primary
# OCF_RESKEY_fail_promote_early_if_peer_primary
#
# meta stuff this agent looks at:
#	OCF_RESKEY_CRM_meta_clone_max
#	OCF_RESKEY_CRM_meta_clone_node_max
#	OCF_RESKEY_CRM_meta_master_max
#	OCF_RESKEY_CRM_meta_master_node_max
#
#	OCF_RESKEY_CRM_meta_interval
#
#	OCF_RESKEY_CRM_meta_notify
#	OCF_RESKEY_CRM_meta_notify_active_uname
#	OCF_RESKEY_CRM_meta_notify_demote_uname
#	OCF_RESKEY_CRM_meta_notify_master_uname
#	OCF_RESKEY_CRM_meta_notify_operation
#	OCF_RESKEY_CRM_meta_notify_promote_uname
#	OCF_RESKEY_CRM_meta_notify_slave_uname
#	OCF_RESKEY_CRM_meta_notify_start_uname
#	OCF_RESKEY_CRM_meta_notify_stop_uname
#	OCF_RESKEY_CRM_meta_notify_type
#

#######################################################################
# Initialization:

# Resource-agents have moved their ocf-shellfuncs file around.
# There are supposed to be symlinks or wrapper files in the old location,
# pointing to the new one, but people seem to get it wrong all the time.
# Try several locations.

: OCF_ROOT          = ${OCF_ROOT:=/usr/lib/ocf}
: OCF_FUNCTIONS_DIR = ${OCF_FUNCTIONS_DIR:=${OCF_ROOT}/lib/heartbeat}

for s in \
	"${OCF_FUNCTIONS_DIR}/ocf-shellfuncs" \
	"${OCF_FUNCTIONS_DIR}/.ocf-shellfuncs" \
	"${OCF_ROOT}/lib/heartbeat/ocf-shellfuncs" \
	"${OCF_ROOT}/resource.d/heartbeat/.ocf-shellfuncs" ;
do
	test -e "$s" || continue
	. "$s"
	set_logtag
	break
done

: __OFC_ACTION=${__OCF_ACTION:=$1} : $1

if ! command -v ocf_is_true &> /dev/null ; then
	# Avoid error messages from init_optional_params below
	# if called without ocf-shellfuncs, for example when trying
	# to "just get the meta-data" for generating a man page.
	ocf_is_true() { 
		case "$1" in
		yes|true|1|YES|TRUE|ja|on|ON) true ;;
		*)	false ;;
		esac
	}
fi

# Defaults
OCF_RESKEY_drbdconf_default="/etc/drbd.conf"
OCF_RESKEY_unfence_extra_args_default="--quiet --flock-required --flock-timeout 0 --unfence-only-if-owner-match"
OCF_RESKEY_require_drbd_module_version_ge_default=8.0.0
OCF_RESKEY_require_drbd_module_version_lt_default=10.0.0

init_optional_params()
{
	remove_master_score_if_peer_primary=false
	remove_master_score_if_peer_primary_only_if_unexpected=false
	if ocf_is_true ${OCF_RESKEY_remove_master_score_if_peer_primary:=false}; then
		remove_master_score_if_peer_primary=true
	elif [[ ${OCF_RESKEY_remove_master_score_if_peer_primary} = "unexpected" ]]; then
		remove_master_score_if_peer_primary=true
		remove_master_score_if_peer_primary_only_if_unexpected=true
	fi

	if ocf_is_true ${OCF_RESKEY_fail_promote_early_if_peer_primary:=false}; then
		fail_promote_early_if_peer_primary=true
	else
		fail_promote_early_if_peer_primary=false
	fi

	if ocf_is_true ${OCF_RESKEY_unfence_if_all_uptodate:=false}; then
		unfence_if_all_uptodate=true
	else
		unfence_if_all_uptodate=false
	fi

	if ocf_is_true ${OCF_RESKEY_connect_only_after_promote:=false}; then
		connect_only_after_promote=true
	else
		connect_only_after_promote=false
	fi

	# if explicitly set to the empty string,
	# or any non-digit, we don't wait at all
	wfc_timeouts=""
	OCF_RESKEY_wfc_timeout_default="5"
	: OCF_RESKEY_wfc_timeout = "${OCF_RESKEY_wfc_timeout:=$OCF_RESKEY_wfc_timeout_default}"
	local t=$OCF_RESKEY_wfc_timeout
	if [[ -n $t ]] ; then
	    if [[ $t != *[!0-9]* ]] ; then
		wfc_timeouts="--wfc-timeout=$t --degr-wfc-timeout=$t --outdated-wfc-timeout=$t"
	    else
		ocf_log debug "non-digits in wfc_timeout ($t), going to not wait at all"
	    fi
	fi
}


# The passed in OCF_CRM_meta_notify_* environment
# is not reliably with pacemaker up to at least
# 1.0.10 and 1.1.4. It should be fixed later.
# Until that is fixed, the "self-outdating feature" would base its actions on
# wrong information, and possibly not outdate when it should, or, even worse,
# outdate the last remaining valid copy.
# Disable.
OCF_RESKEY_stop_outdates_secondary_default="false"

  OCF_RESKEY_adjust_master_score_default="0 10 1000 10000"
#   ignored  |  Consistent    |  Unknown -'  |   |     |
#   ignored  |  NOT UpToDate  |  UpToDate ---'   |     |
# Secondary  |  UpToDate      |  unknown --------'     |
#   ignored  |  UpToDate      |  known   --------------+
#   Primary  |  UpToDate      |  ignored --------------'

: ${OCF_RESKEY_drbdconf:=${OCF_RESKEY_drbdconf_default}}
: ${OCF_RESKEY_stop_outdates_secondary:=${OCF_RESKEY_stop_outdates_secondary_default}}
: ${OCF_RESKEY_adjust_master_score:=${OCF_RESKEY_adjust_master_score_default}}

: ${OCF_RESKEY_require_drbd_module_version_ge:=$OCF_RESKEY_require_drbd_module_version_ge_default}
: ${OCF_RESKEY_require_drbd_module_version_lt:=$OCF_RESKEY_require_drbd_module_version_lt_default}

# Defaults according to "Configuration 1.0 Explained",
# "Multi-state resource configuration options"
: ${OCF_RESKEY_CRM_meta_clone_node_max=1}
: ${OCF_RESKEY_CRM_meta_master_max=1}
: ${OCF_RESKEY_CRM_meta_master_node_max=1}
#######################################################################
# for debugging this RA
DEBUG_LOG_DIR=/tmp/drbd.ocf.ra.debug
DEBUG_LOG=$DEBUG_LOG_DIR/log
USE_DEBUG_LOG=false
ls_stat_is_dir_0700_root() {
	set -- $(command ls -ldn "$1" 2>/dev/null);
	case "$1/$3" in
	drwx?-??-?/0|\
	drwx?-??-?./0)	true ;;
	*)	false ;;
	esac
}
# try to avoid symlink vuln.
if ls_stat_is_dir_0700_root $DEBUG_LOG_DIR &&
   [[ -w "$DEBUG_LOG" && ! -L "$DEBUG_LOG" ]]
then
	USE_DEBUG_LOG=true
	# PS4='+[`date +%F\ %T.%3N`] '
	PS4='+$(date +"%F_%T.%3N"): $HA_LOGTAG $__OCF_ACTION: ${FUNCNAME[0]:+${FUNCNAME[0]}:}${LINENO}: '
	exec 8>>"$DEBUG_LOG"
	(
	printf "%s " "$(date "+%F_%T.%3N: $HA_LOGTAG ")"
	bash -c 'for v in ${!OCF_*}; do printf "%s=%q " "$v" "${!v}"; done;'
	printf " %q" "$0" "$@"
	printf "\n"
	) >&8
else
	exec 8>/dev/null
fi
# end of debugging aid
#######################################################################

meta_data() {
	cat <<END
<?xml version="1.0"?>
<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
<!-- version attribute is version of this resource agent -->
<resource-agent name="drbd" version="LINBIT 1.4">

<!-- Version number of the standard this agent complies with -->
<version>1.0</version>

<longdesc lang="en">
This resource agent manages a DRBD resource as a master/slave resource.
DRBD is a shared-nothing replicated storage device.

NOTE:
To avoid data-divergence, you should enable either
DRBD "quorum" and "on-no-quorum io-error" (recommended),
or configure proper fencing policies in both DRBD
*and* Pacemaker (fencing resource-and-stonith).
This cannot be done from this resource agent alone.

See the DRBD User's Guide for more information.
https://docs.linbit.com/
</longdesc>

<shortdesc lang="en">Manages a DRBD device as a Master/Slave resource</shortdesc>

<parameters>
<parameter name="drbd_resource" unique="1" required="1">
<longdesc lang="en">
The name of the drbd resource from the drbd.conf file.
</longdesc>
<shortdesc lang="en">drbd resource name</shortdesc>
<content type="string"/>
</parameter>

<parameter name="drbdconf">
<longdesc lang="en">
Full path to the drbd.conf file.
</longdesc>
<shortdesc lang="en">Path to drbd.conf</shortdesc>
<content type="string" default="${OCF_RESKEY_drbdconf_default}"/>
</parameter>

<parameter name="adjust_master_score">
<longdesc lang="en">
Space separated list of four master score adjustments for different scenarios:
 - only access to 'consistent' data
 - only remote access to 'uptodate' data
 - currently Secondary, local access to 'uptodate' data, but remote is unknown
 - local access to 'uptodate' data, and currently Primary or remote is known

Numeric values are expected to be non-decreasing.

The first value is 0 by default to prevent pacemaker from trying to promote
while it is unclear whether the data is really the most recent copy.
(DRBD knows it is "consistent", but is unsure about "uptodate"ness).
Please configure proper fencing methods both in DRBD
(fencing resource-and-stonith; appropriate (un)fence-peer handlers)
AND in Pacemaker to make this work reliably.

Advanced use: Adjust the other values to better fit into complex
dependency score calculations.

Intentionally diskless nodes ("Diskless Clients") with access to good data via
some (or all) their peers will use the 3rd or 4th value (minus one) when they
are (Secondary, not all peers up-to-date) or (ALL peers are up-to-date, or they
are Primary themselves). This may need to change if this should become a
frequent use case.

Special considerations:

If a Secondary DRBD is connected to a peer in Primary role, but Pacemaker does
not know about any Primary (using crm_resource --locate), we conclude that
there likely is a cluster-split-brain, and may try to "help" Pacemaker by removing
the master-score.  Also see "remove_master_score_if_peer_primary".
</longdesc>
<shortdesc lang="en">master score adjustments</shortdesc>
<content type="string" default="${OCF_RESKEY_adjust_master_score_default}"/>
</parameter>

<parameter name="stop_outdates_secondary">
<longdesc lang="en">
Recommended setting: leave at default (disabled).

Note that this feature depends on the passed in information in
OCF_RESKEY_CRM_meta_notify_master_uname to be correct, which unfortunately is
not reliable for pacemaker versions up to at least 1.0.10 / 1.1.4.

If a Secondary is stopped (unconfigured), it may be marked as outdated in the
drbd meta data, if we know there is still a Primary running in the cluster.
Note that this does not affect fencing policies set in drbd config,
but is an additional safety feature of this resource agent only.
You can enable this behaviour by setting the parameter to true.

If this feature seems to not do what you expect, make sure you have defined
fencing policies in the drbd configuration as well.
</longdesc>
<shortdesc lang="en">outdate a secondary on stop</shortdesc>
<content type="boolean" default="${OCF_RESKEY_stop_outdates_secondary_default}"/>
</parameter>

<parameter name="ignore_missing_notifications">
<longdesc lang="en">
Some setups do not benefit from notifications.
Allow to disable notifications without patching this resource agent.
</longdesc>
<shortdesc lang="en">ignore missing notify=true</shortdesc>
<content type="boolean" default="false"/>
</parameter>

<parameter name="wfc_timeout">
<longdesc lang="en">
Unless set to the empty string or any non-digits, wait (at most)
this many seconds for the connection(s) to be established
after bringing them up during "start".
</longdesc>
<shortdesc lang="en">wait for connections with timeout</shortdesc>
<content type="integer" default="$OCF_RESKEY_wfc_timeout_default"/>
</parameter>

<parameter name="remove_master_score_if_peer_primary">
<longdesc lang="en">
See also "adjust_master_score" and "fail_promote_early_if_peer_primary".

To prevent a potentially failed promotion attempt in case of cluster split-brain
(Pacemaker communication loss) while DRBD is still connected to a Primary,
you can request to remove any master score while DRBD is connected to a Primary
(and that Primary peer looks like it has all disks up-to-date).

This may delay legitimate failovers after Primary crash by up to some TCP timeout
(until DRBD realizes that the Primary is gone) plus one monitoring interval.

This parameter is interpreted almost as an "ocf boolean",
with the exception of a literal "unexpected", that is:
- (yes|true|1) [actually, according to the OCF spec, also
  (YES|TRUE|True|ja|ON), but please don't go there]: is "true":
  remove (or never assign) master scores, if DRBD appears to see a (healthy) Primary
- "unexpected":
  assign master scores as described under "adjust_master_score",
  while removing it if DRBD appears to see a (healthy) Primary
  that Pacemaker does not know about (as determined by crm_resource --locate).
- everything else is "false":
  ignore the peer role while assigning master scores.
</longdesc>
<shortdesc lang="en">remove master score if we are connected to a primary peer</shortdesc>
<content type="string" default="false"/>
</parameter>

<parameter name="fail_promote_early_if_peer_primary">
<longdesc lang="en">
See also "adjust_master_score" and "remove_master_score_if_peer_primary".

To avoid a useless retry loop during promotion attempts in case of cluster
split-brain (Pacemaker communication loss) while DRBD is still connected to a
Primary, you can chose to give up after the first try if this situation is detected.

If a Primary "vanishes", TCP may not immediately detect this, and an idle DRBD
may take some time until it does in-DRBD-protocol "pings". Pacemaker may well
detect Primary loss earlier than DRBD, and try to promote while DRBD thinks it
can still see a Primary. Which means, in general, trying to promote at least
once is necessary, as that implies an in-DRBD-protocol "peer alive" check.

But if that does not succeed, re-trying until we hit the operation timeout may
not be desired, so you can disable it.
</longdesc>
<shortdesc lang="en">try promote only once when connected to a primary peer</shortdesc>
<content type="boolean" default="false"/>
</parameter>

<parameter name="unfence_if_all_uptodate">
<longdesc lang="en">
If all volumes of this resource report to be UpToDate,
call an unfence script hook, just in case some stale fencing constraint
or similar is still around.
<![CDATA[
- With DRBD utils version <= 8.9.4, this is hardcoded to
  /usr/lib/drbd/crm-unfence-peer.sh -r \$DRBD_RESOURCE
- With DRBD utils version >= 8.9.5, this is dispatched to
   \$DRBDADM unfence-peer \$DRBD_RESOURCE

In any case, the hook itself is responsible to fetch
\$OCF_RESKEY_unfence_extra_args from its environment.
]]>
</longdesc>
<shortdesc lang="en">call unfence hook once fully UpToDate</shortdesc>
<content type="boolean" default="false"/>
</parameter>

<parameter name="unfence_extra_args">
<longdesc lang="en">
This may be used to pass extra hints to the unfence hook.
See description of unfence_if_all_uptodate.
</longdesc>
<shortdesc lang="en">Extra arguments for the unfence hook.</shortdesc>
<content type="boolean" default="$OCF_RESKEY_unfence_extra_args_default"/>
</parameter>

<parameter name="require_drbd_module_version_ge">
<longdesc lang="en">Use this you want to force failure of this resource agent
if the detected DRBD kernel (module) driver version is lower than a required minimum.

Example: use require_drbd_module_version_ge=9.0.16 to fail unless DRBD
module version &gt;= 9.0.16 is available (effectively requires DRBD 9).

The intention of this is to give a more useful failure message
after accidentally downgrading the DRBD version
by installing/upgrading a new kernel.

Note: "ge", "greater-or-equal", inclusive.  Required format: x.y.z

Set empty to skip this check.
</longdesc>
<shortdesc lang="en"></shortdesc>
<content type="string" default="$OCF_RESKEY_require_drbd_module_version_ge_default"/>
</parameter>

<parameter name="require_drbd_module_version_lt">
<longdesc lang="en">Use this you want to force failure of this resource agent
if the detected DRBD kernel (module) driver version is higher than a required maximum.

Example: use require_drbd_module_version_lt=9.0.0 to fail unless DRBD
module version &lt; 9.0 is available (effectively requires DRBD 8.4).

Note: "lt", "less-than", exclusive.  Required format: x.y.z

Set empty to skip this check.
</longdesc>
<content type="string" default="$OCF_RESKEY_require_drbd_module_version_lt_default"/>
</parameter>

<parameter name="connect_only_after_promote">
<longdesc lang="en">
This may be useful for "stacked" setups without proper fencing on the lower
layer (which we obviously do not recommend), to avoid some of the ugly side
effects that may arise after resolving a split-brain on the lower layer.

Keep this DRBD instance disconnected until it is promoted.
After promotion we issue an additional "adjust",
which is supposed to initiate the connection attempts.

This causes a new data generation identifier ("current uuid")
to be generated after the failover of a "healthy" DRBD.
</longdesc>
<content type="boolean" default="false"/>
</parameter>
</parameters>

<actions>
<action name="start"   timeout="240" />
<action name="reload"  timeout="30" />
<action name="promote"	 timeout="90" />
<action name="demote"	timeout="90" />
<action name="notify"	timeout="90" />
<action name="stop"    timeout="100" />
<action name="monitor"  timeout="20" interval="20" role="Slave" />
<action name="monitor"  timeout="20" interval="10" role="Master" />
<action name="meta-data"  timeout="5" />
<action name="validate-all" />
</actions>
</resource-agent>
END
}

# either this pacemaker/lrmd combo does not know about ocf-exit-reason,
# or this resource-agents version already does have ocf_exit_reason(),
# or we define a short version of it here.
if ! command -v ocf_exit_reason &> /dev/null ; then
	if [[ -z $OCF_EXIT_REASON_PREFIX ]]; then
		ocf_exit_reason() { ocf_log err "$*"; }
	else
		ocf_exit_reason() {
			local fmt=$1 ; shift;
			local msg=$(printf "$fmt" "$@")
			>&2 printf "%s%s\n" "$OCF_EXIT_REASON_PREFIX" "$msg";
		}
	fi
fi

__drbd_exit_reason_set=""
drbd_ocf_exit_reason()
{
	__drbd_exit_reason_set=1
	ocf_exit_reason "$@"
}

if $USE_DEBUG_LOG ; then
	dump_stack() { local i n=${#BASH_SOURCE[@]}; for (( i=1; i <= n; i+=1 )); do caller $i ; done; }
	exit() {
		local ex=${1:-$?}
		if [[ -z $__drbd_exit_reason_set ]] ; then
			set -- $(dump_stack)
			drbd_ocf_exit_reason "exit %s at %s\n" "$ex" "$*"
		fi
		command exit "$ex"
	}
fi

do_cmd_success_log_level=""
do_cmd() {
	# Run a command, return its exit code, capture any output, and log
	# everything if appropriate.
	local cmd="$*" cmd_out cmd_err ret=125
	local success_log_level=${do_cmd_success_log_level:-debug}
	local failure_log_level=${do_cmd_failure_log_level:-err}
	ocf_log debug "$DRBD_RESOURCE: Calling $cmd"

	# capture stdout, stderr, and exit code
	eval	"$(exec 3>&1;
			printf "cmd_err=%q\n" \
				"$( exec 2>&1 1>&3 3>&-; \
					out=$( "$@" ); \
					ex=$?; \
				printf "cmd_out=%q\nret=%q\n" "$out" "$ex" )"
		)"

	if [ $ret != 0 ]; then
		ocf_log $failure_log_level "$DRBD_RESOURCE: Called $cmd"
		ocf_log $failure_log_level "$DRBD_RESOURCE: Exit code $ret"
		ocf_log $failure_log_level "$DRBD_RESOURCE: Command output: $cmd_out"
		ocf_log $failure_log_level "$DRBD_RESOURCE: Command stderr: $cmd_err"
	else
		[[ $do_cmd_success_log_level ]] &&
		ocf_log $success_log_level "$DRBD_RESOURCE: Called $cmd"
		ocf_log $success_log_level "$DRBD_RESOURCE: Exit code $ret"
		ocf_log $success_log_level "$DRBD_RESOURCE: Command output: $cmd_out"
		ocf_log $success_log_level "$DRBD_RESOURCE: Command stderr: $cmd_err"
	fi

	[ -n "$cmd_out" ] && echo "$cmd_out"

	return $ret
}

do_drbdadm() {
	local ret
	# Run drbdadm with appropriate command line options, and capture
	# its output.
	# $DRBDADM is defined during drbd_validate as "drbdadm" plus
	# appropriate command line options
	do_cmd $DRBDADM "$@"
	ret=$?

	# having the version mismatch warning once per RA invokation
	# should be enough.
	export DRBD_DONT_WARN_ON_VERSION_MISMATCH=

	return $ret
}

# cached value
unset current_master_score
get_current_master_score()
{
	# only call crm_master once
	[[ ${current_master_score+set} ]] ||
	current_master_score=$(crm_master -q -l reboot -G 2>/dev/null)
	# return value of this function:
	# true if master_score is present
	# false if master_score is not present
	[[ $current_master_score ]]
}

set_master_score() {
	# Use quiet mode (-Q) to quench logging. Actual score updates
	# will get logged by attrd anyway
	if [[ $1 -le 0 ]]; then
		remove_master_score
	else
		do_cmd ${HA_SBIN_DIR}/crm_master -Q -l reboot -v $1 &&
		current_master_score=$1
	fi
}

remove_master_score() {
	do_cmd ${HA_SBIN_DIR}/crm_master -l reboot -D
	current_master_score=""
}

source_drbd_shellfuncs()
{
	local dir=.
	[[ $0 = */* ]] && dir=${0%/*}
	source "$dir/drbd.shellfuncs.sh"
}
source_drbd_shellfuncs

call_unfence()
{
	# It is easier to just call the crm-unfence-peer script, without
	# duplicating part of its code here.  Also adding cibadmin calls to
	# first check if such constraint does even exist would needlessly add
	# even more load on the cib.

	export OCF_RESKEY_unfence_extra_args
	: ${OCF_RESKEY_unfence_extra_args:=$OCF_RESKEY_unfence_extra_args_default}

	$DRBDADM unfence-peer $DRBD_RESOURCE
	: "unfence exit code: $?"
}

# This is not the only fencing mechanism.
# But in addition to the drbd "fence-peer" handler, which should be configured,
# and is expected to place some appropriate constraints, this is used to
# actually store the Outdated information in DRBD on-disk meta data.
#
# called after stop, and from post notification events.
maybe_outdate_self()
{
	# if you claim your right to go online with stale data,
	# there you are.
	ocf_is_true $OCF_RESKEY_stop_outdates_secondary || return 1

	local host stop_uname
	# We ignore $OCF_RESKEY_CRM_meta_notify_promote_uname here
	# because: if demote and promote for a _stacked_ resource
	# (or a "floating" one, where DRBD sits on top of some SAN)
	# happen in the same transition, demote will see the promote
	# hostname here, and voluntarily outdate itself. Which would
	# result in promote failure, as it is using the same meta
	# data, which would then be outdated.
	# If that is not sufficient for you, you probably need to
	# configure fencing policies in the drbd configuration.
	host=$(printf "%s\n" $OCF_RESKEY_CRM_meta_notify_master_uname |
		grep -vix -m1 -e "$HOSTNAME" )
	if [[ -z $host ]] ; then
		# no current master host found, do not outdate myself
		return 1
	fi
	for stop_uname in $OCF_RESKEY_CRM_meta_notify_stop_uname; do
		[[ $host == "$stop_uname" ]] || continue
		# post notification for stop on that host.
		# hrmpf. crm passed in stale master_uname :(
		# ignore
		return 1
	done

	# e.g. post/promote of some other peer.
	# Should not happen, fencing constraints should take care of that.
	# But in case it does, scream out loud.
	if $status_primary; then
		# I am Primary.
		# The other one is Primary (according to OCF_RESKEY_CRM_meta_notify_master_uname).
		# But we cannot talk to each other :( (otherwise this function was not called)
		# One of us has to die.
		# Which one, however, is not ours to decide.

		ocf_log crit "resource internal SPLIT BRAIN: both $HOSTNAME and $host are Primary for $DRBD_RESOURCE, but the replication link is down!"
		return 1
	fi

	# OK, I am not Primary, but there is an other node Primary
	# Outdate myself
	ocf_log notice "outdating $DRBD_RESOURCE: according to OCF_RESKEY_CRM_meta_notify_master_uname, '$host' is still master"
	do_drbdadm outdate $DRBD_RESOURCE

	# on some pacemaker versions, -INFINITY may cause resource instance stop/start.
	# But in this case that is ok, it may even clear the replication link
	# problem.
	set_master_score -INFINITY

	return 0
}

unexpected_primary_rejects_promote()
{
	crm_resource_locate_master

	: "single master config?"
	[[ $OCF_RESKEY_CRM_meta_master_max = 1 ]] || return 1
	: "not primary myself?" # because, if I am, I can no longer reject this...
	! $status_primary  || return 1
	: "some peer is primary?"
	$status_some_peer_primary || return 1
	: "but pacemaker does not know about a Primary?"
	[[ -z $DRBD_PRIMARY_PEER_according_to_pcmk ]] || return 1
	return 0
}

removed_master_score_because_peer_is_primary()
{
	: "remove master score if peer primary?"; $remove_master_score_if_peer_primary || return 1
	: "primary myself?"; $status_primary && return 1
	: "some peer primary?"; $status_some_peer_primary || return 1

	if : "SOME peer ALL up-to-date?"; $status_some_peer_all_up_to_date ; then
		# FIXME this should check if that same primary peer has all healthy
		# disks, not if "some" peer has all healthy disks.
		# But I think this option only makes sense in two node setups,
		# so "some" peer becomes "the" peer anyways.
		: "Okay, peer apparently has healthy disks"
	else
		: "peer is primary, but does not look healthy, use regular master score adjustment"
		return 1
	fi

	if : "remove only if unexpected"; $remove_master_score_if_peer_primary_only_if_unexpected; then
		if  unexpected_primary_rejects_promote; then
			## Do not log "err", if this may be intentional (independent clusters),
			## and this pacemaker was told to not even try to promote
			# case $OCF_RESKEY_CRM_meta_target_role in
			# [Mm]aster|[Ss]tarted|"")
			#
			## But what if they control it via booth and tickets?
			## Or other constraints?

			# do we even have a master score, currently?
			# logging this with every monitoring interval may be too noisy.
			if get_current_master_score ; then
				ocf_log info "I am connected to a Primary that Pacemaker does not know about! Removing master score."
				remove_master_score
			fi
			return 0
		fi
		: "not unexpected, no special treatment"
		return 1
	fi

	: "connected to (apparently healthy) Primary peer, do not allow promote"
	remove_master_score
	return 0
}

drbd_update_master_score() {
	set -- $OCF_RESKEY_adjust_master_score
	local only_consistent=$1 only_remote=$2 local_ok=$3 as_good_as_it_gets=$4

	# NOTE
	# there may be constraint scores from rules on role=Master,
	# that in some ways can add to the node attribute based master score we
	# specify below. If you think you want to add personal preferences,
	# in case the scores given by this RA do not suffice, this is the
	# value space you can work with:
	# -INFINITY: Do not promote. Really. Won't work anyways.
		# Too bad, at least with current (Oktober 2009) Pacemaker,
		# negative master scores cause instance stop; restart cycle :(
	# missing, zero: Do not promote.
	#        I think my data is not good enough.
	#        Though, of course, you may try, and it might even work.
	#     5: please, do not promote, unless this is your only option.
	#    10: promotion is probably a bad idea, our local data is no good,
	#        you'd probably run into severe performance problems, and risk
	#        application crashes or blocking IO in case you lose the
	#        replication connection.
	#  1000: Ok to be promoted, we have good data locally (though we don't
	#        know about the peer, so possibly it has even better data?).
	#        You sould use the crm-fence-peer.sh handler or similar
	#        mechanism to avoid data divergence.
	# 10000: Please promote me/keep me Primary.
	#        I'm confident that my data is as good as it gets.
	#
	# TODO: separately configure the master score for diskless clients
	# For now: if it is "intentionally" diskless, and has access to
	# remote UpToDate, consider it slighly worse than "local_ok".
	#
	if : "have quorum?"; $status_have_quorum ; then
		: "quorate, evaluate status in more detail below"
	else
		: "NOT quorate, should not be master."
		remove_master_score
		return
	fi

	# I'd like to remove the master score, if we find ourselves be
	# connected to an "unexpected" primary:
	# # unexpected_primary_rejects_promote && remove_master_score
	#
	# BUT.
	# Given bad timing, DRBD may still think it is connected
	# but pacemaker already knows the peer is dead.
	# If now a "monitor" squeezes in between "peer known dead"
	# and the soon to be expected "promote", while DRBD still thinks it was
	# connected to a Primary (likely optimistically waiting for generously
	# configured internal timeouts), if we remove the master score from
	# this monitor action, we delay the failover for up to
	# time-for-DRBD-internals-to-declare-peer-dead
	# plus one extra monitor interval.
	#
	# So at least don't do that by default, check for
	# remove_master_score_if_peer_primary and
	# remove_master_score_if_peer_primary_only_if_unexpected
	#
	if removed_master_score_because_peer_is_primary ; then
		return
	fi

	if	: "diskless client?"; $status_diskless_client ; then
		if	: "primary and access to good data?" ;
			$status_primary && $status_some_peer_all_up_to_date; then

			set_master_score $(( as_good_as_it_gets -1 ))
		elif	: "ALL peer-disks up-to-date?"; $status_pdsk_all_up_to_date; then
			set_master_score $(( as_good_as_it_gets -1 ))
		elif	: "SOME peer-disks up-to-date?"; $status_some_peer_all_up_to_date ; then
			set_master_score $(( local_ok - 1 ))
		else	: "Diskless client, without access to good data :("
			remove_master_score
		fi

	elif	: "all disks up-to-date?"; $status_disk_all_up_to_date ; then

		if	: "primary?" ; $status_primary ; then

			# I am Primary, all local disks are UpToDate
			set_master_score $as_good_as_it_gets

			if  : "all peer-disks up-to-date?"; $status_pdsk_all_up_to_date; then
				# I am Primary, all local disks are UpToDate,
				# AND all peer disks are UpToDate
				: == DEBUG == unfence_if_all_uptodate=$unfence_if_all_uptodate
				$unfence_if_all_uptodate && call_unfence

			# else: not so sure about the peer's disks
			fi
		else : "Not primary."
			if : "any peer-disk unknown?"; $status_pdsk_any_unknown ; then
				# all local disks are UpToDate,
				# but I'm not Primary,
				# and I'm not sure about some peer's disk state(s).
				# We may need to outdate ourselves?
				# But if we outdate in a MONITOR, and are disconnected
				# secondary because of a hard primary crash, before CRM noticed
				# that there is no more master, we'd make us utterly useless!
				# Trust that the primary will also notice the disconnect,
				# and will place an appropriate fencing constraint via
				# its fence-peer handler callback.
				set_master_score  $local_ok
			else : "all peer disk states known."
				# We know something about our peer, which means that either the
				# replication link is established, or it was not even
				# consistent last time we talked to each other.
				# Also all our local disks are UpToDate, which means even if we are
				# currently synchronizing, we do so as SyncSource.
				set_master_score $as_good_as_it_gets
			fi
		fi

	elif	: "some peer with all peer-disks up-to-date?" ; $status_some_peer_all_up_to_date ; then

		# At least one of our local disks is not up to date.
		# But at least one of our peers is ALL OK.
		# We can expect to have access to useful
		# data, but with possibly degraded performance,
		# (some) reads need to fetch from the peer.
		set_master_score $only_remote
	
	elif	: "in transitional state?"; $status_disk_transitional_state ; then
		# some transitional state.
		# just don't do anything
		: "ignore"

	elif	: "all disks consistent?" ; $status_disk_all_consistent ; then
		# All local disks seem to be Consistent.
		# They _may_ be up to date, or not.
		# We hope that fencing mechanisms have put constraints in
		# place, so we won't be promoted with stale data.
		# But in case this was a cluster crash,
		# at least allow _someone_ to be promoted.
		set_master_score $only_consistent

	else # not $status_disk_all_consistent and not $status_disk_transitional_state

		# ALWAYS put the cluster in MAINTENANCE MODE
		# if you add a volume to a live replication group,
		# because the new volume will typically come up as Inconsistent
		# the first time, which would cause a monitor to revoke the
		# master score!
		#
		# At least some of our local disks are not really useable.
		# Our peer is not all good either (or some previous case block
		# would have matched).  We have no access to useful data.
		# DRBD would refuse to be promoted, anyways.
		#
		# set_master_score -INFINITY
		# Too bad, at least with current (Oktober 2009) Pacemaker,
		# negative master scores cause instance stop; restart cycle :(
		# Hope that this will suffice.
		remove_master_score
	fi
}

is_drbd_enabled() {
	test -f /proc/drbd
}

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

drbd_usage() {
	echo "\
usage: $0 {start|stop|monitor|validate-all|promote|demote|notify|meta-data}

Expects to have a fully populated OCF RA-compliant environment set."
}

drbd_status() {
	local rc
	local dev
	rc=$OCF_NOT_RUNNING

	is_drbd_enabled || return $rc

	# Not running, if no block devices exist.
	#
	# FIXME what if some do, and some do not exist?
	# Adding/removing volumes to/from existing resources should only be
	# done with maintenance-mode enabled.
	# If someone does manually kill/remove only some of the volumes,
	# we tolerate that here.
	for dev in "${DRBD_DEVICES[@]}" ""; do
		test -b $dev && break
	done
	[[ $dev ]] || return $rc

	# ok, module is loaded, block device nodes exist.
	# lets see the status
	drbd_set_status_variables
	if $status_unconfigured ; then
		rc=$OCF_NOT_RUNNING
	elif $status_primary ; then
		rc=$OCF_RUNNING_MASTER
	else
		rc=$OCF_SUCCESS
	fi

	return $rc
}

# I'm sorry, but there is no $OCF_DEGRADED_MASTER or similar yet.
drbd_monitor() {
	local status
	drbd_status
	status=$?

	if [[ $status = $OCF_NOT_RUNNING ]] && ocf_is_probe ; then
		# see also linux-ha mailing list archives,
		# From: Andrew Beekhof
		# Subject: Re: pacemaker+drbd promotion delay
		# Date: 2012-04-13 01:47:37 GMT
		# e.g.: http://thread.gmane.org/gmane.linux.highavailability.user/37089/focus=37163
		# ---
		: "do nothing" ;
	else
		drbd_update_master_score
	fi

	case $status in
	(0) : "OCF_SUCCESS aka 'running/slave'" ;;
	(1) : "OCF_ERR_GENERIC" ;;
	(2) : "OCF_ERR_ARGS" ;;
	(3) : "OCF_ERR_UNIMPLEMENTED" ;;
	(4) : "OCF_ERR_PERM" ;;
	(5) : "OCF_ERR_INSTALLED" ;;
	(6) : "OCF_ERR_CONFIGURED" ;;
	(7) : "OCF_NOT_RUNNING" ;;
	(8) : "OCF_RUNNING_MASTER" ;;
	(9) : "OCF_FAILED_MASTER" ;;
	(190): "PCMK_OCF_DEGRADED" ;;
	(191): "PCMK_OCF_DEGRADED_MASTER" ;;
	(*) : " WTF? $status " ;;
	esac

	return $status
}

called_crm_resource_locate=false
crm_resource_locate_master()
{
	$called_crm_resource_locate && return
	called_crm_resource_locate=true
	DRBD_PRIMARY_PEER_according_to_pcmk=$(
		crm_resource --resource "$OCF_RESOURCE_INSTANCE" --locate 2>/dev/null |
		sed -ne 's/^.*is running on: \([^ ]*\) Master.*$/\1/p' |
		grep -vix -m1 -e "$HOSTNAME")
}

figure_out_drbd_peer_uname()
{
	# depending on whether or not the peer is currently
	# configured, slave, master, or about to be started,
	# it may be mentioned in various variables (or not at all)
	local x
	# intentionally not cared for stop_uname
	x=$(printf "%s\n" \
		$OCF_RESKEY_CRM_meta_notify_start_uname \
		$OCF_RESKEY_CRM_meta_notify_promote_uname \
		$OCF_RESKEY_CRM_meta_notify_master_uname \
		$OCF_RESKEY_CRM_meta_notify_slave_uname \
		$OCF_RESKEY_CRM_meta_notify_demote_uname |
		grep -vix -m1 -e "$HOSTNAME" )
	# up to utils 9.20, the --peer option was ignored with drbd 9
	# starting with 9.21, it is unknown,
	# and would lead to "drbdadm: unrecognized option ..." error.
	# Only set this for DRBD <= 8.4
	$DRBD_IS_v9 || DRBD_TO_PEER=${x:+ --peer $x}
	crm_resource_locate_master
}

my_udevsettle()
{
	# We have an endless loop below.
	# If caller wants to enforce some timeout,
	# allow to report a useful error message by catching TERM here.
	trap 'drbd_ocf_exit_reason "expected device nodes did not appear (in time)"; exit $OCF_ERR_GENERIC' TERM

	for dev in "${DRBD_DEVICES[@]}"; do
		while ! test -b $dev; do
			sleep 1;
		done
	done
	trap - TERM
	return 0
}
create_device_udev_settle()
{
	local dev
	if $DRBD_HAS_MULTI_VOLUME; then
		if do_drbdadm new-resource $DRBD_RESOURCE &&
		   do_drbdadm new-minor $DRBD_RESOURCE; then
			my_udevsettle
		else
			return 1
		fi
	elif do_drbdadm syncer $DRBD_RESOURCE ; then
		my_udevsettle
	else
		return 1
	fi
}

do_attach()
{
	if $DRBD_IS_v9; then
		# for diskless clients, this may be a no-op.
		do_drbdadm adjust --skip-net $DRBD_RESOURCE
	else
		do_drbdadm attach $DRBD_RESOURCE
	fi
}

wait_connect_84()
{
	[[ -z $wfc_timeouts ]] && return
	local do_cmd_failure_log_level=info
	local dev
	for dev in "${DRBD_DEVICES[@]}"; do
		do_cmd drbdsetup wait-connect $dev $wfc_timeouts
	done
}

wait_connect_90()
{
	[[ -z $wfc_timeouts ]] && return
	local res=$1 peer=$2
	local do_cmd_failure_log_level=info
	# drbdadm wait-connect is not smart enough yet
	# work around our own tools
	# FIXME
	if [[ -n "$peer" ]]; then
		local cmd
		cmd=$(drbdadm -d connect $res:$peer | sed -e 's/ connect / wait-connect-connection /')
		if [[ -n "$cmd" ]] ; then
			do_cmd $cmd $wfc_timeouts
		else
			ocf_log warn "Please use matching node names for hostname, uname, corosync, pacemaker and DRBD! Not a valid DRBD peer: '$peer'"
		fi
	else
		do_cmd drbdsetup wait-connect-resource $res $wfc_timeouts
	fi
}

do_connect()
{
	figure_out_drbd_peer_uname
	if $DRBD_IS_v9; then
		if [[ -n $DRBD_PRIMARY_PEER_according_to_pcmk ]] ; then
			do_drbdadm adjust $DRBD_RESOURCE:$DRBD_PRIMARY_PEER_according_to_pcmk
			wait_connect_90 $DRBD_RESOURCE $DRBD_PRIMARY_PEER_according_to_pcmk
		fi
		do_drbdadm adjust $DRBD_RESOURCE
		wait_connect_90 $DRBD_RESOURCE
	else
		do_drbdadm $DRBD_TO_PEER adjust $DRBD_RESOURCE
		wait_connect_84
	fi
}

drbd_start()
{
	local rc
	local status
	local first_try=true local attach_failed=false

	rc=$OCF_ERR_GENERIC

	if ! is_drbd_enabled; then
		do_cmd modprobe -s drbd `$DRBDADM sh-mod-parms` || {
			drbd_ocf_exit_reason "Cannot load the drbd module.";
			: "$OCF_ERR_INSTALLED = OCF_ERR_INSTALLED"
			return $OCF_ERR_INSTALLED
		}
		ocf_log debug "$DRBD_RESOURCE start: Module loaded."
	fi

	# Keep trying to bring up the resource;
	# wait for the CRM to time us out if this fails
	while :; do
		drbd_status
		status=$?
		case "$status" in
		$OCF_SUCCESS)
			# With DRBD 9, once resource and minor have been
			# created, it reports as "configured Secondary"
			# (which maps to OCF_SUCCESS),
			# so we need to retry the attach here, instead of the
			# second try in the OCF_NOT_RUNNING branch below.
			if $attach_failed && ! do_attach; then
				drbd_ocf_exit_reason "failed to attach"
				exit $OCF_ERR_GENERIC
			fi

			# Just in case we have to adjust something, this is a
			# good place to do it.  Actually, we don't expect to be
			# called to "start" an already "running" resource, so
			# this is probably dead code.
			# Also, ignore the exit code of adjust, as we are
			# "running" already, anyways, right?
			rc=$OCF_SUCCESS
			$connect_only_after_promote && break
			do_connect
			drbd_set_status_variables
			break
			;;
		$OCF_NOT_RUNNING)
			# Check for offline resize. If using internal meta data,
			# we may need to move it first to its expected location.
			$first_try && do_drbdadm check-resize $DRBD_RESOURCE
			figure_out_drbd_peer_uname
			if ! create_device_udev_settle; then
				# We cannot even create the objects.
				# This is supposedly never reached, btw,
				# my_udevsettle is supposed to exit via trap on TERM.
				exit $OCF_ERR_GENERIC
			fi
			if ! do_attach; then
				# If we cannot up it, even on the second try,
				# it is unlikely to get better.  Don't wait for
				# this operation to timeout, but short circuit
				# exit with generic error.
				$first_try || {
					drbd_ocf_exit_reason "failed to attach"
					exit $OCF_ERR_GENERIC
				}
				attach_failed=true
				sleep 1
			fi
			;;
		$OCF_RUNNING_MASTER)
			ocf_log warn "$DRBD_RESOURCE already Primary, demoting."
			do_drbdadm secondary $DRBD_RESOURCE
		esac
		$first_try || sleep 1
		first_try=false
	done
	# in case someone does not configure monitor,
	# we must at least call it once after start.
	drbd_update_master_score

	return $rc
}

drbd_reload() {
	local rc
	local status

	rc=$OCF_ERR_GENERIC

	drbd_status
	status=$?

	case "$status" in
	$OCF_SUCCESS|$OCF_RUNNING_MASTER)
		# Adjust resource just in case reload was requested manually
		# Changes to resource parameters do not require this
		do_drbdadm adjust $DRBD_RESOURCE
		rc=$OCF_SUCCESS
		;;
	$OCF_NOT_RUNNING)
		:
		;;
	esac
	# Update score as adjust_master_score may be changed
	drbd_update_master_score

	return $rc
}

drbd_promote() {
	local rc
	local status
	local first_try=true
	local do_cmd_success_log_level

	rc=$OCF_ERR_GENERIC

	# Keep trying to promote the resource;
	# wait for the CRM to time us out if this fails
	while :; do
		drbd_status
		status=$?
		case "$status" in
		$OCF_SUCCESS)
			if ! $first_try && unexpected_primary_rejects_promote ; then
				if $fail_promote_early_if_peer_primary ; then
					drbd_ocf_exit_reason "Peer node already/still Primary"
					remove_master_score
					break
				else
					ocf_log info "Peer node already/still Primary, promote will likely fail or need several attempts. Retrying anyways."
					do_cmd_success_log_level=info
				fi
			fi
			do_drbdadm primary $DRBD_RESOURCE
			if [[ $? = 17 ]]; then
				# All available disks are inconsistent,
				# or I am consistent, but failed to fence the peer.
				# Cannot become primary.
				# No need to retry indefinitely.
				drbd_ocf_exit_reason "%s" "Refusing to be promoted to Primary without access to UpToDate data"
				break
			fi
			;;
		$OCF_NOT_RUNNING)
			drbd_ocf_exit_reason "%s" "cannot promote a resource that was not started"
			break
			;;
		$OCF_RUNNING_MASTER)
			rc=$OCF_SUCCESS
			if $connect_only_after_promote; then
				figure_out_drbd_peer_uname
				do_drbdadm $DRBD_TO_PEER -v adjust $DRBD_RESOURCE
			fi
			break
		esac
		$first_try || sleep 1
		first_try=false
	done

	# avoid too tight pacemaker driven "recovery" loop,
	# if promotion keeps failing for some reason
	if [[ $rc != 0 ]] && (( $SECONDS < 15 )) ; then
		delay=$(( 15 - SECONDS ))
		ocf_log warn "promotion failed; sleep $delay # to prevent tight recovery loop"
		sleep $delay
	fi
	return $rc
}

drbd_demote() {
	local rc
	local status
	local first_try=true

	rc=$OCF_ERR_GENERIC

	# Keep trying to demote the resource;
	# wait for the CRM to time us out if this fails
	while :; do
		drbd_status
		status=$?
		case "$status" in
		$OCF_SUCCESS)
			rc=$OCF_SUCCESS
			break
			;;
		$OCF_NOT_RUNNING)
			ocf_log error "Trying to promote a resource that was not started"
			break
			;;
		$OCF_RUNNING_MASTER)
			do_drbdadm secondary $DRBD_RESOURCE
		esac
		$first_try || sleep 1
		first_try=false
	done

	return $rc
}

drbd_stop() {
	local rc=$OCF_ERR_GENERIC
	local first_try=true

	# Keep trying to bring down the resource;
	# wait for the CRM to time us out if this fails
	while :; do
		drbd_status
		status=$?
		case "$status" in
		$OCF_SUCCESS)
			do_drbdadm down $DRBD_RESOURCE
			;;
		$OCF_NOT_RUNNING)
			# Just in case, down it anyways, in case it has been
			# deconfigured but not yet removed.
			# Relevant for >= 8.4.
			do_drbdadm down $DRBD_RESOURCE
			# But ignore any return codes,
			# we are not running, so stop is successfull.
			rc=$OCF_SUCCESS
			break
			;;
		$OCF_RUNNING_MASTER)
			ocf_log warn "$DRBD_RESOURCE still Primary, demoting."
			do_drbdadm secondary  $DRBD_RESOURCE
		esac
		$first_try || sleep 1
		first_try=false
	done

	# if there is some Master (Primary) still around,
	# outdate myself in drbd on-disk meta data.
	maybe_outdate_self

	# do not let old master scores laying around.
	# they may confuse crm if this node was set to standby.
	remove_master_score

	return $rc
}


drbd_notify() {
	local n_type=$OCF_RESKEY_CRM_meta_notify_type
	local n_op=$OCF_RESKEY_CRM_meta_notify_operation

	# active_* and *_resource not really interessting
	# : "== DEBUG == active  = $OCF_RESKEY_CRM_meta_notify_active_uname"
	: "== DEBUG == slave   = $OCF_RESKEY_CRM_meta_notify_slave_uname"
	: "== DEBUG == master  = $OCF_RESKEY_CRM_meta_notify_master_uname"
	: "== DEBUG == start   = $OCF_RESKEY_CRM_meta_notify_start_uname"
	: "== DEBUG == promote = $OCF_RESKEY_CRM_meta_notify_promote_uname"
	: "== DEBUG == stop    = $OCF_RESKEY_CRM_meta_notify_stop_uname"
	: "== DEBUG == demote  = $OCF_RESKEY_CRM_meta_notify_demote_uname"

	case $n_type/$n_op in
	*/start)
		# We do not get a /pre/ start notification for ourself.
		# but we get a /pre/ start notification for the other side, unless both
		# are started from the same transition graph.  If there are only two
		# peers (the "classic" two-node DRBD), this adjust is usually a no-op.
		#
		# In case of more than one _possible_ peer, we may still be StandAlone,
		# or configured for a meanwhile failed peer, and should now adjust our
		# network settings during pre-notification of start of the other node.
		#
		# We usually get /post/ notification for ourself and the peer.
		# In both cases adjust should be a no-op.
		if ! $connect_only_after_promote; then
			drbd_set_status_variables
			figure_out_drbd_peer_uname
			do_drbdadm $DRBD_TO_PEER -v adjust $DRBD_RESOURCE
		fi
		;;
	post/*)
		# After something has been done is a good time to
		# recheck our status:
		drbd_set_status_variables
		drbd_update_master_score

		if : "any unknown peer device?"; $status_pdsk_any_unknown ; then
			# Still not properly communicating.
			# Maybe someone else is primary (too)?
			maybe_outdate_self
		fi
	esac

	: "$OCF_SUCCESS = OCF_SUCCESS"
	return $OCF_SUCCESS
}

# "macro" to be able to give useful error messages
# on clone resource configuration error.
meta_expect()
{
	local what=$1 whatvar=OCF_RESKEY_CRM_meta_${1//-/_} op=$2 expect=$3
	local val=${!whatvar}
	if [[ -n $val ]]; then
		# [, not [[, or it won't work ;)
		[ "$val" $op $expect ] && return
	fi
	drbd_ocf_exit_reason "%s" "meta parameter misconfigured, expected $what $op $expect, but found ${val:-unset}."
	return $OCF_ERR_CONFIGURED
}

ls_stat_is_block_maj_147() {
	set -- $(command ls -L -l "$1" 2>/dev/null)
	[[ $1 = b* ]] && [[ $5 == 147,* ]]
}

check_crm_feature_set()
{
	set -- ${OCF_RESKEY_crm_feature_set//[!0-9]/ }
	local a=${1:-0} b=${2:-0} c=${3:-0}
	
	(( a > 3 )) ||
	(( a == 3 && b > 0 )) ||
	(( a == 3 && b == 0 && c > 0 )) ||
	ocf_log warn "You may be disappointed: This RA is intended for pacemaker 1.0 or better!"

	PCMK_OCF_DEGRADED=$OCF_SUCCESS
	PCMK_OCF_DEGRADED_MASTER=$OCF_RUNNING_MASTER

	## pacemaker since crm_feature_set 3.0.10 knows about "degraded" states.
	## But it does not work yet, because LRMD filters the exit codes...
	# if (( a > 3 )) || (( a == 3 && b > 0 )) || (( a == 3 && b == 0 && c >= 10 )); then
	# 	PCMK_OCF_DEGRADED=190
	# 	PCMK_OCF_DEGRADED_MASTER=191
	# fi
}

require_drbd_module_version()
{
	local v k op version_code
	for op in ge lt; do
		k=OCF_RESKEY_require_drbd_module_version_$op
		v=${!k}
		[[ $v ]] || continue
		if ! [[ $v =~ ^([0-9]+)\.([0-9]+)\.([0-9]+)$ ]]; then
			drbd_ocf_exit_reason "require_drbd_module_version_$op: invalid version format '$v'"
			: "$OCF_ERR_CONFIGURED = OCF_ERR_CONFIGURED"
			return $OCF_ERR_CONFIGURED
		fi
		set -- ${BASH_REMATCH[@]:1}
		version_code=$(( 10#$1 << 16 | 10#$2 << 8 | 10#$3 ))
		case $op in
			ge) (( $DRBD_KERNEL_VERSION_CODE >= $version_code )) && continue ;;
			lt) (( $DRBD_KERNEL_VERSION_CODE < $version_code )) && continue ;;
		esac
		drbd_ocf_exit_reason "required DRBD kernel (module) version -$op $v, but found: $DRBD_KERNEL_VERSION_CODE"
		: "$OCF_ERR_INSTALLED = OCF_ERR_INSTALLED"
		return $OCF_ERR_INSTALLED
	done
}

_drbd_validate_all () {
	DRBDADM="drbdadm"
	DRBDSETUP="drbdsetup"
	DRBD_HAS_MULTI_VOLUME=false
	DRBD_HAS_EVENTS2=false
	DRBD_IS_v9=false

	# these will _exit_ if they don't find the binaries
	check_binary $DRBDADM
	check_binary $DRBDSETUP
	# XXX I really take cibadmin, sed, grep, etc. for granted.

	local VERSION DRBD_KERNEL_VERSION_CODE=0 DRBDADM_VERSION_CODE=0
	if VERSION="$($DRBDADM --version 2>/dev/null)"; then
		eval "$VERSION"
	fi
	if (( $DRBD_KERNEL_VERSION_CODE == 0x0 )) ; then
		# Maybe the DRBD module was not loaded (yet).
		# I don't want to load the module here,
		# maybe this is just a probe or stop.
		# It will be loaded on "start", though.
		# Instead, look at modinfo output.
		# Newer drbdadm does this implicitly, but may reexec older
		# drbdadm versions for compatibility reasons.
		DRBD_KERNEL_VERSION_CODE=$(printf "0x%02x%02x%02x" $(
			modinfo -F version drbd |
			sed -ne 's/^\([0-9]\+\)\.\([0-9]\+\)\.\([0-9]\+\).*$/\1 \2 \3/p'))
	fi
	if (( $DRBD_KERNEL_VERSION_CODE == 0x0 )); then
		drbd_ocf_exit_reason "DRBD kernel (module) not available?"
		: "$OCF_ERR_INSTALLED = OCF_ERR_INSTALLED"
		return $OCF_ERR_INSTALLED
	fi
	require_drbd_module_version || return
	if (( $DRBD_KERNEL_VERSION_CODE >= 0x080400 )); then
		DRBD_HAS_MULTI_VOLUME=true
	fi
	if (( $DRBD_KERNEL_VERSION_CODE >= 0x090000 )) ; then
		# "drbdsetup events2" is actually usable since 0x080407.
		# But let's only use it for 9.
		DRBD_HAS_EVENTS2=true
		DRBD_IS_v9=true
	fi
	check_crm_feature_set

	if [[ $__OCF_ACTION != stop ]] ; then
		meta_expect clone-node-max = 1			|| return
		meta_expect master-node-max = 1			|| return

		# Check clone and M/S options.
		# don't try to bump this, even the DRBD 9 kernel module
		# can not reliably handle more than two primaries yet,
		# without risking data corruption.
		# Not even two primaries plus one-or-more-secondaries :-(
		meta_expect master-max -le 2			|| return
		if ! $DRBD_IS_v9 || [[ $OCF_RESKEY_CRM_meta_master_max = 2 ]]; then
			meta_expect clone-max -le 2		|| return
		fi
	fi

	# Rather than returning $OCF_ERR_CONFIGURED, we sometimes return
	# $OCF_ERR_INSTALLED here: the local config may be broken, but some
	# other node may have a valid config.

	# check drbdconf plausibility
	case "$OCF_RESKEY_drbdconf" in
	"")
		# this is actually ok. drbdadm has its own builtin defaults.
		# but as long as we assign an explicit default above,
		# this cannot happen anyways.
		: ;;
	*[!-%+./0-9:=@A-Z_a-z]*)
		# no, I do not trust the configurable cib parameters.
		drbd_ocf_exit_reason "%s" "drbdconf name must only contain [-%+./0-9:=@A-Z_a-z]"
		: "$OCF_ERR_CONFIGURED = OCF_ERR_CONFIGURED"
		return $OCF_ERR_CONFIGURED
		;;
	*)
		# Check if we can read the configuration file.
		if [ ! -r "${OCF_RESKEY_drbdconf}" ]; then
			drbd_ocf_exit_reason "Configuration file ${OCF_RESKEY_drbdconf} does not exist or is not readable!"
			: "$OCF_ERR_INSTALLED = OCF_ERR_INSTALLED"
			return $OCF_ERR_INSTALLED
		fi
		DRBDADM="$DRBDADM -c $OCF_RESKEY_drbdconf"
	esac

	# check drbd_resource plausibility
	case "$OCF_RESKEY_drbd_resource" in
	"")
		drbd_ocf_exit_reason "No resource name specified!"
		: "$OCF_ERR_CONFIGURED = OCF_ERR_CONFIGURED"
		return $OCF_ERR_CONFIGURED
		;;
	*[!-%+./0-9:=@A-Z_a-z]*)
		# no, I do not trust the configurable cib parameters.
		drbd_ocf_exit_reason "%s" "Resource name must only contain [-%+./0-9:=@A-Z_a-z]"
		: "$OCF_ERR_CONFIGURED = OCF_ERR_CONFIGURED"
		return $OCF_ERR_CONFIGURED
	esac

	# The resource should appear in the config file,
	# otherwise something's fishy
	# NOTE
	# since 8.4 has multi volume support,
	# DRBD_DEVICES will be a shell array!
	# FIXME we should double check that we explicitly restrict the set of
	# valid characters in device names...
	if DRBD_DEVICES=($($DRBDADM sh-dev $DRBD_RESOURCE 2>/dev/null)); then
		# Note: If dealing with drbd module version 9 and starting with
		# drbd-utils version 9.5.0, for explicitly listed resources
		# (as is the case here), we don't need to add "--stacked"
		# anymore, even if they are stacked.
		: # nothing to do.
	elif DRBD_DEVICES=($($DRBDADM --stacked sh-dev $DRBD_RESOURCE 2>/dev/null)); then
		# apparently a "stacked" resource. Remember for future DRBDADM calls.
		DRBDADM="$DRBDADM -S"
	else
		if [[ $__OCF_ACTION = "monitor" && $OCF_RESKEY_CRM_meta_interval = 0 ]]; then
			# ok, this was a probe. That may happen on any node,
			# to enforce configuration.
			: "$OCF_NOT_RUNNING = OCF_NOT_RUNNING"
			return $OCF_NOT_RUNNING
		else
			# hm. probably misconfigured constraint somewhere.
			# sorry. don't retry anywhere.
			drbd_ocf_exit_reason "%s" "DRBD resource ${DRBD_RESOURCE} not found in configuration file ${OCF_RESKEY_drbdconf}."
			remove_master_score
			: "$OCF_ERR_INSTALLED = OCF_ERR_INSTALLED"
			return $OCF_ERR_INSTALLED
		fi
	fi

	# check for master-max and allow-two-primaries on start|promote only,
	# so it could be stopped still, if someone re-configured while running.
	case $__OCF_ACTION:$OCF_RESKEY_CRM_meta_master_max in
	start:[!01]|promote:[!01])
		if ! $DRBDADM -d -v dump $DRBD_RESOURCE 2>/dev/null |
			grep -q -Ee '^[[:space:]]*allow-two-primaries([[:space:]]+yes)?;$'
		then
			drbd_ocf_exit_reason "%s" "master-max > 1, but DRBD resource $DRBD_RESOURCE does not allow-two-primaries."
			: "$OCF_ERR_CONFIGURED = OCF_ERR_CONFIGURED"
			return $OCF_ERR_CONFIGURED
		fi
	esac

	# detect whether notify is configured or not.
	# for probes, the meta_notify* namespace is not exported.
	case $__OCF_ACTION in
	stop|monitor|validate-all)
		:;;
	*)
		# Test if the environment variables for either the notify
		# enabled, or one of its effects, are set.
		# If both are unset, we complain.
		if ocf_is_true ${OCF_RESKEY_ignore_missing_notifications:=false} ; then
			: "ignore" ;
		elif ! ocf_is_true ${OCF_RESKEY_CRM_meta_notify} &&
		   [[ ${OCF_RESKEY_CRM_meta_notify_start_uname- NOT SET } = " NOT SET " ]]; then
			drbd_ocf_exit_reason "you really should enable notify when using this RA (or set ignore_missing_notifications=true)"
			: "$OCF_ERR_CONFIGURED = OCF_ERR_CONFIGURED"
			return $OCF_ERR_CONFIGURED
		fi
	esac

	local i j n=0 fallback=false
	for i in $OCF_RESKEY_adjust_master_score; do
		[[ $i = *[!0-9]* ]]   && fallback=true && ocf_log err "BAD adjust_master_score value $i ; falling back to default"
		[[ $j && $i -lt $j ]] && fallback=true && ocf_log err "BAD adjust_master_score value $j > $i ; falling back to default"
		j=$i
		n=$(( n+1 ))
	done
	[[ $n != 4 ]] && fallback=true && ocf_log err "Not enough adjust_master_score values ($n != 4); falling back to default"
	$fallback && OCF_RESKEY_adjust_master_score=$OCF_RESKEY_adjust_master_score_default

	# we use it in various places,
	# just make sure it contains what we expect.
	HOSTNAME=`uname -n`

	: "$OCF_SUCCESS = OCF_SUCCESS"
	return $OCF_SUCCESS
}

drbd_validate_all () {
	local ex
	_drbd_validate_all && return
	ex=$?

	if [[ $__OCF_ACTION = stop ]] ; then
		# try to avoid stop failure and subsequent node level fencing
		# just because someone "accidentally" misconfigured this
		ocf_log err "validate error '$ex' on stop! Trying 'drbdsetup down $DRBD_RESOURCE' directly to avoid fencing..."
		if drbdsetup down "$DRBD_RESOURCE" ; then
			ocf_log err "fencing of this node narrowly avoided. FIX YOUR SETUP!"
			return $OCF_SUCCESS
		else
			ocf_log err "FIX YOUR SETUP! 'drbdsetup down $DRBD_RESOURCE' exit code $?; pacemaker may now try to 'recover' by fencing this node."
			# give the logs some time to be persisted.
			( exec sync ) </dev/null &> /dev/null &
			sleep 5
		fi
	fi
	return $ex
}


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

# exporting this is useful for "drbdsetup show".
# and it makes it all a little bit more readable.
export DRBD_RESOURCE=$OCF_RESKEY_drbd_resource

if [ $# != 1 ]; then
	drbd_usage
	exit $OCF_ERR_ARGS
fi

# if $__OCF_ACTION = monitor, but meta_interval not set,
# this is a "probe". we could change behaviour.
: ${OCF_RESKEY_CRM_meta_interval=0}

init_optional_params

case $__OCF_ACTION in
meta-data)
	meta_data
	exit $OCF_SUCCESS
	;;
usage)
	drbd_usage
	exit $OCF_SUCCESS
esac

if $USE_DEBUG_LOG ; then
	exec 2>&8
	set -x
fi

# Everything except usage and meta-data must pass the validate test
drbd_validate_all || exit

case $__OCF_ACTION in
start)
	drbd_start
	;;
stop)
	drbd_stop
	;;
reload)
	drbd_reload
	;;
notify)
	drbd_notify
	;;
promote)
	drbd_promote
	;;
demote)
	drbd_demote
	;;
status)
	drbd_status
	;;
monitor)
	drbd_monitor
	;;
validate-all)
	;;
*)
	drbd_usage
	exit $OCF_ERR_UNIMPLEMENTED
esac
# exit code is the exit code (return code) of the last command (shell function)