diff options
Diffstat (limited to '')
| -rwxr-xr-x | heartbeat/LVM | 470 | ||||
| -rwxr-xr-x | heartbeat/LVM-activate | 997 |
2 files changed, 1467 insertions, 0 deletions
diff --git a/heartbeat/LVM b/heartbeat/LVM new file mode 100755 index 0000000..b587bd8 --- /dev/null +++ b/heartbeat/LVM @@ -0,0 +1,470 @@ +#!/bin/sh +# +# +# LVM +# +# Description: Manages an LVM volume as an HA resource +# +# +# Author: Alan Robertson +# Support: users@clusterlabs.org +# License: GNU General Public License (GPL) +# Copyright: (C) 2002 - 2005 International Business Machines, Inc. +# +# This code significantly inspired by the LVM resource +# in FailSafe by Lars Marowsky-Bree +# +# +# An example usage in /etc/ha.d/haresources: +# node1 10.0.0.170 ServeRAID::1::1 LVM::myvolname +# +# See usage() function below for more details... +# +# OCF parameters are as below: +# OCF_RESKEY_volgrpname +# +####################################################################### +# Initialization: + +: ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/lib/heartbeat} +. ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs + +# Parameter defaults + +OCF_RESKEY_volgrpname_default="" +OCF_RESKEY_exclusive_default="false" +OCF_RESKEY_tag_default="pacemaker" +OCF_RESKEY_partial_activation_default="false" + +: ${OCF_RESKEY_volgrpname=${OCF_RESKEY_volgrpname_default}} +: ${OCF_RESKEY_exclusive=${OCF_RESKEY_exclusive_default}} +: ${OCF_RESKEY_tag=${OCF_RESKEY_tag_default}} +: ${OCF_RESKEY_partial_activation=${OCF_RESKEY_partial_activation_default}} + +####################################################################### + + +usage() { + methods=`LVM_methods` + methods=`echo $methods | tr ' ' '|'` + cat <<EOF + usage: $0 $methods + + $0 manages an Linux Volume Manager volume (LVM) as an HA resource + + The 'start' operation brings the given volume online + The 'stop' operation takes the given volume offline + The 'status' operation reports whether the volume is available + The 'monitor' operation reports whether the volume seems present + The 'validate-all' operation checks whether the OCF parameters are valid + The 'meta-data' operation show meta data + The 'methods' operation reports on the methods $0 supports + +EOF +} + +meta_data() { + cat <<EOF +<?xml version="1.0"?> +<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd"> +<resource-agent name="LVM" version="1.0"> +<version>1.0</version> + +<longdesc lang="en"> +Resource script for LVM. It manages an Linux Volume Manager volume (LVM) +as an HA resource. +</longdesc> +<shortdesc lang="en">Controls the availability of an LVM Volume Group</shortdesc> + +<parameters> +<parameter name="volgrpname" unique="1" required="1"> +<longdesc lang="en"> +The name of volume group. +</longdesc> +<shortdesc lang="en">Volume group name</shortdesc> +<content type="string" default="${OCF_RESKEY_volgrpname_default}" /> +</parameter> +<parameter name="exclusive" unique="0" required="0"> +<longdesc lang="en"> +If set, the volume group will be activated exclusively. This option works one of +two ways. If the volume group has the cluster attribute set, then the volume group +will be activated exclusively using clvmd across the cluster. If the cluster attribute +is not set, the volume group will be activated exclusively using a tag and the volume_list +filter. When the tag option is in use, the volume_list in lvm.con must be initialized. This +can be as simple as setting 'volume_list = []' depending on your setup. +</longdesc> +<shortdesc lang="en">Exclusive activation</shortdesc> +<content type="boolean" default="${OCF_RESKEY_exclusive_default}" /> +</parameter> + +<parameter name="tag" unique="0" required="0"> +<longdesc lang="en"> +If "exclusive" is set on a non clustered volume group, this overrides the tag to be used. +</longdesc> +<shortdesc lang="en">Exclusive activation tag</shortdesc> +<content type="string" default="${OCF_RESKEY_tag_default}" /> +</parameter> + +<parameter name="partial_activation" unique="0" required="0"> +<longdesc lang="en"> +If set, the volume group will be activated partially even with some +physical volumes missing. It helps to set to true when using mirrored +logical volumes. +</longdesc> +<shortdesc lang="en">Activate VG partially when missing PVs</shortdesc> +<content type="string" default="${OCF_RESKEY_partial_activation_default}" /> +</parameter> + +</parameters> + +<actions> +<action name="start" timeout="30s" /> +<action name="stop" timeout="30s" /> +<action name="status" timeout="30s" /> +<action name="monitor" depth="0" timeout="30s" interval="10s" /> +<action name="methods" timeout="5s" /> +<action name="meta-data" timeout="5s" /> +<action name="validate-all" timeout="5s" /> +</actions> +</resource-agent> +EOF +} + +# +# methods: What methods/operations do we support? +# +LVM_methods() { + cat <<EOF + start + stop + status + monitor + methods + validate-all + meta-data + usage +EOF +} + +## +# +# plain = normal (non-exclusive) local activation +# tag = tagged-exclusive activation +# clvm = clvm-exclusive activation +# +# the mode specific implementation is in lvm-$mode.sh +## + +set_lvm_mode() { + local mode + + if ocf_is_true "$OCF_RESKEY_exclusive"; then + case $(vgs -o attr --noheadings $OCF_RESKEY_volgrpname | tr -d ' ') in + ?????c*) + mode="clvm" ;; + *) + mode="tag" ;; + esac + else + mode="plain" + fi + + . ${OCF_FUNCTIONS_DIR}/lvm-${mode}.sh +} + +# +# Return LVM status (silently) +# +LVM_status() { + local rc=1 + loglevel="debug" + + # Set the log level of the error message + if [ "X${2}" = "X" ]; then + loglevel="err" + if ocf_is_probe; then + loglevel="warn" + else + if [ ${OP_METHOD} = "stop" ]; then + loglevel="info" + fi + fi + fi + + if [ -d /dev/$1 ]; then + test "`cd /dev/$1 && ls`" != "" + rc=$? + if [ $rc -ne 0 ]; then + ocf_exit_reason "VG $1 with no logical volumes is not supported by this RA!" + fi + fi + + if [ $rc -ne 0 ]; then + ocf_log $loglevel "LVM Volume $1 is not available (stopped)" + rc=$OCF_NOT_RUNNING + else + lvm_status + rc=$? + fi + + if [ "X${2}" = "X" ]; then + # status call return + return $rc + fi + + # Report on LVM volume status to stdout... + if [ $rc -eq 0 ]; then + echo "Volume $1 is available (running)" + else + echo "Volume $1 is not available (stopped)" + fi + return $rc +} + +# +# Enable LVM volume +# +LVM_start() { + local vg=$1 + + # systemd drop-in to stop process before storage services during + # shutdown/reboot + if systemd_is_running ; then + systemd_drop_in "99-LVM" "After" "blk-availability.service" + fi + + # TODO: This MUST run vgimport as well + ocf_log info "Activating volume group $vg" + if [ "$LVM_MAJOR" -eq "1" ]; then + ocf_run vgscan $vg + else + ocf_run vgscan + fi + + lvm_pre_activate || exit + ocf_run vgchange $vgchange_activate_options $vg + lvm_post_activate $? + + if LVM_status $vg; then + : OK Volume $vg activated just fine! + return $OCF_SUCCESS + else + ocf_exit_reason "LVM: $vg did not activate correctly" + return $OCF_ERR_GENERIC + fi +} + +# +# Disable the LVM volume +# +LVM_stop() { + local res=$OCF_ERR_GENERIC + local vg=$1 + + if ! vgs $vg > /dev/null 2>&1; then + ocf_log info "Volume group $vg not found" + return $OCF_SUCCESS + fi + + ocf_log info "Deactivating volume group $vg" + + lvm_pre_deactivate || exit + + for i in $(seq 10) + do + ocf_run vgchange $vgchange_deactivate_options $vg + res=$? + if LVM_status $vg; then + ocf_exit_reason "LVM: $vg did not stop correctly" + res=1 + fi + + if [ $res -eq 0 ]; then + break + fi + + res=$OCF_ERR_GENERIC + ocf_log warn "$vg still Active" + ocf_log info "Retry deactivating volume group $vg" + sleep 1 + which udevadm > /dev/null 2>&1 && udevadm settle --timeout=5 + done + + lvm_post_deactivate $res +} + +# +# Check whether the OCF instance parameters are valid +# +LVM_validate_all() { + check_binary $AWK + + ## + # lvmetad is a daemon that caches lvm metadata to improve the + # performance of LVM commands. This daemon should never be used when + # volume groups exist that are being managed by the cluster. The lvmetad + # daemon introduces a response lag, where certain LVM commands look like + # they have completed (like vg activation) when in fact the command + # is still in progress by the lvmetad. This can cause reliability issues + # when managing volume groups in the cluster. For Example, if you have a + # volume group that is a dependency for another application, it is possible + # the cluster will think the volume group is activated and attempt to start + # the application before volume group is really accesible... lvmetad is bad. + ## + lvm dumpconfig global/use_lvmetad | grep 'use_lvmetad.*=.*1' > /dev/null 2>&1 + if [ $? -eq 0 ]; then + # for now warn users that lvmetad is enabled and that they should disable it. In the + # future we may want to consider refusing to start, or killing the lvmetad daemon. + ocf_log warn "Disable lvmetad in lvm.conf. lvmetad should never be enabled in a clustered environment. Set use_lvmetad=0 and kill the lvmetad process" + fi + + ## + # Off-the-shelf tests... + ## + VGOUT=`vgck ${VOLUME} 2>&1` + if [ $? -ne 0 ]; then + # Inconsistency might be due to missing physical volumes, which doesn't + # automatically mean we should fail. If partial_activation=true then + # we should let start try to handle it, or if no PVs are listed as + # "unknown device" then another node may have marked a device missing + # where we have access to all of them and can start without issue. + if vgs -o pv_attr --noheadings $OCF_RESKEY_volgrpname 2>/dev/null | grep 'm' > /dev/null 2>&1; then + case $(vgs -o attr --noheadings $OCF_RESKEY_volgrpname | tr -d ' ') in + ???p??*) + if ! ocf_is_true "$OCF_RESKEY_partial_activation" ; then + # We are missing devices and cannot activate partially + ocf_exit_reason "Volume group [$VOLUME] has devices missing. Consider partial_activation=true to attempt to activate partially" + exit $OCF_ERR_GENERIC + else + # We are missing devices but are allowed to activate partially. + # Assume that caused the vgck failure and carry on + ocf_log warn "Volume group inconsistency detected with missing device(s) and partial_activation enabled. Proceeding with requested action." + fi + ;; + esac + # else the vg is partial but all devices are accounted for, so another + # node must have marked the device missing. Proceed. + else + # vgck failure was for something other than missing devices + ocf_exit_reason "Volume group [$VOLUME] does not exist or contains error! ${VGOUT}" + exit $OCF_ERR_GENERIC + fi + fi + + ## + # Does the Volume Group exist? + ## + if [ "$LVM_MAJOR" = "1" ]; then + VGOUT=`vgdisplay ${VOLUME} 2>&1` + else + VGOUT=`vgdisplay -v ${VOLUME} 2>&1` + fi + if [ $? -ne 0 ]; then + ocf_exit_reason "Volume group [$VOLUME] does not exist or contains error! ${VGOUT}" + exit $OCF_ERR_GENERIC + fi + + if lvs --noheadings -o segtype | grep -q "cache"; then + if ! lvs --noheadings -o cache_mode "$OCF_RESKEY_volgrpname" | grep -q "writethrough"; then + ocf_log warn "LVM CACHE IS NOT IN WRITETHROUGH MODE. THIS IS NOT A SUPPORTED CONFIGURATION." + fi + fi + + if ocf_is_clone && ocf_is_true "$OCF_RESKEY_exclusive"; then + ocf_exit_reason "cloned lvm resources can not be activated exclusively" + exit $OCF_ERR_CONFIGURED + fi + + lvm_validate_all +} + +# +# 'main' starts here... +# + +if + [ $# -ne 1 ] +then + usage + exit $OCF_ERR_ARGS +fi + +case $1 in + meta-data) meta_data + exit $OCF_SUCCESS;; + + methods) LVM_methods + exit $?;; + + usage) usage + exit $OCF_SUCCESS;; + *) ;; +esac + +if + [ -z "$OCF_RESKEY_volgrpname" ] +then + ocf_exit_reason "You must identify the volume group name!" + exit $OCF_ERR_CONFIGURED +fi + +# Get the LVM version number, for this to work we assume(thanks to panjiam): +# +# LVM1 outputs like this +# +# # vgchange --version +# vgchange: Logical Volume Manager 1.0.3 +# Heinz Mauelshagen, Sistina Software 19/02/2002 (IOP 10) +# +# LVM2 and higher versions output in this format +# +# # vgchange --version +# LVM version: 2.00.15 (2004-04-19) +# Library version: 1.00.09-ioctl (2004-03-31) +# Driver version: 4.1.0 + +LVM_VERSION=`vgchange --version 2>&1 | \ + $AWK '/Logical Volume Manager/ {print $5"\n"; exit; } + /LVM version:/ {printf $3"\n"; exit;}'` +rc=$? + +if + ( [ $rc -ne 0 ] || [ -z "$LVM_VERSION" ] ) +then + ocf_exit_reason "LVM: $1 could not determine LVM version. Try 'vgchange --version' manually and modify $0 ?" + exit $OCF_ERR_INSTALLED +fi +LVM_MAJOR="${LVM_VERSION%%.*}" + +VOLUME=$OCF_RESKEY_volgrpname +OP_METHOD=$1 + +set_lvm_mode +lvm_init +if ocf_is_true "$OCF_RESKEY_partial_activation" ; then + vgchange_activate_options="${vgchange_activate_options} --partial" +fi + +# What kind of method was invoked? +case "$1" in + + start) + LVM_validate_all + LVM_start $VOLUME + exit $?;; + + stop) LVM_stop $VOLUME + exit $?;; + + status) LVM_status $VOLUME $1 + exit $?;; + + monitor) LVM_status $VOLUME + exit $?;; + + validate-all) LVM_validate_all + ;; + + *) usage + exit $OCF_ERR_UNIMPLEMENTED;; +esac diff --git a/heartbeat/LVM-activate b/heartbeat/LVM-activate new file mode 100755 index 0000000..f6f24a3 --- /dev/null +++ b/heartbeat/LVM-activate @@ -0,0 +1,997 @@ +#!/bin/sh +# +# +# Copyright (c) 2017 SUSE LINUX, Eric Ren +# 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. +# +# LVM-activate OCF Resource Agent: +# +# Logical volume manager (LVM) provides new features for cluster enviroment: +# lvmlockd and system ID, which aims to replace clvmd and tagged-exclusive +# activation. Accordingly, we have created a new resource agent named "lvmlockd" +# to manage lvmlockd daemon. In addition, this new resource agent "LVM-activate" +# is created to take care of LVM activation/deactivation work. This agent supports +# the new features: lvmlockd and system ID, and also supports the old features: +# clvmd and lvm tag. +# +# Thanks David Teigland! He is the author of these LVM features, giving valuable +# idea/feedback about this resource agent. +############################################################################ + +# Initialization: + +: ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/lib/heartbeat} +. ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs + +# Parameter defaults + +OCF_RESKEY_vgname_default="" +OCF_RESKEY_lvname_default="" +OCF_RESKEY_vg_access_mode_default="" +OCF_RESKEY_activation_mode_default="exclusive" +OCF_RESKEY_tag_default="pacemaker" +OCF_RESKEY_partial_activation_default="false" +OCF_RESKEY_degraded_activation_default="false" +OCF_RESKEY_majority_pvs_default="false" + +: ${OCF_RESKEY_vgname=${OCF_RESKEY_vgname_default}} +: ${OCF_RESKEY_lvname=${OCF_RESKEY_lvname_default}} +: ${OCF_RESKEY_vg_access_mode=${OCF_RESKEY_vg_access_mode_default}} +: ${OCF_RESKEY_activation_mode=${OCF_RESKEY_activation_mode_default}} +: ${OCF_RESKEY_tag=${OCF_RESKEY_tag_default}} +: ${OCF_RESKEY_partial_activation=${OCF_RESKEY_partial_activation_default}} +: ${OCF_RESKEY_degraded_activation=${OCF_RESKEY_degraded_activation_default}} +: ${OCF_RESKEY_majority_pvs=${OCF_RESKEY_majority_pvs_default}} + +# If LV is given, only activate this named LV; otherwise, activate all +# LVs in the named VG. +VG=${OCF_RESKEY_vgname} +LV=${OCF_RESKEY_lvname} + +# How LVM controls access to the VG: +# +# 0: place-holder for any incorrect cases; To be safe, we enforce the VG +# must use any of the following protection methods in cluster environment. +# 1: vg is shared - lvmlockd (new) +# 2: vg is clustered - clvmd (old) +# 3: vg has system_id (new) +# 4: vg has tagging (old) +VG_access_mode=${OCF_RESKEY_vg_access_mode} +VG_access_mode_num=0 + +# Activate LV(s) with "shared" lock for cluster fs +# or "exclusive" lock for local fs +LV_activation_mode=${OCF_RESKEY_activation_mode} + +# For system ID feature +SYSTEM_ID="" + +# For tagging activation mode +OUR_TAG=${OCF_RESKEY_tag} + +####################################################################### + +meta_data() { + cat <<END +<?xml version="1.0"?> +<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd"> + + +<resource-agent name="LVM-activate" version="1.0"> +<version>1.0</version> + +<longdesc lang="en"> +This agent manages LVM activation/deactivation work for a given volume group. + +It supports the following modes, controlled by the vg_access_mode parameter: + +* lvmlockd +* system_id +* clvmd +* tagging + +Notes: + +1. There are two possible configuration combinations: lvmlockd+LVM-activate and +clvm+LVM-activate. However, it is not possible to use both at the same time! + +2. Put all "lvmlockd"/"clvmd" volume groups into auto management by the agent +if using the cluster to manage at least one of them. If you manage some manually, +the stop action of the lvmlockd agent may fail and the node may get fenced, +because some DLM lockspaces might be in use and cannot be closed automatically. + +3. The autoactivation property of volume group will be disabled when vg_access_mode +is set to system_id. + +Option: OCF_CHECK_LEVEL + +The standard monitor operation of depth 0 checks if the VG or LV is valid. +If you want deeper tests, set OCF_CHECK_LEVEL to 10: + + 10: read first 1 byte of the underlying device (raw read) + +If there are many underlying devs in VG, it will only read one of the devs. +This is not perfect solution for detecting underlying devices livable. +e.g. iscsi SAN IO timeout will return EIO, and it makes monitor failed. + +</longdesc> +<shortdesc lang="en">This agent activates/deactivates logical volumes.</shortdesc> + +<parameters> +<parameter name="vgname" unique="0" required="1"> +<longdesc lang="en"> +The volume group name. +</longdesc> +<shortdesc lang="en">The volume group name</shortdesc> +<content type="string" default="${OCF_RESKEY_vgname_default}" /> +</parameter> + +<parameter name="lvname" unique="0" required="0"> +<longdesc lang="en"> +If set, only the specified LV will be activated. +</longdesc> +<shortdesc lang="en">Only activate the given LV</shortdesc> +<content type="string" default="${OCF_RESKEY_lvname_default}" /> +</parameter> + +<parameter name="vg_access_mode" unique="0" required="1"> +<longdesc lang="en"> +This option decides which solution will be used to protect the volume group in +cluster environment. Optional solutions are: lvmlockd, clvmd, system_id and +tagging. +</longdesc> +<shortdesc lang="en">The VG access mode</shortdesc> +<content type="string" default="${OCF_RESKEY_vg_access_mode_default}" /> +</parameter> + +<parameter name="activation_mode" unique="0" required="0"> +<longdesc lang="en"> +The activation mode decides the visibility of logical volumes in the cluster. There +are two different modes: "shared" for cluster filesystem and "exclusive" for local +filesystem. With "shared", an LV can be activated concurrently from multiple nodes. +With "exclusive", an LV can be activated by one node at a time. + +This option only has effect on "lvmlockd"/"clvmd" vg_access_mode. For "system_id" +and "tagging", they always mean exclusive activation. +</longdesc> +<shortdesc lang="en">Logical volume activation mode</shortdesc> +<content type="string" default="${OCF_RESKEY_activation_mode_default}" /> +</parameter> + +<parameter name="tag" unique="0" required="0"> +<longdesc lang="en"> +The tag used for tagging activation mode. +</longdesc> +<shortdesc lang="en">The tag used for tagging activation mode</shortdesc> +<content type="string" default="${OCF_RESKEY_tag_default}" /> +</parameter> + +<parameter name="partial_activation" unique="0" required="0"> +<longdesc lang="en"> +If set, the volume group will be activated partially even with some +physical volumes missing. It helps to set to true when using mirrored +logical volumes. +</longdesc> +<shortdesc lang="en">Activate VG partially when missing PVs</shortdesc> +<content type="string" default="${OCF_RESKEY_partial_activation_default}" /> +</parameter> + +<parameter name="degraded_activation" unique="0" required="0"> +<longdesc lang="en"> +Activate RAID LVs using the "degraded" activation mode. This allows RAID +LVs to be activated with missing PVs if all data can be provided with +RAID redundancy. The RAID level determines the number of PVs that are +required for degraded activation to succeed. If fewer PVs are available, +then degraded activation will fail. Also enable majority_pvs. +</longdesc> +<shortdesc lang="en">Activate RAID LVs in degraded mode when missing PVs</shortdesc> +<content type="string" default="${OCF_RESKEY_degraded_activation_default}" /> +</parameter> + +<parameter name="majority_pvs" unique="0" required="0"> +<longdesc lang="en"> +If set, the VG system ID can be reassigned to a new host if a majority +of PVs in the VG are present. Otherwise, VG failover with system ID +will fail when the VG is missing PVs. Also enable degraded_activation +when RAID LVs are used. +</longdesc> +<shortdesc lang="en">Allow changing the system ID of a VG with a majority of PVs</shortdesc> +<content type="string" default="${OCF_RESKEY_majority_pvs_default}" /> +</parameter> + +</parameters> + +<actions> +<action name="start" timeout="90s" /> +<action name="stop" timeout="90s" /> +<action name="monitor" timeout="90s" interval="30s" depth="0" /> +<action name="meta-data" timeout="10s" /> +<action name="validate-all" timeout="20s" /> +</actions> +</resource-agent> +END +} + +####################################################################### + +usage() { + cat <<END +usage: $0 {start|stop|monitor|validate-all|meta-data} +END +} + +# VG access modes: +# 0: unsafe to activate LV without proper protection in cluster +# 1: vg is shared - lvmlockd (new) +# 2: vg is clustered - clvmd (old) +# 3: vg has system_id (new) +# 4: vg has tagging (old) + +get_VG_access_mode_num() +{ + # Use -o reporting fields to get multiple bits of info from a single command + kvs=$(vgs --foreign --nolocking --noheadings --nameprefixes \ + --rows --config report/compact_output=0 \ + -o name,clustered,lock_type,systemid,tags ${VG} 2>/dev/null | tr -d \') + export ${kvs} + vg_locktype=${LVM2_VG_LOCK_TYPE} + vg_clustered=${LVM2_VG_CLUSTERED} + vg_systemid=${LVM2_VG_SYSTEMID} + vg_tags=${LVM2_VG_TAGS} + + # We know this VG is using lvmlockd if the lock type is dlm. + if [ "$vg_locktype" = "dlm" ]; then + access_mode=1 + elif [ "$vg_clustered" = "clustered" ]; then + access_mode=2 + elif [ -n "$vg_systemid" ]; then + SYSTEM_ID=$(lvm systemid 2>/dev/null | cut -d':' -f2 | tr -d '[:blank:]') + access_mode=3 + elif [ -n "$vg_tags" ]; then + # TODO: + # We don't have reliable way to test if tagging activation is used. + access_mode=4 + else + access_mode=0 + fi + + return $access_mode +} + +# TODO: All tagging activation code is almost copied from LVM RA!!! +# But, the old LVM RA just uses the ordinary tags, not the "hosttag" feature +# which may be a better method for active-inactive cluster scenario. +# +# We have two choice: +# 1. Continue to use the LVM way, which may work well on old system. +# 2. Change to use the real hosttag feature, but it looks very same +# to systemID. +# Anyway, we can easily change this if anyone requests with good reasons. + +# does this vg have our tag +check_tags() +{ + owner=$(vgs -o tags --noheadings ${VG} | tr -d '[:blank:]') + + if [ -z "$owner" ]; then + # No-one owns this VG yet + return 1 + fi + + if [ "$OUR_TAG" = "$owner" ]; then + # yep, this is ours + return 0 + fi + + # some other tag is set on this vg + return 2 +} + +strip_tags() +{ + for tag in $(vgs --noheadings -o tags $OCF_RESKEY_volgrpname | sed s/","/" "/g); do + ocf_log info "Stripping tag, $tag" + + # LVM version 2.02.98 allows changing tags if PARTIAL + vgchange --deltag "$tag" ${VG} + done + + if [ ! -z $(vgs -o tags --noheadings ${VG} | tr -d '[:blank:]') ]; then + ocf_exit_reason "Failed to remove ownership tags from ${VG}" + exit $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +set_tags() +{ + case check_tags in + 0) + # we already own it. + return $OCF_SUCCESS + ;; + 2) + # other tags are set, strip them before setting + if ! strip_tags; then + return $OCF_ERR_GENERIC + fi + ;; + *) + : ;; + esac + + if ! vgchange --addtag $OUR_TAG ${VG} ; then + ocf_exit_reason "Failed to add ownership tag to ${VG}" + return $OCF_ERR_GENERIC + fi + + ocf_log info "New tag \"${OUR_TAG}\" added to ${VG}" + return $OCF_SUCCESS +} + +# Parameters: +# 1st: config item name +# 2nd: expected config item value +config_verify() +{ + name=$1 + expect=$2 + + real=$(lvmconfig "$name" | cut -d'=' -f2) + if [ "$real" != "$expect" ]; then + ocf_exit_reason "config item $name: expect=$expect but real=$real" + exit $OCF_ERR_ARGS + fi + + return $OCF_SUCCESS +} + +lvmlockd_check() +{ + config_verify "global/use_lvmlockd" "1" + + # locking_type was removed from config in v2.03 + ocf_version_cmp "$(lvmconfig --version | awk '/LVM ver/ {sub(/\(.*/, "", $3); print $3}')" "2.03" + if [ "$?" -eq 0 ]; then + config_verify "global/locking_type" "1" + fi + + # We recommend to activate one LV at a time so that this specific volume + # binds to a proper filesystem to protect the data + # TODO: + # Will this warn message be too noisy? + if [ -z "$LV" ]; then + ocf_log warn "You are recommended to activate one LV at a time or use exclusive activation mode." + fi + + # Good: lvmlockd is running, and clvmd is not running + if ! pgrep lvmlockd >/dev/null 2>&1 ; then + if ocf_is_probe; then + ocf_log info "initial probe: lvmlockd is not running yet." + exit $OCF_NOT_RUNNING + fi + + ocf_exit_reason "lvmlockd daemon is not running!" + exit $OCF_ERR_GENERIC + fi + + if pgrep clvmd >/dev/null 2>&1 ; then + ocf_exit_reason "clvmd daemon is running unexpectedly." + exit $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +clvmd_check() +{ + config_verify "global/use_lvmetad" "0" + config_verify "global/use_lvmlockd" "0" + config_verify "global/locking_type" "3" + + # TODO: + # David asked a question: does lvchange -aey works well enough with clvmd? + # + # Corey said: I think it does work well enough. We do a fair amount of exclusive + # activation clvm testing, and my experience is you'll get the LV activated on + # the node you ran the command on. But, I think the specific scenario and issue + # that surprised us all was when the LV was *already* exclusively active on say + # nodeA, and nodeB then attempts to also exclusively activate it as well. Instead + # of failing, the activation succeeds even though nodeB activation didn't occur. + # This is documented in the following bug: + # https://bugzilla.redhat.com/show_bug.cgi?id=1191724#c8 + # Technically, you're not guaranteed to have it activated on the node you run + # the cmd on, but again, that's not been my experience. + # + # Eric: Put the interesting discussion here so that we can be more careful on this. + + # Good: clvmd is running, and lvmlockd is not running + if ! pgrep clvmd >/dev/null 2>&1 ; then + ocf_exit_reason "clvmd daemon is not running!" + exit $OCF_ERR_GENERIC + fi + + if pgrep lvmetad >/dev/null 2>&1 ; then + ocf_exit_reason "Please stop lvmetad daemon when clvmd is running." + exit $OCF_ERR_GENERIC + fi + + if pgrep lvmlockd >/dev/null 2>&1 ; then + ocf_exit_reason "lvmlockd daemon is running unexpectedly." + exit $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +systemid_check() +{ + # system_id_source is set in lvm.conf + source=$(lvmconfig 'global/system_id_source' 2>/dev/null | cut -d"=" -f2) + if [ "$source" = "" ] || [ "$source" = "none" ]; then + ocf_exit_reason "system_id_source in lvm.conf is not set correctly!" + exit $OCF_ERR_ARGS + fi + + if [ -z ${SYSTEM_ID} ]; then + ocf_exit_reason "local/system_id is not set!" + exit $OCF_ERR_ARGS + fi + + return $OCF_SUCCESS +} + +# Verify tags setup +tagging_check() +{ + # The volume_list must be initialized to something in order to + # guarantee our tag will be filtered on startup + if ! lvm dumpconfig activation/volume_list; then + ocf_log err "LVM: Improper setup detected" + ocf_exit_reason "The volume_list filter must be initialized in lvm.conf for exclusive activation without clvmd" + exit $OCF_ERR_ARGS + fi + + # Our tag must _NOT_ be in the volume_list. This agent + # overrides the volume_list during activation using the + # special tag reserved for cluster activation + if lvm dumpconfig activation/volume_list | grep -e "\"@${OUR_TAG}\"" -e "\"${VG}\""; then + ocf_log err "LVM: Improper setup detected" + ocf_exit_reason "The volume_list in lvm.conf must not contain the cluster tag, \"${OUR_TAG}\", or volume group, ${VG}" + exit $OCF_ERR_ARGS + fi + + return $OCF_SUCCESS +} + +read_parameters() +{ + if [ -z "$VG" ] + then + ocf_exit_reason "You must identify the volume group name!" + exit $OCF_ERR_CONFIGURED + fi + + if [ "$LV_activation_mode" != "shared" ] && [ "$LV_activation_mode" != "exclusive" ] + then + ocf_exit_reason "Invalid value for activation_mode: $LV_activation_mode" + exit $OCF_ERR_CONFIGURED + fi + + # Convert VG_access_mode from string to index + case ${VG_access_mode} in + lvmlockd) + VG_access_mode_num=1 + ;; + clvmd) + VG_access_mode_num=2 + ;; + system_id) + VG_access_mode_num=3 + ;; + tagging) + VG_access_mode_num=4 + ;; + *) + # dont exit with error-code here or nodes will get fenced on + # e.g. "pcs resource create" + ocf_exit_reason "You specified an invalid value for vg_access_mode: $VG_access_mode" + ;; + esac +} + +lvm_validate() { + read_parameters + + check_binary pgrep + # Every LVM command is just symlink to lvm binary + check_binary lvm + check_binary dmsetup + + # This is necessary when using system ID to update lvm hints, + # or in older versions of lvm, this is necessary to update the + # lvmetad cache. + pvscan --cache + + if ! vgs --foreign ${VG} >/dev/null 2>&1 ; then + # stop action exits successfully if the VG cannot be accessed... + if [ $__OCF_ACTION = "stop" ]; then + ocf_log warn "VG [${VG}] cannot be accessed, stop action exits successfully." + exit $OCF_SUCCESS + fi + + if ocf_is_probe; then + ocf_log info "initial probe: VG [${VG}] is not found on any block device yet." + exit $OCF_NOT_RUNNING + fi + + # Could be a transient error (e.g., iSCSI connection + # issue) so use OCF_ERR_GENERIC + ocf_exit_reason "Volume group[${VG}] doesn't exist, or not visible on this node!" + exit $OCF_ERR_GENERIC + fi + + vg_missing_pv_count=$(vgs -o missing_pv_count --noheadings ${VG} 2>/dev/null) + + if [ $vg_missing_pv_count -gt 0 ]; then + ocf_log warn "Volume Group ${VG} is missing $vg_missing_pv_count PVs." + + # Setting new system ID will succeed if over half of PVs remain. + # Don't try to calculate here if a majority is present, + # but leave this up to the vgchange command to determine. + if ocf_is_true "$OCF_RESKEY_majority_pvs" ; then + ocf_log warn "Attempting fail over with missing PVs (majority.)" + + # Setting new system ID will fail, and behavior is undefined for + # other access modes. + elif ocf_is_true "$OCF_RESKEY_partial_activation" ; then + ocf_log warn "Attempting fail over with missing PVs (partial.)" + + else + ocf_exit_reason "Volume group [$VG] has devices missing. Consider majority_pvs=true" + exit $OCF_ERR_GENERIC + fi + fi + + # Get the access mode from VG metadata and check if it matches the input + # value. Skip to check "tagging" mode because there's no reliable way to + # automatically check if "tagging" mode is being used. + get_VG_access_mode_num + mode=$? + if [ $VG_access_mode_num -ne 4 ] && [ $mode -ne $VG_access_mode_num ]; then + ocf_exit_reason "The specified vg_access_mode doesn't match the lock_type on VG metadata!" + exit $OCF_ERR_CONFIGURED + fi + + # Nothing to do if the VG has no logical volume + lv_count=$(vgs --foreign -o lv_count --noheadings ${VG} 2>/dev/null) + if [ $lv_count -lt 1 ]; then + ocf_exit_reason "Volume group [$VG] doesn't contain any logical volume!" + exit $OCF_ERR_CONFIGURED + fi + + # Check if the given $LV is in the $VG + if [ -n "$LV" ]; then + output=$(lvs --foreign --noheadings ${VG}/${LV} 2>&1) + if [ $? -ne 0 ]; then + ocf_log err "lvs: ${output}" + ocf_exit_reason "LV ($LV) is not in the given VG ($VG)." + exit $OCF_ERR_CONFIGURED + fi + fi + + # VG_access_mode_num specific checking goes here + case ${VG_access_mode_num} in + 1) + lvmlockd_check + ;; + 2) + clvmd_check + ;; + 3) + systemid_check + ;; + 4) + tagging_check + ;; + *) + ocf_exit_reason "Incorrect VG access mode detected!" + exit $OCF_ERR_CONFIGURED + esac + + if [ $? -ne $OCF_SUCCESS ]; then + ocf_exit_reason "Improper configuration issue is detected!" + exit $OCF_ERR_CONFIGURED + fi + + return $OCF_SUCCESS +} + +# To activate LV(s) with different "activation mode" parameters +do_activate() { + do_activate_opt=$1 + + if ocf_is_true "$OCF_RESKEY_degraded_activation" ; then + # This will allow a RAID LV to be activated if sufficient + # devices are available to allow the LV to be usable + do_activate_opt="${do_activate_opt} --activationmode degraded" + + elif ocf_is_true "$OCF_RESKEY_partial_activation" ; then + # This will allow a mirror LV to be activated if any + # devices are missing, but the activated LV may not be + # usable, so it is not recommended. Also, other LV + # types without data redundancy will be activated + # when partial is set. + # RAID LVs and degraded_activation should be used instead. + do_activate_opt="${do_activate_opt} --partial" + fi + + # Only activate the specific LV if it's given + if [ -n "$LV" ]; then + ocf_run lvchange $do_activate_opt ${VG}/${LV} + if [ $? -ne $OCF_SUCCESS ]; then + return $OCF_ERR_GENERIC + fi + else + ocf_run lvchange $do_activate_opt ${VG} + if [ $? -ne $OCF_SUCCESS ]; then + return $OCF_ERR_GENERIC + fi + fi + + return $OCF_SUCCESS +} + +lvmlockd_activate() { + if [ "$LV_activation_mode" = "shared" ]; then + activate_opt="-asy" + else + activate_opt="-aey" + fi + + # lvmlockd requires shared VGs to be started before they're used + ocf_run vgchange --lockstart ${VG} + rc=$? + if [ $rc -ne $OCF_SUCCESS ]; then + ocf_exit_reason "Failed to start shared VG(s), exit code: $rc" + return $OCF_ERR_GENERIC + fi + + do_activate "$activate_opt" + if [ $? -ne $OCF_SUCCESS ]; then + return $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +# clvmd must be running to activate clustered VG +clvmd_activate() { + if [ "$LV_activation_mode" = "shared" ]; then + activate_opt="-asy" + else + activate_opt="-aey" + fi + + do_activate "$activate_opt" + if [ $? -ne $OCF_SUCCESS ]; then + return $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +systemid_activate() { + majority_opt="" + set_autoactivation=0 + cur_systemid=$(vgs --foreign --noheadings -o systemid ${VG} | tr -d '[:blank:]') + + if ocf_is_true "$OCF_RESKEY_majority_pvs" ; then + vgchange --help | grep '\--majoritypvs' >/dev/null 2>&1 && majority_opt="--majoritypvs" + fi + + # Put our system ID on the VG + vgchange -y $majority_opt --config "local/extra_system_ids=[\"${cur_systemid}\"]" \ + --systemid ${SYSTEM_ID} ${VG} + vgchange --help | grep '\--setautoactivation' >/dev/null 2>&1 && set_autoactivation=1 + + if [ $set_autoactivation -ne 0 ]; then + if vgs -o autoactivation ${VG} | grep enabled >/dev/null 2>&1 ; then + ocf_log info "disable the autoactivation property of ${VG}" + ocf_run vgchange --setautoactivation n ${VG} + fi + fi + + do_activate "-ay" + if [ $? -ne $OCF_SUCCESS ]; then + return $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +tagging_activate() { + if ! set_tags ; then + ocf_exit_reason "Failed to set tags on ${VG}." + return $OCF_ERR_GENERIC + fi + + do_activate "-ay --config activation{volume_list=[\"@${OUR_TAG}\"]}" + if [ $? -ne $OCF_SUCCESS ]; then + return $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +lvmlockd_deactivate() { + do_activate "-an" + if [ $? -ne $OCF_SUCCESS ]; then + return $OCF_ERR_GENERIC + fi + + OUT=$(lvs --noheadings -S lv_active=active ${VG} 2>/dev/null) + [ -n "$OUT" ] && return $OCF_SUCCESS + + # Close the lockspace of this VG if there is no active LV + ocf_run vgchange --lockstop ${VG} + rc=$? + if [ $rc -ne $OCF_SUCCESS ]; then + ocf_exit_reason "Failed to close the shared VG lockspace, exit code: $rc" + return $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +clvmd_deactivate() { + do_activate "-an" + if [ $? -ne $OCF_SUCCESS ]; then + return $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +systemid_deactivate() { + do_activate "-an" + if [ $? -ne $OCF_SUCCESS ]; then + return $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +tagging_deactivate() { + do_activate "-an --config activation{volume_list=[\"@${OUR_TAG}\"]}" + if [ $? -ne $OCF_SUCCESS ]; then + return $OCF_ERR_GENERIC + fi + + if ! strip_tags ; then + ocf_exit_reason "Failed to remove tags on ${VG}." + return $OCF_ERR_GENERIC + fi + + return $OCF_SUCCESS +} + +# TODO: +# How can we accurately check if LVs in the given VG are all active? +# +# David: +# If we wanted to check that all LVs in the VG are active, then we would +# probably need to use the lvs/lv_live_table command here since dmsetup +# won't know about inactive LVs that should be active. +# +# Eric: +# But, lvs/lv_live_table command doesn't work well now. I tried the following +# method: +# +# lv_count=$(vgs --foreign -o lv_count --noheadings ${VG} 2>/dev/null | tr -d '[:blank:]') +# dm_count=$(dmsetup --noheadings info -c -S "vg_name=${VG}" 2>/dev/null | grep -c "${VG}-") +# test $lv_count -eq $dm_count +# +# It works, but we cannot afford to use LVM command in lvm_status. LVM command is expensive +# because it may potencially scan all disks on the system, update the metadata even using +# lvs/vgs when the metadata is somehow inconsistent. +# +# So, we have to make compromise that the VG is assumably active if any LV of the VG is active. +# +# Paul: +# VGS + LVS with "-" in their name get mangled with double dashes in dmsetup. +# Switching to wc and just counting lines while depending on the vgname + lvname filter +# in dmsetup gets around the issue with dmsetup reporting correctly but grep failing. +# +# Logic for both test cases and dmsetup calls changed so they match too. +# +# This is AllBad but there isn't a better way that I'm aware of yet. +lvm_status() { + if [ -n "${LV}" ]; then + # dmsetup ls? It cannot accept device name. It's + # too heavy to list all DM devices. + dm_count=$(dmsetup info --noheadings --noflush -c -S "vg_name=${VG} && lv_name=${LV}" | grep -c -v '^No devices found') + else + dm_count=$(dmsetup info --noheadings --noflush -c -S "vg_name=${VG}" | grep -c -v '^No devices found') + fi + + if [ $dm_count -eq 0 ]; then + if ocf_is_probe ;then + return $OCF_NOT_RUNNING + else + return $OCF_ERR_GENERIC + fi + fi + + case "$OCF_CHECK_LEVEL" in + 0) + ;; + 10) + # if there are many lv in vg dir, pick the first name + dm_name="/dev/${VG}/$(ls -1 /dev/${VG} | head -n 1)" + + # read 1 byte to check the dev is alive + dd if=${dm_name} of=/dev/null bs=1 count=1 >/dev/null \ + 2>&1 + if [ $? -ne 0 ]; then + return $OCF_ERR_GENERIC + fi + return $OCF_SUCCESS + ;; + *) + ocf_exit_reason "unsupported monitor level $OCF_CHECK_LEVEL" + return $OCF_ERR_CONFIGURED + ;; + esac +} + +lvm_start() { + if systemd_is_running ; then + # Create drop-in to deactivate VG before stopping + # storage services during shutdown/reboot. + systemctl show resource-agents-deps.target \ + --property=After | cut -d'=' -f2 | \ + grep -qE "(^|\s)blk-availability.service(\s|$)" + + if [ "$?" -ne 0 ]; then + systemd_drop_in "99-LVM-activate" "After" \ + "blk-availability.service" + fi + + # If blk-availability isn't started, the "After=" + # directive has no effect. + if ! systemctl is-active blk-availability.service ; then + systemctl start blk-availability.service + fi + fi + + if lvm_status ; then + ocf_log info "${vol}: is already active." + return $OCF_SUCCESS + fi + + [ -z ${LV} ] && vol=${VG} || vol=${VG}/${LV} + ocf_log info "Activating ${vol}" + + case ${VG_access_mode_num} in + 1) + lvmlockd_activate + ;; + 2) + clvmd_activate + ;; + 3) + systemid_activate + ;; + 4) + tagging_activate + ;; + *) + ocf_exit_reason "VG [${VG}] is not properly configured in cluster. It's unsafe!" + exit $OCF_ERR_CONFIGURED + ;; + esac + + rc=$? + if lvm_status ; then + ocf_log info "${vol}: activated successfully." + return $OCF_SUCCESS + else + ocf_exit_reason "${vol}: failed to activate." + return $rc + fi +} + +# Deactivate LVM volume(s) +lvm_stop() { + [ -z ${LV} ] && vol=${VG} || vol=${VG}/${LV} + + if ! lvm_status ; then + ocf_log info "${vol}: has already been deactivated." + return $OCF_SUCCESS + fi + + ocf_log info "Deactivating ${vol}" + + case ${VG_access_mode_num} in + 1) + lvmlockd_deactivate + ;; + 2) + clvmd_deactivate + ;; + 3) + systemid_deactivate + ;; + 4) + tagging_deactivate + ;; + *) + ocf_exit_reason "VG [${VG}] is not properly configured in cluster. It's unsafe!" + exit $OCF_SUCCESS + ;; + esac + + if ! lvm_status ; then + ocf_log info "${vol}: deactivated successfully." + return $OCF_SUCCESS + else + ocf_exit_reason "${vol}: failed to deactivate." + return $OCF_ERR_GENERIC + fi +} + +# +# MAIN +# + +case $__OCF_ACTION in +start) + lvm_validate + lvm_start + ;; +stop) + read_parameters + lvm_stop + ;; +monitor) + lvm_status + ;; +validate-all) + lvm_validate + ;; +meta-data) + meta_data + ;; +usage|help) + usage + ;; +*) + usage + exit $OCF_ERR_UNIMPLEMENTED + ;; +esac +rc=$? + +ocf_log debug "${OCF_RESOURCE_INSTANCE} $__OCF_ACTION : $rc" +exit $rc |