diff --git a/heartbeat/LVM-activate b/heartbeat/LVM-activate index c2239d881..3c462c75c 100755 --- a/heartbeat/LVM-activate +++ b/heartbeat/LVM-activate @@ -1,850 +1,855 @@ #!/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} . ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs # If LV is given, only activate this named LV; otherwise, activate all # LVs in the named VG. VG=${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=${OCF_RESKEY_activation_mode:-exclusive} # For system ID feature SYSTEM_ID="" # For tagging activation mode OUR_TAG=${OCF_RESKEY_tag:-pacemaker} ####################################################################### meta_data() { cat < 1.0 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. This agent activates/deactivates logical volumes. The volume group name. The volume group name If set, only the specified LV will be activated. Only activate the given LV 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. The VG access mode 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. Logical volume activation mode The tag used for tagging activation mode. The tag used for tagging activation mode END } ####################################################################### usage() { cat </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 $access_mode } # TODO: All tagging activation code is almost copied from LVM RA!!! # But, the old LVM RA just uses the ordinary tags, not the "hosttag" feature # which may be a better method for active-inactive cluster scenario. # # We have two choice: # 1. Continue to use the LVM way, which may work well on old system. # 2. Change to use the real hosttag feature, but it looks very same # to systemID. # Anyway, we can easily change this if anyone requests with good reasons. # does this vg have our tag check_tags() { local owner=$(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() { local tag 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_log err "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() { local name=$1 local expect=$2 local real="" real=$(lvmconfig "$name" | cut -d'=' -f2) if [ "$real" != "$expect" ]; then ocf_exit_reason "config item $name: expect=$expect but real=$real" exit $OCF_ERR_CONFIGURED fi return $OCF_SUCCESS } lvmlockd_check() { config_verify "global/use_lvmlockd" "1" - config_verify "global/locking_type" "1" + + # locking_type was removed from config in v2.03 + ocf_version_cmp "$(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 exit $OCF_NOT_RUNNING fi ocf_exit_reason "lvmlockd daemon is not running!" exit $OCF_ERR_CONFIGURED fi if pgrep clvmd >/dev/null 2>&1 ; then ocf_exit_reason "clvmd daemon is running unexpectedly." exit $OCF_ERR_CONFIGURED 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_CONFIGURED fi if pgrep lvmetad >/dev/null 2>&1 ; then ocf_exit_reason "Please stop lvmetad daemon when clvmd is running." exit $OCF_ERR_CONFIGURED fi if pgrep lvmlockd >/dev/null 2>&1 ; then ocf_exit_reason "lvmlockd daemon is running unexpectedly." exit $OCF_ERR_CONFIGURED fi return $OCF_SUCCESS } systemid_check() { local source # system_id_source is set in lvm.conf source=$(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_CONFIGURED fi if [ -z ${SYSTEM_ID} ]; then ocf_exit_reason "local/system_id is not set!" exit $OCF_ERR_CONFIGURED 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 $OCF_ERR_CONFIGURED fi # Our tag must _NOT_ be in the volume_list. This agent # overrides the volume_list during activation using the # special tag reserved for cluster activation if lvm dumpconfig activation/volume_list | grep -e "\"@${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_CONFIGURED fi return $OCF_SUCCESS } read_parameters() { if [ -z "$VG" ] then ocf_exit_reason "You must identify the volume group name!" exit $OCF_ERR_ARGS fi if [ "$LV_activation_mode" != "shared" ] && [ "$LV_activation_mode" != "exclusive" ] then ocf_exit_reason "Invalid value for activation_mode: $LV_activation_mode" exit $OCF_ERR_ARGS 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() { local lv_count local mode read_parameters check_binary pgrep # Every LVM command is just symlink to lvm binary check_binary lvm check_binary dmsetup 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 $OCF_SUCCESS fi ocf_exit_reason "Volume group[${VG}] doesn't exist, or not visible on this node!" exit $OCF_ERR_CONFIGURED fi # Get the access mode from VG metadata and check if it matches the input # value. Skip to check "tagging" mode because there's no reliable way to # automatically check if "tagging" mode is being used. get_VG_access_mode_num mode=$? 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 $OCF_ERR_ARGS fi # Nothing to do if the VG has no logical volume lv_count=$(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 OUT=$(lvs --foreign --noheadings ${VG}/${LV} 2>&1) if [ $? -ne 0 ]; then ocf_log err "lvs: ${OUT}" ocf_exit_reason "LV ($LV) is not in the given VG ($VG)." exit $OCF_ERR_ARGS 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() { local activate_opt=$1 # Only activate the specific LV if it's given if [ -n "$LV" ]; then ocf_run lvchange $activate_opt ${VG}/${LV} if [ $? -ne $OCF_SUCCESS ]; then return $OCF_ERR_GENERIC fi else ocf_run lvchange $activate_opt ${VG} if [ $? -ne $OCF_SUCCESS ]; then return $OCF_ERR_GENERIC fi fi return $OCF_SUCCESS } lvmlockd_activate() { # activation opt local activate_opt if [ "$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_log err "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 $OCF_SUCCESS } # clvmd must be running to activate clustered VG clvmd_activate() { local activate_opt if [ "$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 $OCF_SUCCESS } systemid_activate() { local cur_systemid pvscan --cache cur_systemid=$(vgs --foreign --noheadings -o systemid ${VG} | tr -d '[:blank:]') # Put our system ID on the VG vgchange -y --config "local/extra_system_ids=[\"${cur_systemid}\"]" \ --systemid ${SYSTEM_ID} ${VG} do_activate "-ay" if [ $? -ne $OCF_SUCCESS ]; then return $OCF_ERR_GENERIC fi return $OCF_SUCCESS } tagging_activate() { if ! set_tags ; then ocf_log err "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_log err "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_log err "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 "vgname=${VG}" 2>/dev/null | grep -c "${VG}-") # test $lv_count -eq $dm_count # # It works, but we cannot afford to use LVM command in lvm_status. LVM command is expensive # because it may potencially scan all disks on the system, update the metadata even using # lvs/vgs when the metadata is somehow inconsistent. # # So, we have to make compromise that the VG is assumably active if any LV of the VG is active. # # Paul: # VGS + LVS with "-" in their name get mangled with double dashes in dmsetup. # Switching to wc and just counting lines while depending on the vgname + lvname filter # in dmsetup gets around the issue with dmsetup reporting correctly but grep failing. # # Logic for both test cases and dmsetup calls changed so they match too. # # This is AllBad but there isn't a better way that I'm aware of yet. lvm_status() { local dm_count if [ -n "${LV}" ]; then # dmsetup ls? It cannot accept device name. It's # too heavy to list all DM devices. dm_count=$(dmsetup info --noheadings --noflush -c -S "vgname=${VG} && lvname=${LV}" | grep -c -v '^No devices found') else dm_count=$(dmsetup info --noheadings --noflush -c -S "vgname=${VG}" | grep -c -v '^No devices found') fi if [ $dm_count -eq 0 ]; then return $OCF_NOT_RUNNING fi return $OCF_SUCCESS } lvm_start() { local rc local vol if lvm_status ; then ocf_log info "${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() { local vol [ -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 ${VG_access_mode_num} in 1) lvmlockd_deactivate ;; 2) clvmd_deactivate ;; 3) systemid_deactivate ;; 4) tagging_deactivate ;; *) ocf_log err "VG [${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_validate 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