conf-sanity.sh

lustre 1.6.5 source code

#!/bin/bash
# requirement:
#	add uml1 uml2 uml3 in your /etc/hosts

# FIXME - there is no reason to use all of these different
#   return codes, espcially when most of them are mapped to something
#   else anyway.  The combination of test number and return code
#   figure out what failed.

set -e

ONLY=${ONLY:-"$*"}

# These tests don't apply to mountconf
#              xml xml xml xml xml xml dumb
MOUNTCONFSKIP="10  11  12  13  13b 14  15 "

# bug number for skipped test:                     13369
ALWAYS_EXCEPT=" $CONF_SANITY_EXCEPT $MOUNTCONFSKIP 34a"
# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!

SRCDIR=`dirname $0`
PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH

SAVE_PWD=$PWD
LUSTRE=${LUSTRE:-`dirname $0`/..}
RLUSTRE=${RLUSTRE:-$LUSTRE}
HOSTNAME=`hostname`

. $LUSTRE/tests/test-framework.sh
init_test_env $@
# use small MDS + OST size to speed formatting time
MDSSIZE=40000
OSTSIZE=40000
. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}

#
[ "$SLOW" = "no" ] && EXCEPT_SLOW="0 1 2 3 6 7 15 18 24b 25 30 31 32 33 34a "

reformat() {
        formatall
}

writeconf() {
    local facet=mds
    shift
    stop ${facet} -f
    rm -f ${facet}active
    # who knows if/where $TUNEFS is installed?  Better reformat if it fails...
    do_facet ${facet} "$TUNEFS --writeconf $MDSDEV" || echo "tunefs failed, reformatting instead" && reformat
}

gen_config() {
        reformat
        # The MGS must be started before the OSTs for a new fs, so start
        # and stop to generate the startup logs. 
	start_mds
	start_ost
	sleep 5
	stop_ost
	stop_mds
}

start_mds() {
	echo "start mds service on `facet_active_host mds`"
	start mds $MDSDEV $MDS_MOUNT_OPTS || return 94
}

stop_mds() {
	echo "stop mds service on `facet_active_host mds`"
	# These tests all use non-failover stop
	stop mds -f  || return 97
}

start_ost() {
	echo "start ost1 service on `facet_active_host ost1`"
	start ost1 `ostdevname 1` $OST_MOUNT_OPTS || return 95
}

stop_ost() {
	echo "stop ost1 service on `facet_active_host ost1`"
	# These tests all use non-failover stop
	stop ost1 -f  || return 98
}

start_ost2() {
	echo "start ost2 service on `facet_active_host ost2`"
	start ost2 `ostdevname 2` $OST_MOUNT_OPTS || return 92
}

stop_ost2() {
	echo "stop ost2 service on `facet_active_host ost2`"
	# These tests all use non-failover stop
	stop ost2 -f  || return 93
}

start_client() {
	echo "start client on `facet_active_host client`"
	start client || return 99 
}

stop_client() {
	echo "stop client on `facet_active_host client`"
	stop client || return 100 
}

mount_client() {
	local MOUNTPATH=$1
	echo "mount $FSNAME on ${MOUNTPATH}....."
	zconf_mount `hostname` $MOUNTPATH  || return 96
}

remount_client() {
	local SAVEMOUNTOPT=$MOUNTOPT
	MOUNTOPT="remount,$1"
	local MOUNTPATH=$2
	echo "remount '$1' lustre on ${MOUNTPATH}....."
	zconf_mount `hostname`  $MOUNTPATH  || return 96
	MOUNTOPT=$SAVEMOUNTOPT
}

umount_client() {
	local MOUNTPATH=$1
	echo "umount lustre on ${MOUNTPATH}....."
	zconf_umount `hostname` $MOUNTPATH || return 97
}

manual_umount_client(){
	local rc
	local FORCE=$1
	echo "manual umount lustre on ${MOUNT}...."
	do_facet client "umount -d ${FORCE} $MOUNT"
	rc=$?
	return $rc
}

setup() {
	start_ost
	start_mds
	mount_client $MOUNT
}

cleanup_nocli() {
	stop_mds || return 201
	stop_ost || return 202
	unload_modules || return 203
}

cleanup() {
 	umount_client $MOUNT || return 200
	cleanup_nocli || return $?
}

check_mount() {
	do_facet client "cp /etc/passwd $DIR/a" || return 71
	do_facet client "rm $DIR/a" || return 72
	# make sure lustre is actually mounted (touch will block, 
        # but grep won't, so do it after) 
        do_facet client "grep $MOUNT' ' /proc/mounts > /dev/null" || return 73
	echo "setup single mount lustre success"
}

check_mount2() {
	do_facet client "touch $DIR/a" || return 71	
	do_facet client "rm $DIR/a" || return 72	
	do_facet client "touch $DIR2/a" || return 73	
	do_facet client "rm $DIR2/a" || return 74	
	echo "setup double mount lustre success"
}

build_test_filter

if [ "$ONLY" == "setup" ]; then
	setup
	exit
fi

if [ "$ONLY" == "cleanup" ]; then
	cleanup
	exit
fi

#create single point mountpoint

gen_config


test_0() {
        setup
	check_mount || return 41
	cleanup || return $?
}
run_test 0 "single mount setup"

test_1() {
	start_ost
	echo "start ost second time..."
	setup
	check_mount || return 42
	cleanup || return $?
}
run_test 1 "start up ost twice (should return errors)"

test_2() {
	start_ost
	start_mds	
	echo "start mds second time.."
	start_mds
	mount_client $MOUNT
	check_mount || return 43
	cleanup || return $?
}
run_test 2 "start up mds twice (should return err)"

test_3() {
	setup
	#mount.lustre returns an error if already in mtab
	mount_client $MOUNT && return $?
	check_mount || return 44
	cleanup || return $?
}
run_test 3 "mount client twice (should return err)"

test_4() {
	setup
	touch $DIR/$tfile || return 85
	stop_ost -f
	cleanup
	eno=$?
	# ok for ost to fail shutdown
	if [ 202 -ne $eno ]; then
		return $eno;
	fi
	return 0
}
run_test 4 "force cleanup ost, then cleanup"

test_5a() {	# was test_5
	setup
	touch $DIR/$tfile || return 1
	fuser -m -v $MOUNT && echo "$MOUNT is in use by user space process."

	stop_mds -f || return 2

	# cleanup may return an error from the failed
	# disconnects; for now I'll consider this successful
	# if all the modules have unloaded.
 	umount -d $MOUNT &
	UMOUNT_PID=$!
	sleep 6
	echo "killing umount"
	kill -TERM $UMOUNT_PID
	echo "waiting for umount to finish"
	wait $UMOUNT_PID
	if grep " $MOUNT " /proc/mounts; then
		echo "test 5: /proc/mounts after failed umount"
		umount $MOUNT &
		UMOUNT_PID=$!
		sleep 2
		echo "killing umount"
		kill -TERM $UMOUNT_PID
		echo "waiting for umount to finish"
		wait $UMOUNT_PID
		grep " $MOUNT " /proc/mounts && echo "test 5: /proc/mounts after second umount" && return 11
	fi

	manual_umount_client
	# stop_mds is a no-op here, and should not fail
	cleanup_nocli || return $?
	# df may have lingering entry
	manual_umount_client
	# mtab may have lingering entry
	local WAIT=0
	local MAX_WAIT=20
	local sleep=1
	while [ "$WAIT" -ne "$MAX_WAIT" ]; do
		sleep $sleep
		grep -q $MOUNT" " /etc/mtab || break
        	echo "Waiting /etc/mtab updated ... "
		WAIT=$(( WAIT + sleep))
	done
	[ "$WAIT" -eq "$MAX_WAIT" ] && error "/etc/mtab is not updated in $WAIT secs"
	echo "/etc/mtab updated in $WAIT secs"
}
run_test 5a "force cleanup mds, then cleanup"

test_5b() {
	start_ost
	[ -d $MOUNT ] || mkdir -p $MOUNT
	grep " $MOUNT " /etc/mtab && echo "test 5b: mtab before mount" && return 10
	mount_client $MOUNT && return 1
	grep " $MOUNT " /etc/mtab && echo "test 5b: mtab after failed mount" && return 11
	umount_client $MOUNT	
	# stop_mds is a no-op here, and should not fail
	cleanup_nocli || return $?
	return 0
}
run_test 5b "mds down, cleanup after failed mount (bug 2712) (should return errs)"

test_5c() {
	start_ost
	start_mds
	[ -d $MOUNT ] || mkdir -p $MOUNT
	grep " $MOUNT " /etc/mtab && echo "test 5c: mtab before mount" && return 10
	local oldfs="${FSNAME}"
	FSNAME="wrong.${FSNAME}"
	mount_client $MOUNT || :
	FSNAME=${oldfs}
	grep " $MOUNT " /etc/mtab && echo "test 5c: mtab after failed mount" && return 11
	umount_client $MOUNT
	cleanup_nocli  || return $?
}
run_test 5c "cleanup after failed mount (bug 2712) (should return errs)"

test_5d() {
	start_ost
	start_mds
	stop_ost -f
	grep " $MOUNT " /etc/mtab && echo "test 5d: mtab before mount" && return 10
	mount_client $MOUNT || return 1
	cleanup  || return $?
	grep " $MOUNT " /etc/mtab && echo "test 5d: mtab after unmount" && return 11
	return 0
}
run_test 5d "mount with ost down"

test_5e() {
	start_ost
	start_mds

#define OBD_FAIL_PTLRPC_DELAY_SEND       0x506
	do_facet client "sysctl -w lustre.fail_loc=0x80000506"
	grep " $MOUNT " /etc/mtab && echo "test 5e: mtab before mount" && return 10
	mount_client $MOUNT || echo "mount failed (not fatal)"
	cleanup  || return $?
	grep " $MOUNT " /etc/mtab && echo "test 5e: mtab after unmount" && return 11
	return 0
}
run_test 5e "delayed connect, don't crash (bug 10268)"

test_6() {
	setup
	manual_umount_client
	mount_client ${MOUNT} || return 87
	touch $DIR/a || return 86
	cleanup  || return $?
}
run_test 6 "manual umount, then mount again"

test_7() {
	setup
	manual_umount_client
	cleanup_nocli || return $?
}
run_test 7 "manual umount, then cleanup"

test_8() {
	setup
	mount_client $MOUNT2
	check_mount2 || return 45
	umount_client $MOUNT2
	cleanup  || return $?
}
run_test 8 "double mount setup"

test_9() {
        start_ost

	do_facet ost1 sysctl lnet.debug=\'inode trace\' || return 1
	do_facet ost1 sysctl lnet.subsystem_debug=\'mds ost\' || return 1

        CHECK_PTLDEBUG="`do_facet ost1 sysctl -n lnet.debug`"
        if [ "$CHECK_PTLDEBUG" ] && [ "$CHECK_PTLDEBUG" = "trace inode" ];then
           echo "lnet.debug success"
        else
           echo "lnet.debug: want 'trace inode', have '$CHECK_PTLDEBUG'"
           return 1
        fi
        CHECK_SUBSYS="`do_facet ost1 sysctl -n lnet.subsystem_debug`"
        if [ "$CHECK_SUBSYS" ] && [ "$CHECK_SUBSYS" = "mds ost" ]; then
           echo "lnet.subsystem_debug success"
        else
           echo "lnet.subsystem_debug: want 'mds ost', have '$CHECK_SUBSYS'"
           return 1
        fi
        stop_ost || return $?
}

run_test 9 "test ptldebug and subsystem for mkfs"

test_10() {
        echo "generate configuration with the same name for node and mds"
        OLDXMLCONFIG=$XMLCONFIG
        XMLCONFIG="broken.xml"
        [ -f "$XMLCONFIG" ] && rm -f $XMLCONFIG