conf-sanity.sh

lustre 1.6.5 source code

	# With a new good MDT failover nid, we should be able to mount a client
	# (but it cant talk to OST)
        local OLDMOUNTOPT=$MOUNTOPT
        MOUNTOPT="exclude=lustre-OST0000"
	mount_client $MOUNT
        MOUNTOPT=$OLDMOUNTOPT
	set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "lustre-MDT0000.mdc.max_rpcs_in_flight" ||
		return 11

	zconf_umount `hostname` $MOUNT -f
	cleanup_nocli
	load_modules

        # mount a second time to make sure we didnt leave upgrade flag on
	load_modules
        $TUNEFS --dryrun $TMP/$tdir/mds || error "tunefs failed"
	load_modules
        start mds $TMP/$tdir/mds "-o loop,exclude=lustre-OST0000" || return 12
        cleanup_nocli

	[ -d $TMP/$tdir ] && { rm -rf $TMP/$tdir || true; }	# true is only for TMP on NFS
}
run_test 32a "Upgrade from 1.4 (not live)"

test_32b() {
        # XXX - make this run on client-only systems with real hardware on
        #       the OST and MDT
        #       there appears to be a lot of assumption here about loopback
        #       devices
        # or maybe this test is just totally useless on a client-only system
        [ "$mds_HOST" = "`hostname`" ] || { skip "remote MDS" && return 0; }
        [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
		{ skip "remote OST" && return 0; }

        [ -z "$TUNEFS" ] && skip "No tunefs" && return
	local DISK1_4=$LUSTRE/tests/disk1_4.zip
        [ ! -r $DISK1_4 ] && skip "Cant find $DISK1_4, skipping" && return
	mkdir -p $TMP/$tdir
	unzip -o -j -d $TMP/$tdir $DISK1_4 || { skip "Cant unzip $DISK1_4, skipping" && return ; }
	load_modules
	sysctl lnet.debug=$PTLDEBUG
	NEWNAME=sofia

	# writeconf will cause servers to register with their current nids
	$TUNEFS --writeconf --fsname=$NEWNAME $TMP/$tdir/mds || error "tunefs failed"
	start mds $TMP/$tdir/mds "-o loop" || return 3
        local UUID=$(lctl get_param -n mds.${NEWNAME}-MDT0000.uuid)
	echo MDS uuid $UUID
	[ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID" 

	$TUNEFS --mgsnode=`hostname` --fsname=$NEWNAME --writeconf $TMP/$tdir/ost1 || error "tunefs failed"
	start ost1 $TMP/$tdir/ost1 "-o loop" || return 5
        UUID=$(lctl get_param -n obdfilter.${NEWNAME}-OST0000.uuid)
	echo OST uuid $UUID
	[ "$UUID" == "ost1_UUID" ] || error "UUID is wrong: $UUID"

	echo "OSC changes should succeed:" 
	$LCTL conf_param ${NEWNAME}-OST0000.osc.max_dirty_mb=15 || return 7
	$LCTL conf_param ${NEWNAME}-OST0000.failover.node=$NID || return 8
	echo "ok."
	echo "MDC changes should succeed:" 
	$LCTL conf_param ${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight=9 || return 9
	echo "ok."

	# MDT and OST should have registered with new nids, so we should have
	# a fully-functioning client
	echo "Check client and old fs contents"
	OLDFS=$FSNAME
	FSNAME=$NEWNAME
	mount_client $MOUNT
	FSNAME=$OLDFS
	set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight" || return 11
	[ "$(cksum $MOUNT/passwd | cut -d' ' -f 1,2)" == "2479747619 779" ] || return 12  
	echo "ok."

	cleanup
	[ -d $TMP/$tdir ] && { rm -rf $TMP/$tdir || true; }	# true is only for TMP on NFS
}
run_test 32b "Upgrade from 1.4 with writeconf"

test_33a() { # bug 12333, was test_33
        local rc=0
        local FSNAME2=test-123
        [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST

        if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then
                do_facet mds [ -b "$MDSDEV" ] && \
                skip "mixed loopback and real device not working" && return
        fi

        local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
        local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
        add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10
        add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --index=8191 --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10

        start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT
        start fs2ost $fs2ostdev $OST_MOUNT_OPTS
        do_facet mds "$LCTL conf_param $FSNAME2.sys.timeout=200" || rc=1
        mkdir -p $MOUNT2
        mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || rc=2
        echo "ok."

        umount -d $MOUNT2
        stop fs2ost -f
        stop fs2mds -f
        rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev
        cleanup_nocli || rc=6
        return $rc
}
run_test 33a "Mount ost with a large index number"

test_33b() {	# was test_33a
        setup

        do_facet client dd if=/dev/zero of=$MOUNT/24 bs=1024k count=1
        # Drop lock cancelation reply during umount
	#define OBD_FAIL_LDLM_CANCEL             0x304
        do_facet client sysctl -w lustre.fail_loc=0x80000304
        #sysctl -w lnet.debug=-1
        umount_client $MOUNT
        cleanup
}
run_test 33b "Drop cancel during umount"

test_34a() {
        setup
	do_facet client "sh runmultiop_bg_pause $DIR/file O_c"
	manual_umount_client
	rc=$?
	do_facet client killall -USR1 multiop
	if [ $rc -eq 0 ]; then
		error "umount not fail!"
	fi
	sleep 1
        cleanup
}
run_test 34a "umount with opened file should be fail"


test_34b() {
	setup
	touch $DIR/$tfile || return 1
	stop_mds --force || return 2

 	manual_umount_client --force
	rc=$?
	if [ $rc -ne 0 ]; then
		error "mtab after failed umount - rc $rc"
	fi

	cleanup
	return 0	
}
run_test 34b "force umount with failed mds should be normal"

test_34c() {
	setup
	touch $DIR/$tfile || return 1
	stop_ost --force || return 2

 	manual_umount_client --force
	rc=$?
	if [ $rc -ne 0 ]; then
		error "mtab after failed umount - rc $rc"
	fi

	cleanup
	return 0	
}
run_test 34c "force umount with failed ost should be normal"

test_35() { # bug 12459
	setup

	debugsave
	sysctl -w lnet.debug="ha"

	log "Set up a fake failnode for the MDS"
	FAKENID="127.0.0.2"
	do_facet mds $LCTL conf_param ${FSNAME}-MDT0000.failover.node=$FAKENID || return 4

	log "Wait for RECONNECT_INTERVAL seconds (10s)"
	sleep 10

	MSG="conf-sanity.sh test_35 `date +%F%kh%Mm%Ss`"
	$LCTL clear
	log "$MSG"
	log "Stopping the MDT:"
	stop_mds || return 5

	df $MOUNT > /dev/null 2>&1 &
	DFPID=$!
	log "Restarting the MDT:"
	start_mds || return 6
	log "Wait for df ($DFPID) ... "
	wait $DFPID
	log "done"
	debugrestore

	# retrieve from the log the first server that the client tried to
	# contact after the connection loss
	$LCTL dk $TMP/lustre-log-$TESTNAME.log
	NEXTCONN=`awk "/${MSG}/ {start = 1;}
		       /import_select_connection.*${FSNAME}-MDT0000-mdc.* using connection/ {
				if (start) {
					if (\\\$NF ~ /$FAKENID/)
						print \\\$NF;
					else
						print 0;
					exit;
				}
		       }" $TMP/lustre-log-$TESTNAME.log`
	[ "$NEXTCONN" != "0" ] && log "The client didn't try to reconnect to the last active server (tried ${NEXTCONN} instead)" && return 7
	cleanup
}
run_test 35 "Reconnect to the last active server first"

test_36() { # 12743
        local rc
        local FSNAME2=test1234
        local fs3ost_HOST=$ost_HOST

        [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST && fs3ost_HOST=$ost1_HOST
        rc=0

        if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" -o -z "$fs3ost_DEV" ]; then
		do_facet mds [ -b "$MDSDEV" ] && \
		skip "mixed loopback and real device not working" && return
        fi
        [ $OSTCOUNT -lt 2 ] && skip "skipping test for single OST" && return

	[ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
		{ skip "remote OST" && return 0; }

        local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
        local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
        local fs3ostdev=${fs3ost_DEV:-$(ostdevname 2)_2}
        add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10
        # XXX after we support non 4K disk blocksize, change following --mkfsoptions with
        # other argument
        add fs2ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10
        add fs3ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=$MGSNID --reformat $fs3ostdev || exit 10

        start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS
        start fs2ost $fs2ostdev $OST_MOUNT_OPTS
        start fs3ost $fs3ostdev $OST_MOUNT_OPTS
        mkdir -p $MOUNT2
        mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1

        sleep 5 # until 11778 fixed

        dd if=/dev/zero of=$MOUNT2/$tfile bs=1M count=7 || return 2

        BKTOTAL=`lctl get_param -n obdfilter.*.kbytestotal | awk 'BEGIN{total=0}; {total+=$1}; END{print total}'`
        BKFREE=`lctl get_param -n obdfilter.*.kbytesfree | awk 'BEGIN{free=0}; {free+=$1}; END{print free}'`
        BKAVAIL=`lctl get_param -n obdfilter.*.kbytesavail | awk 'BEGIN{avail=0}; {avail+=$1}; END{print avail}'`
        STRING=`df -P $MOUNT2 | tail -n 1 | awk '{print $2","$3","$4}'`
        DFTOTAL=`echo $STRING | cut -d, -f1`
        DFUSED=`echo $STRING  | cut -d, -f2`
        DFAVAIL=`echo $STRING | cut -d, -f3`
        DFFREE=$(($DFTOTAL - $DFUSED))

        ALLOWANCE=$((64 * $OSTCOUNT))

        if [ $DFTOTAL -lt $(($BKTOTAL - $ALLOWANCE)) ] ||  
           [ $DFTOTAL -gt $(($BKTOTAL + $ALLOWANCE)) ] ; then
                echo "**** FAIL: df total($DFTOTAL) mismatch OST total($BKTOTAL)"
                rc=1
        fi
        if [ $DFFREE -lt $(($BKFREE - $ALLOWANCE)) ] || 
           [ $DFFREE -gt $(($BKFREE + $ALLOWANCE)) ] ; then
                echo "**** FAIL: df free($DFFREE) mismatch OST free($BKFREE)"
                rc=2
        fi
        if [ $DFAVAIL -lt $(($BKAVAIL - $ALLOWANCE)) ] || 
           [ $DFAVAIL -gt $(($BKAVAIL + $ALLOWANCE)) ] ; then
                echo "**** FAIL: df avail($DFAVAIL) mismatch OST avail($BKAVAIL)"
                rc=3
       fi

        umount -d $MOUNT2
        stop fs3ost -f || return 200
        stop fs2ost -f || return 201
        stop fs2mds -f || return 202
        rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev $fs3ostdev
        unload_modules || return 203
        return $rc
}
run_test 36 "df report consistency on OSTs with different block size"

test_37() {
	[ -n "$CLIENTONLY" -o -n "$CLIENTMODSONLY" ] && skip "client only testing" && return 0
	LOCAL_MDSDEV="$TMP/mdt.img"
	SYM_MDSDEV="$TMP/sym_mdt.img"

	echo "MDS :     $LOCAL_MDSDEV"
	echo "SYMLINK : $SYM_MDSDEV"
	rm -f $LOCAL_MDSDEV

	touch $LOCAL_MDSDEV
	mkfs.lustre --reformat --fsname=lustre --mdt --mgs --device-size=9000 $LOCAL_MDSDEV ||
		error "mkfs.lustre $LOCAL_MDSDEV failed"
	ln -s $LOCAL_MDSDEV $SYM_MDSDEV

	echo "mount symlink device - $SYM_MDSDEV"

	mount_op=`mount -v -t lustre -o loop $SYM_MDSDEV ${MOUNT%/*}/mds 2>&1 | grep "unable to set tunable"`
	umount -d ${MOUNT%/*}/mds
	rm -f $LOCAL_MDSDEV $SYM_MDSDEV

	if [ -n "$mount_op" ]; then
		error "**** FAIL: set tunables failed for symlink device"
	fi
	return 0
}
run_test 37 "verify set tunables works for symlink device"

test_38() { # bug 14222
	setup
	# like runtests
	COUNT=10
	SRC="/etc /bin"
	FILES=`find $SRC -type f -mtime +1 | head -n $COUNT`
	log "copying $(echo $FILES | wc -w) files to $DIR/$tdir"
	mkdir -p $DIR/$tdir
	tar cf - $FILES | tar xf - -C $DIR/$tdir || \
		error "copying $SRC to $DIR/$tdir"
	sync
	umount_client $MOUNT
	stop_mds
	log "rename lov_objid file on MDS"
	rm -f $TMP/lov_objid.orig
	do_facet mds "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.orig\\\" $MDSDEV"
	do_facet mds "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV"

	do_facet mds "od -Ax -td8 $TMP/lov_objid.orig"
	# check create in mds_lov_connect
	start_mds
	mount_client $MOUNT
	for f in $FILES; do
		[ $V ] && log "verifying $DIR/$tdir/$f"
		diff -q $f $DIR/$tdir/$f || ERROR=y
	done
	do_facet mds "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new\\\"  $MDSDEV"
	do_facet mds "od -Ax -td8 $TMP/lov_objid.new"
	[ "$ERROR" = "y" ] && error "old and new files are different after connect" || true
	
	
	# check it's updates in sync
	umount_client $MOUNT
	stop_mds
	
	do_facet mds dd if=/dev/zero of=$TMP/lov_objid.clear bs=4096 count=1
	do_facet mds "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV"
	do_facet mds "debugfs -w -R \\\"write $TMP/lov_objid.clear lov_objid\\\" $MDSDEV "

	start_mds
	mount_client $MOUNT
	for f in $FILES; do
		[ $V ] && log "verifying $DIR/$tdir/$f"
		diff -q $f $DIR/$tdir/$f || ERROR=y
	done
        do_facet mds "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new1\\\" $MDSDEV"
	do_facet mds "od -Ax -td8 $TMP/lov_objid.new1"
	umount_client $MOUNT
	stop_mds
	[ "$ERROR" = "y" ] && error "old and new files are different after sync" || true
	
	log "files compared the same"
	cleanup
}
run_test 38 "MDS recreates missing lov_objid file from OST data"

test_39() { #bug 14413
        PTLDEBUG=+malloc
        setup
        cleanup
        perl $SRCDIR/leak_finder.pl $TMP/debug 2>&1 | egrep '*** Leak:' && 
                error "memory leak detected" || true
}
run_test 39 "leak_finder recognizes both LUSTRE and LNET malloc messages"

test_40() { # bug 15759
	start_ost
	#define OBD_FAIL_TGT_TOOMANY_THREADS     0x706
	do_facet mds "sysctl -w lustre.fail_loc=0x80000706"
	start_mds
	cleanup
}
run_test 40 "race during service thread startup"

equals_msg `basename $0`: test complete
[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true