X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Ftest-framework.sh;h=acaf0e86889cc2da5c2db8275cfb7f0bb7e79041;hp=743d017c5931ca57ca8752bcc5e16f5612bd0c16;hb=a6e6e1557180f89f0360e3cb9b9e32b465114b5f;hpb=77aa3f2e38e94d1ac83afa2f7ef18a1ca9de92af

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 743d017..acaf0e8 100644
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -185,6 +185,7 @@ init_test_env() {
 	export ZFS=${ZFS:-zfs}
 	export ZPOOL=${ZPOOL:-zpool}
 	export ZDB=${ZDB:-zdb}
+	export PARTPROBE=${PARTPROBE:-partprobe}
 
     #[ -d /r ] && export ROOT=${ROOT:-/r}
     export TMP=${TMP:-$ROOT/tmp}
@@ -297,6 +298,7 @@ init_test_env() {
     fi
 
     export SHUTDOWN_ATTEMPTS=${SHUTDOWN_ATTEMPTS:-3}
+	export OSD_TRACK_DECLARES_LBUG=${OSD_TRACK_DECLARES_LBUG:-"yes"}
 
     # command line
 
@@ -490,6 +492,7 @@ load_modules_local() {
         grep -q crc16 $SYMLIST || { modprobe crc16 2>/dev/null || true; }
         grep -q -w jbd $SYMLIST || { modprobe jbd 2>/dev/null || true; }
         grep -q -w jbd2 $SYMLIST || { modprobe jbd2 2>/dev/null || true; }
+		load_module lfsck/lfsck
 		[ "$LQUOTA" != "no" ] && load_module quota/lquota $LQUOTAOPTS
 		if [[ $(node_fstypes $HOSTNAME) == *zfs* ]]; then
 			modprobe zfs
@@ -591,7 +594,7 @@ fs_log_size() {
 	local size=0
 	case $fstype in
 		ldiskfs) size=50;; # largest seen is 44, leave some headroom
-		zfs)     size=256;;
+		zfs)     size=400;; # largest seen is 384
 	esac
 
 	echo -n $size
@@ -808,6 +811,140 @@ ostdevlabel() {
 }
 
 #
+# Get the device of a facet.
+#
+facet_device() {
+	local facet=$1
+	local device
+
+	case $facet in
+		mgs) device=$(mgsdevname) ;;
+		mds*) device=$(mdsdevname $(facet_number $facet)) ;;
+		ost*) device=$(ostdevname $(facet_number $facet)) ;;
+		fs2mds) device=$(mdsdevname 1_2) ;;
+		fs2ost) device=$(ostdevname 1_2) ;;
+		fs3ost) device=$(ostdevname 2_2) ;;
+		*) ;;
+	esac
+
+	echo -n $device
+}
+
+#
+# Get the virtual device of a facet.
+#
+facet_vdevice() {
+	local facet=$1
+	local device
+
+	case $facet in
+		mgs) device=$(mgsvdevname) ;;
+		mds*) device=$(mdsvdevname $(facet_number $facet)) ;;
+		ost*) device=$(ostvdevname $(facet_number $facet)) ;;
+		fs2mds) device=$(mdsvdevname 1_2) ;;
+		fs2ost) device=$(ostvdevname 1_2) ;;
+		fs3ost) device=$(ostvdevname 2_2) ;;
+		*) ;;
+	esac
+
+	echo -n $device
+}
+
+#
+# Re-read the partition table on failover partner host.
+# After a ZFS storage pool is created on a shared device, the partition table
+# on the device may change. However, the operating system on the failover
+# host may not notice the change automatically. Without the up-to-date partition
+# block devices, 'zpool import ..' cannot find the labels, whose positions are
+# relative to partition rather than disk beginnings.
+#
+# This function performs partprobe on the failover host to make it re-read the
+# partition table.
+#
+refresh_partition_table() {
+	local facet=$1
+	local device=$2
+	local host
+
+	host=$(facet_passive_host $facet)
+	if [[ -n "$host" ]]; then
+		do_node $host "$PARTPROBE $device"
+	fi
+}
+
+#
+# Get ZFS storage pool name.
+#
+zpool_name() {
+	local facet=$1
+	local device
+	local poolname
+
+	device=$(facet_device $facet)
+	# poolname is string before "/"
+	poolname="${device%%/*}"
+
+	echo -n $poolname
+}
+
+#
+# Export ZFS storage pool.
+# Before exporting the pool, all datasets within the pool should be unmounted.
+#
+export_zpool() {
+	local facet=$1
+	shift
+	local opts="$@"
+	local poolname
+
+	poolname=$(zpool_name $facet)
+
+	if [[ -n "$poolname" ]]; then
+		do_facet $facet "! $ZPOOL list -H $poolname >/dev/null 2>&1 ||
+			grep -q ^$poolname/ /proc/mounts ||
+			$ZPOOL export $opts $poolname"
+	fi
+}
+
+#
+# Import ZFS storage pool.
+# Force importing, even if the pool appears to be potentially active.
+#
+import_zpool() {
+	local facet=$1
+	shift
+	local opts=${@:-"-o cachefile=none"}
+	local poolname
+
+	poolname=$(zpool_name $facet)
+
+	if [[ -n "$poolname" ]]; then
+		opts+=" -d $(dirname $(facet_vdevice $facet))"
+		do_facet $facet "$ZPOOL list -H $poolname >/dev/null 2>&1 ||
+			$ZPOOL import -f $opts $poolname"
+	fi
+}
+
+#
+# Set the "cachefile=none" property on ZFS storage pool so that the pool
+# is not automatically imported on system startup.
+#
+# In a failover environment, this will provide resource level fencing which
+# will ensure that the same ZFS storage pool will not be imported concurrently
+# on different nodes.
+#
+disable_zpool_cache() {
+	local facet=$1
+	local poolname
+
+	poolname=$(zpool_name $facet)
+
+	if [[ -n "$poolname" ]]; then
+		do_facet $facet "$ZPOOL set cachefile=none $poolname"
+	fi
+}
+
+#
 # This and set_osd_param() shall be used to access OSD parameters
 # once existed under "obdfilter":
 #
@@ -935,6 +1072,11 @@ mount_facet() {
 		opts=$(csa_add "$opts" -o loop)
 	fi
 
+	if [[ $(facet_fstype $facet) == zfs ]]; then
+		# import ZFS storage pool
+		import_zpool $facet || return ${PIPESTATUS[0]}
+	fi
+
 	echo "Starting ${facet}: $opts ${!dev} $mntpt"
 	# for testing LU-482 error handling in mount_facets() and test_0a()
 	if [ -f $TMP/test-lu482-trigger ]; then
@@ -981,39 +1123,6 @@ start() {
     return $RC
 }
 
-#
-# When a ZFS OSD is made read-only by replay_barrier(), its pool is "freezed".
-# Because stopping corresponding target may not clear this in-memory state, we
-# need to zap the pool from memory by exporting and reimporting the pool.
-#
-# Although the uberblocks are not updated when a pool is freezed, transactions
-# are still written to the disks.  Modified blocks may be cached in memory when
-# tests try reading them back.  The export-and-reimport process also evicts any
-# cached pool data from memory to provide the correct "data loss" semantics.
-#
-refresh_disk() {
-	local facet=$1
-	local fstype=$(facet_fstype $facet)
-	local _dev
-	local dev
-	local poolname
-
-	if [ "${fstype}" == "zfs" ]; then
-		_dev=$(facet_active $facet)_dev
-		dev=${!_dev} # expand _dev to its value, e.g. ${mds1_dev}
-		poolname="${dev%%/*}" # poolname is string before "/"
-
-		if [ "${poolname}" == "" ]; then
-			echo "invalid dataset name: $dev"
-			return
-		fi
-		do_facet $facet "cp /etc/zfs/zpool.cache /tmp/zpool.cache.back"
-		do_facet $facet "$ZPOOL export ${poolname}"
-		do_facet $facet "$ZPOOL import -f -c /tmp/zpool.cache.back \
-		                 ${poolname}"
-	fi
-}
-
 stop() {
     local running
     local facet=$1
@@ -1028,9 +1137,14 @@ stop() {
         do_facet ${facet} umount -d $@ $mntpt
     fi
 
-    # umount should block, but we should wait for unrelated obd's
-    # like the MGS or MGC to also stop.
-    wait_exit_ST ${facet}
+	# umount should block, but we should wait for unrelated obd's
+	# like the MGS or MGC to also stop.
+	wait_exit_ST ${facet} || return ${PIPESTATUS[0]}
+
+	if [[ $(facet_fstype $facet) == zfs ]]; then
+		# export ZFS storage pool
+		export_zpool $facet
+	fi
 }
 
 # save quota version (both administrative and operational quotas)
@@ -1488,7 +1602,6 @@ reboot_facet() {
 	if [ "$FAILURE_MODE" = HARD ]; then
 		reboot_node $(facet_active_host $facet)
 	else
-		refresh_disk ${facet}
 		sleep 10
 	fi
 }
@@ -1551,6 +1664,7 @@ TESTLOG_PREFIX=$TESTLOG_PREFIX \
 TESTNAME=$TESTNAME \
 DBENCH_LIB=$DBENCH_LIB \
 DBENCH_SRC=$DBENCH_SRC \
+CLIENT_COUNT=$((CLIENTCOUNT - 1)) \
 LFS=$LFS \
 run_${load}.sh" &
     local ppid=$!
@@ -1760,30 +1874,43 @@ cleanup_check() {
 }
 
 wait_update () {
-    local node=$1
-    local TEST=$2
-    local FINAL=$3
-    local MAX=${4:-90}
-
-        local RESULT
-        local WAIT=0
-        local sleep=1
-        local print=10
-        while [ true ]; do
-            RESULT=$(do_node $node "$TEST")
-            if [ "$RESULT" == "$FINAL" ]; then
-                [ -z "$RESULT" -o $WAIT -le $sleep ] ||
-                    echo "Updated after ${WAIT}s: wanted '$FINAL' got '$RESULT'"
-                return 0
-            fi
-            [ $WAIT -ge $MAX ] && break
-            [ $((WAIT % print)) -eq 0 ] &&
-                echo "Waiting $((MAX - WAIT)) secs for update"
-            WAIT=$((WAIT + sleep))
-            sleep $sleep
-        done
-        echo "Update not seen after ${MAX}s: wanted '$FINAL' got '$RESULT'"
-        return 3
+	local verbose=false
+	if [[ "$1" == "--verbose" ]]; then
+		shift
+		verbose=true
+	fi
+
+	local node=$1
+	local TEST=$2
+	local FINAL=$3
+	local MAX=${4:-90}
+	local RESULT
+	local PREV_RESULT
+	local WAIT=0
+	local sleep=1
+	local print=10
+
+	while [ true ]; do
+		RESULT=$(do_node $node "$TEST")
+		if [[ "$RESULT" == "$FINAL" ]]; then
+			[[ -z "$RESULT" || $WAIT -le $sleep ]] ||
+				echo "Updated after ${WAIT}s: wanted '$FINAL'"\
+				     "got '$RESULT'"
+			return 0
+		fi
+		if [[ $verbose && "$RESULT" != "$PREV_RESULT" ]]; then
+			echo "Changed after ${WAIT}s: from '$PREV_RESULT'"\
+			     "to '$RESULT'"
+			PREV_RESULT=$RESULT
+		fi
+		[[ $WAIT -ge $MAX ]] && break
+		[[ $((WAIT % print)) -eq 0 ]] &&
+			echo "Waiting $((MAX - WAIT)) secs for update"
+		WAIT=$((WAIT + sleep))
+		sleep $sleep
+	done
+	echo "Update not seen after ${MAX}s: wanted '$FINAL' got '$RESULT'"
+	return 3
 }
 
 wait_update_facet() {
@@ -2151,7 +2278,7 @@ replay_barrier() {
 	do_facet $facet "sync; sync; sync"
 	df $MOUNT
 
-        # make sure there will be no seq change
+	# make sure there will be no seq change
 	local clients=${CLIENTS:-$HOSTNAME}
 	local f=fsa-\\\$\(hostname\)
 	do_nodes $clients "mcreate $MOUNT/$f; rm $MOUNT/$f"
@@ -2159,6 +2286,21 @@ replay_barrier() {
 
 	local svc=${facet}_svc
 	do_facet $facet $LCTL --device ${!svc} notransno
+	#
+	# If a ZFS OSD is made read-only here, its pool is "freezed". This
+	# in-memory state has to be cleared by either rebooting the host or
+	# exporting and reimporting the pool.
+	#
+	# Although the uberblocks are not updated when a pool is freezed,
+	# transactions are still written to the disks. Modified blocks may be
+	# cached in memory when tests try reading them back. The
+	# export-and-reimport process also evicts any cached pool data from
+	# memory to provide the correct "data loss" semantics.
+	#
+	# In the test framework, the exporting and importing operations are
+	# handled by stop() and mount_facet() separately, which are used
+	# inside fail() and fail_abort().
+	#
 	do_facet $facet $LCTL --device ${!svc} readonly
 	do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
 	$LCTL mark "local REPLAY BARRIER on ${!svc}"
@@ -2212,7 +2354,6 @@ fail_nodf() {
 fail_abort() {
 	local facet=$1
 	stop $facet
-	refresh_disk ${facet}
 	change_active $facet
 	wait_for_facet $facet
 	mount_facet $facet -o abort_recovery
@@ -2638,13 +2779,26 @@ do_nodesv() {
 }
 
 add() {
-    local facet=$1
-    shift
-    # make sure its not already running
-    stop ${facet} -f
-    rm -f $TMP/${facet}active
-    [[ $facet = mds1 ]] && combined_mgs_mds && rm -f $TMP/mgsactive
-    do_facet ${facet} $MKFS $*
+	local facet=$1
+	shift
+	# make sure its not already running
+	stop ${facet} -f
+	rm -f $TMP/${facet}active
+	[[ $facet = mds1 ]] && combined_mgs_mds && rm -f $TMP/mgsactive
+	do_facet ${facet} $MKFS $* || return ${PIPESTATUS[0]}
+
+	if [[ $(facet_fstype $facet) == zfs ]]; then
+		#
+		# After formatting a ZFS target, "cachefile=none" property will
+		# be set on the ZFS storage pool so that the pool is not
+		# automatically imported on system startup. And then the pool
+		# will be exported so as to leave the importing and exporting
+		# operations handled by mount_facet() and stop() separately.
+		#
+		refresh_partition_table $facet $(facet_vdevice $facet)
+		disable_zpool_cache $facet
+		export_zpool $facet
+	fi
 }
 
 ostdevname() {
@@ -3393,10 +3547,17 @@ is_empty_dir() {
 
 # empty lustre filesystem may have empty directories lost+found and .lustre
 is_empty_fs() {
+	# exclude .lustre & lost+found
 	[ $(find $1 -maxdepth 1 -name lost+found -o -name .lustre -prune -o \
 		-print | wc -l) = 1 ] || return 1
 	[ ! -d $1/lost+found ] || is_empty_dir $1/lost+found || return 1
-	[ ! -d $1/.lustre ] || is_empty_dir $1/.lustre || return 1
+	if [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.0) ]; then
+		# exclude .lustre/fid (LU-2780)
+		[ $(find $1/.lustre -maxdepth 1 -name fid -prune -o \
+			-print | wc -l) = 1 ] || return 1
+	else
+		[ ! -d $1/.lustre ] || is_empty_dir $1/.lustre || return 1
+	fi
 	return 0
 }
 
@@ -3458,6 +3619,15 @@ check_and_setup_lustre() {
         set_default_debug_nodes $(comma_list $(nodes_list))
     fi
 
+	if [ $(lower $OSD_TRACK_DECLARES_LBUG) == 'yes' ] ; then
+		local facets="$(get_facets OST),$(get_facets MDS),mgs"
+		local nodes="$(facets_hosts ${facets})"
+		if [ -n "$nodes" ] ; then
+			do_nodes $nodes "$LCTL set_param \
+				 osd-ldiskfs.track_declares_assert=1 || true"
+		fi
+	fi
+
 	init_gss
 	if $GSS; then
 		set_flavor_all $SEC
@@ -4030,28 +4200,33 @@ log_trace_dump() {
 ##################################
 
 error_noexit() {
-    local TYPE=${TYPE:-"FAIL"}
+	local TYPE=${TYPE:-"FAIL"}
 
-    local dump=true
-    # do not dump logs if $1=false
-    if [ "x$1" = "xfalse" ]; then
-        shift
-        dump=false
-    fi
+	local dump=true
+	# do not dump logs if $1=false
+	if [ "x$1" = "xfalse" ]; then
+		shift
+		dump=false
+	fi
 
-    log " ${TESTSUITE} ${TESTNAME}: @@@@@@ ${TYPE}: $@ "
-    log_trace_dump
 
-    mkdir -p $LOGDIR
-    # We need to dump the logs on all nodes
-    if $dump; then
-        gather_logs $(comma_list $(nodes_list))
-    fi
+	log " ${TESTSUITE} ${TESTNAME}: @@@@@@ ${TYPE}: $@ "
+	log_trace_dump
+
+	mkdir -p $LOGDIR
+	# We need to dump the logs on all nodes
+	if $dump; then
+		gather_logs $(comma_list $(nodes_list))
+	fi
 
 	debugrestore
 	[ "$TESTSUITELOG" ] &&
 		echo "$TESTSUITE: $TYPE: $TESTNAME $@" >> $TESTSUITELOG
-	echo "$@" > $LOGDIR/err
+	if [ -z "$*" ]; then
+		echo "error() without useful message, please fix" > $LOGDIR/err
+	else
+		echo "$@" > $LOGDIR/err
+	fi
 }
 
 exit_status () {
@@ -4063,21 +4238,21 @@ exit_status () {
 }
 
 error() {
-    error_noexit "$@"
-    exit 1
+	error_noexit "$@"
+	exit 1
 }
 
 error_exit() {
-    error "$@"
+	error "$@"
 }
 
 # use only if we are ignoring failures for this test, bugno required.
 # (like ALWAYS_EXCEPT, but run the test and ignore the results.)
 # e.g. error_ignore 5494 "your message"
 error_ignore() {
-    local TYPE="IGNORE (bz$1)"
-    shift
-    error_noexit "$@"
+	local TYPE="IGNORE (bz$1)"
+	shift
+	error_noexit "$@"
 }
 
 error_and_remount() {
@@ -4087,7 +4262,7 @@ error_and_remount() {
 }
 
 skip_env () {
-    $FAIL_ON_SKIP_ENV && error false $@ || skip $@
+	$FAIL_ON_SKIP_ENV && error false $@ || skip $@
 }
 
 skip() {
@@ -4738,10 +4913,11 @@ generate_machine_file() {
 }
 
 get_stripe () {
-    local file=$1/stripe
-    touch $file
-    $LFS getstripe -v $file || error
-    rm -f $file
+	local file=$1/stripe
+
+	touch $file
+	$LFS getstripe -v $file || error "getstripe $file failed"
+	rm -f $file
 }
 
 setstripe_nfsserver () {
@@ -5911,6 +6087,33 @@ run_llverfs()
         llverfs $partial_arg $llverfs_opts $dir
 }
 
+#Remove objects from OST
+remove_ost_objects() {
+	local facet=$1
+	local ostdev=$2
+	local group=$3
+	shift 3
+	local objids="$@"
+	local mntpt=$(facet_mntpt $facet)
+	local opts=$OST_MOUNT_OPTS
+	local i
+	local rc
+
+	echo "removing objects from $ostdev on $facet: $objids"
+	if ! test -b $ostdev; then
+		opts=$(csa_add "$opts" -o loop)
+	fi
+	mount -t $(facet_fstype $facet) $opts $ostdev $mntpt ||
+		return $?
+	rc=0
+	for i in $objids; do
+		rm $mntpt/O/$group/d$((i % 32))/$i || { rc=$?; break; }
+	done
+	umount -f $mntpt || return $?
+	return $rc
+}
+
+#Remove files from MDT
 remove_mdt_files() {
 	local facet=$1
 	local mdtdev=$2
@@ -5926,7 +6129,7 @@ remove_mdt_files() {
 	fi
 	mount -t $(facet_fstype $facet) $opts $mdtdev $mntpt ||
 		return $?
-	rc=0;
+	rc=0
 	for f in $files; do
 		rm $mntpt/ROOT/$f || { rc=$?; break; }
 	done
@@ -6031,7 +6234,8 @@ large_xattr_enabled() {
 
 	local mds_dev=$(mdsdevname ${SINGLEMDS//mds/})
 
-	do_facet $SINGLEMDS "$DUMPE2FS -h $mds_dev 2>&1 | grep -q large_xattr"
+	do_facet $SINGLEMDS "$DUMPE2FS -h $mds_dev 2>&1 |
+		grep -E -q '(ea_inode|large_xattr)'"
 	return ${PIPESTATUS[0]}
 }
 
@@ -6184,41 +6388,28 @@ generate_logname() {
 	echo "$TESTLOG_PREFIX.$TESTNAME.$logname.$(hostname -s).log"
 }
 
-# mkdir directory on different MDTs
+# make directory on different MDTs
 test_mkdir() {
 	local option
 	local parent
 	local child
 	local path
-	local dir
 	local rc=0
 
-	if [ $# -eq 2 ]; then
-		option=$1
-		path=$2
-	else
-		path=$1
-	fi
-
-	child=${path##*/}
-	parent=${path%/*}
+	case $# in
+		1) path=$1;;
+		2) option=$1
+		   path=$2;;
+		*) error "Only creating single directory is supported";;
+	esac
 
-	if [ "$parent" == "$child" ]; then
-		parent=$(pwd)
-	fi
+	child=$(basename $path)
+	parent=$(dirname $path)
 
-	if [ "$option" == "-p" -a -d ${parent}/${child} ]; then
+	if [ "$option" == "-p" -a -d $parent/$child ]; then
 		return $rc
 	fi
 
-	# it needs to check whether there is further / in child
-	dir=$(echo $child | awk -F '/' '{print $2}')
-	if [ ! -z "$dir" ]; then
-		local subparent=$(echo $child | awk -F '/' '{ print $1 }')
-		parent=${parent}"/"${subparent}
-		child=$dir
-	fi
-
 	if [ ! -d ${parent} ]; then
 		if [ "$option" == "-p" ]; then
 			mkdir -p ${parent}
@@ -6228,19 +6419,18 @@ test_mkdir() {
 	fi
 
 	if [ $MDSCOUNT -le 1 ]; then
-		mkdir $option ${parent}/${child} || rc=$?
+		mkdir $option $parent/$child || rc=$?
 	else
 		local mdt_idx=$($LFS getstripe -M $parent)
+		local test_num=$(echo $testnum | sed -e 's/[^0-9]*//g')
 
 		if [ "$mdt_idx" -ne 0 ]; then
-			mkdir $option ${parent}/${child} || rc=$?
-			return $rc
+			mkdir $option $parent/$child || rc=$?
+		else
+			mdt_idx=$((test_num % MDSCOUNT))
+			echo "mkdir $mdt_idx for $parent/$child"
+			$LFS setdirstripe -i $mdt_idx $parent/$child || rc=$?
 		fi
-
-		local test_num=$(echo $testnum | sed -e 's/[^0-9]*//g')
-		local mdt_idx=$((test_num % MDSCOUNT))
-		echo "mkdir $mdt_idx for ${parent}/${child}"
-		$LFS setdirstripe -i $mdt_idx ${parent}/${child} || rc=$?
 	fi
 	return $rc
 }