X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fconf-sanity.sh;h=546780d939e3d4a7bfa30193a68d3988896f03ba;hb=11317196203c26f8c9e4d2a8261859b1c7a76c9d;hp=93dfacbc2af948b3ed4c5c14354d4c1cec12a01c;hpb=ff353fef1eca3cc4d4b492b9591c819c00f4e8dc;p=fs%2Flustre-release.git

diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh
index 93dfacb..546780d 100644
--- a/lustre/tests/conf-sanity.sh
+++ b/lustre/tests/conf-sanity.sh
@@ -1,9 +1,9 @@
 #!/bin/bash
 
-# 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.
+# 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 tests should use error() to describe the failure more clearly,
+#   and reduce the need to look into the tests to see what failed.
 
 set -e
 
@@ -75,8 +75,8 @@ if [[ "$LDISKFS_MKFS_OPTS" != *lazy_itable_init* ]]; then
 fi
 
 [ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test:        LU-2778 LU-2059
-	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 57b     50h"
+# bug number for skipped test:        LU-2778 LU-2059 LU-4444
+	ALWAYS_EXCEPT="$ALWAYS_EXCEPT 57b     50h     69"
 
 init_logging
 
@@ -84,7 +84,7 @@ init_logging
 require_dsh_mds || exit 0
 require_dsh_ost || exit 0
 #
-[ "$SLOW" = "no" ] && EXCEPT_SLOW="30a 31 45"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="30a 31 45 69"
 
 
 assert_DIR
@@ -1062,6 +1062,78 @@ test_28() {
 }
 run_test 28 "permanent parameter setting"
 
+test_28a() { # LU-4221
+	[[ $(lustre_version_code ost1) -ge $(version_code 2.5.52) ]] ||
+		{ skip "Need OST version at least 2.5.52" && return 0; }
+	[ "$(facet_fstype ost1)" = "zfs" ] &&
+		skip "LU-4221: no such proc params for ZFS OSTs" && return
+
+	local name
+	local param
+	local cmd
+	local old
+	local new
+	local device="$FSNAME-OST0000"
+
+	setup
+
+	# In this test we will set three kinds of proc parameters with
+	# lctl conf_param:
+	# 1. the ones moved from the OFD to the OSD, and only their
+	#    symlinks kept in obdfilter
+	# 2. non-symlink ones in the OFD
+	# 3. non-symlink ones in the OSD
+
+	# Check 1.
+	# prepare a symlink parameter in the OFD
+	name="writethrough_cache_enable"
+	param="$device.ost.$name"
+	cmd="$LCTL get_param -n obdfilter.$device.$name"
+
+	# conf_param the symlink parameter in the OFD
+	old=$(do_facet ost1 $cmd)
+	new=$(((old + 1) % 2))
+	set_conf_param_and_check ost1 "$cmd" "$param" $new ||
+		error "lctl conf_param $device.ost.$param=$new failed"
+
+	# conf_param the target parameter in the OSD
+	param="$device.osd.$name"
+	cmd="$LCTL get_param -n osd-*.$device.$name"
+	set_conf_param_and_check ost1 "$cmd" "$param" $old ||
+		error "lctl conf_param $device.osd.$param=$old failed"
+
+	# Check 2.
+	# prepare a non-symlink parameter in the OFD
+	name="client_cache_seconds"
+	param="$device.ost.$name"
+	cmd="$LCTL get_param -n obdfilter.$device.$name"
+
+	# conf_param the parameter in the OFD
+	old=$(do_facet ost1 $cmd)
+	new=$((old * 2))
+	set_conf_param_and_check ost1 "$cmd" "$param" $new ||
+		error "lctl conf_param $device.ost.$param=$new failed"
+	set_conf_param_and_check ost1 "$cmd" "$param" $old ||
+		error "lctl conf_param $device.ost.$param=$old failed"
+
+	# Check 3.
+	# prepare a non-symlink parameter in the OSD
+	name="lma_self_repair"
+	param="$device.osd.$name"
+	cmd="$LCTL get_param -n osd-*.$device.$name"
+
+	# conf_param the parameter in the OSD
+	old=$(do_facet ost1 $cmd)
+	new=$(((old + 1) % 2))
+	set_conf_param_and_check ost1 "$cmd" "$param" $new ||
+		error "lctl conf_param $device.osd.$param=$new failed"
+	set_conf_param_and_check ost1 "$cmd" "$param" $old ||
+		error "lctl conf_param $device.osd.$param=$old failed"
+
+	cleanup
+}
+run_test 28a "set symlink parameters permanently with conf_param"
+
 test_29() {
 	[ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2, skipping" && return
         setup > /dev/null 2>&1
@@ -1243,7 +1315,9 @@ test_32newtarball() {
 		count=$(($T32_BLIMIT / 1024 / 2))
 	chown $T32_QID.$T32_QID $tmp/src/t32_qf_old
 
-	formatall
+	# format ost with comma-separated NIDs to verify LU-4460
+	local failnid="$(h2$NETTYPE 1.2.3.4),$(h2$NETTYPE 4.3.2.1)"
+	MGSNID="$MGSNID,$MGSNID" OSTOPT="--failnode=$failnid" formatall
 
 	setupall
 
@@ -1335,6 +1409,7 @@ t32_check() {
 
 t32_test_cleanup() {
 	local tmp=$TMP/t32
+	local fstype=$(facet_fstype $SINGLEMDS)
 	local rc=$?
 
 	if $shall_cleanup_lustre; then
@@ -1352,6 +1427,10 @@ t32_test_cleanup() {
 
 	$r rm -rf $tmp
 	rm -rf $tmp
+	if [ $fstype == "zfs" ]; then
+		$r $ZPOOL destroy t32fs-mdt1 || rc=$?
+		$r $ZPOOL destroy t32fs-ost1 || rc=$?
+	fi
 	return $rc
 }
 
@@ -1391,16 +1470,20 @@ t32_reload_modules() {
 t32_wait_til_devices_gone() {
 	local node=$1
 	local devices
+	local loops
 	local i=0
 
 	echo wait for devices to go
 	while ((i < 20)); do
 		devices=$(do_rpc_nodes $node $LCTL device_list | wc -l)
-		((devices == 0)) && return 0
+		loops=$(do_rpc_nodes $node losetup -a | grep -c t32)
+		((devices == 0 && loops == 0)) && return 0
 		sleep 5
 		i=$((i + 1))
 	done
-	echo "waiting for devices on $node: Given up"
+	echo "waiting for dev on $node: dev $devices loop $loops given up"
+	do_rpc_nodes $node "losetup -a"
+	do_rpc_nodes $node "$LCTL devices_list"
 	return 1
 }
 
@@ -1510,6 +1593,8 @@ t32_test() {
 	local nrpcs
 	local list
 	local fstype=$(facet_fstype $SINGLEMDS)
+	local mdt_dev=$tmp/mdt
+	local ost_dev=$tmp/ost
 
 	trap 'trap - RETURN; t32_test_cleanup' RETURN
 
@@ -1532,16 +1617,32 @@ t32_test() {
 	local version=$(version_code $img_commit)
 	[[ $version -gt $(version_code 2.4.0) ]] && ff_convert="no"
 
+	if [ $fstype == "zfs" ]; then
+		# import pool first
+		$r $ZPOOL import -f -d $tmp t32fs-mdt1
+		$r $ZPOOL import -f -d $tmp t32fs-ost1
+		mdt_dev=t32fs-mdt1/mdt1
+		ost_dev=t32fs-ost1/ost1
+		wait_update_facet $SINGLEMDS "$ZPOOL list |
+			awk '/^t32fs-mdt1/ { print \\\$1 }'" "t32fs-mdt1" || {
+				error_noexit "import zfs pool failed"
+				return 1
+			}
+	fi
+
 	$r $LCTL set_param debug="$PTLDEBUG"
 
-	$r $TUNEFS --dryrun $tmp/mdt || {
+	$r $TUNEFS --dryrun $mdt_dev || {
+		$r losetup -a
 		error_noexit "tunefs.lustre before mounting the MDT"
 		return 1
 	}
 	if [ "$writeconf" ]; then
-		mopts=loop,writeconf
+		mopts=writeconf
 		if [ $fstype == "ldiskfs" ]; then
-			$r $TUNEFS --quota $tmp/mdt || {
+			mopts="loop,$mopts"
+			$r $TUNEFS --quota $mdt_dev || {
+				$r losetup -a
 				error_noexit "Enable mdt quota feature"
 				return 1
 			}
@@ -1555,18 +1656,26 @@ t32_test() {
 			local osthost=$(facet_active_host ost1)
 			local ostnid=$(do_node $osthost $LCTL list_nids | head -1)
 
-			$r mount -t lustre -o loop,nosvc $tmp/mdt $tmp/mnt/mdt
+			mopts=nosvc
+			if [ $fstype == "ldiskfs" ]; then
+				mopts="loop,$mopts"
+			fi
+			$r mount -t lustre -o $mopts $mdt_dev $tmp/mnt/mdt
 			$r lctl replace_nids $fsname-OST0000 $ostnid
 			$r lctl replace_nids $fsname-MDT0000 $nid
 			$r umount -d $tmp/mnt/mdt
 		fi
 
-		mopts=loop,exclude=$fsname-OST0000
+		mopts=exclude=$fsname-OST0000
+		if [ $fstype == "ldiskfs" ]; then
+			mopts="loop,$mopts"
+		fi
 	fi
 
 	t32_wait_til_devices_gone $node
 
-	$r mount -t lustre -o $mopts $tmp/mdt $tmp/mnt/mdt || {
+	$r mount -t lustre -o $mopts $mdt_dev $tmp/mnt/mdt || {
+		$r losetup -a
 		error_noexit "Mounting the MDT"
 		return 1
 	}
@@ -1610,22 +1719,27 @@ t32_test() {
 		return 1
 	fi
 
-	$r $TUNEFS --dryrun $tmp/ost || {
+	$r $TUNEFS --dryrun $ost_dev || {
 		error_noexit "tunefs.lustre before mounting the OST"
 		return 1
 	}
 	if [ "$writeconf" ]; then
-		mopts=loop,mgsnode=$nid,$writeconf
+		mopts=mgsnode=$nid,$writeconf
 		if [ $fstype == "ldiskfs" ]; then
-			$r $TUNEFS --quota $tmp/ost || {
+			mopts="loop,$mopts"
+			$r $TUNEFS --quota $ost_dev || {
+				$r losetup -a
 				error_noexit "Enable ost quota feature"
 				return 1
 			}
 		fi
 	else
-		mopts=loop,mgsnode=$nid
+		mopts=mgsnode=$nid
+		if [ $fstype == "ldiskfs" ]; then
+			mopts="loop,$mopts"
+		fi
 	fi
-	$r mount -t lustre -o $mopts $tmp/ost $tmp/mnt/ost || {
+	$r mount -t lustre -o $mopts $ost_dev $tmp/mnt/ost || {
 		error_noexit "Mounting the OST"
 		return 1
 	}
@@ -1865,12 +1979,17 @@ t32_test() {
 		}
 
 		# mount a second time to make sure we didnt leave upgrade flag on
-		$r $TUNEFS --dryrun $tmp/mdt || {
+		$r $TUNEFS --dryrun $mdt_dev || {
+			$r losetup -a
 			error_noexit "tunefs.lustre before remounting the MDT"
 			return 1
 		}
-		$r mount -t lustre -o loop,exclude=$fsname-OST0000 $tmp/mdt \
-				 $tmp/mnt/mdt || {
+
+		mopts=exclude=$fsname-OST0000
+		if [ $fstype == "ldiskfs" ]; then
+			mopts="loop,$mopts"
+		fi
+		$r mount -t lustre -o $mopts $mdt_dev $tmp/mnt/mdt || {
 			error_noexit "Remounting the MDT"
 			return 1
 		}
@@ -2726,7 +2845,7 @@ test_48() { # bug 17636
 run_test 48 "too many acls on file"
 
 # check PARAM_SYS_LDLM_TIMEOUT option of MKFS.LUSTRE
-test_49() { # bug 17710
+test_49a() { # bug 17710
 	local timeout_orig=$TIMEOUT
 	local ldlm_timeout_orig=$LDLM_TIMEOUT
 	local LOCAL_TIMEOUT=20
@@ -2736,49 +2855,60 @@ test_49() { # bug 17710
 
 	reformat
 	setup_noconfig
-	check_mount || return 1
+	check_mount || error "client mount failed"
 
 	echo "check ldlm_timout..."
-	LDLM_MDS="`do_facet $SINGLEMDS lctl get_param -n ldlm_timeout`"
-	LDLM_OST1="`do_facet ost1 lctl get_param -n ldlm_timeout`"
-	LDLM_CLIENT="`do_facet client lctl get_param -n ldlm_timeout`"
+	local LDLM_MDS="$(do_facet $SINGLEMDS lctl get_param -n ldlm_timeout)"
+	local LDLM_OST1="$(do_facet ost1 lctl get_param -n ldlm_timeout)"
+	local LDLM_CLIENT="$(do_facet client lctl get_param -n ldlm_timeout)"
 
-	if [ $LDLM_MDS -ne $LDLM_OST1 ] || [ $LDLM_MDS -ne $LDLM_CLIENT ]; then
+	if [ $LDLM_MDS -ne $LDLM_OST1 -o $LDLM_MDS -ne $LDLM_CLIENT ]; then
 		error "Different LDLM_TIMEOUT:$LDLM_MDS $LDLM_OST1 $LDLM_CLIENT"
 	fi
 
 	if [ $LDLM_MDS -ne $((LOCAL_TIMEOUT / 3)) ]; then
-		error "LDLM_TIMEOUT($LDLM_MDS) is not correct"
+		error "LDLM_TIMEOUT($LDLM_MDS) is not $((LOCAL_TIMEOUT / 3))"
 	fi
 
 	umount_client $MOUNT
-	stop_ost || return 2
-	stop_mds || return 3
+	stop_ost || error "problem stopping OSS"
+	stop_mds || error "problem stopping MDS"
+
+	LDLM_TIMEOUT=$ldlm_timeout_orig
+	TIMEOUT=$timeout_orig
+}
+run_test 49a "check PARAM_SYS_LDLM_TIMEOUT option of mkfs.lustre"
+
+test_49b() { # bug 17710
+	local timeout_orig=$TIMEOUT
+	local ldlm_timeout_orig=$LDLM_TIMEOUT
+	local LOCAL_TIMEOUT=20
 
 	LDLM_TIMEOUT=$((LOCAL_TIMEOUT - 1))
+	TIMEOUT=$LOCAL_TIMEOUT
 
 	reformat
 	setup_noconfig
-	check_mount || return 7
+	check_mount || error "client mount failed"
 
-	LDLM_MDS="`do_facet $SINGLEMDS lctl get_param -n ldlm_timeout`"
-	LDLM_OST1="`do_facet ost1 lctl get_param -n ldlm_timeout`"
-	LDLM_CLIENT="`do_facet client lctl get_param -n ldlm_timeout`"
+	local LDLM_MDS="$(do_facet $SINGLEMDS lctl get_param -n ldlm_timeout)"
+	local LDLM_OST1="$(do_facet ost1 lctl get_param -n ldlm_timeout)"
+	local LDLM_CLIENT="$(do_facet client lctl get_param -n ldlm_timeout)"
 
-	if [ $LDLM_MDS -ne $LDLM_OST1 ] || [ $LDLM_MDS -ne $LDLM_CLIENT ]; then
+	if [ $LDLM_MDS -ne $LDLM_OST1 -o $LDLM_MDS -ne $LDLM_CLIENT ]; then
 		error "Different LDLM_TIMEOUT:$LDLM_MDS $LDLM_OST1 $LDLM_CLIENT"
 	fi
 
 	if [ $LDLM_MDS -ne $((LOCAL_TIMEOUT - 1)) ]; then
-		error "LDLM_TIMEOUT($LDLM_MDS) is not correct"
+		error "LDLM_TIMEOUT($LDLM_MDS) is not $((LOCAL_TIMEOUT - 1))"
 	fi
 
-	cleanup || return $?
+	cleanup || error "cleanup failed"
 
 	LDLM_TIMEOUT=$ldlm_timeout_orig
 	TIMEOUT=$timeout_orig
 }
-run_test 49 "check PARAM_SYS_LDLM_TIMEOUT option of MKFS.LUSTRE"
+run_test 49b "check PARAM_SYS_LDLM_TIMEOUT option of mkfs.lustre"
 
 lazystatfs() {
         # Test both statfs and lfs df and fail if either one fails
@@ -3163,7 +3293,13 @@ test_52() {
 	do_node $ost1node 'mv '$objects' '${ost1mnt}'/lost+found'
 	[ $? -eq 0 ] || { error "Unable to move objects"; return 14; }
 
+	# recover objects dry-run
+	echo "ll_recover_lost_found_objs dry_run"
+	do_node $ost1node "ll_recover_lost_found_objs -n -d $ost1mnt/O"
+	[ $? -eq 0 ] || { error "ll_recover_lost_found_objs failed"; return 15; }
+
 	# recover objects
+	echo "ll_recover_lost_found_objs fix run"
 	do_node $ost1node "ll_recover_lost_found_objs -d $ost1mnt/lost+found"
 	[ $? -eq 0 ] || { error "ll_recover_lost_found_objs failed"; return 15; }
 
@@ -3861,6 +3997,15 @@ test_68() {
 run_test 68 "be able to reserve specific sequences in FLDB"
 
 test_69() {
+	local server_version=$(lustre_version_code $SINGLEMDS)
+
+	[[ $server_version -lt $(version_code 2.4.2) ]] &&
+		skip "Need MDS version at least 2.4.2" && return
+
+	[[ $server_version -ge $(version_code 2.4.50) ]] &&
+	[[ $server_version -lt $(version_code 2.5.0) ]] &&
+		skip "Need MDS version at least 2.5.0" && return
+
 	setup
 
 	# use OST0000 since it probably has the most creations
@@ -3874,27 +4019,30 @@ test_69() {
 	# not, then the OST will refuse to allow the MDS connect
 	# because the LAST_ID value is too different from the MDS
 	#define OST_MAX_PRECREATE=20000
-	local num_create=$((20000 * 3/2 - $last_id + 100))
+	local num_create=$((20000 * 5))
 
-	mkdir $DIR/$tdir
+	mkdir -p $DIR/$tdir
 	$LFS setstripe -i 0 $DIR/$tdir
-	createmany $DIR/$tdir/$tfile- $num_create
+	createmany -o $DIR/$tdir/$tfile- $num_create ||
+		error "createmany: failed to create $num_create files: $?"
 	# delete all of the files with objects on OST0 so the
 	# filesystem is not inconsistent later on
-	$LFS find $MOUNT --index 0 -print0 | xargs -0 unlink
+	$LFS find $MOUNT --ost 0 | xargs rm
 
 	stop_ost || error "OST0 stop failure"
-	add ost1 $(mkfs_opts ost1 $ostdev) --reformat --replace $ostdev ||
+	add ost1 $(mkfs_opts ost1 $(ostdevname 1)) --reformat --replace \
+		$(ostdevname 1) $(ostvdevname 1) ||
 		error "reformat and replace $ostdev failed"
 	start_ost || error "OST0 restart failure"
+	wait_osc_import_state mds ost FULL
 
 	touch $DIR/$tdir/$tfile-last || error "create file after reformat"
-	local idx=$($LFS getstripe -c $DIR/$tdir/$tfile-last)
+	local idx=$($LFS getstripe -i $DIR/$tdir/$tfile-last)
 	[ $idx -ne 0 ] && error "$DIR/$tdir/$tfile-last on $idx not 0" || true
 
 	cleanup
 }
-run_test 68 "replace an OST with the same index"
+run_test 69 "replace an OST with the same index"
 
 test_70a() {
 	[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return