ONLY=${ONLY:-"$*"}
-#Bug number for excepting test
-ALWAYS_EXCEPT="$SANITY_LFSCK_EXCEPT"
-
-[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
-# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
-
-LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
+LUSTRE=${LUSTRE:-$(dirname $0)/..}
. $LUSTRE/tests/test-framework.sh
init_test_env $@
-. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
init_logging
+# bug number for skipped test:
+ALWAYS_EXCEPT="$SANITY_LFSCK_EXCEPT "
+# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
+
+[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
+build_test_filter
+
require_dsh_mds || exit 0
load_modules
exit 0
fi
-[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.60) ]] &&
- skip "Need MDS version at least 2.3.60" && exit 0
+(( $MDS1_VERSION >= $(version_code 2.3.60) )) ||
+ skip "Need MDS version at least 2.3.60"
LTIME=${LTIME:-120}
# use small MDS + OST size to speed formatting time
# do not use too small MDSSIZE/OSTSIZE, which affect the default journal size
MDSSIZE=100000
+[ "$mds1_FSTYPE" == zfs ] && MDSSIZE=300000
OSTSIZE=100000
+[ "$ost1_FSTYPE" == zfs ] && OSTSIZE=300000
+
# no need too many OSTs, to reduce the format/start/stop overhead
+cleanupall
[ $OSTCOUNT -gt 4 ] && OSTCOUNT=4
# build up a clean test environment.
-formatall
-setupall
-
-[[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.4.90) ]] &&
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2c"
-
-[[ $(lustre_version_code ost1) -lt $(version_code 2.5.55) ]] &&
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 12 13 14 15 16 17 18 19 20 21"
-
-[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.50) ]] &&
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2d 2e 3 22 23 24 25 26 27 28 29 30 31"
-
-# DNE does not support striped directory on zfs-based backend yet.
-[ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 31"
-
-build_test_filter
+REFORMAT="yes" check_and_setup_lustre
MDT_DEV="${FSNAME}-MDT0000"
OST_DEV="${FSNAME}-OST0000"
$LCTL get_param -n mdd.${MDT_DEV}.lfsck_layout"
SHOW_LAYOUT_ON_OST="do_facet ost1 \
$LCTL get_param -n obdfilter.${OST_DEV}.lfsck_layout"
-MOUNT_OPTS_SCRUB="-o user_xattr"
-MOUNT_OPTS_NOSCRUB="-o user_xattr,noscrub"
-MOUNT_OPTS_SKIP_LFSCK="-o user_xattr,skip_lfsck"
+MOUNT_OPTS_SCRUB="$MDS_MOUNT_OPTS -o user_xattr"
+MOUNT_OPTS_NOSCRUB="$MDS_MOUNT_OPTS -o user_xattr,noscrub"
+MOUNT_OPTS_SKIP_LFSCK="$MDS_MOUNT_OPTS -o user_xattr,skip_lfsck"
lfsck_prep() {
local ndirs=$1
}
run_e2fsck_on_mdt0() {
- [ $(facet_fstype $SINGLEMDS) != ldiskfs ] && return
+ [ $mds1_FSTYPE == ldiskfs ] || return 0
stop $SINGLEMDS > /dev/null || error "(0) Fail to the stop MDT0"
run_e2fsck $(facet_active_host $SINGLEMDS) $(mdsdevname 1) "-n" |
run_test 0 "Control LFSCK manually"
test_1a() {
- [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
- skip "OI Scrub not implemented for ZFS" && return
-
lfsck_prep 1 1
#define OBD_FAIL_FID_INDIR 0x1501
test_1b()
{
- [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
- skip "OI Scrub not implemented for ZFS" && return
+ [ "$mds1_FSTYPE" != ldiskfs ] &&
+ skip "OI Scrub not implemented for ZFS"
lfsck_prep 1 1
}
run_test 1b "LFSCK can find out and repair the missing FID-in-LMA"
+test_1c() {
+ lfsck_prep 1 1
+
+ #define OBD_FAIL_FID_IGIF 0x1504
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1504
+ touch $DIR/$tdir/dummy
+
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+ umount_client $MOUNT
+ $START_NAMESPACE -r || error "(3) Fail to start LFSCK for namespace!"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ $SHOW_NAMESPACE
+ error "(4) unexpected status"
+ }
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^dirent_repaired/ { print $2 }')
+ # for interop with old server
+ [ -z "$repaired" ] &&
+ repaired=$($SHOW_NAMESPACE |
+ awk '/^updated_phase1/ { print $2 }')
+
+ [ $repaired -eq 1 ] ||
+ error "(5) Fail to repair lost FID-in-dirent: $repaired"
+
+ run_e2fsck_on_mdt0
+
+ mount_client $MOUNT || error "(6) Fail to start client!"
+
+ #define OBD_FAIL_FID_LOOKUP 0x1505
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1505
+ ls $DIR/$tdir/ > /dev/null || error "(7) no FID-in-dirent."
+
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+}
+run_test 1c "LFSCK can find out and repair lost FID-in-dirent"
+
+test_1d() {
+ [ $MDS1_VERSION -lt $(version_code 2.13.57) ] &&
+ skip "MDS older than 2.13.57"
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+
+ check_mount_and_prep
+
+ touch $DIR/$tdir/$tfile
+ mkdir $DIR/$tdir/subdir
+ $LFS mkdir -i 1 $DIR/$tdir/remotedir
+ $LFS path2fid $DIR/$tdir
+ ll_decode_linkea $DIR/$tdir/$tfile
+ ll_decode_linkea $DIR/$tdir/subdir
+ ll_decode_linkea $DIR/$tdir/remotedir
+
+ local mntpt=$(facet_mntpt mds1)
+
+ # unlink OI files to remove the stale entry
+ local saved_opts=$MDS_MOUNT_OPTS
+
+ stopall
+ mount_fstype mds1 $mntpt
+ # increase $tdir FID oid in LMA
+ do_facet mds1 "getfattr -d -m trusted.lma -e hex \
+ --absolute-names $mntpt/ROOT/$tdir | \
+ sed -E 's/0(.{8})$/1\1/' | setfattr --restore=-"
+ unmount_fstype mds1 $mntpt
+ setupall
+
+ # the FID oid in LMA was increased above, and it's not in OI table,
+ # run scrub first to generate mapping in OI, so the following namespace
+ # check can fix linkea correctly, this is not necessary normally.
+ do_facet mds1 $LCTL lfsck_start -M ${MDT_DEV} -t scrub ||
+ error "failed to start LFSCK for scrub!"
+ wait_update_facet mds1 "$LCTL get_param -n \
+ osd-*.$(facet_svc mds1).oi_scrub |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ error "unexpected status"
+
+ $START_NAMESPACE -r -A || error "fail to start LFSCK for namespace!"
+ wait_update_facet mds1 "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ $SHOW_NAMESPACE
+ error "unexpected status"
+ }
+ $LFS path2fid $DIR/$tdir
+ ll_decode_linkea $DIR/$tdir/$tfile
+ ll_decode_linkea $DIR/$tdir/subdir
+ ll_decode_linkea $DIR/$tdir/remotedir
+
+ local pfid
+ local fid
+
+ fid=$($LFS path2fid $DIR/$tdir)
+ for f in $tfile subdir remotedir; do
+ pfid=$(ll_decode_linkea $DIR/$tdir/$f |
+ awk '/pfid/ { print $3 }')
+ pfid=${pfid%,}
+ [ "$pfid" == "$fid" ] || error "$fid in LMA != $pfid in linkea"
+ done
+}
+run_test 1d "LFSCK can fix mismatch of FID in LMA and FID in child linkea"
+
test_2a() {
lfsck_prep 1 1
test_2c()
{
+ (( $MDS1_VERSION > $(version_code 2.4.90) )) ||
+ skip "MDS older than 2.4.90"
+
lfsck_prep 1 1
#define OBD_FAIL_LFSCK_LINKEA_MORE2 0x1605
test_2d()
{
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-4788"
+
lfsck_prep 1 1
#define OBD_FAIL_LFSCK_NO_LINKEA 0x161d
test_2e()
{
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5511"
check_mount_and_prep
test_3()
{
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-4788"
+
lfsck_prep 4 4
mkdir $DIR/$tdir/dummy || error "(1) Fail to mkdir"
test_4()
{
- [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
- skip "OI Scrub not implemented for ZFS" && return
+ [ "$mds1_FSTYPE" != ldiskfs ] &&
+ skip "OI Scrub not implemented for ZFS"
lfsck_prep 3 3
cleanup_mount $MOUNT || error "(0.1) Fail to stop client!"
test_5()
{
- [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
- skip "OI Scrub not implemented for ZFS" && return
+ [ "$mds1_FSTYPE" != ldiskfs ] &&
+ skip "OI Scrub not implemented for ZFS"
lfsck_prep 1 1 1
cleanup_mount $MOUNT || error "(0.1) Fail to stop client!"
local D_POS1=$($SHOW_NAMESPACE |
awk '/^latest_start_position/ { print $4 }')
- if [ "$D_POS0" == "N/A" -o "$D_POS1" == "N/A" ]; then
+ echo "Additional debug for 6b"
+ $SHOW_NAMESPACE
+ if [ "$D_POS0" == "N/A" -o "$D_POS0" == "0x0" \
+ -o "$D_POS1" == "0x0" -o "$D_POS1" == "N/A" ]; then
[[ $O_POS0 -lt $O_POS1 ]] ||
error "(7.1) $O_POS1 is not larger than $O_POS0"
else
return 0
fi
- [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.50) ]] ||
- { skip "Need MDS version >= 2.7.50"; return; }
-
check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/lfsck || error "(1) Fail to mkdir lfsck"
$LFS setstripe -c 1 -i -1 $DIR/$tdir/lfsck
# There may be time error, normally it should be less than 2 seconds.
# We allow another 20% schedule error.
local TIME_DIFF=2
- # MAX_MARGIN = 1.2 = 12 / 10
+ # MAX_MARGIN = 1.3 = 13 / 10
local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) / \
- RUN_TIME1 * 12 / 10))
+ RUN_TIME1 * 13 / 10))
[ $SPEED -lt $MAX_SPEED ] || {
$SHOW_LAYOUT
log "speed1: $BASE_SPEED1 time1: $RUN_TIME1"
- error_ignore LU-9887 "(4) Speed $SPEED, expected < $MAX_SPEED"
+ error "(4) Speed $SPEED, expected < $MAX_SPEED"
}
# adjust speed limit
sleep $RUN_TIME2
SPEED=$($SHOW_LAYOUT | awk '/^average_speed_phase1/ { print $2 }')
- # MIN_MARGIN = 0.8 = 8 / 10
+ # MIN_MARGIN = 0.7 = 7 / 10
local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
- (RUN_TIME1 + RUN_TIME2) * 8 / 10))
+ (RUN_TIME1 + RUN_TIME2) * 7 / 10))
[ $SPEED -gt $MIN_SPEED ] || {
- if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+ if [ $mds1_FSTYPE != ldiskfs ]; then
error_ignore LU-5624 \
"(5.1) Got speed $SPEED, expected more than $MIN_SPEED"
else
fi
}
- # MAX_MARGIN = 1.2 = 12 / 10
+ # MAX_MARGIN = 1.3 = 13 / 10
MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
- (RUN_TIME1 + RUN_TIME2) * 12 / 10))
+ (RUN_TIME1 + RUN_TIME2) * 13 / 10))
[ $SPEED -lt $MAX_SPEED ] || {
$SHOW_LAYOUT
log "speed1: $BASE_SPEED1 time1: $RUN_TIME1"
log "speed2: $BASE_SPEED2 time2: $RUN_TIME2"
- error_ignore LU-9887 "(6) Speed $SPEED, expected < $MAX_SPEED"
+ error "(6) Speed $SPEED, expected < $MAX_SPEED"
}
- do_facet $SINGLEMDS \
- $LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
+ do_nodes $(comma_list $(mdts_nodes)) \
+ $LCTL set_param -n mdd.*.lfsck_speed_limit 0
+ do_nodes $(comma_list $(osts_nodes)) \
+ $LCTL set_param -n obdfilter.*.lfsck_speed_limit 0
wait_update_facet $SINGLEMDS \
"$LCTL get_param -n mdd.${MDT_DEV}.lfsck_layout |
return 0
fi
- [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.50) ]] ||
- { skip "Need MDS version >= 2.7.50"; return; }
-
lfsck_prep 0 0
echo "Preparing another 50 * 50 files (with error) at $(date)."
# There may be time error, normally it should be less than 2 seconds.
# We allow another 20% schedule error.
local TIME_DIFF=2
- # MAX_MARGIN = 1.2 = 12 / 10
+ # MAX_MARGIN = 1.3 = 13 / 10
local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) / \
- RUN_TIME1 * 12 / 10))
+ RUN_TIME1 * 13 / 10))
[ $SPEED -lt $MAX_SPEED ] || {
$SHOW_NAMESPACE
log "speed1: $BASE_SPEED1 time1: $RUN_TIME1"
- error_ignore LU-9887 "(8) Speed $SPEED, expected < $MAX_SPEED"
+ error "(8) Speed $SPEED, expected < $MAX_SPEED"
}
# adjust speed limit
sleep $RUN_TIME2
SPEED=$($SHOW_NAMESPACE | awk '/^average_speed_phase2/ { print $2 }')
- # MIN_MARGIN = 0.8 = 8 / 10
+ # MIN_MARGIN = 0.7 = 7 / 10
local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
- (RUN_TIME1 + RUN_TIME2) * 8 / 10))
+ (RUN_TIME1 + RUN_TIME2) * 7 / 10))
[ $SPEED -gt $MIN_SPEED ] || {
- if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+ if [ $mds1_FSTYPE != ldiskfs ]; then
error_ignore LU-5624 \
"(9.1) Got speed $SPEED, expected more than $MIN_SPEED"
else
fi
}
- # MAX_MARGIN = 1.2 = 12 / 10
+ # MAX_MARGIN = 1.3 = 13 / 10
MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
- (RUN_TIME1 + RUN_TIME2) * 12 / 10))
+ (RUN_TIME1 + RUN_TIME2) * 13 / 10))
[ $SPEED -lt $MAX_SPEED ] || {
$SHOW_NAMESPACE
log "speed1: $BASE_SPEED1 time1: $RUN_TIME1"
log "speed2: $BASE_SPEED2 time2: $RUN_TIME2"
- error_ignore LU-9887 "(10) Speed $SPEED, expected < $MAX_SPEED"
+ error "(10) Speed $SPEED, expected < $MAX_SPEED"
}
- do_facet $SINGLEMDS \
- $LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
+ do_nodes $(comma_list $(mdts_nodes)) \
+ $LCTL set_param -n mdd.*.lfsck_speed_limit 0
+ do_nodes $(comma_list $(osts_nodes)) \
+ $LCTL set_param -n obdfilter.*.lfsck_speed_limit 0
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_namespace |
awk '/^status/ { print \\\$2 }'" "completed" 32 || {
test_10()
{
- [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
- skip "lookup(..)/linkea on ZFS issue" && return
+ [[ $mds1_FSTYPE == ldiskfs ]] || skip "lookup(..)/linkea on ZFS issue"
lfsck_prep 1 1
[ "$STATUS" == "scanning-phase1" ] ||
error "(15) Expect 'scanning-phase1', but got '$STATUS'"
- do_facet $SINGLEMDS \
- $LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
+ do_nodes $(comma_list $(mdts_nodes)) \
+ $LCTL set_param -n mdd.*.lfsck_speed_limit 0
+ do_nodes $(comma_list $(osts_nodes)) \
+ $LCTL set_param -n obdfilter.*.lfsck_speed_limit 0
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_namespace |
awk '/^status/ { print \\\$2 }'" "completed" 32 || {
}
test_11a() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-1267"
+
check_mount_and_prep
- $SETSTRIPE -c 1 -i 0 $DIR/$tdir
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir
createmany -o $DIR/$tdir/f 64 || error "(0) Fail to create 64 files."
echo "stopall"
run_test 11a "LFSCK can rebuild lost last_id"
test_11b() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-1267"
+
check_mount_and_prep
- $SETSTRIPE -c 1 -i 0 $DIR/$tdir
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir
echo "set fail_loc=0x160d to skip the updating LAST_ID on-disk"
#define OBD_FAIL_LFSCK_SKIP_LASTID 0x160d
local proc_path="${FSNAME}-OST0000-osc-MDT0000"
local seq=$(do_facet mds1 $LCTL get_param -n \
osp.${proc_path}.prealloc_last_seq)
- local lastid1=$(do_facet ost1 "lctl get_param -n \
- obdfilter.${ost1_svc}.last_id" | grep $seq |
- awk -F: '{ print $2 }')
+ local id_used=$(do_facet mds1 $LCTL get_param -n \
+ osp.${proc_path}.prealloc_last_id)
umount_client $MOUNT
stop ost1 || error "(1) Fail to stop ost1"
error "(2) Fail to start ost1"
for ((i = 0; i < 60; i++)); do
- lastid2=$(do_facet ost1 "lctl get_param -n \
- obdfilter.${ost1_svc}.last_id" | grep $seq |
- awk -F: '{ print $2 }')
- [ ! -z $lastid2 ] && break;
+ id_ost1=$(do_facet ost1 \
+ "$LCTL get_param -n obdfilter.$ost1_svc.last_id" |
+ awk -F: "/$seq/ { print \$2 }")
+ [ -n "$id_ost1" ] && break
sleep 1
done
echo "the on-disk LAST_ID should be smaller than the expected one"
- [ $lastid1 -gt $lastid2 ] ||
- error "(4) expect lastid1 [ $lastid1 ] > lastid2 [ $lastid2 ]"
+ [ $id_used -gt $id_ost1 ] ||
+ error "(4) expect id_used '$id_used' > id_ost1 '$id_ost1'"
echo "trigger LFSCK for layout on ost1 to rebuild the on-disk LAST_ID"
$START_LAYOUT_ON_OST -r || error "(5) Fail to start LFSCK on OST!"
- wait_update_facet ost1 "$LCTL get_param -n \
- obdfilter.${OST_DEV}.lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ wait_update_facet ost1 \
+ "$LCTL get_param -n obdfilter.$ost1_svc.lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 || {
$SHOW_LAYOUT_ON_OST
error "(6) unexpected status"
}
error "(8) Fail to start ost1"
echo "the on-disk LAST_ID should have been rebuilt"
- wait_update_facet ost1 "$LCTL get_param -n \
- obdfilter.${ost1_svc}.last_id | grep $seq |
- awk -F: '{ print \\\$2 }'" "$lastid1" 60 || {
- do_facet ost1 $LCTL get_param -n \
- obdfilter.${ost1_svc}.last_id
- error "(9) expect lastid1 $seq:$lastid1"
+ # last_id may be larger than $id_used if objects were created/skipped
+ wait_update_facet_cond ost1 \
+ "$LCTL get_param -n obdfilter.$ost1_svc.last_id |
+ awk -F: '/$seq/ { print \\\$2 }'" "-ge" "$id_used" 60 || {
+ do_facet ost1 $LCTL get_param obdfilter.$ost1_svc.last_id
+ error "(9) expect last_id >= id_used $seq:$id_used"
}
do_facet ost1 $LCTL set_param fail_loc=0
run_test 11b "LFSCK can rebuild crashed last_id"
test_12a() {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3950"
check_mount_and_prep
for k in $(seq $MDSCOUNT); do
run_test 12a "single command to trigger LFSCK on all devices"
test_12b() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3950"
+
check_mount_and_prep
echo "Start LFSCK without '-M' specified."
run_test 12b "auto detect Lustre device"
test_13() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3593"
+
echo "#####"
echo "The lmm_oi in layout EA should be consistent with the MDT-object"
echo "FID; otherwise, the LFSCK should re-generate the lmm_oi from the"
#define OBD_FAIL_LFSCK_BAD_LMMOI 0x160f
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160f
createmany -o $DIR/$tdir/f 1
- $LFS setstripe -E 1M -E -1 $DIR/$tdir/f1 ||
+ $LFS setstripe -E 1M -S 1M -E -1 $DIR/$tdir/f1 ||
error "(0) Fail to create PFL $DIR/$tdir/f1"
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
run_test 13 "LFSCK can repair crashed lmm_oi"
test_14a() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3590"
+
echo "#####"
echo "The OST-object referenced by the MDT-object should be there;"
echo "otherwise, the LFSCK should re-create the missing OST-object."
run_test 14a "LFSCK can repair MDT-object with dangling LOV EA reference (1)"
test_14b() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3590"
+
echo "#####"
echo "The OST-object referenced by the MDT-object should be there;"
echo "otherwise, the LFSCK should re-create the missing OST-object."
run_test 14b "LFSCK can repair MDT-object with dangling LOV EA reference (2)"
test_15a() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3591"
+
echo "#####"
echo "If the OST-object referenced by the MDT-object back points"
echo "to some non-exist MDT-object, then the LFSCK should repair"
run_test 15a "LFSCK can repair unmatched MDT-object/OST-object pairs (1)"
test_15b() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3591"
+
echo "#####"
echo "If the OST-object referenced by the MDT-object back points"
echo "to other MDT-object that doesn't recognize the OST-object,"
run_test 15b "LFSCK can repair unmatched MDT-object/OST-object pairs (2)"
test_15c() {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
-
- [ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.55) ] &&
- skip "Skip the test after 2.7.55 see LU-6437" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ (( $MDS1_VERSION < $(version_code 2.7.55) )) ||
+ skip "MDS newer than 2.7.55, LU-6475"
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3591"
echo "#####"
echo "According to current metadata migration implementation,"
run_test 15c "LFSCK can repair unmatched MDT-object/OST-object pairs (3)"
test_16() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3594"
+
echo "#####"
echo "If the OST-object's owner information does not match the owner"
echo "information stored in the MDT-object, then the LFSCK trust the"
dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1
cancel_lru_locks osc
+ # created but no setattr or write to the file.
+ mkdir $DIR/$tdir/d1
+ chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/d1
+ $RUNAS createmany -o $DIR/$tdir/d1/o 100 || error "create failed"
+
echo "Inject failure stub to skip OST-object owner changing"
#define OBD_FAIL_LFSCK_BAD_OWNER 0x1613
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1613
run_test 16 "LFSCK can repair inconsistent MDT-object/OST-object owner"
test_17() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3594"
+
echo "#####"
echo "If more than one MDT-objects reference the same OST-object,"
echo "and the OST-object only recognizes one MDT-object, then the"
$LCTL set_param debug=+cache > /dev/null
test_18a() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3336"
+
echo "#####"
echo "The target MDT-object is there, but related stripe information"
echo "is lost or partly lost. The LFSCK should regenerate the missing"
run_test 18a "Find out orphan OST-object and repair it (1)"
test_18b() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3336"
+
echo "#####"
echo "The target MDT-object is lost. The LFSCK should re-create the"
echo "MDT-object under .lustre/lost+found/MDTxxxx. The admin should"
cancel_lru_locks mdc
cancel_lru_locks osc
+ # dryrun mode only check orphans, not repaie
+ echo "Trigger layout LFSCK --dryrun to find out orphan OST-object"
+ $START_LAYOUT --dryrun -o -r ||
+ error "Fail to start layout LFSCK in dryrun mode"
+ wait_all_targets_blocked layout completed 2
+
+ local PARAMS=$($SHOW_LAYOUT | awk '/^param/ { print $2 }')
+ [ "$PARAMS" == "dryrun,all_targets,orphan" ] ||
+ error "Expect 'dryrun,all_targets,orphan', got '$PARAMS'"
+
+ local orphans=$(do_facet mds1 $LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_layout |
+ awk '/^inconsistent_orphan/ { print $2 }')
+ [ $orphans -eq 3 ] ||
+ error "Expect 3 found on mds1, but got: $orphans"
+
+ # orphan parents should not be created
+ local subdir
+ for subdir in $MOUNT/.lustre/lost+found/*; do
+ [ ! "$(ls -A $subdir)" ] || error "$subdir not empty"
+ done
+
echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
$START_LAYOUT -r -o || error "(1) Fail to start LFSCK for layout!"
run_test 18b "Find out orphan OST-object and repair it (2)"
test_18c() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3336"
+
echo "#####"
echo "The target MDT-object is lost, and the OST-object FID is missing."
echo "The LFSCK should re-create the MDT-object with new FID under the "
run_test 18c "Find out orphan OST-object and repair it (3)"
test_18d() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3336"
+
echo "#####"
echo "The target MDT-object layout EA is corrupted, but the right"
echo "OST-object is still alive as orphan. The layout LFSCK will"
run_test 18d "Find out orphan OST-object and repair it (4)"
test_18e() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3336"
+
echo "#####"
echo "The target MDT-object layout EA slot is occpuied by some new"
echo "created OST-object when repair dangling reference case. Such"
run_test 18f "Skip the failed OST(s) when handle orphan OST-objects"
test_18g() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+
echo "#####"
echo "The target MDT-object is lost, but related OI mapping is there"
echo "The LFSCK should recreate the lost MDT-object without affected"
check_mount_and_prep
- $LFS setstripe -E 2M -c 1 -E -1 $DIR/$tdir/f0 ||
+ $LFS setstripe -E 2M -S 1M -c 1 -E -1 $DIR/$tdir/f0 ||
error "(0) Fail to create PFL $DIR/$tdir/f0"
cat $LUSTRE/tests/test-framework.sh > $DIR/$tdir/f0 ||
$LCTL set_param debug=-cache > /dev/null
test_19a() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3951"
+
check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
run_test 19a "OST-object inconsistency self detect"
test_19b() {
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-3951"
+
check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
run_test 19b "OST-object inconsistency self repair"
PATTERN_WITH_HOLE="40000001"
-PATTERN_WITHOUT_HOLE="1"
+PATTERN_WITHOUT_HOLE="raid0"
test_20a() {
[ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs" && return
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-4887"
echo "#####"
echo "The target MDT-object and some of its OST-object are lost."
test_20b() {
[ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs" && return
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ (( $MDS1_VERSION > $(version_code 2.5.55) )) ||
+ skip "MDS older than 2.5.55, LU-4887"
echo "#####"
echo "The target MDT-object and some of its OST-object are lost."
run_test 20b "Handle the orphan with dummy LOV EA slot properly - PFL case"
test_21() {
- [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.59) ]] &&
- skip "ignore the test if MDS is older than 2.5.59" && return
+ (( $MDS1_VERSION > $(version_code 2.5.59) )) ||
+ skip "MDS older than 2.5.59, LU-4887"
check_mount_and_prep
createmany -o $DIR/$tdir/f 100 || error "(0) Fail to create 100 files"
test_22a() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5511"
echo "#####"
echo "The parent_A references the child directory via some name entry,"
test_22b() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5511"
echo "#####"
echo "The parent_A references the child directory via the name entry_B,"
test_23a() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5512"
echo "#####"
echo "The name entry is there, but the MDT-object for such name "
run_test 23a "LFSCK can repair dangling name entry (1)"
test_23b() {
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5512"
+
echo "#####"
echo "The objectA has multiple hard links, one of them corresponding"
echo "to the name entry_B. But there is something wrong for the name"
}
run_test 23b "LFSCK can repair dangling name entry (2)"
+cleanup_23c() {
+ do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ $SHOW_NAMESPACE
+ error "(10) unexpected status"
+ }
+
+ stop_full_debug_logging
+}
+
test_23c() {
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5512"
+
echo "#####"
echo "The objectA has multiple hard links, one of them corresponding"
echo "to the name entry_B. But there is something wrong for the name"
check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0 on MDT0"
- $LFS path2fid $DIR/$tdir/d0
+ parent_fid="$($LFS path2fid $DIR/$tdir/d0)"
+ echo "parent_fid=$parent_fid"
createmany -o $DIR/$tdir/d0/t 10 || error "(1.5) Fail to creatmany"
echo "dummy" > $DIR/$tdir/d0/f0 || error "(2) Fail to touch on MDT0"
- $LFS path2fid $DIR/$tdir/d0/f0
+ f0_fid="$($LFS path2fid $DIR/$tdir/d0/f0)"
+ echo "f0_fid=$f0_fid"
echo "dead" > $DIR/$tdir/d0/f1 || error "(3) Fail to touch on MDT0"
- $LFS path2fid $DIR/$tdir/d0/f1
+ f1_fid="$($LFS path2fid $DIR/$tdir/d0/f1)"
+ echo "f1_fid=$f1_fid"
- local SEQ0=$($LFS path2fid $DIR/$tdir/d0/f0 | awk -F':' '{print $1}')
- local SEQ1=$($LFS path2fid $DIR/$tdir/d0/f1 | awk -F':' '{print $1}')
-
- if [ "$SEQ0" != "$SEQ1" ]; then
+ if [ "${fid_f0/:.*/}" != "${fid_f1/:.*/}" ]; then
# To guarantee that the f0 and f1 are in the same FID seq
rm -f $DIR/$tdir/d0/f0 ||
error "(3.1) Fail to unlink $DIR/$tdir/d0/f0"
echo "dummy" > $DIR/$tdir/d0/f0 ||
error "(3.2) Fail to touch on MDT0"
- $LFS path2fid $DIR/$tdir/d0/f0
+ f0_fid="$($LFS path2fid $DIR/$tdir/d0/f0)"
+ echo "f0_fid=$f0_fid (replaced)"
fi
- local OID=$($LFS path2fid $DIR/$tdir/d0/f1 | awk -F':' '{print $2}')
- OID=$(printf %d $OID)
+ local oid=$(awk -F':' '{ printf $2 }' <<< $f1_fid)
echo "Inject failure stub on MDT0 to simulate dangling name entry"
#define OBD_FAIL_LFSCK_DANGLING3 0x1621
- do_facet $SINGLEMDS $LCTL set_param fail_val=$OID fail_loc=0x1621
+ do_facet $SINGLEMDS $LCTL set_param fail_val=$oid fail_loc=0x1621
ln $DIR/$tdir/d0/f0 $DIR/$tdir/d0/foo || error "(4) Fail to hard link"
do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
$START_NAMESPACE -r -C ||
error "(7) Fail to start LFSCK for namespace"
- wait_update_facet client "stat $DIR/$tdir/d0/foo |
- awk '/Size/ { print \\\$2 }'" "0" $LTIME || {
+ wait_update_facet client "stat -c%s $DIR/$tdir/d0/foo" "0" $LTIME || {
+ # While unexpected by the test, it is valid for LFSCK to repair
+ # the link to the original object before any data is written.
+ local size=$(stat -c %s $DIR/$tdir/d0/foo)
+
+ if [ "$size" = "6" -a "$(<$DIR/$tdir/d0/foo)" = "dummy" ]; then
+ log "LFSCK repaired file prematurely"
+ cleanup_23c
+ return 0
+ fi
+
stat $DIR/$tdir/d0/foo
$SHOW_NAMESPACE
error "(8) unexpected size"
echo "data" >> $DIR/$tdir/d0/foo || error "(9) Fail to write"
cancel_lru_locks osc
- do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
- wait_update_facet $SINGLEMDS "$LCTL get_param -n \
- mdd.${MDT_DEV}.lfsck_namespace |
- awk '/^status/ { print \\\$2 }'" "completed" 32 || {
- $SHOW_NAMESPACE
- error "(10) unexpected status"
- }
-
- stop_full_debug_logging
+ cleanup_23c
local repaired=$($SHOW_NAMESPACE |
awk '/^dangling_repaired/ { print $2 }')
test_24() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5513"
echo "#####"
echo "Two MDT-objects back reference the same name entry via their"
$LFS path2fid $DIR/$tdir/d0/dummy
local pfid
- if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+ if [ $mds1_FSTYPE != ldiskfs ]; then
pfid=$($LFS path2fid $DIR/$tdir/d0/guard)
else
pfid=$($LFS path2fid $DIR/$tdir/d0/dummy)
run_test 24 "LFSCK can repair multiple-referenced name entry"
test_25() {
- [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
- skip "ldiskfs only test" && return
+ [[ $mds1_FSTYPE == ldiskfs ]] || skip "only ldiskfs fixes dirent type"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5515"
echo "#####"
echo "The file type in the name entry does not match the file type"
run_test 25 "LFSCK can repair bad file type in the name entry"
test_26a() {
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5516"
+
echo "#####"
echo "The local name entry back referenced by the MDT-object is lost."
echo "The namespace LFSCK will add the missing local name entry back"
run_test 26a "LFSCK can add the missing local name entry back to the namespace"
test_26b() {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5516"
echo "#####"
echo "The remote name entry back referenced by the MDT-object is lost."
run_test 26b "LFSCK can add the missing remote name entry back to the namespace"
test_27a() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5516"
+
echo "#####"
echo "The local parent referenced by the MDT-object linkEA is lost."
echo "The namespace LFSCK will re-create the lost parent as orphan."
run_test 27a "LFSCK can recreate the lost local parent directory as orphan"
test_27b() {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5516"
echo "#####"
echo "The remote parent referenced by the MDT-object linkEA is lost."
run_test 27b "LFSCK can recreate the lost remote parent directory as orphan"
test_28() {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5506"
echo "#####"
echo "The target name entry is lost. The LFSCK should insert the"
run_test 28 "Skip the failed MDT(s) when handle orphan MDT-objects"
test_29a() {
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5517"
+
echo "#####"
echo "The object's nlink attribute is larger than the object's known"
echo "name entries count. The LFSCK will repair the object's nlink"
#run_test 29a "LFSCK can repair bad nlink count (1)"
test_29b() {
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5517"
+
echo "#####"
echo "The object's nlink attribute is smaller than the object's known"
echo "name entries count. The LFSCK will repair the object's nlink"
test_29c()
{
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5517"
+
echo "#####"
echo "The namespace LFSCK will create many hard links to the target"
echo "file as to exceed the linkEA size limitation. Under such case"
cancel_lru_locks mdc
if [ $MDSCOUNT -ge 2 ]; then
- $LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null ||
- error "(3.1) Migrate failure"
+ $LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null &&
+ error "(3.1) Migrate should fail"
echo "The object with linkEA overflow should NOT be migrated"
local newfid=$($LFS path2fid $DIR/$tdir/guard/f0)
unlinkmany $DIR/$tdir/foo/ttttttttttt 100 || error "(4) Fail to unlink"
if [ $MDSCOUNT -ge 2 ]; then
- $LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null ||
- error "(5.1) Migrate failure"
+ $LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null &&
+ error "(5.1) Migrate should fail"
# The overflow timestamp is still there, so migration will fail.
local newfid=$($LFS path2fid $DIR/$tdir/guard/f0)
run_test 29c "verify linkEA size limitation"
test_30() {
- [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
- skip "ldiskfs only test" && return
+ [[ $mds1_FSTYPE == ldiskfs ]] || skip "only ldiskfs has lost+found"
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5518"
echo "#####"
echo "The namespace LFSCK will move the orphans from backend"
mkdir $DIR/$tdir/foo/d0 || error "(3) Fail to mkdir d0"
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+ local pfid=$($LFS path2fid $DIR/$tdir/foo)
+ local cfid=$($LFS path2fid $DIR/$tdir/foo/d0)
+
touch $DIR/$tdir/foo/d0/f1 || error "(4) Fail to touch f1"
mkdir $DIR/$tdir/foo/d0/d1 || error "(5) Fail to mkdir d1"
ls -ail $MOUNT/.lustre/lost+found/MDT0000/
- cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-*-D-*)
+ local cname=$MOUNT/.lustre/lost+found/MDT0000/${cfid}-${pfid}-D-0
[ ! -z "$cname" ] || error "(20) d0 is not recovered"
- stat ${cname}/d1 || error "(21) d0 is not recovered"
+ stat ${cname}/d1 || error "(21) d1 is not recovered"
stat ${cname}/f1 || error "(22) f1 is not recovered"
}
run_test 30 "LFSCK can recover the orphans from backend /lost+found"
test_31a() {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5519"
echo "#####"
echo "For the name entry under a striped directory, if the name"
[ $repaired -ge 1 ] ||
error "(5) Fail to repair bad name hash: $repaired"
+ local rc=$($LFS find -H badtype $DIR/$tdir/striped_dir | wc -l)
+ [ $rc -ge 1 ] ||
+ error "Fail to find flag bad type: $rc"
+
umount_client $MOUNT || error "(6) umount failed"
mount_client $MOUNT || error "(7) mount failed"
run_test 31a "The LFSCK can find/repair the name entry with bad name hash (1)"
test_31b() {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5519"
echo "#####"
echo "For the name entry under a striped directory, if the name"
#define OBD_FAIL_LFSCK_BAD_NAME_HASH 0x1628
$LCTL set_param fail_loc=0x1628 fail_val=1
- createmany -d $DIR/$tdir/striped_dir/d $MDSCOUNT ||
+ createmany -d $DIR/$tdir/striped_dir/d $((MDSCOUNT * 5)) ||
error "(2) Fail to create file under striped directory"
$LCTL set_param fail_loc=0 fail_val=0
local repaired=$(do_facet mds2 $LCTL get_param -n \
mdd.$(facet_svc mds2).lfsck_namespace |
awk '/^name_hash_repaired/ { print $2 }')
+ echo "repaired $repaired name entries with bad hash"
[ $repaired -ge 1 ] ||
error "(5) Fail to repair bad name hash: $repaired"
umount_client $MOUNT || error "(6) umount failed"
mount_client $MOUNT || error "(7) mount failed"
- for ((i = 0; i < $MDSCOUNT; i++)); do
+ for ((i = 0; i < $((MDSCOUNT * 5)); i++)); do
stat $DIR/$tdir/striped_dir/d$i ||
error "(8) Fail to stat d$i after LFSCK"
rmdir $DIR/$tdir/striped_dir/d$i ||
run_test 31b "The LFSCK can find/repair the name entry with bad name hash (2)"
test_31c() {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5519"
echo "#####"
echo "For some reason, the master MDT-object of the striped directory"
[ $repaired -eq 1 ] ||
error "(4) Fail to re-generate master LMV EA: $repaired"
+ local rc=$($LFS find -H lostlmv $DIR/$tdir/striped_dir | wc -l)
+ [ $rc -eq 1 ] || error "Fail to find flag lost LMV: $rc"
+
umount_client $MOUNT || error "(5) umount failed"
mount_client $MOUNT || error "(6) mount failed"
run_test 31c "Re-generate the lost master LMV EA for striped directory"
test_31d() {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5519"
echo "#####"
echo "For some reason, the master MDT-object of the striped directory"
run_test 31d "Set broken striped directory (modified after broken) as read-only"
test_31e() {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5519"
echo "#####"
echo "For some reason, the slave MDT-object of the striped directory"
test_31f() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5519"
echo "#####"
echo "For some reason, the slave MDT-object of the striped directory"
test_31g() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5519"
echo "#####"
echo "For some reason, the stripe index in the slave LMV EA is"
test_31h() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ (( $MDS1_VERSION > $(version_code 2.6.50) )) ||
+ skip "MDS older than 2.6.50, LU-5519"
echo "#####"
echo "For some reason, the shard's name entry in the striped"
}
run_test 31h "Repair the corrupted shard's name entry"
-test_32()
+test_32a()
{
lfsck_prep 5 5
umount_client $MOUNT
- #define OBD_FAIL_LFSCK_ASSISTANT_DIRECT 0x162d
+ #define OBD_FAIL_LFSCK_ENGINE_DELAY 0x162d
do_facet $SINGLEMDS $LCTL set_param fail_val=3 fail_loc=0x162d
- $START_LAYOUT -r || error "(2) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
local STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
- error "(3) Expect 'scanning-phase1', but got '$STATUS'"
+ error "(2) Expect 'scanning-phase1', but got '$STATUS'"
echo "stop ost1"
- stop ost1 > /dev/null || error "(4) Fail to stop OST1!"
+ stop ost1 > /dev/null || error "(3) Fail to stop OST1!"
do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
- sleep 1
+ sleep 4
echo "stop LFSCK"
- $STOP_LFSCK || error "(5) Fail to stop LFSCK!"
+ $STOP_LFSCK || error "(4) Fail to stop LFSCK!"
+
+ start ost1 $(ostdevname 1) $MOUNT_OPTS_NOSCRUB > /dev/null ||
+ error "(5) Fail to start ost1"
}
-run_test 32 "stop LFSCK when some OST failed"
+run_test 32a "stop LFSCK when some OST failed"
+
+test_32b()
+{
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+ lfsck_prep 5 5
+ $LFS mkdir -i 1 $DIR/$tdir/dp ||
+ error "(1) Fail to create $DIR/$tdir/dp"
+ $LFS mkdir -i 0 -c $MDSCOUNT $DIR/$tdir/dp/dc1 ||
+ error "(2) Fail to create $DIR/$tdir/dp/dc1"
+ $LFS mkdir -i 0 -c $MDSCOUNT $DIR/$tdir/dp/dc2 ||
+ error "(3) Fail to create $DIR/$tdir/dp/dc2"
+ umount_client $MOUNT
+
+ #define OBD_FAIL_LFSCK_ENGINE_DELAY 0x162d
+ do_facet $SINGLEMDS $LCTL set_param fail_val=3 fail_loc=0x162d
+ $START_NAMESPACE -r -A || error "(4) Fail to start LFSCK for namespace!"
+
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "scanning-phase1" 32 || {
+ $SHOW_NAMESPACE
+ error "(5) unexpected status"
+ }
+
+ echo "stop mds2"
+ stop mds2 > /dev/null || error "(6) Fail to stop MDT2!"
+
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
+ sleep 4
+
+ echo "stop LFSCK"
+ $STOP_LFSCK || error "(7) Fail to stop LFSCK!"
+
+ start mds2 $(mdsdevname 2) $MOUNT_OPTS_NOSCRUB > /dev/null ||
+ error "(8) Fail to start MDT2"
+}
+run_test 32b "stop LFSCK when some MDT failed"
test_33()
{
}
run_test 33 "check LFSCK paramters"
+test_34()
+{
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ [ "$mds1_FSTYPE" != zfs ] && skip "Only valid for ZFS backend"
+
+ lfsck_prep 1 1
+
+ #define OBD_FAIL_LFSCK_NO_AGENTOBJ 0x1630
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1630
+ $LFS mkdir -i 1 $DIR/$tdir/dummy ||
+ error "(1) Fail to create $DIR/$tdir/dummy"
+
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+ $START_NAMESPACE -r || error "(2) Fail to start LFSCK for namespace!"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ $SHOW_NAMESPACE
+ error "(3) unexpected status"
+ }
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^dirent_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(4) Fail to repair the lost agent object: $repaired"
+
+ $START_NAMESPACE -r || error "(5) Fail to start LFSCK for namespace!"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ $SHOW_NAMESPACE
+ error "(6) unexpected status"
+ }
+
+ repaired=$($SHOW_NAMESPACE | awk '/^dirent_repaired/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(7) Unexpected repairing: $repaired"
+}
+run_test 34 "LFSCK can rebuild the lost agent object"
+
+test_35()
+{
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+ lfsck_prep 1 1
+
+ #define OBD_FAIL_LFSCK_NO_AGENTENT 0x1631
+ do_facet mds2 $LCTL set_param fail_loc=0x1631
+ $LFS mkdir -i 1 $DIR/$tdir/dummy ||
+ error "(1) Fail to create $DIR/$tdir/dummy"
+
+ sync; sleep 3
+ do_facet mds2 $LCTL set_param fail_loc=0
+ $START_NAMESPACE -A -r || error "(2) Fail to start LFSCK for namespace!"
+ wait_update_facet mds2 "$LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
+ error "(3) MDS${k} is not the expected 'completed'"
+
+ local repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^agent_entries_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(4) Fail to repair the lost agent entry: $repaired"
+
+ echo "stopall to cleanup object cache"
+ stopall > /dev/null
+ echo "setupall"
+ setupall > /dev/null
+
+ $START_NAMESPACE -A -r || error "(5) Fail to start LFSCK for namespace!"
+ wait_update_facet mds2 "$LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
+ error "(6) MDS${k} is not the expected 'completed'"
+
+ repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^agent_entries_repaired/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(7) Unexpected repairing: $repaired"
+}
+run_test 35 "LFSCK can rebuild the lost agent entry"
+
+test_36a() {
+ [ $OSTCOUNT -lt 3 ] && skip "needs >= 3 OSTs" && return
+
+ echo "#####"
+ echo "The target MDT-object's LOV EA corrupted as to lose one of the "
+ echo "mirrors information. The layout LFSCK should rebuild the LOV EA "
+ echo "with the PFID EA of related OST-object(s) belong to the mirror."
+ echo "#####"
+
+ check_mount_and_prep
+
+ lfs df $DIR
+ lfs df -i $DIR
+ lctl get_param osc.*.*grant*
+ stack_trap "lfs df $DIR; lfs df -i $DIR; lctl get_param osc.*.*grant*"
+
+ $LFS setstripe -N -E 1M -o 0,1 -E -1 -o 2 -N -E 2M -o 1,2 -E -1 -o 0 \
+ -N -E 3M -o 2,0 -E -1 -o 1 $DIR/$tdir/f0 ||
+ error "(0) Fail to create mirror file $DIR/$tdir/f0"
+ $LFS setstripe -N -E 1M -o 0,1 -E -1 -o 2 -N -E 2M -o 1,2 -E -1 -o 0 \
+ -N -E 3M -o 2,0 -E -1 -o 1 $DIR/$tdir/f1 ||
+ error "(1) Fail to create mirror file $DIR/$tdir/f1"
+ $LFS setstripe -N -E 1M -o 0,1 -E -1 -o 2 -N -E 2M -o 1,2 -E -1 -o 0 \
+ -N -E 3M -o 2,0 -E -1 -o 1 $DIR/$tdir/f2 ||
+ error "(2) Fail to create mirror file $DIR/$tdir/f2"
+
+ dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=4 ||
+ error "(3) Fail to write $DIR/$tdir/f0"
+ dd if=/dev/zero of=$DIR/$tdir/f1 bs=1M count=4 ||
+ error "(4) Fail to write $DIR/$tdir/f1"
+ dd if=/dev/zero of=$DIR/$tdir/f2 bs=1M count=4 ||
+ error "(5) Fail to write $DIR/$tdir/f2"
+
+ $LFS mirror resync $DIR/$tdir/f0 ||
+ error "(6) Fail to resync $DIR/$tdir/f0"
+ $LFS mirror resync $DIR/$tdir/f1 ||
+ error "(7) Fail to resync $DIR/$tdir/f1"
+ $LFS mirror resync $DIR/$tdir/f2 ||
+ error "(8) Fail to resync $DIR/$tdir/f2"
+
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
+
+ $LFS getstripe $DIR/$tdir/f0 ||
+ error "(9) Fail to getstripe for $DIR/$tdir/f0"
+ $LFS getstripe $DIR/$tdir/f1 ||
+ error "(10) Fail to getstripe for $DIR/$tdir/f1"
+ $LFS getstripe $DIR/$tdir/f2 ||
+ error "(11) Fail to getstripe for $DIR/$tdir/f2"
+
+ echo "Inject failure, to simulate the case of missing one mirror in LOV"
+ #define OBD_FAIL_LFSCK_LOST_MDTOBJ 0x1616
+ do_facet mds1 $LCTL set_param fail_loc=0x1616
+
+ $LFS mirror split --mirror-id 1 -d $DIR/$tdir/f0 ||
+ error "(12) Fail to split 1st mirror from $DIR/$tdir/f0"
+ $LFS mirror split --mirror-id 2 -d $DIR/$tdir/f1 ||
+ error "(13) Fail to split 2nd mirror from $DIR/$tdir/f1"
+ $LFS mirror split --mirror-id 3 -d $DIR/$tdir/f2 ||
+ error "(14) Fail to split 3rd mirror from $DIR/$tdir/f2"
+
+ sync
+ sleep 2
+ do_facet mds1 $LCTL set_param fail_loc=0
+
+ $LFS getstripe $DIR/$tdir/f0 | grep "lcme_mirror_id:.*1" &&
+ error "(15) The 1st of mirror is not destroyed"
+ $LFS getstripe $DIR/$tdir/f1 | grep "lcme_mirror_id:.*2" &&
+ error "(16) The 2nd of mirror is not destroyed"
+ $LFS getstripe $DIR/$tdir/f2 | grep "lcme_mirror_id:.*3" &&
+ error "(17) The 3rd of mirror is not destroyed"
+
+ local mirrors
+
+ mirrors=$($LFS getstripe -N $DIR/$tdir/f0)
+ [ $mirrors -eq 2 ] || error "(18) $DIR/$tdir/f0 has $mirrors mirrors"
+ mirrors=$($LFS getstripe -N $DIR/$tdir/f1)
+ [ $mirrors -eq 2 ] || error "(19) $DIR/$tdir/f1 has $mirrors mirrors"
+ mirrors=$($LFS getstripe -N $DIR/$tdir/f2)
+ [ $mirrors -eq 2 ] || error "(20) $DIR/$tdir/f2 has $mirrors mirrors"
+
+ echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+ $START_LAYOUT -r -o || error "(21) Fail to start LFSCK for layout!"
+
+ for k in $(seq $MDSCOUNT); do
+ # The LFSCK status query internal is 30 seconds. For the case
+ # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
+ # time to guarantee the status sync up.
+ wait_update_facet mds${k} "$LCTL get_param -n \
+ mdd.$(facet_svc mds${k}).lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ error "(22) MDS${k} is not the expected 'completed'"
+ done
+
+ for k in $(seq $OSTCOUNT); do
+ local cur_status=$(do_facet ost${k} $LCTL get_param -n \
+ obdfilter.$(facet_svc ost${k}).lfsck_layout |
+ awk '/^status/ { print $2 }')
+ [ "$cur_status" == "completed" ] ||
+ error "(23) OST${k} Expect 'completed', but got '$cur_status'"
+ done
+
+ local repaired=$(do_facet mds1 $LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_layout |
+ awk '/^repaired_orphan/ { print $2 }')
+ [ $repaired -eq 9 ] ||
+ error "(24) Expect 9 fixed on mds1, but got: $repaired"
+
+ mirrors=$($LFS getstripe -N $DIR/$tdir/f0)
+ [ $mirrors -eq 3 ] || error "(25) $DIR/$tdir/f0 has $mirrors mirrors"
+ mirrors=$($LFS getstripe -N $DIR/$tdir/f1)
+ [ $mirrors -eq 3 ] || error "(26) $DIR/$tdir/f1 has $mirrors mirrors"
+ mirrors=$($LFS getstripe -N $DIR/$tdir/f2)
+ [ $mirrors -eq 3 ] || error "(27) $DIR/$tdir/f2 has $mirrors mirrors"
+
+ $LFS getstripe $DIR/$tdir/f0 | grep "lcme_mirror_id:.*1" || {
+ $LFS getstripe $DIR/$tdir/f0
+ error "(28) The 1st of mirror is not recovered"
+ }
+
+ $LFS getstripe $DIR/$tdir/f1 | grep "lcme_mirror_id:.*2" || {
+ $LFS getstripe $DIR/$tdir/f1
+ error "(29) The 2nd of mirror is not recovered"
+ }
+
+ $LFS getstripe $DIR/$tdir/f2 | grep "lcme_mirror_id:.*3" || {
+ $LFS getstripe $DIR/$tdir/f2
+ error "(30) The 3rd of mirror is not recovered"
+ }
+}
+run_test 36a "rebuild LOV EA for mirrored file (1)"
+
+test_36b() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ [ $OSTCOUNT -lt 3 ] && skip "needs >= 3 OSTs" && return
+
+ echo "#####"
+ echo "The mirrored file lost its MDT-object, but relatd OST-objects "
+ echo "are still there. The layout LFSCK should rebuild the LOV EA "
+ echo "with the PFID EA of related OST-object(s) belong to the file. "
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS setstripe -N -E 1M -o 0,1 -E -1 -o 2 -N -E 2M -o 1,2 -E -1 -o 0 \
+ -N -E 3M -o 2,0 -E -1 -o 1 $DIR/$tdir/f0 ||
+ error "(0) Fail to create mirror file $DIR/$tdir/f0"
+
+ local fid=$($LFS path2fid $DIR/$tdir/f0)
+
+ dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=4 ||
+ error "(1) Fail to write $DIR/$tdir/f0"
+ $LFS mirror resync $DIR/$tdir/f0 ||
+ error "(2) Fail to resync $DIR/$tdir/f0"
+
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
+
+ $LFS getstripe $DIR/$tdir/f0 ||
+ error "(3) Fail to getstripe for $DIR/$tdir/f0"
+
+ echo "Inject failure, to simulate the case of missing the MDT-object"
+ #define OBD_FAIL_LFSCK_LOST_MDTOBJ 0x1616
+ do_facet mds1 $LCTL set_param fail_loc=0x1616
+ rm -f $DIR/$tdir/f0 || error "(4) Fail to remove $DIR/$tdir/f0"
+
+ sync
+ sleep 2
+ do_facet mds1 $LCTL set_param fail_loc=0
+
+ echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+ $START_LAYOUT -r -o || error "(5) Fail to start LFSCK for layout!"
+
+ for k in $(seq $MDSCOUNT); do
+ # The LFSCK status query internal is 30 seconds. For the case
+ # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
+ # time to guarantee the status sync up.
+ wait_update_facet mds${k} "$LCTL get_param -n \
+ mdd.$(facet_svc mds${k}).lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ error "(6) MDS${k} is not the expected 'completed'"
+ done
+
+ for k in $(seq $OSTCOUNT); do
+ local cur_status=$(do_facet ost${k} $LCTL get_param -n \
+ obdfilter.$(facet_svc ost${k}).lfsck_layout |
+ awk '/^status/ { print $2 }')
+ [ "$cur_status" == "completed" ] ||
+ error "(7) OST${k} Expect 'completed', but got '$cur_status'"
+ done
+
+ local count=$(do_facet mds1 $LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_layout |
+ awk '/^repaired_orphan/ { print $2 }')
+ [ $count -eq 9 ] || error "(8) Expect 9 fixed on mds1, but got: $count"
+
+ local name=$MOUNT/.lustre/lost+found/MDT0000/${fid}-R-0
+ count=$($LFS getstripe --mirror-count $name)
+ [ $count -eq 3 ] || error "(9) $DIR/$tdir/f0 has $count mirrors"
+
+ count=$($LFS getstripe --component-count $name)
+ [ $count -eq 6 ] || error "(10) $DIR/$tdir/f0 has $count components"
+
+ $LFS getstripe $name | grep "lcme_mirror_id:.*1" || {
+ $LFS getstripe $name
+ error "(11) The 1st of mirror is not recovered"
+ }
+
+ $LFS getstripe $name | grep "lcme_mirror_id:.*2" || {
+ $LFS getstripe $name
+ error "(12) The 2nd of mirror is not recovered"
+ }
+
+ $LFS getstripe $name | grep "lcme_mirror_id:.*3" || {
+ $LFS getstripe $name
+ error "(13) The 3rd of mirror is not recovered"
+ }
+}
+run_test 36b "rebuild LOV EA for mirrored file (2)"
+
+test_36c() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+ [ $OSTCOUNT -lt 3 ] && skip "needs >= 3 OSTs" && return
+
+ echo "#####"
+ echo "The mirrored file has been modified, not resynced yet, then "
+ echo "lost its MDT-object, but relatd OST-objects are still there. "
+ echo "The layout LFSCK should rebuild the LOV EA and relatd status "
+ echo "with the PFID EA of related OST-object(s) belong to the file. "
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS setstripe -N -E 1M -o 0,1 -E -1 -o 2 -N -E 2M -o 1,2 -E -1 -o 0 \
+ $DIR/$tdir/f0 ||
+ error "(0) Fail to create mirror file $DIR/$tdir/f0"
+
+ local fid=$($LFS path2fid $DIR/$tdir/f0)
+
+ # The 1st dd && resync makes all related OST-objects have been written
+ dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=4 ||
+ error "(1.1) Fail to write $DIR/$tdir/f0"
+ $LFS mirror resync $DIR/$tdir/f0 ||
+ error "(1.2) Fail to resync $DIR/$tdir/f0"
+ # The 2nd dd makes one mirror to be stale
+ dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=4 ||
+ error "(1.3) Fail to write $DIR/$tdir/f0"
+
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
+
+ $LFS getstripe $DIR/$tdir/f0 ||
+ error "(2) Fail to getstripe for $DIR/$tdir/f0"
+
+ local saved_flags1=$($LFS getstripe $DIR/$tdir/f0 | head -n 10 |
+ awk '/lcme_flags/ { print $2 }')
+ local saved_flags2=$($LFS getstripe $DIR/$tdir/f0 | tail -n 10 |
+ awk '/lcme_flags/ { print $2 }')
+
+ echo "Inject failure, to simulate the case of missing the MDT-object"
+ #define OBD_FAIL_LFSCK_LOST_MDTOBJ 0x1616
+ do_facet mds1 $LCTL set_param fail_loc=0x1616
+ rm -f $DIR/$tdir/f0 || error "(3) Fail to remove $DIR/$tdir/f0"
+
+ sync
+ sleep 2
+ do_facet mds1 $LCTL set_param fail_loc=0
+
+ echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+ $START_LAYOUT -r -o || error "(4) Fail to start LFSCK for layout!"
+
+ for k in $(seq $MDSCOUNT); do
+ # The LFSCK status query internal is 30 seconds. For the case
+ # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
+ # time to guarantee the status sync up.
+ wait_update_facet mds${k} "$LCTL get_param -n \
+ mdd.$(facet_svc mds${k}).lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ error "(5) MDS${k} is not the expected 'completed'"
+ done
+
+ for k in $(seq $OSTCOUNT); do
+ local cur_status=$(do_facet ost${k} $LCTL get_param -n \
+ obdfilter.$(facet_svc ost${k}).lfsck_layout |
+ awk '/^status/ { print $2 }')
+ [ "$cur_status" == "completed" ] ||
+ error "(6) OST${k} Expect 'completed', but got '$cur_status'"
+ done
+
+ local count=$(do_facet mds1 $LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_layout |
+ awk '/^repaired_orphan/ { print $2 }')
+ [ $count -eq 6 ] || error "(7) Expect 9 fixed on mds1, but got: $count"
+
+ local name=$MOUNT/.lustre/lost+found/MDT0000/${fid}-R-0
+ count=$($LFS getstripe --mirror-count $name)
+ [ $count -eq 2 ] || error "(8) $DIR/$tdir/f0 has $count mirrors"
+
+ count=$($LFS getstripe --component-count $name)
+ [ $count -eq 4 ] || error "(9) $DIR/$tdir/f0 has $count components"
+
+ local flags=$($LFS getstripe $name | head -n 10 |
+ awk '/lcme_flags/ { print $2 }')
+ [ "$flags" == "$saved_flags1" ] || {
+ $LFS getstripe $name
+ error "(10) expect flags $saved_flags1, got $flags"
+ }
+
+ flags=$($LFS getstripe $name | tail -n 10 |
+ awk '/lcme_flags/ { print $2 }')
+ [ "$flags" == "$saved_flags2" ] || {
+ $LFS getstripe $name
+ error "(11) expect flags $saved_flags2, got $flags"
+ }
+}
+run_test 36c "rebuild LOV EA for mirrored file (3)"
+
+test_37()
+{
+ local PID
+ local rc
+ local t_dir="$DIR/$tdir/d0"
+ check_mount_and_prep
+
+ $LFS mkdir -i 0 $t_dir || error "(2) Fail to mkdir $t_dir on MDT0"
+ multiop_bg_pause $t_dir D_c || { error "multiop failed: $?"; return 1; }
+ PID=$!
+ rmdir $t_dir
+
+ $START_NAMESPACE -r -A || {
+ error "(3) Fail to start LFSCK for namespace!"; kill -USR1 $PID; }
+
+ wait_all_targets_blocked namespace completed 4
+ stat $t_dir && rc=1
+ kill -USR1 $PID
+ return $rc
+}
+run_test 37 "LFSCK must skip a ORPHAN"
+
+test_38()
+{
+ [[ "$MDS1_VERSION" -le $(version_code 2.12.51) ]] &&
+ skip "Need MDS version newer than 2.12.51"
+
+ test_mkdir $DIR/$tdir
+ local uuid1=$(cat /proc/sys/kernel/random/uuid)
+ local uuid2=$(cat /proc/sys/kernel/random/uuid)
+
+ # create foreign file
+ $LFS setstripe --foreign=none --flags 0xda05 \
+ -x "${uuid1}@${uuid2}" $DIR/$tdir/$tfile ||
+ error "$DIR/$tdir/$tfile: create failed"
+
+ $LFS getstripe -v $DIR/$tdir/$tfile |
+ grep "lfm_magic:.*0x0BD70BD0" ||
+ error "$DIR/$tdir/$tfile: invalid LOV EA foreign magic"
+ # lfm_length is LOV EA size - sizeof(lfm_magic) - sizeof(lfm_length)
+ $LFS getstripe -v $DIR/$tdir/$tfile | grep "lfm_length:.*73" ||
+ error "$DIR/$tdir/$tfile: invalid LOV EA foreign size"
+ $LFS getstripe -v $DIR/$tdir/$tfile | grep "lfm_type:.*none" ||
+ error "$DIR/$tdir/$tfile: invalid LOV EA foreign type"
+ $LFS getstripe -v $DIR/$tdir/$tfile |
+ grep "lfm_flags:.*0x0000DA05" ||
+ error "$DIR/$tdir/$tfile: invalid LOV EA foreign flags"
+ $LFS getstripe $DIR/$tdir/$tfile |
+ grep "lfm_value:.*${uuid1}@${uuid2}" ||
+ error "$DIR/$tdir/$tfile: invalid LOV EA foreign value"
+
+ # modify striping should fail
+ $LFS setstripe -c 2 $DIR/$tdir/$tfile &&
+ error "$DIR/$tdir/$tfile: setstripe should fail"
+
+ $START_NAMESPACE -r -A || error "Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 1
+
+ # check that "global" namespace_repaired == 0 !!!
+ local repaired=$(do_facet mds1 \
+ "$LCTL lfsck_query -t all -M ${FSNAME}-MDT0000 |
+ awk '/^namespace_repaired/ { print \\\$2 }'")
+ [ $repaired -eq 0 ] ||
+ error "(2) Expect no namespace repair, but got: $repaired"
+
+ $START_LAYOUT -A -r || error "Fail to start LFSCK for layout"
+
+ wait_all_targets_blocked layout completed 2
+
+ # check that "global" layout_repaired == 0 !!!
+ local repaired=$(do_facet mds1 \
+ "$LCTL lfsck_query -t all -M ${FSNAME}-MDT0000 |
+ awk '/^layout_repaired/ { print \\\$2 }'")
+ [ $repaired -eq 0 ] ||
+ error "(2) Expect no layout repair, but got: $repaired"
+
+ echo "post-lfsck checks of foreign file"
+
+ $LFS getstripe -v $DIR/$tdir/$tfile |
+ grep "lfm_magic:.*0x0BD70BD0" ||
+ error "$DIR/$tdir/$tfile: invalid LOV EA foreign magic"
+ # lfm_length is LOV EA size - sizeof(lfm_magic) - sizeof(lfm_length)
+ $LFS getstripe -v $DIR/$tdir/$tfile | grep "lfm_length:.*73" ||
+ error "$DIR/$tdir/$tfile: invalid LOV EA foreign size"
+ $LFS getstripe -v $DIR/$tdir/$tfile | grep "lfm_type:.*none" ||
+ error "$DIR/$tdir/$tfile: invalid LOV EA foreign type"
+ $LFS getstripe -v $DIR/$tdir/$tfile |
+ grep "lfm_flags:.*0x0000DA05" ||
+ error "$DIR/$tdir/$tfile: invalid LOV EA foreign flags"
+ $LFS getstripe $DIR/$tdir/$tfile |
+ grep "lfm_value:.*${uuid1}@${uuid2}" ||
+ error "$DIR/$tdir/$tfile: invalid LOV EA foreign value"
+
+ # modify striping should fail
+ $LFS setstripe -c 2 $DIR/$tdir/$tfile &&
+ error "$DIR/$tdir/$tfile: setstripe should fail"
+
+ # R/W should fail
+ cat $DIR/$tdir/$tfile && "$DIR/$tdir/$tfile: read should fail"
+ cat /etc/passwd > $DIR/$tdir/$tfile &&
+ error "$DIR/$tdir/$tfile: write should fail"
+
+ #remove foreign file
+ rm $DIR/$tdir/$tfile ||
+ error "$DIR/$tdir/$tfile: remove of foreign file has failed"
+}
+run_test 38 "LFSCK does not break foreign file and reverse is also true"
+
+test_39()
+{
+ [[ "$MDS1_VERSION" -le $(version_code 2.12.51) ]] &&
+ skip "Need MDS version newer than 2.12.51"
+
+ test_mkdir $DIR/$tdir
+ local uuid1=$(cat /proc/sys/kernel/random/uuid)
+ local uuid2=$(cat /proc/sys/kernel/random/uuid)
+
+ # create foreign dir
+ $LFS mkdir --foreign=none --xattr="${uuid1}@${uuid2}" --flags=0xda05 \
+ $DIR/$tdir/${tdir}2 ||
+ error "$DIR/$tdir/${tdir}2: create failed"
+
+ $LFS getdirstripe -v $DIR/$tdir/${tdir}2 |
+ grep "lfm_magic:.*0x0CD50CD0" ||
+ error "$DIR/$tdir/${tdir}2: invalid LMV EA magic"
+ # lfm_length is LMV EA size - sizeof(lfm_magic) - sizeof(lfm_length)
+ # - sizeof(lfm_type) - sizeof(lfm_flags)
+ $LFS getdirstripe -v $DIR/$tdir/${tdir}2 | grep "lfm_length:.*73" ||
+ error "$DIR/$tdir/${tdir}2: invalid LMV EA size"
+ $LFS getdirstripe -v $DIR/$tdir/${tdir}2 | grep "lfm_type:.*none" ||
+ error "$DIR/$tdir/${tdir}2: invalid LMV EA type"
+ $LFS getdirstripe -v $DIR/$tdir/${tdir}2 |
+ grep "lfm_flags:.*0x0000DA05" ||
+ error "$DIR/$tdir/${tdir}2: invalid LMV EA flags"
+ $LFS getdirstripe $DIR/$tdir/${tdir}2 |
+ grep "lfm_value.*${uuid1}@${uuid2}" ||
+ error "$DIR/$tdir/${tdir}2: invalid LMV EA value"
+
+ # file create in dir should fail
+ touch $DIR/$tdir/${tdir}2/$tfile &&
+ "$DIR/${tdir}2: file create should fail"
+
+ # chmod should work
+ chmod 777 $DIR/$tdir/${tdir}2 ||
+ error "$DIR/${tdir}2: chmod failed"
+
+ # chown should work
+ chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/${tdir}2 ||
+ error "$DIR/${tdir}2: chown failed"
+
+ $START_NAMESPACE -r -A || error "Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 1
+
+ # check that "global" namespace_repaired == 0 !!!
+ local repaired=$(do_facet mds1 \
+ "$LCTL lfsck_query -t all -M ${FSNAME}-MDT0000 |
+ awk '/^namespace_repaired/ { print \\\$2 }'")
+ [ $repaired -eq 0 ] ||
+ error "(2) Expect nothing to be repaired, but got: $repaired"
+
+ $START_LAYOUT -A -r || error "Fail to start LFSCK for layout"
+
+ wait_all_targets_blocked layout completed 2
+
+ # check that "global" layout_repaired == 0 !!!
+ local repaired=$(do_facet mds1 \
+ "$LCTL lfsck_query -t all -M ${FSNAME}-MDT0000 |
+ awk '/^layout_repaired/ { print \\\$2 }'")
+ [ $repaired -eq 0 ] ||
+ error "(2) Expect no layout repair, but got: $repaired"
+
+ echo "post-lfsck checks of foreign dir"
+
+ $LFS getdirstripe -v $DIR/$tdir/${tdir}2 |
+ grep "lfm_magic:.*0x0CD50CD0" ||
+ error "$DIR/$tdir/${tdir}2: invalid LMV EA magic"
+ # lfm_length is LMV EA size - sizeof(lfm_magic) - sizeof(lfm_length)
+ # - sizeof(lfm_type) - sizeof(lfm_flags)
+ $LFS getdirstripe -v $DIR/$tdir/${tdir}2 | grep "lfm_length:.*73" ||
+ error "$DIR/$tdir/${tdir}2: invalid LMV EA size"
+ $LFS getdirstripe -v $DIR/$tdir/${tdir}2 | grep "lfm_type:.*none" ||
+ error "$DIR/$tdir/${tdir}2: invalid LMV EA type"
+ $LFS getdirstripe -v $DIR/$tdir/${tdir}2 |
+ grep "lfm_flags:.*0x0000DA05" ||
+ error "$DIR/$tdir/${tdir}2: invalid LMV EA flags"
+ $LFS getdirstripe $DIR/$tdir/${tdir}2 |
+ grep "lfm_value.*${uuid1}@${uuid2}" ||
+ error "$DIR/$tdir/${tdir}2: invalid LMV EA value"
+
+ # file create in dir should fail
+ touch $DIR/$tdir/${tdir}2/$tfile &&
+ "$DIR/${tdir}2: file create should fail"
+
+ # chmod should work
+ chmod 777 $DIR/$tdir/${tdir}2 ||
+ error "$DIR/${tdir}2: chmod failed"
+
+ # chown should work
+ chown $RUNAS_ID:$RUNAS_GID $DIR/$tdir/${tdir}2 ||
+ error "$DIR/${tdir}2: chown failed"
+
+ #remove foreign dir
+ rmdir $DIR/$tdir/${tdir}2 ||
+ error "$DIR/$tdir/${tdir}2: remove of foreign dir has failed"
+}
+run_test 39 "LFSCK does not break foreign dir and reverse is also true"
+
+test_40a() {
+ [[ $MDSCOUNT -ge 2 ]] || skip "needs >= 2 MDTs"
+
+ check_mount_and_prep
+ $LFS mkdir -i 1 $DIR/$tdir/dir1
+ $LFS setstripe -E 1M -c1 -S 1M -E 128M -c2 -S 4M -E eof $DIR/$tdir/dir1
+
+ touch $DIR/$tdir/dir1/f1
+ local layout1=$(get_layout_param $DIR/$tdir/dir1/f1)
+
+ echo "Migrate $DIR/$tdir/dir1 from MDT1 to MDT0"
+ $LFS migrate -m 0 $DIR/$tdir/dir1
+
+ echo "trigger LFSCK for layout"
+ do_facet $SINGLEMDS $LCTL lfsck_start -M ${MDT_DEV} -t layout -r
+
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ $SHOW_LAYOUT
+ error "(2) unexpected status"
+ }
+
+ local layout2=$(get_layout_param $DIR/$tdir/dir1/f1)
+
+ [[ "$layout1" == "$layout2" ]] || error "layout lost after lfsck"
+}
+run_test 40a "LFSCK correctly fixes lmm_oi in composite layout"
+
+test_41()
+{
+ local old_debug=$(do_facet $SINGLEMDS $LCTL get_param -n debug)
+
+ do_facet $SINGLEMDS $LCTL set_param debug=+lfsck
+ $LFS setstripe -E 1G -z 64M -E -1 -z 128M $DIR/$tfile
+ do_facet $SINGLEMDS $LCTL dk > /dev/null
+
+ echo "trigger LFSCK for SEL layout"
+ do_facet $SINGLEMDS $LCTL lfsck_start -M ${MDT_DEV} -A -t all -r -n on
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ $SHOW_LAYOUT
+ error "(2) unexpected status"
+ }
+
+ local errors=$(do_facet $SINGLEMDS $LCTL dk |
+ grep "lfsck_layout_verify_header")
+
+ [[ "x$errors" == "x" ]] || {
+ echo "$errors"
+ error "lfsck failed"
+ }
+
+ do_facet $SINGLEMDS "$LCTL set_param debug='$old_debug'"
+}
+run_test 41 "SEL support in LFSCK"
+
# restore MDS/OST size
MDSSIZE=${SAVED_MDSSIZE}
OSTSIZE=${SAVED_OSTSIZE}
OSTCOUNT=${SAVED_OSTCOUNT}
# cleanup the system at last
-formatall
+REFORMAT="yes" cleanup_and_setup_lustre
complete $SECONDS
+check_and_cleanup_lustre
exit_status