set -e
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!
require_dsh_mds || exit 0
+load_modules
+
+if ! check_versions; then
+ skip "It is NOT necessary to test lfsck under interoperation mode"
+ exit 0
+fi
+
+[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.60) ]] &&
+ skip "Need MDS version at least 2.3.60" && exit 0
+
+LTIME=${LTIME:-120}
+
SAVED_MDSSIZE=${MDSSIZE}
SAVED_OSTSIZE=${OSTSIZE}
SAVED_OSTCOUNT=${OSTCOUNT}
# do not use too small MDSSIZE/OSTSIZE, which affect the default journal size
MDSSIZE=100000
OSTSIZE=100000
-# no need too much OSTs, to reduce the format/start/stop overhead
+# no need too many OSTs, to reduce the format/start/stop overhead
[ $OSTCOUNT -gt 4 ] && OSTCOUNT=4
# build up a clean test environment.
formatall
setupall
-[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.60) ]] &&
- skip "Need MDS version at least 2.3.60" && check_and_cleanup_lustre &&
- exit 0
-
[[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.4.90) ]] &&
ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2c"
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"
+ ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2d 2e 3 22 23 24 25 26 27 28 29 30 31"
-build_test_filter
+# DNE does not support striped directory on zfs-based backend yet.
+[ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
+ ALWAYS_EXCEPT="$ALWAYS_EXCEPT 31"
-$LCTL set_param debug=+lfsck > /dev/null || true
+build_test_filter
MDT_DEV="${FSNAME}-MDT0000"
OST_DEV="${FSNAME}-OST0000"
$LCTL get_param -n obdfilter.${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"
lfsck_prep() {
local ndirs=$1
echo "prepared $(date)."
}
+run_e2fsck_on_mdt0() {
+ [ $(facet_fstype $SINGLEMDS) != ldiskfs ] && return
+
+ stop $SINGLEMDS > /dev/null || error "(0) Fail to the stop MDT0"
+ run_e2fsck $(facet_active_host $SINGLEMDS) $(mdsdevname 1) "-n" |
+ grep "Fix? no" && {
+ run_e2fsck $(facet_active_host $SINGLEMDS) $(mdsdevname 1) "-n"
+ error "(2) Detected inconsistency on MDT0"
+ }
+ start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
+ error "(3) Fail to start MDT0"
+}
+
+wait_all_targets_blocked() {
+ local com=$1
+ local status=$2
+ local err=$3
+
+ local count=$(do_facet mds1 \
+ "$LCTL lfsck_query -t $com -M ${FSNAME}-MDT0000 -w |
+ awk '/^${com}_mdts_${status}/ { print \\\$2 }'")
+ [[ $count -eq $MDSCOUNT ]] || {
+ do_facet mds1 "$LCTL lfsck_query -t $com -M ${FSNAME}-MDT0000"
+ error "($err) only $count of $MDSCOUNT MDTs are in ${status}"
+ }
+}
+
+wait_all_targets() {
+ local com=$1
+ local status=$2
+ local err=$3
+
+ wait_update_facet mds1 "$LCTL lfsck_query -t $com -M ${FSNAME}-MDT0000 |
+ awk '/^${com}_mdts_${status}/ { print \\\$2 }'" \
+ "$MDSCOUNT" $LTIME || {
+ do_facet mds1 "$LCTL lfsck_query -t $com -M ${FSNAME}-MDT0000"
+ error "($err) some MDTs are not in ${status}"
+ }
+}
+
test_0() {
lfsck_prep 3 3
[ $repaired -eq 1 ] ||
error "(5) Fail to repair crashed FID-in-dirent: $repaired"
+ run_e2fsck_on_mdt0
+
mount_client $MOUNT || error "(6) Fail to start client!"
#define OBD_FAIL_FID_LOOKUP 0x1505
awk '/^updated_phase1/ { print $2 }')
[ $repaired -eq 1 ] ||
- error "(5) Fail to repair missed FID-in-LMA: $repaired"
+ error "(5) Fail to repair the missing FID-in-LMA: $repaired"
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+ 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=0
}
-run_test 1b "LFSCK can find out and repair missed FID-in-LMA"
+run_test 1b "LFSCK can find out and repair the missing FID-in-LMA"
test_2a() {
lfsck_prep 1 1
[ $repaired -eq 1 ] ||
error "(5) Fail to repair crashed linkEA: $repaired"
+ run_e2fsck_on_mdt0
+
mount_client $MOUNT || error "(6) Fail to start client!"
stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
[ $repaired -eq 1 ] ||
error "(5) Fail to repair crashed linkEA: $repaired"
+ run_e2fsck_on_mdt0
+
mount_client $MOUNT || error "(6) Fail to start client!"
stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
[ $repaired -eq 1 ] ||
error "(5) Fail to repair crashed linkEA: $repaired"
+ run_e2fsck_on_mdt0
+
mount_client $MOUNT || error "(6) Fail to start client!"
stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
[ $repaired -eq 1 ] ||
error "(5) Fail to repair crashed linkEA: $repaired"
+ run_e2fsck_on_mdt0
+
mount_client $MOUNT || error "(6) Fail to start client!"
stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
[ "$dummyname" == "$DIR/$tdir/dummy" ] ||
error "(8) Fail to repair linkEA: $dummyfid $dummyname"
}
-run_test 2d "LFSCK can recover the missed linkEA entry"
+run_test 2d "LFSCK can recover the missing linkEA entry"
test_2e()
{
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
$START_NAMESPACE -r -A || 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 || {
- $SHOW_NAMESPACE
- error "(4) unexpected status"
- }
+
+ wait_all_targets_blocked namespace completed 4
local repaired=$($SHOW_NAMESPACE |
awk '/^linkea_repaired/ { print $2 }')
[ $repaired -ge 9 ] ||
error "(9) Fail to re-generate FID-in-dirent: $repaired"
+ run_e2fsck_on_mdt0
+
mount_client $MOUNT || error "(10) Fail to start client!"
#define OBD_FAIL_FID_LOOKUP 0x1505
[ $repaired -ge 2 ] ||
error "(9) Fail to generate FID-in-dirent for IGIF: $repaired"
+ run_e2fsck_on_mdt0
+
mount_client $MOUNT || error "(10) Fail to start client!"
#define OBD_FAIL_FID_LOOKUP 0x1505
echo "stop $SINGLEMDS"
stop $SINGLEMDS > /dev/null || error "(4) Fail to stop MDS!"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
echo "start $SINGLEMDS"
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
error "(5) Fail to start MDS!"
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_namespace |
awk '/^status/ { print \\\$2 }'" "completed" 30 || {
error "(4) unexpected status"
}
+ umount_client $MOUNT
echo "stop $SINGLEMDS"
stop $SINGLEMDS > /dev/null || error "(5) Fail to stop MDS!"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
echo "start $SINGLEMDS"
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
error "(6) Fail to start MDS!"
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_namespace |
awk '/^status/ { print \\\$2 }'" "completed" 30 || {
[ "$STATUS" == "paused" ] ||
error "(20) Expect 'paused', but got '$STATUS'"
+ echo "stop $SINGLEMDS"
+ stop $SINGLEMDS > /dev/null || error "(20.1) Fail to stop MDS!"
+
+ echo "start $SINGLEMDS without resume LFSCK"
+ start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SKIP_LFSCK > /dev/null ||
+ error "(20.2) Fail to start MDS!"
+
+ timer=0
+ while [ $timer -lt $timeout ]; do
+ STATUS=$(do_facet $SINGLEMDS "$LCTL get_param -n \
+ mdt.${MDT_DEV}.recovery_status |
+ awk '/^status/ { print \\\$2 }'")
+ [ "$STATUS" != "RECOVERING" ] && break;
+ sleep 1
+ timer=$((timer + 1))
+ done
+
+ [ $timer != $timeout ] ||
+ error "(20.3) recovery timeout"
+
+ STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
+ [ "$STATUS" == "paused" ] ||
+ error "(20.4) Expect 'paused', but got '$STATUS'"
+
#define OBD_FAIL_LFSCK_DELAY3 0x1602
do_facet $SINGLEMDS $LCTL set_param fail_val=2 fail_loc=0x1602
return 0
fi
- lfsck_prep 70 70
+ [[ $server_version -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
+ createmany -o $DIR/$tdir/lfsck/f 5000
local BASE_SPEED1=100
local RUN_TIME1=10
- $START_NAMESPACE -r -s $BASE_SPEED1 || error "(3) Fail to start LFSCK!"
+ $START_LAYOUT -r -s $BASE_SPEED1 || error "(2) Fail to start LFSCK!"
sleep $RUN_TIME1
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
+ STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
error "(3) Expect 'scanning-phase1', but got '$STATUS'"
- local SPEED=$($SHOW_NAMESPACE |
+ local SPEED=$($SHOW_LAYOUT |
awk '/^average_speed_phase1/ { print $2 }')
# There may be time error, normally it should be less than 2 seconds.
$LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit $BASE_SPEED2
sleep $RUN_TIME2
- SPEED=$($SHOW_NAMESPACE | awk '/^average_speed_phase1/ { print $2 }')
+ SPEED=$($SHOW_LAYOUT | awk '/^average_speed_phase1/ { print $2 }')
# MIN_MARGIN = 0.8 = 8 / 10
local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
(RUN_TIME1 + RUN_TIME2) * 8 / 10))
- # Account for slow ZFS performance - LU-4934
- [ $SPEED -gt $MIN_SPEED ] || [ $(facet_fstype $SINGLEMDS) -eq zfs ] ||
- error "(5) Got speed $SPEED, expected more than $MIN_SPEED"
+ [ $SPEED -gt $MIN_SPEED ] || {
+ if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+ error_ignore LU-5624 \
+ "(5.1) Got speed $SPEED, expected more than $MIN_SPEED"
+ else
+ error \
+ "(5.2) Got speed $SPEED, expected more than $MIN_SPEED"
+ fi
+ }
# MAX_MARGIN = 1.2 = 12 / 10
MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
$LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
wait_update_facet $SINGLEMDS \
- "$LCTL get_param -n mdd.${MDT_DEV}.lfsck_namespace|\
- awk '/^status/ { print \\\$2 }'" "completed" 30 ||
+ "$LCTL get_param -n mdd.${MDT_DEV}.lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 30 ||
error "(7) Failed to get expected 'completed'"
}
run_test 9a "LFSCK speed control (1)"
return 0
fi
+ [[ $server_version -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 $(date)."
local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
(RUN_TIME1 + RUN_TIME2) * 8 / 10))
- [ $SPEED -gt $MIN_SPEED ] ||[ $(facet_fstype $SINGLEMDS) -eq zfs ] ||
- error "(9) Got speed $SPEED, expected more than $MIN_SPEED"
+ [ $SPEED -gt $MIN_SPEED ] || {
+ if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+ error_ignore LU-5624 \
+ "(9.1) Got speed $SPEED, expected more than $MIN_SPEED"
+ else
+ error \
+ "(9.2) Got speed $SPEED, expected more than $MIN_SPEED"
+ fi
+ }
# MAX_MARGIN = 1.2 = 12 / 10
MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
echo "set fail_loc=0x160d to skip the updating LAST_ID on-disk"
#define OBD_FAIL_LFSCK_SKIP_LASTID 0x160d
do_facet ost1 $LCTL set_param fail_loc=0x160d
- createmany -o $DIR/$tdir/f 64
+
+ local count=$(precreated_ost_obj_count 0 0)
+
+ createmany -o $DIR/$tdir/f $((count + 32))
+
+ local proc_path="${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 0x100000000 |
+ obdfilter.${ost1_svc}.last_id" | grep $seq |
awk -F: '{ print $2 }')
umount_client $MOUNT
for ((i = 0; i < 60; i++)); do
lastid2=$(do_facet ost1 "lctl get_param -n \
- obdfilter.${ost1_svc}.last_id" | grep 0x100000000 |
+ obdfilter.${ost1_svc}.last_id" | grep $seq |
awk -F: '{ print $2 }')
[ ! -z $lastid2 ] && break;
sleep 1
echo "the on-disk LAST_ID should have been rebuilt"
wait_update_facet ost1 "$LCTL get_param -n \
- obdfilter.${ost1_svc}.last_id | grep 0x100000000 |
+ obdfilter.${ost1_svc}.last_id | grep $seq |
awk -F: '{ print \\\$2 }'" "$lastid1" 60 || {
- $LCTL get_param -n obdfilter.${ost1_svc}.last_id
- error "(9) expect lastid1 0x100000000:$lastid1"
+ do_facet ost1 $LCTL get_param -n \
+ obdfilter.${ost1_svc}.last_id
+ error "(9) expect lastid1 $seq:$lastid1"
}
do_facet ost1 $LCTL set_param fail_loc=0
-s 1 -r || error "(2) Fail to start LFSCK on all devices!"
echo "All the LFSCK targets should be in 'scanning-phase1' status."
- for k in $(seq $MDSCOUNT); do
- local STATUS=$(do_facet mds${k} $LCTL get_param -n \
- mdd.$(facet_svc mds${k}).lfsck_namespace |
- awk '/^status/ { print $2 }')
- [ "$STATUS" == "scanning-phase1" ] ||
- error "(3) MDS${k} Expect 'scanning-phase1', but got '$STATUS'"
- done
+ wait_all_targets namespace scanning-phase1 3
echo "Stop namespace LFSCK on all targets by single lctl command."
do_facet mds1 $LCTL lfsck_stop -M ${FSNAME}-MDT0000 -A ||
error "(4) Fail to stop LFSCK on all devices!"
echo "All the LFSCK targets should be in 'stopped' status."
- for k in $(seq $MDSCOUNT); do
- local STATUS=$(do_facet mds${k} $LCTL get_param -n \
- mdd.$(facet_svc mds${k}).lfsck_namespace |
- awk '/^status/ { print $2 }')
- [ "$STATUS" == "stopped" ] ||
- error "(5) MDS${k} Expect 'stopped', but got '$STATUS'"
- done
+ wait_all_targets_blocked namespace stopped 5
echo "Re-start namespace LFSCK on all targets by single command (-s 0)."
do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -t namespace -A \
-s 0 -r || error "(6) Fail to start LFSCK on all devices!"
echo "All the LFSCK targets should be in 'completed' status."
- for k in $(seq $MDSCOUNT); do
- wait_update_facet mds${k} "$LCTL get_param -n \
- mdd.$(facet_svc mds${k}).lfsck_namespace |
- awk '/^status/ { print \\\$2 }'" "completed" 8 ||
- error "(7) MDS${k} is not the expected 'completed'"
- done
+ wait_all_targets_blocked namespace completed 7
+
+ start_full_debug_logging
echo "Start layout LFSCK on all targets by single command (-s 1)."
do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -t layout -A \
-s 1 -r || error "(8) Fail to start LFSCK on all devices!"
echo "All the LFSCK targets should be in 'scanning-phase1' status."
- for k in $(seq $MDSCOUNT); do
- local STATUS=$(do_facet mds${k} $LCTL get_param -n \
- mdd.$(facet_svc mds${k}).lfsck_layout |
- awk '/^status/ { print $2 }')
- [ "$STATUS" == "scanning-phase1" ] ||
- error "(9) MDS${k} Expect 'scanning-phase1', but got '$STATUS'"
- done
+ wait_all_targets layout scanning-phase1 9
echo "Stop layout LFSCK on all targets by single lctl command."
do_facet mds1 $LCTL lfsck_stop -M ${FSNAME}-MDT0000 -A ||
error "(10) Fail to stop LFSCK on all devices!"
echo "All the LFSCK targets should be in 'stopped' status."
- for k in $(seq $MDSCOUNT); do
- local STATUS=$(do_facet mds${k} $LCTL get_param -n \
- mdd.$(facet_svc mds${k}).lfsck_layout |
- awk '/^status/ { print $2 }')
- [ "$STATUS" == "stopped" ] ||
- error "(11) MDS${k} Expect 'stopped', but got '$STATUS'"
- done
+ wait_all_targets_blocked layout stopped 11
for k in $(seq $OSTCOUNT); do
local STATUS=$(do_facet ost${k} $LCTL get_param -n \
-s 0 -r || error "(13) Fail to start LFSCK on all devices!"
echo "All the LFSCK targets should be in 'completed' status."
- 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${k}).lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 32 ||
- error "(14) MDS${k} is not the expected 'completed'"
- done
+ wait_all_targets_blocked layout completed 14
+
+ stop_full_debug_logging
}
run_test 12 "single command to trigger LFSCK on all devices"
test_14() {
echo "#####"
echo "The OST-object referenced by the MDT-object should be there;"
- echo "otherwise, the LFSCK should re-create the missed OST-object."
+ echo "otherwise, the LFSCK should re-create the missing OST-object."
echo "#####"
check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
- local count=$(precreated_ost_obj_count 0 0)
-
echo "Inject failure stub to simulate dangling referenced MDT-object"
#define OBD_FAIL_LFSCK_DANGLING 0x1610
do_facet ost1 $LCTL set_param fail_loc=0x1610
+ local count=$(precreated_ost_obj_count 0 0)
+
createmany -o $DIR/$tdir/f $((count + 31))
touch $DIR/$tdir/guard
do_facet ost1 $LCTL set_param fail_loc=0
}
run_test 15b "LFSCK can repair unmatched MDT-object/OST-object pairs (2)"
+test_15c() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "We need at least 2 MDSes for this test" && return
+
+ [ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.55) ] &&
+ skip "Skip the test after 2.7.55 see LU-6437" && return
+
+ echo "#####"
+ echo "According to current metadata migration implementation,"
+ echo "before the old MDT-object is removed, both the new MDT-object"
+ echo "and old MDT-object will reference the same LOV layout. Then if"
+ echo "the layout LFSCK finds the new MDT-object by race, it will"
+ echo "regard related OST-object(s) as multiple referenced case, and"
+ echo "will try to create new OST-object(s) for the new MDT-object."
+ echo "To avoid such trouble, the layout LFSCK needs to lock the old"
+ echo "MDT-object before confirm the multiple referenced case."
+ echo "#####"
+
+ check_mount_and_prep
+ $LFS mkdir -i 1 $DIR/$tdir/a1
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir/a1
+ dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=1
+ cancel_lru_locks osc
+
+ echo "Inject failure stub on MDT1 to delay the migration"
+
+ #define OBD_FAIL_MIGRATE_DELAY 0x1803
+ do_facet mds2 $LCTL set_param fail_val=5 fail_loc=0x1803
+ echo "Migrate $DIR/$tdir/a1 from MDT1 to MDT0 with delay"
+ $LFS migrate -m 0 $DIR/$tdir/a1 &
+
+ sleep 1
+ echo "Trigger layout LFSCK to race with the migration"
+ $START_LAYOUT -A -r || error "(1) Fail to start layout LFSCK!"
+
+ wait_all_targets_blocked layout completed 2
+
+ do_facet mds2 $LCTL set_param fail_loc=0 fail_val=0
+ local repaired=$($SHOW_LAYOUT |
+ awk '/^repaired_unmatched_pair/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(3) Fail to repair unmatched pair: $repaired"
+
+ repaired=$($SHOW_LAYOUT |
+ awk '/^repaired_multiple_referenced/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(4) Unexpectedly repaird multiple references: $repaired"
+}
+run_test 15c "LFSCK can repair unmatched MDT-object/OST-object pairs (3)"
+
test_16() {
echo "#####"
echo "If the OST-object's owner information does not match the owner"
}
run_test 17 "LFSCK can repair multiple references"
+$LCTL set_param debug=+cache > /dev/null
+
test_18a() {
echo "#####"
echo "The target MDT-object is there, but related stripe information"
- echo "is lost or partly lost. The LFSCK should regenerate the missed"
+ echo "is lost or partly lost. The LFSCK should regenerate the missing"
echo "layout EA entries."
echo "#####"
check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/a1
- $LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
+ $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
local saved_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
if [ $MDSCOUNT -ge 2 ]; then
$LFS mkdir -i 1 $DIR/$tdir/a2
- $LFS setstripe -c 2 -i 1 -s 1M $DIR/$tdir/a2
+ $LFS setstripe -c 2 -i 1 -S 1M $DIR/$tdir/a2
dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
$LFS path2fid $DIR/$tdir/a2/f2
$LFS getstripe $DIR/$tdir/a2/f2
# 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 ||
+ awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
error "(4) MDS${k} is not the expected 'completed'"
done
check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/a1
- $LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
+ $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
local saved_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
local fid1=$($LFS path2fid $DIR/$tdir/a1/f1)
if [ $MDSCOUNT -ge 2 ]; then
$LFS mkdir -i 1 $DIR/$tdir/a2
- $LFS setstripe -c 2 -i 1 -s 1M $DIR/$tdir/a2
+ $LFS setstripe -c 2 -i 1 -S 1M $DIR/$tdir/a2
dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
fid2=$($LFS path2fid $DIR/$tdir/a2/f2)
echo ${fid2}
# time to guarantee the status sync up.
wait_update_facet mds${k} "$LCTL get_param -n \
mdd.$(facet_svc mds${k}).lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
error "(2) MDS${k} is not the expected 'completed'"
done
check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/a1
- $LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
+ $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
echo "Inject failure, to simulate the case of missing parent FID"
#define OBD_FAIL_LFSCK_NOPFID 0x1617
if [ $MDSCOUNT -ge 2 ]; then
$LFS mkdir -i 1 $DIR/$tdir/a2
- $LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a2
+ $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a2
dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
$LFS getstripe $DIR/$tdir/a2/f2
fi
# 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 ||
+ awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
error "(2) MDS${k} is not the expected 'completed'"
done
check_mount_and_prep
mkdir $DIR/$tdir/a1
- $LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
+ $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
echo "guard" > $DIR/$tdir/a1/f1
echo "foo" > $DIR/$tdir/a1/f2
local saved_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
wait_update_facet mds1 "$LCTL get_param -n \
mdd.$(facet_svc mds1).lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "scanning-phase2" 32 ||
+ awk '/^status/ { print \\\$2 }'" "scanning-phase2" $LTIME ||
error "(3.0) MDS1 is not the expected 'scanning-phase2'"
do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
# time to guarantee the status sync up.
wait_update_facet mds${k} "$LCTL get_param -n \
mdd.$(facet_svc mds${k}).lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
error "(3) MDS${k} is not the expected 'completed'"
done
check_mount_and_prep
mkdir $DIR/$tdir/a1
- $LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
+ $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
echo "guard" > $DIR/$tdir/a1/f1
echo "foo" > $DIR/$tdir/a1/f2
local saved_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
#define OBD_FAIL_LFSCK_DELAY3 0x1602
do_facet $SINGLEMDS $LCTL set_param fail_val=10 fail_loc=0x1602
+ start_full_debug_logging
+
echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
$START_LAYOUT -r -o -c || error "(2) Fail to start LFSCK for layout!"
wait_update_facet mds1 "$LCTL get_param -n \
mdd.$(facet_svc mds1).lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "scanning-phase2" 32 ||
+ awk '/^status/ { print \\\$2 }'" "scanning-phase2" $LTIME ||
error "(3) MDS1 is not the expected 'scanning-phase2'"
# to guarantee all updates are synced.
# time to guarantee the status sync up.
wait_update_facet mds${k} "$LCTL get_param -n \
mdd.$(facet_svc mds${k}).lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
error "(4) MDS${k} is not the expected 'completed'"
done
error "(5) OST${k} Expect 'completed', but got '$cur_status'"
done
+ stop_full_debug_logging
+
local repaired=$(do_facet $SINGLEMDS $LCTL get_param -n \
mdd.$(facet_svc $SINGLEMDS).lfsck_layout |
awk '/^repaired_orphan/ { print $2 }')
check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/a1
- $LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
+ $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
dd if=/dev/zero of=$DIR/$tdir/a1/guard bs=1M count=2
dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
$LFS mkdir -i 0 $DIR/$tdir/a2
- $LFS setstripe -c 2 -i 0 -s 1M $DIR/$tdir/a2
+ $LFS setstripe -c 2 -i 0 -S 1M $DIR/$tdir/a2
dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
$LFS getstripe $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a2/f2
if [ $MDSCOUNT -ge 2 ]; then
$LFS mkdir -i 1 $DIR/$tdir/a3
- $LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a3
+ $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a3
dd if=/dev/zero of=$DIR/$tdir/a3/guard bs=1M count=2
dd if=/dev/zero of=$DIR/$tdir/a3/f3 bs=1M count=2
$LFS mkdir -i 1 $DIR/$tdir/a4
- $LFS setstripe -c 2 -i 0 -s 1M $DIR/$tdir/a4
+ $LFS setstripe -c 2 -i 0 -S 1M $DIR/$tdir/a4
dd if=/dev/zero of=$DIR/$tdir/a4/f4 bs=1M count=2
$LFS getstripe $DIR/$tdir/a3/f3
$LFS getstripe $DIR/$tdir/a4/f4
# 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 }'" "partial" 32 ||
+ awk '/^status/ { print \\\$2 }'" "partial" $LTIME ||
error "(2) MDS${k} is not the expected 'partial'"
done
wait_update_facet ost1 "$LCTL get_param -n \
obdfilter.$(facet_svc ost1).lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "partial" 32 || {
+ awk '/^status/ { print \\\$2 }'" "partial" $LTIME || {
error "(3) OST1 is not the expected 'partial'"
}
wait_update_facet ost2 "$LCTL get_param -n \
obdfilter.$(facet_svc ost2).lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ awk '/^status/ { print \\\$2 }'" "completed" $LTIME || {
error "(4) OST2 is not the expected 'completed'"
}
# time to guarantee the status sync up.
wait_update_facet mds${k} "$LCTL get_param -n \
mdd.$(facet_svc mds${k}).lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
error "(8) MDS${k} is not the expected 'completed'"
done
}
run_test 18f "Skip the failed OST(s) when handle orphan OST-objects"
+$LCTL set_param debug=-cache > /dev/null
+
test_19a() {
check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/a1
if [ $OSTCOUNT -gt 2 ]; then
- $LFS setstripe -c 3 -i 0 -s 1M $DIR/$tdir/a1
+ $LFS setstripe -c 3 -i 0 -S 1M $DIR/$tdir/a1
bcount=513
else
- $LFS setstripe -c 2 -i 0 -s 1M $DIR/$tdir/a1
+ $LFS setstripe -c 2 -i 0 -S 1M $DIR/$tdir/a1
bcount=257
fi
echo "#####"
echo "The parent_A references the child directory via some name entry,"
echo "but the child directory back references another parent_B via its"
- echo "".." name entry. The parent_A does not exist. Then the namesapce"
+ echo "".." name entry. The parent_B does not exist. Then the namespace"
echo "LFSCK will repair the child directory's ".." name entry."
echo "#####"
$START_NAMESPACE -A -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 || {
- $SHOW_NAMESPACE
- error "(6) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 6
local repaired=$($SHOW_NAMESPACE |
awk '/^unmatched_pairs_repaired/ { print $2 }')
echo "The parent_A references the child directory via the name entry_B,"
echo "but the child directory back references another parent_C via its"
echo "".." name entry. The parent_C exists, but there is no the name"
- echo "entry_B under the parent_B. Then the namesapce LFSCK will repair"
+ echo "entry_B under the parent_C. Then the namespace LFSCK will repair"
echo "the child directory's ".." name entry and its linkEA."
echo "#####"
echo "Inject failure stub on MDT0 to simulate bad dotdot name entry"
echo "and bad linkEA. The dummy's dotdot name entry references the"
echo "guard. The dummy's linkEA references n non-exist name entry."
- #define OBD_FAIL_LFSCK_BAD_PARENT2 0x161f
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x161f
+ #define OBD_FAIL_LFSCK_BAD_PARENT 0x161e
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x161e
$LFS mkdir -i 0 $DIR/$tdir/foo/dummy ||
error "(3) Fail to mkdir on MDT0"
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
$START_NAMESPACE -A -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 || {
- $SHOW_NAMESPACE
- error "(6) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 6
local repaired=$($SHOW_NAMESPACE |
awk '/^unmatched_pairs_repaired/ { print $2 }')
$START_NAMESPACE -A -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 || {
- $SHOW_NAMESPACE
- error "(6) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 6
local repaired=$($SHOW_NAMESPACE |
awk '/^dangling_repaired/ { print $2 }')
$START_NAMESPACE -A -r -C ||
error "(9) 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 || {
- $SHOW_NAMESPACE
- error "(10) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 10
repaired=$($SHOW_NAMESPACE |
awk '/^dangling_repaired/ { print $2 }')
check_mount_and_prep
+ [[ -d $MOUNT/.lustre/lost+found/MDT0000 ]] || {
+ # Trigger LFSCK firstly, that will generate the
+ # .lustre/lost+found/MDTxxxx in advance to avoid
+ # reusing the local object for the dangling name
+ # entry. LU-7429
+ $START_NAMESPACE -r ||
+ error "(0) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 0.1
+ }
+
$LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0 on MDT0"
+ $LFS path2fid $DIR/$tdir/d0
+
echo "dummy" > $DIR/$tdir/d0/f0 || error "(2) Fail to touch on MDT0"
+ $LFS path2fid $DIR/$tdir/d0/f0
+
echo "dead" > $DIR/$tdir/d0/f1 || error "(3) Fail to touch on MDT0"
+ $LFS path2fid $DIR/$tdir/d0/f1
+
+ local OID=$($LFS path2fid $DIR/$tdir/d0/f1 | awk -F':' '{print $2}')
+ OID=$(printf %d $OID)
+
+ if [ $OID -eq 1 ]; 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
+ fi
echo "Inject failure stub on MDT0 to simulate dangling name entry"
#define OBD_FAIL_LFSCK_DANGLING3 0x1621
- do_facet $SINGLEMDS $LCTL set_param 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_loc=0
+ do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
rm -f $DIR/$tdir/d0/f1 || error "(5) Fail to unlink $DIR/$tdir/d0/f1"
echo "LFSCK cannot replace it."
echo "#####"
+ start_full_debug_logging
+
check_mount_and_prep
+ [[ -d $MOUNT/.lustre/lost+found/MDT0000 ]] || {
+ # Trigger LFSCK firstly, that will generate the
+ # .lustre/lost+found/MDTxxxx in advance to avoid
+ # reusing the local object for the dangling name
+ # entry. LU-7429
+ $START_NAMESPACE -r ||
+ error "(0) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 0.1
+ }
+
$LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0 on MDT0"
+ $LFS path2fid $DIR/$tdir/d0
+
echo "dummy" > $DIR/$tdir/d0/f0 || error "(2) Fail to touch on MDT0"
+ $LFS path2fid $DIR/$tdir/d0/f0
+
echo "dead" > $DIR/$tdir/d0/f1 || error "(3) Fail to touch on MDT0"
+ $LFS path2fid $DIR/$tdir/d0/f1
+
+ local OID=$($LFS path2fid $DIR/$tdir/d0/f1 | awk -F':' '{print $2}')
+ OID=$(printf %d $OID)
+
+ if [ $OID -eq 1 ]; 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
+ fi
echo "Inject failure stub on MDT0 to simulate dangling name entry"
#define OBD_FAIL_LFSCK_DANGLING3 0x1621
- do_facet $SINGLEMDS $LCTL set_param 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_loc=0
+ do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
rm -f $DIR/$tdir/d0/f1 || error "(5) Fail to unlink $DIR/$tdir/d0/f1"
error "(7) Fail to start LFSCK for namespace"
wait_update_facet client "stat $DIR/$tdir/d0/foo |
- awk '/Size/ { print \\\$2 }'" "0" 32 || {
+ awk '/Size/ { print \\\$2 }'" "0" $LTIME || {
stat $DIR/$tdir/guard
$SHOW_NAMESPACE
error "(8) unexpected size"
error "(10) unexpected status"
}
+ stop_full_debug_logging
+
local repaired=$($SHOW_NAMESPACE |
awk '/^dangling_repaired/ { print $2 }')
[ $repaired -eq 1 ] ||
}
run_test 23c "LFSCK can repair dangling name entry (3)"
-$LCTL set_param debug=-lfsck > /dev/null || true
+test_24() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "We need at least 2 MDSes for this test" && return
+
+ echo "#####"
+ echo "Two MDT-objects back reference the same name entry via their"
+ echo "each own linkEA entry, but the name entry only references one"
+ echo "MDT-object. The namespace LFSCK will remove the linkEA entry"
+ echo "for the MDT-object that is not recognized. If such MDT-object"
+ echo "has no other linkEA entry after the removing, then the LFSCK"
+ echo "will add it as orphan under the .lustre/lost+found/MDTxxxx/."
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 1 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
+
+ mkdir $DIR/$tdir/d0/guard || error "(1) Fail to mkdir guard"
+ $LFS path2fid $DIR/$tdir/d0/guard
+
+ mkdir $DIR/$tdir/d0/dummy || error "(2) Fail to mkdir dummy"
+ $LFS path2fid $DIR/$tdir/d0/dummy
+
+ local pfid
+ if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+ pfid=$($LFS path2fid $DIR/$tdir/d0/guard)
+ else
+ pfid=$($LFS path2fid $DIR/$tdir/d0/dummy)
+ fi
+
+ touch $DIR/$tdir/d0/guard/foo ||
+ error "(3) Fail to touch $DIR/$tdir/d0/guard/foo"
+
+ echo "Inject failure stub on MDT0 to simulate the case that"
+ echo "the $DIR/$tdir/d0/dummy/foo has the 'bad' linkEA entry"
+ echo "that references $DIR/$tdir/d0/guard/foo."
+ echo "Then remove the name entry $DIR/$tdir/d0/dummy/foo."
+ echo "So the MDT-object $DIR/$tdir/d0/dummy/foo will be left"
+ echo "there with the same linkEA entry as another MDT-object"
+ echo "$DIR/$tdir/d0/guard/foo has"
+
+ #define OBD_FAIL_LFSCK_MUL_REF 0x1622
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1622
+ $LFS mkdir -i 0 $DIR/$tdir/d0/dummy/foo ||
+ error "(4) Fail to mkdir $DIR/$tdir/d0/dummy/foo"
+ $LFS path2fid $DIR/$tdir/d0/dummy/foo
+ local cfid=$($LFS path2fid $DIR/$tdir/d0/dummy/foo)
+ rmdir $DIR/$tdir/d0/dummy/foo ||
+ error "(5) Fail to remove $DIR/$tdir/d0/dummy/foo name entry"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ echo "stat $DIR/$tdir/d0/dummy/foo should fail"
+ stat $DIR/$tdir/d0/dummy/foo > /dev/null 2>&1 &&
+ error "(6) stat successfully unexpectedly"
+
+ echo "Trigger namespace LFSCK to repair multiple-referenced name entry"
+ $START_NAMESPACE -A -r ||
+ error "(7) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 8
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^multiple_referenced_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(9) Fail to repair multiple referenced name entry: $repaired"
+
+ echo "There should be an orphan under .lustre/lost+found/MDT0000/"
+ [ -d $MOUNT/.lustre/lost+found/MDT0000 ] ||
+ error "(10) $MOUNT/.lustre/lost+found/MDT0000/ should be there"
+
+ local cname="$cfid-$pfid-D-0"
+ ls -ail $MOUNT/.lustre/lost+found/MDT0000/$cname ||
+ error "(11) .lustre/lost+found/MDT0000/ should not be empty"
+}
+run_test 24 "LFSCK can repair multiple-referenced name entry"
+
+test_25() {
+ [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
+ skip "Only support to inject failure on ldiskfs" && return
+
+ echo "#####"
+ echo "The file type in the name entry does not match the file type"
+ echo "claimed by the referenced object. Then the LFSCK will update"
+ echo "the file type in the name entry."
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
+
+ echo "Inject failure stub on MDT0 to simulate the case that"
+ echo "the file type stored in the name entry is wrong."
+
+ #define OBD_FAIL_LFSCK_BAD_TYPE 0x1623
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1623
+ touch $DIR/$tdir/d0/foo || error "(2) Fail to touch $DIR/$tdir/d0/foo"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ echo "Trigger namespace LFSCK to repair bad file type in the name entry"
+ $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 '/^bad_file_type_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(5) Fail to repair bad file type in name entry: $repaired"
+
+ ls -ail $DIR/$tdir/d0 || error "(6) Fail to 'ls' the $DIR/$tdir/d0"
+}
+run_test 25 "LFSCK can repair bad file type in the name entry"
+
+test_26a() {
+ 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"
+ echo "to the normal namespace."
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
+ touch $DIR/$tdir/d0/foo || error "(2) Fail to create foo"
+ local foofid=$($LFS path2fid $DIR/$tdir/d0/foo)
+
+ ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/dummy ||
+ error "(3) Fail to hard link to $DIR/$tdir/d0/foo"
+
+ echo "Inject failure stub on MDT0 to simulate the case that"
+ echo "foo's name entry will be removed, but the foo's object"
+ echo "and its linkEA are kept in the system."
+
+ #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1624
+ rm -f $DIR/$tdir/d0/foo || error "(4) Fail to unlink $DIR/$tdir/d0/foo"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ ls -ail $DIR/$tdir/d0/foo > /dev/null 2>&1 &&
+ error "(5) 'ls' should fail"
+
+ echo "Trigger namespace LFSCK to repair the missing remote name entry"
+ $START_NAMESPACE -r -A ||
+ error "(6) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 7
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(8) Fail to repair lost dirent: $repaired"
+
+ ls -ail $DIR/$tdir/d0/foo ||
+ error "(9) Fail to 'ls' $DIR/$tdir/d0/foo"
+
+ local foofid2=$($LFS path2fid $DIR/$tdir/d0/foo)
+ [ "$foofid" == "$foofid2" ] ||
+ error "(10) foo's FID changed: $foofid, $foofid2"
+}
+run_test 26a "LFSCK can add the missing local name entry back to the namespace"
+
+test_26b() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "We need at least 2 MDSes for this test" && return
+
+ echo "#####"
+ echo "The remote name entry back referenced by the MDT-object is lost."
+ echo "The namespace LFSCK will add the missing remote name entry back"
+ echo "to the normal namespace."
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 1 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
+ $LFS mkdir -i 0 $DIR/$tdir/d0/foo || error "(2) Fail to mkdir foo"
+ local foofid=$($LFS path2fid $DIR/$tdir/d0/foo)
+
+ echo "Inject failure stub on MDT0 to simulate the case that"
+ echo "foo's name entry will be removed, but the foo's object"
+ echo "and its linkEA are kept in the system."
+
+ #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1624
+ rmdir $DIR/$tdir/d0/foo || error "(3) Fail to rmdir $DIR/$tdir/d0/foo"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ ls -ail $DIR/$tdir/d0/foo > /dev/null 2>&1 &&
+ error "(4) 'ls' should fail"
+
+ echo "Trigger namespace LFSCK to repair the missing remote name entry"
+ $START_NAMESPACE -r -A ||
+ error "(5) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 6
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(7) Fail to repair lost dirent: $repaired"
+
+ ls -ail $DIR/$tdir/d0/foo ||
+ error "(8) Fail to 'ls' $DIR/$tdir/d0/foo"
+
+ local foofid2=$($LFS path2fid $DIR/$tdir/d0/foo)
+ [ "$foofid" == "$foofid2" ] ||
+ error "(9) foo's FID changed: $foofid, $foofid2"
+}
+run_test 26b "LFSCK can add the missing remote name entry back to the namespace"
+
+test_27a() {
+ 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."
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
+ touch $DIR/$tdir/d0/foo || error "(2) Fail to create foo"
+ ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/dummy ||
+ error "(3) Fail to hard link to $DIR/$tdir/d0/foo"
+
+ echo "Inject failure stub on MDT0 to simulate the case that"
+ echo "foo's name entry will be removed, but the foo's object"
+ echo "and its linkEA are kept in the system. And then remove"
+ echo "another hard link and the parent directory."
+
+ #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1624
+ rm -f $DIR/$tdir/d0/foo ||
+ error "(4) Fail to unlink $DIR/$tdir/d0/foo"
+ rm -f $DIR/$tdir/d0/dummy ||
+ error "(5) Fail to unlink $DIR/$tdir/d0/dummy"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ rm -rf $DIR/$tdir/d0 || error "(5) Fail to unlink the dir d0"
+ ls -ail $DIR/$tdir/d0 > /dev/null 2>&1 && error "(6) 'ls' should fail"
+
+ echo "Trigger namespace LFSCK to repair the lost parent"
+ $START_NAMESPACE -r -A ||
+ error "(6) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 7
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(8) Fail to repair lost dirent: $repaired"
+
+ echo "There should be an orphan under .lustre/lost+found/MDT0000/"
+ [ -d $MOUNT/.lustre/lost+found/MDT0000 ] ||
+ error "(9) $MOUNT/.lustre/lost+found/MDT0000/ should be there"
+
+ ls -ail $MOUNT/.lustre/lost+found/MDT0000/
+
+ cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-P-*)
+ [ ! -z "$cname" ] ||
+ error "(10) .lustre/lost+found/MDT0000/ should not be empty"
+}
+run_test 27a "LFSCK can recreate the lost local parent directory as orphan"
+
+test_27b() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "We need at least 2 MDSes for this test" && return
+
+ echo "#####"
+ echo "The remote parent referenced by the MDT-object linkEA is lost."
+ echo "The namespace LFSCK will re-create the lost parent as orphan."
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 1 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
+ $LFS mkdir -i 0 $DIR/$tdir/d0/foo || error "(2) Fail to mkdir foo"
+
+ $LFS path2fid $DIR/$tdir/d0
+
+ echo "Inject failure stub on MDT0 to simulate the case that"
+ echo "foo's name entry will be removed, but the foo's object"
+ echo "and its linkEA are kept in the system. And then remove"
+ echo "the parent directory."
+
+ #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1624
+ rmdir $DIR/$tdir/d0/foo || error "(3) Fail to rmdir $DIR/$tdir/d0/foo"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ rmdir $DIR/$tdir/d0 || error "(4) Fail to unlink the dir d0"
+ ls -ail $DIR/$tdir/d0 > /dev/null 2>&1 && error "(5) 'ls' should fail"
+
+ echo "Trigger namespace LFSCK to repair the missing remote name entry"
+ $START_NAMESPACE -r -A ||
+ error "(6) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 7
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(8) Fail to repair lost dirent: $repaired"
+
+ ls -ail $MOUNT/.lustre/lost+found/
+
+ echo "There should be an orphan under .lustre/lost+found/MDT0001/"
+ [ -d $MOUNT/.lustre/lost+found/MDT0001 ] ||
+ error "(9) $MOUNT/.lustre/lost+found/MDT0001/ should be there"
+
+ ls -ail $MOUNT/.lustre/lost+found/MDT0001/
+
+ cname=$(find $MOUNT/.lustre/lost+found/MDT0001/ -name *-P-*)
+ [ ! -z "$cname" ] ||
+ error "(10) .lustre/lost+found/MDT0001/ should not be empty"
+}
+run_test 27b "LFSCK can recreate the lost remote parent directory as orphan"
+
+test_28() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "The test needs at least 2 MDTs" && return
+
+ echo "#####"
+ echo "The target name entry is lost. The LFSCK should insert the"
+ echo "orphan MDT-object under .lustre/lost+found/MDTxxxx. But if"
+ echo "the MDT (on which the orphan MDT-object resides) has ever"
+ echo "failed to respond some name entry verification during the"
+ echo "first stage-scanning, then the LFSCK should skip to handle"
+ echo "orphan MDT-object on this MDT. But other MDTs should not"
+ echo "be affected."
+ echo "#####"
+
+ check_mount_and_prep
+ $LFS mkdir -i 0 $DIR/$tdir/d1
+ $LFS mkdir -i 1 $DIR/$tdir/d1/a1
+ $LFS mkdir -i 1 $DIR/$tdir/d1/a2
+
+ $LFS mkdir -i 1 $DIR/$tdir/d2
+ $LFS mkdir -i 0 $DIR/$tdir/d2/a1
+ $LFS mkdir -i 0 $DIR/$tdir/d2/a2
+
+ echo "Inject failure stub on MDT0 to simulate the case that"
+ echo "d1/a1's name entry will be removed, but the d1/a1's object"
+ echo "and its linkEA are kept in the system. And the case that"
+ echo "d2/a2's name entry will be removed, but the d2/a2's object"
+ echo "and its linkEA are kept in the system."
+
+ #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
+ do_facet mds1 $LCTL set_param fail_loc=0x1624
+ do_facet mds2 $LCTL set_param fail_loc=0x1624
+ rmdir $DIR/$tdir/d1/a1 || error "(1) Fail to rmdir $DIR/$tdir/d1/a1"
+ rmdir $DIR/$tdir/d2/a2 || error "(2) Fail to rmdir $DIR/$tdir/d2/a2"
+ do_facet mds1 $LCTL set_param fail_loc=0
+ do_facet mds2 $LCTL set_param fail_loc=0
+
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
+
+ echo "Inject failure, to simulate the MDT0 fail to handle"
+ echo "MDT1 LFSCK request during the first-stage scanning."
+ #define OBD_FAIL_LFSCK_BAD_NETWORK 0x161c
+ do_facet mds2 $LCTL set_param fail_loc=0x161c fail_val=0
+
+ echo "Trigger namespace LFSCK on all devices to find out orphan object"
+ $START_NAMESPACE -r -A ||
+ error "(3) Fail to start LFSCK for namespace"
+
+ wait_update_facet mds1 "$LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "partial" 32 || {
+ error "(4) mds1 is not the expected 'partial'"
+ }
+
+ wait_update_facet mds2 "$LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ error "(5) mds2 is not the expected 'completed'"
+ }
+
+ do_facet mds2 $LCTL set_param fail_loc=0 fail_val=0
+
+ local repaired=$(do_facet mds1 $LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_namespace |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(6) Expect 0 fixed on mds1, but got: $repaired"
+
+ repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(7) Expect 1 fixed on mds2, but got: $repaired"
+
+ echo "Trigger namespace LFSCK on all devices again to cleanup"
+ $START_NAMESPACE -r -A ||
+ error "(8) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 9
+
+ local repaired=$(do_facet mds1 $LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_namespace |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(10) Expect 1 fixed on mds1, but got: $repaired"
+
+ repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(11) Expect 0 fixed on mds2, but got: $repaired"
+}
+run_test 28 "Skip the failed MDT(s) when handle orphan MDT-objects"
+
+test_29a() {
+ 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"
+ echo "attribute to match the known name entries count"
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
+ touch $DIR/$tdir/d0/foo || error "(2) Fail to create foo"
+
+ echo "Inject failure stub on MDT0 to simulate the case that foo's"
+ echo "nlink attribute is larger than its name entries count."
+
+ #define OBD_FAIL_LFSCK_MORE_NLINK 0x1625
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1625
+ ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/h1 ||
+ error "(3) Fail to hard link to $DIR/$tdir/d0/foo"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ cancel_lru_locks mdc
+ local count=$(stat --format=%h $DIR/$tdir/d0/foo)
+ [ $count -eq 3 ] || error "(4) Cannot inject error: $count"
+
+ echo "Trigger namespace LFSCK to repair the nlink count"
+ $START_NAMESPACE -r -A ||
+ error "(5) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 6
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^nlinks_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(7) Fail to repair nlink count: $repaired"
+
+ cancel_lru_locks mdc
+ count=$(stat --format=%h $DIR/$tdir/d0/foo)
+ [ $count -eq 2 ] || error "(8) Fail to repair nlink count: $count"
+}
+run_test 29a "LFSCK can repair bad nlink count (1)"
+
+test_29b() {
+ 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"
+ echo "attribute to match the known name entries count"
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
+ touch $DIR/$tdir/d0/foo || error "(2) Fail to create foo"
+
+ echo "Inject failure stub on MDT0 to simulate the case that foo's"
+ echo "nlink attribute is smaller than its name entries count."
+
+ #define OBD_FAIL_LFSCK_LESS_NLINK 0x1626
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1626
+ ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/h1 ||
+ error "(3) Fail to hard link to $DIR/$tdir/d0/foo"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ cancel_lru_locks mdc
+ local count=$(stat --format=%h $DIR/$tdir/d0/foo)
+ [ $count -eq 1 ] || error "(4) Cannot inject error: $count"
+
+ echo "Trigger namespace LFSCK to repair the nlink count"
+ $START_NAMESPACE -r -A ||
+ error "(5) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 6
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^nlinks_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(7) Fail to repair nlink count: $repaired"
+
+ cancel_lru_locks mdc
+ count=$(stat --format=%h $DIR/$tdir/d0/foo)
+ [ $count -eq 2 ] || error "(8) Fail to repair nlink count: $count"
+}
+run_test 29b "LFSCK can repair bad nlink count (2)"
+
+test_29c() {
+ echo "#####"
+ echo "There are too many hard links to the object, and exceeds the"
+ echo "object's linkEA limitation, as to NOT all the known name entries"
+ echo "will be recorded in the linkEA. Under such case, LFSCK should"
+ echo "skip the nlink verification for this object."
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
+ touch $DIR/$tdir/d0/foo || error "(2) Fail to create foo"
+ ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/h1 ||
+ error "(3) Fail to hard link to $DIR/$tdir/d0/foo"
+
+ echo "Inject failure stub on MDT0 to simulate the case that"
+ echo "foo's hard links exceed the object's linkEA limitation."
+
+ #define OBD_FAIL_LFSCK_LINKEA_OVERFLOW 0x1627
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1627
+ ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/h2 ||
+ error "(4) Fail to hard link to $DIR/$tdir/d0/foo"
+
+ cancel_lru_locks mdc
+
+ local count1=$(stat --format=%h $DIR/$tdir/d0/foo)
+ [ $count1 -eq 3 ] || error "(5) Stat failure: $count1"
+
+ local foofid=$($LFS path2fid $DIR/$tdir/d0/foo)
+ $LFS fid2path $DIR $foofid
+ local count2=$($LFS fid2path $DIR $foofid | wc -l)
+ [ $count2 -eq 2 ] || error "(6) Fail to inject error: $count2"
+
+ echo "Trigger namespace LFSCK to repair the nlink count"
+ $START_NAMESPACE -r -A ||
+ error "(7) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 8
+
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^nlinks_repaired/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(9) Repair nlink count unexpcetedly: $repaired"
+
+ cancel_lru_locks mdc
+
+ count1=$(stat --format=%h $DIR/$tdir/d0/foo)
+ [ $count1 -eq 3 ] || error "(10) Stat failure: $count1"
+
+ count2=$($LFS fid2path $DIR $foofid | wc -l)
+ [ $count2 -eq 2 ] ||
+ error "(11) Repaired something unexpectedly: $count2"
+}
+run_test 29c "Not verify nlink attr if hark links exceed linkEA limitation"
+
+test_30() {
+ [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
+ skip "Only support backend /lost+found for ldiskfs" && return
+
+ echo "#####"
+ echo "The namespace LFSCK will move the orphans from backend"
+ echo "/lost+found directory to normal client visible namespace"
+ echo "or to global visible ./lustre/lost+found/MDTxxxx/ directory"
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 0 $DIR/$tdir/foo || error "(1) Fail to mkdir foo"
+ touch $DIR/$tdir/foo/f0 || error "(2) Fail to touch f1"
+
+ echo "Inject failure stub on MDT0 to simulate the case that"
+ echo "directory d0 has no linkEA entry, then the LFSCK will"
+ echo "move it into .lustre/lost+found/MDTxxxx/ later."
+
+ #define OBD_FAIL_LFSCK_NO_LINKEA 0x161d
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x161d
+ mkdir $DIR/$tdir/foo/d0 || error "(3) Fail to mkdir d0"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ touch $DIR/$tdir/foo/d0/f1 || error "(4) Fail to touch f1"
+ mkdir $DIR/$tdir/foo/d0/d1 || error "(5) Fail to mkdir d1"
+
+ echo "Inject failure stub on MDT0 to simulate the case that the"
+ echo "object's name entry will be removed, but not destroy the"
+ echo "object. Then backend e2fsck will handle it as orphan and"
+ echo "add them into the backend /lost+found directory."
+
+ #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1624
+ rmdir $DIR/$tdir/foo/d0/d1 || error "(6) Fail to rmdir d1"
+ rm -f $DIR/$tdir/foo/d0/f1 || error "(7) Fail to unlink f1"
+ rmdir $DIR/$tdir/foo/d0 || error "(8) Fail to rmdir d0"
+ rm -f $DIR/$tdir/foo/f0 || error "(9) Fail to unlink f0"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ umount_client $MOUNT || error "(10) Fail to stop client!"
+
+ stop $SINGLEMDS || error "(11) Fail to stop MDT0"
+
+ echo "run e2fsck"
+ run_e2fsck $(facet_host $SINGLEMDS) $MDT_DEVNAME "-y" ||
+ error "(12) Fail to run e2fsck"
+
+ start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
+ error "(13) Fail to start MDT0"
+
+ echo "Trigger namespace LFSCK to recover backend orphans"
+ $START_NAMESPACE -r -A ||
+ error "(14) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 15
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^local_lost_found_moved/ { print $2 }')
+ [ $repaired -ge 4 ] ||
+ error "(16) Fail to recover backend orphans: $repaired"
+
+ mount_client $MOUNT || error "(17) Fail to start client!"
+
+ stat $DIR/$tdir/foo/f0 || error "(18) f0 is not recovered"
+
+ ls -ail $MOUNT/.lustre/lost+found/
+
+ echo "d0 should become orphan under .lustre/lost+found/MDT0000/"
+ [ -d $MOUNT/.lustre/lost+found/MDT0000 ] ||
+ error "(19) $MOUNT/.lustre/lost+found/MDT0000/ should be there"
+
+ ls -ail $MOUNT/.lustre/lost+found/MDT0000/
+
+ cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-*-D-*)
+ [ ! -z "$cname" ] || error "(20) d0 is not recovered"
+
+ stat ${cname}/d1 || error "(21) d0 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 "The test needs at least 2 MDTs" && return
+
+ echo "#####"
+ echo "For the name entry under a striped directory, if the name"
+ echo "hash does not match the shard, then the LFSCK will repair"
+ echo "the bad name entry"
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
+ error "(1) Fail to create striped directory"
+
+ echo "Inject failure stub on client to simulate the case that"
+ echo "some name entry should be inserted into other non-first"
+ echo "shard, but inserted into the first shard by wrong"
+
+ #define OBD_FAIL_LFSCK_BAD_NAME_HASH 0x1628
+ $LCTL set_param fail_loc=0x1628 fail_val=0
+ createmany -d $DIR/$tdir/striped_dir/d $MDSCOUNT ||
+ error "(2) Fail to create file under striped directory"
+ $LCTL set_param fail_loc=0 fail_val=0
+
+ echo "Trigger namespace LFSCK to repair bad name hash"
+ $START_NAMESPACE -r -A ||
+ error "(3) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 4
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^name_hash_repaired/ { print $2 }')
+ [ $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
+ stat $DIR/$tdir/striped_dir/d$i ||
+ error "(8) Fail to stat d$i after LFSCK"
+ rmdir $DIR/$tdir/striped_dir/d$i ||
+ error "(9) Fail to unlink d$i after LFSCK"
+ done
+
+ rmdir $DIR/$tdir/striped_dir ||
+ error "(10) Fail to remove the striped directory after LFSCK"
+}
+run_test 31a "The LFSCK can find/repair the name entry with bad name hash (1)"
+
+test_31b() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "The test needs at least 2 MDTs" && return
+
+ echo "#####"
+ echo "For the name entry under a striped directory, if the name"
+ echo "hash does not match the shard, then the LFSCK will repair"
+ echo "the bad name entry"
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
+ error "(1) Fail to create striped directory"
+
+ echo "Inject failure stub on client to simulate the case that"
+ echo "some name entry should be inserted into other non-second"
+ echo "shard, but inserted into the secod shard by wrong"
+
+ #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 ||
+ error "(2) Fail to create file under striped directory"
+ $LCTL set_param fail_loc=0 fail_val=0
+
+ echo "Trigger namespace LFSCK to repair bad name hash"
+ $START_NAMESPACE -r -A ||
+ error "(3) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 4
+
+ local repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^name_hash_repaired/ { print $2 }')
+ [ $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
+ stat $DIR/$tdir/striped_dir/d$i ||
+ error "(8) Fail to stat d$i after LFSCK"
+ rmdir $DIR/$tdir/striped_dir/d$i ||
+ error "(9) Fail to unlink d$i after LFSCK"
+ done
+
+ rmdir $DIR/$tdir/striped_dir ||
+ error "(10) Fail to remove the striped directory after LFSCK"
+}
+run_test 31b "The LFSCK can find/repair the name entry with bad name hash (2)"
+
+test_31c() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "The test needs at least 2 MDTs" && return
+
+ echo "#####"
+ echo "For some reason, the master MDT-object of the striped directory"
+ echo "may lost its master LMV EA. If nobody created files under the"
+ echo "master directly after the master LMV EA lost, then the LFSCK"
+ echo "should re-generate the master LMV EA."
+ echo "#####"
+
+ check_mount_and_prep
+
+ echo "Inject failure stub on MDT0 to simulate the case that the"
+ echo "master MDT-object of the striped directory lost the LMV EA."
+
+ #define OBD_FAIL_LFSCK_LOST_MASTER_LMV 0x1629
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1629
+ $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
+ error "(1) Fail to create striped directory"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ echo "Trigger namespace LFSCK to re-generate master LMV EA"
+ $START_NAMESPACE -r -A ||
+ error "(2) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 3
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^striped_dirs_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(4) Fail to re-generate master LMV EA: $repaired"
+
+ umount_client $MOUNT || error "(5) umount failed"
+ mount_client $MOUNT || error "(6) mount failed"
+
+ local empty=$(ls $DIR/$tdir/striped_dir/)
+ [ -z "$empty" ] || error "(7) The master LMV EA is not repaired: $empty"
+
+ rmdir $DIR/$tdir/striped_dir ||
+ error "(8) Fail to remove the striped directory after LFSCK"
+}
+run_test 31c "Re-generate the lost master LMV EA for striped directory"
+
+test_31d() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "The test needs at least 2 MDTs" && return
+
+ echo "#####"
+ echo "For some reason, the master MDT-object of the striped directory"
+ echo "may lost its master LMV EA. If somebody created files under the"
+ echo "master directly after the master LMV EA lost, then the LFSCK"
+ echo "should NOT re-generate the master LMV EA, instead, it should"
+ echo "change the broken striped dirctory as read-only to prevent"
+ echo "further damage"
+ echo "#####"
+
+ check_mount_and_prep
+
+ echo "Inject failure stub on MDT0 to simulate the case that the"
+ echo "master MDT-object of the striped directory lost the LMV EA."
+
+ #define OBD_FAIL_LFSCK_LOST_MASTER_LMV 0x1629
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1629
+ $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
+ error "(1) Fail to create striped directory"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0
+
+ umount_client $MOUNT || error "(2) umount failed"
+ mount_client $MOUNT || error "(3) mount failed"
+
+ touch $DIR/$tdir/striped_dir/dummy ||
+ error "(4) Fail to touch under broken striped directory"
+
+ echo "Trigger namespace LFSCK to find out the inconsistency"
+ $START_NAMESPACE -r -A ||
+ error "(5) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 6
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^striped_dirs_repaired/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(7) Re-generate master LMV EA unexpected: $repaired"
+
+ stat $DIR/$tdir/striped_dir/dummy ||
+ error "(8) Fail to stat $DIR/$tdir/striped_dir/dummy"
+
+ touch $DIR/$tdir/striped_dir/foo &&
+ error "(9) The broken striped directory should be read-only"
+
+ chattr -i $DIR/$tdir/striped_dir ||
+ error "(10) Fail to chattr on the broken striped directory"
+
+ rmdir $DIR/$tdir/striped_dir ||
+ error "(11) Fail to remove the striped directory after LFSCK"
+}
+run_test 31d "Set broken striped directory (modified after broken) as read-only"
+
+test_31e() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "The test needs at least 2 MDTs" && return
+
+ echo "#####"
+ echo "For some reason, the slave MDT-object of the striped directory"
+ echo "may lost its slave LMV EA. The LFSCK should re-generate the"
+ echo "slave LMV EA."
+ echo "#####"
+
+ check_mount_and_prep
+
+ echo "Inject failure stub on MDT0 to simulate the case that the"
+ echo "slave MDT-object (that resides on the same MDT as the master"
+ echo "MDT-object resides on) lost the LMV EA."
+
+ #define OBD_FAIL_LFSCK_LOST_SLAVE_LMV 0x162a
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x162a fail_val=0
+ $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
+ error "(1) Fail to create striped directory"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0 fail_val=0
+
+ echo "Trigger namespace LFSCK to re-generate slave LMV EA"
+ $START_NAMESPACE -r -A ||
+ error "(2) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 3
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^striped_shards_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(4) Fail to re-generate slave LMV EA: $repaired"
+
+ rmdir $DIR/$tdir/striped_dir ||
+ error "(5) Fail to remove the striped directory after LFSCK"
+}
+run_test 31e "Re-generate the lost slave LMV EA for striped directory (1)"
+
+test_31f() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "The test needs at least 2 MDTs" && return
+
+ echo "#####"
+ echo "For some reason, the slave MDT-object of the striped directory"
+ echo "may lost its slave LMV EA. The LFSCK should re-generate the"
+ echo "slave LMV EA."
+ echo "#####"
+
+ check_mount_and_prep
+
+ echo "Inject failure stub on MDT0 to simulate the case that the"
+ echo "slave MDT-object (that resides on different MDT as the master"
+ echo "MDT-object resides on) lost the LMV EA."
+
+ #define OBD_FAIL_LFSCK_LOST_SLAVE_LMV 0x162a
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x162a fail_val=1
+ $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
+ error "(1) Fail to create striped directory"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0 fail_val=0
+
+ echo "Trigger namespace LFSCK to re-generate slave LMV EA"
+ $START_NAMESPACE -r -A ||
+ error "(2) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 3
+
+ local repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^striped_shards_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(4) Fail to re-generate slave LMV EA: $repaired"
+
+ rmdir $DIR/$tdir/striped_dir ||
+ error "(5) Fail to remove the striped directory after LFSCK"
+}
+run_test 31f "Re-generate the lost slave LMV EA for striped directory (2)"
+
+test_31g() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "The test needs at least 2 MDTs" && return
+
+ echo "#####"
+ echo "For some reason, the stripe index in the slave LMV EA is"
+ echo "corrupted. The LFSCK should repair the slave LMV EA."
+ echo "#####"
+
+ check_mount_and_prep
+
+ echo "Inject failure stub on MDT0 to simulate the case that the"
+ echo "slave LMV EA on the first shard of the striped directory"
+ echo "claims the same index as the second shard claims"
+
+ #define OBD_FAIL_LFSCK_BAD_SLAVE_LMV 0x162b
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x162b fail_val=0
+ $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
+ error "(1) Fail to create striped directory"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0 fail_val=0
+
+ echo "Trigger namespace LFSCK to repair the slave LMV EA"
+ $START_NAMESPACE -r -A ||
+ error "(2) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 3
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^striped_shards_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(4) Fail to repair slave LMV EA: $repaired"
+
+ umount_client $MOUNT || error "(5) umount failed"
+ mount_client $MOUNT || error "(6) mount failed"
+
+ touch $DIR/$tdir/striped_dir/foo ||
+ error "(7) Fail to touch file after the LFSCK"
+
+ rm -f $DIR/$tdir/striped_dir/foo ||
+ error "(8) Fail to unlink file after the LFSCK"
+
+ rmdir $DIR/$tdir/striped_dir ||
+ error "(9) Fail to remove the striped directory after LFSCK"
+}
+run_test 31g "Repair the corrupted slave LMV EA"
+
+test_31h() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "The test needs at least 2 MDTs" && return
+
+ echo "#####"
+ echo "For some reason, the shard's name entry in the striped"
+ echo "directory may be corrupted. The LFSCK should repair the"
+ echo "bad shard's name entry."
+ echo "#####"
+
+ check_mount_and_prep
+
+ echo "Inject failure stub on MDT0 to simulate the case that the"
+ echo "first shard's name entry in the striped directory claims"
+ echo "the same index as the second shard's name entry claims."
+
+ #define OBD_FAIL_LFSCK_BAD_SLAVE_NAME 0x162c
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x162c fail_val=0
+ $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
+ error "(1) Fail to create striped directory"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0 fail_val=0
+
+ echo "Trigger namespace LFSCK to repair the shard's name entry"
+ $START_NAMESPACE -r -A ||
+ error "(2) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 3
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^dirent_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(4) Fail to repair shard's name entry: $repaired"
+
+ umount_client $MOUNT || error "(5) umount failed"
+ mount_client $MOUNT || error "(6) mount failed"
+
+ touch $DIR/$tdir/striped_dir/foo ||
+ error "(7) Fail to touch file after the LFSCK"
+
+ rm -f $DIR/$tdir/striped_dir/foo ||
+ error "(8) Fail to unlink file after the LFSCK"
+
+ rmdir $DIR/$tdir/striped_dir ||
+ error "(9) Fail to remove the striped directory after LFSCK"
+}
+run_test 31h "Repair the corrupted shard's name entry"
+
+test_32()
+{
+ lfsck_prep 5 5
+ umount_client $MOUNT
+
+ #define OBD_FAIL_LFSCK_ASSISTANT_DIRECT 0x162d
+ do_facet $SINGLEMDS $LCTL set_param fail_val=3 fail_loc=0x162d
+ $START_LAYOUT -r || error "(2) Fail to start LFSCK for layout!"
+
+ local STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
+ [ "$STATUS" == "scanning-phase1" ] ||
+ error "(3) Expect 'scanning-phase1', but got '$STATUS'"
+
+ echo "stop ost1"
+ stop ost1 > /dev/null || error "(4) Fail to stop OST1!"
+
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
+ sleep 1
+
+ echo "stop LFSCK"
+ $STOP_LFSCK || error "(5) Fail to stop LFSCK!"
+}
+run_test 32 "stop LFSCK when some OST failed"
# restore MDS/OST size
MDSSIZE=${SAVED_MDSSIZE}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff