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}
# use small MDS + OST size to speed formatting time
# do not use too small MDSSIZE/OSTSIZE, which affect the default journal size
MDSSIZE=100000
+[ $(facet_fstype $SINGLEMDS) == zfs ] && MDSSIZE=300000
OSTSIZE=100000
+[ $(facet_fstype ost1) == 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) -lt $(version_code 2.3.60) ]] &&
- skip "Need MDS version at least 2.3.60" && check_and_cleanup_lustre &&
- exit 0
+REFORMAT="yes" check_and_setup_lustre
[[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.4.90) ]] &&
ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2c"
$LCTL get_param -n obdfilter.${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
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
[ $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
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
}
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_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 ||
test_2e()
{
- [ $MDSCOUNT -lt 2 ] &&
- skip "We need at least 2 MDSes for this test" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
check_mount_and_prep
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 $MOUNT || error "(10) Fail to start client!"
#define OBD_FAIL_FID_LOOKUP 0x1505
local D_POS1=$($SHOW_NAMESPACE |
awk '/^latest_start_position/ { print $4 }')
+ echo "Additional debug for 6b"
+ $SHOW_NAMESPACE
if [ "$D_POS0" == "N/A" -o "$D_POS1" == "N/A" ]; then
[[ $O_POS0 -lt $O_POS1 ]] ||
error "(7.1) $O_POS1 is not larger than $O_POS0"
[ "$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 $LCTL set_param fail_val=2 fail_loc=0x1602
# There may be time error, normally it should be less than 2 seconds.
# We allow another 20% schedule error.
local TIME_DIFF=2
- # MAX_MARGIN = 1.2 = 12 / 10
+ # MAX_MARGIN = 1.3 = 13 / 10
local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) / \
- RUN_TIME1 * 12 / 10))
- [ $SPEED -lt $MAX_SPEED ] ||
- error "(4) Got speed $SPEED, expected less than $MAX_SPEED"
+ RUN_TIME1 * 13 / 10))
+ [ $SPEED -lt $MAX_SPEED ] || {
+ $SHOW_LAYOUT
+ log "speed1: $BASE_SPEED1 time1: $RUN_TIME1"
+ error "(4) Speed $SPEED, expected < $MAX_SPEED"
+ }
# adjust speed limit
local BASE_SPEED2=300
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 $SINGLEMDS) != ldiskfs ]; then
error_ignore LU-5624 \
fi
}
- # MAX_MARGIN = 1.2 = 12 / 10
+ # MAX_MARGIN = 1.3 = 13 / 10
MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
- (RUN_TIME1 + RUN_TIME2) * 12 / 10))
- [ $SPEED -lt $MAX_SPEED ] ||
- error "(6) Got speed $SPEED, expected less than $MAX_SPEED"
+ (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 "(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.${MDT_DEV}.lfsck_layout |
# There may be time error, normally it should be less than 2 seconds.
# We allow another 20% schedule error.
local TIME_DIFF=2
- # MAX_MARGIN = 1.2 = 12 / 10
+ # MAX_MARGIN = 1.3 = 13 / 10
local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) / \
- RUN_TIME1 * 12 / 10))
- [ $SPEED -lt $MAX_SPEED ] ||
- error "(8) Got speed $SPEED, expected less than $MAX_SPEED"
+ RUN_TIME1 * 13 / 10))
+ [ $SPEED -lt $MAX_SPEED ] || {
+ $SHOW_NAMESPACE
+ log "speed1: $BASE_SPEED1 time1: $RUN_TIME1"
+ error "(8) Speed $SPEED, expected < $MAX_SPEED"
+ }
# adjust speed limit
local BASE_SPEED2=150
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 $SINGLEMDS) != ldiskfs ]; then
error_ignore LU-5624 \
fi
}
- # MAX_MARGIN = 1.2 = 12 / 10
+ # MAX_MARGIN = 1.3 = 13 / 10
MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
- (RUN_TIME1 + RUN_TIME2) * 12 / 10))
- [ $SPEED -lt $MAX_SPEED ] ||
- error "(10) Got speed $SPEED, expected less than $MAX_SPEED"
+ (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 "(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.${MDT_DEV}.lfsck_namespace |
awk '/^status/ { print \\\$2 }'" "completed" 32 || {
[ "$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 || {
}
run_test 11b "LFSCK can rebuild crashed last_id"
-test_12() {
- [ $MDSCOUNT -lt 2 ] &&
- skip "We need at least 2 MDSes for test_12" && return
+test_12a() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
check_mount_and_prep
for k in $(seq $MDSCOUNT); do
-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
-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"
+run_test 12a "single command to trigger LFSCK on all devices"
+
+test_12b() {
+ check_mount_and_prep
+
+ echo "Start LFSCK without '-M' specified."
+ do_facet mds1 $LCTL lfsck_start -A -r ||
+ error "(0) Fail to start LFSCK without '-M'"
+
+ wait_all_targets_blocked namespace completed 1
+ wait_all_targets_blocked layout completed 2
+
+ local count=$(do_facet mds1 $LCTL dl |
+ awk '{ print $3 }' | grep mdt | wc -l)
+ if [ $count -gt 1 ]; then
+ echo
+ echo "Start layout LFSCK on the node with multipe targets,"
+ echo "but not specify '-M'/'-A' option. Should get failure."
+ echo
+ do_facet mds1 $LCTL lfsck_start -t layout -r &&
+ error "(3) Start layout LFSCK should fail" || true
+ fi
+}
+run_test 12b "auto detect Lustre device"
test_13() {
echo "#####"
echo "Inject failure stub to simulate bad lmm_oi"
#define OBD_FAIL_LFSCK_BAD_LMMOI 0x160f
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160f
- createmany -o $DIR/$tdir/f 32
+ createmany -o $DIR/$tdir/f 1
+ $LFS setstripe -E 1M -S 1M -E -1 $DIR/$tdir/f1 ||
+ error "(0) Fail to create PFL $DIR/$tdir/f1"
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
echo "Trigger layout LFSCK to find out the bad lmm_oi and fix them"
local repaired=$($SHOW_LAYOUT |
awk '/^repaired_others/ { print $2 }')
- [ $repaired -eq 32 ] ||
+ [ $repaired -eq 2 ] ||
error "(3) Fail to repair crashed lmm_oi: $repaired"
}
run_test 13 "LFSCK can repair crashed lmm_oi"
-test_14() {
+test_14a() {
echo "#####"
echo "The OST-object referenced by the MDT-object should be there;"
echo "otherwise, the LFSCK should re-create the missing OST-object."
+ echo "without '--delay-create-ostobj' option."
echo "#####"
check_mount_and_prep
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
+ createmany -o $DIR/$tdir/f $((count + 16)) ||
+ error "(0.1) Fail to create $DIR/$tdir/fx"
+ touch $DIR/$tdir/guard0
+
+ for ((i = 0; i < 16; i++)); do
+ $LFS setstripe -E 512K -S 256K -o 0 -E 2M \
+ $DIR/$tdir/f_comp${i} ||
+ error "(0.2) Fail to create $DIR/$tdir/f_comp${i}"
+ done
+ touch $DIR/$tdir/guard1
+
do_facet ost1 $LCTL set_param fail_loc=0
start_full_debug_logging
# exhaust other pre-created dangling cases
count=$(precreated_ost_obj_count 0 0)
createmany -o $DIR/$tdir/a $count ||
- error "(0) Fail to create $count files."
+ error "(0.5) Fail to create $count files."
echo "'ls' should fail because of dangling referenced MDT-object"
ls -ail $DIR/$tdir > /dev/null 2>&1 && error "(1) ls should fail."
error "(4) Fail to repair dangling reference: $repaired"
echo "'stat' should fail because of not repair dangling by default"
- stat $DIR/$tdir/guard > /dev/null 2>&1 && error "(5) stat should fail"
+ stat $DIR/$tdir/guard0 > /dev/null 2>&1 &&
+ error "(5.1) stat should fail"
+ stat $DIR/$tdir/guard1 > /dev/null 2>&1 &&
+ error "(5.2) stat should fail"
echo "Trigger layout LFSCK to repair dangling reference"
$START_LAYOUT -r -c || error "(6) Fail to start LFSCK for layout!"
# a while until the target reparation has been done. LU-4970.
echo "'stat' should success after layout LFSCK repairing"
+ wait_update_facet client "stat $DIR/$tdir/guard0 |
+ awk '/Size/ { print \\\$2 }'" "0" 32 || {
+ stat $DIR/$tdir/guard0
+ $SHOW_LAYOUT
+ error "(8.1) unexpected size"
+ }
+
+ wait_update_facet client "stat $DIR/$tdir/guard1 |
+ awk '/Size/ { print \\\$2 }'" "0" 32 || {
+ stat $DIR/$tdir/guard1
+ $SHOW_LAYOUT
+ error "(8.2) unexpected size"
+ }
+
+ repaired=$($SHOW_LAYOUT |
+ awk '/^repaired_dangling/ { print $2 }')
+ [ $repaired -ge 32 ] ||
+ error "(9) Fail to repair dangling reference: $repaired"
+
+ stop_full_debug_logging
+
+ echo "stopall to cleanup object cache"
+ stopall > /dev/null
+ echo "setupall"
+ setupall > /dev/null
+}
+run_test 14a "LFSCK can repair MDT-object with dangling LOV EA reference (1)"
+
+test_14b() {
+ echo "#####"
+ echo "The OST-object referenced by the MDT-object should be there;"
+ echo "otherwise, the LFSCK should re-create the missing OST-object."
+ echo "with '--delay-create-ostobj' option."
+ echo "#####"
+
+ check_mount_and_prep
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir
+
+ 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 $LCTL set_param fail_loc=0
+
+ start_full_debug_logging
+
+ # exhaust other pre-created dangling cases
+ count=$(precreated_ost_obj_count 0 0)
+ createmany -o $DIR/$tdir/a $count ||
+ error "(0) Fail to create $count files."
+
+ echo "'ls' should fail because of dangling referenced MDT-object"
+ ls -ail $DIR/$tdir > /dev/null 2>&1 && error "(1) ls should fail."
+
+ echo "Trigger layout LFSCK to find out dangling reference"
+ $START_LAYOUT -r -o -d || error "(2) Fail to start LFSCK for layout!"
+
+ wait_all_targets_blocked layout completed 3
+
+ local repaired=$($SHOW_LAYOUT |
+ awk '/^repaired_dangling/ { print $2 }')
+ [ $repaired -ge 32 ] ||
+ error "(4) Fail to repair dangling reference: $repaired"
+
+ echo "'stat' should fail because of not repair dangling by default"
+ stat $DIR/$tdir/guard > /dev/null 2>&1 && error "(5) stat should fail"
+
+ echo "Trigger layout LFSCK to repair dangling reference"
+ $START_LAYOUT -r -o -c -d || error "(6) Fail to start LFSCK for layout!"
+
+ wait_all_targets_blocked layout completed 7
+
+ # There may be some async LFSCK updates in processing, wait for
+ # a while until the target reparation has been done. LU-4970.
+
+ echo "'stat' should success after layout LFSCK repairing"
wait_update_facet client "stat $DIR/$tdir/guard |
awk '/Size/ { print \\\$2 }'" "0" 32 || {
stat $DIR/$tdir/guard
error "(9) Fail to repair dangling reference: $repaired"
stop_full_debug_logging
+
+ echo "stopall to cleanup object cache"
+ stopall > /dev/null
+ echo "setupall"
+ setupall > /dev/null
}
-run_test 14 "LFSCK can repair MDT-object with dangling reference"
+run_test 14b "LFSCK can repair MDT-object with dangling LOV EA reference (2)"
test_15a() {
echo "#####"
echo "non-exist MDT-object."
#define OBD_FAIL_LFSCK_UNMATCHED_PAIR1 0x1611
- do_facet ost1 $LCTL set_param fail_loc=0x1611
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param fail_loc=0x1611
dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1
+ $LFS setstripe -E 1M -S 256K -c 1 -E -1 -S 512K -c $OSTCOUNT \
+ $DIR/$tdir/f1 ||
+ error "(0) Fail to create PFL $DIR/$tdir/f1"
+ # 'dd' will trigger punch RPC firstly on every OST-objects.
+ # So even though some OST-object will not be write by 'dd',
+ # as long as it is allocated (may be NOT allocated in pfl_3b)
+ # its layout information will be set also.
+ dd if=/dev/zero of=$DIR/$tdir/f1 bs=4K count=257
cancel_lru_locks osc
- do_facet ost1 $LCTL set_param fail_loc=0
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param fail_loc=0
echo "Trigger layout LFSCK to find out unmatched pairs and fix them"
$START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
local repaired=$($SHOW_LAYOUT |
awk '/^repaired_unmatched_pair/ { print $2 }')
- [ $repaired -eq 1 ] ||
+ [ $repaired -ge 3 ] ||
error "(3) Fail to repair unmatched pair: $repaired"
}
run_test 15a "LFSCK can repair unmatched MDT-object/OST-object pairs (1)"
echo "#####"
check_mount_and_prep
- $LFS setstripe -c 1 -i 0 $DIR/$tdir
- dd if=/dev/zero of=$DIR/$tdir/guard bs=1M count=1
+ mkdir -p $DIR/$tdir/0
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir/0
+ dd if=/dev/zero of=$DIR/$tdir/0/guard bs=1M count=1
cancel_lru_locks osc
echo "Inject failure stub to make the OST-object to back point to"
echo "other MDT-object"
+ local stripes=1
+ [ $OSTCOUNT -ge 2 ] && stripes=2
+
#define OBD_FAIL_LFSCK_UNMATCHED_PAIR2 0x1612
- do_facet ost1 $LCTL set_param fail_loc=0x1612
- dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param fail_loc=0x1612
+ dd if=/dev/zero of=$DIR/$tdir/0/f0 bs=1M count=1
+ $LFS setstripe -E 1M -S 256K -c $stripes -E 2M -S 512K -c 1 \
+ $DIR/$tdir/f1 ||
+ error "(0) Fail to create PFL $DIR/$tdir/f1"
+ dd if=/dev/zero of=$DIR/$tdir/f1 bs=1M count=2
cancel_lru_locks osc
- do_facet ost1 $LCTL set_param fail_loc=0
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param fail_loc=0
echo "Trigger layout LFSCK to find out unmatched pairs and fix them"
$START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
local repaired=$($SHOW_LAYOUT |
awk '/^repaired_unmatched_pair/ { print $2 }')
- [ $repaired -eq 1 ] ||
+ [ $repaired -eq 4 ] ||
error "(3) Fail to repair unmatched pair: $repaired"
}
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
+ [ $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
echo "#####"
echo "According to current metadata migration implementation,"
#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 mv -M 0 $DIR/$tdir/a1 &
+ $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!"
- 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" $LTIME ||
- error "(2) MDS${k} is not the expected 'completed'"
- done
+ wait_all_targets_blocked layout completed 2
do_facet mds2 $LCTL set_param fail_loc=0 fail_val=0
local repaired=$($SHOW_LAYOUT |
dd if=/dev/zero of=$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
#define OBD_FAIL_LFSCK_MULTIPLE_REF 0x1614
do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0x1614
-
dd if=/dev/zero of=$DIR/$tdir/guard bs=1M count=1
+ cancel_lru_locks mdc
cancel_lru_locks osc
createmany -o $DIR/$tdir/f 1
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
-
+ $LFS setstripe -E 2M -S 256K -o 0 -E 4M -S 512K -o 0 \
+ $DIR/$tdir/f1 ||
+ error "(0) Fail to create PFL $DIR/$tdir/f1"
cancel_lru_locks mdc
cancel_lru_locks osc
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
echo "$DIR/$tdir/f0 and $DIR/$tdir/guard use the same OST-objects"
+ echo "$DIR/$tdir/f1 and $DIR/$tdir/guard use the same OST-objects"
local size=$(ls -l $DIR/$tdir/f0 | awk '{ print $5 }')
[ $size -eq 1048576 ] ||
- error "(1) f0 (wrong) size should be 1048576, but got $size"
+ error "(1.1) f0 (wrong) size should be 1048576, but got $size"
+
+ size=$(ls -l $DIR/$tdir/f1 | awk '{ print $5 }')
+ [ $size -eq 1048576 ] ||
+ error "(1.2) f1 (wrong) size should be 1048576, but got $size"
echo "Trigger layout LFSCK to find out multiple refenced MDT-objects"
echo "and fix them"
local repaired=$($SHOW_LAYOUT |
awk '/^repaired_multiple_referenced/ { print $2 }')
- [ $repaired -eq 1 ] ||
+ [ $repaired -eq 2 ] ||
error "(4) Fail to repair multiple references: $repaired"
echo "$DIR/$tdir/f0 and $DIR/$tdir/guard should use diff OST-objects"
size=$(ls -l $DIR/$tdir/guard | awk '{ print $5 }')
[ $size -eq 1048576 ] ||
error "(6) guard size should be 1048576, but got $size"
+
+ echo "$DIR/$tdir/f1 and $DIR/$tdir/guard should use diff OST-objects"
+ dd if=/dev/zero of=$DIR/$tdir/f1 bs=1M count=2 ||
+ error "(7) Fail to write f1."
+ size=$(ls -l $DIR/$tdir/guard | awk '{ print $5 }')
+ [ $size -eq 1048576 ] ||
+ error "(8) guard size should be 1048576, but got $size"
}
run_test 17 "LFSCK can repair multiple references"
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 $DIR/$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 $6 }')
+ local saved_size1=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
$LFS path2fid $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a2/f2
fi
+ $LFS setstripe -E 1M -S 1M -o 0 -E -1 -S 1M $DIR/$tdir/f3 ||
+ error "(0) Fail to create PFL $DIR/$tdir/f3"
+
+ dd if=/dev/zero of=$DIR/$tdir/f3 bs=1M count=2
+
+ local saved_size2=$(ls -il $DIR/$tdir/f3 | awk '{ print $6 }')
+
+ $LFS path2fid $DIR/$tdir/f3
+ $LFS getstripe $DIR/$tdir/f3
+
cancel_lru_locks osc
echo "Inject failure, to make the MDT-object lost its layout EA"
chown 1.1 $DIR/$tdir/a2/f2
fi
+ chown 1.1 $DIR/$tdir/f3
+
sync
sleep 2
echo "The file size should be incorrect since layout EA is lost"
local cur_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
- [ "$cur_size" != "$saved_size" ] ||
+ [ "$cur_size" != "$saved_size1" ] ||
error "(1) Expect incorrect file1 size"
if [ $MDSCOUNT -ge 2 ]; then
cur_size=$(ls -il $DIR/$tdir/a2/f2 | awk '{ print $6 }')
- [ "$cur_size" != "$saved_size" ] ||
+ [ "$cur_size" != "$saved_size1" ] ||
error "(2) Expect incorrect file2 size"
fi
+ cur_size=$(ls -il $DIR/$tdir/f3 | awk '{ print $6 }')
+ [ "$cur_size" != "$saved_size2" ] ||
+ error "(1.2) Expect incorrect file3 size"
+
echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
$START_LAYOUT -r -o || error "(3) Fail to start LFSCK for layout!"
local repaired=$(do_facet mds1 $LCTL get_param -n \
mdd.$(facet_svc mds1).lfsck_layout |
awk '/^repaired_orphan/ { print $2 }')
- [ $repaired -eq 1 ] ||
- error "(6.1) Expect 1 fixed on mds1, but got: $repaired"
+ [ $repaired -eq 3 ] ||
+ error "(6.1) Expect 3 fixed on mds1, but got: $repaired"
if [ $MDSCOUNT -ge 2 ]; then
repaired=$(do_facet mds2 $LCTL get_param -n \
$LFS getstripe $DIR/$tdir/a2/f2
fi
+ $LFS path2fid $DIR/$tdir/f3
+ $LFS getstripe $DIR/$tdir/f3
+
echo "The file size should be correct after layout LFSCK scanning"
cur_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
- [ "$cur_size" == "$saved_size" ] ||
- error "(7) Expect file1 size $saved_size, but got $cur_size"
+ [ "$cur_size" == "$saved_size1" ] ||
+ error "(7) Expect file1 size $saved_size1, but got $cur_size"
if [ $MDSCOUNT -ge 2 ]; then
cur_size=$(ls -il $DIR/$tdir/a2/f2 | awk '{ print $6 }')
- [ "$cur_size" == "$saved_size" ] ||
- error "(8) Expect file2 size $saved_size, but got $cur_size"
+ [ "$cur_size" == "$saved_size1" ] ||
+ error "(8) Expect file2 size $saved_size1, but got $cur_size"
fi
+
+ cur_size=$(ls -il $DIR/$tdir/f3 | awk '{ print $6 }')
+ [ "$cur_size" == "$saved_size2" ] ||
+ error "(9) Expect file1 size $saved_size2, but got $cur_size"
}
run_test 18a "Find out orphan OST-object and repair it (1)"
test_18b() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+
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"
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 $DIR/$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 $6 }')
+ local saved_size1=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
local fid1=$($LFS path2fid $DIR/$tdir/a1/f1)
echo ${fid1}
$LFS getstripe $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a2/f2
fi
+ $LFS setstripe -E 1M -S 1M -o 0 -E -1 -S 1M $DIR/$tdir/f3 ||
+ error "(0) Fail to create PFL $DIR/$tdir/f3"
+
+ dd if=/dev/zero of=$DIR/$tdir/f3 bs=1M count=2
+
+ local saved_size2=$(ls -il $DIR/$tdir/f3 | awk '{ print $6 }')
+ local fid3=$($LFS path2fid $DIR/$tdir/f3)
+ echo ${fid3}
+ $LFS getstripe $DIR/$tdir/f3
+
cancel_lru_locks osc
echo "Inject failure, to simulate the case of missing the MDT-object"
rm -f $DIR/$tdir/a2/f2
fi
+ rm -f $DIR/$tdir/f3
+
sync
sleep 2
local repaired=$(do_facet mds1 $LCTL get_param -n \
mdd.$(facet_svc mds1).lfsck_layout |
awk '/^repaired_orphan/ { print $2 }')
- [ $repaired -eq 1 ] ||
- error "(4.1) Expect 1 fixed on mds1, but got: $repaired"
+ [ $repaired -eq 3 ] ||
+ error "(4.1) Expect 3 fixed on mds1, but got: $repaired"
if [ $MDSCOUNT -ge 2 ]; then
repaired=$(do_facet mds2 $LCTL get_param -n \
mv $name $DIR/$tdir/a2/f2 || error "(6) Fail to move $name"
fi
+ mv $MOUNT/.lustre/lost+found/MDT0000/${fid3}-R-0 $DIR/$tdir/f3 ||
+ error "(5) Fail to move $MOUNT/.lustre/lost+found/MDT0000/${fid3}-R-0"
+
$LFS path2fid $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a2/f2
fi
+ $LFS path2fid $DIR/$tdir/f3
+ $LFS getstripe $DIR/$tdir/f3
+
echo "The file size should be correct after layout LFSCK scanning"
local cur_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
- [ "$cur_size" == "$saved_size" ] ||
- error "(7) Expect file1 size $saved_size, but got $cur_size"
+ [ "$cur_size" == "$saved_size1" ] ||
+ error "(7) Expect file1 size $saved_size1, but got $cur_size"
if [ $MDSCOUNT -ge 2 ]; then
cur_size=$(ls -il $DIR/$tdir/a2/f2 | awk '{ print $6 }')
- [ "$cur_size" == "$saved_size" ] ||
- error "(8) Expect file2 size $saved_size, but got $cur_size"
+ [ "$cur_size" == "$saved_size1" ] ||
+ error "(8) Expect file2 size $saved_size1, but got $cur_size"
fi
+
+ cur_size=$(ls -il $DIR/$tdir/f3 | awk '{ print $6 }')
+ [ "$cur_size" == "$saved_size2" ] ||
+ error "(9) Expect file1 size $saved_size2, but got $cur_size"
}
run_test 18b "Find out orphan OST-object and repair it (2)"
test_18c() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+
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 "
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 $DIR/$tdir/a1
echo "Inject failure, to simulate the case of missing parent FID"
#define OBD_FAIL_LFSCK_NOPFID 0x1617
- do_facet ost1 $LCTL set_param fail_loc=0x1617
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param fail_loc=0x1617
dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
$LFS getstripe $DIR/$tdir/a1/f1
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 $DIR/$tdir/a2
dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
$LFS getstripe $DIR/$tdir/a2/f2
fi
+ $LFS setstripe -E 1M -S 1M -o 0 -E -1 -S 1M $DIR/$tdir/f3 ||
+ error "(0) Fail to create PFL $DIR/$tdir/f3"
+
+ dd if=/dev/zero of=$DIR/$tdir/f3 bs=1M count=2
+ $LFS getstripe $DIR/$tdir/f3
+
cancel_lru_locks osc
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param fail_loc=0
echo "Inject failure, to simulate the case of missing the MDT-object"
#define OBD_FAIL_LFSCK_LOST_MDTOBJ 0x1616
rm -f $DIR/$tdir/a2/f2
fi
+ rm -f $DIR/$tdir/f3
+
sync
sleep 2
done
if [ $MDSCOUNT -ge 2 ]; then
- expected=2
+ expected=4
else
- expected=1
+ expected=3
fi
local repaired=$(do_facet mds1 $LCTL get_param -n \
test_18d() {
echo "#####"
- echo "The target MDT-object layout EA slot is occpuied by some new"
- echo "created OST-object when repair dangling reference case. Such"
- echo "conflict OST-object has never been modified. Then when found"
- echo "the orphan OST-object, LFSCK will replace it with the orphan"
- echo "OST-object."
+ echo "The target MDT-object layout EA is corrupted, but the right"
+ echo "OST-object is still alive as orphan. The layout LFSCK will"
+ echo "not create new OST-object to occupy such slot."
echo "#####"
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 $DIR/$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 }')
+
+ echo "guard" > $DIR/$tdir/a1/f3
+ $LFS setstripe -E 1M -S 1M -o 0 -E -1 -S 1M $DIR/$tdir/a1/f4 ||
+ error "(0) Fail to create PFL $DIR/$tdir/a1/f4"
+ echo "foo" > $DIR/$tdir/a1/f4
+
+ local saved_size1=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
+ local saved_size2=$(ls -il $DIR/$tdir/a1/f4 | awk '{ print $6 }')
$LFS path2fid $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a1/f1
$LFS path2fid $DIR/$tdir/a1/f2
$LFS getstripe $DIR/$tdir/a1/f2
+ $LFS path2fid $DIR/$tdir/a1/f3
+ $LFS getstripe $DIR/$tdir/a1/f3
+ $LFS path2fid $DIR/$tdir/a1/f4
+ $LFS getstripe $DIR/$tdir/a1/f4
cancel_lru_locks osc
echo "Inject failure to make $DIR/$tdir/a1/f1 and $DIR/$tdir/a1/f2"
echo "to reference the same OST-object (which is f1's OST-obejct)."
echo "Then drop $DIR/$tdir/a1/f1 and its OST-object, so f2 becomes"
echo "dangling reference case, but f2's old OST-object is there."
+
+ echo "The failure also makes $DIR/$tdir/a1/f3 and $DIR/$tdir/a1/f4"
+ echo "to reference the same OST-object (which is f3's OST-obejct)."
+ echo "Then drop $DIR/$tdir/a1/f3 and its OST-object, so f4 becomes"
+ echo "dangling reference case, but f4's old OST-object is there."
echo
#define OBD_FAIL_LFSCK_CHANGE_STRIPE 0x1618
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1618
chown 1.1 $DIR/$tdir/a1/f2
+ chown 1.1 $DIR/$tdir/a1/f4
rm -f $DIR/$tdir/a1/f1
+ rm -f $DIR/$tdir/a1/f3
sync
sleep 2
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
echo "setupall"
setupall > /dev/null
- echo "The file size should be incorrect since dangling referenced"
- local cur_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
- [ "$cur_size" != "$saved_size" ] ||
- error "(1) Expect incorrect file2 size"
-
- #define OBD_FAIL_LFSCK_DELAY3 0x1602
- do_facet $SINGLEMDS $LCTL set_param fail_val=5 fail_loc=0x1602
-
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" $LTIME ||
- error "(3.0) MDS1 is not the expected 'scanning-phase2'"
-
- do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
+ $START_LAYOUT -r -o -c -d || error "(2) Fail to start LFSCK for layout!"
for k in $(seq $MDSCOUNT); do
# The LFSCK status query internal is 30 seconds. For the case
local repaired=$(do_facet $SINGLEMDS $LCTL get_param -n \
mdd.$(facet_svc $SINGLEMDS).lfsck_layout |
awk '/^repaired_orphan/ { print $2 }')
- [ $repaired -eq 1 ] ||
- error "(5) Expect 1 orphan has been fixed, but got: $repaired"
+ [ $repaired -eq 2 ] ||
+ error "(5) Expect 2 orphans have been fixed, but got: $repaired"
+
+ repaired=$(do_facet $SINGLEMDS $LCTL get_param -n \
+ mdd.$(facet_svc $SINGLEMDS).lfsck_layout |
+ awk '/^repaired_dangling/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(6) Expect 0 dangling has been fixed, but got: $repaired"
echo "The file size should be correct after layout LFSCK scanning"
- cur_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
- [ "$cur_size" == "$saved_size" ] ||
- error "(6) Expect file2 size $saved_size, but got $cur_size"
+ local cur_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
+ [ "$cur_size" == "$saved_size1" ] ||
+ error "(7) Expect file2 size $saved_size1, but got $cur_size"
+
+ cur_size=$(ls -il $DIR/$tdir/a1/f4 | awk '{ print $6 }')
+ [ "$cur_size" == "$saved_size2" ] ||
+ error "(8) Expect file4 size $saved_size2, but got $cur_size"
echo "The LFSCK should find back the original data."
cat $DIR/$tdir/a1/f2
$LFS path2fid $DIR/$tdir/a1/f2
$LFS getstripe $DIR/$tdir/a1/f2
+ cat $DIR/$tdir/a1/f4
+ $LFS path2fid $DIR/$tdir/a1/f4
+ $LFS getstripe $DIR/$tdir/a1/f4
}
run_test 18d "Find out orphan OST-object and repair it (4)"
test_18e() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+
echo "#####"
echo "The target MDT-object layout EA slot is occpuied by some new"
echo "created OST-object when repair dangling reference case. Such"
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 $DIR/$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 }')
+
+ echo "guard" > $DIR/$tdir/a1/f3
+ $LFS setstripe -E 1M -S 1M -o 0 -E -1 -S 1M $DIR/$tdir/a1/f4 ||
+ error "(0) Fail to create PFL $DIR/$tdir/a1/f4"
+ echo "foo" > $DIR/$tdir/a1/f4
+
+ local saved_size1=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
+ local saved_size2=$(ls -il $DIR/$tdir/a1/f4 | awk '{ print $6 }')
+
$LFS path2fid $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a1/f1
$LFS path2fid $DIR/$tdir/a1/f2
$LFS getstripe $DIR/$tdir/a1/f2
+ $LFS path2fid $DIR/$tdir/a1/f3
+ $LFS getstripe $DIR/$tdir/a1/f3
+ $LFS path2fid $DIR/$tdir/a1/f4
+ $LFS getstripe $DIR/$tdir/a1/f4
cancel_lru_locks osc
echo "Inject failure to make $DIR/$tdir/a1/f1 and $DIR/$tdir/a1/f2"
echo "to reference the same OST-object (which is f1's OST-obejct)."
echo "Then drop $DIR/$tdir/a1/f1 and its OST-object, so f2 becomes"
echo "dangling reference case, but f2's old OST-object is there."
+
+ echo "Also the failure makes $DIR/$tdir/a1/f3 and $DIR/$tdir/a1/f4"
+ echo "to reference the same OST-object (which is f3's OST-obejct)."
+ echo "Then drop $DIR/$tdir/a1/f3 and its OST-object, so f4 becomes"
+ echo "dangling reference case, but f4's old OST-object is there."
echo
#define OBD_FAIL_LFSCK_CHANGE_STRIPE 0x1618
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1618
chown 1.1 $DIR/$tdir/a1/f2
+ chown 1.1 $DIR/$tdir/a1/f4
rm -f $DIR/$tdir/a1/f1
+ rm -f $DIR/$tdir/a1/f3
sync
sleep 2
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
echo "setupall"
setupall > /dev/null
- echo "The file size should be incorrect since dangling referenced"
- local cur_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
- [ "$cur_size" != "$saved_size" ] ||
- error "(1) Expect incorrect file2 size"
-
#define OBD_FAIL_LFSCK_DELAY3 0x1602
do_facet $SINGLEMDS $LCTL set_param fail_val=10 fail_loc=0x1602
sync
sleep 2
- echo "Write new data to f2 to modify the new created OST-object."
- echo "dummy" >> $DIR/$tdir/a1/f2
+ echo "Write new data to f2/f4 to modify the new created OST-object."
+ echo "dummy" >> $DIR/$tdir/a1/f2 || error "write a1/f2 failed"
+ echo "dummy" >> $DIR/$tdir/a1/f4 || error "write a1/f4 failed"
do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
local repaired=$(do_facet $SINGLEMDS $LCTL get_param -n \
mdd.$(facet_svc $SINGLEMDS).lfsck_layout |
awk '/^repaired_orphan/ { print $2 }')
- [ $repaired -eq 1 ] ||
- error "(6) Expect 1 orphan has been fixed, but got: $repaired"
+ [ $repaired -eq 2 ] ||
+ error "(6) Expect 2 orphans have been fixed, but got: $repaired"
echo "There should be stub file under .lustre/lost+found/MDT0000/"
[ -d $MOUNT/.lustre/lost+found/MDT0000 ] ||
error "(7) $MOUNT/.lustre/lost+found/MDT0000/ should be there"
- cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-C-*)
- [ ! -z "$cname" ] ||
- error "(8) .lustre/lost+found/MDT0000/ should not be empty"
+ local count=$(ls -l $MOUNT/.lustre/lost+found/MDT0000/*-C-* | wc -l)
+ if [ $count -ne 2 ]; then
+ ls -l $MOUNT/.lustre/lost+found/MDT0000/*-C-*
+ error "(8) Expect 2 stubs under lost+found, but got $count"
+ fi
+
+ echo "The stub file should keep the original f2 or f4 data"
+ cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-C-* | head -n 1)
+ local cur_size=$(ls -il $cname | awk '{ print $6 }')
+ [ "$cur_size" != "$saved_size1" -a "$cur_size" != "$saved_size2" ] &&
+ error "(9) Got unexpected $cur_size"
+
+ cat $cname
+ $LFS path2fid $cname
+ $LFS getstripe $cname
- echo "The stub file should keep the original f2 data"
+ cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-C-* | tail -n 1)
cur_size=$(ls -il $cname | awk '{ print $6 }')
- [ "$cur_size" == "$saved_size" ] ||
- error "(9) Expect file2 size $saved_size, but got $cur_size"
+ [ "$cur_size" != "$saved_size1" -a "$cur_size" != "$saved_size2" ] &&
+ error "(10) Got unexpected $cur_size"
cat $cname
$LFS path2fid $cname
$LFS getstripe $cname
- echo "The f2 should contains new data."
+ echo "The f2/f4 should contains new data."
cat $DIR/$tdir/a1/f2
$LFS path2fid $DIR/$tdir/a1/f2
$LFS getstripe $DIR/$tdir/a1/f2
+ cat $DIR/$tdir/a1/f4
+ $LFS path2fid $DIR/$tdir/a1/f4
+ $LFS getstripe $DIR/$tdir/a1/f4
}
run_test 18e "Find out orphan OST-object and repair it (5)"
test_18f() {
- [ $OSTCOUNT -lt 2 ] &&
- skip "The test needs at least 2 OSTs" && return
+ [ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs" && return
echo "#####"
echo "The target MDT-object is lost. The LFSCK should re-create the"
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 $DIR/$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
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 $DIR/$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
}
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"
+ echo "by the stale OI mapping."
+ echo "#####"
+
+ check_mount_and_prep
+ $LFS mkdir -i 0 $DIR/$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=$OSTCOUNT
+ local fid1=$($LFS path2fid $DIR/$tdir/a1/f1)
+ echo ${fid1}
+ $LFS getstripe $DIR/$tdir/a1/f1
+ cancel_lru_locks osc
+
+ echo "Inject failure to simulate lost MDT-object but keep OI mapping"
+ #define OBD_FAIL_LFSCK_LOST_MDTOBJ2 0x162e
+ do_facet mds1 $LCTL set_param fail_loc=0x162e
+ rm -f $DIR/$tdir/a1/f1
+
+ do_facet mds1 $LCTL set_param fail_loc=0
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
+
+ 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!"
+
+ 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" $LTIME ||
+ error "(2) MDS${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 "(3) 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 $OSTCOUNT ] ||
+ error "(4) Expect $OSTCOUNT fixed, but got: $repaired"
+
+ echo "Move the files from ./lustre/lost+found/MDTxxxx to namespace"
+ mv $MOUNT/.lustre/lost+found/MDT0000/${fid1}-R-0 $DIR/$tdir/a1/f1 ||
+ error "(5) Fail to move $MOUNT/.lustre/lost+found/MDT0000/${fid1}-R-0"
+
+ $LFS path2fid $DIR/$tdir/a1/f1
+ $LFS getstripe $DIR/$tdir/a1/f1
+}
+run_test 18g "Find out orphan OST-object and repair it (7)"
+
+test_18h() {
+ echo "#####"
+ echo "The PFL extent crashed. During the first cycle LFSCK scanning,"
+ echo "the layout LFSCK will keep the bad PFL file(s) there without"
+ echo "scanning its OST-object(s). Then in the second stage scanning,"
+ echo "the OST will return related OST-object(s) to the MDT as orphan."
+ echo "And then the LFSCK on the MDT can rebuild the PFL extent with"
+ echo "the 'orphan(s)' stripe information."
+ echo "#####"
+
+ check_mount_and_prep
+
+ $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 ||
+ error "(1.1) Fail to write $DIR/$tdir/f0"
+
+ dd if=$LUSTRE/tests/test-framework.sh of=$DIR/$tdir/f0 bs=1M seek=2 ||
+ error "(1.2) Fail to write $DIR/$tdir/f0"
+
+ cp $DIR/$tdir/f0 $DIR/$tdir/guard
+
+ echo "Inject failure stub to simulate bad PFL extent range"
+ #define OBD_FAIL_LFSCK_BAD_PFL_RANGE 0x162f
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x162f
+
+ chown 1.1 $DIR/$tdir/f0
+
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1 &&
+ error "(2) Write to bad PFL file should fail"
+
+ echo "Trigger layout LFSCK to find out the bad lmm_oi and fix them"
+ $START_LAYOUT -r -o || error "(3) 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" $LTIME ||
+ error "(4.1) MDS${k} is not the expected 'completed'"
+ done
+
+ for k in $(seq $OSTCOUNT); do
+ cur_status=$(do_facet ost${k} $LCTL get_param -n \
+ obdfilter.$(facet_svc ost${k}).lfsck_layout |
+ awk '/^status/ { print $2 }')
+ [ "$cur_status" == "completed" ] ||
+ error "(4.2) OST${k} Expect 'completed', but got '$cur_status'"
+
+ done
+
+ local repaired=$($SHOW_LAYOUT |
+ awk '/^repaired_orphan/ { print $2 }')
+ [ $repaired -eq 2 ] ||
+ error "(5) Fail to repair crashed PFL range: $repaired"
+
+ echo "Data in $DIR/$tdir/f0 should not be broken"
+ diff $DIR/$tdir/f0 $DIR/$tdir/guard ||
+ error "(6) Data in $DIR/$tdir/f0 is broken"
+
+ echo "Write should succeed after LFSCK repairing the bad PFL range"
+ dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1 ||
+ error "(7) Write should succeed after LFSCK"
+}
+run_test 18h "LFSCK can repair crashed PFL extent range"
+
$LCTL set_param debug=-cache > /dev/null
test_19a() {
check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
- echo "foo" > $DIR/$tdir/a0
- echo "guard" > $DIR/$tdir/a1
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param -n \
+ obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid 0
+
+ echo "foo1" > $DIR/$tdir/a0
+ $LFS setstripe -E 512K -S 512K -o 0 -E -1 -S 1M $DIR/$tdir/a1 ||
+ error "(0) Fail to create PFL $DIR/$tdir/a1"
+ echo "foo2" > $DIR/$tdir/a1
+ echo "guard" > $DIR/$tdir/a2
cancel_lru_locks osc
echo "Inject failure, then client will offer wrong parent FID when read"
- do_facet ost1 $LCTL set_param -n \
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param -n \
obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid 1
+
#define OBD_FAIL_LFSCK_INVALID_PFID 0x1619
$LCTL set_param fail_loc=0x1619
echo "Read RPC with wrong parent FID should be denied"
- cat $DIR/$tdir/a0 && error "(3) Read should be denied!"
+ cat $DIR/$tdir/a0 && error "(3.1) Read a0 should be denied!"
+ cat $DIR/$tdir/a1 && error "(3.2) Read a1 should be denied!"
$LCTL set_param fail_loc=0
}
run_test 19a "OST-object inconsistency self detect"
echo "Inject failure stub to make the OST-object to back point to"
echo "non-exist MDT-object"
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param -n \
+ obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid 0
+
#define OBD_FAIL_LFSCK_UNMATCHED_PAIR1 0x1611
- do_facet ost1 $LCTL set_param fail_loc=0x1611
- echo "foo" > $DIR/$tdir/f0
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param fail_loc=0x1611
+ echo "foo1" > $DIR/$tdir/f0
+ $LFS setstripe -E 1M -S 1M -o 0 -E 4M -S 256K $DIR/$tdir/f1 ||
+ error "(0) Fail to create PFL $DIR/$tdir/f1"
+ echo "foo2" > $DIR/$tdir/f1
cancel_lru_locks osc
- do_facet ost1 $LCTL set_param fail_loc=0
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param fail_loc=0
+ do_facet ost1 $LCTL set_param -n \
+ obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid 0
echo "Nothing should be fixed since self detect and repair is disabled"
local repaired=$(do_facet ost1 $LCTL get_param -n \
obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid |
echo "Read RPC with right parent FID should be accepted,"
echo "and cause parent FID on OST to be fixed"
- do_facet ost1 $LCTL set_param -n \
+ do_nodes $(comma_list $(osts_nodes)) $LCTL set_param -n \
obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid 1
- cat $DIR/$tdir/f0 || error "(2) Read should not be denied!"
+
+ cat $DIR/$tdir/f0 || error "(2.1) Read f0 should not be denied!"
+ cat $DIR/$tdir/f1 || error "(2.2) Read f1 should not be denied!"
repaired=$(do_facet ost1 $LCTL get_param -n \
obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid |
awk '/^repaired/ { print $2 }')
- [ $repaired -eq 1 ] ||
+ [ $repaired -eq 2 ] ||
error "(3) Expected 1 repaired, but got $repaired"
}
run_test 19b "OST-object inconsistency self repair"
-test_20() {
- [ $OSTCOUNT -lt 2 ] &&
- skip "The test needs at least 2 OSTs" && return
+PATTERN_WITH_HOLE="40000001"
+PATTERN_WITHOUT_HOLE="raid0"
+
+test_20a() {
+ [ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs" && return
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
echo "#####"
echo "The target MDT-object and some of its OST-object are lost."
sleep 2
do_facet mds1 $LCTL set_param fail_loc=0 fail_val=0
- echo "Inject failure to slow down the LFSCK on OST0"
- #define OBD_FAIL_LFSCK_DELAY5 0x161b
- do_facet ost1 $LCTL set_param fail_loc=0x161b
-
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!"
- sleep 3
- do_facet ost1 $LCTL set_param fail_loc=0
-
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
$LFS getstripe -v $name || error "(5.1) cannot getstripe on $name"
- local pattern=0x$($LFS getstripe -L $name)
- [[ $((pattern & LOV_PATTERN_F_HOLE)) -eq 0 ]] ||
+ local pattern=$($LFS getstripe -L $name)
+ [[ "$pattern" = "$PATTERN_WITHOUT_HOLE" ]] ||
error "(5.2) NOT expect pattern flag hole, but got $pattern"
local stripes=$($LFS getstripe -c $name)
$LFS getstripe -v $name || error "(6.1) cannot getstripe on $name"
- pattern=0x$($LFS getstripe -L $name)
- [[ $((pattern & LOV_PATTERN_F_HOLE)) -ne 0 ]] ||
+ pattern=$($LFS getstripe -L $name)
+ [[ "$pattern" = "$PATTERN_WITH_HOLE" ]] ||
error "(6.2) expect pattern flag hole, but got $pattern"
stripes=$($LFS getstripe -c $name)
$LFS getstripe -v $name || error "(7.1) cannot getstripe on $name"
- pattern=0x$($LFS getstripe -L $name)
+ pattern=$($LFS getstripe -L $name)
+ [[ "$pattern" = "$PATTERN_WITH_HOLE" ]] ||
+ error "(7.2) expect pattern flag hole, but got $pattern"
+
stripes=$($LFS getstripe -c $name)
size=$(stat $name | awk '/Size:/ { print $2 }')
if [ $OSTCOUNT -gt 2 ]; then
- [[ $((pattern & LOV_PATTERN_F_HOLE)) -ne 0 ]] ||
- error "(7.2.1) expect pattern flag hole, but got $pattern"
-
[ $stripes -eq 3 ] ||
error "(7.3.1) expect the stripe count is 3, but got $stripes"
touch $name || error "(7.10.1) cannot touch $name"
else
- [[ $((pattern & LOV_PATTERN_F_HOLE)) -eq 0 ]] ||
- error "(7.2.2) NOT expect pattern flag hole, but got $pattern"
-
- [ $stripes -eq 1 ] ||
- error "(7.3.2) expect the stripe count is 1, but got $stripes"
+ [ $stripes -eq 2 ] ||
+ error "(7.3.2) expect the stripe count is 2, but got $stripes"
# stripe1 is dummy
[ $size -eq $((4096 * (256 + 0))) ] ||
error "(7.4.2) expect the size $((4096 * 256)), but got $size"
- cat $name > /dev/null || error "(7.5.2) cannot read $name"
+ cat $name > /dev/null &&
+ error "(7.5.2) normal read $name should fail"
+
+ failures=$(dd if=$name of=$DIR/$tdir/dump conv=sync,noerror \
+ bs=4096 2>&1 | grep "Input/output error" | wc -l)
+ [ $failures -eq 256 ] ||
+ error "(7.6.2) expect 256 IO failures, but get $failures"
+
+ bcount=$((256 * 2))
+ size=$(stat $DIR/$tdir/dump | awk '/Size:/ { print $2 }')
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(7.7.2) expect the size $((4096 * $bcount)), got $size"
- echo "dummy" >> $name || error "(7.8.2) cannot write $name"
+ dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 \
+ seek=256 && error "(7.8.2) write to the LOV EA hole should fail"
chown $RUNAS_ID:$RUNAS_GID $name ||
error "(7.9.2) cannot chown on $name"
$LFS getstripe -v $name || error "(8.1) cannot getstripe on $name"
- pattern=0x$($LFS getstripe -L $name)
- [[ $((pattern & LOV_PATTERN_F_HOLE)) -eq 0 ]] ||
- error "(8.2) NOT expect pattern flag hole, but got $pattern"
+ pattern=$($LFS getstripe -L $name)
+ [[ "$pattern" = "$PATTERN_WITH_HOLE" ]] ||
+ error "(8.2) expect pattern flag hole, but got $pattern"
stripes=$($LFS getstripe -c $name)
- # LFSCK does not know the old f3 had 3 stripes.
- # It only tries to find as much as possible.
- # The stripe count depends on the last stripe's offset.
- [ $stripes -eq 2 ] ||
- error "(8.3) expect the stripe count is 2, but got $stripes"
+ [ $stripes -eq 3 ] ||
+ error "(8.3) expect the stripe count is 3, but got $stripes"
size=$(stat $name | awk '/Size:/ { print $2 }')
# stripe2 is lost
[ $size -eq $((4096 * (256 + 256 + 0))) ] ||
error "(8.4) expect the size $((4096 * 512)), but got $size"
- cat $name > /dev/null || error "(8.5) cannot read $name"
-
- echo "dummy" >> $name || error "(8.6) cannot write $name"
-
- chown $RUNAS_ID:$RUNAS_GID $name ||
- error "(8.7) cannot chown on $name"
-
- touch $name || error "(8.8) cannot touch $name"
-
- rm -f $name || error "(8.9) cannot unlink $name"
-}
-run_test 20 "Handle the orphan with dummy LOV EA slot properly"
+ cat $name > /dev/null &&
+ error "(8.5) normal read $name should fail"
-test_21() {
- [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.59) ]] &&
- skip "ignore the test if MDS is older than 2.5.59" && return
+ failures=$(dd if=$name of=$DIR/$tdir/dump conv=sync,noerror \
+ bs=4096 2>&1 | grep "Input/output error" | wc -l)
+ # stripe2 is dummy
+ [ $failures -eq 256 ] ||
+ error "(8.6) expect 256 IO failures, but get $failures"
- check_mount_and_prep
- createmany -o $DIR/$tdir/f 100 || error "(0) Fail to create 100 files"
+ bcount=$((256 * 3))
+ size=$(stat $DIR/$tdir/dump | awk '/Size:/ { print $2 }')
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(8.7) expect the size $((4096 * $bcount)), but got $size"
- echo "Start all LFSCK components by default (-s 1)"
- do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -s 1 -r ||
- error "Fail to start LFSCK"
+ dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 \
+ seek=512 && error "(8.8) write to the LOV EA hole should fail"
- echo "namespace LFSCK should be in 'scanning-phase1' status"
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "scanning-phase1" ] ||
- error "Expect namespace 'scanning-phase1', but got '$STATUS'"
+ chown $RUNAS_ID:$RUNAS_GID $name ||
+ error "(8.9) cannot chown on $name"
- echo "layout LFSCK should be in 'scanning-phase1' status"
- STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
- [ "$STATUS" == "scanning-phase1" ] ||
- error "Expect layout 'scanning-phase1', but got '$STATUS'"
+ touch $name || error "(8.10) cannot touch $name"
- echo "Stop all LFSCK components by default"
- do_facet mds1 $LCTL lfsck_stop -M ${FSNAME}-MDT0000 ||
- error "Fail to stop LFSCK"
+ rm -f $name || error "(8.11) cannot unlink $name"
}
-run_test 21 "run all LFSCK components by default"
+run_test 20a "Handle the orphan with dummy LOV EA slot properly"
-test_22a() {
- [ $MDSCOUNT -lt 2 ] &&
- skip "We need at least 2 MDSes for this test" && return
+test_20b() {
+ [ $OSTCOUNT -lt 2 ] && skip "needs >= 2 OSTs" && return
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
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_B does not exist. Then the namespace"
- echo "LFSCK will repair the child directory's ".." name entry."
+ echo "The target MDT-object and some of its OST-object are lost."
+ echo "The LFSCK should find out the left OST-objects and re-create"
+ echo "the MDT-object under the direcotry .lustre/lost+found/MDTxxxx/"
+ echo "with the partial OST-objects (LOV EA hole)."
+
+ echo "New client can access the file with LOV EA hole via normal"
+ echo "system tools or commands without crash the system - PFL case."
echo "#####"
check_mount_and_prep
- $LFS mkdir -i 1 $DIR/$tdir/guard || error "(1) Fail to mkdir on MDT1"
- $LFS mkdir -i 1 $DIR/$tdir/foo || error "(2) Fail to mkdir on MDT1"
+ $LFS setstripe -E 2M -S 1M -c 2 -E -1 -S 1M -c 2 $DIR/$tdir/f0 ||
+ error "(0) Fail to create PFL file $DIR/$tdir/f0"
+ $LFS setstripe -E 2M -S 1M -c 2 -E -1 -S 1M -c 2 $DIR/$tdir/f1 ||
+ error "(1) Fail to create PFL file $DIR/$tdir/f1"
+ $LFS setstripe -E 2M -S 1M -c 2 -E -1 -S 1M -c 2 $DIR/$tdir/f2 ||
+ error "(2) Fail to create PFL file $DIR/$tdir/f2"
- echo "Inject failure stub on MDT0 to simulate bad dotdot name entry"
- echo "The dummy's dotdot name entry references the guard."
- #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 $LCTL set_param fail_loc=0
+ local bcount=$((256 * 3 + 1))
- rmdir $DIR/$tdir/guard || error "(4) Fail to rmdir $DIR/$tdir/guard"
+ dd if=/dev/zero of=$DIR/$tdir/f0 bs=4096 count=$bcount
+ dd if=/dev/zero of=$DIR/$tdir/f1 bs=4096 count=$bcount
+ dd if=/dev/zero of=$DIR/$tdir/f2 bs=4096 count=$bcount
- echo "Trigger namespace LFSCK to repair unmatched pairs"
- $START_NAMESPACE -A -r ||
- error "(5) Fail to start LFSCK for namespace"
+ local fid0=$($LFS path2fid $DIR/$tdir/f0)
+ local fid1=$($LFS path2fid $DIR/$tdir/f1)
+ local fid2=$($LFS path2fid $DIR/$tdir/f2)
- 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"
- }
+ echo ${fid0}
+ $LFS getstripe $DIR/$tdir/f0
+ echo ${fid1}
+ $LFS getstripe $DIR/$tdir/f1
+ echo ${fid2}
+ $LFS getstripe $DIR/$tdir/f2
- local repaired=$($SHOW_NAMESPACE |
- awk '/^unmatched_pairs_repaired/ { print $2 }')
- [ $repaired -eq 1 ] ||
- error "(7) Fail to repair unmatched pairs: $repaired"
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
- echo "'ls' should success after namespace LFSCK repairing"
- ls -ail $DIR/$tdir/foo/dummy > /dev/null ||
- error "(8) ls should success."
-}
-run_test 22a "LFSCK can repair unmatched pairs (1)"
+ echo "Inject failure..."
+ echo "To simulate f0 lost MDT-object"
+ #define OBD_FAIL_LFSCK_LOST_MDTOBJ 0x1616
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1616
+ rm -f $DIR/$tdir/f0
+
+ echo "To simulate the case of f1 lost MDT-object and "
+ echo "the first OST-object in each PFL component"
+ #define OBD_FAIL_LFSCK_LOST_SPEOBJ 0x161a
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x161a
+ rm -f $DIR/$tdir/f1
+
+ echo "To simulate the case of f2 lost MDT-object and "
+ echo "the second OST-object in each PFL component"
+ do_facet $SINGLEMDS $LCTL set_param fail_val=1
+ rm -f $DIR/$tdir/f2
+
+ sync
+ sleep 2
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
+
+ echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+ $START_LAYOUT -r -o || error "(3) 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${k}).lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ error "(4) MDS${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 "(5) 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 8 ] ||
+ error "(6) Expect 8 fixed on mds1, but got: $repaired"
+
+ #
+ # ${fid0}-R-0 is the old f0
+ #
+ local name="$MOUNT/.lustre/lost+found/MDT0000/${fid0}-R-0"
+ echo "Check $name, which is the old f0"
+
+ $LFS getstripe -v $name || error "(7.1) cannot getstripe on $name"
+
+ local pattern=$($LFS getstripe -L -I1 $name)
+ [[ "$pattern" = "$PATTERN_WITHOUT_HOLE" ]] ||
+ error "(7.2.1) NOT expect pattern flag hole, but got $pattern"
+
+ pattern=$($LFS getstripe -L -I2 $name)
+ [[ "$pattern" = "$PATTERN_WITHOUT_HOLE" ]] ||
+ error "(7.2.2) NOT expect pattern flag hole, but got $pattern"
+
+ local stripes=$($LFS getstripe -c -I1 $name)
+ [ $stripes -eq 2 ] ||
+ error "(7.3.1) expect 2 stripes, but got $stripes"
+
+ stripes=$($LFS getstripe -c -I2 $name)
+ [ $stripes -eq 2 ] ||
+ error "(7.3.2) expect 2 stripes, but got $stripes"
+
+ local e_start=$($LFS getstripe -I1 $name |
+ awk '/lcme_extent.e_start:/ { print $2 }')
+ [ $e_start -eq 0 ] ||
+ error "(7.4.1) expect the COMP1 start at 0, got $e_start"
+
+ local e_end=$($LFS getstripe -I1 $name |
+ awk '/lcme_extent.e_end:/ { print $2 }')
+ [ $e_end -eq 2097152 ] ||
+ error "(7.4.2) expect the COMP1 end at 2097152, got $e_end"
+
+ e_start=$($LFS getstripe -I2 $name |
+ awk '/lcme_extent.e_start:/ { print $2 }')
+ [ $e_start -eq 2097152 ] ||
+ error "(7.5.1) expect the COMP2 start at 2097152, got $e_start"
+
+ e_end=$($LFS getstripe -I2 $name |
+ awk '/lcme_extent.e_end:/ { print $2 }')
+ [ "$e_end" = "EOF" ] ||
+ error "(7.5.2) expect the COMP2 end at (EOF), got $e_end"
+
+ local size=$(stat $name | awk '/Size:/ { print $2 }')
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(7.6) expect the size $((4096 * $bcount)), but got $size"
+
+ cat $name > /dev/null || error "(7.7) cannot read $name"
+
+ echo "dummy" >> $name || error "(7.8) cannot write $name"
+
+ chown $RUNAS_ID:$RUNAS_GID $name || error "(7.9) cannot chown on $name"
+
+ touch $name || error "(7.10) cannot touch $name"
+
+ rm -f $name || error "(7.11) cannot unlink $name"
+
+ #
+ # ${fid1}-R-0 contains the old f1's second stripe in each COMP
+ #
+ name="$MOUNT/.lustre/lost+found/MDT0000/${fid1}-R-0"
+ echo "Check $name, it contains f1's second OST-object in each COMP"
+
+ $LFS getstripe -v $name || error "(8.1) cannot getstripe on $name"
+
+ pattern=$($LFS getstripe -L -I1 $name)
+ [[ "$pattern" = "$PATTERN_WITH_HOLE" ]] ||
+ error "(8.2.1) expect pattern flag hole, but got $pattern"
+
+ pattern=$($LFS getstripe -L -I2 $name)
+ [[ "$pattern" = "$PATTERN_WITH_HOLE" ]] ||
+ error "(8.2.2) expect pattern flag hole, but got $pattern"
+
+ stripes=$($LFS getstripe -c -I1 $name)
+ [ $stripes -eq 2 ] ||
+ error "(8.3.2) expect 2 stripes, but got $stripes"
+
+ stripes=$($LFS getstripe -c -I2 $name)
+ [ $stripes -eq 2 ] ||
+ error "(8.3.2) expect 2 stripes, but got $stripes"
+
+ e_start=$($LFS getstripe -I1 $name |
+ awk '/lcme_extent.e_start:/ { print $2 }')
+ [ $e_start -eq 0 ] ||
+ error "(8.4.1) expect the COMP1 start at 0, got $e_start"
+
+ e_end=$($LFS getstripe -I1 $name |
+ awk '/lcme_extent.e_end:/ { print $2 }')
+ [ $e_end -eq 2097152 ] ||
+ error "(8.4.2) expect the COMP1 end at 2097152, got $e_end"
+
+ e_start=$($LFS getstripe -I2 $name |
+ awk '/lcme_extent.e_start:/ { print $2 }')
+ [ $e_start -eq 2097152 ] ||
+ error "(8.5.1) expect the COMP2 start at 2097152, got $e_start"
+
+ e_end=$($LFS getstripe -I2 $name |
+ awk '/lcme_extent.e_end:/ { print $2 }')
+ [ "$e_end" = "EOF" ] ||
+ error "(8.5.2) expect the COMP2 end at (EOF), got $e_end"
+
+ size=$(stat $name | awk '/Size:/ { print $2 }')
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(8.6) expect the size $((4096 * $bcount)), but got $size"
+
+ cat $name > /dev/null && error "(8.7) normal read $name should fail"
+
+ local failures=$(dd if=$name of=$DIR/$tdir/dump conv=sync,noerror \
+ bs=4096 2>&1 | grep "Input/output error" | wc -l)
+
+ # The first stripe in each COMP was lost
+ [ $failures -eq 512 ] ||
+ error "(8.8) expect 512 IO failures, but get $failures"
+
+ size=$(stat $DIR/$tdir/dump | awk '/Size:/ { print $2 }')
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(8.9) expect the size $((4096 * $bcount)), but got $size"
+
+ dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 &&
+ error "(8.10) write to the LOV EA hole should fail"
+
+ dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 seek=300 ||
+ error "(8.11) write to normal stripe should NOT fail"
+
+ echo "foo" >> $name && error "(8.12) append write $name should fail"
+
+ chown $RUNAS_ID:$RUNAS_GID $name || error "(8.13) cannot chown on $name"
+
+ touch $name || error "(8.14) cannot touch $name"
+
+ rm -f $name || error "(8.15) cannot unlink $name"
+
+ #
+ # ${fid2}-R-0 contains the old f2's first stripe in each COMP
+ #
+ name="$MOUNT/.lustre/lost+found/MDT0000/${fid2}-R-0"
+ echo "Check $name, it contains f2's first stripe in each COMP"
+
+ $LFS getstripe -v $name || error "(9.1) cannot getstripe on $name"
+
+ pattern=$($LFS getstripe -L -I1 $name)
+ [[ "$pattern" = "$PATTERN_WITH_HOLE" ]] ||
+ error "(9.2.1) expect pattern flag hole, but got $pattern"
+
+ pattern=$($LFS getstripe -L -I2 $name)
+ [[ "$pattern" = "$PATTERN_WITH_HOLE" ]] ||
+ error "(9.2.2) expect pattern flag hole, but got $pattern"
+
+ stripes=$($LFS getstripe -c -I1 $name)
+ [ $stripes -eq 2 ] ||
+ error "(9.3.2) expect 2 stripes, but got $stripes"
+
+ stripes=$($LFS getstripe -c -I2 $name)
+ [ $stripes -eq 2 ] ||
+ error "(9.3.2) expect 2 stripes, but got $stripes"
+
+ e_start=$($LFS getstripe -I1 $name |
+ awk '/lcme_extent.e_start:/ { print $2 }')
+ [ $e_start -eq 0 ] ||
+ error "(9.4.1) expect the COMP1 start at 0, got $e_start"
+
+ e_end=$($LFS getstripe -I1 $name |
+ awk '/lcme_extent.e_end:/ { print $2 }')
+ [ $e_end -eq 2097152 ] ||
+ error "(9.4.2) expect the COMP1 end at 2097152, got $e_end"
+
+ e_start=$($LFS getstripe -I2 $name |
+ awk '/lcme_extent.e_start:/ { print $2 }')
+ [ $e_start -eq 2097152 ] ||
+ error "(9.5.1) expect the COMP2 start at 2097152, got $e_start"
+
+ e_end=$($LFS getstripe -I2 $name |
+ awk '/lcme_extent.e_end:/ { print $2 }')
+ [ "$e_end" = "EOF" ] ||
+ error "(9.5.2) expect the COMP2 end at (EOF), got $e_end"
+
+ size=$(stat $name | awk '/Size:/ { print $2 }')
+ # The second stripe in COMP was lost, so we do not know there
+ # have ever been some data before. 'stat' will regard it as
+ # no data on the lost stripe.
+ bcount=$((256 * 3))
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(9.6) expect size $((4096 * $bcount)), but got $size"
+
+ cat $name > /dev/null &&
+ error "(9.7) normal read $name should fail"
+
+ failures=$(dd if=$name of=$DIR/$tdir/dump conv=sync,noerror \
+ bs=4096 2>&1 | grep "Input/output error" | wc -l)
+ [ $failures -eq 512 ] ||
+ error "(9.8) expect 256 IO failures, but get $failures"
+
+ size=$(stat $DIR/$tdir/dump | awk '/Size:/ { print $2 }')
+ # The second stripe in COMP was lost, so we do not know there
+ # have ever been some data before. Since 'dd' skip failure,
+ # it will regard the lost stripe contains data.
+ bcount=$((256 * 4))
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(9.9) expect the size $((4096 * $bcount)), but got $size"
+
+ dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 \
+ seek=300 && error "(9.10) write to the LOV EA hole should fail"
+
+ dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 ||
+ error "(9.11) write to normal stripe should NOT fail"
+
+ echo "foo" >> $name &&
+ error "(9.12) append write $name should fail"
+
+ chown $RUNAS_ID:$RUNAS_GID $name ||
+ error "(9.13) cannot chown on $name"
+
+ touch $name || error "(9.14) cannot touch $name"
+
+ rm -f $name || error "(7.15) cannot unlink $name"
+}
+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
+
+ check_mount_and_prep
+ createmany -o $DIR/$tdir/f 100 || error "(0) Fail to create 100 files"
+
+ echo "Start all LFSCK components by default (-s 1)"
+ do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -s 1 -r ||
+ error "Fail to start LFSCK"
+
+ echo "namespace LFSCK should be in 'scanning-phase1' status"
+ local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
+ [ "$STATUS" == "scanning-phase1" ] ||
+ error "Expect namespace 'scanning-phase1', but got '$STATUS'"
+
+ echo "layout LFSCK should be in 'scanning-phase1' status"
+ STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
+ [ "$STATUS" == "scanning-phase1" ] ||
+ error "Expect layout 'scanning-phase1', but got '$STATUS'"
+
+ echo "Stop all LFSCK components by default"
+ do_facet mds1 $LCTL lfsck_stop -M ${FSNAME}-MDT0000 ||
+ error "Fail to stop LFSCK"
+}
+run_test 21 "run all LFSCK components by default"
+
+test_22a() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+
+ 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_B does not exist. Then the namespace"
+ echo "LFSCK will repair the child directory's ".." name entry."
+ echo "#####"
+
+ check_mount_and_prep
+
+ $LFS mkdir -i 1 $DIR/$tdir/guard || error "(1) Fail to mkdir on MDT1"
+ $LFS mkdir -i 1 $DIR/$tdir/foo || error "(2) Fail to mkdir on MDT1"
+
+ echo "Inject failure stub on MDT0 to simulate bad dotdot name entry"
+ echo "The dummy's dotdot name entry references the guard."
+ #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
+
+ rmdir $DIR/$tdir/guard || error "(4) Fail to rmdir $DIR/$tdir/guard"
+
+ echo "Trigger namespace LFSCK to repair unmatched pairs"
+ $START_NAMESPACE -A -r ||
+ error "(5) Fail to start LFSCK for namespace"
+
+ wait_all_targets_blocked namespace completed 6
+
+ local repaired=$($SHOW_NAMESPACE |
+ awk '/^unmatched_pairs_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(7) Fail to repair unmatched pairs: $repaired"
+
+ echo "'ls' should success after namespace LFSCK repairing"
+ ls -ail $DIR/$tdir/foo/dummy > /dev/null ||
+ error "(8) ls should success."
+}
+run_test 22a "LFSCK can repair unmatched pairs (1)"
test_22b() {
- [ $MDSCOUNT -lt 2 ] &&
- skip "We need at least 2 MDSes for this test" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "The parent_A references the child directory via the name entry_B,"
$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 }')
run_test 22b "LFSCK can repair unmatched pairs (2)"
test_23a() {
- [ $MDSCOUNT -lt 2 ] &&
- skip "We need at least 2 MDSes for this test" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "The name entry is there, but the MDT-object for such name "
$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
$LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0 on MDT0"
+ $LFS path2fid $DIR/$tdir/d0
+
+ 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
+
echo "dead" > $DIR/$tdir/d0/f1 || error "(3) Fail to touch on MDT0"
+ $LFS path2fid $DIR/$tdir/d0/f1
+
+ 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
+ # 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
+
+ local OID=$($LFS path2fid $DIR/$tdir/d0/f1 | awk -F':' '{print $2}')
+ OID=$(printf %d $OID)
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
+
+ # If there is creation after the dangling injection, it may re-use
+ # the just released local object (inode) that is referenced by the
+ # dangling name entry. It will fail the dangling injection.
+ # So before deleting the target object for the dangling name entry,
+ # remove some other objects to avoid the target object being reused
+ # by some potential creations. LU-7429
+ unlinkmany $DIR/$tdir/d0/t 10 || error "(5.0) Fail to unlinkmany"
rm -f $DIR/$tdir/d0/f1 || error "(5) Fail to unlink $DIR/$tdir/d0/f1"
}
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() {
echo "#####"
echo "The objectA has multiple hard links, one of them corresponding"
echo "LFSCK cannot replace it."
echo "#####"
+ start_full_debug_logging
+
check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0 on MDT0"
+ 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"
+ 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"
+ f1_fid="$($LFS path2fid $DIR/$tdir/d0/f1)"
+ echo "f1_fid=$f1_fid"
+
+ if [ "${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"
+ f0_fid="$($LFS path2fid $DIR/$tdir/d0/f0)"
+ echo "f0_fid=$f0_fid (replaced)"
+ fi
+
+ local oid=$(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_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
+
+ # If there is creation after the dangling injection, it may re-use
+ # the just released local object (inode) that is referenced by the
+ # dangling name entry. It will fail the dangling injection.
+ # So before deleting the target object for the dangling name entry,
+ # remove some other objects to avoid the target object being reused
+ # by some potential creations. LU-7429
+ unlinkmany $DIR/$tdir/d0/t 10 || error "(5.0) Fail to unlinkmany"
rm -f $DIR/$tdir/d0/f1 || error "(5) Fail to unlink $DIR/$tdir/d0/f1"
$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" 32 || {
- stat $DIR/$tdir/guard
+ 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 || {
- $SHOW_NAMESPACE
- error "(10) unexpected status"
- }
+ cleanup_23c
local repaired=$($SHOW_NAMESPACE |
awk '/^dangling_repaired/ { print $2 }')
run_test 23c "LFSCK can repair dangling name entry (3)"
test_24() {
- [ $MDSCOUNT -lt 2 ] &&
- skip "We need at least 2 MDSes for this test" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
echo "#####"
echo "Two MDT-objects back reference the same name entry via their"
$START_NAMESPACE -A -r ||
error "(7) 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 || {
- $SHOW_NAMESPACE
- error "(8) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 8
local repaired=$($SHOW_NAMESPACE |
awk '/^multiple_referenced_repaired/ { print $2 }')
test_25() {
[ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
- skip "Only support to inject failure on ldiskfs" && return
+ skip "ldiskfs only test" && return
echo "#####"
echo "The file type in the name entry does not match the file type"
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 && "(5) 'ls' should fail"
+ 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_update_facet $SINGLEMDS "$LCTL get_param -n \
- mdd.${MDT_DEV}.lfsck_namespace |
- awk '/^status/ { print \\\$2 }'" "completed" 32 || {
- $SHOW_NAMESPACE
- error "(7) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 7
local repaired=$($SHOW_NAMESPACE |
awk '/^lost_dirent_repaired/ { print $2 }')
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
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "The remote name entry back referenced by the MDT-object is lost."
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 && "(4) 'ls' should fail"
+ 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_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 '/^lost_dirent_repaired/ { print $2 }')
run_test 26b "LFSCK can add the missing remote name entry back to the namespace"
test_27a() {
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
+
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."
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 && "(6) 'ls' should fail"
+ 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_update_facet $SINGLEMDS "$LCTL get_param -n \
- mdd.${MDT_DEV}.lfsck_namespace |
- awk '/^status/ { print \\\$2 }'" "completed" 32 || {
- $SHOW_NAMESPACE
- error "(7) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 7
local repaired=$($SHOW_NAMESPACE |
awk '/^lost_dirent_repaired/ { print $2 }')
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
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
echo "#####"
echo "The remote parent referenced by the MDT-object linkEA is lost."
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 && "(5) 'ls' should fail"
+ 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_update_facet $SINGLEMDS "$LCTL get_param -n \
- mdd.${MDT_DEV}.lfsck_namespace |
- awk '/^status/ { print \\\$2 }'" "completed" 32 || {
- $SHOW_NAMESPACE
- error "(7) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 7
local repaired=$($SHOW_NAMESPACE |
awk '/^lost_dirent_repaired/ { print $2 }')
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
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "The target name entry is lost. The LFSCK should insert the"
$START_NAMESPACE -r -A ||
error "(8) Fail to start LFSCK for namespace"
- 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${k}).lfsck_namespace |
- awk '/^status/ { print \\\$2 }'" "completed" 32 ||
- error "(9) MDS${k} is not the expected 'completed'"
- done
+ wait_all_targets_blocked namespace completed 9
local repaired=$(do_facet mds1 $LCTL get_param -n \
mdd.$(facet_svc mds1).lfsck_namespace |
$START_NAMESPACE -r -A ||
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 '/^nlinks_repaired/ { print $2 }')
count=$(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)"
+# Disable 29a, we only allow nlink to be updated if the known linkEA
+# entries is larger than nlink count.
+#
+#run_test 29a "LFSCK can repair bad nlink count (1)"
test_29b() {
echo "#####"
$START_NAMESPACE -r -A ||
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 '/^nlinks_repaired/ { print $2 }')
}
run_test 29b "LFSCK can repair bad nlink count (2)"
-test_29c() {
+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 "The namespace LFSCK will create many hard links to the target"
+ echo "file as to exceed the linkEA size limitation. Under such case"
+ echo "the linkEA will be marked as overflow that will prevent the"
+ echo "target file to be migrated. Then remove some hard links to"
+ echo "make the left hard links to be held within the linkEA size"
+ echo "limitation. But before the namespace LFSCK adding all the"
+ echo "missed linkEA entries back, the overflow mark (timestamp)"
+ echo "will not be cleared."
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"
+ mkdir -p $DIR/$tdir/guard || error "(0.1) Fail to mkdir"
+ $LFS mkdir -i $((MDSCOUNT - 1)) $DIR/$tdir/foo ||
+ error "(0.2) Fail to mkdir"
+ touch $DIR/$tdir/guard/f0 || error "(1) Fail to create"
+ local oldfid=$($LFS path2fid $DIR/$tdir/guard/f0)
+
+ # define MAX_LINKEA_SIZE 4096
+ # sizeof(link_ea_header) = 24
+ # sizeof(link_ea_entry) = 18
+ # nlink_min=$(((MAX_LINKEA_SIZE - sizeof(link_ea_header)) /
+ # (sizeof(link_ea_entry) + name_length))
+ # If the average name length is 12 bytes, then 150 hard links
+ # is totally enough to overflow the linkEA
+ echo "Create 150 hard links should succeed although the linkEA overflow"
+ createmany -l $DIR/$tdir/guard/f0 $DIR/$tdir/foo/ttttttttttt 150 ||
+ error "(2) Fail to hard link"
- echo "Inject failure stub on MDT0 to simulate the case that"
- echo "foo's hard links exceed the object's linkEA limitation."
+ cancel_lru_locks mdc
+ if [ $MDSCOUNT -ge 2 ]; then
+ $LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null &&
+ error "(3.1) Migrate should fail"
- #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"
+ echo "The object with linkEA overflow should NOT be migrated"
+ local newfid=$($LFS path2fid $DIR/$tdir/guard/f0)
+ [ "$newfid" == "$oldfid" ] ||
+ error "(3.2) Migrate should fail: $newfid != $oldfid"
+ fi
- cancel_lru_locks mdc
+ # Remove 100 hard links, then the linkEA should have space
+ # to hold the missed linkEA entries.
+ echo "Remove 100 hard links to save space for the missed linkEA entries"
+ unlinkmany $DIR/$tdir/foo/ttttttttttt 100 || error "(4) Fail to unlink"
- local count1=$(stat --format=%h $DIR/$tdir/d0/foo)
- [ $count1 -eq 3 ] || error "(5) Stat failure: $count1"
+ if [ $MDSCOUNT -ge 2 ]; then
+ $LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null &&
+ error "(5.1) Migrate should fail"
- local foofid=$($LFS path2fid $DIR/$tdir/d0/foo)
- $LFS fid2path $DIR $foofid
- local count2=$($LFS fid2path $DIR $foofid | wc -l)
- [ $count2 -eq 2 ] || "(6) Fail to inject error: $count2"
+ # The overflow timestamp is still there, so migration will fail.
+ local newfid=$($LFS path2fid $DIR/$tdir/guard/f0)
+ [ "$newfid" == "$oldfid" ] ||
+ error "(5.2) Migrate should fail: $newfid != $oldfid"
+ fi
- echo "Trigger namespace LFSCK to repair the nlink count"
+ # sleep 3 seconds to guarantee that the overflow is recognized
+ sleep 3
+
+ echo "Trigger namespace LFSCK to clear the overflow timestamp"
$START_NAMESPACE -r -A ||
- error "(7) Fail to start LFSCK for namespace"
+ error "(6) 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 "(8) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 7
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
local repaired=$($SHOW_NAMESPACE |
- awk '/^nlinks_repaired/ { print $2 }')
+ awk '/^linkea_overflow_cleared/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(8) Fail to clear linkea overflow: $repaired"
+
+ repaired=$($SHOW_NAMESPACE |
+ awk '/^nlinks_repaired/ { print $2 }')
[ $repaired -eq 0 ] ||
- error "(9) Repair nlink count unexpcetedly: $repaired"
+ error "(9) Unexpected nlink repaired: $repaired"
- cancel_lru_locks mdc
+ if [ $MDSCOUNT -ge 2 ]; then
+ $LFS migrate -m 1 $DIR/$tdir/guard 2>/dev/null ||
+ error "(10.1) Migrate failure"
- count1=$(stat --format=%h $DIR/$tdir/d0/foo)
- [ $count1 -eq 3 ] || error "(10) Stat failure: $count1"
+ # Migration should succeed after clear the overflow timestamp.
+ local newfid=$($LFS path2fid $DIR/$tdir/guard/f0)
+ [ "$newfid" != "$oldfid" ] ||
+ error "(10.2) Migrate should succeed"
- count2=$($LFS fid2path $DIR $foofid | wc -l)
- [ $count2 -eq 2 ] ||
- error "(11) Repaired something unexpectedly: $count2"
+ ls -l $DIR/$tdir/foo > /dev/null ||
+ error "(11) 'ls' failed after migration"
+ fi
+
+ rm -f $DIR/$tdir/guard/f0 || error "(12) Fail to unlink f0"
+ rm -rf $DIR/$tdir/foo || error "(13) Fail to rmdir foo"
}
-run_test 29c "Not verify nlink attr if hark links exceed linkEA limitation"
+run_test 29c "verify linkEA size limitation"
test_30() {
[ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
- skip "Only support backend /lost+found for ldiskfs" && return
+ skip "ldiskfs only test" && return
+ [ -n "$FILESET" ] && skip "Not functional for FILESET set"
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"
$START_NAMESPACE -r -A ||
error "(14) 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 || {
- $SHOW_NAMESPACE
- error "(15) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 15
local repaired=$($SHOW_NAMESPACE |
awk '/^local_lost_found_moved/ { print $2 }')
mount_client $MOUNT || error "(17) Fail to start client!"
- stat $DIR/$tdir/foo/f0 || "(18) f0 is not recovered"
+ stat $DIR/$tdir/foo/f0 || error "(18) f0 is not recovered"
ls -ail $MOUNT/.lustre/lost+found/
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 "The test needs at least 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "For the name entry under a striped directory, if the name"
$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 '/^name_hash_repaired/ { print $2 }')
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
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "For the name entry under a striped directory, if the name"
$START_NAMESPACE -r -A ||
error "(3) Fail to start LFSCK for namespace"
- wait_update_facet mds2 "$LCTL get_param -n \
- mdd.$(facet_svc mds2).lfsck_namespace |
- awk '/^status/ { print \\\$2 }'" "completed" 32 ||
- error "(4) unexpected status"
+ wait_all_targets_blocked namespace completed 4
local repaired=$(do_facet mds2 $LCTL get_param -n \
mdd.$(facet_svc mds2).lfsck_namespace |
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
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "For some reason, the master MDT-object of the striped directory"
$START_NAMESPACE -r -A ||
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 || {
- $SHOW_NAMESPACE
- error "(3) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 3
local repaired=$($SHOW_NAMESPACE |
awk '/^striped_dirs_repaired/ { print $2 }')
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
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "For some reason, the master MDT-object of the striped directory"
$START_NAMESPACE -r -A ||
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 '/^striped_dirs_repaired/ { print $2 }')
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
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "For some reason, the slave MDT-object of the striped directory"
$START_NAMESPACE -r -A ||
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 || {
- $SHOW_NAMESPACE
- error "(3) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 3
local repaired=$($SHOW_NAMESPACE |
awk '/^striped_shards_repaired/ { print $2 }')
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
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "For some reason, the slave MDT-object of the striped directory"
check_mount_and_prep
echo "Inject failure stub on MDT0 to simulate the case that the"
- echo "slave MDT-object (that resides on differnt MDT as the master"
+ 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
$START_NAMESPACE -r -A ||
error "(2) Fail to start LFSCK for namespace"
- wait_update_facet mds2 "$LCTL get_param -n \
- mdd.$(facet_svc mds2).lfsck_namespace |
- awk '/^status/ { print \\\$2 }'" "completed" 32 ||
- error "(3) unexpected status"
+ wait_all_targets_blocked namespace completed 3
local repaired=$(do_facet mds2 $LCTL get_param -n \
mdd.$(facet_svc mds2).lfsck_namespace |
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
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "For some reason, the stripe index in the slave LMV EA is"
$START_NAMESPACE -r -A ||
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 || {
- $SHOW_NAMESPACE
- error "(3) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 3
local repaired=$($SHOW_NAMESPACE |
awk '/^striped_shards_repaired/ { print $2 }')
run_test 31g "Repair the corrupted slave LMV EA"
test_31h() {
- [ $MDSCOUNT -lt 2 ] &&
- skip "The test needs at least 2 MDTs" && return
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
echo "#####"
echo "For some reason, the shard's name entry in the striped"
$START_NAMESPACE -r -A ||
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 || {
- $SHOW_NAMESPACE
- error "(3) unexpected status"
- }
+ wait_all_targets_blocked namespace completed 3
local repaired=$($SHOW_NAMESPACE |
awk '/^dirent_repaired/ { print $2 }')
}
run_test 31h "Repair the corrupted shard's name entry"
+test_32a()
+{
+ lfsck_prep 5 5
+ umount_client $MOUNT
+
+ #define OBD_FAIL_LFSCK_ENGINE_DELAY 0x162d
+ do_facet $SINGLEMDS $LCTL set_param fail_val=3 fail_loc=0x162d
+ $START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
+
+ local STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
+ [ "$STATUS" == "scanning-phase1" ] ||
+ error "(2) Expect 'scanning-phase1', but got '$STATUS'"
+
+ echo "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 4
+
+ echo "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 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()
+{
+ lfsck_prep 5 5
+
+ $START_LAYOUT --dryrun -o -r ||
+ error "(1) Fail to start layout LFSCK"
+ wait_all_targets_blocked layout completed 2
+
+ local PARAMS=$($SHOW_LAYOUT | awk '/^param/ { print $2 }')
+ [ "$PARAMS" == "dryrun,all_targets,orphan" ] ||
+ error "(3) Expect 'dryrun,all_targets,orphan', got '$PARAMS'"
+
+ $START_NAMESPACE -e abort -A -r ||
+ error "(4) Fail to start namespace LFSCK"
+ wait_all_targets_blocked namespace completed 5
+
+ PARAMS=$($SHOW_NAMESPACE | awk '/^param/ { print $2 }')
+ [ "$PARAMS" == "failout,all_targets" ] ||
+ error "(6) Expect 'failout,all_targets', got '$PARAMS'"
+}
+run_test 33 "check LFSCK paramters"
+
+test_34()
+{
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
+ [ $(facet_fstype $SINGLEMDS) != zfs ] &&
+ skip "Only valid for ZFS backend" && return
+
+ 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 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() {
+ [ $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() {
+ [ $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 $rc
+}
+run_test 37 "LFSCK must skip a ORPHAN"
+
+
# restore MDS/OST size
MDSSIZE=${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
git://git.whamcloud.com / fs / lustre-release.git / blobdiff