set -e
ONLY=${ONLY:-"$*"}
+
+#Bug number for excepting test
ALWAYS_EXCEPT="$SANITY_LFSCK_EXCEPT"
+
[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
require_dsh_mds || exit 0
+load_modules
+
+if ! check_versions; then
+ skip "It is NOT necessary to test lfsck under interoperation mode"
+ exit 0
+fi
+
+[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.60) ]] &&
+ skip "Need MDS version at least 2.3.60" && exit 0
+
LTIME=${LTIME:-120}
SAVED_MDSSIZE=${MDSSIZE}
formatall
setupall
-[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.60) ]] &&
- skip "Need MDS version at least 2.3.60" && check_and_cleanup_lustre &&
- exit 0
-
[[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.4.90) ]] &&
ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2c"
$LCTL get_param -n obdfilter.${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"
+}
+
test_0() {
lfsck_prep 3 3
[ $repaired -eq 1 ] ||
error "(5) Fail to repair crashed FID-in-dirent: $repaired"
+ run_e2fsck_on_mdt0
+
mount_client $MOUNT || error "(6) Fail to start client!"
#define OBD_FAIL_FID_LOOKUP 0x1505
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
[ $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 ||
[ $repaired -ge 9 ] ||
error "(9) Fail to re-generate FID-in-dirent: $repaired"
+ run_e2fsck_on_mdt0
+
mount_client $MOUNT || error "(10) Fail to start client!"
#define OBD_FAIL_FID_LOOKUP 0x1505
[ $repaired -ge 2 ] ||
error "(9) Fail to generate FID-in-dirent for IGIF: $repaired"
+ run_e2fsck_on_mdt0
+
mount_client $MOUNT || error "(10) Fail to start client!"
#define OBD_FAIL_FID_LOOKUP 0x1505
echo "stop $SINGLEMDS"
stop $SINGLEMDS > /dev/null || error "(4) Fail to stop MDS!"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
echo "start $SINGLEMDS"
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
error "(5) Fail to start MDS!"
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_namespace |
awk '/^status/ { print \\\$2 }'" "completed" 30 || {
error "(4) unexpected status"
}
+ umount_client $MOUNT
echo "stop $SINGLEMDS"
stop $SINGLEMDS > /dev/null || error "(5) Fail to stop MDS!"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
echo "start $SINGLEMDS"
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
error "(6) Fail to start MDS!"
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_namespace |
awk '/^status/ { print \\\$2 }'" "completed" 30 || {
[ "$STATUS" == "paused" ] ||
error "(20) Expect 'paused', but got '$STATUS'"
+ echo "stop $SINGLEMDS"
+ stop $SINGLEMDS > /dev/null || error "(20.1) Fail to stop MDS!"
+
+ echo "start $SINGLEMDS without resume LFSCK"
+ start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SKIP_LFSCK > /dev/null ||
+ error "(20.2) Fail to start MDS!"
+
+ timer=0
+ while [ $timer -lt $timeout ]; do
+ STATUS=$(do_facet $SINGLEMDS "$LCTL get_param -n \
+ mdt.${MDT_DEV}.recovery_status |
+ awk '/^status/ { print \\\$2 }'")
+ [ "$STATUS" != "RECOVERING" ] && break;
+ sleep 1
+ timer=$((timer + 1))
+ done
+
+ [ $timer != $timeout ] ||
+ error "(20.3) recovery timeout"
+
+ STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
+ [ "$STATUS" == "paused" ] ||
+ error "(20.4) Expect 'paused', but got '$STATUS'"
+
#define OBD_FAIL_LFSCK_DELAY3 0x1602
do_facet $SINGLEMDS $LCTL set_param fail_val=2 fail_loc=0x1602
return 0
fi
- lfsck_prep 70 70
+ [[ $server_version -ge $(version_code 2.7.50) ]] ||
+ { skip "Need MDS version >= 2.7.50"; return; }
+
+ check_mount_and_prep
+ $LFS mkdir -i 0 $DIR/$tdir/lfsck || error "(1) Fail to mkdir lfsck"
+ $LFS setstripe -c 1 -i -1 $DIR/$tdir/lfsck
+ createmany -o $DIR/$tdir/lfsck/f 5000
local BASE_SPEED1=100
local RUN_TIME1=10
- $START_NAMESPACE -r -s $BASE_SPEED1 || error "(3) Fail to start LFSCK!"
+ $START_LAYOUT -r -s $BASE_SPEED1 || error "(2) Fail to start LFSCK!"
sleep $RUN_TIME1
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
+ STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
error "(3) Expect 'scanning-phase1', but got '$STATUS'"
- local SPEED=$($SHOW_NAMESPACE |
+ local SPEED=$($SHOW_LAYOUT |
awk '/^average_speed_phase1/ { print $2 }')
# There may be time error, normally it should be less than 2 seconds.
$LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit $BASE_SPEED2
sleep $RUN_TIME2
- SPEED=$($SHOW_NAMESPACE | awk '/^average_speed_phase1/ { print $2 }')
+ SPEED=$($SHOW_LAYOUT | awk '/^average_speed_phase1/ { print $2 }')
# MIN_MARGIN = 0.8 = 8 / 10
local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
$LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
wait_update_facet $SINGLEMDS \
- "$LCTL get_param -n mdd.${MDT_DEV}.lfsck_namespace|\
- awk '/^status/ { print \\\$2 }'" "completed" 30 ||
+ "$LCTL get_param -n mdd.${MDT_DEV}.lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 30 ||
error "(7) Failed to get expected 'completed'"
}
run_test 9a "LFSCK speed control (1)"
return 0
fi
+ [[ $server_version -ge $(version_code 2.7.50) ]] ||
+ { skip "Need MDS version >= 2.7.50"; return; }
+
lfsck_prep 0 0
echo "Preparing another 50 * 50 files (with error) at $(date)."
echo "set fail_loc=0x160d to skip the updating LAST_ID on-disk"
#define OBD_FAIL_LFSCK_SKIP_LASTID 0x160d
do_facet ost1 $LCTL set_param fail_loc=0x160d
- createmany -o $DIR/$tdir/f 64
+
+ local count=$(precreated_ost_obj_count 0 0)
+
+ createmany -o $DIR/$tdir/f $((count + 32))
+
+ local proc_path="${FSNAME}-OST0000-osc-MDT0000"
+ local seq=$(do_facet mds1 $LCTL get_param -n \
+ osp.${proc_path}.prealloc_last_seq)
local lastid1=$(do_facet ost1 "lctl get_param -n \
- obdfilter.${ost1_svc}.last_id" | grep 0x100000000 |
+ obdfilter.${ost1_svc}.last_id" | grep $seq |
awk -F: '{ print $2 }')
umount_client $MOUNT
for ((i = 0; i < 60; i++)); do
lastid2=$(do_facet ost1 "lctl get_param -n \
- obdfilter.${ost1_svc}.last_id" | grep 0x100000000 |
+ obdfilter.${ost1_svc}.last_id" | grep $seq |
awk -F: '{ print $2 }')
[ ! -z $lastid2 ] && break;
sleep 1
echo "the on-disk LAST_ID should have been rebuilt"
wait_update_facet ost1 "$LCTL get_param -n \
- obdfilter.${ost1_svc}.last_id | grep 0x100000000 |
+ obdfilter.${ost1_svc}.last_id | grep $seq |
awk -F: '{ print \\\$2 }'" "$lastid1" 60 || {
- $LCTL get_param -n obdfilter.${ost1_svc}.last_id
- error "(9) expect lastid1 0x100000000:$lastid1"
+ do_facet ost1 $LCTL get_param -n \
+ obdfilter.${ost1_svc}.last_id
+ error "(9) expect lastid1 $seq:$lastid1"
}
do_facet ost1 $LCTL set_param fail_loc=0
error "(7) MDS${k} is not the expected 'completed'"
done
+ start_full_debug_logging
+
echo "Start layout LFSCK on all targets by single command (-s 1)."
do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -t layout -A \
-s 1 -r || error "(8) Fail to start LFSCK on all devices!"
awk '/^status/ { print \\\$2 }'" "completed" 32 ||
error "(14) MDS${k} is not the expected 'completed'"
done
+
+ stop_full_debug_logging
}
run_test 12 "single command to trigger LFSCK on all devices"
}
run_test 15b "LFSCK can repair unmatched MDT-object/OST-object pairs (2)"
+test_15c() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "We need at least 2 MDSes for this test" && return
+
+ [ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.55) ] &&
+ skip "Skip the test after 2.7.55 see LU-6437" && return
+
+ echo "#####"
+ echo "According to current metadata migration implementation,"
+ echo "before the old MDT-object is removed, both the new MDT-object"
+ echo "and old MDT-object will reference the same LOV layout. Then if"
+ echo "the layout LFSCK finds the new MDT-object by race, it will"
+ echo "regard related OST-object(s) as multiple referenced case, and"
+ echo "will try to create new OST-object(s) for the new MDT-object."
+ echo "To avoid such trouble, the layout LFSCK needs to lock the old"
+ echo "MDT-object before confirm the multiple referenced case."
+ echo "#####"
+
+ check_mount_and_prep
+ $LFS mkdir -i 1 $DIR/$tdir/a1
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir/a1
+ dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=1
+ cancel_lru_locks osc
+
+ echo "Inject failure stub on MDT1 to delay the migration"
+
+ #define OBD_FAIL_MIGRATE_DELAY 0x1803
+ do_facet mds2 $LCTL set_param fail_val=5 fail_loc=0x1803
+ echo "Migrate $DIR/$tdir/a1 from MDT1 to MDT0 with delay"
+ $LFS migrate -m 0 $DIR/$tdir/a1 &
+
+ sleep 1
+ echo "Trigger layout LFSCK to race with the migration"
+ $START_LAYOUT -A -r || error "(1) Fail to start layout LFSCK!"
+
+ 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
+
+ do_facet mds2 $LCTL set_param fail_loc=0 fail_val=0
+ local repaired=$($SHOW_LAYOUT |
+ awk '/^repaired_unmatched_pair/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(3) Fail to repair unmatched pair: $repaired"
+
+ repaired=$($SHOW_LAYOUT |
+ awk '/^repaired_multiple_referenced/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(4) Unexpectedly repaird multiple references: $repaired"
+}
+run_test 15c "LFSCK can repair unmatched MDT-object/OST-object pairs (3)"
+
test_16() {
echo "#####"
echo "If the OST-object's owner information does not match the owner"
#define OBD_FAIL_LFSCK_DELAY3 0x1602
do_facet $SINGLEMDS $LCTL set_param fail_val=10 fail_loc=0x1602
+ start_full_debug_logging
+
echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
$START_LAYOUT -r -o -c || error "(2) Fail to start LFSCK for layout!"
error "(5) OST${k} Expect 'completed', but got '$cur_status'"
done
+ stop_full_debug_logging
+
local repaired=$(do_facet $SINGLEMDS $LCTL get_param -n \
mdd.$(facet_svc $SINGLEMDS).lfsck_layout |
awk '/^repaired_orphan/ { print $2 }')
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 namesapce"
+ echo "".." name entry. The parent_B does not exist. Then the namespace"
echo "LFSCK will repair the child directory's ".." name entry."
echo "#####"
echo "The parent_A references the child directory via the name entry_B,"
echo "but the child directory back references another parent_C via its"
echo "".." name entry. The parent_C exists, but there is no the name"
- echo "entry_B under the parent_C. Then the namesapce LFSCK will repair"
+ echo "entry_B under the parent_C. Then the namespace LFSCK will repair"
echo "the child directory's ".." name entry and its linkEA."
echo "#####"
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