MCREATE=${MCREATE:-mcreate}
SAVED_MDSSIZE=${MDSSIZE}
SAVED_OSTSIZE=${OSTSIZE}
+SAVED_OSTCOUNT=${OSTCOUNT}
# use small MDS + OST size to speed formatting time
# do not use too small MDSSIZE/OSTSIZE, which affect the default journal size
MDSSIZE=100000
OSTSIZE=100000
+# no need too much OSTs, to reduce the format/start/stop overhead
+[ $OSTCOUNT -gt 4 ] && OSTCOUNT=4
-check_and_setup_lustre
+# build up a clean test environment.
+formatall
+setupall
[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.60) ]] &&
skip "Need MDS version at least 2.3.60" && check_and_cleanup_lustre &&
ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2c"
[[ $(lustre_version_code ost1) -lt $(version_code 2.5.55) ]] &&
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 12 13 14 15 16 17 18 19"
+ ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 12 13 14 15 16 17 18 19 20 21"
+
+[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
+# bug number for skipped test: LU-4970
+ ALWAYS_EXCEPT="$ALWAYS_EXCEPT 14"
build_test_filter
local nfiles=$2
local igif=$3
- echo "formatall"
- formatall > /dev/null
-
- echo "setupall"
- setupall > /dev/null
+ check_mount_and_prep
+ echo "preparing... $nfiles * $ndirs files will be created $(date)."
if [ ! -z $igif ]; then
#define OBD_FAIL_FID_IGIF 0x1504
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1504
fi
- echo "preparing... ${nfiles} * ${ndirs} files will be created."
- mkdir -p $DIR/$tdir
- cp $LUSTRE/tests/*.sh $DIR/
- for ((i = 0; i < ${ndirs}; i++)); do
- mkdir $DIR/$tdir/d${i}
- touch $DIR/$tdir/f${i}
- for ((j = 0; j < ${nfiles}; j++)); do
- touch $DIR/$tdir/d${i}/f${j}
- done
- mkdir $DIR/$tdir/e${i}
- done
+ cp $LUSTRE/tests/*.sh $DIR/$tdir/
+ if [ $ndirs -gt 0 ]; then
+ createmany -d $DIR/$tdir/d $ndirs
+ createmany -m $DIR/$tdir/f $ndirs
+ if [ $nfiles -gt 0 ]; then
+ for ((i = 0; i < $ndirs; i++)); do
+ createmany -m $DIR/$tdir/d${i}/f $nfiles > \
+ /dev/null || error "createmany $nfiles"
+ done
+ fi
+ createmany -d $DIR/$tdir/e $ndirs
+ fi
if [ ! -z $igif ]; then
touch $DIR/$tdir/dummy
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
fi
- echo "prepared."
- cleanup_mount $MOUNT > /dev/null || error "Fail to stop client!"
- echo "stop $SINGLEMDS"
- stop $SINGLEMDS > /dev/null || error "Fail to stop MDS!"
+ echo "prepared $(date)."
}
test_0() {
- lfsck_prep 10 10
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
+ lfsck_prep 3 3
#define OBD_FAIL_LFSCK_DELAY1 0x1600
- do_facet $SINGLEMDS $LCTL set_param fail_val=3
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1600
- $START_NAMESPACE || error "(2) Fail to start LFSCK for namespace!"
+ do_facet $SINGLEMDS $LCTL set_param fail_val=3 fail_loc=0x1600
+ $START_NAMESPACE -r || error "(2) Fail to start LFSCK for namespace!"
$SHOW_NAMESPACE || error "Fail to monitor LFSCK (3)"
[ "$STATUS" == "scanning-phase1" ] ||
error "(8) Expect 'scanning-phase1', but got '$STATUS'"
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
- wait_update_facet $SINGLEMDS \
- "$LCTL get_param -n mdd.${MDT_DEV}.lfsck_namespace | \
- awk '/^status/ { print \\\$2 }'" "completed" 20 || \
+ 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" 6 || {
+ $SHOW_NAMESPACE
error "(9) unexpected status"
+ }
local repaired=$($SHOW_NAMESPACE |
awk '/^updated_phase1/ { print $2 }')
local scanned1=$($SHOW_NAMESPACE | awk '/^success_count/ { print $2 }')
$START_NAMESPACE -r || error "(11) Fail to reset LFSCK!"
- wait_update_facet $SINGLEMDS \
- "$LCTL get_param -n mdd.${MDT_DEV}.lfsck_namespace | \
- awk '/^status/ { print \\\$2 }'" "completed" 20 || \
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
+ $SHOW_NAMESPACE
error "(12) unexpected status"
+ }
local scanned2=$($SHOW_NAMESPACE | awk '/^success_count/ { print $2 }')
[ $((scanned1 + 1)) -eq $scanned2 ] ||
error "(13) Expect success $((scanned1 + 1)), but got $scanned2"
echo "stopall, should NOT crash LU-3649"
- stopall > /dev/null
+ stopall || error "(14) Fail to stopall"
}
run_test 0 "Control LFSCK manually"
skip "OI Scrub not implemented for ZFS" && return
lfsck_prep 1 1
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
-
- mount_client $MOUNT || error "(2) Fail to start client!"
#define OBD_FAIL_FID_INDIR 0x1501
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1501
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
umount_client $MOUNT
- $START_NAMESPACE || error "(3) Fail to start LFSCK for namespace!"
-
- sleep 3
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(4) Expect 'completed', but got '$STATUS'"
+ $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" 6 || {
+ $SHOW_NAMESPACE
+ error "(4) unexpected status"
+ }
local repaired=$($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 crashed FID-in-dirent: $repaired"
skip "OI Scrub not implemented for ZFS" && return
lfsck_prep 1 1
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
-
- mount_client $MOUNT || error "(2) Fail to start client!"
#define OBD_FAIL_FID_INLMA 0x1502
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1502
umount_client $MOUNT
#define OBD_FAIL_FID_NOLMA 0x1506
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1506
- $START_NAMESPACE || error "(3) Fail to start LFSCK for namespace!"
-
- sleep 3
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(4) Expect 'completed', but got '$STATUS'"
+ $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" 6 || {
+ $SHOW_NAMESPACE
+ error "(4) unexpected status"
+ }
local repaired=$($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 missed FID-in-LMA: $repaired"
test_2a() {
lfsck_prep 1 1
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
-
- mount_client $MOUNT || error "(2) Fail to start client!"
#define OBD_FAIL_LFSCK_LINKEA_CRASH 0x1603
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1603
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
umount_client $MOUNT
- $START_NAMESPACE || error "(3) Fail to start LFSCK for namespace!"
-
- sleep 3
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(4) Expect 'completed', but got '$STATUS'"
+ $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" 6 || {
+ $SHOW_NAMESPACE
+ error "(4) unexpected status"
+ }
local repaired=$($SHOW_NAMESPACE |
+ awk '/^linkea_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 crashed linkEA: $repaired"
test_2b()
{
lfsck_prep 1 1
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
-
- mount_client $MOUNT || error "(2) Fail to start client!"
#define OBD_FAIL_LFSCK_LINKEA_MORE 0x1604
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1604
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
umount_client $MOUNT
- $START_NAMESPACE || error "(3) Fail to start LFSCK for namespace!"
-
- sleep 3
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(4) Expect 'completed', but got '$STATUS'"
+ $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" 6 || {
+ $SHOW_NAMESPACE
+ error "(4) unexpected status"
+ }
local repaired=$($SHOW_NAMESPACE |
awk '/^updated_phase2/ { print $2 }')
test_2c()
{
lfsck_prep 1 1
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
-
- mount_client $MOUNT || error "(2) Fail to start client!"
#define OBD_FAIL_LFSCK_LINKEA_MORE2 0x1605
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1605
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
umount_client $MOUNT
- $START_NAMESPACE || error "(3) Fail to start LFSCK for namespace!"
-
- sleep 3
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(4) Expect 'completed', but got '$STATUS'"
+ $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" 6 || {
+ $SHOW_NAMESPACE
+ error "(4) unexpected status"
+ }
local repaired=$($SHOW_NAMESPACE |
awk '/^updated_phase2/ { print $2 }')
skip "OI Scrub not implemented for ZFS" && return
lfsck_prep 3 3
+ cleanup_mount $MOUNT || error "(0.1) Fail to stop client!"
+ stop $SINGLEMDS > /dev/null || error "(0.2) Fail to stop MDS!"
+
mds_backup_restore $SINGLEMDS || error "(1) Fail to backup/restore!"
echo "start $SINGLEMDS with disabling OI scrub"
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
error "(2) Fail to start MDS!"
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "init" ] ||
- error "(3) Expect 'init', but got '$STATUS'"
-
#define OBD_FAIL_LFSCK_DELAY2 0x1601
- do_facet $SINGLEMDS $LCTL set_param fail_val=1
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1601
- $START_NAMESPACE || error "(4) Fail to start LFSCK for namespace!"
+ do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
+ $START_NAMESPACE -r || error "(4) Fail to start LFSCK for namespace!"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^flags/ { print \\\$2 }'" "inconsistent" 6 || {
+ $SHOW_NAMESPACE
+ error "(5) unexpected status"
+ }
- sleep 5
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
+ local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
- error "(5) Expect 'scanning-phase1', but got '$STATUS'"
-
- local FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
- [ "$FLAGS" == "inconsistent" ] ||
- error "(6) Expect 'inconsistent', but got '$FLAGS'"
+ error "(6) Expect 'scanning-phase1', but got '$STATUS'"
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
- sleep 3
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(7) Expect 'completed', but got '$STATUS'"
+ 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" 6 || {
+ $SHOW_NAMESPACE
+ error "(7) unexpected status"
+ }
FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
[ -z "$FLAGS" ] || error "(8) Expect empty flags, but got '$FLAGS'"
local repaired=$($SHOW_NAMESPACE |
+ awk '/^dirent_repaired/ { print $2 }')
+ # for interop with old server
+ [ -z "$repaired" ] &&
+ repaired=$($SHOW_NAMESPACE |
awk '/^updated_phase1/ { print $2 }')
+
[ $repaired -ge 9 ] ||
- error "(9) Fail to repair crashed linkEA: $repaired"
+ error "(9) Fail to re-generate FID-in-dirent: $repaired"
mount_client $MOUNT || error "(10) 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 "(11) no FID-in-dirent."
-
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
}
run_test 4 "FID-in-dirent can be rebuilt after MDT file-level backup/restore"
skip "OI Scrub not implemented for ZFS" && return
lfsck_prep 1 1 1
+ cleanup_mount $MOUNT || error "(0.1) Fail to stop client!"
+ stop $SINGLEMDS > /dev/null || error "(0.2) Fail to stop MDS!"
+
mds_backup_restore $SINGLEMDS 1 || error "(1) Fail to backup/restore!"
echo "start $SINGLEMDS with disabling OI scrub"
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
error "(2) Fail to start MDS!"
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "init" ] ||
- error "(3) Expect 'init', but got '$STATUS'"
-
#define OBD_FAIL_LFSCK_DELAY2 0x1601
- do_facet $SINGLEMDS $LCTL set_param fail_val=1
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1601
- $START_NAMESPACE || error "(4) Fail to start LFSCK for namespace!"
+ do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
+ $START_NAMESPACE -r || error "(4) Fail to start LFSCK for namespace!"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^flags/ { print \\\$2 }'" "inconsistent,upgrade" 6 || {
+ $SHOW_NAMESPACE
+ error "(5) unexpected status"
+ }
- sleep 5
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
+ local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
- error "(5) Expect 'scanning-phase1', but got '$STATUS'"
-
- local FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
- [ "$FLAGS" == "inconsistent,upgrade" ] ||
- error "(6) Expect 'inconsistent,upgrade', but got '$FLAGS'"
+ error "(6) Expect 'scanning-phase1', but got '$STATUS'"
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
- sleep 3
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(7) Expect 'completed', but got '$STATUS'"
+ 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" 6 || {
+ $SHOW_NAMESPACE
+ error "(7) unexpected status"
+ }
FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
[ -z "$FLAGS" ] || error "(8) Expect empty flags, but got '$FLAGS'"
local repaired=$($SHOW_NAMESPACE |
+ awk '/^dirent_repaired/ { print $2 }')
+ # for interop with old server
+ [ -z "$repaired" ] &&
+ repaired=$($SHOW_NAMESPACE |
awk '/^updated_phase1/ { print $2 }')
+
[ $repaired -ge 2 ] ||
- error "(9) Fail to repair crashed linkEA: $repaired"
+ error "(9) Fail to generate FID-in-dirent for IGIF: $repaired"
mount_client $MOUNT || error "(10) Fail to start client!"
[ "$dummyname" == "$DIR/$tdir/dummy" ] ||
error "(13) Fail to generate linkEA: $dummyfid $dummyname"
}
-run_test 5 "LFSCK can handle IFIG object upgrading"
+run_test 5 "LFSCK can handle IGIF object upgrading"
test_6a() {
- lfsck_prep 10 10
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
+ lfsck_prep 5 5
#define OBD_FAIL_LFSCK_DELAY1 0x1600
- do_facet $SINGLEMDS $LCTL set_param fail_val=1
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1600
- $START_NAMESPACE || error "(2) Fail to start LFSCK for namespace!"
+ do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1600
+ $START_NAMESPACE -r || error "(2) Fail to start LFSCK for namespace!"
local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
# Fail the LFSCK to guarantee there is at least one checkpoint
#define OBD_FAIL_LFSCK_FATAL1 0x1608
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x80001608
- sleep 3
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "failed" ] ||
- error "(4) Expect 'failed', but got '$STATUS'"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "failed" 6 || {
+ $SHOW_NAMESPACE
+ error "(4) unexpected status"
+ }
- local POSITION0=$($SHOW_NAMESPACE |
- awk '/^last_checkpoint_position/ { print $2 }' |
- tr -d ',')
+ local POS0=$($SHOW_NAMESPACE |
+ awk '/^last_checkpoint_position/ { print $2 }' |
+ tr -d ',')
#define OBD_FAIL_LFSCK_DELAY1 0x1600
- do_facet $SINGLEMDS $LCTL set_param fail_val=1
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1600
+ do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1600
$START_NAMESPACE || error "(5) Fail to start LFSCK for namespace!"
STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
error "(6) Expect 'scanning-phase1', but got '$STATUS'"
- local POSITION1=$($SHOW_NAMESPACE |
- awk '/^latest_start_position/ { print $2 }' |
- tr -d ',')
- [ $POSITION0 -lt $POSITION1 ] ||
- error "(7) Expect larger than: $POSITION0, but got $POSITION1"
+ local POS1=$($SHOW_NAMESPACE |
+ awk '/^latest_start_position/ { print $2 }' |
+ tr -d ',')
+ [[ $POS0 -lt $POS1 ]] ||
+ error "(7) Expect larger than: $POS0, but got $POS1"
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
- sleep 3
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(8) Expect 'completed', but got '$STATUS'"
+ 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" 6 || {
+ $SHOW_NAMESPACE
+ error "(8) unexpected status"
+ }
}
run_test 6a "LFSCK resumes from last checkpoint (1)"
test_6b() {
- lfsck_prep 10 10
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
+ lfsck_prep 5 5
#define OBD_FAIL_LFSCK_DELAY2 0x1601
- do_facet $SINGLEMDS $LCTL set_param fail_val=1
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1601
- $START_NAMESPACE || error "(2) Fail to start LFSCK for namespace!"
+ do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
+ $START_NAMESPACE -r || error "(2) Fail to start LFSCK for namespace!"
local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
error "(3) Expect 'scanning-phase1', but got '$STATUS'"
- # Sleep 3 sec to guarantee at least one object processed by LFSCK
- sleep 3
+ # Sleep 5 sec to guarantee that we are in the directory scanning
+ sleep 5
# Fail the LFSCK to guarantee there is at least one checkpoint
#define OBD_FAIL_LFSCK_FATAL2 0x1609
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x80001609
- sleep 3
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "failed" ] ||
- error "(4) Expect 'failed', but got '$STATUS'"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "failed" 6 || {
+ $SHOW_NAMESPACE
+ error "(4) unexpected status"
+ }
- local POSITION0=$($SHOW_NAMESPACE |
- awk '/^last_checkpoint_position/ { print $4 }')
+ local O_POS0=$($SHOW_NAMESPACE |
+ awk '/^last_checkpoint_position/ { print $2 }' |
+ tr -d ',')
+
+ local D_POS0=$($SHOW_NAMESPACE |
+ awk '/^last_checkpoint_position/ { print $4 }')
#define OBD_FAIL_LFSCK_DELAY2 0x1601
- do_facet $SINGLEMDS $LCTL set_param fail_val=1
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1601
+ do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
$START_NAMESPACE || error "(5) Fail to start LFSCK for namespace!"
STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
error "(6) Expect 'scanning-phase1', but got '$STATUS'"
- local POSITION1=$($SHOW_NAMESPACE |
- awk '/^latest_start_position/ { print $4 }')
- if [ $POSITION0 -gt $POSITION1 ]; then
- [ $POSITION1 -eq 0 -a $POSITION0 -eq $((POSITION1 + 1)) ] ||
- error "(7) Expect larger than: $POSITION0, but got $POSITION1"
+ local O_POS1=$($SHOW_NAMESPACE |
+ awk '/^latest_start_position/ { print $2 }' |
+ tr -d ',')
+ local D_POS1=$($SHOW_NAMESPACE |
+ awk '/^latest_start_position/ { print $4 }')
+
+ if [ "$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"
+ else
+ [[ $D_POS0 -lt $D_POS1 ]] ||
+ error "(7.2) $D_POS1 is not larger than $D_POS0"
fi
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
- sleep 3
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(8) Expect 'completed', but got '$STATUS'"
+ 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" 6 || {
+ $SHOW_NAMESPACE
+ error "(8) unexpected status"
+ }
}
run_test 6b "LFSCK resumes from last checkpoint (2)"
test_7a()
{
- lfsck_prep 10 10
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
+ lfsck_prep 5 5
+ umount_client $MOUNT
#define OBD_FAIL_LFSCK_DELAY2 0x1601
- do_facet $SINGLEMDS $LCTL set_param fail_val=1
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1601
- $START_NAMESPACE || error "(2) Fail to start LFSCK for namespace!"
+ do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
+ $START_NAMESPACE -r || error "(2) Fail to start LFSCK for namespace!"
local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
error "(5) Fail to start MDS!"
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "scanning-phase1" ] ||
- error "(6) Expect 'scanning-phase1', but got '$STATUS'"
-
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
- sleep 3
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(7) Expect 'completed', but got '$STATUS'"
+ 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 || {
+ $SHOW_NAMESPACE
+ error "(6) unexpected status"
+ }
}
run_test 7a "non-stopped LFSCK should auto restarts after MDS remount (1)"
test_7b()
{
lfsck_prep 2 2
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
-
- mount_client $MOUNT || error "(2) Fail to start client!"
#define OBD_FAIL_LFSCK_LINKEA_MORE 0x1604
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1604
done
#define OBD_FAIL_LFSCK_DELAY3 0x1602
- do_facet $SINGLEMDS $LCTL set_param fail_val=1
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1602
- $START_NAMESPACE || error "(3) Fail to start LFSCK for namespace!"
-
- sleep 3
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "scanning-phase2" ] ||
- error "(4) Expect 'scanning-phase2', but got '$STATUS'"
+ do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1602
+ $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 }'" "scanning-phase2" 6 || {
+ $SHOW_NAMESPACE
+ error "(4) unexpected status"
+ }
echo "stop $SINGLEMDS"
stop $SINGLEMDS > /dev/null || error "(5) Fail to stop MDS!"
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
error "(6) Fail to start MDS!"
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "scanning-phase2" ] ||
- error "(7) Expect 'scanning-phase2', but got '$STATUS'"
-
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
- sleep 3
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(8) Expect 'completed', but got '$STATUS'"
+ 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 || {
+ $SHOW_NAMESPACE
+ error "(7) unexpected status"
+ }
}
run_test 7b "non-stopped LFSCK should auto restarts after MDS remount (2)"
test_8()
{
+ echo "formatall"
+ formatall > /dev/null
+ echo "setupall"
+ setupall > /dev/null
+
lfsck_prep 20 20
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
[ "$STATUS" == "init" ] ||
error "(2) Expect 'init', but got '$STATUS'"
- mount_client $MOUNT || error "(3) Fail to start client!"
-
#define OBD_FAIL_LFSCK_LINKEA_CRASH 0x1603
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1603
mkdir $DIR/$tdir/crashed
touch $DIR/$tdir/dummy${i}
done
+ umount_client $MOUNT || error "(3) Fail to stop client!"
+
#define OBD_FAIL_LFSCK_DELAY2 0x1601
- do_facet $SINGLEMDS $LCTL set_param fail_val=2
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1601
+ do_facet $SINGLEMDS $LCTL set_param fail_val=2 fail_loc=0x1601
$START_NAMESPACE || error "(4) Fail to start LFSCK for namespace!"
STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
#define OBD_FAIL_LFSCK_FATAL2 0x1609
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x80001609
- sleep 3
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "failed" ] ||
- error "(10) Expect 'failed', but got '$STATUS'"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "failed" 6 || {
+ $SHOW_NAMESPACE
+ error "(10) unexpected status"
+ }
#define OBD_FAIL_LFSCK_DELAY1 0x1600
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1600
error "(20) Expect 'paused', but got '$STATUS'"
#define OBD_FAIL_LFSCK_DELAY3 0x1602
- do_facet $SINGLEMDS $LCTL set_param fail_val=2
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1602
+ do_facet $SINGLEMDS $LCTL set_param fail_val=2 fail_loc=0x1602
$START_NAMESPACE || error "(21) Fail to start LFSCK for namespace!"
- sleep 2
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "scanning-phase2" ] ||
- error "(22) Expect 'scanning-phase2', but got '$STATUS'"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "scanning-phase2" 6 || {
+ $SHOW_NAMESPACE
+ error "(22) unexpected status"
+ }
local FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
[ "$FLAGS" == "scanned-once,inconsistent" ] ||
error "(23) Expect 'scanned-once,inconsistent',but got '$FLAGS'"
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
- sleep 2
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(24) Expect 'completed', but got '$STATUS'"
+ 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" 6 || {
+ $SHOW_NAMESPACE
+ error "(24) unexpected status"
+ }
FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
[ -z "$FLAGS" ] || error "(25) Expect empty flags, but got '$FLAGS'"
-
}
run_test 8 "LFSCK state machine"
fi
lfsck_prep 70 70
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
-
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "init" ] ||
- error "(2) Expect 'init', but got '$STATUS'"
local BASE_SPEED1=100
local RUN_TIME1=10
- $START_NAMESPACE -s $BASE_SPEED1 || error "(3) Fail to start LFSCK!"
+ $START_NAMESPACE -r -s $BASE_SPEED1 || error "(3) Fail to start LFSCK!"
sleep $RUN_TIME1
STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
(RUN_TIME1 + RUN_TIME2) * 8 / 10))
- [ $SPEED -gt $MIN_SPEED ] ||
+ # Account for slow ZFS performance - LU-4934
+ [ $SPEED -gt $MIN_SPEED ] || [ $(facet_fstype $SINGLEMDS) -eq zfs ] ||
error "(5) Got speed $SPEED, expected more than $MIN_SPEED"
# MAX_MARGIN = 1.2 = 12 / 10
fi
lfsck_prep 0 0
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
- mount_client $MOUNT || error "(2) Fail to start client!"
-
- echo "Another preparing... 50 * 50 files (with error) will be created."
+ echo "Preparing another 50 * 50 files (with error) at $(date)."
#define OBD_FAIL_LFSCK_LINKEA_MORE 0x1604
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1604
+ createmany -d $DIR/$tdir/d 50
+ createmany -m $DIR/$tdir/f 50
for ((i = 0; i < 50; i++)); do
- mkdir -p $DIR/$tdir/d${i}
- touch $DIR/$tdir/f${i}
- for ((j = 0; j < 50; j++)); do
- touch $DIR/$tdir/d${i}/f${j}
- done
+ createmany -m $DIR/$tdir/d${i}/f 50 > /dev/null
done
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "init" ] ||
- error "(3) Expect 'init', but got '$STATUS'"
-
#define OBD_FAIL_LFSCK_NO_DOUBLESCAN 0x160c
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160c
- $START_NAMESPACE || error "(4) Fail to start LFSCK!"
-
- sleep 10
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "stopped" ] ||
- error "(5) Expect 'stopped', but got '$STATUS'"
+ $START_NAMESPACE -r || error "(4) Fail to start LFSCK!"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "stopped" 10 || {
+ $SHOW_NAMESPACE
+ error "(5) unexpected status"
+ }
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+ echo "Prepared at $(date)."
local BASE_SPEED1=50
local RUN_TIME1=10
local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
(RUN_TIME1 + RUN_TIME2) * 8 / 10))
- [ $SPEED -gt $MIN_SPEED ] ||
+ [ $SPEED -gt $MIN_SPEED ] ||[ $(facet_fstype $SINGLEMDS) -eq zfs ] ||
error "(9) Got speed $SPEED, expected more than $MIN_SPEED"
# MAX_MARGIN = 1.2 = 12 / 10
do_facet $SINGLEMDS \
$LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
- sleep 5
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(11) Expect 'completed', but got '$STATUS'"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
+ $SHOW_NAMESPACE
+ error "(11) unexpected status"
+ }
}
run_test 9b "LFSCK speed control (2)"
skip "lookup(..)/linkea on ZFS issue" && return
lfsck_prep 1 1
- echo "start $SINGLEMDS"
- start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
- error "(1) Fail to start MDS!"
-
- mount_client $MOUNT || error "(2) Fail to start client!"
+ echo "Preparing more files with error at $(date)."
#define OBD_FAIL_LFSCK_LINKEA_CRASH 0x1603
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1603
+
for ((i = 0; i < 1000; i = $((i+2)))); do
mkdir -p $DIR/$tdir/d${i}
touch $DIR/$tdir/f${i}
- for ((j = 0; j < 5; j++)); do
- touch $DIR/$tdir/d${i}/f${j}
- done
+ createmany -m $DIR/$tdir/d${i}/f 5 > /dev/null
done
#define OBD_FAIL_LFSCK_LINKEA_MORE 0x1604
do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1604
+
for ((i = 1; i < 1000; i = $((i+2)))); do
mkdir -p $DIR/$tdir/d${i}
touch $DIR/$tdir/f${i}
- for ((j = 0; j < 5; j++)); do
- touch $DIR/$tdir/d${i}/f${j}
- done
+ createmany -m $DIR/$tdir/d${i}/f 5 > /dev/null
done
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+ echo "Prepared at $(date)."
+
ln $DIR/$tdir/f200 $DIR/$tdir/d200/dummy
umount_client $MOUNT
mount_client $MOUNT || error "(3) Fail to start client!"
- local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "init" ] ||
- error "(4) Expect 'init', but got '$STATUS'"
-
- $START_NAMESPACE -s 100 || error "(5) Fail to start LFSCK!"
+ $START_NAMESPACE -r -s 100 || error "(5) Fail to start LFSCK!"
sleep 10
STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
do_facet $SINGLEMDS \
$LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
- umount_client $MOUNT
- sleep 10
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(16) Expect 'completed', but got '$STATUS'"
+ wait_update_facet $SINGLEMDS "$LCTL get_param -n \
+ mdd.${MDT_DEV}.lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
+ $SHOW_NAMESPACE
+ error "(16) unexpected status"
+ }
}
run_test 10 "System is available during LFSCK scanning"
}
test_11a() {
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$SETSTRIPE -c 1 -i 0 $DIR/$tdir
- createmany -o $DIR/$tdir/f 64
+ createmany -o $DIR/$tdir/f 64 || error "(0) Fail to create 64 files."
echo "stopall"
stopall > /dev/null
ost_remove_lastid 1 0 || error "(1) Fail to remove LAST_ID"
- echo "start ost1"
start ost1 $(ostdevname 1) $MOUNT_OPTS_NOSCRUB > /dev/null ||
error "(2) Fail to start ost1"
- local STATUS=$($SHOW_LAYOUT_ON_OST | awk '/^status/ { print $2 }')
- [ "$STATUS" == "init" ] ||
- error "(3) Expect 'init', but got '$STATUS'"
-
#define OBD_FAIL_LFSCK_DELAY4 0x160e
- do_facet ost1 $LCTL set_param fail_val=3
- do_facet ost1 $LCTL set_param fail_loc=0x160e
+ do_facet ost1 $LCTL set_param fail_val=3 fail_loc=0x160e
echo "trigger LFSCK for layout on ost1 to rebuild the LAST_ID(s)"
- $START_LAYOUT_ON_OST || error "(4) Fail to start LFSCK on OST!"
+ $START_LAYOUT_ON_OST -r || error "(4) Fail to start LFSCK on OST!"
wait_update_facet ost1 "$LCTL get_param -n \
obdfilter.${OST_DEV}.lfsck_layout |
awk '/^flags/ { print \\\$2 }'" "crashed_lastid" 60 || {
$SHOW_LAYOUT_ON_OST
- return 5
+ error "(5) unexpected status"
}
- do_facet ost1 $LCTL set_param fail_val=0
- do_facet ost1 $LCTL set_param fail_loc=0
+ do_facet ost1 $LCTL set_param fail_val=0 fail_loc=0
wait_update_facet ost1 "$LCTL get_param -n \
obdfilter.${OST_DEV}.lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 3 || {
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
$SHOW_LAYOUT_ON_OST
- return 6
+ error "(6) unexpected status"
}
echo "the LAST_ID(s) should have been rebuilt"
run_test 11a "LFSCK can rebuild lost last_id"
test_11b() {
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$SETSTRIPE -c 1 -i 0 $DIR/$tdir
echo "set fail_loc=0x160d to skip the updating LAST_ID on-disk"
awk -F: '{ print $2 }')
umount_client $MOUNT
- echo "stop ost1"
stop ost1 || error "(1) Fail to stop ost1"
#define OBD_FAIL_OST_ENOSPC 0x215
do_facet ost1 $LCTL set_param fail_loc=0x215
- echo "start ost1"
start ost1 $(ostdevname 1) $OST_MOUNT_OPTS ||
error "(2) Fail to start ost1"
- local STATUS=$($SHOW_LAYOUT_ON_OST | awk '/^status/ { print $2 }')
- [ "$STATUS" == "init" ] ||
- error "(3) Expect 'init', but got '$STATUS'"
-
for ((i = 0; i < 60; i++)); do
lastid2=$(do_facet ost1 "lctl get_param -n \
obdfilter.${ost1_svc}.last_id" | grep 0x100000000 |
error "(4) expect lastid1 [ $lastid1 ] > lastid2 [ $lastid2 ]"
echo "trigger LFSCK for layout on ost1 to rebuild the on-disk LAST_ID"
- $START_LAYOUT_ON_OST || error "(5) Fail to start LFSCK on OST!"
+ $START_LAYOUT_ON_OST -r || error "(5) Fail to start LFSCK on OST!"
wait_update_facet ost1 "$LCTL get_param -n \
obdfilter.${OST_DEV}.lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 3 || {
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
$SHOW_LAYOUT_ON_OST
- return 6
+ error "(6) unexpected status"
}
- echo "stop ost1"
stop ost1 || error "(7) Fail to stop ost1"
- echo "start ost1"
start ost1 $(ostdevname 1) $OST_MOUNT_OPTS ||
error "(8) Fail to start ost1"
}
do_facet ost1 $LCTL set_param fail_loc=0
+ stopall || error "(10) Fail to stopall"
}
run_test 11b "LFSCK can rebuild crashed last_id"
[ $MDSCOUNT -lt 2 ] &&
skip "We need at least 2 MDSes for test_12" && exit 0
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
-
- echo "All the LFSCK targets should be in 'init' status."
+ check_mount_and_prep
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" == "init" ] ||
- error "(1) MDS${k} Expect 'init', but got '$STATUS'"
-
$LFS mkdir -i $((k - 1)) $DIR/$tdir/${k}
- createmany -o $DIR/$tdir/${k}/f 100
+ createmany -o $DIR/$tdir/${k}/f 100 ||
+ error "(0) Fail to create 100 files."
done
echo "Start namespace LFSCK on all targets by single command (-s 1)."
do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -t namespace -A \
- -s 1 || error "(2) Fail to start LFSCK on all devices!"
+ -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
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 || error "(8) Fail to start LFSCK on all devices!"
+ -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
echo "MDT-object FID."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
echo "Inject failure stub to simulate bad lmm_oi"
#define OBD_FAIL_LFSCK_BAD_LMMOI 0x160f
createmany -o $DIR/$tdir/f 32
do_facet $SINGLEMDS $LCTL set_param fail_loc=0
- echo "stopall to cleanup object cache"
- stopall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
echo "Trigger layout LFSCK to find out the bad lmm_oi and fix them"
- $START_LAYOUT || error "(1) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 3 || return 2
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
+ $SHOW_LAYOUT
+ error "(2) unexpected status"
+ }
local repaired=$($SHOW_LAYOUT |
awk '/^repaired_others/ { print $2 }')
echo "otherwise, the LFSCK should re-create the missed OST-object."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
+ local count=$(precreated_ost_obj_count 0 0)
+
echo "Inject failure stub to simulate dangling referenced MDT-object"
#define OBD_FAIL_LFSCK_DANGLING 0x1610
do_facet ost1 $LCTL set_param fail_loc=0x1610
- createmany -o $DIR/$tdir/f 64
+ createmany -o $DIR/$tdir/f $((count + 32))
do_facet ost1 $LCTL set_param fail_loc=0
- echo "stopall to cleanup object cache"
- stopall > /dev/null
- echo "setupall"
- setupall > /dev/null
+ start_full_debug_logging
+
+ # exhaust other pre-created dangling cases
+ count=$(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 || error "(2) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r || error "(2) Fail to start LFSCK for layout!"
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 6 || return 3
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
+ $SHOW_LAYOUT
+ error "(3) unexpected status"
+ }
local repaired=$($SHOW_LAYOUT |
awk '/^repaired_dangling/ { print $2 }')
- [ $repaired -eq 32 ] ||
+ [ $repaired -ge 32 ] ||
error "(4) Fail to repair dangling reference: $repaired"
echo "'ls' should fail because it will not repair dangling by default"
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 6 || return 3
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
+ $SHOW_LAYOUT
+ error "(7) unexpected status"
+ }
- local repaired=$($SHOW_LAYOUT |
+ repaired=$($SHOW_LAYOUT |
awk '/^repaired_dangling/ { print $2 }')
- [ $repaired -eq 32 ] ||
- error "(7) Fail to repair dangling reference: $repaired"
+ [ $repaired -ge 32 ] ||
+ error "(8) Fail to repair dangling reference: $repaired"
echo "'ls' should success after layout LFSCK repairing"
- ls -ail $DIR/$tdir > /dev/null || error "(8) ls should success."
+ ls -ail $DIR/$tdir > /dev/null || error "(9) ls should success."
+ stop_full_debug_logging
}
run_test 14 "LFSCK can repair MDT-object with dangling reference"
echo "the OST-object to back point to the right MDT-object."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
echo "Inject failure stub to make the OST-object to back point to"
do_facet ost1 $LCTL set_param fail_loc=0x1611
dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1
cancel_lru_locks osc
- sync
- sleep 2
do_facet ost1 $LCTL set_param fail_loc=0
- echo "stopall to cleanup object cache"
- stopall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
echo "Trigger layout LFSCK to find out unmatched pairs and fix them"
- $START_LAYOUT || error "(1) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 3 || return 2
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
+ $SHOW_LAYOUT
+ error "(2) unexpected status"
+ }
local repaired=$($SHOW_LAYOUT |
awk '/^repaired_unmatched_pair/ { print $2 }')
echo "MDT-object (the first one)."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
- dd if=/dev/zero of=$DIR/$tdir/guard bs=1K count=1
+ dd if=/dev/zero of=$DIR/$tdir/guard bs=1M count=1
cancel_lru_locks osc
echo "Inject failure stub to make the OST-object to back point to"
do_facet ost1 $LCTL set_param fail_loc=0x1612
dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1
cancel_lru_locks osc
- sync
- sleep 2
do_facet ost1 $LCTL set_param fail_loc=0
- echo "stopall to cleanup object cache"
- stopall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
echo "Trigger layout LFSCK to find out unmatched pairs and fix them"
- $START_LAYOUT || error "(1) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 3 || return 2
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
+ $SHOW_LAYOUT
+ error "(2) unexpected status"
+ }
local repaired=$($SHOW_LAYOUT |
awk '/^repaired_unmatched_pair/ { print $2 }')
echo "MDT-object and update the OST-object's owner information."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1
cancel_lru_locks osc
- sync
- sleep 2
echo "Inject failure stub to skip OST-object owner changing"
#define OBD_FAIL_LFSCK_BAD_OWNER 0x1613
echo "Trigger layout LFSCK to find out inconsistent OST-object owner"
echo "and fix them"
- $START_LAYOUT || error "(1) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 3 || return 2
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
+ $SHOW_LAYOUT
+ error "(2) unexpected status"
+ }
local repaired=$($SHOW_LAYOUT |
awk '/^repaired_inconsistent_owner/ { print $2 }')
echo "MDT-objects."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
echo "Inject failure stub to make two MDT-objects to refernce"
echo "the OST-object"
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
#define OBD_FAIL_LFSCK_MULTIPLE_REF 0x1614
- do_facet $SINGLEMDS $LCTL set_param fail_loc=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 osc
- sync
- sleep 2
- createmany -o $DIR/$tdir/f 1 > /dev/null 2>&1
+ createmany -o $DIR/$tdir/f 1
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
- echo "stopall to cleanup object cache"
- stopall > /dev/null
- echo "setupall"
- setupall > /dev/null
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
echo "$DIR/$tdir/f0 and $DIR/$tdir/guard use the same OST-objects"
local size=$(ls -l $DIR/$tdir/f0 | awk '{ print $5 }')
echo "Trigger layout LFSCK to find out multiple refenced MDT-objects"
echo "and fix them"
- $START_LAYOUT || error "(2) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r || error "(2) Fail to start LFSCK for layout!"
wait_update_facet $SINGLEMDS "$LCTL get_param -n \
mdd.${MDT_DEV}.lfsck_layout |
- awk '/^status/ { print \\\$2 }'" "completed" 3 || return 3
+ awk '/^status/ { print \\\$2 }'" "completed" 6 || {
+ $SHOW_LAYOUT
+ error "(3) unexpected status"
+ }
local repaired=$($SHOW_LAYOUT |
awk '/^repaired_multiple_referenced/ { print $2 }')
run_test 17 "LFSCK can repair multiple references"
test_18a() {
- [ $MDSCOUNT -lt 2 ] &&
- skip "We need at least 2 MDSes for test_18a" && exit 0
-
- [ $OSTCOUNT -lt 2 ] &&
- skip "We need at least 2 OSTs for test_18a" && exit 0
-
echo "#####"
echo "The target MDT-object is there, but related stripe information"
echo "is lost or partly lost. The LFSCK should regenerate the missed"
echo "layout EA entries."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/a1
- $LFS mkdir -i 1 $DIR/$tdir/a2
$LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
- $LFS setstripe -c 2 -i 1 -s 1M $DIR/$tdir/a2
dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
- dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
local saved_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
$LFS path2fid $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a1/f1
- $LFS path2fid $DIR/$tdir/a2/f2
- $LFS getstripe $DIR/$tdir/a2/f2
- sync
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ $LFS mkdir -i 1 $DIR/$tdir/a2
+ $LFS setstripe -c 2 -i 1 -s 1M $DIR/$tdir/a2
+ dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
+ $LFS path2fid $DIR/$tdir/a2/f2
+ $LFS getstripe $DIR/$tdir/a2/f2
+ fi
+
cancel_lru_locks osc
echo "Inject failure, to make the MDT-object lost its layout EA"
#define OBD_FAIL_LFSCK_LOST_STRIPE 0x1615
do_facet mds1 $LCTL set_param fail_loc=0x1615
chown 1.1 $DIR/$tdir/a1/f1
- do_facet mds2 $LCTL set_param fail_loc=0x1615
- chown 1.1 $DIR/$tdir/a2/f2
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ do_facet mds2 $LCTL set_param fail_loc=0x1615
+ chown 1.1 $DIR/$tdir/a2/f2
+ fi
+
sync
sleep 2
+
do_facet mds1 $LCTL set_param fail_loc=0
- do_facet mds2 $LCTL set_param fail_loc=0
+ if [ $MDSCOUNT -ge 2 ]; then
+ do_facet mds2 $LCTL set_param fail_loc=0
+ fi
- echo "stopall to cleanup object cache"
- stopall > /dev/null
- echo "setupall"
- setupall > /dev/null
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
echo "The file size should be incorrect since layout EA is lost"
local cur_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
[ "$cur_size" != "$saved_size" ] ||
error "(1) Expect incorrect file1 size"
- cur_size=$(ls -il $DIR/$tdir/a2/f2 | awk '{ print $6 }')
- [ "$cur_size" != "$saved_size" ] ||
- error "(2) Expect incorrect file2 size"
+ if [ $MDSCOUNT -ge 2 ]; then
+ cur_size=$(ls -il $DIR/$tdir/a2/f2 | awk '{ print $6 }')
+ [ "$cur_size" != "$saved_size" ] ||
+ error "(2) Expect incorrect file2 size"
+ fi
echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
- $START_LAYOUT -o || error "(3) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r -o || error "(3) Fail to start LFSCK for layout!"
for k in $(seq $MDSCOUNT); do
# The LFSCK status query internal is 30 seconds. For the case
error "(5) OST${k} Expect 'completed', but got '$cur_status'"
done
- for k in 1 2; do
- local repaired=$(do_facet mds${k} $LCTL get_param -n \
- mdd.$(facet_svc mds${k}).lfsck_layout |
+ local repaired=$(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"
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_layout |
awk '/^repaired_orphan/ { print $2 }')
- [ $repaired -eq ${k} ] ||
- error "(6) Expect ${k} fixed on mds${k}, but got: $repaired"
- done
+ [ $repaired -eq 2 ] ||
+ error "(6.2) Expect 2 fixed on mds2, but got: $repaired"
+ fi
$LFS path2fid $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a1/f1
- $LFS path2fid $DIR/$tdir/a2/f2
- $LFS getstripe $DIR/$tdir/a2/f2
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ $LFS path2fid $DIR/$tdir/a2/f2
+ $LFS getstripe $DIR/$tdir/a2/f2
+ fi
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 $cur_size"
- cur_size=$(ls -il $DIR/$tdir/a2/f2 | awk '{ print $6 }')
- [ "$cur_size" == "$saved_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"
+ fi
}
run_test 18a "Find out orphan OST-object and repair it (1)"
test_18b() {
- [ $MDSCOUNT -lt 2 ] &&
- skip "We need at least 2 MDSes for test_18b" && exit 0
-
- [ $OSTCOUNT -lt 2 ] &&
- skip "We need at least 2 OSTs for test_18b" && exit 0
-
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"
echo "can move it back to normal namespace manually."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/a1
- $LFS mkdir -i 1 $DIR/$tdir/a2
$LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
- $LFS setstripe -c 2 -i 1 -s 1M $DIR/$tdir/a2
dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
- dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
local saved_size=$(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
- local fid2=$($LFS path2fid $DIR/$tdir/a2/f2)
- echo ${fid2}
- $LFS getstripe $DIR/$tdir/a2/f2
- sync
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ $LFS mkdir -i 1 $DIR/$tdir/a2
+ $LFS setstripe -c 2 -i 1 -s 1M $DIR/$tdir/a2
+ dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
+ fid2=$($LFS path2fid $DIR/$tdir/a2/f2)
+ echo ${fid2}
+ $LFS getstripe $DIR/$tdir/a2/f2
+ fi
+
cancel_lru_locks osc
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/a1/f1
- do_facet mds2 $LCTL set_param fail_loc=0x1616
- rm -f $DIR/$tdir/a2/f2
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ do_facet mds2 $LCTL set_param fail_loc=0x1616
+ rm -f $DIR/$tdir/a2/f2
+ fi
+
sync
sleep 2
+
do_facet mds1 $LCTL set_param fail_loc=0
- do_facet mds2 $LCTL set_param fail_loc=0
+ if [ $MDSCOUNT -ge 2 ]; then
+ do_facet mds2 $LCTL set_param fail_loc=0
+ fi
- echo "stopall to cleanup object cache"
- stopall > /dev/null
- echo "setupall"
- setupall > /dev/null
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
- $START_LAYOUT -o || error "(1) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r -o || error "(1) Fail to start LFSCK for layout!"
for k in $(seq $MDSCOUNT); do
# The LFSCK status query internal is 30 seconds. For the case
error "(3) OST${k} Expect 'completed', but got '$cur_status'"
done
- for k in 1 2; do
- local repaired=$(do_facet mds${k} $LCTL get_param -n \
- mdd.$(facet_svc mds${k}).lfsck_layout |
- awk '/^repaired_orphan/ { print $2 }')
- [ $repaired -eq ${k} ] ||
- error "(4) Expect ${k} fixed on mds${k}, but got: $repaired"
- done
+ local repaired=$(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"
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_layout |
+ awk '/^repaired_orphan/ { print $2 }')
+ [ $repaired -eq 2 ] ||
+ error "(4.2) Expect 2 fixed on mds2, but got: $repaired"
+ fi
echo "Move the files from ./lustre/lost+found/MDTxxxx to namespace"
- mv $MOUNT/.lustre/lost+found/MDT0000/R-${fid1} $DIR/$tdir/a1/f1 ||
- error "(5) Fail to move $MOUNT/.lustre/lost+found/MDT0000/R-${fid1}"
+ 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"
- mv $MOUNT/.lustre/lost+found/MDT0001/R-${fid2} $DIR/$tdir/a2/f2 ||
- error "(6) Fail to move $MOUNT/.lustre/lost+found/MDT0001/R-${fid2}"
+ if [ $MDSCOUNT -ge 2 ]; then
+ local name=$MOUNT/.lustre/lost+found/MDT0001/${fid2}-R-0
+ mv $name $DIR/$tdir/a2/f2 || error "(6) Fail to move $name"
+ fi
$LFS path2fid $DIR/$tdir/a1/f1
$LFS getstripe $DIR/$tdir/a1/f1
- $LFS path2fid $DIR/$tdir/a2/f2
- $LFS getstripe $DIR/$tdir/a2/f2
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ $LFS path2fid $DIR/$tdir/a2/f2
+ $LFS getstripe $DIR/$tdir/a2/f2
+ fi
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 $cur_size"
- cur_size=$(ls -il $DIR/$tdir/a2/f2 | awk '{ print $6 }')
- [ "$cur_size" == "$saved_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"
+ fi
}
run_test 18b "Find out orphan OST-object and repair it (2)"
test_18c() {
- [ $MDSCOUNT -lt 2 ] &&
- skip "We need at least 2 MDSes for test_18c" && exit 0
-
- [ $OSTCOUNT -lt 2 ] &&
- skip "We need at least 2 OSTs for test_18c" && exit 0
-
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 "
echo "directory .lustre/lost+found/MDTxxxx."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$LFS mkdir -i 0 $DIR/$tdir/a1
- $LFS mkdir -i 1 $DIR/$tdir/a2
$LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
- $LFS setstripe -c 2 -i 1 -s 1M $DIR/$tdir/a2
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_facet ost2 $LCTL set_param fail_loc=0x1617
dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
- dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
$LFS getstripe $DIR/$tdir/a1/f1
- $LFS getstripe $DIR/$tdir/a2/f2
- sync
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ $LFS mkdir -i 1 $DIR/$tdir/a2
+ $LFS setstripe -c 2 -i 1 -s 1M $DIR/$tdir/a2
+ do_facet ost2 $LCTL set_param fail_loc=0x1617
+ dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
+ $LFS getstripe $DIR/$tdir/a2/f2
+ fi
+
cancel_lru_locks osc
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/a1/f1
- do_facet mds2 $LCTL set_param fail_loc=0x1616
- rm -f $DIR/$tdir/a2/f2
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ do_facet mds2 $LCTL set_param fail_loc=0x1616
+ rm -f $DIR/$tdir/a2/f2
+ fi
+
sync
sleep 2
+
do_facet mds1 $LCTL set_param fail_loc=0
- do_facet mds2 $LCTL set_param fail_loc=0
+ if [ $MDSCOUNT -ge 2 ]; then
+ do_facet mds2 $LCTL set_param fail_loc=0
+ fi
- echo "stopall to cleanup object cache"
- stopall > /dev/null
- echo "setupall"
- setupall > /dev/null
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
- $START_LAYOUT -o || error "(1) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r -o || error "(1) Fail to start LFSCK for layout!"
for k in $(seq $MDSCOUNT); do
# The LFSCK status query internal is 30 seconds. For the case
error "(3) OST${k} Expect 'completed', but got '$cur_status'"
done
+ if [ $MDSCOUNT -ge 2 ]; then
+ expected=3
+ else
+ expected=1
+ fi
+
local repaired=$(do_facet mds1 $LCTL get_param -n \
mdd.$(facet_svc mds1).lfsck_layout |
awk '/^repaired_orphan/ { print $2 }')
- [ $repaired -eq 3 ] ||
- error "(4) Expect 3 fixed on mds1, but got: $repaired"
-
- repaired=$(do_facet mds2 $LCTL get_param -n \
- mdd.$(facet_svc mds2).lfsck_layout |
- awk '/^repaired_orphan/ { print $2 }')
- [ $repaired -eq 0 ] ||
- error "(5) Expect 0 fixed on mds2, but got: $repaired"
+ [ $repaired -eq $expected ] ||
+ error "(4) Expect $expected fixed on mds1, but got: $repaired"
+
+ if [ $MDSCOUNT -ge 2 ]; then
+ repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_layout |
+ awk '/^repaired_orphan/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(5) Expect 0 fixed on mds2, but got: $repaired"
+ fi
echo "There should be some stub under .lustre/lost+found/MDT0001/"
- ls -ail $MOUNT/.lustre/lost+found/MDT0001/N-* &&
+ ls -ail $MOUNT/.lustre/lost+found/MDT0001/*-N-0 &&
error "(6) .lustre/lost+found/MDT0001/ should be empty"
echo "There should be some stub under .lustre/lost+found/MDT0000/"
- ls -ail $MOUNT/.lustre/lost+found/MDT0000/N-* ||
+ ls -ail $MOUNT/.lustre/lost+found/MDT0000/*-N-0 ||
error "(7) .lustre/lost+found/MDT0000/ should not be empty"
}
run_test 18c "Find out orphan OST-object and repair it (3)"
echo "OST-object."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir/a1
+ check_mount_and_prep
+ mkdir $DIR/$tdir/a1
$LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
echo "guard" > $DIR/$tdir/a1/f1
echo "foo" > $DIR/$tdir/a1/f2
$LFS getstripe $DIR/$tdir/a1/f1
$LFS path2fid $DIR/$tdir/a1/f2
$LFS getstripe $DIR/$tdir/a1/f2
- sync
cancel_lru_locks osc
echo "Inject failure to make $DIR/$tdir/a1/f1 and $DIR/$tdir/a1/f2"
[ "$cur_size" != "$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 -o -c || error "(2) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r -o -c || error "(2) Fail to start LFSCK for layout!"
+
+ wait_update_facet mds1 "$LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "scanning-phase2" 6 ||
+ error "(3.0) MDS1 is not the expected 'scanning-phase2'"
+
+ do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
for k in $(seq $MDSCOUNT); do
# The LFSCK status query internal is 30 seconds. For the case
[ "$cur_size" == "$saved_size" ] ||
error "(6) Expect file2 size $saved_size, but got $cur_size"
- echo "There should be some stub under .lustre/lost+found/MDT0000/"
- ls -ail $MOUNT/.lustre/lost+found/MDT0000/ &&
- error "(7) .lustre/lost+found/MDT0000/ should be empty"
-
echo "The LFSCK should find back the original data."
cat $DIR/$tdir/a1/f2
$LFS path2fid $DIR/$tdir/a1/f2
echo "old orphan OST-object."
echo "#####"
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir/a1
+ check_mount_and_prep
+ mkdir $DIR/$tdir/a1
$LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
echo "guard" > $DIR/$tdir/a1/f1
echo "foo" > $DIR/$tdir/a1/f2
$LFS getstripe $DIR/$tdir/a1/f1
$LFS path2fid $DIR/$tdir/a1/f2
$LFS getstripe $DIR/$tdir/a1/f2
- sync
cancel_lru_locks osc
echo "Inject failure to make $DIR/$tdir/a1/f1 and $DIR/$tdir/a1/f2"
error "(1) Expect incorrect file2 size"
#define OBD_FAIL_LFSCK_DELAY3 0x1602
- do_facet $SINGLEMDS $LCTL set_param fail_val=10
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1602
+ do_facet $SINGLEMDS $LCTL set_param fail_val=10 fail_loc=0x1602
echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
- $START_LAYOUT -o -c || error "(2) Fail to start LFSCK for layout!"
+ $START_LAYOUT -r -o -c || error "(2) Fail to start LFSCK for layout!"
wait_update_facet mds1 "$LCTL get_param -n \
mdd.$(facet_svc mds1).lfsck_layout |
echo "Write new data to f2 to modify the new created OST-object."
echo "dummy" >> $DIR/$tdir/a1/f2
- do_facet $SINGLEMDS $LCTL set_param fail_val=0
- do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+ do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
for k in $(seq $MDSCOUNT); do
# The LFSCK status query internal is 30 seconds. For the case
error "(6) Expect 1 orphan has been fixed, but got: $repaired"
echo "There should be stub file under .lustre/lost+found/MDT0000/"
- local cname=$(ls $MOUNT/.lustre/lost+found/MDT0000/C-*)
+ local cname=$(ls $MOUNT/.lustre/lost+found/MDT0000/*-C-0)
[ ! -z $name ] ||
error "(7) .lustre/lost+found/MDT0000/ should not be empty"
run_test 18e "Find out orphan OST-object and repair it (5)"
test_19a() {
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
echo "foo" > $DIR/$tdir/a0
echo "guard" > $DIR/$tdir/a1
-
cancel_lru_locks osc
- umount_client $MOUNT || error "(1) Fail to stop client!"
- mount_client $MOUNT || error "(2) Fail to start client!"
echo "Inject failure, then client will offer wrong parent FID when read"
do_facet ost1 $LCTL set_param -n \
run_test 19a "OST-object inconsistency self detect"
test_19b() {
- echo "stopall"
- stopall > /dev/null
- echo "formatall"
- formatall > /dev/null
- echo "setupall"
- setupall > /dev/null
-
- mkdir -p $DIR/$tdir
+ check_mount_and_prep
$LFS setstripe -c 1 -i 0 $DIR/$tdir
echo "Inject failure stub to make the OST-object to back point to"
do_facet ost1 $LCTL set_param fail_loc=0x1611
echo "foo" > $DIR/$tdir/f0
cancel_lru_locks osc
- sync
- sleep 2
do_facet ost1 $LCTL set_param fail_loc=0
echo "Nothing should be fixed since self detect and repair is disabled"
}
run_test 19b "OST-object inconsistency self repair"
+test_20() {
+ [ $OSTCOUNT -lt 2 ] &&
+ skip "The test needs at least 2 OSTs" && return
+
+ echo "#####"
+ 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."
+
+ echo "For old client, even though it cannot access the file with"
+ echo "LOV EA hole, it should not cause the system crash."
+ echo "#####"
+
+ check_mount_and_prep
+ $LFS mkdir -i 0 $DIR/$tdir/a1
+ if [ $OSTCOUNT -gt 2 ]; then
+ $LFS setstripe -c 3 -i 0 -s 1M $DIR/$tdir/a1
+ bcount=513
+ else
+ $LFS setstripe -c 2 -i 0 -s 1M $DIR/$tdir/a1
+ bcount=257
+ fi
+
+ # 256 blocks on the stripe0.
+ # 1 block on the stripe1 for 2 OSTs case.
+ # 256 blocks on the stripe1 for other cases.
+ # 1 block on the stripe2 if OSTs > 2
+ dd if=/dev/zero of=$DIR/$tdir/a1/f0 bs=4096 count=$bcount
+ dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=4096 count=$bcount
+ dd if=/dev/zero of=$DIR/$tdir/a1/f2 bs=4096 count=$bcount
+
+ local fid0=$($LFS path2fid $DIR/$tdir/a1/f0)
+ local fid1=$($LFS path2fid $DIR/$tdir/a1/f1)
+ local fid2=$($LFS path2fid $DIR/$tdir/a1/f2)
+
+ echo ${fid0}
+ $LFS getstripe $DIR/$tdir/a1/f0
+ echo ${fid1}
+ $LFS getstripe $DIR/$tdir/a1/f1
+ echo ${fid2}
+ $LFS getstripe $DIR/$tdir/a1/f2
+
+ if [ $OSTCOUNT -gt 2 ]; then
+ dd if=/dev/zero of=$DIR/$tdir/a1/f3 bs=4096 count=$bcount
+ fid3=$($LFS path2fid $DIR/$tdir/a1/f3)
+ echo ${fid3}
+ $LFS getstripe $DIR/$tdir/a1/f3
+ fi
+
+ cancel_lru_locks osc
+
+ echo "Inject failure..."
+ echo "To simulate f0 lost MDT-object"
+ #define OBD_FAIL_LFSCK_LOST_MDTOBJ 0x1616
+ do_facet mds1 $LCTL set_param fail_loc=0x1616
+ rm -f $DIR/$tdir/a1/f0
+
+ echo "To simulate f1 lost MDT-object and OST-object0"
+ #define OBD_FAIL_LFSCK_LOST_SPEOBJ 0x161a
+ do_facet mds1 $LCTL set_param fail_loc=0x161a
+ rm -f $DIR/$tdir/a1/f1
+
+ echo "To simulate f2 lost MDT-object and OST-object1"
+ do_facet mds1 $LCTL set_param fail_val=1
+ rm -f $DIR/$tdir/a1/f2
+
+ if [ $OSTCOUNT -gt 2 ]; then
+ echo "To simulate f3 lost MDT-object and OST-object2"
+ do_facet mds1 $LCTL set_param fail_val=2
+ rm -f $DIR/$tdir/a1/f3
+ fi
+
+ umount_client $MOUNT
+ sync
+ 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 $MDSCOUNT); do
+ # The LFSCK status query internal is 30 seconds. For the case
+ # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
+ # time to guarantee the status sync up.
+ wait_update_facet mds${k} "$LCTL get_param -n \
+ mdd.$(facet_svc mds${k}).lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ error "(2) MDS${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 }')
+ if [ $OSTCOUNT -gt 2 ]; then
+ [ $repaired -eq 9 ] ||
+ error "(4.1) Expect 9 fixed on mds1, but got: $repaired"
+ else
+ [ $repaired -eq 4 ] ||
+ error "(4.2) Expect 4 fixed on mds1, but got: $repaired"
+ fi
+
+ mount_client $MOUNT || error "(5.0) Fail to start client!"
+
+ LOV_PATTERN_F_HOLE=0x40000000
+
+ #
+ # ${fid0}-R-0 is the old f0
+ #
+ local name="$MOUNT/.lustre/lost+found/MDT0000/${fid0}-R-0"
+ echo "Check $name, which is the old f0"
+
+ $LFS getstripe -v $name || error "(5.1) cannot getstripe on $name"
+
+ local pattern=0x$($LFS getstripe -L $name)
+ [[ $((pattern & LOV_PATTERN_F_HOLE)) -eq 0 ]] ||
+ error "(5.2) NOT expect pattern flag hole, but got $pattern"
+
+ local stripes=$($LFS getstripe -c $name)
+ if [ $OSTCOUNT -gt 2 ]; then
+ [ $stripes -eq 3 ] ||
+ error "(5.3.1) expect the stripe count is 3, but got $stripes"
+ else
+ [ $stripes -eq 2 ] ||
+ error "(5.3.2) expect the stripe count is 2, but got $stripes"
+ fi
+
+ local size=$(stat $name | awk '/Size:/ { print $2 }')
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(5.4) expect the size $((4096 * $bcount)), but got $size"
+
+ cat $name > /dev/null || error "(5.5) cannot read $name"
+
+ echo "dummy" >> $name || error "(5.6) cannot write $name"
+
+ chown $RUNAS_ID:$RUNAS_GID $name || error "(5.7) cannot chown on $name"
+
+ touch $name || error "(5.8) cannot touch $name"
+
+ rm -f $name || error "(5.9) cannot unlink $name"
+
+ #
+ # ${fid1}-R-0 contains the old f1's stripe1 (and stripe2 if OSTs > 2)
+ #
+ name="$MOUNT/.lustre/lost+found/MDT0000/${fid1}-R-0"
+ if [ $OSTCOUNT -gt 2 ]; then
+ echo "Check $name, it contains the old f1's stripe1 and stripe2"
+ else
+ echo "Check $name, it contains the old f1's stripe1"
+ fi
+
+ $LFS getstripe -v $name || error "(6.1) cannot getstripe on $name"
+
+ pattern=0x$($LFS getstripe -L $name)
+ [[ $((pattern & LOV_PATTERN_F_HOLE)) -ne 0 ]] ||
+ error "(6.2) expect pattern flag hole, but got $pattern"
+
+ stripes=$($LFS getstripe -c $name)
+ if [ $OSTCOUNT -gt 2 ]; then
+ [ $stripes -eq 3 ] ||
+ error "(6.3.1) expect the stripe count is 3, but got $stripes"
+ else
+ [ $stripes -eq 2 ] ||
+ error "(6.3.2) expect the stripe count is 2, but got $stripes"
+ fi
+
+ size=$(stat $name | awk '/Size:/ { print $2 }')
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(6.4) expect the size $((4096 * $bcount)), but got $size"
+
+ cat $name > /dev/null && error "(6.5) 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)
+
+ # stripe0 is dummy
+ [ $failures -eq 256 ] ||
+ error "(6.6) expect 256 IO failures, but get $failures"
+
+ size=$(stat $DIR/$tdir/dump | awk '/Size:/ { print $2 }')
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(6.7) expect the size $((4096 * $bcount)), but got $size"
+
+ dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 &&
+ error "(6.8) write to the LOV EA hole should fail"
+
+ dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 seek=300 ||
+ error "(6.9) write to normal stripe should NOT fail"
+
+ echo "foo" >> $name && error "(6.10) append write $name should fail"
+
+ chown $RUNAS_ID:$RUNAS_GID $name || error "(6.11) cannot chown on $name"
+
+ touch $name || error "(6.12) cannot touch $name"
+
+ rm -f $name || error "(6.13) cannot unlink $name"
+
+ #
+ # ${fid2}-R-0 it contains the old f2's stripe0 (and stripe2 if OSTs > 2)
+ #
+ name="$MOUNT/.lustre/lost+found/MDT0000/${fid2}-R-0"
+ if [ $OSTCOUNT -gt 2 ]; then
+ echo "Check $name, it contains the old f2's stripe0 and stripe2"
+ else
+ echo "Check $name, it contains the old f2's stripe0"
+ fi
+
+ $LFS getstripe -v $name || error "(7.1) cannot getstripe on $name"
+
+ pattern=0x$($LFS getstripe -L $name)
+ stripes=$($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"
+
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(7.4.1) expect size $((4096 * $bcount)), but got $size"
+
+ cat $name > /dev/null &&
+ error "(7.5.1) 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)
+ # stripe1 is dummy
+ [ $failures -eq 256 ] ||
+ error "(7.6) expect 256 IO failures, but get $failures"
+
+ size=$(stat $DIR/$tdir/dump | awk '/Size:/ { print $2 }')
+ [ $size -eq $((4096 * $bcount)) ] ||
+ error "(7.7) expect the size $((4096 * $bcount)), but got $size"
+
+ dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 \
+ seek=300 && error "(7.8.0) write to the LOV EA hole should fail"
+
+ dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 ||
+ error "(7.8.1) write to normal stripe should NOT fail"
+
+ echo "foo" >> $name &&
+ error "(7.8.3) append write $name should fail"
+
+ chown $RUNAS_ID:$RUNAS_GID $name ||
+ error "(7.9.1) cannot chown on $name"
+
+ 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"
+
+ # 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"
+
+ echo "dummy" >> $name || error "(7.8.2) cannot write $name"
+
+ chown $RUNAS_ID:$RUNAS_GID $name ||
+ error "(7.9.2) cannot chown on $name"
+
+ touch $name || error "(7.10.2) cannot touch $name"
+ fi
+
+ rm -f $name || error "(7.11) cannot unlink $name"
+
+ [ $OSTCOUNT -le 2 ] && return
+
+ #
+ # ${fid3}-R-0 should contains the old f3's stripe0 and stripe1
+ #
+ name="$MOUNT/.lustre/lost+found/MDT0000/${fid3}-R-0"
+ echo "Check $name, which contains the old f3's stripe0 and stripe1"
+
+ $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 $pattern"
+
+ stripes=$($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 $stripes"
+
+ size=$(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"
+
+test_21() {
+ [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.59) ]] &&
+ skip "ignore the test if MDS is older than 2.5.59" && exit 0
+
+ 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"
+
$LCTL set_param debug=-lfsck > /dev/null || true
# restore MDS/OST size
MDSSIZE=${SAVED_MDSSIZE}
OSTSIZE=${SAVED_OSTSIZE}
+OSTCOUNT=${SAVED_OSTCOUNT}
# cleanup the system at last
formatall