. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
init_logging
-# remove the check when ZFS backend iteration is ready
-[ $(facet_fstype $SINGLEMDS) != "ldiskfs" ] &&
- skip "test LFSCK only for ldiskfs" && exit 0
-[ $(facet_fstype ost1) != ldiskfs ] &&
- skip "test LFSCK only for ldiskfs" && exit 0
-
require_dsh_mds || exit 0
MCREATE=${MCREATE:-mcreate}
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"
+ ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 12 13 14 15 16 17 18 19"
build_test_filter
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 "(9) Expect 'completed', but got '$STATUS'"
+ wait_update_facet $SINGLEMDS \
+ "$LCTL get_param -n mdd.${MDT_DEV}.lfsck_namespace | \
+ awk '/^status/ { print \\\$2 }'" "completed" 20 || \
+ error "(9) unexpected status"
local repaired=$($SHOW_NAMESPACE |
awk '/^updated_phase1/ { print $2 }')
local scanned1=$($SHOW_NAMESPACE | awk '/^success_count/ { print $2 }')
$START_NAMESPACE -r || error "(11) Fail to reset LFSCK!"
- sleep 3
-
- STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
- [ "$STATUS" == "completed" ] ||
- error "(12) Expect 'completed', but got '$STATUS'"
+ wait_update_facet $SINGLEMDS \
+ "$LCTL get_param -n mdd.${MDT_DEV}.lfsck_namespace | \
+ awk '/^status/ { print \\\$2 }'" "completed" 20 || \
+ error "(12) unexpected status"
local scanned2=$($SHOW_NAMESPACE | awk '/^success_count/ { print $2 }')
[ $((scanned1 + 1)) -eq $scanned2 ] ||
run_test 0 "Control LFSCK manually"
test_1a() {
+ [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
+ skip "OI Scrub not implemented for ZFS" && return
+
lfsck_prep 1 1
echo "start $SINGLEMDS"
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
test_1b()
{
+ [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
+ skip "OI Scrub not implemented for ZFS" && return
+
lfsck_prep 1 1
echo "start $SINGLEMDS"
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
test_4()
{
+ [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
+ skip "OI Scrub not implemented for ZFS" && return
+
lfsck_prep 3 3
mds_backup_restore $SINGLEMDS || error "(1) Fail to backup/restore!"
echo "start $SINGLEMDS with disabling OI scrub"
test_5()
{
+ [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
+ skip "OI Scrub not implemented for ZFS" && return
+
lfsck_prep 1 1 1
mds_backup_restore $SINGLEMDS 1 || error "(1) Fail to backup/restore!"
echo "start $SINGLEMDS with disabling OI scrub"
test_10()
{
+ [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
+ skip "lookup(..)/linkea on ZFS issue" && return
+
lfsck_prep 1 1
echo "start $SINGLEMDS"
start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
# step 1: local mount
mount_fstype ost${ost} || return 1
# step 2: remove the specified LAST_ID
- ${rcmd} rm -fv $(facet_mntpt ost${ost})/O/${idx}/LAST_ID
+ ${rcmd} rm -fv $(facet_mntpt ost${ost})/O/${idx}/{LAST_ID,d0/0}
# step 3: umount
unmount_fstype ost${ost} || return 2
}
echo "setupall"
setupall > /dev/null
+ mkdir -p $DIR/$tdir
+
echo "All the LFSCK targets should be in 'init' status."
for k in $(seq $MDSCOUNT); do
local STATUS=$(do_facet mds${k} $LCTL get_param -n \
[ "$STATUS" == "init" ] ||
error "(1) MDS${k} Expect 'init', but got '$STATUS'"
- $LFS mkdir -i $((k - 1)) $DIR/${k}
- createmany -o $DIR/${k}/f 100
+ $LFS mkdir -i $((k - 1)) $DIR/$tdir/${k}
+ createmany -o $DIR/$tdir/${k}/f 100
+ 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!"
+
+ 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
+
+ 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
+
+ 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
- echo "Trigger LFSCK on all targets by single command (limited speed)."
+ 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 10 || error "(2) Fail to start LFSCK on all devices!"
+ -s 1 || error "(8) Fail to start LFSCK on all devices!"
echo "All the LFSCK targets should be in 'scanning-phase1' status."
for k in $(seq $MDSCOUNT); do
mdd.$(facet_svc mds${k}).lfsck_layout |
awk '/^status/ { print $2 }')
[ "$STATUS" == "scanning-phase1" ] ||
- error "(3) MDS${k} Expect 'scanning-phase1', but got '$STATUS'"
+ error "(9) MDS${k} Expect 'scanning-phase1', but got '$STATUS'"
done
echo "Stop layout 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!"
+ error "(10) Fail to stop LFSCK on all devices!"
echo "All the LFSCK targets should be in 'stopped' status."
for k in $(seq $MDSCOUNT); do
mdd.$(facet_svc mds${k}).lfsck_layout |
awk '/^status/ { print $2 }')
[ "$STATUS" == "stopped" ] ||
- error "(5) MDS${k} Expect 'stopped', but got '$STATUS'"
+ error "(11) MDS${k} Expect 'stopped', but got '$STATUS'"
done
for k in $(seq $OSTCOUNT); do
obdfilter.$(facet_svc ost${k}).lfsck_layout |
awk '/^status/ { print $2 }')
[ "$STATUS" == "stopped" ] ||
- error "(6) OST${k} Expect 'stopped', but got '$STATUS'"
+ error "(12) OST${k} Expect 'stopped', but got '$STATUS'"
done
- echo "Re-trigger LFSCK on all targets by single command (full speed)."
+ echo "Re-start layout LFSCK on all targets by single command (-s 0)."
do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -t layout -A \
- -s 0 || error "(7) Fail to start LFSCK on all devices!"
+ -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 $MDSCOUNT); do
wait_update_facet mds${k} "$LCTL get_param -n \
mdd.$(facet_svc mds${k}).lfsck_layout |
awk '/^status/ { print \\\$2 }'" "completed" 32 ||
- error "(8) MDS${k} is not the expected 'completed'"
+ error "(14) MDS${k} is not the expected 'completed'"
done
}
run_test 12 "single command to trigger LFSCK on all devices"
}
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
+ $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
+ 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
+ sync
+ sleep 2
+ do_facet mds1 $LCTL set_param fail_loc=0
+ do_facet mds2 $LCTL set_param fail_loc=0
+
+ echo "stopall to cleanup object cache"
+ stopall > /dev/null
+ echo "setupall"
+ setupall > /dev/null
+
+ echo "The file size should be incorrect since layout EA is lost"
+ local cur_size=$(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"
+
+ echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+ $START_LAYOUT -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
+
+ 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 "(6) Expect ${k} fixed on mds${k}, but got: $repaired"
+ done
+
+ $LFS path2fid $DIR/$tdir/a1/f1
+ $LFS getstripe $DIR/$tdir/a1/f1
+ $LFS path2fid $DIR/$tdir/a2/f2
+ $LFS getstripe $DIR/$tdir/a2/f2
+
+ 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" ] ||
+ error "(8) Expect file2 size $saved_size, but got $cur_size"
+}
+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
+ $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
+ 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
+ sync
+ sleep 2
+ do_facet mds1 $LCTL set_param fail_loc=0
+ do_facet mds2 $LCTL set_param fail_loc=0
+
+ echo "stopall to cleanup object cache"
+ stopall > /dev/null
+ echo "setupall"
+ setupall > /dev/null
+
+ echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+ $START_LAYOUT -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${k}).lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ error "(2) MDS${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
+
+ 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
+
+ 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 $MOUNT/.lustre/lost+found/MDT0001/R-${fid2} $DIR/$tdir/a2/f2 ||
+ error "(6) Fail to move $MOUNT/.lustre/lost+found/MDT0001/R-${fid2}"
+
+ $LFS path2fid $DIR/$tdir/a1/f1
+ $LFS getstripe $DIR/$tdir/a1/f1
+ $LFS path2fid $DIR/$tdir/a2/f2
+ $LFS getstripe $DIR/$tdir/a2/f2
+
+ 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" ] ||
+ error "(8) Expect file2 size $saved_size, but got $cur_size"
+}
+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
+ $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
+ 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
+ sync
+ sleep 2
+ do_facet mds1 $LCTL set_param fail_loc=0
+ do_facet mds2 $LCTL set_param fail_loc=0
+
+ echo "stopall to cleanup object cache"
+ stopall > /dev/null
+ echo "setupall"
+ setupall > /dev/null
+
+ echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+ $START_LAYOUT -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${k}).lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ error "(2) MDS${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 }')
+ [ $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"
+
+ echo "There should be some stub under .lustre/lost+found/MDT0001/"
+ ls -ail $MOUNT/.lustre/lost+found/MDT0001/N-* &&
+ 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-* ||
+ error "(7) .lustre/lost+found/MDT0000/ should not be empty"
+}
+run_test 18c "Find out orphan OST-object and repair it (3)"
+
+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 "#####"
+
+ echo "stopall"
+ stopall > /dev/null
+ echo "formatall"
+ formatall > /dev/null
+ echo "setupall"
+ setupall > /dev/null
+
+ mkdir -p $DIR/$tdir/a1
+ $LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
+ echo "guard" > $DIR/$tdir/a1/f1
+ echo "foo" > $DIR/$tdir/a1/f2
+ local saved_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
+ $LFS path2fid $DIR/$tdir/a1/f1
+ $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"
+ 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
+
+ #define OBD_FAIL_LFSCK_CHANGE_STRIPE 0x1618
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1618
+ chown 1.1 $DIR/$tdir/a1/f2
+ rm -f $DIR/$tdir/a1/f1
+ sync
+ sleep 2
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ echo "stopall to cleanup object cache"
+ stopall > /dev/null
+ echo "setupall"
+ setupall > /dev/null
+
+ echo "The file size should be incorrect since dangling referenced"
+ local cur_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
+ [ "$cur_size" != "$saved_size" ] ||
+ error "(1) Expect incorrect file2 size"
+
+ echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+ $START_LAYOUT -o || error "(2) Fail to start LFSCK for layout!"
+
+ for k in $(seq $MDSCOUNT); do
+ # The LFSCK status query internal is 30 seconds. For the case
+ # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
+ # time to guarantee the status sync up.
+ wait_update_facet mds${k} "$LCTL get_param -n \
+ mdd.$(facet_svc mds${k}).lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ error "(3) MDS${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 "(4) OST${k} Expect 'completed', but got '$cur_status'"
+ done
+
+ 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: $repaired"
+
+ echo "The file size should be correct after layout LFSCK scanning"
+ cur_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
+ [ "$cur_size" == "$saved_size" ] ||
+ error "(6) Expect file2 size $saved_size, but got $cur_size"
+
+ 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
+ $LFS getstripe $DIR/$tdir/a1/f2
+}
+run_test 18d "Find out orphan OST-object and repair it (4)"
+
+test_18e() {
+ 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 been modified by others. To keep the"
+ echo "new data, the LFSCK will create a new file to refernece this"
+ 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
+ $LFS setstripe -c 1 -i 0 -s 1M $DIR/$tdir/a1
+ echo "guard" > $DIR/$tdir/a1/f1
+ echo "foo" > $DIR/$tdir/a1/f2
+ local saved_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
+ $LFS path2fid $DIR/$tdir/a1/f1
+ $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"
+ 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
+
+ #define OBD_FAIL_LFSCK_CHANGE_STRIPE 0x1618
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1618
+ chown 1.1 $DIR/$tdir/a1/f2
+ rm -f $DIR/$tdir/a1/f1
+ sync
+ sleep 2
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+
+ echo "stopall to cleanup object cache"
+ stopall > /dev/null
+ echo "setupall"
+ setupall > /dev/null
+
+ echo "The file size should be incorrect since dangling referenced"
+ local cur_size=$(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
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1602
+
+ echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
+ $START_LAYOUT -o || error "(2) Fail to start LFSCK for layout!"
+
+ wait_update_facet mds1 "$LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_layout |
+ awk '/^status/ { print \\\$2 }'" "scanning-phase2" 6 ||
+ error "(3) MDS1 is not the expected 'scanning-phase2'"
+
+ # to guarantee all updates are synced.
+ sync
+ sleep 2
+
+ 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
+
+ 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 $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: $repaired"
+
+ echo "There should be stub file under .lustre/lost+found/MDT0000/"
+ local cname=$(ls $MOUNT/.lustre/lost+found/MDT0000/C-*)
+ [ ! -z $name ] ||
+ error "(7) .lustre/lost+found/MDT0000/ should not be empty"
+
+ echo "The stub file should keep the original f2 data"
+ cur_size=$(ls -il $cname | awk '{ print $6 }')
+ [ "$cur_size" == "$saved_size" ] ||
+ error "(8) Expect file2 size $saved_size, but got $cur_size"
+
+ cat $cname
+ $LFS path2fid $cname
+ $LFS getstripe $cname
+
+ echo "The f2 should contains new data."
+ cat $DIR/$tdir/a1/f2
+ $LFS path2fid $DIR/$tdir/a1/f2
+ $LFS getstripe $DIR/$tdir/a1/f2
+}
+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
+ $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 $LCTL set_param -n \
+ obdfilter.${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!"
+ $LCTL set_param fail_loc=0
+}
+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
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir
+
+ echo "Inject failure stub to make the OST-object to back point to"
+ echo "non-exist MDT-object"
+
+ #define OBD_FAIL_LFSCK_UNMATCHED_PAIR1 0x1611
+ do_facet ost1 $LCTL set_param fail_loc=0x1611
+ echo "foo" > $DIR/$tdir/f0
+ cancel_lru_locks osc
+ sync
+ sleep 2
+ do_facet ost1 $LCTL set_param fail_loc=0
+
+ echo "Nothing should be fixed since self detect and repair is disabled"
+ local repaired=$(do_facet ost1 $LCTL get_param -n \
+ obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid |
+ awk '/^repaired/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(1) Expected 0 repaired, but got $repaired"
+
+ echo "Read RPC with right parent FID should be accepted,"
+ echo "and cause parent FID on OST to be fixed"
+
+ do_facet ost1 $LCTL set_param -n \
+ obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid 1
+ cat $DIR/$tdir/f0 || error "(2) Read 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 ] ||
+ error "(3) Expected 1 repaired, but got $repaired"
+}
+run_test 19b "OST-object inconsistency self repair"
+
$LCTL set_param debug=-lfsck > /dev/null || true
# restore MDS/OST size
git://git.whamcloud.com / fs / lustre-release.git / blobdiff