LU-7349 lfsck: skip auto resume lfsck when mount

[fs/lustre-release.git] / lustre / tests / sanity-lfsck.sh

diff --git a/lustre/tests/sanity-lfsck.sh b/lustre/tests/sanity-lfsck.sh
index da57b5f..10a892e 100644 (file)
--- a/lustre/tests/sanity-lfsck.sh
+++ b/lustre/tests/sanity-lfsck.sh
@@ -22,6 +22,16 @@ init_logging
 
 require_dsh_mds || exit 0
 
+load_modules
+
+if ! check_versions; then
+       skip "It is NOT necessary to test lfsck under interoperation mode"
+       exit 0
+fi
+
+[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.60) ]] &&
+       skip "Need MDS version at least 2.3.60" && exit 0
+
 LTIME=${LTIME:-120}
 
 SAVED_MDSSIZE=${MDSSIZE}
@@ -38,10 +48,6 @@ OSTSIZE=100000
 formatall
 setupall
 
-[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.60) ]] &&
-       skip "Need MDS version at least 2.3.60" && check_and_cleanup_lustre &&
-       exit 0
-
 [[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.4.90) ]] &&
        ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2c"
 
@@ -74,6 +80,7 @@ SHOW_LAYOUT_ON_OST="do_facet ost1 \
                $LCTL get_param -n obdfilter.${OST_DEV}.lfsck_layout"
 MOUNT_OPTS_SCRUB="-o user_xattr"
 MOUNT_OPTS_NOSCRUB="-o user_xattr,noscrub"
+MOUNT_OPTS_SKIP_LFSCK="-o user_xattr,skip_lfsck"
 
 lfsck_prep() {
        local ndirs=$1
@@ -109,6 +116,19 @@ lfsck_prep() {
        echo "prepared $(date)."
 }
 
+run_e2fsck_on_mdt0() {
+       [ $(facet_fstype $SINGLEMDS) != ldiskfs ] && return
+
+       stop $SINGLEMDS > /dev/null || error "(0) Fail to the stop MDT0"
+       run_e2fsck $(facet_active_host $SINGLEMDS) $(mdsdevname 1) "-n" |
+               grep "Fix? no" && {
+               run_e2fsck $(facet_active_host $SINGLEMDS) $(mdsdevname 1) "-n"
+               error "(2) Detected inconsistency on MDT0"
+       }
+       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
+               error "(3) Fail to start MDT0"
+}
+
 test_0() {
        lfsck_prep 3 3
 
@@ -195,6 +215,8 @@ test_1a() {
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair crashed FID-in-dirent: $repaired"
 
+       run_e2fsck_on_mdt0
+
        mount_client $MOUNT || error "(6) Fail to start client!"
 
        #define OBD_FAIL_FID_LOOKUP     0x1505
@@ -239,6 +261,8 @@ test_1b()
                error "(5) Fail to repair the missing FID-in-LMA: $repaired"
 
        do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+       run_e2fsck_on_mdt0
+
        mount_client $MOUNT || error "(6) Fail to start client!"
 
        #define OBD_FAIL_FID_LOOKUP     0x1505
@@ -276,6 +300,8 @@ test_2a() {
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair crashed linkEA: $repaired"
 
+       run_e2fsck_on_mdt0
+
        mount_client $MOUNT || error "(6) Fail to start client!"
 
        stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
@@ -311,6 +337,8 @@ test_2b()
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair crashed linkEA: $repaired"
 
+       run_e2fsck_on_mdt0
+
        mount_client $MOUNT || error "(6) Fail to start client!"
 
        stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
@@ -346,6 +374,8 @@ test_2c()
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair crashed linkEA: $repaired"
 
+       run_e2fsck_on_mdt0
+
        mount_client $MOUNT || error "(6) Fail to start client!"
 
        stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
@@ -381,6 +411,8 @@ test_2d()
        [ $repaired -eq 1 ] ||
                error "(5) Fail to repair crashed linkEA: $repaired"
 
+       run_e2fsck_on_mdt0
+
        mount_client $MOUNT || error "(6) Fail to start client!"
 
        stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
@@ -518,6 +550,8 @@ test_4()
        [ $repaired -ge 9 ] ||
                error "(9) Fail to re-generate FID-in-dirent: $repaired"
 
+       run_e2fsck_on_mdt0
+
        mount_client $MOUNT || error "(10) Fail to start client!"
 
        #define OBD_FAIL_FID_LOOKUP     0x1505
@@ -576,6 +610,8 @@ test_5()
        [ $repaired -ge 2 ] ||
                error "(9) Fail to generate FID-in-dirent for IGIF: $repaired"
 
+       run_e2fsck_on_mdt0
+
        mount_client $MOUNT || error "(10) Fail to start client!"
 
        #define OBD_FAIL_FID_LOOKUP     0x1505
@@ -905,6 +941,30 @@ test_8()
        [ "$STATUS" == "paused" ] ||
                error "(20) Expect 'paused', but got '$STATUS'"
 
+       echo "stop $SINGLEMDS"
+       stop $SINGLEMDS > /dev/null || error "(20.1) Fail to stop MDS!"
+
+       echo "start $SINGLEMDS without resume LFSCK"
+       start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SKIP_LFSCK > /dev/null ||
+               error "(20.2) Fail to start MDS!"
+
+       timer=0
+       while [ $timer -lt $timeout ]; do
+               STATUS=$(do_facet $SINGLEMDS "$LCTL get_param -n \
+                       mdt.${MDT_DEV}.recovery_status |
+                       awk '/^status/ { print \\\$2 }'")
+               [ "$STATUS" != "RECOVERING" ] && break;
+               sleep 1
+               timer=$((timer + 1))
+       done
+
+       [ $timer != $timeout ] ||
+               error "(20.3) recovery timeout"
+
+       STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
+       [ "$STATUS" == "paused" ] ||
+               error "(20.4) Expect 'paused', but got '$STATUS'"
+
        #define OBD_FAIL_LFSCK_DELAY3           0x1602
        do_facet $SINGLEMDS $LCTL set_param fail_val=2 fail_loc=0x1602
 
@@ -1590,6 +1650,9 @@ test_15c() {
        [ $MDSCOUNT -lt 2 ] &&
                skip "We need at least 2 MDSes for this test" && return
 
+       [ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.55) ] &&
+               skip "Skip the test after 2.7.55 see LU-6437" && return
+
        echo "#####"
        echo "According to current metadata migration implementation,"
        echo "before the old MDT-object is removed, both the new MDT-object"
@@ -1612,7 +1675,7 @@ test_15c() {
        #define OBD_FAIL_MIGRATE_DELAY                  0x1803
        do_facet mds2 $LCTL set_param fail_val=5 fail_loc=0x1803
        echo "Migrate $DIR/$tdir/a1 from MDT1 to MDT0 with delay"
-       $LFS mv -M 0 $DIR/$tdir/a1 &
+       $LFS migrate -m 0 $DIR/$tdir/a1 &
 
        sleep 1
        echo "Trigger layout LFSCK to race with the migration"
@@ -4096,7 +4159,7 @@ test_31f() {
        check_mount_and_prep
 
        echo "Inject failure stub on MDT0 to simulate the case that the"
-       echo "slave MDT-object (that resides on differnt MDT as the master"
+       echo "slave MDT-object (that resides on different MDT as the master"
        echo "MDT-object resides on) lost the LMV EA."
 
        #define OBD_FAIL_LFSCK_LOST_SLAVE_LMV   0x162a