+# check PARAM_SYS_LDLM_TIMEOUT option of MKFS.LUSTRE
+test_49() { # bug 17710
+ local OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS
+ local OLD_OST_MKFS_OPTS=$OST_MKFS_OPTS
+ local LOCAL_TIMEOUT=20
+
+
+ OST_MKFS_OPTS="--ost --fsname=$FSNAME --device-size=$OSTSIZE --mgsnode=$MGSNID --param sys.timeout=$LOCAL_TIMEOUT --param sys.ldlm_timeout=$LOCAL_TIMEOUT $MKFSOPT $OSTOPT"
+
+ reformat
+ start_mds
+ start_ost
+ mount_client $MOUNT
+ check_mount || return 1
+
+ echo "check ldlm_timout..."
+ LDLM_MDS="`do_facet mds lctl get_param -n ldlm_timeout`"
+ LDLM_OST1="`do_facet ost1 lctl get_param -n ldlm_timeout`"
+ LDLM_CLIENT="`do_facet client lctl get_param -n ldlm_timeout`"
+
+ if [ $LDLM_MDS -ne $LDLM_OST1 ] || [ $LDLM_MDS -ne $LDLM_CLIENT ]; then
+ error "Different LDLM_TIMEOUT:$LDLM_MDS $LDLM_OST1 $LDLM_CLIENT"
+ fi
+
+ if [ $LDLM_MDS -ne $((LOCAL_TIMEOUT / 3)) ]; then
+ error "LDLM_TIMEOUT($LDLM_MDS) is not correct"
+ fi
+
+ umount_client $MOUNT
+ stop_ost || return 2
+ stop_mds || return 3
+
+ OST_MKFS_OPTS="--ost --fsname=$FSNAME --device-size=$OSTSIZE --mgsnode=$MGSNID --param sys.timeout=$LOCAL_TIMEOUT --param sys.ldlm_timeout=$((LOCAL_TIMEOUT - 1)) $MKFSOPT $OSTOPT"
+
+ reformat
+ start_mds || return 4
+ start_ost || return 5
+ mount_client $MOUNT || return 6
+ check_mount || return 7
+
+ LDLM_MDS="`do_facet mds lctl get_param -n ldlm_timeout`"
+ LDLM_OST1="`do_facet ost1 lctl get_param -n ldlm_timeout`"
+ LDLM_CLIENT="`do_facet client lctl get_param -n ldlm_timeout`"
+
+ if [ $LDLM_MDS -ne $LDLM_OST1 ] || [ $LDLM_MDS -ne $LDLM_CLIENT ]; then
+ error "Different LDLM_TIMEOUT:$LDLM_MDS $LDLM_OST1 $LDLM_CLIENT"
+ fi
+
+ if [ $LDLM_MDS -ne $((LOCAL_TIMEOUT - 1)) ]; then
+ error "LDLM_TIMEOUT($LDLM_MDS) is not correct"
+ fi
+
+ cleanup || return $?
+
+ MDS_MKFS_OPTS=$OLD_MDS_MKFS_OPTS
+ OST_MKFS_OPTS=$OLD_OST_MKFS_OPTS
+}
+run_test 49 "check PARAM_SYS_LDLM_TIMEOUT option of MKFS.LUSTRE"
+
+lazystatfs() {
+ # Test both statfs and lfs df and fail if either one fails
+ multiop_bg_pause $1 f_
+ RC1=$?
+ PID=$!
+ killall -USR1 multiop
+ [ $RC1 -ne 0 ] && log "lazystatfs multiop failed"
+ wait $PID || { RC1=$?; log "multiop return error "; }
+
+ $LFS df &
+ PID=$!
+ sleep 5
+ kill -s 0 $PID
+ RC2=$?
+ if [ $RC2 -eq 0 ]; then
+ kill -s 9 $PID
+ log "lazystatfs df failed"
+ fi
+
+ RC=0
+ [[ $RC1 -ne 0 || $RC2 -eq 0 ]] && RC=1
+ return $RC
+}
+
+test_50a() {
+ setup
+ lctl set_param llite.$FSNAME-*.lazystatfs=1
+ touch $DIR/$tfile
+
+ lazystatfs $MOUNT || error "lazystatfs failed but no down servers"
+
+ cleanup || return $?
+}
+run_test 50a "lazystatfs all servers available =========================="
+
+test_50b() {
+ setup
+ lctl set_param llite.$FSNAME-*.lazystatfs=1
+ touch $DIR/$tfile
+
+ # Wait for client to detect down OST
+ stop_ost || error "Unable to stop OST1"
+ wait_osc_import_state mds ost DISCONN
+
+ lazystatfs $MOUNT || error "lazystatfs should don't have returned EIO"
+
+ umount_client $MOUNT || error "Unable to unmount client"
+ stop_mds || error "Unable to stop MDS"
+}
+run_test 50b "lazystatfs all servers down =========================="
+
+test_50c() {
+ start_mds || error "Unable to start MDS"
+ start_ost || error "Unable to start OST1"
+ start_ost2 || error "Unable to start OST2"
+ mount_client $MOUNT || error "Unable to mount client"
+ lctl set_param llite.$FSNAME-*.lazystatfs=1
+ touch $DIR/$tfile
+
+ # Wait for client to detect down OST
+ stop_ost || error "Unable to stop OST1"
+ wait_osc_import_state mds ost DISCONN
+ lazystatfs $MOUNT || error "lazystatfs failed with one down server"
+
+ umount_client $MOUNT || error "Unable to unmount client"
+ stop_ost2 || error "Unable to stop OST2"
+ stop_mds || error "Unable to stop MDS"
+}
+run_test 50c "lazystatfs one server down =========================="
+
+test_50d() {
+ start_mds || error "Unable to start MDS"
+ start_ost || error "Unable to start OST1"
+ start_ost2 || error "Unable to start OST2"
+ mount_client $MOUNT || error "Unable to mount client"
+ lctl set_param llite.$FSNAME-*.lazystatfs=1
+ touch $DIR/$tfile
+
+ # Issue the statfs during the window where the client still
+ # belives the OST to be available but it is in fact down.
+ # No failure just a statfs which hangs for a timeout interval.
+ stop_ost || error "Unable to stop OST1"
+ lazystatfs $MOUNT || error "lazystatfs failed with one down server"
+
+ umount_client $MOUNT || error "Unable to unmount client"
+ stop_ost2 || error "Unable to stop OST2"
+ stop_mds || error "Unable to stop MDS"
+}
+run_test 50d "lazystatfs client/server conn race =========================="
+
+test_50e() {
+ local RC1
+ local pid
+
+ reformat_and_config
+ start_mds || return 1
+ #first client should see only one ost
+ start_ost || return 2
+ wait_osc_import_state mds ost FULL
+
+ # Wait for client to detect down OST
+ stop_ost || error "Unable to stop OST1"
+ wait_osc_import_state mds ost DISCONN
+
+ mount_client $MOUNT || error "Unable to mount client"
+ lctl set_param llite.$FSNAME-*.lazystatfs=0
+
+ multiop_bg_pause $MOUNT _f
+ RC1=$?
+ pid=$!
+
+ if [ $RC1 -ne 0 ]; then
+ log "multiop failed $RC1"
+ else
+ kill -USR1 $pid
+ sleep $(( $TIMEOUT+1 ))
+ kill -0 $pid
+ [ $? -ne 0 ] && error "process isn't sleep"
+ start_ost || error "Unable to start OST1"
+ wait $pid || error "statfs failed"
+ fi
+
+ umount_client $MOUNT || error "Unable to unmount client"
+ stop_ost || error "Unable to stop OST1"
+ stop_mds || error "Unable to stop MDS"
+}
+run_test 50e "normal statfs all servers down =========================="
+
+test_50f() {
+ local RC1
+ local pid
+ CONN_PROC="osc.$FSNAME-OST0001-osc-[M]*.ost_server_uuid"
+
+ start_mds || error "Unable to start mds"
+ #first client should see only one ost
+ start_ost || error "Unable to start OST1"
+ wait_osc_import_state mds ost FULL
+
+ start_ost2 || error "Unable to start OST2"
+ wait_osc_import_state mds ost2 FULL
+
+ # Wait for client to detect down OST
+ stop_ost2 || error "Unable to stop OST2"
+
+ wait_osc_import_state mds ost2 DISCONN
+ mount_client $MOUNT || error "Unable to mount client"
+ lctl set_param llite.$FSNAME-*.lazystatfs=0
+
+ multiop_bg_pause $MOUNT _f
+ RC1=$?
+ pid=$!
+
+ if [ $RC1 -ne 0 ]; then
+ log "lazystatfs multiop failed $RC1"
+ else
+ kill -USR1 $pid
+ sleep $(( $TIMEOUT+1 ))
+ kill -0 $pid
+ [ $? -ne 0 ] && error "process isn't sleep"
+ start_ost2 || error "Unable to start OST2"
+ wait $pid || error "statfs failed"
+ stop_ost2 || error "Unable to stop OST2"
+ fi
+
+ umount_client $MOUNT || error "Unable to unmount client"
+ stop_ost || error "Unable to stop OST1"
+ stop_mds || error "Unable to stop MDS"
+ writeconf
+}
+run_test 50f "normal statfs one server in down =========================="
+
+test_51() {
+ local LOCAL_TIMEOUT=20
+
+ reformat
+ start_mds
+ start_ost
+ mount_client $MOUNT
+ check_mount || return 1
+
+ mkdir $MOUNT/d1
+ $LFS setstripe -c -1 $MOUNT/d1
+ #define OBD_FAIL_MDS_REINT_DELAY 0x142
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x142"
+ touch $MOUNT/d1/f1 &
+ local pid=$!
+ sleep 2
+ start_ost2 || return 2
+ wait $pid
+ stop_ost2 || return 3
+ cleanup
+}
+run_test 51 "Verify that mdt_reint handles RMF_MDT_MD correctly when an OST is added"