2 # -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
3 # vim:shiftwidth=4:softtabstop=4:tabstop=4:
5 # Tests for multiple mount protection (MMP) feature.
7 # Run select tests by setting ONLY, or as arguments to the script.
8 # Skip specific tests by setting EXCEPT.
10 # e.g. ONLY="5 6" or ONLY="`seq 8 11`" or EXCEPT="7"
15 # bug number for skipped test:
16 ALWAYS_EXCEPT=${ALWAYS_EXCEPT:-"$MMP_EXCEPT"}
17 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
19 SRCDIR=$(cd $(dirname $0); echo $PWD)
20 export PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH:/sbin
22 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
23 . $LUSTRE/tests/test-framework.sh
25 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
28 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
29 remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
31 # unmount and cleanup the Lustre filesystem
32 MMP_RESTORE_MOUNT=false
33 if is_mounted $MOUNT || is_mounted $MOUNT2; then
35 MMP_RESTORE_MOUNT=true
38 SAVED_FAIL_ON_ERROR=$FAIL_ON_ERROR
43 # Get the failover facet.
44 get_failover_facet() {
46 local failover_facet=${facet}failover
48 local host=$(facet_host $facet)
49 local failover_host=$(facet_host $failover_facet)
51 [ -z "$failover_host" -o "$host" = "$failover_host" ] && \
57 # Initiate the variables for Lustre servers and targets.
59 MMP_MDS=${MMP_MDS:-$SINGLEMDS}
60 MMP_MDS_FAILOVER=$(get_failover_facet $MMP_MDS)
62 local mds_num=$(echo $MMP_MDS | tr -d "mds")
63 MMP_MDSDEV=$(mdsdevname $mds_num)
65 MMP_OSS=${MMP_OSS:-ost1}
66 MMP_OSS_FAILOVER=$(get_failover_facet $MMP_OSS)
68 local oss_num=$(echo $MMP_OSS | tr -d "ost")
69 MMP_OSTDEV=$(ostdevname $oss_num)
72 # Stop the MDS and OSS services on the primary or failover servers.
80 if [ "$flavor" = "failover" ]; then
81 mds_facet=$MMP_MDS_FAILOVER
82 oss_facet=$MMP_OSS_FAILOVER
88 stop $mds_facet $opts || return ${PIPESTATUS[0]}
89 stop $oss_facet $opts || return ${PIPESTATUS[0]}
92 # Enable the MMP feature.
97 do_facet $facet "$TUNE2FS -O mmp $device"
98 return ${PIPESTATUS[0]}
101 # Disable the MMP feature.
106 do_facet $facet "$TUNE2FS -O ^mmp $device"
107 return ${PIPESTATUS[0]}
110 # Set the MMP block to 'fsck' state
115 do_facet $facet "$LUSTRE/tests/mmp_mark.sh $device"
116 return ${PIPESTATUS[0]}
119 # Reset the MMP block (if any) back to the clean state.
124 do_facet $facet "$TUNE2FS -f -E clear-mmp $device"
125 return ${PIPESTATUS[0]}
128 # Check whether the MMP feature is enabled or not.
133 do_facet $facet "$DUMPE2FS -h $device | grep mmp"
134 return ${PIPESTATUS[0]}
137 # Get MMP update interval (in seconds) from the Lustre server target.
138 get_mmp_update_interval() {
143 interval=$(do_facet $facet \
144 "$DEBUGFS -c -R dump_mmp $device 2>/dev/null" |
145 awk 'tolower($0) ~ /update.interval/ { print $NF }')
146 [ -z "$interval" ] && interval=5
151 # Get MMP check interval (in seconds) from the Lustre server target.
152 get_mmp_check_interval() {
157 interval=$(do_facet $facet \
158 "$DEBUGFS -c -R dump_mmp $device 2>/dev/null" |
159 awk 'tolower($0) ~ /check.interval/ { print $NF }')
160 [ -z "$interval" ] && interval=5
165 # Adjust the MMP update interval (in seconds) on the Lustre server target.
166 # Specifying an interval of 0 means to use the default interval.
167 set_mmp_update_interval() {
170 local interval=${3:-0}
172 do_facet $facet "$TUNE2FS -E mmp_update_interval=$interval $device"
173 return ${PIPESTATUS[0]}
176 # Enable the MMP feature on the Lustre server targets.
180 if [ $(facet_fstype $MMP_MDS) != ldiskfs ]; then
181 skip "Only applicable to ldiskfs-based MDTs"
185 if [ $(facet_fstype $MMP_OSS) != ldiskfs ]; then
186 skip "Only applicable to ldiskfs-based OSTs"
190 # The MMP feature is automatically enabled by mkfs.lustre for
191 # new file system at format time if failover is being used.
192 # Otherwise, the Lustre administrator has to manually enable
193 # this feature when the file system is unmounted.
195 local var=${MMP_MDS}failover_HOST
196 if [ -z "${!var}" ]; then
197 log "Failover is not used on MDS, enabling MMP manually..."
198 enable_mmp $MMP_MDS $MMP_MDSDEV || \
199 error "failed to enable MMP on $MMP_MDSDEV on $MMP_MDS"
202 var=${MMP_OSS}failover_HOST
203 if [ -z "${!var}" ]; then
204 log "Failover is not used on OSS, enabling MMP manually..."
205 enable_mmp $MMP_OSS $MMP_OSTDEV || \
206 error "failed to enable MMP on $MMP_OSTDEV on $MMP_OSS"
209 # check whether the MMP feature is enabled or not
210 mmp_is_enabled $MMP_MDS $MMP_MDSDEV || \
211 error "MMP was not enabled on $MMP_MDSDEV on $MMP_MDS"
213 mmp_is_enabled $MMP_OSS $MMP_OSTDEV || \
214 error "MMP was not enabled on $MMP_OSTDEV on $MMP_OSS"
217 # Disable the MMP feature on the Lustre server targets
218 # which did not use failover.
221 local var=${MMP_MDS}failover_HOST
222 if [ -z "${!var}" ]; then
223 log "Failover is not used on MDS, disabling MMP manually..."
224 disable_mmp $MMP_MDS $MMP_MDSDEV || \
225 error "failed to disable MMP on $MMP_MDSDEV on $MMP_MDS"
226 mmp_is_enabled $MMP_MDS $MMP_MDSDEV && \
227 error "MMP was not disabled on $MMP_MDSDEV on $MMP_MDS"
230 var=${MMP_OSS}failover_HOST
231 if [ -z "${!var}" ]; then
232 log "Failover is not used on OSS, disabling MMP manually..."
233 disable_mmp $MMP_OSS $MMP_OSTDEV || \
234 error "failed to disable MMP on $MMP_OSTDEV on $MMP_OSS"
235 mmp_is_enabled $MMP_OSS $MMP_OSTDEV && \
236 error "MMP was not disabled on $MMP_OSTDEV on $MMP_OSS"
242 # Mount the shared target on the failover server after some interval it's
243 # mounted on the primary server.
244 mount_after_interval_sub() {
252 local failover_facet=$(get_failover_facet $facet)
255 local first_mount_rc=0
256 local second_mount_rc=0
258 log "Mounting $device on $facet..."
259 start $facet $device $opts &
262 if [ $interval -ne 0 ]; then
263 log "sleep $interval..."
267 log "Mounting $device on $failover_facet..."
268 start $failover_facet $device $opts
269 second_mount_rc=${PIPESTATUS[0]}
272 first_mount_rc=${PIPESTATUS[0]}
274 if [ $second_mount_rc -eq 0 -a $first_mount_rc -eq 0 ]; then
275 error_noexit "one mount delayed by mmp interval $interval should fail"
276 stop $facet || return ${PIPESTATUS[0]}
277 [ "$failover_facet" != "$facet" ] && stop $failover_facet || \
278 return ${PIPESTATUS[0]}
280 elif [ $second_mount_rc -ne 0 -a $first_mount_rc -ne 0 ]; then
281 error_noexit "mount failure on failover pair $facet,$failover_facet"
282 return $first_mount_rc
288 mount_after_interval() {
289 local mdt_interval=$1
290 local ost_interval=$2
293 mount_after_interval_sub $mdt_interval $MMP_MDSDEV $MMP_MDS \
294 $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
297 mount_after_interval_sub $ost_interval $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
299 if [ $rc -ne 0 ]; then
307 # Mount the shared target on the failover server
308 # during unmounting it on the primary server.
309 mount_during_unmount() {
315 local failover_facet=$(get_failover_facet $facet)
321 log "Mounting $device on $facet..."
322 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
324 log "Unmounting $device on $facet..."
328 log "Mounting $device on $failover_facet..."
329 start $failover_facet $device $mnt_opts
330 mount_rc=${PIPESTATUS[0]}
333 unmount_rc=${PIPESTATUS[0]}
335 if [ $mount_rc -eq 0 ]; then
336 error_noexit "mount during unmount of the first filesystem should fail"
337 stop $failover_facet || return ${PIPESTATUS[0]}
341 if [ $unmount_rc -ne 0 ]; then
342 error_noexit "unmount the $device on $facet should succeed"
349 # Mount the shared target on the failover server
350 # after clean unmounting it on the primary server.
351 mount_after_unmount() {
357 local failover_facet=$(get_failover_facet $facet)
359 log "Mounting $device on $facet..."
360 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
362 log "Unmounting $device on $facet..."
363 stop $facet || return ${PIPESTATUS[0]}
365 log "Mounting $device on $failover_facet..."
366 start $failover_facet $device $mnt_opts || return ${PIPESTATUS[0]}
371 # Mount the shared target on the failover server after rebooting
372 # the primary server.
373 mount_after_reboot() {
379 local failover_facet=$(get_failover_facet $facet)
382 log "Mounting $device on $facet..."
383 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
385 if [ "$FAILURE_MODE" = "HARD" ]; then
386 shutdown_facet $facet
388 wait_for_facet $facet
390 replay_barrier_nodf $facet
393 log "Mounting $device on $failover_facet..."
394 start $failover_facet $device $mnt_opts
396 if [ $rc -ne 0 ]; then
397 error_noexit "mount $device on $failover_facet should succeed"
398 stop $facet || return ${PIPESTATUS[0]}
405 # Run e2fsck on the Lustre server target.
413 echo "Running e2fsck on the device $device on $facet..."
414 do_facet $facet "$E2FSCK $opts $device"
415 return ${PIPESTATUS[0]}
418 # Check whether there are failover pairs for MDS and OSS servers.
419 check_failover_pair() {
420 [ "$MMP_MDS" = "$MMP_MDS_FAILOVER" -o "$MMP_OSS" = "$MMP_OSS_FAILOVER" ] \
421 && { skip_env "failover pair is needed" && return 1; }
427 # Test 1 - two mounts at the same time.
429 check_failover_pair || return 0
431 mount_after_interval 0 0 || return ${PIPESTATUS[0]}
432 stop_services primary || return ${PIPESTATUS[0]}
434 run_test 1 "two mounts at the same time"
436 # Test 2 - one mount delayed by mmp update interval.
438 check_failover_pair || return 0
440 local mdt_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
441 local ost_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
443 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
444 stop_services primary || return ${PIPESTATUS[0]}
446 run_test 2 "one mount delayed by mmp update interval"
448 # Test 3 - one mount delayed by 2x mmp check interval.
450 check_failover_pair || return 0
452 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
453 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
455 mdt_interval=$((2 * $mdt_interval + 1))
456 ost_interval=$((2 * $ost_interval + 1))
458 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
459 stop_services primary || return ${PIPESTATUS[0]}
461 run_test 3 "one mount delayed by 2x mmp check interval"
463 # Test 4 - one mount delayed by > 2x mmp check interval.
465 check_failover_pair || return 0
467 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
468 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
470 mdt_interval=$((4 * $mdt_interval))
471 ost_interval=$((4 * $ost_interval))
473 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
474 stop_services primary || return ${PIPESTATUS[0]}
476 run_test 4 "one mount delayed by > 2x mmp check interval"
478 # Test 5 - mount during unmount of the first filesystem.
481 check_failover_pair || return 0
483 mount_during_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
484 return ${PIPESTATUS[0]}
487 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
488 mount_during_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
490 if [ $rc -ne 0 ]; then
491 stop $MMP_MDS || return ${PIPESTATUS[0]}
495 stop $MMP_MDS || return ${PIPESTATUS[0]}
497 run_test 5 "mount during unmount of the first filesystem"
499 # Test 6 - mount after clean unmount.
502 check_failover_pair || return 0
504 mount_after_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
505 return ${PIPESTATUS[0]}
508 mount_after_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
510 if [ $rc -ne 0 ]; then
511 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
515 stop_services failover || return ${PIPESTATUS[0]}
517 run_test 6 "mount after clean unmount"
519 # Test 7 - mount after reboot.
522 check_failover_pair || return 0
524 mount_after_reboot $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
525 return ${PIPESTATUS[0]}
528 mount_after_reboot $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
530 if [ $rc -ne 0 ]; then
531 stop $MMP_MDS || return ${PIPESTATUS[0]}
532 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
536 stop_services failover || return ${PIPESTATUS[0]}
537 stop_services primary || return ${PIPESTATUS[0]}
539 run_test 7 "mount after reboot"
541 # Test 8 - mount during e2fsck (should never succeed).
547 # After writing a new sequence number into the MMP block, e2fsck will
548 # sleep at least (2 * new_interval + 1) seconds before it goes into
553 saved_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
554 set_mmp_update_interval $MMP_MDS $MMP_MDSDEV $new_interval
556 run_e2fsck $MMP_MDS $MMP_MDSDEV "-fy" &
560 if start $MMP_MDS_FAILOVER $MMP_MDSDEV $MDS_MOUNT_OPTS; then
562 "mount $MMP_MDSDEV on $MMP_MDS_FAILOVER should fail"
563 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
564 set_mmp_update_interval $MMP_MDS $MMP_MDSDEV $saved_interval
569 set_mmp_update_interval $MMP_MDS $MMP_MDSDEV $saved_interval
573 saved_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
574 set_mmp_update_interval $MMP_OSS $MMP_OSTDEV $new_interval
576 run_e2fsck $MMP_OSS $MMP_OSTDEV "-fy" &
580 if start $MMP_OSS_FAILOVER $MMP_OSTDEV $OST_MOUNT_OPTS; then
582 "mount $MMP_OSTDEV on $MMP_OSS_FAILOVER should fail"
583 stop $MMP_OSS_FAILOVER || return ${PIPESTATUS[0]}
584 set_mmp_update_interval $MMP_OSS $MMP_OSTDEV $saved_interval
589 set_mmp_update_interval $MMP_OSS $MMP_OSTDEV $saved_interval
592 run_test 8 "mount during e2fsck"
594 # Test 9 - mount after aborted e2fsck (should never succeed).
596 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
597 if ! start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
598 local rc=${PIPESTATUS[0]}
599 stop $MMP_MDS || return ${PIPESTATUS[0]}
602 stop_services primary || return ${PIPESTATUS[0]}
604 mark_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
606 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
607 if start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS; then
608 error_noexit "mount $MMP_MDSDEV on $MMP_MDS should fail"
609 stop $MMP_MDS || return ${PIPESTATUS[0]}
613 reset_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
615 mark_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
617 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
618 if start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
619 error_noexit "mount $MMP_OSTDEV on $MMP_OSS should fail"
620 stop $MMP_OSS || return ${PIPESTATUS[0]}
624 reset_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
627 run_test 9 "mount after aborted e2fsck"
629 # Test 10 - e2fsck with mounted filesystem.
633 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
634 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
636 run_e2fsck $MMP_MDS_FAILOVER $MMP_MDSDEV "-fn"
639 # e2fsck is called with -n option (Open the filesystem read-only), so
640 # 0 (No errors) and 4 (File system errors left uncorrected) are the only
641 # acceptable exit codes in this case
642 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
643 error_noexit "e2fsck $MMP_MDSDEV on $MMP_MDS_FAILOVER returned $rc"
644 stop $MMP_MDS || return ${PIPESTATUS[0]}
648 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
649 start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS
651 if [ $rc -ne 0 ]; then
652 stop $MMP_MDS || return ${PIPESTATUS[0]}
656 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fn"
658 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
659 error_noexit "e2fsck $MMP_OSTDEV on $MMP_OSS_FAILOVER returned $rc"
662 stop_services primary || return ${PIPESTATUS[0]}
665 run_test 10 "e2fsck with mounted filesystem"
668 FAIL_ON_ERROR=$SAVED_FAIL_ON_ERROR
671 $MMP_RESTORE_MOUNT && setupall