2 # vim:expandtab:shiftwidth=4:softtabstop=4:tabstop=4:
4 # Tests for multiple mount protection (MMP) feature.
6 # Run select tests by setting ONLY, or as arguments to the script.
7 # Skip specific tests by setting EXCEPT.
9 # e.g. ONLY="5 6" or ONLY="`seq 8 11`" or EXCEPT="7"
14 # bug number for skipped test:
15 ALWAYS_EXCEPT=${ALWAYS_EXCEPT:-"$MMP_EXCEPT"}
16 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
18 SRCDIR=$(cd $(dirname $0); echo $PWD)
19 export PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH:/sbin
21 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
22 . $LUSTRE/tests/test-framework.sh
24 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
27 require_dsh_mds || exit 0
28 require_dsh_ost || exit 0
30 # unmount and cleanup the Lustre filesystem
31 MMP_RESTORE_MOUNT=false
32 if is_mounted $MOUNT || is_mounted $MOUNT2; then
34 MMP_RESTORE_MOUNT=true
37 SAVED_FAIL_ON_ERROR=$FAIL_ON_ERROR
42 # Get the failover facet.
43 get_failover_facet() {
45 local failover_facet=${facet}failover
47 local host=$(facet_host $facet)
48 local failover_host=$(facet_host $failover_facet)
50 [ -z "$failover_host" -o "$host" = "$failover_host" ] && \
56 # Initiate the variables for Lustre servers and targets.
59 MMP_MDS_FAILOVER=$(get_failover_facet $MMP_MDS)
63 MMP_OSS=${MMP_OSS:-ost1}
64 MMP_OSS_FAILOVER=$(get_failover_facet $MMP_OSS)
66 local oss_num=$(echo $MMP_OSS | tr -d "ost")
67 MMP_OSTDEV=$(ostdevname $oss_num)
70 # Stop the MDS and OSS services on the primary or failover servers.
78 if [ "$flavor" = "failover" ]; then
79 mds_facet=$MMP_MDS_FAILOVER
80 oss_facet=$MMP_OSS_FAILOVER
86 stop $mds_facet $opts || return ${PIPESTATUS[0]}
87 stop $oss_facet $opts || return ${PIPESTATUS[0]}
90 # Enable the MMP feature.
95 do_facet $facet "$TUNE2FS -O mmp $device"
96 return ${PIPESTATUS[0]}
99 # Disable the MMP feature.
104 do_facet $facet "$TUNE2FS -O ^mmp $device"
105 return ${PIPESTATUS[0]}
108 # Reset the MMP block (if any) back to the clean state.
113 do_facet $facet "$TUNE2FS -f -E clear-mmp $device"
114 return ${PIPESTATUS[0]}
117 # Check whether the MMP feature is enabled or not.
122 do_facet $facet "$DUMPE2FS -h $device | grep mmp"
123 return ${PIPESTATUS[0]}
126 # Get MMP update interval (in seconds) from the Lustre server target.
127 get_mmp_update_interval() {
132 interval=$(do_facet $facet "$DEBUGFS -c -R dump_mmp $device 2>/dev/null \
133 | grep 'MMP Update Interval' | cut -d' ' -f4")
134 [ -z "$interval" ] && interval=1
139 # Get MMP check interval (in seconds) from the Lustre server target.
140 get_mmp_check_interval() {
145 interval=$(do_facet $facet "$DEBUGFS -c -R dump_mmp $device 2>/dev/null \
146 | grep 'MMP Check Interval' | cut -d' ' -f4")
147 [ -z "$interval" ] && interval=5
152 # Enable the MMP feature on the Lustre server targets.
156 # The MMP feature is automatically enabled by mkfs.lustre for
157 # new file system at format time if failover is being used.
158 # Otherwise, the Lustre administrator has to manually enable
159 # this feature when the file system is unmounted.
161 if [ -z "$mdsfailover_HOST" ]; then
162 log "Failover is not used on MDS, enabling MMP manually..."
163 enable_mmp $MMP_MDS $MMP_MDSDEV || \
164 error "failed to enable MMP on $MMP_MDSDEV on $MMP_MDS"
167 local var=${MMP_OSS}failover_HOST
169 if [ -z "${!var}" ]; then
170 log "Failover is not used on OSS, enabling MMP manually..."
171 enable_mmp $MMP_OSS $MMP_OSTDEV || \
172 error "failed to enable MMP on $MMP_OSTDEV on $MMP_OSS"
175 # check whether the MMP feature is enabled or not
176 mmp_is_enabled $MMP_MDS $MMP_MDSDEV || \
177 error "MMP was not enabled on $MMP_MDSDEV on $MMP_MDS"
179 mmp_is_enabled $MMP_OSS $MMP_OSTDEV || \
180 error "MMP was not enabled on $MMP_OSTDEV on $MMP_OSS"
183 # Disable the MMP feature on the Lustre server targets
184 # which did not use failover.
187 if [ -z "$mdsfailover_HOST" ]; then
188 log "Failover is not used on MDS, disabling MMP manually..."
189 disable_mmp $MMP_MDS $MMP_MDSDEV || \
190 error "failed to disable MMP on $MMP_MDSDEV on $MMP_MDS"
191 mmp_is_enabled $MMP_MDS $MMP_MDSDEV && \
192 error "MMP was not disabled on $MMP_MDSDEV on $MMP_MDS"
195 local var=${MMP_OSS}failover_HOST
197 if [ -z "${!var}" ]; then
198 log "Failover is not used on OSS, disabling MMP manually..."
199 disable_mmp $MMP_OSS $MMP_OSTDEV || \
200 error "failed to disable MMP on $MMP_OSTDEV on $MMP_OSS"
201 mmp_is_enabled $MMP_OSS $MMP_OSTDEV && \
202 error "MMP was not disabled on $MMP_OSTDEV on $MMP_OSS"
208 # Mount the shared target on the failover server after some interval it's
209 # mounted on the primary server.
210 mount_after_interval_sub() {
218 local failover_facet=$(get_failover_facet $facet)
221 local first_mount_rc=0
222 local second_mount_rc=0
224 log "Mounting $device on $facet..."
225 start $facet $device $opts &
228 if [ $interval -ne 0 ]; then
229 log "sleep $interval..."
233 log "Mounting $device on $failover_facet..."
234 start $failover_facet $device $opts
235 second_mount_rc=${PIPESTATUS[0]}
238 first_mount_rc=${PIPESTATUS[0]}
240 if [ $second_mount_rc -eq 0 -a $first_mount_rc -eq 0 ]; then
241 error_noexit "one mount delayed by mmp interval $interval should fail"
242 stop $facet || return ${PIPESTATUS[0]}
243 [ "$failover_facet" != "$facet" ] && stop $failover_facet || \
244 return ${PIPESTATUS[0]}
246 elif [ $second_mount_rc -ne 0 -a $first_mount_rc -ne 0 ]; then
247 error_noexit "failed to mount on the failover pair $facet,$failover_facet"
248 return $first_mount_rc
254 mount_after_interval() {
255 local mdt_interval=$1
256 local ost_interval=$2
259 mount_after_interval_sub $mdt_interval $MMP_MDSDEV $MMP_MDS \
260 $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
263 mount_after_interval_sub $ost_interval $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
265 if [ $rc -ne 0 ]; then
273 # Mount the shared target on the failover server
274 # during unmounting it on the primary server.
275 mount_during_unmount() {
281 local failover_facet=$(get_failover_facet $facet)
287 log "Mounting $device on $facet..."
288 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
293 start $failover_facet $device $mnt_opts
294 mount_rc=${PIPESTATUS[0]}
297 unmount_rc=${PIPESTATUS[0]}
299 if [ $mount_rc -eq 0 ]; then
300 error_noexit "mount during unmount of the first filesystem should fail"
301 stop $failover_facet || return ${PIPESTATUS[0]}
305 if [ $unmount_rc -ne 0 ]; then
306 error_noexit "unmount the $device on $facet should succeed"
313 # Mount the shared target on the failover server
314 # after clean unmounting it on the primary server.
315 mount_after_unmount() {
321 local failover_facet=$(get_failover_facet $facet)
323 log "Mounting $device on $facet..."
324 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
326 log "Unmounting $device on $facet..."
327 stop $facet || return ${PIPESTATUS[0]}
329 log "Mounting $device on $failover_facet..."
330 start $failover_facet $device $mnt_opts || return ${PIPESTATUS[0]}
335 # Mount the shared target on the failover server after rebooting
336 # the primary server.
337 mount_after_reboot() {
343 local failover_facet=$(get_failover_facet $facet)
346 log "Mounting $device on $facet..."
347 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
349 if [ "$FAILURE_MODE" = "HARD" ]; then
350 shutdown_facet $facet
352 wait_for_facet $facet
354 replay_barrier_nodf $facet
357 log "Mounting $device on $failover_facet..."
358 start $failover_facet $device $mnt_opts
360 if [ $rc -ne 0 ]; then
361 error_noexit "mount $device on $failover_facet should succeed"
362 stop $facet || return ${PIPESTATUS[0]}
369 # Run e2fsck on the Lustre server target.
377 log "Running e2fsck on the device $device on $facet..."
378 do_facet $facet "$E2FSCK $opts $device"
379 return ${PIPESTATUS[0]}
382 # Check whether there are failover pairs for MDS and OSS servers.
383 check_failover_pair() {
384 [ "$MMP_MDS" = "$MMP_MDS_FAILOVER" -o "$MMP_OSS" = "$MMP_OSS_FAILOVER" ] \
385 && { skip_env "failover pair is needed" && return 1; }
391 # Test 1 - two mounts at the same time.
393 check_failover_pair || return 0
395 mount_after_interval 0 0 || return ${PIPESTATUS[0]}
396 stop_services primary || return ${PIPESTATUS[0]}
398 run_test 1 "two mounts at the same time"
400 # Test 2 - one mount delayed by mmp update interval.
402 check_failover_pair || return 0
404 local mdt_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
405 local ost_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
407 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
408 stop_services primary || return ${PIPESTATUS[0]}
410 run_test 2 "one mount delayed by mmp update interval"
412 # Test 3 - one mount delayed by 2x mmp check interval.
414 check_failover_pair || return 0
416 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
417 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
419 mdt_interval=$((2 * $mdt_interval + 1))
420 ost_interval=$((2 * $ost_interval + 1))
422 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
423 stop_services primary || return ${PIPESTATUS[0]}
425 run_test 3 "one mount delayed by 2x mmp check interval"
427 # Test 4 - one mount delayed by > 2x mmp check interval.
429 check_failover_pair || return 0
431 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
432 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
434 mdt_interval=$((4 * $mdt_interval))
435 ost_interval=$((4 * $ost_interval))
437 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
438 stop_services primary || return ${PIPESTATUS[0]}
440 run_test 4 "one mount delayed by > 2x mmp check interval"
442 # Test 5 - mount during unmount of the first filesystem.
445 check_failover_pair || return 0
447 mount_during_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
448 return ${PIPESTATUS[0]}
451 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
452 mount_during_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
454 if [ $rc -ne 0 ]; then
455 stop $MMP_MDS || return ${PIPESTATUS[0]}
459 stop $MMP_MDS || return ${PIPESTATUS[0]}
461 run_test 5 "mount during unmount of the first filesystem"
463 # Test 6 - mount after clean unmount.
466 check_failover_pair || return 0
468 mount_after_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
469 return ${PIPESTATUS[0]}
472 mount_after_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
474 if [ $rc -ne 0 ]; then
475 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
479 stop_services failover || return ${PIPESTATUS[0]}
481 run_test 6 "mount after clean unmount"
483 # Test 7 - mount after reboot.
486 check_failover_pair || return 0
488 mount_after_reboot $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
489 return ${PIPESTATUS[0]}
492 mount_after_reboot $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
494 if [ $rc -ne 0 ]; then
495 stop $MMP_MDS || return ${PIPESTATUS[0]}
496 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
500 stop_services failover || return ${PIPESTATUS[0]}
501 stop_services primary || return ${PIPESTATUS[0]}
503 run_test 7 "mount after reboot"
505 # Test 8 - mount during e2fsck (should never succeed).
509 run_e2fsck $MMP_MDS $MMP_MDSDEV "-fy" &
513 log "Mounting $MMP_MDSDEV on $MMP_MDS_FAILOVER..."
514 if start $MMP_MDS_FAILOVER $MMP_MDSDEV $MDS_MOUNT_OPTS; then
515 error_noexit "mount $MMP_MDSDEV on $MMP_MDS_FAILOVER should fail"
516 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
523 run_e2fsck $MMP_OSS $MMP_OSTDEV "-fy" &
527 log "Mounting $MMP_OSTDEV on $MMP_OSS_FAILOVER..."
528 if start $MMP_OSS_FAILOVER $MMP_OSTDEV $OST_MOUNT_OPTS; then
529 error_noexit "mount $MMP_OSTDEV on $MMP_OSS_FAILOVER should fail"
530 stop $MMP_OSS_FAILOVER || return ${PIPESTATUS[0]}
537 run_test 8 "mount during e2fsck"
539 # Test 9 - mount after aborted e2fsck (should never succeed).
542 local mdt_mmp_check_interval
543 local ost_mmp_check_interval
545 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
546 if ! start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
547 local rc=${PIPESTATUS[0]}
548 stop $MMP_MDS || return ${PIPESTATUS[0]}
551 stop_services primary || return ${PIPESTATUS[0]}
553 mdt_mmp_check_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
554 run_e2fsck $MMP_MDS_FAILOVER $MMP_MDSDEV "-fy" &
556 sleep $((2 * $mdt_mmp_check_interval))
557 kill -s ABRT $e2fsck_pid
559 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
560 if start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS; then
561 error_noexit "mount $MMP_MDSDEV on $MMP_MDS should fail"
562 stop $MMP_MDS || return ${PIPESTATUS[0]}
566 reset_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
569 ost_mmp_check_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
570 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fy" &
572 sleep $((2 * $ost_mmp_check_interval))
573 kill -s ABRT $e2fsck_pid
575 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
576 if start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
577 error_noexit "mount $MMP_OSTDEV on $MMP_OSS should fail"
578 stop $MMP_OSS || return ${PIPESTATUS[0]}
582 reset_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
585 run_test 9 "mount after aborted e2fsck"
587 # Test 10 - e2fsck with mounted filesystem.
591 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
592 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
594 run_e2fsck $MMP_MDS_FAILOVER $MMP_MDSDEV "-fn"
597 # e2fsck is always called with -n, i.e.
598 # 0 (No errors) and 4 (File system errors left uncorrected) are the only acceptable
599 # e2fsck exit codes in case
600 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
601 error_noexit "e2fsck $MMP_MDSDEV on $MMP_MDS_FAILOVER returned $rc"
602 stop $MMP_MDS || return ${PIPESTATUS[0]}
606 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
607 start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS
609 if [ $rc -ne 0 ]; then
610 stop $MMP_MDS || return ${PIPESTATUS[0]}
614 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fn"
616 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
617 error_noexit "e2fsck $MMP_OSTDEV on $MMP_OSS_FAILOVER returned $rc"
620 stop_services primary || return ${PIPESTATUS[0]}
623 run_test 10 "e2fsck with mounted filesystem"
626 FAIL_ON_ERROR=$SAVED_FAIL_ON_ERROR
628 complete $(basename $0) $SECONDS
629 $MMP_RESTORE_MOUNT && setupall