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 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
28 remote_ost_nodsh && skip "remote OST with nodsh" && 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.
58 MMP_MDS=${MMP_MDS:-$SINGLEMDS}
59 MMP_MDS_FAILOVER=$(get_failover_facet $MMP_MDS)
61 local mds_num=$(echo $MMP_MDS | tr -d "mds")
62 MMP_MDSDEV=$(mdsdevname $mds_num)
64 MMP_OSS=${MMP_OSS:-ost1}
65 MMP_OSS_FAILOVER=$(get_failover_facet $MMP_OSS)
67 local oss_num=$(echo $MMP_OSS | tr -d "ost")
68 MMP_OSTDEV=$(ostdevname $oss_num)
71 # Stop the MDS and OSS services on the primary or failover servers.
79 if [ "$flavor" = "failover" ]; then
80 mds_facet=$MMP_MDS_FAILOVER
81 oss_facet=$MMP_OSS_FAILOVER
87 stop $mds_facet $opts || return ${PIPESTATUS[0]}
88 stop $oss_facet $opts || return ${PIPESTATUS[0]}
91 # Enable the MMP feature.
96 do_facet $facet "$TUNE2FS -O mmp $device"
97 return ${PIPESTATUS[0]}
100 # Disable the MMP feature.
105 do_facet $facet "$TUNE2FS -O ^mmp $device"
106 return ${PIPESTATUS[0]}
109 # Set the MMP block to 'fsck' state
114 do_facet $facet "$LUSTRE/tests/mmp_mark.sh $device"
115 return ${PIPESTATUS[0]}
118 # Reset the MMP block (if any) back to the clean state.
123 do_facet $facet "$TUNE2FS -f -E clear-mmp $device"
124 return ${PIPESTATUS[0]}
127 # Check whether the MMP feature is enabled or not.
132 do_facet $facet "$DUMPE2FS -h $device | grep mmp"
133 return ${PIPESTATUS[0]}
136 # Get MMP update interval (in seconds) from the Lustre server target.
137 get_mmp_update_interval() {
142 interval=$(do_facet $facet "$DEBUGFS -c -R dump_mmp $device 2>/dev/null \
143 | grep 'MMP Update Interval' | cut -d' ' -f4")
144 [ -z "$interval" ] && interval=1
149 # Get MMP check interval (in seconds) from the Lustre server target.
150 get_mmp_check_interval() {
155 interval=$(do_facet $facet "$DEBUGFS -c -R dump_mmp $device 2>/dev/null \
156 | grep 'MMP Check Interval' | cut -d' ' -f4")
157 [ -z "$interval" ] && interval=5
162 # Enable the MMP feature on the Lustre server targets.
166 # The MMP feature is automatically enabled by mkfs.lustre for
167 # new file system at format time if failover is being used.
168 # Otherwise, the Lustre administrator has to manually enable
169 # this feature when the file system is unmounted.
171 local var=${MMP_MDS}failover_HOST
172 if [ -z "${!var}" ]; then
173 log "Failover is not used on MDS, enabling MMP manually..."
174 enable_mmp $MMP_MDS $MMP_MDSDEV || \
175 error "failed to enable MMP on $MMP_MDSDEV on $MMP_MDS"
178 var=${MMP_OSS}failover_HOST
179 if [ -z "${!var}" ]; then
180 log "Failover is not used on OSS, enabling MMP manually..."
181 enable_mmp $MMP_OSS $MMP_OSTDEV || \
182 error "failed to enable MMP on $MMP_OSTDEV on $MMP_OSS"
185 # check whether the MMP feature is enabled or not
186 mmp_is_enabled $MMP_MDS $MMP_MDSDEV || \
187 error "MMP was not enabled on $MMP_MDSDEV on $MMP_MDS"
189 mmp_is_enabled $MMP_OSS $MMP_OSTDEV || \
190 error "MMP was not enabled on $MMP_OSTDEV on $MMP_OSS"
193 # Disable the MMP feature on the Lustre server targets
194 # which did not use failover.
197 local var=${MMP_MDS}failover_HOST
198 if [ -z "${!var}" ]; then
199 log "Failover is not used on MDS, disabling MMP manually..."
200 disable_mmp $MMP_MDS $MMP_MDSDEV || \
201 error "failed to disable MMP on $MMP_MDSDEV on $MMP_MDS"
202 mmp_is_enabled $MMP_MDS $MMP_MDSDEV && \
203 error "MMP was not disabled on $MMP_MDSDEV on $MMP_MDS"
206 var=${MMP_OSS}failover_HOST
207 if [ -z "${!var}" ]; then
208 log "Failover is not used on OSS, disabling MMP manually..."
209 disable_mmp $MMP_OSS $MMP_OSTDEV || \
210 error "failed to disable MMP on $MMP_OSTDEV on $MMP_OSS"
211 mmp_is_enabled $MMP_OSS $MMP_OSTDEV && \
212 error "MMP was not disabled on $MMP_OSTDEV on $MMP_OSS"
218 # Mount the shared target on the failover server after some interval it's
219 # mounted on the primary server.
220 mount_after_interval_sub() {
228 local failover_facet=$(get_failover_facet $facet)
231 local first_mount_rc=0
232 local second_mount_rc=0
234 log "Mounting $device on $facet..."
235 start $facet $device $opts &
238 if [ $interval -ne 0 ]; then
239 log "sleep $interval..."
243 log "Mounting $device on $failover_facet..."
244 start $failover_facet $device $opts
245 second_mount_rc=${PIPESTATUS[0]}
248 first_mount_rc=${PIPESTATUS[0]}
250 if [ $second_mount_rc -eq 0 -a $first_mount_rc -eq 0 ]; then
251 error_noexit "one mount delayed by mmp interval $interval should fail"
252 stop $facet || return ${PIPESTATUS[0]}
253 [ "$failover_facet" != "$facet" ] && stop $failover_facet || \
254 return ${PIPESTATUS[0]}
256 elif [ $second_mount_rc -ne 0 -a $first_mount_rc -ne 0 ]; then
257 error_noexit "failed to mount on the failover pair $facet,$failover_facet"
258 return $first_mount_rc
264 mount_after_interval() {
265 local mdt_interval=$1
266 local ost_interval=$2
269 mount_after_interval_sub $mdt_interval $MMP_MDSDEV $MMP_MDS \
270 $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
273 mount_after_interval_sub $ost_interval $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
275 if [ $rc -ne 0 ]; then
283 # Mount the shared target on the failover server
284 # during unmounting it on the primary server.
285 mount_during_unmount() {
291 local failover_facet=$(get_failover_facet $facet)
297 log "Mounting $device on $facet..."
298 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
300 log "Unmounting $device on $facet..."
304 log "Mounting $device on $failover_facet..."
305 start $failover_facet $device $mnt_opts
306 mount_rc=${PIPESTATUS[0]}
309 unmount_rc=${PIPESTATUS[0]}
311 if [ $mount_rc -eq 0 ]; then
312 error_noexit "mount during unmount of the first filesystem should fail"
313 stop $failover_facet || return ${PIPESTATUS[0]}
317 if [ $unmount_rc -ne 0 ]; then
318 error_noexit "unmount the $device on $facet should succeed"
325 # Mount the shared target on the failover server
326 # after clean unmounting it on the primary server.
327 mount_after_unmount() {
333 local failover_facet=$(get_failover_facet $facet)
335 log "Mounting $device on $facet..."
336 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
338 log "Unmounting $device on $facet..."
339 stop $facet || return ${PIPESTATUS[0]}
341 log "Mounting $device on $failover_facet..."
342 start $failover_facet $device $mnt_opts || return ${PIPESTATUS[0]}
347 # Mount the shared target on the failover server after rebooting
348 # the primary server.
349 mount_after_reboot() {
355 local failover_facet=$(get_failover_facet $facet)
358 log "Mounting $device on $facet..."
359 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
361 if [ "$FAILURE_MODE" = "HARD" ]; then
362 shutdown_facet $facet
364 wait_for_facet $facet
366 replay_barrier_nodf $facet
369 log "Mounting $device on $failover_facet..."
370 start $failover_facet $device $mnt_opts
372 if [ $rc -ne 0 ]; then
373 error_noexit "mount $device on $failover_facet should succeed"
374 stop $facet || return ${PIPESTATUS[0]}
381 # Run e2fsck on the Lustre server target.
389 log "Running e2fsck on the device $device on $facet..."
390 do_facet $facet "$E2FSCK $opts $device"
391 return ${PIPESTATUS[0]}
394 # Check whether there are failover pairs for MDS and OSS servers.
395 check_failover_pair() {
396 [ "$MMP_MDS" = "$MMP_MDS_FAILOVER" -o "$MMP_OSS" = "$MMP_OSS_FAILOVER" ] \
397 && { skip_env "failover pair is needed" && return 1; }
403 # Test 1 - two mounts at the same time.
405 check_failover_pair || return 0
407 mount_after_interval 0 0 || return ${PIPESTATUS[0]}
408 stop_services primary || return ${PIPESTATUS[0]}
410 run_test 1 "two mounts at the same time"
412 # Test 2 - one mount delayed by mmp update interval.
414 check_failover_pair || return 0
416 local mdt_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
417 local ost_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
419 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
420 stop_services primary || return ${PIPESTATUS[0]}
422 run_test 2 "one mount delayed by mmp update interval"
424 # Test 3 - one mount delayed by 2x mmp check interval.
426 check_failover_pair || return 0
428 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
429 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
431 mdt_interval=$((2 * $mdt_interval + 1))
432 ost_interval=$((2 * $ost_interval + 1))
434 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
435 stop_services primary || return ${PIPESTATUS[0]}
437 run_test 3 "one mount delayed by 2x mmp check interval"
439 # Test 4 - one mount delayed by > 2x mmp check interval.
441 check_failover_pair || return 0
443 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
444 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
446 mdt_interval=$((4 * $mdt_interval))
447 ost_interval=$((4 * $ost_interval))
449 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
450 stop_services primary || return ${PIPESTATUS[0]}
452 run_test 4 "one mount delayed by > 2x mmp check interval"
454 # Test 5 - mount during unmount of the first filesystem.
457 check_failover_pair || return 0
459 mount_during_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
460 return ${PIPESTATUS[0]}
463 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
464 mount_during_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
466 if [ $rc -ne 0 ]; then
467 stop $MMP_MDS || return ${PIPESTATUS[0]}
471 stop $MMP_MDS || return ${PIPESTATUS[0]}
473 run_test 5 "mount during unmount of the first filesystem"
475 # Test 6 - mount after clean unmount.
478 check_failover_pair || return 0
480 mount_after_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
481 return ${PIPESTATUS[0]}
484 mount_after_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
486 if [ $rc -ne 0 ]; then
487 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
491 stop_services failover || return ${PIPESTATUS[0]}
493 run_test 6 "mount after clean unmount"
495 # Test 7 - mount after reboot.
498 check_failover_pair || return 0
500 mount_after_reboot $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
501 return ${PIPESTATUS[0]}
504 mount_after_reboot $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
506 if [ $rc -ne 0 ]; then
507 stop $MMP_MDS || return ${PIPESTATUS[0]}
508 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
512 stop_services failover || return ${PIPESTATUS[0]}
513 stop_services primary || return ${PIPESTATUS[0]}
515 run_test 7 "mount after reboot"
517 # Test 8 - mount during e2fsck (should never succeed).
521 run_e2fsck $MMP_MDS $MMP_MDSDEV "-fy" &
525 log "Mounting $MMP_MDSDEV on $MMP_MDS_FAILOVER..."
526 if start $MMP_MDS_FAILOVER $MMP_MDSDEV $MDS_MOUNT_OPTS; then
527 error_noexit "mount $MMP_MDSDEV on $MMP_MDS_FAILOVER should fail"
528 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
535 run_e2fsck $MMP_OSS $MMP_OSTDEV "-fy" &
539 log "Mounting $MMP_OSTDEV on $MMP_OSS_FAILOVER..."
540 if start $MMP_OSS_FAILOVER $MMP_OSTDEV $OST_MOUNT_OPTS; then
541 error_noexit "mount $MMP_OSTDEV on $MMP_OSS_FAILOVER should fail"
542 stop $MMP_OSS_FAILOVER || return ${PIPESTATUS[0]}
549 run_test 8 "mount during e2fsck"
551 # Test 9 - mount after aborted e2fsck (should never succeed).
553 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
554 if ! start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
555 local rc=${PIPESTATUS[0]}
556 stop $MMP_MDS || return ${PIPESTATUS[0]}
559 stop_services primary || return ${PIPESTATUS[0]}
561 mark_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
563 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
564 if start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS; then
565 error_noexit "mount $MMP_MDSDEV on $MMP_MDS should fail"
566 stop $MMP_MDS || return ${PIPESTATUS[0]}
570 reset_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
572 mark_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
574 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
575 if start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
576 error_noexit "mount $MMP_OSTDEV on $MMP_OSS should fail"
577 stop $MMP_OSS || return ${PIPESTATUS[0]}
581 reset_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
584 run_test 9 "mount after aborted e2fsck"
586 # Test 10 - e2fsck with mounted filesystem.
590 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
591 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
593 run_e2fsck $MMP_MDS_FAILOVER $MMP_MDSDEV "-fn"
596 # e2fsck is called with -n option (Open the filesystem read-only), so
597 # 0 (No errors) and 4 (File system errors left uncorrected) are the only
598 # acceptable exit codes in this case
599 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
600 error_noexit "e2fsck $MMP_MDSDEV on $MMP_MDS_FAILOVER returned $rc"
601 stop $MMP_MDS || return ${PIPESTATUS[0]}
605 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
606 start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS
608 if [ $rc -ne 0 ]; then
609 stop $MMP_MDS || return ${PIPESTATUS[0]}
613 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fn"
615 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
616 error_noexit "e2fsck $MMP_OSTDEV on $MMP_OSS_FAILOVER returned $rc"
619 stop_services primary || return ${PIPESTATUS[0]}
622 run_test 10 "e2fsck with mounted filesystem"
625 FAIL_ON_ERROR=$SAVED_FAIL_ON_ERROR
627 complete $(basename $0) $SECONDS
628 $MMP_RESTORE_MOUNT && setupall