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 # Reset the MMP block (if any) back to the clean state.
114 do_facet $facet "$TUNE2FS -f -E clear-mmp $device"
115 return ${PIPESTATUS[0]}
118 # Check whether the MMP feature is enabled or not.
123 do_facet $facet "$DUMPE2FS -h $device | grep mmp"
124 return ${PIPESTATUS[0]}
127 # Get MMP update interval (in seconds) from the Lustre server target.
128 get_mmp_update_interval() {
133 interval=$(do_facet $facet "$DEBUGFS -c -R dump_mmp $device 2>/dev/null \
134 | grep 'MMP Update Interval' | cut -d' ' -f4")
135 [ -z "$interval" ] && interval=1
140 # Get MMP check interval (in seconds) from the Lustre server target.
141 get_mmp_check_interval() {
146 interval=$(do_facet $facet "$DEBUGFS -c -R dump_mmp $device 2>/dev/null \
147 | grep 'MMP Check Interval' | cut -d' ' -f4")
148 [ -z "$interval" ] && interval=5
153 # Enable the MMP feature on the Lustre server targets.
157 # The MMP feature is automatically enabled by mkfs.lustre for
158 # new file system at format time if failover is being used.
159 # Otherwise, the Lustre administrator has to manually enable
160 # this feature when the file system is unmounted.
162 local var=${MMP_MDS}failover_HOST
163 if [ -z "${!var}" ]; then
164 log "Failover is not used on MDS, enabling MMP manually..."
165 enable_mmp $MMP_MDS $MMP_MDSDEV || \
166 error "failed to enable MMP on $MMP_MDSDEV on $MMP_MDS"
169 var=${MMP_OSS}failover_HOST
170 if [ -z "${!var}" ]; then
171 log "Failover is not used on OSS, enabling MMP manually..."
172 enable_mmp $MMP_OSS $MMP_OSTDEV || \
173 error "failed to enable MMP on $MMP_OSTDEV on $MMP_OSS"
176 # check whether the MMP feature is enabled or not
177 mmp_is_enabled $MMP_MDS $MMP_MDSDEV || \
178 error "MMP was not enabled on $MMP_MDSDEV on $MMP_MDS"
180 mmp_is_enabled $MMP_OSS $MMP_OSTDEV || \
181 error "MMP was not enabled on $MMP_OSTDEV on $MMP_OSS"
184 # Disable the MMP feature on the Lustre server targets
185 # which did not use failover.
188 local var=${MMP_MDS}failover_HOST
189 if [ -z "${!var}" ]; then
190 log "Failover is not used on MDS, disabling MMP manually..."
191 disable_mmp $MMP_MDS $MMP_MDSDEV || \
192 error "failed to disable MMP on $MMP_MDSDEV on $MMP_MDS"
193 mmp_is_enabled $MMP_MDS $MMP_MDSDEV && \
194 error "MMP was not disabled on $MMP_MDSDEV on $MMP_MDS"
197 var=${MMP_OSS}failover_HOST
198 if [ -z "${!var}" ]; then
199 log "Failover is not used on OSS, disabling MMP manually..."
200 disable_mmp $MMP_OSS $MMP_OSTDEV || \
201 error "failed to disable MMP on $MMP_OSTDEV on $MMP_OSS"
202 mmp_is_enabled $MMP_OSS $MMP_OSTDEV && \
203 error "MMP was not disabled on $MMP_OSTDEV on $MMP_OSS"
209 # Mount the shared target on the failover server after some interval it's
210 # mounted on the primary server.
211 mount_after_interval_sub() {
219 local failover_facet=$(get_failover_facet $facet)
222 local first_mount_rc=0
223 local second_mount_rc=0
225 log "Mounting $device on $facet..."
226 start $facet $device $opts &
229 if [ $interval -ne 0 ]; then
230 log "sleep $interval..."
234 log "Mounting $device on $failover_facet..."
235 start $failover_facet $device $opts
236 second_mount_rc=${PIPESTATUS[0]}
239 first_mount_rc=${PIPESTATUS[0]}
241 if [ $second_mount_rc -eq 0 -a $first_mount_rc -eq 0 ]; then
242 error_noexit "one mount delayed by mmp interval $interval should fail"
243 stop $facet || return ${PIPESTATUS[0]}
244 [ "$failover_facet" != "$facet" ] && stop $failover_facet || \
245 return ${PIPESTATUS[0]}
247 elif [ $second_mount_rc -ne 0 -a $first_mount_rc -ne 0 ]; then
248 error_noexit "failed to mount on the failover pair $facet,$failover_facet"
249 return $first_mount_rc
255 mount_after_interval() {
256 local mdt_interval=$1
257 local ost_interval=$2
260 mount_after_interval_sub $mdt_interval $MMP_MDSDEV $MMP_MDS \
261 $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
264 mount_after_interval_sub $ost_interval $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
266 if [ $rc -ne 0 ]; then
274 # Mount the shared target on the failover server
275 # during unmounting it on the primary server.
276 mount_during_unmount() {
282 local failover_facet=$(get_failover_facet $facet)
288 log "Mounting $device on $facet..."
289 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
291 log "Unmounting $device on $facet..."
295 log "Mounting $device on $failover_facet..."
296 start $failover_facet $device $mnt_opts
297 mount_rc=${PIPESTATUS[0]}
300 unmount_rc=${PIPESTATUS[0]}
302 if [ $mount_rc -eq 0 ]; then
303 error_noexit "mount during unmount of the first filesystem should fail"
304 stop $failover_facet || return ${PIPESTATUS[0]}
308 if [ $unmount_rc -ne 0 ]; then
309 error_noexit "unmount the $device on $facet should succeed"
316 # Mount the shared target on the failover server
317 # after clean unmounting it on the primary server.
318 mount_after_unmount() {
324 local failover_facet=$(get_failover_facet $facet)
326 log "Mounting $device on $facet..."
327 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
329 log "Unmounting $device on $facet..."
330 stop $facet || return ${PIPESTATUS[0]}
332 log "Mounting $device on $failover_facet..."
333 start $failover_facet $device $mnt_opts || return ${PIPESTATUS[0]}
338 # Mount the shared target on the failover server after rebooting
339 # the primary server.
340 mount_after_reboot() {
346 local failover_facet=$(get_failover_facet $facet)
349 log "Mounting $device on $facet..."
350 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
352 if [ "$FAILURE_MODE" = "HARD" ]; then
353 shutdown_facet $facet
355 wait_for_facet $facet
357 replay_barrier_nodf $facet
360 log "Mounting $device on $failover_facet..."
361 start $failover_facet $device $mnt_opts
363 if [ $rc -ne 0 ]; then
364 error_noexit "mount $device on $failover_facet should succeed"
365 stop $facet || return ${PIPESTATUS[0]}
372 # Run e2fsck on the Lustre server target.
380 log "Running e2fsck on the device $device on $facet..."
381 do_facet $facet "$E2FSCK $opts $device"
382 return ${PIPESTATUS[0]}
385 # Check whether there are failover pairs for MDS and OSS servers.
386 check_failover_pair() {
387 [ "$MMP_MDS" = "$MMP_MDS_FAILOVER" -o "$MMP_OSS" = "$MMP_OSS_FAILOVER" ] \
388 && { skip_env "failover pair is needed" && return 1; }
394 # Test 1 - two mounts at the same time.
396 check_failover_pair || return 0
398 mount_after_interval 0 0 || return ${PIPESTATUS[0]}
399 stop_services primary || return ${PIPESTATUS[0]}
401 run_test 1 "two mounts at the same time"
403 # Test 2 - one mount delayed by mmp update interval.
405 check_failover_pair || return 0
407 local mdt_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
408 local ost_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
410 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
411 stop_services primary || return ${PIPESTATUS[0]}
413 run_test 2 "one mount delayed by mmp update interval"
415 # Test 3 - one mount delayed by 2x mmp check interval.
417 check_failover_pair || return 0
419 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
420 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
422 mdt_interval=$((2 * $mdt_interval + 1))
423 ost_interval=$((2 * $ost_interval + 1))
425 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
426 stop_services primary || return ${PIPESTATUS[0]}
428 run_test 3 "one mount delayed by 2x mmp check interval"
430 # Test 4 - one mount delayed by > 2x mmp check interval.
432 check_failover_pair || return 0
434 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
435 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
437 mdt_interval=$((4 * $mdt_interval))
438 ost_interval=$((4 * $ost_interval))
440 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
441 stop_services primary || return ${PIPESTATUS[0]}
443 run_test 4 "one mount delayed by > 2x mmp check interval"
445 # Test 5 - mount during unmount of the first filesystem.
448 check_failover_pair || return 0
450 mount_during_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
451 return ${PIPESTATUS[0]}
454 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
455 mount_during_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
457 if [ $rc -ne 0 ]; then
458 stop $MMP_MDS || return ${PIPESTATUS[0]}
462 stop $MMP_MDS || return ${PIPESTATUS[0]}
464 run_test 5 "mount during unmount of the first filesystem"
466 # Test 6 - mount after clean unmount.
469 check_failover_pair || return 0
471 mount_after_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
472 return ${PIPESTATUS[0]}
475 mount_after_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
477 if [ $rc -ne 0 ]; then
478 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
482 stop_services failover || return ${PIPESTATUS[0]}
484 run_test 6 "mount after clean unmount"
486 # Test 7 - mount after reboot.
489 check_failover_pair || return 0
491 mount_after_reboot $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
492 return ${PIPESTATUS[0]}
495 mount_after_reboot $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
497 if [ $rc -ne 0 ]; then
498 stop $MMP_MDS || return ${PIPESTATUS[0]}
499 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
503 stop_services failover || return ${PIPESTATUS[0]}
504 stop_services primary || return ${PIPESTATUS[0]}
506 run_test 7 "mount after reboot"
508 # Test 8 - mount during e2fsck (should never succeed).
512 run_e2fsck $MMP_MDS $MMP_MDSDEV "-fy" &
516 log "Mounting $MMP_MDSDEV on $MMP_MDS_FAILOVER..."
517 if start $MMP_MDS_FAILOVER $MMP_MDSDEV $MDS_MOUNT_OPTS; then
518 error_noexit "mount $MMP_MDSDEV on $MMP_MDS_FAILOVER should fail"
519 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
526 run_e2fsck $MMP_OSS $MMP_OSTDEV "-fy" &
530 log "Mounting $MMP_OSTDEV on $MMP_OSS_FAILOVER..."
531 if start $MMP_OSS_FAILOVER $MMP_OSTDEV $OST_MOUNT_OPTS; then
532 error_noexit "mount $MMP_OSTDEV on $MMP_OSS_FAILOVER should fail"
533 stop $MMP_OSS_FAILOVER || return ${PIPESTATUS[0]}
540 run_test 8 "mount during e2fsck"
542 # Test 9 - mount after aborted e2fsck (should never succeed).
545 local mdt_mmp_check_interval
546 local ost_mmp_check_interval
548 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
549 if ! start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
550 local rc=${PIPESTATUS[0]}
551 stop $MMP_MDS || return ${PIPESTATUS[0]}
554 stop_services primary || return ${PIPESTATUS[0]}
556 mdt_mmp_check_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
557 run_e2fsck $MMP_MDS_FAILOVER $MMP_MDSDEV "-fy" &
559 sleep $((2 * $mdt_mmp_check_interval))
560 kill -s ABRT $e2fsck_pid
562 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
563 if start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS; then
564 error_noexit "mount $MMP_MDSDEV on $MMP_MDS should fail"
565 stop $MMP_MDS || return ${PIPESTATUS[0]}
569 reset_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
572 ost_mmp_check_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
573 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fy" &
575 sleep $((2 * $ost_mmp_check_interval))
576 kill -s ABRT $e2fsck_pid
578 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
579 if start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
580 error_noexit "mount $MMP_OSTDEV on $MMP_OSS should fail"
581 stop $MMP_OSS || return ${PIPESTATUS[0]}
585 reset_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
588 run_test 9 "mount after aborted e2fsck"
590 # Test 10 - e2fsck with mounted filesystem.
594 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
595 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
597 run_e2fsck $MMP_MDS_FAILOVER $MMP_MDSDEV "-fn"
600 # e2fsck is called with -n option (Open the filesystem read-only), so
601 # 0 (No errors) and 4 (File system errors left uncorrected) are the only
602 # acceptable exit codes in this case
603 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
604 error_noexit "e2fsck $MMP_MDSDEV on $MMP_MDS_FAILOVER returned $rc"
605 stop $MMP_MDS || return ${PIPESTATUS[0]}
609 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
610 start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS
612 if [ $rc -ne 0 ]; then
613 stop $MMP_MDS || return ${PIPESTATUS[0]}
617 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fn"
619 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
620 error_noexit "e2fsck $MMP_OSTDEV on $MMP_OSS_FAILOVER returned $rc"
623 stop_services primary || return ${PIPESTATUS[0]}
626 run_test 10 "e2fsck with mounted filesystem"
629 FAIL_ON_ERROR=$SAVED_FAIL_ON_ERROR
631 complete $(basename $0) $SECONDS
632 $MMP_RESTORE_MOUNT && setupall