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 $device 2>$TMP/mmp.debugfs.msg" |
145 awk 'tolower($0) ~ /update.interval/ { print $NF }')
146 [ -z "$interval" ] && interval=5 &&
147 do_facet $facet cat $TMP/mmp.debugfs.msg &&
148 echo "$facet:$device: assume update interval=$interval" 1>&2 ||
149 echo "$facet:$device: got actual update interval=$interval" 1>&2
154 # Get MMP check interval (in seconds) from the Lustre server target.
155 get_mmp_check_interval() {
160 interval=$(do_facet $facet \
161 "$DEBUGFS -c -R dump_mmp $device 2>$TMP/mmp.debugfs.msg" |
162 awk 'tolower($0) ~ /check.interval/ { print $NF }')
163 [ -z "$interval" ] && interval=5 &&
164 do_facet $facet cat $TMP/mmp.debugfs.msg &&
165 echo "$facet:$device: assume check interval=$interval" 1>&2 ||
166 echo "$facet:$device: got actual check interval=$interval" 1>&2
171 # Adjust the MMP update interval (in seconds) on the Lustre server target.
172 # Specifying an interval of 0 means to use the default interval.
173 set_mmp_update_interval() {
176 local interval=${3:-0}
178 do_facet $facet "$TUNE2FS -E mmp_update_interval=$interval $device"
179 return ${PIPESTATUS[0]}
182 # Enable the MMP feature on the Lustre server targets.
186 if [ $(facet_fstype $MMP_MDS) != ldiskfs ]; then
187 skip "Only applicable to ldiskfs-based MDTs"
191 if [ $(facet_fstype $MMP_OSS) != ldiskfs ]; then
192 skip "Only applicable to ldiskfs-based OSTs"
196 # The MMP feature is automatically enabled by mkfs.lustre for
197 # new file system at format time if failover is being used.
198 # Otherwise, the Lustre administrator has to manually enable
199 # this feature when the file system is unmounted.
201 local var=${MMP_MDS}failover_HOST
202 if [ -z "${!var}" ]; then
203 log "Failover is not used on MDS, enabling MMP manually..."
204 enable_mmp $MMP_MDS $MMP_MDSDEV || \
205 error "failed to enable MMP on $MMP_MDSDEV on $MMP_MDS"
208 var=${MMP_OSS}failover_HOST
209 if [ -z "${!var}" ]; then
210 log "Failover is not used on OSS, enabling MMP manually..."
211 enable_mmp $MMP_OSS $MMP_OSTDEV || \
212 error "failed to enable MMP on $MMP_OSTDEV on $MMP_OSS"
215 # check whether the MMP feature is enabled or not
216 mmp_is_enabled $MMP_MDS $MMP_MDSDEV || \
217 error "MMP was not enabled on $MMP_MDSDEV on $MMP_MDS"
219 mmp_is_enabled $MMP_OSS $MMP_OSTDEV || \
221 do_facet $MMP_OSS "$DUMPE2FS -h $MMP_OSTDEV"
222 log "Try to enable again:"
223 enable_mmp $MMP_OSS $MMP_OSTDEV
225 do_facet $MMP_OSS "$DUMPE2FS -h $MMP_OSTDEV"
226 error "MMP was not enabled on $MMP_OSTDEV on $MMP_OSS"
230 # Disable the MMP feature on the Lustre server targets
231 # which did not use failover.
234 local var=${MMP_MDS}failover_HOST
235 if [ -z "${!var}" ]; then
236 log "Failover is not used on MDS, disabling MMP manually..."
237 disable_mmp $MMP_MDS $MMP_MDSDEV || \
238 error "failed to disable MMP on $MMP_MDSDEV on $MMP_MDS"
239 mmp_is_enabled $MMP_MDS $MMP_MDSDEV && \
240 error "MMP was not disabled on $MMP_MDSDEV on $MMP_MDS"
243 var=${MMP_OSS}failover_HOST
244 if [ -z "${!var}" ]; then
245 log "Failover is not used on OSS, disabling MMP manually..."
246 disable_mmp $MMP_OSS $MMP_OSTDEV || \
247 error "failed to disable MMP on $MMP_OSTDEV on $MMP_OSS"
248 mmp_is_enabled $MMP_OSS $MMP_OSTDEV && \
249 error "MMP was not disabled on $MMP_OSTDEV on $MMP_OSS"
255 # Mount the shared target on the failover server after some interval it's
256 # mounted on the primary server.
257 mount_after_interval_sub() {
265 local failover_facet=$(get_failover_facet $facet)
268 local first_mount_rc=0
269 local second_mount_rc=0
271 log "Mounting $device on $facet..."
272 start $facet $device $opts &
275 if [ $interval -ne 0 ]; then
276 log "sleep $interval..."
280 log "Mounting $device on $failover_facet..."
281 start $failover_facet $device $opts
282 second_mount_rc=${PIPESTATUS[0]}
285 first_mount_rc=${PIPESTATUS[0]}
287 if [ $second_mount_rc -eq 0 -a $first_mount_rc -eq 0 ]; then
288 error_noexit "one mount delayed by mmp interval $interval should fail"
289 stop $facet || return ${PIPESTATUS[0]}
290 [ "$failover_facet" != "$facet" ] && stop $failover_facet || \
291 return ${PIPESTATUS[0]}
293 elif [ $second_mount_rc -ne 0 -a $first_mount_rc -ne 0 ]; then
294 error_noexit "mount failure on failover pair $facet,$failover_facet"
295 return $first_mount_rc
301 mount_after_interval() {
302 local mdt_interval=$1
303 local ost_interval=$2
306 mount_after_interval_sub $mdt_interval $MMP_MDSDEV $MMP_MDS \
307 $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
310 mount_after_interval_sub $ost_interval $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
312 if [ $rc -ne 0 ]; then
320 # Mount the shared target on the failover server
321 # during unmounting it on the primary server.
322 mount_during_unmount() {
328 local failover_facet=$(get_failover_facet $facet)
334 log "Mounting $device on $facet..."
335 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
337 log "Unmounting $device on $facet..."
341 log "Mounting $device on $failover_facet..."
342 start $failover_facet $device $mnt_opts
343 mount_rc=${PIPESTATUS[0]}
346 unmount_rc=${PIPESTATUS[0]}
348 if [ $mount_rc -eq 0 ]; then
349 error_noexit "mount during unmount of the first filesystem should fail"
350 stop $failover_facet || return ${PIPESTATUS[0]}
354 if [ $unmount_rc -ne 0 ]; then
355 error_noexit "unmount the $device on $facet should succeed"
362 # Mount the shared target on the failover server
363 # after clean unmounting it on the primary server.
364 mount_after_unmount() {
370 local failover_facet=$(get_failover_facet $facet)
372 log "Mounting $device on $facet..."
373 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
375 log "Unmounting $device on $facet..."
376 stop $facet || return ${PIPESTATUS[0]}
378 log "Mounting $device on $failover_facet..."
379 start $failover_facet $device $mnt_opts || return ${PIPESTATUS[0]}
384 # Mount the shared target on the failover server after rebooting
385 # the primary server.
386 mount_after_reboot() {
392 local failover_facet=$(get_failover_facet $facet)
395 log "Mounting $device on $facet..."
396 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
398 if [ "$FAILURE_MODE" = "HARD" ]; then
399 shutdown_facet $facet
401 wait_for_facet $facet
403 replay_barrier_nodf $facet
406 log "Mounting $device on $failover_facet..."
407 start $failover_facet $device $mnt_opts
409 if [ $rc -ne 0 ]; then
410 error_noexit "mount $device on $failover_facet should succeed"
411 stop $facet || return ${PIPESTATUS[0]}
418 # Run e2fsck on the Lustre server target.
426 echo "Running e2fsck on the device $device on $facet..."
427 do_facet $facet "$E2FSCK $opts $device"
428 return ${PIPESTATUS[0]}
431 # Check whether there are failover pairs for MDS and OSS servers.
432 check_failover_pair() {
433 [ "$MMP_MDS" = "$MMP_MDS_FAILOVER" -o "$MMP_OSS" = "$MMP_OSS_FAILOVER" ] \
434 && { skip_env "failover pair is needed" && return 1; }
440 # Test 1 - two mounts at the same time.
442 check_failover_pair || return 0
444 mount_after_interval 0 0 || return ${PIPESTATUS[0]}
445 stop_services primary || return ${PIPESTATUS[0]}
447 run_test 1 "two mounts at the same time"
449 # Test 2 - one mount delayed by mmp update interval.
451 check_failover_pair || return 0
453 local mdt_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
454 local ost_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
456 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
457 stop_services primary || return ${PIPESTATUS[0]}
459 run_test 2 "one mount delayed by mmp update interval"
461 # Test 3 - one mount delayed by 2x mmp check interval.
463 check_failover_pair || return 0
465 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
466 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
468 mdt_interval=$((2 * $mdt_interval + 1))
469 ost_interval=$((2 * $ost_interval + 1))
471 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
472 stop_services primary || return ${PIPESTATUS[0]}
474 run_test 3 "one mount delayed by 2x mmp check interval"
476 # Test 4 - one mount delayed by > 2x mmp check interval.
478 check_failover_pair || return 0
480 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
481 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
483 mdt_interval=$((4 * $mdt_interval))
484 ost_interval=$((4 * $ost_interval))
486 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
487 stop_services primary || return ${PIPESTATUS[0]}
489 run_test 4 "one mount delayed by > 2x mmp check interval"
491 # Test 5 - mount during unmount of the first filesystem.
494 check_failover_pair || return 0
496 mount_during_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
497 return ${PIPESTATUS[0]}
500 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
501 mount_during_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
503 if [ $rc -ne 0 ]; then
504 stop $MMP_MDS || return ${PIPESTATUS[0]}
508 stop $MMP_MDS || return ${PIPESTATUS[0]}
510 run_test 5 "mount during unmount of the first filesystem"
512 # Test 6 - mount after clean unmount.
515 check_failover_pair || return 0
517 mount_after_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
518 return ${PIPESTATUS[0]}
521 mount_after_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
523 if [ $rc -ne 0 ]; then
524 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
528 stop_services failover || return ${PIPESTATUS[0]}
530 run_test 6 "mount after clean unmount"
532 # Test 7 - mount after reboot.
535 check_failover_pair || return 0
537 mount_after_reboot $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
538 return ${PIPESTATUS[0]}
541 mount_after_reboot $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
543 if [ $rc -ne 0 ]; then
544 stop $MMP_MDS || return ${PIPESTATUS[0]}
545 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
549 stop_services failover || return ${PIPESTATUS[0]}
550 stop_services primary || return ${PIPESTATUS[0]}
552 run_test 7 "mount after reboot"
554 # Test 8 - mount during e2fsck (should never succeed).
560 # After writing a new sequence number into the MMP block, e2fsck will
561 # sleep at least (2 * new_interval + 1) seconds before it goes into
566 saved_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
567 set_mmp_update_interval $MMP_MDS $MMP_MDSDEV $new_interval
569 run_e2fsck $MMP_MDS $MMP_MDSDEV "-fy" &
573 if start $MMP_MDS_FAILOVER $MMP_MDSDEV $MDS_MOUNT_OPTS; then
575 "mount $MMP_MDSDEV on $MMP_MDS_FAILOVER should fail"
576 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
577 set_mmp_update_interval $MMP_MDS $MMP_MDSDEV $saved_interval
582 set_mmp_update_interval $MMP_MDS $MMP_MDSDEV $saved_interval
586 saved_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
587 set_mmp_update_interval $MMP_OSS $MMP_OSTDEV $new_interval
589 run_e2fsck $MMP_OSS $MMP_OSTDEV "-fy" &
593 if start $MMP_OSS_FAILOVER $MMP_OSTDEV $OST_MOUNT_OPTS; then
595 "mount $MMP_OSTDEV on $MMP_OSS_FAILOVER should fail"
596 stop $MMP_OSS_FAILOVER || return ${PIPESTATUS[0]}
597 set_mmp_update_interval $MMP_OSS $MMP_OSTDEV $saved_interval
602 set_mmp_update_interval $MMP_OSS $MMP_OSTDEV $saved_interval
605 run_test 8 "mount during e2fsck"
607 # Test 9 - mount after aborted e2fsck (should never succeed).
609 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
610 if ! start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
611 local rc=${PIPESTATUS[0]}
612 stop $MMP_MDS || return ${PIPESTATUS[0]}
615 stop_services primary || return ${PIPESTATUS[0]}
617 mark_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
619 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
620 if start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS; then
621 error_noexit "mount $MMP_MDSDEV on $MMP_MDS should fail"
622 stop $MMP_MDS || return ${PIPESTATUS[0]}
626 reset_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
628 mark_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
630 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
631 if start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
632 error_noexit "mount $MMP_OSTDEV on $MMP_OSS should fail"
633 stop $MMP_OSS || return ${PIPESTATUS[0]}
637 reset_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
640 run_test 9 "mount after aborted e2fsck"
642 # Test 10 - e2fsck with mounted filesystem.
646 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
647 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
649 run_e2fsck $MMP_MDS_FAILOVER $MMP_MDSDEV "-fn"
652 # e2fsck is called with -n option (Open the filesystem read-only), so
653 # 0 (No errors) and 4 (File system errors left uncorrected) are the only
654 # acceptable exit codes in this case
655 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
656 error_noexit "e2fsck $MMP_MDSDEV on $MMP_MDS_FAILOVER returned $rc"
657 stop $MMP_MDS || return ${PIPESTATUS[0]}
661 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
662 start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS
664 if [ $rc -ne 0 ]; then
665 stop $MMP_MDS || return ${PIPESTATUS[0]}
669 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fn"
671 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
672 error_noexit "e2fsck $MMP_OSTDEV on $MMP_OSS_FAILOVER returned $rc"
675 stop_services primary || return ${PIPESTATUS[0]}
678 run_test 10 "e2fsck with mounted filesystem"
681 FAIL_ON_ERROR=$SAVED_FAIL_ON_ERROR
684 $MMP_RESTORE_MOUNT && setupall