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 || \
220 error "MMP was not enabled on $MMP_OSTDEV on $MMP_OSS"
223 # Disable the MMP feature on the Lustre server targets
224 # which did not use failover.
227 local var=${MMP_MDS}failover_HOST
228 if [ -z "${!var}" ]; then
229 log "Failover is not used on MDS, disabling MMP manually..."
230 disable_mmp $MMP_MDS $MMP_MDSDEV || \
231 error "failed to disable MMP on $MMP_MDSDEV on $MMP_MDS"
232 mmp_is_enabled $MMP_MDS $MMP_MDSDEV && \
233 error "MMP was not disabled on $MMP_MDSDEV on $MMP_MDS"
236 var=${MMP_OSS}failover_HOST
237 if [ -z "${!var}" ]; then
238 log "Failover is not used on OSS, disabling MMP manually..."
239 disable_mmp $MMP_OSS $MMP_OSTDEV || \
240 error "failed to disable MMP on $MMP_OSTDEV on $MMP_OSS"
241 mmp_is_enabled $MMP_OSS $MMP_OSTDEV && \
242 error "MMP was not disabled on $MMP_OSTDEV on $MMP_OSS"
248 # Mount the shared target on the failover server after some interval it's
249 # mounted on the primary server.
250 mount_after_interval_sub() {
258 local failover_facet=$(get_failover_facet $facet)
261 local first_mount_rc=0
262 local second_mount_rc=0
264 log "Mounting $device on $facet..."
265 start $facet $device $opts &
268 if [ $interval -ne 0 ]; then
269 log "sleep $interval..."
273 log "Mounting $device on $failover_facet..."
274 start $failover_facet $device $opts
275 second_mount_rc=${PIPESTATUS[0]}
278 first_mount_rc=${PIPESTATUS[0]}
280 if [ $second_mount_rc -eq 0 -a $first_mount_rc -eq 0 ]; then
281 error_noexit "one mount delayed by mmp interval $interval should fail"
282 stop $facet || return ${PIPESTATUS[0]}
283 [ "$failover_facet" != "$facet" ] && stop $failover_facet || \
284 return ${PIPESTATUS[0]}
286 elif [ $second_mount_rc -ne 0 -a $first_mount_rc -ne 0 ]; then
287 error_noexit "mount failure on failover pair $facet,$failover_facet"
288 return $first_mount_rc
294 mount_after_interval() {
295 local mdt_interval=$1
296 local ost_interval=$2
299 mount_after_interval_sub $mdt_interval $MMP_MDSDEV $MMP_MDS \
300 $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
303 mount_after_interval_sub $ost_interval $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
305 if [ $rc -ne 0 ]; then
313 # Mount the shared target on the failover server
314 # during unmounting it on the primary server.
315 mount_during_unmount() {
321 local failover_facet=$(get_failover_facet $facet)
327 log "Mounting $device on $facet..."
328 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
330 log "Unmounting $device on $facet..."
334 log "Mounting $device on $failover_facet..."
335 start $failover_facet $device $mnt_opts
336 mount_rc=${PIPESTATUS[0]}
339 unmount_rc=${PIPESTATUS[0]}
341 if [ $mount_rc -eq 0 ]; then
342 error_noexit "mount during unmount of the first filesystem should fail"
343 stop $failover_facet || return ${PIPESTATUS[0]}
347 if [ $unmount_rc -ne 0 ]; then
348 error_noexit "unmount the $device on $facet should succeed"
355 # Mount the shared target on the failover server
356 # after clean unmounting it on the primary server.
357 mount_after_unmount() {
363 local failover_facet=$(get_failover_facet $facet)
365 log "Mounting $device on $facet..."
366 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
368 log "Unmounting $device on $facet..."
369 stop $facet || return ${PIPESTATUS[0]}
371 log "Mounting $device on $failover_facet..."
372 start $failover_facet $device $mnt_opts || return ${PIPESTATUS[0]}
377 # Mount the shared target on the failover server after rebooting
378 # the primary server.
379 mount_after_reboot() {
385 local failover_facet=$(get_failover_facet $facet)
388 log "Mounting $device on $facet..."
389 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
391 if [ "$FAILURE_MODE" = "HARD" ]; then
392 shutdown_facet $facet
394 wait_for_facet $facet
396 replay_barrier_nodf $facet
399 log "Mounting $device on $failover_facet..."
400 start $failover_facet $device $mnt_opts
402 if [ $rc -ne 0 ]; then
403 error_noexit "mount $device on $failover_facet should succeed"
404 stop $facet || return ${PIPESTATUS[0]}
411 # Run e2fsck on the Lustre server target.
419 echo "Running e2fsck on the device $device on $facet..."
420 do_facet $facet "$E2FSCK $opts $device"
421 return ${PIPESTATUS[0]}
424 # Check whether there are failover pairs for MDS and OSS servers.
425 check_failover_pair() {
426 [ "$MMP_MDS" = "$MMP_MDS_FAILOVER" -o "$MMP_OSS" = "$MMP_OSS_FAILOVER" ] \
427 && { skip_env "failover pair is needed" && return 1; }
433 # Test 1 - two mounts at the same time.
435 check_failover_pair || return 0
437 mount_after_interval 0 0 || return ${PIPESTATUS[0]}
438 stop_services primary || return ${PIPESTATUS[0]}
440 run_test 1 "two mounts at the same time"
442 # Test 2 - one mount delayed by mmp update interval.
444 check_failover_pair || return 0
446 local mdt_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
447 local ost_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
449 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
450 stop_services primary || return ${PIPESTATUS[0]}
452 run_test 2 "one mount delayed by mmp update interval"
454 # Test 3 - one mount delayed by 2x mmp check interval.
456 check_failover_pair || return 0
458 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
459 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
461 mdt_interval=$((2 * $mdt_interval + 1))
462 ost_interval=$((2 * $ost_interval + 1))
464 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
465 stop_services primary || return ${PIPESTATUS[0]}
467 run_test 3 "one mount delayed by 2x mmp check interval"
469 # Test 4 - one mount delayed by > 2x mmp check interval.
471 check_failover_pair || return 0
473 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
474 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
476 mdt_interval=$((4 * $mdt_interval))
477 ost_interval=$((4 * $ost_interval))
479 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
480 stop_services primary || return ${PIPESTATUS[0]}
482 run_test 4 "one mount delayed by > 2x mmp check interval"
484 # Test 5 - mount during unmount of the first filesystem.
487 check_failover_pair || return 0
489 mount_during_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
490 return ${PIPESTATUS[0]}
493 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
494 mount_during_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
496 if [ $rc -ne 0 ]; then
497 stop $MMP_MDS || return ${PIPESTATUS[0]}
501 stop $MMP_MDS || return ${PIPESTATUS[0]}
503 run_test 5 "mount during unmount of the first filesystem"
505 # Test 6 - mount after clean unmount.
508 check_failover_pair || return 0
510 mount_after_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
511 return ${PIPESTATUS[0]}
514 mount_after_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
516 if [ $rc -ne 0 ]; then
517 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
521 stop_services failover || return ${PIPESTATUS[0]}
523 run_test 6 "mount after clean unmount"
525 # Test 7 - mount after reboot.
528 check_failover_pair || return 0
530 mount_after_reboot $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
531 return ${PIPESTATUS[0]}
534 mount_after_reboot $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
536 if [ $rc -ne 0 ]; then
537 stop $MMP_MDS || return ${PIPESTATUS[0]}
538 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
542 stop_services failover || return ${PIPESTATUS[0]}
543 stop_services primary || return ${PIPESTATUS[0]}
545 run_test 7 "mount after reboot"
547 # Test 8 - mount during e2fsck (should never succeed).
553 # After writing a new sequence number into the MMP block, e2fsck will
554 # sleep at least (2 * new_interval + 1) seconds before it goes into
559 saved_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
560 set_mmp_update_interval $MMP_MDS $MMP_MDSDEV $new_interval
562 run_e2fsck $MMP_MDS $MMP_MDSDEV "-fy" &
566 if start $MMP_MDS_FAILOVER $MMP_MDSDEV $MDS_MOUNT_OPTS; then
568 "mount $MMP_MDSDEV on $MMP_MDS_FAILOVER should fail"
569 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
570 set_mmp_update_interval $MMP_MDS $MMP_MDSDEV $saved_interval
575 set_mmp_update_interval $MMP_MDS $MMP_MDSDEV $saved_interval
579 saved_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
580 set_mmp_update_interval $MMP_OSS $MMP_OSTDEV $new_interval
582 run_e2fsck $MMP_OSS $MMP_OSTDEV "-fy" &
586 if start $MMP_OSS_FAILOVER $MMP_OSTDEV $OST_MOUNT_OPTS; then
588 "mount $MMP_OSTDEV on $MMP_OSS_FAILOVER should fail"
589 stop $MMP_OSS_FAILOVER || return ${PIPESTATUS[0]}
590 set_mmp_update_interval $MMP_OSS $MMP_OSTDEV $saved_interval
595 set_mmp_update_interval $MMP_OSS $MMP_OSTDEV $saved_interval
598 run_test 8 "mount during e2fsck"
600 # Test 9 - mount after aborted e2fsck (should never succeed).
602 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
603 if ! start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
604 local rc=${PIPESTATUS[0]}
605 stop $MMP_MDS || return ${PIPESTATUS[0]}
608 stop_services primary || return ${PIPESTATUS[0]}
610 mark_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
612 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
613 if start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS; then
614 error_noexit "mount $MMP_MDSDEV on $MMP_MDS should fail"
615 stop $MMP_MDS || return ${PIPESTATUS[0]}
619 reset_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
621 mark_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
623 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
624 if start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
625 error_noexit "mount $MMP_OSTDEV on $MMP_OSS should fail"
626 stop $MMP_OSS || return ${PIPESTATUS[0]}
630 reset_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
633 run_test 9 "mount after aborted e2fsck"
635 # Test 10 - e2fsck with mounted filesystem.
639 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
640 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
642 run_e2fsck $MMP_MDS_FAILOVER $MMP_MDSDEV "-fn"
645 # e2fsck is called with -n option (Open the filesystem read-only), so
646 # 0 (No errors) and 4 (File system errors left uncorrected) are the only
647 # acceptable exit codes in this case
648 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
649 error_noexit "e2fsck $MMP_MDSDEV on $MMP_MDS_FAILOVER returned $rc"
650 stop $MMP_MDS || return ${PIPESTATUS[0]}
654 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
655 start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS
657 if [ $rc -ne 0 ]; then
658 stop $MMP_MDS || return ${PIPESTATUS[0]}
662 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fn"
664 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
665 error_noexit "e2fsck $MMP_OSTDEV on $MMP_OSS_FAILOVER returned $rc"
668 stop_services primary || return ${PIPESTATUS[0]}
671 run_test 10 "e2fsck with mounted filesystem"
674 FAIL_ON_ERROR=$SAVED_FAIL_ON_ERROR
677 $MMP_RESTORE_MOUNT && setupall