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 $facet "$DEBUGFS -c -R dump_mmp $device 2>/dev/null \
144 | grep 'MMP Update Interval' | cut -d' ' -f4")
145 [ -z "$interval" ] && interval=1
150 # Get MMP check interval (in seconds) from the Lustre server target.
151 get_mmp_check_interval() {
156 interval=$(do_facet $facet "$DEBUGFS -c -R dump_mmp $device 2>/dev/null \
157 | grep 'MMP Check Interval' | cut -d' ' -f4")
158 [ -z "$interval" ] && interval=5
163 # Enable the MMP feature on the Lustre server targets.
167 if [ $(facet_fstype $MMP_MDS) != ldiskfs ]; then
168 skip "Only applicable to ldiskfs-based MDTs"
172 if [ $(facet_fstype $MMP_OSS) != ldiskfs ]; then
173 skip "Only applicable to ldiskfs-based OSTs"
177 # The MMP feature is automatically enabled by mkfs.lustre for
178 # new file system at format time if failover is being used.
179 # Otherwise, the Lustre administrator has to manually enable
180 # this feature when the file system is unmounted.
182 local var=${MMP_MDS}failover_HOST
183 if [ -z "${!var}" ]; then
184 log "Failover is not used on MDS, enabling MMP manually..."
185 enable_mmp $MMP_MDS $MMP_MDSDEV || \
186 error "failed to enable MMP on $MMP_MDSDEV on $MMP_MDS"
189 var=${MMP_OSS}failover_HOST
190 if [ -z "${!var}" ]; then
191 log "Failover is not used on OSS, enabling MMP manually..."
192 enable_mmp $MMP_OSS $MMP_OSTDEV || \
193 error "failed to enable MMP on $MMP_OSTDEV on $MMP_OSS"
196 # check whether the MMP feature is enabled or not
197 mmp_is_enabled $MMP_MDS $MMP_MDSDEV || \
198 error "MMP was not enabled on $MMP_MDSDEV on $MMP_MDS"
200 mmp_is_enabled $MMP_OSS $MMP_OSTDEV || \
201 error "MMP was not enabled on $MMP_OSTDEV on $MMP_OSS"
204 # Disable the MMP feature on the Lustre server targets
205 # which did not use failover.
208 local var=${MMP_MDS}failover_HOST
209 if [ -z "${!var}" ]; then
210 log "Failover is not used on MDS, disabling MMP manually..."
211 disable_mmp $MMP_MDS $MMP_MDSDEV || \
212 error "failed to disable MMP on $MMP_MDSDEV on $MMP_MDS"
213 mmp_is_enabled $MMP_MDS $MMP_MDSDEV && \
214 error "MMP was not disabled on $MMP_MDSDEV on $MMP_MDS"
217 var=${MMP_OSS}failover_HOST
218 if [ -z "${!var}" ]; then
219 log "Failover is not used on OSS, disabling MMP manually..."
220 disable_mmp $MMP_OSS $MMP_OSTDEV || \
221 error "failed to disable MMP on $MMP_OSTDEV on $MMP_OSS"
222 mmp_is_enabled $MMP_OSS $MMP_OSTDEV && \
223 error "MMP was not disabled on $MMP_OSTDEV on $MMP_OSS"
229 # Mount the shared target on the failover server after some interval it's
230 # mounted on the primary server.
231 mount_after_interval_sub() {
239 local failover_facet=$(get_failover_facet $facet)
242 local first_mount_rc=0
243 local second_mount_rc=0
245 log "Mounting $device on $facet..."
246 start $facet $device $opts &
249 if [ $interval -ne 0 ]; then
250 log "sleep $interval..."
254 log "Mounting $device on $failover_facet..."
255 start $failover_facet $device $opts
256 second_mount_rc=${PIPESTATUS[0]}
259 first_mount_rc=${PIPESTATUS[0]}
261 if [ $second_mount_rc -eq 0 -a $first_mount_rc -eq 0 ]; then
262 error_noexit "one mount delayed by mmp interval $interval should fail"
263 stop $facet || return ${PIPESTATUS[0]}
264 [ "$failover_facet" != "$facet" ] && stop $failover_facet || \
265 return ${PIPESTATUS[0]}
267 elif [ $second_mount_rc -ne 0 -a $first_mount_rc -ne 0 ]; then
268 error_noexit "failed to mount on the failover pair $facet,$failover_facet"
269 return $first_mount_rc
275 mount_after_interval() {
276 local mdt_interval=$1
277 local ost_interval=$2
280 mount_after_interval_sub $mdt_interval $MMP_MDSDEV $MMP_MDS \
281 $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
284 mount_after_interval_sub $ost_interval $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
286 if [ $rc -ne 0 ]; then
294 # Mount the shared target on the failover server
295 # during unmounting it on the primary server.
296 mount_during_unmount() {
302 local failover_facet=$(get_failover_facet $facet)
308 log "Mounting $device on $facet..."
309 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
311 log "Unmounting $device on $facet..."
315 log "Mounting $device on $failover_facet..."
316 start $failover_facet $device $mnt_opts
317 mount_rc=${PIPESTATUS[0]}
320 unmount_rc=${PIPESTATUS[0]}
322 if [ $mount_rc -eq 0 ]; then
323 error_noexit "mount during unmount of the first filesystem should fail"
324 stop $failover_facet || return ${PIPESTATUS[0]}
328 if [ $unmount_rc -ne 0 ]; then
329 error_noexit "unmount the $device on $facet should succeed"
336 # Mount the shared target on the failover server
337 # after clean unmounting it on the primary server.
338 mount_after_unmount() {
344 local failover_facet=$(get_failover_facet $facet)
346 log "Mounting $device on $facet..."
347 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
349 log "Unmounting $device on $facet..."
350 stop $facet || return ${PIPESTATUS[0]}
352 log "Mounting $device on $failover_facet..."
353 start $failover_facet $device $mnt_opts || return ${PIPESTATUS[0]}
358 # Mount the shared target on the failover server after rebooting
359 # the primary server.
360 mount_after_reboot() {
366 local failover_facet=$(get_failover_facet $facet)
369 log "Mounting $device on $facet..."
370 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
372 if [ "$FAILURE_MODE" = "HARD" ]; then
373 shutdown_facet $facet
375 wait_for_facet $facet
377 replay_barrier_nodf $facet
380 log "Mounting $device on $failover_facet..."
381 start $failover_facet $device $mnt_opts
383 if [ $rc -ne 0 ]; then
384 error_noexit "mount $device on $failover_facet should succeed"
385 stop $facet || return ${PIPESTATUS[0]}
392 # Run e2fsck on the Lustre server target.
400 log "Running e2fsck on the device $device on $facet..."
401 do_facet $facet "$E2FSCK $opts $device"
402 return ${PIPESTATUS[0]}
405 # Check whether there are failover pairs for MDS and OSS servers.
406 check_failover_pair() {
407 [ "$MMP_MDS" = "$MMP_MDS_FAILOVER" -o "$MMP_OSS" = "$MMP_OSS_FAILOVER" ] \
408 && { skip_env "failover pair is needed" && return 1; }
414 # Test 1 - two mounts at the same time.
416 check_failover_pair || return 0
418 mount_after_interval 0 0 || return ${PIPESTATUS[0]}
419 stop_services primary || return ${PIPESTATUS[0]}
421 run_test 1 "two mounts at the same time"
423 # Test 2 - one mount delayed by mmp update interval.
425 check_failover_pair || return 0
427 local mdt_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
428 local ost_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
430 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
431 stop_services primary || return ${PIPESTATUS[0]}
433 run_test 2 "one mount delayed by mmp update interval"
435 # Test 3 - one mount delayed by 2x mmp check interval.
437 check_failover_pair || return 0
439 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
440 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
442 mdt_interval=$((2 * $mdt_interval + 1))
443 ost_interval=$((2 * $ost_interval + 1))
445 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
446 stop_services primary || return ${PIPESTATUS[0]}
448 run_test 3 "one mount delayed by 2x mmp check interval"
450 # Test 4 - one mount delayed by > 2x mmp check interval.
452 check_failover_pair || return 0
454 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
455 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
457 mdt_interval=$((4 * $mdt_interval))
458 ost_interval=$((4 * $ost_interval))
460 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
461 stop_services primary || return ${PIPESTATUS[0]}
463 run_test 4 "one mount delayed by > 2x mmp check interval"
465 # Test 5 - mount during unmount of the first filesystem.
468 check_failover_pair || return 0
470 mount_during_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
471 return ${PIPESTATUS[0]}
474 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
475 mount_during_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
477 if [ $rc -ne 0 ]; then
478 stop $MMP_MDS || return ${PIPESTATUS[0]}
482 stop $MMP_MDS || return ${PIPESTATUS[0]}
484 run_test 5 "mount during unmount of the first filesystem"
486 # Test 6 - mount after clean unmount.
489 check_failover_pair || return 0
491 mount_after_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
492 return ${PIPESTATUS[0]}
495 mount_after_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
497 if [ $rc -ne 0 ]; then
498 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
502 stop_services failover || return ${PIPESTATUS[0]}
504 run_test 6 "mount after clean unmount"
506 # Test 7 - mount after reboot.
509 check_failover_pair || return 0
511 mount_after_reboot $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
512 return ${PIPESTATUS[0]}
515 mount_after_reboot $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
517 if [ $rc -ne 0 ]; then
518 stop $MMP_MDS || return ${PIPESTATUS[0]}
519 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
523 stop_services failover || return ${PIPESTATUS[0]}
524 stop_services primary || return ${PIPESTATUS[0]}
526 run_test 7 "mount after reboot"
528 # Test 8 - mount during e2fsck (should never succeed).
532 run_e2fsck $MMP_MDS $MMP_MDSDEV "-fy" &
536 log "Mounting $MMP_MDSDEV on $MMP_MDS_FAILOVER..."
537 if start $MMP_MDS_FAILOVER $MMP_MDSDEV $MDS_MOUNT_OPTS; then
538 error_noexit "mount $MMP_MDSDEV on $MMP_MDS_FAILOVER should fail"
539 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
546 run_e2fsck $MMP_OSS $MMP_OSTDEV "-fy" &
550 log "Mounting $MMP_OSTDEV on $MMP_OSS_FAILOVER..."
551 if start $MMP_OSS_FAILOVER $MMP_OSTDEV $OST_MOUNT_OPTS; then
552 error_noexit "mount $MMP_OSTDEV on $MMP_OSS_FAILOVER should fail"
553 stop $MMP_OSS_FAILOVER || return ${PIPESTATUS[0]}
560 run_test 8 "mount during e2fsck"
562 # Test 9 - mount after aborted e2fsck (should never succeed).
564 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
565 if ! start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
566 local rc=${PIPESTATUS[0]}
567 stop $MMP_MDS || return ${PIPESTATUS[0]}
570 stop_services primary || return ${PIPESTATUS[0]}
572 mark_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
574 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
575 if start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS; then
576 error_noexit "mount $MMP_MDSDEV on $MMP_MDS should fail"
577 stop $MMP_MDS || return ${PIPESTATUS[0]}
581 reset_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
583 mark_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
585 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
586 if start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
587 error_noexit "mount $MMP_OSTDEV on $MMP_OSS should fail"
588 stop $MMP_OSS || return ${PIPESTATUS[0]}
592 reset_mmp_block $MMP_OSS $MMP_OSTDEV || return ${PIPESTATUS[0]}
595 run_test 9 "mount after aborted e2fsck"
597 # Test 10 - e2fsck with mounted filesystem.
601 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
602 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
604 run_e2fsck $MMP_MDS_FAILOVER $MMP_MDSDEV "-fn"
607 # e2fsck is called with -n option (Open the filesystem read-only), so
608 # 0 (No errors) and 4 (File system errors left uncorrected) are the only
609 # acceptable exit codes in this case
610 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
611 error_noexit "e2fsck $MMP_MDSDEV on $MMP_MDS_FAILOVER returned $rc"
612 stop $MMP_MDS || return ${PIPESTATUS[0]}
616 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
617 start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS
619 if [ $rc -ne 0 ]; then
620 stop $MMP_MDS || return ${PIPESTATUS[0]}
624 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fn"
626 if [ $rc -ne 0 ] && [ $rc -ne 4 ]; then
627 error_noexit "e2fsck $MMP_OSTDEV on $MMP_OSS_FAILOVER returned $rc"
630 stop_services primary || return ${PIPESTATUS[0]}
633 run_test 10 "e2fsck with mounted filesystem"
636 FAIL_ON_ERROR=$SAVED_FAIL_ON_ERROR
638 complete $(basename $0) $SECONDS
639 $MMP_RESTORE_MOUNT && setupall