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 if [ -z "$mdsfailover_HOST" ]; then
163 log "Failover is not used on MDS, enabling MMP manually..."
164 enable_mmp $MMP_MDS $MMP_MDSDEV || \
165 error "failed to enable MMP on $MMP_MDSDEV on $MMP_MDS"
168 if [ -z "$ostfailover_HOST" ]; then
169 log "Failover is not used on OSS, enabling MMP manually..."
170 enable_mmp $MMP_OSS $MMP_OSTDEV || \
171 error "failed to enable MMP on $MMP_OSTDEV on $MMP_OSS"
174 # check whether the MMP feature is enabled or not
175 mmp_is_enabled $MMP_MDS $MMP_MDSDEV || \
176 error "MMP was not enabled on $MMP_MDSDEV on $MMP_MDS"
178 mmp_is_enabled $MMP_OSS $MMP_OSTDEV || \
179 error "MMP was not enabled on $MMP_OSTDEV on $MMP_OSS"
182 # Disable the MMP feature on the Lustre server targets
183 # which did not use failover.
186 if [ -z "$mdsfailover_HOST" ]; then
187 log "Failover is not used on MDS, disabling MMP manually..."
188 disable_mmp $MMP_MDS $MMP_MDSDEV || \
189 error "failed to disable MMP on $MMP_MDSDEV on $MMP_MDS"
190 mmp_is_enabled $MMP_MDS $MMP_MDSDEV && \
191 error "MMP was not disabled on $MMP_MDSDEV on $MMP_MDS"
194 if [ -z "$ostfailover_HOST" ]; then
195 log "Failover is not used on OSS, disabling MMP manually..."
196 disable_mmp $MMP_OSS $MMP_OSTDEV || \
197 error "failed to disable MMP on $MMP_OSTDEV on $MMP_OSS"
198 mmp_is_enabled $MMP_OSS $MMP_OSTDEV && \
199 error "MMP was not disabled on $MMP_OSTDEV on $MMP_OSS"
205 # Mount the shared target on the failover server after some interval it's
206 # mounted on the primary server.
207 mount_after_interval_sub() {
215 local failover_facet=$(get_failover_facet $facet)
218 local first_mount_rc=0
219 local second_mount_rc=0
221 log "Mounting $device on $facet..."
222 start $facet $device $opts &
225 if [ $interval -ne 0 ]; then
226 log "sleep $interval..."
230 log "Mounting $device on $failover_facet..."
231 start $failover_facet $device $opts
232 second_mount_rc=${PIPESTATUS[0]}
235 first_mount_rc=${PIPESTATUS[0]}
237 if [ $second_mount_rc -eq 0 -a $first_mount_rc -eq 0 ]; then
238 error_noexit "one mount delayed by mmp interval $interval should fail"
239 stop $facet || return ${PIPESTATUS[0]}
240 [ "$failover_facet" != "$facet" ] && stop $failover_facet || \
241 return ${PIPESTATUS[0]}
243 elif [ $second_mount_rc -ne 0 -a $first_mount_rc -ne 0 ]; then
244 error_noexit "failed to mount on the failover pair $facet,$failover_facet"
245 return $first_mount_rc
251 mount_after_interval() {
252 local mdt_interval=$1
253 local ost_interval=$2
256 mount_after_interval_sub $mdt_interval $MMP_MDSDEV $MMP_MDS \
257 $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
260 mount_after_interval_sub $ost_interval $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
262 if [ $rc -ne 0 ]; then
270 # Mount the shared target on the failover server
271 # during unmounting it on the primary server.
272 mount_during_unmount() {
278 local failover_facet=$(get_failover_facet $facet)
284 log "Mounting $device on $facet..."
285 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
287 log "Unmounting $device on $facet..."
291 log "Mounting $device on $failover_facet..."
292 start $failover_facet $device $mnt_opts
293 mount_rc=${PIPESTATUS[0]}
296 unmount_rc=${PIPESTATUS[0]}
298 if [ $mount_rc -eq 0 ]; then
299 error_noexit "mount during unmount of the first filesystem should fail"
300 stop $failover_facet || return ${PIPESTATUS[0]}
304 if [ $unmount_rc -ne 0 ]; then
305 error_noexit "unmount the $device on $facet should succeed"
312 # Mount the shared target on the failover server
313 # after clean unmounting it on the primary server.
314 mount_after_unmount() {
320 local failover_facet=$(get_failover_facet $facet)
322 log "Mounting $device on $facet..."
323 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
325 log "Unmounting $device on $facet..."
326 stop $facet || return ${PIPESTATUS[0]}
328 log "Mounting $device on $failover_facet..."
329 start $failover_facet $device $mnt_opts || return ${PIPESTATUS[0]}
334 # Mount the shared target on the failover server after rebooting
335 # the primary server.
336 mount_after_reboot() {
342 local failover_facet=$(get_failover_facet $facet)
345 log "Mounting $device on $facet..."
346 start $facet $device $mnt_opts || return ${PIPESTATUS[0]}
348 if [ "$FAILURE_MODE" = "HARD" ]; then
349 shutdown_facet $facet
353 replay_barrier_nodf $facet
356 log "Mounting $device on $failover_facet..."
357 start $failover_facet $device $mnt_opts
359 if [ $rc -ne 0 ]; then
360 error_noexit "mount $device on $failover_facet should succeed"
361 stop $facet || return ${PIPESTATUS[0]}
368 # Run e2fsck on the Lustre server target.
376 log "Running e2fsck on the device $device on $facet..."
377 do_facet $facet "$E2FSCK $opts $device"
378 return ${PIPESTATUS[0]}
381 # Check whether there are failover pairs for MDS and OSS servers.
382 check_failover_pair() {
383 [ "$MMP_MDS" = "$MMP_MDS_FAILOVER" -o "$MMP_OSS" = "$MMP_OSS_FAILOVER" ] \
384 && { skip_env "failover pair is needed" && return 1; }
390 # Test 1 - two mounts at the same time.
392 check_failover_pair || return 0
394 mount_after_interval 0 0 || return ${PIPESTATUS[0]}
395 stop_services primary || return ${PIPESTATUS[0]}
397 run_test 1 "two mounts at the same time"
399 # Test 2 - one mount delayed by mmp update interval.
401 check_failover_pair || return 0
403 local mdt_interval=$(get_mmp_update_interval $MMP_MDS $MMP_MDSDEV)
404 local ost_interval=$(get_mmp_update_interval $MMP_OSS $MMP_OSTDEV)
406 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
407 stop_services primary || return ${PIPESTATUS[0]}
409 run_test 2 "one mount delayed by mmp update interval"
411 # Test 3 - one mount delayed by 2x mmp check interval.
413 check_failover_pair || return 0
415 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
416 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
418 mdt_interval=$((2 * $mdt_interval + 1))
419 ost_interval=$((2 * $ost_interval + 1))
421 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
422 stop_services primary || return ${PIPESTATUS[0]}
424 run_test 3 "one mount delayed by 2x mmp check interval"
426 # Test 4 - one mount delayed by > 2x mmp check interval.
428 check_failover_pair || return 0
430 local mdt_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
431 local ost_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
433 mdt_interval=$((4 * $mdt_interval))
434 ost_interval=$((4 * $ost_interval))
436 mount_after_interval $mdt_interval $ost_interval || return ${PIPESTATUS[0]}
437 stop_services primary || return ${PIPESTATUS[0]}
439 run_test 4 "one mount delayed by > 2x mmp check interval"
441 # Test 5 - mount during unmount of the first filesystem.
444 check_failover_pair || return 0
446 mount_during_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
447 return ${PIPESTATUS[0]}
450 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
451 mount_during_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
453 if [ $rc -ne 0 ]; then
454 stop $MMP_MDS || return ${PIPESTATUS[0]}
458 stop $MMP_MDS || return ${PIPESTATUS[0]}
460 run_test 5 "mount during unmount of the first filesystem"
462 # Test 6 - mount after clean unmount.
465 check_failover_pair || return 0
467 mount_after_unmount $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
468 return ${PIPESTATUS[0]}
471 mount_after_unmount $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
473 if [ $rc -ne 0 ]; then
474 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
478 stop_services failover || return ${PIPESTATUS[0]}
480 run_test 6 "mount after clean unmount"
482 # Test 7 - mount after reboot.
485 check_failover_pair || return 0
487 mount_after_reboot $MMP_MDSDEV $MMP_MDS $MDS_MOUNT_OPTS || \
488 return ${PIPESTATUS[0]}
491 mount_after_reboot $MMP_OSTDEV $MMP_OSS $OST_MOUNT_OPTS
493 if [ $rc -ne 0 ]; then
494 stop $MMP_MDS || return ${PIPESTATUS[0]}
495 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
499 stop_services failover || return ${PIPESTATUS[0]}
500 stop_services primary || return ${PIPESTATUS[0]}
502 run_test 7 "mount after reboot"
504 # Test 8 - mount during e2fsck (should never succeed).
508 run_e2fsck $MMP_MDS $MMP_MDSDEV "-fy" &
512 log "Mounting $MMP_MDSDEV on $MMP_MDS_FAILOVER..."
513 if start $MMP_MDS_FAILOVER $MMP_MDSDEV $MDS_MOUNT_OPTS; then
514 error_noexit "mount $MMP_MDSDEV on $MMP_MDS_FAILOVER should fail"
515 stop $MMP_MDS_FAILOVER || return ${PIPESTATUS[0]}
522 run_e2fsck $MMP_OSS $MMP_OSTDEV "-fy" &
526 log "Mounting $MMP_OSTDEV on $MMP_OSS_FAILOVER..."
527 if start $MMP_OSS_FAILOVER $MMP_OSTDEV $OST_MOUNT_OPTS; then
528 error_noexit "mount $MMP_OSTDEV on $MMP_OSS_FAILOVER should fail"
529 stop $MMP_OSS_FAILOVER || return ${PIPESTATUS[0]}
536 run_test 8 "mount during e2fsck"
538 # Test 9 - mount after aborted e2fsck (should never succeed).
541 local mdt_mmp_check_interval
542 local ost_mmp_check_interval
544 start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS || return ${PIPESTATUS[0]}
545 if ! start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS; then
546 local rc=${PIPESTATUS[0]}
547 stop $MMP_MDS || return ${PIPESTATUS[0]}
550 stop_services primary || return ${PIPESTATUS[0]}
552 mdt_mmp_check_interval=$(get_mmp_check_interval $MMP_MDS $MMP_MDSDEV)
553 run_e2fsck $MMP_MDS_FAILOVER $MMP_MDSDEV "-fy" &
555 sleep $((2 * $mdt_mmp_check_interval + 1))
556 kill -s ABRT $e2fsck_pid
558 log "Mounting $MMP_MDSDEV on $MMP_MDS..."
559 if start $MMP_MDS $MMP_MDSDEV $MDS_MOUNT_OPTS; then
560 error_noexit "mount $MMP_MDSDEV on $MMP_MDS should fail"
561 stop $MMP_MDS || return ${PIPESTATUS[0]}
565 reset_mmp_block $MMP_MDS $MMP_MDSDEV || return ${PIPESTATUS[0]}
568 ost_mmp_check_interval=$(get_mmp_check_interval $MMP_OSS $MMP_OSTDEV)
569 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fy" &
571 sleep $((2 * $ost_mmp_check_interval + 1))
572 kill -s ABRT $e2fsck_pid
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"
595 if [ $rc -ne 8 ]; then
596 error_noexit "e2fsck $MMP_MDSDEV on $MMP_MDS_FAILOVER should return 8"
597 stop $MMP_MDS || return ${PIPESTATUS[0]}
598 [ $rc -ne 0 ] && return $rc || return 1
601 log "Mounting $MMP_OSTDEV on $MMP_OSS..."
602 start $MMP_OSS $MMP_OSTDEV $OST_MOUNT_OPTS
604 if [ $rc -ne 0 ]; then
605 stop $MMP_MDS || return ${PIPESTATUS[0]}
609 run_e2fsck $MMP_OSS_FAILOVER $MMP_OSTDEV "-fn"
611 if [ $rc -ne 8 ]; then
612 error_noexit "e2fsck $MMP_OSTDEV on $MMP_OSS_FAILOVER should return 8"
613 stop_services primary || return ${PIPESTATUS[0]}
614 [ $rc -ne 0 ] && return $rc || return 1
617 stop_services primary || return ${PIPESTATUS[0]}
620 run_test 10 "e2fsck with mounted filesystem"
623 FAIL_ON_ERROR=$SAVED_FAIL_ON_ERROR
625 equals_msg $(basename $0): test complete
626 $MMP_RESTORE_MOUNT && setupall
627 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && \
628 grep -q FAIL $TESTSUITELOG && exit 1 || true