3 # add uml1 uml2 uml3 in your /etc/hosts
5 # FIXME - there is no reason to use all of these different
6 # return codes, espcially when most of them are mapped to something
7 # else anyway. The combination of test number and return code
8 # figure out what failed.
14 # bug number for skipped test: 13739
15 HEAD_EXCEPT=" 32a 32b"
17 # bug number for skipped test:
18 ALWAYS_EXCEPT=" $CONF_SANITY_EXCEPT $HEAD_EXCEPT"
19 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
22 PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH
24 PTLDEBUG=${PTLDEBUG:--1}
26 LUSTRE=${LUSTRE:-`dirname $0`/..}
27 RLUSTRE=${RLUSTRE:-$LUSTRE}
29 . $LUSTRE/tests/test-framework.sh
31 # STORED_MDSSIZE is used in test_18
32 if [ -n "$MDSSIZE" ]; then
33 STORED_MDSSIZE=$MDSSIZE
35 # use small MDS + OST size to speed formatting time
38 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
40 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
41 remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
44 [ "$SLOW" = "no" ] && EXCEPT_SLOW="0 1 2 3 6 7 15 18 24b 25 30 31 32 33 34a 45"
53 local facet=$SINGLEMDS
54 local dev=${facet}_dev
58 # who knows if/where $TUNEFS is installed? Better reformat if it fails...
59 do_facet ${facet} "$TUNEFS --writeconf ${!dev}" || echo "tunefs failed, reformatting instead" && reformat
64 # The MGS must be started before the OSTs for a new fs, so start
65 # and stop to generate the startup logs.
74 local facet=$SINGLEMDS
75 # we can not use MDSDEV1 here because SINGLEMDS could be set not to mds1 only
76 local num=$(echo $facet | tr -d "mds")
77 local dev=$(mdsdevname $num)
78 echo "start mds service on `facet_active_host $facet`"
79 start $facet ${dev} $MDS_MOUNT_OPTS || return 94
83 echo "stop mds service on `facet_active_host $SINGLEMDS`"
84 # These tests all use non-failover stop
85 stop $SINGLEMDS -f || return 97
89 echo "start ost1 service on `facet_active_host ost1`"
90 start ost1 `ostdevname 1` $OST_MOUNT_OPTS || return 95
94 echo "stop ost1 service on `facet_active_host ost1`"
95 # These tests all use non-failover stop
96 stop ost1 -f || return 98
100 echo "start ost2 service on `facet_active_host ost2`"
101 start ost2 `ostdevname 2` $OST_MOUNT_OPTS || return 92
105 echo "stop ost2 service on `facet_active_host ost2`"
106 # These tests all use non-failover stop
107 stop ost2 -f || return 93
112 echo "mount $FSNAME on ${MOUNTPATH}....."
113 zconf_mount `hostname` $MOUNTPATH || return 96
117 local SAVEMOUNTOPT=$MOUNTOPT
118 MOUNTOPT="remount,$1"
120 echo "remount '$1' lustre on ${MOUNTPATH}....."
121 zconf_mount `hostname` $MOUNTPATH || return 96
122 MOUNTOPT=$SAVEMOUNTOPT
127 echo "umount lustre on ${MOUNTPATH}....."
128 zconf_umount `hostname` $MOUNTPATH || return 97
131 manual_umount_client(){
134 echo "manual umount lustre on ${MOUNT}...."
135 do_facet client "umount -d ${FORCE} $MOUNT"
147 stop_mds || return 201
148 stop_ost || return 202
149 unload_modules || return 203
153 umount_client $MOUNT || return 200
154 cleanup_nocli || return $?
158 do_facet client "cp /etc/passwd $DIR/a" || return 71
159 do_facet client "rm $DIR/a" || return 72
160 # make sure lustre is actually mounted (touch will block,
161 # but grep won't, so do it after)
162 do_facet client "grep $MOUNT' ' /proc/mounts > /dev/null" || return 73
163 echo "setup single mount lustre success"
167 do_facet client "touch $DIR/a" || return 71
168 do_facet client "rm $DIR/a" || return 72
169 do_facet client "touch $DIR2/a" || return 73
170 do_facet client "rm $DIR2/a" || return 74
171 echo "setup double mount lustre success"
176 if [ "$ONLY" == "setup" ]; then
181 if [ "$ONLY" == "cleanup" ]; then
186 #create single point mountpoint
194 check_mount || return 41
197 run_test 0 "single mount setup"
201 echo "start ost second time..."
203 check_mount || return 42
206 run_test 1 "start up ost twice (should return errors)"
211 echo "start mds second time.."
214 check_mount || return 43
217 run_test 2 "start up mds twice (should return err)"
221 #mount.lustre returns an error if already in mtab
222 mount_client $MOUNT && return $?
223 check_mount || return 44
226 run_test 3 "mount client twice (should return err)"
230 touch $DIR/$tfile || return 85
234 # ok for ost to fail shutdown
235 if [ 202 -ne $eno ]; then
240 run_test 4 "force cleanup ost, then cleanup"
242 test_5a() { # was test_5
244 touch $DIR/$tfile || return 1
245 fuser -m -v $MOUNT && echo "$MOUNT is in use by user space process."
247 stop_mds -f || return 2
249 # cleanup may return an error from the failed
250 # disconnects; for now I'll consider this successful
251 # if all the modules have unloaded.
255 echo "killing umount"
256 kill -TERM $UMOUNT_PID
257 echo "waiting for umount to finish"
259 if grep " $MOUNT " /proc/mounts; then
260 echo "test 5: /proc/mounts after failed umount"
264 echo "killing umount"
265 kill -TERM $UMOUNT_PID
266 echo "waiting for umount to finish"
268 grep " $MOUNT " /proc/mounts && echo "test 5: /proc/mounts after second umount" && return 11
272 # stop_mds is a no-op here, and should not fail
273 cleanup_nocli || return $?
274 # df may have lingering entry
276 # mtab may have lingering entry
280 while [ "$WAIT" -ne "$MAX_WAIT" ]; do
282 grep -q $MOUNT" " /etc/mtab || break
283 echo "Waiting /etc/mtab updated ... "
284 WAIT=$(( WAIT + sleep))
286 [ "$WAIT" -eq "$MAX_WAIT" ] && error "/etc/mtab is not updated in $WAIT secs"
287 echo "/etc/mtab updated in $WAIT secs"
289 run_test 5a "force cleanup mds, then cleanup"
293 [ -d $MOUNT ] || mkdir -p $MOUNT
294 grep " $MOUNT " /etc/mtab && echo "test 5b: mtab before mount" && return 10
295 mount_client $MOUNT && return 1
296 grep " $MOUNT " /etc/mtab && echo "test 5b: mtab after failed mount" && return 11
298 # stop_mds is a no-op here, and should not fail
299 cleanup_nocli || return $?
302 run_test 5b "mds down, cleanup after failed mount (bug 2712) (should return errs)"
307 [ -d $MOUNT ] || mkdir -p $MOUNT
308 grep " $MOUNT " /etc/mtab && echo "test 5c: mtab before mount" && return 10
309 local oldfs="${FSNAME}"
310 FSNAME="wrong.${FSNAME}"
311 mount_client $MOUNT || :
313 grep " $MOUNT " /etc/mtab && echo "test 5c: mtab after failed mount" && return 11
315 cleanup_nocli || return $?
317 run_test 5c "cleanup after failed mount (bug 2712) (should return errs)"
323 grep " $MOUNT " /etc/mtab && echo "test 5d: mtab before mount" && return 10
324 mount_client $MOUNT || return 1
326 grep " $MOUNT " /etc/mtab && echo "test 5d: mtab after unmount" && return 11
329 run_test 5d "mount with ost down"
335 #define OBD_FAIL_PTLRPC_DELAY_SEND 0x506
336 do_facet client "lctl set_param fail_loc=0x80000506"
337 grep " $MOUNT " /etc/mtab && echo "test 5e: mtab before mount" && return 10
338 mount_client $MOUNT || echo "mount failed (not fatal)"
340 grep " $MOUNT " /etc/mtab && echo "test 5e: mtab after unmount" && return 11
343 run_test 5e "delayed connect, don't crash (bug 10268)"
348 mount_client ${MOUNT} || return 87
349 touch $DIR/a || return 86
352 run_test 6 "manual umount, then mount again"
357 cleanup_nocli || return $?
359 run_test 7 "manual umount, then cleanup"
364 check_mount2 || return 45
365 umount_client $MOUNT2
368 run_test 8 "double mount setup"
373 do_facet ost1 lctl set_param debug=\'inode trace\' || return 1
374 do_facet ost1 lctl set_param subsystem_debug=\'mds ost\' || return 1
376 CHECK_PTLDEBUG="`do_facet ost1 lctl get_param -n debug`"
377 if [ "$CHECK_PTLDEBUG" ] && [ "$CHECK_PTLDEBUG" = "trace inode" ];then
378 echo "lnet.debug success"
380 echo "lnet.debug: want 'trace inode', have '$CHECK_PTLDEBUG'"
383 CHECK_SUBSYS="`do_facet ost1 lctl get_param -n subsystem_debug`"
384 if [ "$CHECK_SUBSYS" ] && [ "$CHECK_SUBSYS" = "mds ost" ]; then
385 echo "lnet.subsystem_debug success"
387 echo "lnet.subsystem_debug: want 'mds ost', have '$CHECK_SUBSYS'"
390 stop_ost || return $?
393 run_test 9 "test ptldebug and subsystem for mkfs"
395 # LOGS/PENDING do not exist anymore since CMD3
397 local TMPMTPT="${TMP}/conf16"
398 local dev=${SINGLEMDS}_dev
400 if [ ! -e "$MDSDEV" ]; then
401 log "no $MDSDEV existing, so mount Lustre to create one"
403 check_mount || return 41
407 [ -f "$MDSDEV" ] && LOOPOPT="-o loop"
409 log "change the mode of $MDSDEV/OBJECTS to 555"
410 do_facet $SINGLEMDS "mkdir -p $TMPMTPT &&
411 mount $LOOPOPT -t $FSTYPE $MDSDEV $TMPMTPT &&
412 chmod 555 $TMPMTPT/OBJECTS &&
413 umount $TMPMTPT" || return $?
415 log "mount Lustre to change the mode of OBJECTS, then umount Lustre"
417 check_mount || return 41
420 log "read the mode of OBJECTS and check if they has been changed properly"
421 EXPECTEDOBJECTSMODE=`do_facet $SINGLEMDS "debugfs -R 'stat OBJECTS' $MDSDEV 2> /dev/null" | grep 'Mode: ' | sed -e "s/.*Mode: *//" -e "s/ *Flags:.*//"`
423 if [ "$EXPECTEDOBJECTSMODE" = "0777" ]; then
424 log "Success:Lustre change the mode of OBJECTS correctly"
426 error "Lustre does not change mode of OBJECTS properly"
429 run_test 16 "verify that lustre will correct the mode of OBJECTS"
432 local dev=${SINGLEMDS}_dev
435 if [ ! -e "$MDSDEV" ]; then
436 echo "no $MDSDEV existing, so mount Lustre to create one"
438 check_mount || return 41
442 echo "Remove mds config log"
443 do_facet $SINGLEMDS "debugfs -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MDSDEV || return \$?" || return $?
446 start_mds && return 42
449 run_test 17 "Verify failed mds_postsetup won't fail assertion (2936) (should return errs)"
452 [ "$FSTYPE" != "ldiskfs" ] && skip "not needed for FSTYPE=$FSTYPE" && return
454 local dev=${SINGLEMDS}_dev
460 # check if current MDSSIZE is large enough
461 [ $MDSSIZE -ge $MIN ] && OK=1 && myMDSSIZE=$MDSSIZE && \
462 log "use MDSSIZE=$MDSSIZE"
464 # check if the global config has a large enough MDSSIZE
465 [ -z "$OK" -a ! -z "$STORED_MDSSIZE" ] && [ $STORED_MDSSIZE -ge $MIN ] && \
466 OK=1 && myMDSSIZE=$STORED_MDSSIZE && \
467 log "use STORED_MDSSIZE=$STORED_MDSSIZE"
469 # check if the block device is large enough
470 [ -z "$OK" -a -b $MDSDEV ] && \
471 [ "$(dd if=$MDSDEV of=/dev/null bs=1k count=1 skip=$MIN 2>&1 |
472 awk '($3 == "in") { print $1 }')" = "1+0" ] && OK=1 && \
473 myMDSSIZE=$MIN && log "use device $MDSDEV with MIN=$MIN"
475 # check if a loopback device has enough space for fs metadata (5%)
476 [ -z "$OK" ] && [ -f $MDSDEV -o ! -e $MDSDEV ] &&
477 SPACE=$(df -P $(dirname $MDSDEV) |
478 awk '($1 != "Filesystem") {print $4}') &&
479 [ $SPACE -gt $((MIN / 20)) ] && OK=1 && myMDSSIZE=$MIN && \
480 log "use file $MDSDEV with MIN=$MIN"
482 [ -z "$OK" ] && skip "$MDSDEV too small for ${MIN}kB MDS" && return
485 echo "mount mds with large journal..."
486 local OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS
488 MDS_MKFS_OPTS="--mgs --mdt --fsname=$FSNAME --device-size=$myMDSSIZE --param sys.timeout=$TIMEOUT $MDSOPT"
491 echo "mount lustre system..."
493 check_mount || return 41
495 echo "check journal size..."
496 local FOUNDSIZE=`do_facet mds "debugfs -c -R 'stat <8>' $MDSDEV" | awk '/Size: / { print $NF; exit;}'`
497 if [ $FOUNDSIZE -gt $((32 * 1024 * 1024)) ]; then
498 log "Success: mkfs creates large journals. Size: $((FOUNDSIZE >> 20))M"
500 error "expected journal size > 32M, found $((FOUNDSIZE >> 20))M"
505 MDS_MKFS_OPTS=$OLD_MDS_MKFS_OPTS
508 run_test 18 "check mkfs creates large journals"
511 start_mds || return 1
512 stop_mds -f || return 2
514 run_test 19a "start/stop MDS without OSTs"
517 start_ost || return 1
518 stop_ost -f || return 2
520 run_test 19b "start/stop OSTs without MDS"
523 # first format the ost/mdt
527 check_mount || return 43
529 remount_client ro $MOUNT || return 44
530 touch $DIR/$tfile && echo "$DIR/$tfile created incorrectly" && return 45
531 [ -e $DIR/$tfile ] && echo "$DIR/$tfile exists incorrectly" && return 46
532 remount_client rw $MOUNT || return 47
534 [ ! -f $DIR/$tfile ] && echo "$DIR/$tfile missing" && return 48
535 MCNT=`grep -c $MOUNT /etc/mtab`
536 [ "$MCNT" -ne 1 ] && echo "$MOUNT in /etc/mtab $MCNT times" && return 49
541 run_test 20 "remount ro,rw mounts work and doesn't break /etc/mtab"
549 run_test 21a "start mds before ost, stop ost first"
557 run_test 21b "start ost before mds, stop mds first"
567 run_test 21c "start mds between two osts, stop mds last"
570 #reformat to remove all logs
573 echo Client mount before any osts are in the logs
575 check_mount && return 41
578 echo Client mount with ost in logs, but none running
582 # check_mount will block trying to contact ost
586 echo Client mount with a running ost
589 # if gss enabled, wait full time to let connection from
590 # mds to ost be established, due to the mismatch between
591 # initial connect timeout and gss context negotiation timeout.
592 # This perhaps could be remove after AT landed.
593 echo "sleep $((TIMEOUT + TIMEOUT + TIMEOUT))s"
594 sleep $((TIMEOUT + TIMEOUT + TIMEOUT))
597 check_mount || return 41
602 run_test 22 "start a client before osts (should return errs)"
604 test_23a() { # was test_23
608 # force down client so that recovering mds waits for reconnect
609 local running=$(grep -c $MOUNT /proc/mounts) || true
610 if [ $running -ne 0 ]; then
611 echo "Stopping client $MOUNT (opts: -f)"
615 # enter recovery on mds
617 # try to start a new client
618 mount_client $MOUNT &
620 MOUNT_PID=$(ps -ef | grep "t lustre" | grep -v grep | awk '{print $2}')
621 MOUNT_LUSTRE_PID=`ps -ef | grep mount.lustre | grep -v grep | awk '{print $2}'`
622 echo mount pid is ${MOUNT_PID}, mount.lustre pid is ${MOUNT_LUSTRE_PID}
624 ps --ppid $MOUNT_LUSTRE_PID
625 # FIXME why o why can't I kill these? Manual "ctrl-c" works...
626 kill -TERM $MOUNT_LUSTRE_PID
627 echo "waiting for mount to finish"
629 # we can not wait $MOUNT_PID because it is not a child of this shell
635 while [ "$WAIT" -lt "$MAX_WAIT" ]; do
637 PID1=$(ps -ef | awk '{print $2}' | grep -w $MOUNT_PID)
638 PID2=$(ps -ef | awk '{print $2}' | grep -w $MOUNT_LUSTRE_PID)
641 [ -z "$PID1" -a -z "$PID2" ] && break
642 echo "waiting for mount to finish ... "
643 WAIT=$(( WAIT + sleep))
645 [ "$WAIT" -eq "$MAX_WAIT" ] && error "MOUNT_PID $MOUNT_PID and \
646 MOUNT__LUSTRE_PID $MOUNT__LUSTRE_PID still not killed in $WAIT secs"
651 run_test 23a "interrupt client during recovery mount delay"
656 test_23b() { # was test_23
659 # Simulate -EINTR during mount OBD_FAIL_LDLM_CLOSE_THREAD
660 lctl set_param fail_loc=0x80000313
664 run_test 23b "Simulate -EINTR during mount"
666 fs2mds_HOST=$mds_HOST
667 fs2ost_HOST=$ost_HOST
671 echo "umount $MOUNT2 ..."
672 umount $MOUNT2 || true
673 echo "stopping fs2mds ..."
674 stop fs2mds -f || true
675 echo "stopping fs2ost ..."
676 stop fs2ost -f || true
683 [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST
684 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then
685 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
686 skip "mixed loopback and real device not working" && return
689 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
690 local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
692 # test 8-char fsname as well
693 local FSNAME2=test1234
694 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --nomgs --mgsnode=$MGSNID --reformat $fs2mdsdev || exit 10
696 add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2ostdev || exit 10
699 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT
700 start fs2ost $fs2ostdev $OST_MOUNT_OPTS
702 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1
704 check_mount || return 2
705 # files written on 1 should not show up on 2
706 cp /etc/passwd $DIR/$tfile
708 [ -e $MOUNT2/$tfile ] && error "File bleed" && return 7
711 cp /etc/passwd $MOUNT2/b || return 3
712 rm $MOUNT2/b || return 4
713 # 2 is actually mounted
714 grep $MOUNT2' ' /proc/mounts > /dev/null || return 5
716 facet_failover fs2mds
717 facet_failover fs2ost
720 # the MDS must remain up until last MDT
722 MDS=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1)
723 [ -z "$MDS" ] && error "No MDT" && return 8
725 cleanup_nocli || return 6
727 run_test 24a "Multiple MDTs on a single node"
730 if [ -z "$fs2mds_DEV" ]; then
731 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
732 skip "mixed loopback and real device not working" && return
735 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
737 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME}2 --mgs --reformat $fs2mdsdev || exit 10
739 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && return 2
742 run_test 24b "Multiple MGSs on a single node (should return err)"
746 check_mount || return 2
747 local MODULES=$($LCTL modules | awk '{ print $2 }')
748 rmmod $MODULES 2>/dev/null || true
751 run_test 25 "Verify modules are referenced"
755 # we need modules before mount for sysctl, so make sure...
756 do_facet $SINGLEMDS "lsmod | grep -q lustre || modprobe lustre"
757 #define OBD_FAIL_MDS_FS_SETUP 0x135
758 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000135"
759 start_mds && echo MDS started && return 1
760 lctl get_param -n devices
761 DEVS=$(lctl get_param -n devices | wc -l)
762 [ $DEVS -gt 0 ] && return 2
763 unload_modules || return 203
765 run_test 26 "MDT startup failure cleans LOV (should return errs)"
771 local ORIG=$(do_facet $myfacet "$TEST")
772 if [ $# -gt 3 ]; then
778 echo "Setting $PARAM from $ORIG to $FINAL"
779 do_facet $SINGLEMDS "$LCTL conf_param $PARAM=$FINAL" || error conf_param failed
785 RESULT=$(do_facet $myfacet "$TEST")
786 if [ $RESULT -eq $FINAL ]; then
787 echo "Updated config after $WAIT sec (got $RESULT)"
791 if [ $WAIT -eq $MAX ]; then
792 echo "Config update not seen: wanted $FINAL got $RESULT"
795 echo "Waiting $(($MAX - $WAIT)) secs for config update"
800 start_ost || return 1
801 start_mds || return 2
802 echo "Requeue thread should have started: "
803 ps -e | grep ll_cfg_requeue
804 set_and_check ost1 "lctl get_param -n obdfilter.$FSNAME-OST0000.client_cache_seconds" "$FSNAME-OST0000.ost.client_cache_seconds" || return 3
807 run_test 27a "Reacquire MGS lock if OST started first"
812 local device=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }')
814 facet_failover $SINGLEMDS
815 set_and_check $SINGLEMDS "lctl get_param -n mdt.$device.identity_acquire_expire" "$device.mdt.identity_acquire_expire" || return 3
816 set_and_check client "lctl get_param -n mdc.$device-mdc-*.max_rpcs_in_flight" "$device.mdc.max_rpcs_in_flight" || return 4
820 run_test 27b "Reacquire MGS lock after failover"
824 TEST="lctl get_param -n llite.$FSNAME-*.max_read_ahead_whole_mb"
825 PARAM="$FSNAME.llite.max_read_ahead_whole_mb"
828 set_and_check client "$TEST" "$PARAM" $FINAL || return 3
829 FINAL=$(($FINAL + 1))
830 set_and_check client "$TEST" "$PARAM" $FINAL || return 4
831 umount_client $MOUNT || return 200
834 if [ $RESULT -ne $FINAL ]; then
835 echo "New config not seen: wanted $FINAL got $RESULT"
838 echo "New config success: got $RESULT"
840 set_and_check client "$TEST" "$PARAM" $ORIG || return 5
843 run_test 28 "permanent parameter setting"
846 [ "$OSTCOUNT" -lt "2" ] && skip "$OSTCOUNT < 2, skipping" && return
847 setup > /dev/null 2>&1
851 local PARAM="$FSNAME-OST0001.osc.active"
852 local PROC_ACT="osc.$FSNAME-OST0001-osc-[^M]*.active"
853 local PROC_UUID="osc.$FSNAME-OST0001-osc-[^M]*.ost_server_uuid"
855 ACTV=$(lctl get_param -n $PROC_ACT)
857 set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $DEAC || return 2
858 # also check ost_server_uuid status
859 RESULT=$(lctl get_param -n $PROC_UUID | grep DEACTIV)
860 if [ -z "$RESULT" ]; then
861 echo "Live client not deactivated: $(lctl get_param -n $PROC_UUID)"
864 echo "Live client success: got $RESULT"
868 local MPROC="osc.$FSNAME-OST0001-osc-[M]*.active"
873 RESULT=`do_facet $SINGLEMDS " lctl get_param -n $MPROC"`
874 [ ${PIPESTATUS[0]} = 0 ] || error "Can't read $MPROC"
875 if [ $RESULT -eq $DEAC ]; then
876 echo "MDT deactivated also after $WAIT sec (got $RESULT)"
880 if [ $WAIT -eq $MAX ]; then
881 echo "MDT not deactivated: wanted $DEAC got $RESULT"
884 echo "Waiting $(($MAX - $WAIT)) secs for MDT deactivated"
887 # test new client starts deactivated
888 umount_client $MOUNT || return 200
890 RESULT=$(lctl get_param -n $PROC_UUID | grep DEACTIV | grep NEW)
891 if [ -z "$RESULT" ]; then
892 echo "New client not deactivated from start: $(lctl get_param -n $PROC_UUID)"
895 echo "New client success: got $RESULT"
898 # make sure it reactivates
899 set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $ACTV || return 6
904 #writeconf to remove all ost2 traces for subsequent tests
910 run_test 29 "permanently remove an OST"
915 TEST="lctl get_param -n llite.$FSNAME-*.max_read_ahead_whole_mb"
917 LIST=(1 2 3 4 5 4 3 2 1 2 3 4 5 4 3 2 1 2 3 4 5)
918 for i in ${LIST[@]}; do
919 set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $i || return 3
921 # make sure client restart still works
923 mount_client $MOUNT || return 4
924 [ "$($TEST)" -ne "$i" ] && return 5
925 set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $ORIG || return 6
928 run_test 30 "Big config llog"
930 test_31() { # bug 10734
931 # ipaddr must not exist
932 mount -t lustre 4.3.2.1@tcp:/lustre $MOUNT || true
935 run_test 31 "Connect to non-existent node (shouldn't crash)"
938 # XXX - make this test verify 1.8 -> 2.0 upgrade is working
939 # XXX - make this run on client-only systems with real hardware on
941 # there appears to be a lot of assumption here about loopback
943 # or maybe this test is just totally useless on a client-only system
944 [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
945 [ "$mds_HOST" = "`hostname`" ] || { skip "remote MDS" && return 0; }
946 [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
947 { skip "remote OST" && return 0; }
949 [ -z "$TUNEFS" ] && skip "No tunefs" && return
950 local DISK1_8=$LUSTRE/tests/disk1_8.tgz
951 [ ! -r $DISK1_8 ] && skip "Cannot find $DISK1_8" && return 0
954 tar xjvf $DISK1_8 -C $TMP/$tdir || \
955 { skip "Cannot untar $DISK1_8" && return 0; }
958 lctl set_param debug=$PTLDEBUG
960 $TUNEFS $tmpdir/mds || error "tunefs failed"
961 # nids are wrong, so client wont work, but server should start
962 start mds $tmpdir/mds "-o loop,exclude=lustre-OST0000" || return 3
963 local UUID=$(lctl get_param -n mdt.lustre-MDT0000.uuid)
965 [ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID"
967 $TUNEFS --mgsnode=`hostname` $tmpdir/ost1 || error "tunefs failed"
968 start ost1 $tmpdir/ost1 "-o loop" || return 5
969 UUID=$(lctl get_param -n obdfilter.lustre-OST0000.uuid)
971 [ "$UUID" == "ost1_UUID" ] || error "UUID is wrong: $UUID"
973 local NID=$($LCTL list_nids | head -1)
975 echo "OSC changes should return err:"
976 $LCTL conf_param lustre-OST0000.osc.max_dirty_mb=15 && return 7
977 $LCTL conf_param lustre-OST0000.failover.node=$NID && return 8
979 echo "MDC changes should succeed:"
980 $LCTL conf_param lustre-MDT0000.mdc.max_rpcs_in_flight=9 || return 9
981 $LCTL conf_param lustre-MDT0000.failover.node=$NID || return 10
984 # With a new good MDT failover nid, we should be able to mount a client
985 # (but it cant talk to OST)
986 local OLDMOUNTOPT=$MOUNTOPT
987 MOUNTOPT="exclude=lustre-OST0000"
989 MOUNTOPT=$OLDMOUNTOPT
990 set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "lustre-MDT0000.mdc.max_rpcs_in_flight" ||
993 zconf_umount `hostname` $MOUNT -f
997 # mount a second time to make sure we didnt leave upgrade flag on
999 $TUNEFS --dryrun $tmpdir/mds || error "tunefs failed"
1001 start mds $tmpdir/mds "-o loop,exclude=lustre-OST0000" || return 12
1004 rm -rf $tmpdir || true # true is only for TMP on NFS
1006 run_test 32a "Upgrade from 1.8 (not live)"
1009 # XXX - make this test verify 1.8 -> 2.0 upgrade is working
1010 # XXX - make this run on client-only systems with real hardware on
1012 # there appears to be a lot of assumption here about loopback
1014 # or maybe this test is just totally useless on a client-only system
1015 [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
1016 [ "$mds_HOST" = "`hostname`" ] || { skip "remote MDS" && return 0; }
1017 [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
1018 { skip "remote OST" && return 0; }
1020 [ -z "$TUNEFS" ] && skip "No tunefs" && return
1021 local DISK1_8=$LUSTRE/tests/disk1_8.tgz
1022 [ ! -r $DISK1_8 ] && skip "Cannot find $DISK1_8" && return 0
1024 tar xjvf $DISK1_8 -C $TMP/$tdir || \
1025 { skip "Cannot untar $DISK1_8" && return ; }
1028 lctl set_param debug=$PTLDEBUG
1031 # writeconf will cause servers to register with their current nids
1032 $TUNEFS --writeconf --fsname=$NEWNAME $tmpdir/mds || error "tunefs failed"
1033 start mds $tmpdir/mds "-o loop" || return 3
1034 local UUID=$(lctl get_param -n mdt.${NEWNAME}-MDT0000.uuid)
1036 [ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID"
1038 $TUNEFS --mgsnode=`hostname` --fsname=$NEWNAME --writeconf $tmpdir/ost1 || error "tunefs failed"
1039 start ost1 $tmpdir/ost1 "-o loop" || return 5
1040 UUID=$(lctl get_param -n obdfilter.${NEWNAME}-OST0000.uuid)
1042 [ "$UUID" == "ost1_UUID" ] || error "UUID is wrong: $UUID"
1044 echo "OSC changes should succeed:"
1045 $LCTL conf_param ${NEWNAME}-OST0000.osc.max_dirty_mb=15 || return 7
1046 $LCTL conf_param ${NEWNAME}-OST0000.failover.node=$NID || return 8
1048 echo "MDC changes should succeed:"
1049 $LCTL conf_param ${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight=9 || return 9
1052 # MDT and OST should have registered with new nids, so we should have
1053 # a fully-functioning client
1054 echo "Check client and old fs contents"
1059 set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight" || return 11
1060 [ "$(cksum $MOUNT/passwd | cut -d' ' -f 1,2)" == "2479747619 779" ] || return 12
1064 rm -rf $tmpdir || true # true is only for TMP on NFS
1066 run_test 32b "Upgrade from 1.8 with writeconf"
1068 test_33a() { # bug 12333, was test_33
1070 local FSNAME2=test-123
1071 [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST
1073 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then
1074 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
1075 skip "mixed loopback and real device not working" && return
1078 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
1079 local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
1080 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10
1081 add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --index=8191 --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10
1083 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT
1084 start fs2ost $fs2ostdev $OST_MOUNT_OPTS
1085 do_facet $SINGLEMDS "$LCTL conf_param $FSNAME2.sys.timeout=200" || rc=1
1087 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || rc=2
1090 cp /etc/hosts $MOUNT2/ || rc=3
1091 $LFS getstripe $MOUNT2/hosts
1096 rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev
1097 cleanup_nocli || rc=6
1100 run_test 33a "Mount ost with a large index number"
1102 test_33b() { # was test_34
1105 do_facet client dd if=/dev/zero of=$MOUNT/24 bs=1024k count=1
1106 # Drop lock cancelation reply during umount
1107 #define OBD_FAIL_LDLM_CANCEL 0x304
1108 do_facet client lctl set_param fail_loc=0x80000304
1109 #lctl set_param debug=-1
1110 umount_client $MOUNT
1113 run_test 33b "Drop cancel during umount"
1117 do_facet client "sh runmultiop_bg_pause $DIR/file O_c"
1118 manual_umount_client
1120 do_facet client killall -USR1 multiop
1121 if [ $rc -eq 0 ]; then
1122 error "umount not fail!"
1127 run_test 34a "umount with opened file should be fail"
1132 touch $DIR/$tfile || return 1
1133 stop_mds --force || return 2
1135 manual_umount_client --force
1137 if [ $rc -ne 0 ]; then
1138 error "mtab after failed umount - rc $rc"
1144 run_test 34b "force umount with failed mds should be normal"
1148 touch $DIR/$tfile || return 1
1149 stop_ost --force || return 2
1151 manual_umount_client --force
1153 if [ $rc -ne 0 ]; then
1154 error "mtab after failed umount - rc $rc"
1160 run_test 34c "force umount with failed ost should be normal"
1162 test_35() { # bug 12459
1165 DBG_SAVE="`lctl get_param -n debug`"
1166 lctl set_param debug="ha"
1168 log "Set up a fake failnode for the MDS"
1170 local device=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1)
1171 do_facet $SINGLEMDS $LCTL conf_param ${device}.failover.node=$FAKENID || return 4
1173 log "Wait for RECONNECT_INTERVAL seconds (10s)"
1176 MSG="conf-sanity.sh test_35 `date +%F%kh%Mm%Ss`"
1179 log "Stopping the MDT:"
1180 stop_mds || return 5
1182 df $MOUNT > /dev/null 2>&1 &
1184 log "Restarting the MDT:"
1185 start_mds || return 6
1186 log "Wait for df ($DFPID) ... "
1189 lctl set_param debug="$DBG_SAVE"
1191 # retrieve from the log the first server that the client tried to
1192 # contact after the connection loss
1193 $LCTL dk $TMP/lustre-log-$TESTNAME.log
1194 NEXTCONN=`awk "/${MSG}/ {start = 1;}
1195 /import_select_connection.$device-mdc.* using connection/ {
1197 if (\\\$NF ~ /$FAKENID/)
1203 }" $TMP/lustre-log-$TESTNAME.log`
1204 [ "$NEXTCONN" != "0" ] && log "The client didn't try to reconnect to the last active server (tried ${NEXTCONN} instead)" && return 7
1207 run_test 35 "Reconnect to the last active server first"
1211 local FSNAME2=test1234
1212 local fs3ost_HOST=$ost_HOST
1214 [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST && fs3ost_HOST=$ost1_HOST
1217 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" -o -z "$fs3ost_DEV" ]; then
1218 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
1219 skip "mixed loopback and real device not working" && return
1221 [ $OSTCOUNT -lt 2 ] && skip "skipping test for single OST" && return
1223 [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
1224 { skip "remote OST" && return 0; }
1226 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
1227 local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
1228 local fs3ostdev=${fs3ost_DEV:-$(ostdevname 2)_2}
1229 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10
1230 # XXX after we support non 4K disk blocksize, change following --mkfsoptions with
1232 add fs2ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10
1233 add fs3ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=$MGSNID --reformat $fs3ostdev || exit 10
1235 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS
1236 start fs2ost $fs2ostdev $OST_MOUNT_OPTS
1237 start fs3ost $fs3ostdev $OST_MOUNT_OPTS
1239 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1
1241 sleep 5 # until 11778 fixed
1243 dd if=/dev/zero of=$MOUNT2/$tfile bs=1M count=7 || return 2
1245 BKTOTAL=`lctl get_param -n obdfilter.*.kbytestotal | awk 'BEGIN{total=0}; {total+=$1}; END{print total}'`
1246 BKFREE=`lctl get_param -n obdfilter.*.kbytesfree | awk 'BEGIN{free=0}; {free+=$1}; END{print free}'`
1247 BKAVAIL=`lctl get_param -n obdfilter.*.kbytesavail | awk 'BEGIN{avail=0}; {avail+=$1}; END{print avail}'`
1248 STRING=`df -P $MOUNT2 | tail -n 1 | awk '{print $2","$3","$4}'`
1249 DFTOTAL=`echo $STRING | cut -d, -f1`
1250 DFUSED=`echo $STRING | cut -d, -f2`
1251 DFAVAIL=`echo $STRING | cut -d, -f3`
1252 DFFREE=$(($DFTOTAL - $DFUSED))
1254 ALLOWANCE=$((64 * $OSTCOUNT))
1256 if [ $DFTOTAL -lt $(($BKTOTAL - $ALLOWANCE)) ] ||
1257 [ $DFTOTAL -gt $(($BKTOTAL + $ALLOWANCE)) ] ; then
1258 echo "**** FAIL: df total($DFTOTAL) mismatch OST total($BKTOTAL)"
1261 if [ $DFFREE -lt $(($BKFREE - $ALLOWANCE)) ] ||
1262 [ $DFFREE -gt $(($BKFREE + $ALLOWANCE)) ] ; then
1263 echo "**** FAIL: df free($DFFREE) mismatch OST free($BKFREE)"
1266 if [ $DFAVAIL -lt $(($BKAVAIL - $ALLOWANCE)) ] ||
1267 [ $DFAVAIL -gt $(($BKAVAIL + $ALLOWANCE)) ] ; then
1268 echo "**** FAIL: df avail($DFAVAIL) mismatch OST avail($BKAVAIL)"
1273 stop fs3ost -f || return 200
1274 stop fs2ost -f || return 201
1275 stop fs2mds -f || return 202
1276 rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev $fs3ostdev
1277 unload_modules || return 203
1280 run_test 36 "df report consistency on OSTs with different block size"
1283 [ -n "$CLIENTONLY" -o -n "$CLIENTMODSONLY" ] && skip "client only testing" && return 0
1284 LOCAL_MDSDEV="$TMP/mdt.img"
1285 SYM_MDSDEV="$TMP/sym_mdt.img"
1287 echo "MDS : $LOCAL_MDSDEV"
1288 echo "SYMLINK : $SYM_MDSDEV"
1292 mkfs.lustre --reformat --fsname=lustre --mdt --mgs --device-size=9000 $LOCAL_MDSDEV ||
1293 error "mkfs.lustre $LOCAL_MDSDEV failed"
1294 ln -s $LOCAL_MDSDEV $SYM_MDSDEV
1296 echo "mount symlink device - $SYM_MDSDEV"
1298 mount_op=`mount -v -t lustre -o loop $SYM_MDSDEV ${MOUNT%/*}/mds 2>&1 | grep "unable to set tunable"`
1299 umount -d ${MOUNT%/*}/mds
1300 rm -f $LOCAL_MDSDEV $SYM_MDSDEV
1302 if [ -n "$mount_op" ]; then
1303 error "**** FAIL: set tunables failed for symlink device"
1307 run_test 37 "verify set tunables works for symlink device"
1309 test_38() { # bug 14222
1314 FILES=`find $SRC -type f -mtime +1 | head -n $COUNT`
1315 log "copying $(echo $FILES | wc -w) files to $DIR/$tdir"
1317 tar cf - $FILES | tar xf - -C $DIR/$tdir || \
1318 error "copying $SRC to $DIR/$tdir"
1320 umount_client $MOUNT
1322 log "rename lov_objid file on MDS"
1323 rm -f $TMP/lov_objid.orig
1325 local dev=${SINGLEMDS}_dev
1326 local MDSDEV=${!dev}
1327 do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.orig\\\" $MDSDEV"
1328 do_facet $SINGLEMDS "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV"
1330 do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.orig"
1331 # check create in mds_lov_connect
1335 [ $V ] && log "verifying $DIR/$tdir/$f"
1336 diff -q $f $DIR/$tdir/$f || ERROR=y
1338 do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new\\\" $MDSDEV"
1339 do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.new"
1340 [ "$ERROR" = "y" ] && error "old and new files are different after connect" || true
1342 # check it's updates in sync
1343 umount_client $MOUNT
1346 do_facet $SINGLEMDS dd if=/dev/zero of=$TMP/lov_objid.clear bs=4096 count=1
1347 do_facet $SINGLEMDS "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV"
1348 do_facet $SINGLEMDS "debugfs -w -R \\\"write $TMP/lov_objid.clear lov_objid\\\" $MDSDEV "
1353 [ $V ] && log "verifying $DIR/$tdir/$f"
1354 diff -q $f $DIR/$tdir/$f || ERROR=y
1356 do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new1\\\" $MDSDEV"
1357 do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.new1"
1358 umount_client $MOUNT
1360 [ "$ERROR" = "y" ] && error "old and new files are different after sync" || true
1362 log "files compared the same"
1365 run_test 38 "MDS recreates missing lov_objid file from OST data"
1371 perl $SRCDIR/leak_finder.pl $TMP/debug 2>&1 | egrep '*** Leak:' &&
1372 error "memory leak detected" || true
1374 run_test 39 "leak_finder recognizes both LUSTRE and LNET malloc messages"
1376 test_40() { # bug 15759
1378 #define OBD_FAIL_TGT_TOOMANY_THREADS 0x706
1379 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000706"
1383 run_test 40 "race during service thread startup"
1385 test_41() { #bug 14134
1387 local dev=${SINGLEMDS}_dev
1388 local MDSDEV=${!dev}
1390 start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nosvc -n
1391 start ost1 `ostdevname 1` $OST_MOUNT_OPTS
1392 start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nomgs
1394 mount_client $MOUNT || return 1
1397 echo "blah blah" > $MOUNT/$tfile
1400 umount_client $MOUNT
1401 stop ost1 -f || return 201
1402 stop_mds -f || return 202
1403 stop_mds -f || return 203
1404 unload_modules || return 204
1407 run_test 41 "mount mds with --nosvc and --nomgs"
1409 test_42() { #bug 14693
1411 check_mount || return 2
1412 do_facet client lctl conf_param lustre.llite.some_wrong_param=10
1413 umount_client $MOUNT
1414 mount_client $MOUNT || return 1
1418 run_test 42 "invalid config param should not prevent client from mounting"
1420 umount_client $MOUNT
1426 check_mount || return 2
1431 #define OBD_FAIL_PTLRPC_LONG_UNLINK 0x50f
1432 do_facet client "lctl set_param fail_loc=0x50f"
1435 manual_umount_client --force || return 3
1436 do_facet client "lctl set_param fail_loc=0x0"
1438 mount_client $MOUNT || return 4
1442 run_test 45 "long unlink handling in ptlrpcd"
1445 equals_msg `basename $0`: test complete
1446 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true