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/$NAME.sh}
41 [ "$SLOW" = "no" ] && EXCEPT_SLOW="0 1 2 3 6 7 15 18 24b 25 30 31 32 33 34a "
50 local facet=$SINGLEMDS
51 local dev=${facet}_dev
55 # who knows if/where $TUNEFS is installed? Better reformat if it fails...
56 do_facet ${facet} "$TUNEFS --writeconf ${!dev}" || echo "tunefs failed, reformatting instead" && reformat
61 # The MGS must be started before the OSTs for a new fs, so start
62 # and stop to generate the startup logs.
71 local facet=$SINGLEMDS
72 # we can not use MDSDEV1 here because SINGLEMDS could be set not to mds1 only
73 local num=$(echo $facet | tr -d "mds")
74 local dev=$(mdsdevname $num)
75 echo "start mds service on `facet_active_host $facet`"
76 start $facet ${dev} $MDS_MOUNT_OPTS || return 94
80 echo "stop mds service on `facet_active_host $SINGLEMDS`"
81 # These tests all use non-failover stop
82 stop $SINGLEMDS -f || return 97
86 echo "start ost1 service on `facet_active_host ost1`"
87 start ost1 `ostdevname 1` $OST_MOUNT_OPTS || return 95
91 echo "stop ost1 service on `facet_active_host ost1`"
92 # These tests all use non-failover stop
93 stop ost1 -f || return 98
97 echo "start ost2 service on `facet_active_host ost2`"
98 start ost2 `ostdevname 2` $OST_MOUNT_OPTS || return 92
102 echo "stop ost2 service on `facet_active_host ost2`"
103 # These tests all use non-failover stop
104 stop ost2 -f || return 93
109 echo "mount $FSNAME on ${MOUNTPATH}....."
110 zconf_mount `hostname` $MOUNTPATH || return 96
114 local SAVEMOUNTOPT=$MOUNTOPT
115 MOUNTOPT="remount,$1"
117 echo "remount '$1' lustre on ${MOUNTPATH}....."
118 zconf_mount `hostname` $MOUNTPATH || return 96
119 MOUNTOPT=$SAVEMOUNTOPT
124 echo "umount lustre on ${MOUNTPATH}....."
125 zconf_umount `hostname` $MOUNTPATH || return 97
128 manual_umount_client(){
129 echo "manual umount lustre on ${MOUNT}...."
130 do_facet client "umount -d $MOUNT"
140 stop_mds || return 201
141 stop_ost || return 202
142 unload_modules || return 203
146 umount_client $MOUNT || return 200
147 cleanup_nocli || return $?
151 do_facet client "cp /etc/passwd $DIR/a" || return 71
152 do_facet client "rm $DIR/a" || return 72
153 # make sure lustre is actually mounted (touch will block,
154 # but grep won't, so do it after)
155 do_facet client "grep $MOUNT' ' /proc/mounts > /dev/null" || return 73
156 echo "setup single mount lustre success"
160 do_facet client "touch $DIR/a" || return 71
161 do_facet client "rm $DIR/a" || return 72
162 do_facet client "touch $DIR2/a" || return 73
163 do_facet client "rm $DIR2/a" || return 74
164 echo "setup double mount lustre success"
169 if [ "$ONLY" == "setup" ]; then
174 if [ "$ONLY" == "cleanup" ]; then
179 #create single point mountpoint
187 check_mount || return 41
190 run_test 0 "single mount setup"
194 echo "start ost second time..."
196 check_mount || return 42
199 run_test 1 "start up ost twice (should return errors)"
204 echo "start mds second time.."
207 check_mount || return 43
210 run_test 2 "start up mds twice (should return err)"
214 #mount.lustre returns an error if already in mtab
215 mount_client $MOUNT && return $?
216 check_mount || return 44
219 run_test 3 "mount client twice (should return err)"
223 touch $DIR/$tfile || return 85
227 # ok for ost to fail shutdown
228 if [ 202 -ne $eno ]; then
233 run_test 4 "force cleanup ost, then cleanup"
235 test_5a() { # was test_5
237 touch $DIR/$tfile || return 1
238 fuser -m -v $MOUNT && echo "$MOUNT is in use by user space process."
240 stop_mds -f || return 2
242 # cleanup may return an error from the failed
243 # disconnects; for now I'll consider this successful
244 # if all the modules have unloaded.
248 echo "killing umount"
249 kill -TERM $UMOUNT_PID
250 echo "waiting for umount to finish"
252 if grep " $MOUNT " /proc/mounts; then
253 echo "test 5: /proc/mounts after failed umount"
257 echo "killing umount"
258 kill -TERM $UMOUNT_PID
259 echo "waiting for umount to finish"
261 grep " $MOUNT " /proc/mounts && echo "test 5: /proc/mounts after second umount" && return 11
265 # stop_mds is a no-op here, and should not fail
266 cleanup_nocli || return $?
267 # df may have lingering entry
269 # mtab may have lingering entry
273 while [ "$WAIT" -ne "$MAX_WAIT" ]; do
275 grep -q $MOUNT" " /etc/mtab || break
276 echo "Waiting /etc/mtab updated ... "
277 WAIT=$(( WAIT + sleep))
279 [ "$WAIT" -eq "$MAX_WAIT" ] && error "/etc/mtab is not updated in $WAIT secs"
280 echo "/etc/mtab updated in $WAIT secs"
282 run_test 5a "force cleanup mds, then cleanup"
286 [ -d $MOUNT ] || mkdir -p $MOUNT
287 grep " $MOUNT " /etc/mtab && echo "test 5b: mtab before mount" && return 10
288 mount_client $MOUNT && return 1
289 grep " $MOUNT " /etc/mtab && echo "test 5b: mtab after failed mount" && return 11
291 # stop_mds is a no-op here, and should not fail
292 cleanup_nocli || return $?
295 run_test 5b "mds down, cleanup after failed mount (bug 2712) (should return errs)"
300 [ -d $MOUNT ] || mkdir -p $MOUNT
301 grep " $MOUNT " /etc/mtab && echo "test 5c: mtab before mount" && return 10
302 local oldfs="${FSNAME}"
303 FSNAME="wrong.${FSNAME}"
304 mount_client $MOUNT || :
306 grep " $MOUNT " /etc/mtab && echo "test 5c: mtab after failed mount" && return 11
308 cleanup_nocli || return $?
310 run_test 5c "cleanup after failed mount (bug 2712) (should return errs)"
316 grep " $MOUNT " /etc/mtab && echo "test 5d: mtab before mount" && return 10
317 mount_client $MOUNT || return 1
319 grep " $MOUNT " /etc/mtab && echo "test 5d: mtab after unmount" && return 11
322 run_test 5d "mount with ost down"
328 #define OBD_FAIL_PTLRPC_DELAY_SEND 0x506
329 do_facet client "lctl set_param fail_loc=0x80000506"
330 grep " $MOUNT " /etc/mtab && echo "test 5e: mtab before mount" && return 10
331 mount_client $MOUNT || echo "mount failed (not fatal)"
333 grep " $MOUNT " /etc/mtab && echo "test 5e: mtab after unmount" && return 11
336 run_test 5e "delayed connect, don't crash (bug 10268)"
341 mount_client ${MOUNT} || return 87
342 touch $DIR/a || return 86
345 run_test 6 "manual umount, then mount again"
350 cleanup_nocli || return $?
352 run_test 7 "manual umount, then cleanup"
357 check_mount2 || return 45
358 umount_client $MOUNT2
361 run_test 8 "double mount setup"
366 do_facet ost1 lctl set_param debug=\'inode trace\' || return 1
367 do_facet ost1 lctl set_param subsystem_debug=\'mds ost\' || return 1
369 CHECK_PTLDEBUG="`do_facet ost1 lctl get_param -n debug`"
370 if [ "$CHECK_PTLDEBUG" ] && [ "$CHECK_PTLDEBUG" = "trace inode" ];then
371 echo "lnet.debug success"
373 echo "lnet.debug: want 'trace inode', have '$CHECK_PTLDEBUG'"
376 CHECK_SUBSYS="`do_facet ost1 lctl get_param -n subsystem_debug`"
377 if [ "$CHECK_SUBSYS" ] && [ "$CHECK_SUBSYS" = "mds ost" ]; then
378 echo "lnet.subsystem_debug success"
380 echo "lnet.subsystem_debug: want 'mds ost', have '$CHECK_SUBSYS'"
383 stop_ost || return $?
386 run_test 9 "test ptldebug and subsystem for mkfs"
388 # LOGS/PENDING do not exist anymore since CMD3
390 local TMPMTPT="${TMP}/conf16"
391 local dev=${SINGLEMDS}_dev
393 if [ ! -e "$MDSDEV" ]; then
394 log "no $MDSDEV existing, so mount Lustre to create one"
396 check_mount || return 41
400 [ -f "$MDSDEV" ] && LOOPOPT="-o loop"
402 log "change the mode of $MDSDEV/OBJECTS to 555"
403 do_facet $SINGLEMDS "mkdir -p $TMPMTPT &&
404 mount $LOOPOPT -t $FSTYPE $MDSDEV $TMPMTPT &&
405 chmod 555 $TMPMTPT/OBJECTS &&
406 umount $TMPMTPT" || return $?
408 log "mount Lustre to change the mode of OBJECTS, then umount Lustre"
410 check_mount || return 41
413 log "read the mode of OBJECTS and check if they has been changed properly"
414 EXPECTEDOBJECTSMODE=`do_facet $SINGLEMDS "debugfs -R 'stat OBJECTS' $MDSDEV 2> /dev/null" | grep 'Mode: ' | sed -e "s/.*Mode: *//" -e "s/ *Flags:.*//"`
416 if [ "$EXPECTEDOBJECTSMODE" = "0777" ]; then
417 log "Success:Lustre change the mode of OBJECTS correctly"
419 error "Lustre does not change mode of OBJECTS properly"
422 run_test 16 "verify that lustre will correct the mode of OBJECTS"
425 local dev=${SINGLEMDS}_dev
428 if [ ! -e "$MDSDEV" ]; then
429 echo "no $MDSDEV existing, so mount Lustre to create one"
431 check_mount || return 41
435 echo "Remove mds config log"
436 do_facet $SINGLEMDS "debugfs -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MDSDEV || return \$?" || return $?
439 start_mds && return 42
442 run_test 17 "Verify failed mds_postsetup won't fail assertion (2936) (should return errs)"
445 [ "$FSTYPE" != "ldiskfs" ] && skip "not needed for FSTYPE=$FSTYPE" && return
447 local dev=${SINGLEMDS}_dev
453 # check if current MDSSIZE is large enough
454 [ $MDSSIZE -ge $MIN ] && OK=1 && myMDSSIZE=$MDSSIZE && \
455 log "use MDSSIZE=$MDSSIZE"
457 # check if the global config has a large enough MDSSIZE
458 [ -z "$OK" -a ! -z "$STORED_MDSSIZE" ] && [ $STORED_MDSSIZE -ge $MIN ] && \
459 OK=1 && myMDSSIZE=$STORED_MDSSIZE && \
460 log "use STORED_MDSSIZE=$STORED_MDSSIZE"
462 # check if the block device is large enough
463 [ -z "$OK" -a -b $MDSDEV ] && \
464 [ "$(dd if=$MDSDEV of=/dev/null bs=1k count=1 skip=$MIN 2>&1 |
465 awk '($3 == "in") { print $1 }')" = "1+0" ] && OK=1 && \
466 myMDSSIZE=$MIN && log "use device $MDSDEV with MIN=$MIN"
468 # check if a loopback device has enough space for fs metadata (5%)
469 [ -z "$OK" ] && [ -f $MDSDEV -o ! -e $MDSDEV ] &&
470 SPACE=$(df -P $(dirname $MDSDEV) |
471 awk '($1 != "Filesystem") {print $4}') &&
472 [ $SPACE -gt $((MIN / 20)) ] && OK=1 && myMDSSIZE=$MIN && \
473 log "use file $MDSDEV with MIN=$MIN"
475 [ -z "$OK" ] && skip "$MDSDEV too small for ${MIN}kB MDS" && return
478 echo "mount mds with large journal..."
479 local OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS
481 MDS_MKFS_OPTS="--mgs --mdt --fsname=$FSNAME --device-size=$myMDSSIZE --param sys.timeout=$TIMEOUT $MDSOPT"
484 echo "mount lustre system..."
486 check_mount || return 41
488 echo "check journal size..."
489 local FOUNDSIZE=`do_facet mds "debugfs -c -R 'stat <8>' $MDSDEV" | awk '/Size: / { print $NF; exit;}'`
490 if [ $FOUNDSIZE -gt $((32 * 1024 * 1024)) ]; then
491 log "Success: mkfs creates large journals. Size: $((FOUNDSIZE >> 20))M"
493 error "expected journal size > 32M, found $((FOUNDSIZE >> 20))M"
498 MDS_MKFS_OPTS=$OLD_MDS_MKFS_OPTS
501 run_test 18 "check mkfs creates large journals"
504 start_mds || return 1
505 stop_mds -f || return 2
507 run_test 19a "start/stop MDS without OSTs"
510 start_ost || return 1
511 stop_ost -f || return 2
513 run_test 19b "start/stop OSTs without MDS"
516 # first format the ost/mdt
520 check_mount || return 43
522 remount_client ro $MOUNT || return 44
523 touch $DIR/$tfile && echo "$DIR/$tfile created incorrectly" && return 45
524 [ -e $DIR/$tfile ] && echo "$DIR/$tfile exists incorrectly" && return 46
525 remount_client rw $MOUNT || return 47
527 [ ! -f $DIR/$tfile ] && echo "$DIR/$tfile missing" && return 48
528 MCNT=`grep -c $MOUNT /etc/mtab`
529 [ "$MCNT" -ne 1 ] && echo "$MOUNT in /etc/mtab $MCNT times" && return 49
534 run_test 20 "remount ro,rw mounts work and doesn't break /etc/mtab"
542 run_test 21a "start mds before ost, stop ost first"
550 run_test 21b "start ost before mds, stop mds first"
560 run_test 21c "start mds between two osts, stop mds last"
563 #reformat to remove all logs
566 echo Client mount before any osts are in the logs
568 check_mount && return 41
571 echo Client mount with ost in logs, but none running
575 # check_mount will block trying to contact ost
579 echo Client mount with a running ost
582 # if gss enabled, wait full time to let connection from
583 # mds to ost be established, due to the mismatch between
584 # initial connect timeout and gss context negotiation timeout.
585 # This perhaps could be remove after AT landed.
586 echo "sleep $((TIMEOUT + TIMEOUT + TIMEOUT))s"
587 sleep $((TIMEOUT + TIMEOUT + TIMEOUT))
590 check_mount || return 41
595 run_test 22 "start a client before osts (should return errs)"
597 test_23a() { # was test_23
601 # force down client so that recovering mds waits for reconnect
602 local running=$(grep -c $MOUNT /proc/mounts) || true
603 if [ $running -ne 0 ]; then
604 echo "Stopping client $MOUNT (opts: -f)"
608 # enter recovery on mds
610 # try to start a new client
611 mount_client $MOUNT &
613 MOUNT_PID=$(ps -ef | grep "t lustre" | grep -v grep | awk '{print $2}')
614 MOUNT_LUSTRE_PID=`ps -ef | grep mount.lustre | grep -v grep | awk '{print $2}'`
615 echo mount pid is ${MOUNT_PID}, mount.lustre pid is ${MOUNT_LUSTRE_PID}
617 ps --ppid $MOUNT_LUSTRE_PID
618 # FIXME why o why can't I kill these? Manual "ctrl-c" works...
619 kill -TERM $MOUNT_LUSTRE_PID
620 echo "waiting for mount to finish"
622 # we can not wait $MOUNT_PID because it is not a child of this shell
628 while [ "$WAIT" -lt "$MAX_WAIT" ]; do
630 PID1=$(ps -ef | awk '{print $2}' | grep -w $MOUNT_PID)
631 PID2=$(ps -ef | awk '{print $2}' | grep -w $MOUNT_LUSTRE_PID)
634 [ -z "$PID1" -a -z "$PID2" ] && break
635 echo "waiting for mount to finish ... "
636 WAIT=$(( WAIT + sleep))
638 [ "$WAIT" -eq "$MAX_WAIT" ] && error "MOUNT_PID $MOUNT_PID and \
639 MOUNT__LUSTRE_PID $MOUNT__LUSTRE_PID still not killed in $WAIT secs"
644 run_test 23a "interrupt client during recovery mount delay"
649 test_23b() { # was test_23
652 # Simulate -EINTR during mount OBD_FAIL_LDLM_CLOSE_THREAD
653 lctl set_param fail_loc=0x80000313
657 run_test 23b "Simulate -EINTR during mount"
659 fs2mds_HOST=$mds_HOST
660 fs2ost_HOST=$ost_HOST
664 echo "umount $MOUNT2 ..."
665 umount $MOUNT2 || true
666 echo "stopping fs2mds ..."
667 stop fs2mds -f || true
668 echo "stopping fs2ost ..."
669 stop fs2ost -f || true
676 [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST
677 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then
678 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
679 skip "mixed loopback and real device not working" && return
682 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
683 local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
685 # test 8-char fsname as well
686 local FSNAME2=test1234
687 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --nomgs --mgsnode=$MGSNID --reformat $fs2mdsdev || exit 10
689 add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2ostdev || exit 10
692 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT
693 start fs2ost $fs2ostdev $OST_MOUNT_OPTS
695 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1
697 check_mount || return 2
698 # files written on 1 should not show up on 2
699 cp /etc/passwd $DIR/$tfile
701 [ -e $MOUNT2/$tfile ] && error "File bleed" && return 7
704 cp /etc/passwd $MOUNT2/b || return 3
705 rm $MOUNT2/b || return 4
706 # 2 is actually mounted
707 grep $MOUNT2' ' /proc/mounts > /dev/null || return 5
709 facet_failover fs2mds
710 facet_failover fs2ost
713 # the MDS must remain up until last MDT
715 MDS=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1)
716 [ -z "$MDS" ] && error "No MDT" && return 8
718 cleanup_nocli || return 6
720 run_test 24a "Multiple MDTs on a single node"
723 if [ -z "$fs2mds_DEV" ]; then
724 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
725 skip "mixed loopback and real device not working" && return
728 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
730 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME}2 --mgs --reformat $fs2mdsdev || exit 10
732 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && return 2
735 run_test 24b "Multiple MGSs on a single node (should return err)"
739 check_mount || return 2
740 local MODULES=$($LCTL modules | awk '{ print $2 }')
741 rmmod $MODULES 2>/dev/null || true
744 run_test 25 "Verify modules are referenced"
748 # we need modules before mount for sysctl, so make sure...
749 do_facet $SINGLEMDS "lsmod | grep -q lustre || modprobe lustre"
750 #define OBD_FAIL_MDS_FS_SETUP 0x135
751 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000135"
752 start_mds && echo MDS started && return 1
753 lctl get_param -n devices
754 DEVS=$(lctl get_param -n devices | wc -l)
755 [ $DEVS -gt 0 ] && return 2
756 unload_modules || return 203
758 run_test 26 "MDT startup failure cleans LOV (should return errs)"
764 local ORIG=$(do_facet $myfacet "$TEST")
765 if [ $# -gt 3 ]; then
771 echo "Setting $PARAM from $ORIG to $FINAL"
772 do_facet $SINGLEMDS "$LCTL conf_param $PARAM=$FINAL" || error conf_param failed
778 RESULT=$(do_facet $myfacet "$TEST")
779 if [ $RESULT -eq $FINAL ]; then
780 echo "Updated config after $WAIT sec (got $RESULT)"
784 if [ $WAIT -eq $MAX ]; then
785 echo "Config update not seen: wanted $FINAL got $RESULT"
788 echo "Waiting $(($MAX - $WAIT)) secs for config update"
793 start_ost || return 1
794 start_mds || return 2
795 echo "Requeue thread should have started: "
796 ps -e | grep ll_cfg_requeue
797 set_and_check ost1 "lctl get_param -n obdfilter.$FSNAME-OST0000.client_cache_seconds" "$FSNAME-OST0000.ost.client_cache_seconds" || return 3
800 run_test 27a "Reacquire MGS lock if OST started first"
805 local device=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }')
807 facet_failover $SINGLEMDS
808 set_and_check $SINGLEMDS "lctl get_param -n mdt.$device.identity_acquire_expire" "$device.mdt.identity_acquire_expire" || return 3
809 set_and_check client "lctl get_param -n mdc.$device-mdc-*.max_rpcs_in_flight" "$device.mdc.max_rpcs_in_flight" || return 4
813 run_test 27b "Reacquire MGS lock after failover"
817 TEST="lctl get_param -n llite.$FSNAME-*.max_read_ahead_whole_mb"
818 PARAM="$FSNAME.llite.max_read_ahead_whole_mb"
821 set_and_check client "$TEST" "$PARAM" $FINAL || return 3
822 FINAL=$(($FINAL + 1))
823 set_and_check client "$TEST" "$PARAM" $FINAL || return 4
824 umount_client $MOUNT || return 200
827 if [ $RESULT -ne $FINAL ]; then
828 echo "New config not seen: wanted $FINAL got $RESULT"
831 echo "New config success: got $RESULT"
833 set_and_check client "$TEST" "$PARAM" $ORIG || return 5
836 run_test 28 "permanent parameter setting"
839 [ "$OSTCOUNT" -lt "2" ] && skip "$OSTCOUNT < 2, skipping" && return
840 setup > /dev/null 2>&1
844 local PARAM="$FSNAME-OST0001.osc.active"
845 local PROC_ACT="osc.$FSNAME-OST0001-osc-[^M]*.active"
846 local PROC_UUID="osc.$FSNAME-OST0001-osc-[^M]*.ost_server_uuid"
848 ACTV=$(lctl get_param -n $PROC_ACT)
850 set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $DEAC || return 2
851 # also check ost_server_uuid status
852 RESULT=$(lctl get_param -n $PROC_UUID | grep DEACTIV)
853 if [ -z "$RESULT" ]; then
854 echo "Live client not deactivated: $(lctl get_param -n $PROC_UUID)"
857 echo "Live client success: got $RESULT"
861 local MPROC="osc.$FSNAME-OST0001-osc-[M]*.active"
866 RESULT=`do_facet $SINGLEMDS " lctl get_param -n $MPROC"`
867 [ ${PIPESTATUS[0]} = 0 ] || error "Can't read $MPROC"
868 if [ $RESULT -eq $DEAC ]; then
869 echo "MDT deactivated also after $WAIT sec (got $RESULT)"
873 if [ $WAIT -eq $MAX ]; then
874 echo "MDT not deactivated: wanted $DEAC got $RESULT"
877 echo "Waiting $(($MAX - $WAIT)) secs for MDT deactivated"
880 # test new client starts deactivated
881 umount_client $MOUNT || return 200
883 RESULT=$(lctl get_param -n $PROC_UUID | grep DEACTIV | grep NEW)
884 if [ -z "$RESULT" ]; then
885 echo "New client not deactivated from start: $(lctl get_param -n $PROC_UUID)"
888 echo "New client success: got $RESULT"
891 # make sure it reactivates
892 set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $ACTV || return 6
897 #writeconf to remove all ost2 traces for subsequent tests
903 run_test 29 "permanently remove an OST"
908 TEST="lctl get_param -n llite.$FSNAME-*.max_read_ahead_whole_mb"
910 LIST=(1 2 3 4 5 4 3 2 1 2 3 4 5 4 3 2 1 2 3 4 5)
911 for i in ${LIST[@]}; do
912 set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $i || return 3
914 # make sure client restart still works
916 mount_client $MOUNT || return 4
917 [ "$($TEST)" -ne "$i" ] && return 5
918 set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $ORIG || return 6
921 run_test 30 "Big config llog"
923 test_31() { # bug 10734
924 # ipaddr must not exist
925 mount -t lustre 4.3.2.1@tcp:/lustre $MOUNT || true
928 run_test 31 "Connect to non-existent node (shouldn't crash)"
931 # XXX - make this run on client-only systems with real hardware on
933 # there appears to be a lot of assumption here about loopback
935 # or maybe this test is just totally useless on a client-only system
936 [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
937 [ "$mds_HOST" = "`hostname`" ] || { skip "remote MDS" && return 0; }
938 [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
939 { skip "remote OST" && return 0; }
941 [ -z "$TUNEFS" ] && skip "No tunefs" && return
942 local DISK1_4=$LUSTRE/tests/disk1_4.zip
943 [ ! -r $DISK1_4 ] && skip "Cant find $DISK1_4, skipping" && return
945 local tmpdir=$TMP/conf32a
946 unzip -o -j -d $tmpdir $DISK1_4 || { skip "Cant unzip $DISK1_4, skipping" && return ; }
948 lctl set_param debug=$PTLDEBUG
950 $TUNEFS $tmpdir/mds || error "tunefs failed"
951 # nids are wrong, so client wont work, but server should start
952 start mds $tmpdir/mds "-o loop,exclude=lustre-OST0000" || return 3
953 local UUID=$(lctl get_param -n mdt.lustre-MDT0000.uuid)
955 [ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID"
957 $TUNEFS --mgsnode=`hostname` $tmpdir/ost1 || error "tunefs failed"
958 start ost1 $tmpdir/ost1 "-o loop" || return 5
959 UUID=$(lctl get_param -n obdfilter.lustre-OST0000.uuid)
961 [ "$UUID" == "ost1_UUID" ] || error "UUID is wrong: $UUID"
963 local NID=$($LCTL list_nids | head -1)
965 echo "OSC changes should return err:"
966 $LCTL conf_param lustre-OST0000.osc.max_dirty_mb=15 && return 7
967 $LCTL conf_param lustre-OST0000.failover.node=$NID && return 8
969 echo "MDC changes should succeed:"
970 $LCTL conf_param lustre-MDT0000.mdc.max_rpcs_in_flight=9 || return 9
971 $LCTL conf_param lustre-MDT0000.failover.node=$NID || return 10
974 # With a new good MDT failover nid, we should be able to mount a client
975 # (but it cant talk to OST)
976 local OLDMOUNTOPT=$MOUNTOPT
977 MOUNTOPT="exclude=lustre-OST0000"
979 MOUNTOPT=$OLDMOUNTOPT
980 set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "lustre-MDT0000.mdc.max_rpcs_in_flight" ||
983 zconf_umount `hostname` $MOUNT -f
987 # mount a second time to make sure we didnt leave upgrade flag on
989 $TUNEFS --dryrun $tmpdir/mds || error "tunefs failed"
991 start mds $tmpdir/mds "-o loop,exclude=lustre-OST0000" || return 12
994 rm -rf $tmpdir || true # true is only for TMP on NFS
996 run_test 32a "Upgrade from 1.4 (not live)"
999 # XXX - make this run on client-only systems with real hardware on
1001 # there appears to be a lot of assumption here about loopback
1003 # or maybe this test is just totally useless on a client-only system
1004 [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
1005 [ "$mds_HOST" = "`hostname`" ] || { skip "remote MDS" && return 0; }
1006 [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
1007 { skip "remote OST" && return 0; }
1009 [ -z "$TUNEFS" ] && skip "No tunefs" && return
1010 local DISK1_4=$LUSTRE/tests/disk1_4.zip
1011 [ ! -r $DISK1_4 ] && skip "Cant find $DISK1_4, skipping" && return
1013 local tmpdir=$TMP/conf32b
1014 unzip -o -j -d $tmpdir $DISK1_4 || { skip "Cant unzip $DISK1_4, skipping" && return ; }
1016 lctl set_param debug=$PTLDEBUG
1019 # writeconf will cause servers to register with their current nids
1020 $TUNEFS --writeconf --fsname=$NEWNAME $tmpdir/mds || error "tunefs failed"
1021 start mds $tmpdir/mds "-o loop" || return 3
1022 local UUID=$(lctl get_param -n mdt.${NEWNAME}-MDT0000.uuid)
1024 [ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID"
1026 $TUNEFS --mgsnode=`hostname` --fsname=$NEWNAME --writeconf $tmpdir/ost1 || error "tunefs failed"
1027 start ost1 $tmpdir/ost1 "-o loop" || return 5
1028 UUID=$(lctl get_param -n obdfilter.${NEWNAME}-OST0000.uuid)
1030 [ "$UUID" == "ost1_UUID" ] || error "UUID is wrong: $UUID"
1032 echo "OSC changes should succeed:"
1033 $LCTL conf_param ${NEWNAME}-OST0000.osc.max_dirty_mb=15 || return 7
1034 $LCTL conf_param ${NEWNAME}-OST0000.failover.node=$NID || return 8
1036 echo "MDC changes should succeed:"
1037 $LCTL conf_param ${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight=9 || return 9
1040 # MDT and OST should have registered with new nids, so we should have
1041 # a fully-functioning client
1042 echo "Check client and old fs contents"
1047 set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight" || return 11
1048 [ "$(cksum $MOUNT/passwd | cut -d' ' -f 1,2)" == "2479747619 779" ] || return 12
1052 rm -rf $tmpdir || true # true is only for TMP on NFS
1054 run_test 32b "Upgrade from 1.4 with writeconf"
1056 test_33a() { # bug 12333, was test_33
1058 local FSNAME2=test-123
1059 [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST
1061 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then
1062 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
1063 skip "mixed loopback and real device not working" && return
1066 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
1067 local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
1068 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10
1069 add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --index=8191 --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10
1071 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT
1072 start fs2ost $fs2ostdev $OST_MOUNT_OPTS
1073 do_facet $SINGLEMDS "$LCTL conf_param $FSNAME2.sys.timeout=200" || rc=1
1075 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || rc=2
1078 cp /etc/hosts $MOUNT2/ || rc=3
1079 $LFS getstripe $MOUNT2/hosts
1084 rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev
1085 cleanup_nocli || rc=6
1088 run_test 33a "Mount ost with a large index number"
1090 test_33b() { # was test_34
1093 do_facet client dd if=/dev/zero of=$MOUNT/24 bs=1024k count=1
1094 # Drop lock cancelation reply during umount
1095 #define OBD_FAIL_LDLM_CANCEL 0x304
1096 do_facet client lctl set_param fail_loc=0x80000304
1097 #lctl set_param debug=-1
1098 umount_client $MOUNT
1101 run_test 33b "Drop cancel during umount"
1105 do_facet client "sh runmultiop_bg_pause $DIR/file O_c"
1106 manual_umount_client
1108 do_facet client killall -USR1 multiop
1109 if [ $rc -eq 0 ]; then
1110 error "umount not fail!"
1115 run_test 34a "umount with opened file should be fail"
1120 touch $DIR/$tfile || return 1
1121 stop_mds --force || return 2
1123 manual_umount_client --force
1125 if [ $rc -ne 0 ]; then
1126 error "mtab after failed umount - rc $rc"
1132 run_test 34b "force umount with failed mds should be normal"
1136 touch $DIR/$tfile || return 1
1137 stop_ost --force || return 2
1139 manual_umount_client --force
1141 if [ $rc -ne 0 ]; then
1142 error "mtab after failed umount - rc $rc"
1148 run_test 34c "force umount with failed ost should be normal"
1150 test_35() { # bug 12459
1153 DBG_SAVE="`lctl get_param -n debug`"
1154 lctl set_param debug="ha"
1156 log "Set up a fake failnode for the MDS"
1158 local device=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1)
1159 do_facet $SINGLEMDS $LCTL conf_param ${device}.failover.node=$FAKENID || return 4
1161 log "Wait for RECONNECT_INTERVAL seconds (10s)"
1164 MSG="conf-sanity.sh test_35 `date +%F%kh%Mm%Ss`"
1167 log "Stopping the MDT:"
1168 stop_mds || return 5
1170 df $MOUNT > /dev/null 2>&1 &
1172 log "Restarting the MDT:"
1173 start_mds || return 6
1174 log "Wait for df ($DFPID) ... "
1177 lctl set_param debug="$DBG_SAVE"
1179 # retrieve from the log the first server that the client tried to
1180 # contact after the connection loss
1181 $LCTL dk $TMP/lustre-log-$TESTNAME.log
1182 NEXTCONN=`awk "/${MSG}/ {start = 1;}
1183 /import_select_connection.$device-mdc.* using connection/ {
1185 if (\\\$NF ~ /$FAKENID/)
1191 }" $TMP/lustre-log-$TESTNAME.log`
1192 [ "$NEXTCONN" != "0" ] && log "The client didn't try to reconnect to the last active server (tried ${NEXTCONN} instead)" && return 7
1195 run_test 35 "Reconnect to the last active server first"
1199 local FSNAME2=test1234
1200 local fs3ost_HOST=$ost_HOST
1202 [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST && fs3ost_HOST=$ost1_HOST
1205 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" -o -z "$fs3ost_DEV" ]; then
1206 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
1207 skip "mixed loopback and real device not working" && return
1209 [ $OSTCOUNT -lt 2 ] && skip "skipping test for single OST" && return
1211 [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
1212 { skip "remote OST" && return 0; }
1214 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
1215 local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
1216 local fs3ostdev=${fs3ost_DEV:-$(ostdevname 2)_2}
1217 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10
1218 # XXX after we support non 4K disk blocksize, change following --mkfsoptions with
1220 add fs2ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10
1221 add fs3ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=$MGSNID --reformat $fs3ostdev || exit 10
1223 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS
1224 start fs2ost $fs2ostdev $OST_MOUNT_OPTS
1225 start fs3ost $fs3ostdev $OST_MOUNT_OPTS
1227 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1
1229 sleep 5 # until 11778 fixed
1231 dd if=/dev/zero of=$MOUNT2/$tfile bs=1M count=7 || return 2
1233 BKTOTAL=`lctl get_param -n obdfilter.*.kbytestotal | awk 'BEGIN{total=0}; {total+=$1}; END{print total}'`
1234 BKFREE=`lctl get_param -n obdfilter.*.kbytesfree | awk 'BEGIN{free=0}; {free+=$1}; END{print free}'`
1235 BKAVAIL=`lctl get_param -n obdfilter.*.kbytesavail | awk 'BEGIN{avail=0}; {avail+=$1}; END{print avail}'`
1236 STRING=`df -P $MOUNT2 | tail -n 1 | awk '{print $2","$3","$4}'`
1237 DFTOTAL=`echo $STRING | cut -d, -f1`
1238 DFUSED=`echo $STRING | cut -d, -f2`
1239 DFAVAIL=`echo $STRING | cut -d, -f3`
1240 DFFREE=$(($DFTOTAL - $DFUSED))
1242 ALLOWANCE=$((64 * $OSTCOUNT))
1244 if [ $DFTOTAL -lt $(($BKTOTAL - $ALLOWANCE)) ] ||
1245 [ $DFTOTAL -gt $(($BKTOTAL + $ALLOWANCE)) ] ; then
1246 echo "**** FAIL: df total($DFTOTAL) mismatch OST total($BKTOTAL)"
1249 if [ $DFFREE -lt $(($BKFREE - $ALLOWANCE)) ] ||
1250 [ $DFFREE -gt $(($BKFREE + $ALLOWANCE)) ] ; then
1251 echo "**** FAIL: df free($DFFREE) mismatch OST free($BKFREE)"
1254 if [ $DFAVAIL -lt $(($BKAVAIL - $ALLOWANCE)) ] ||
1255 [ $DFAVAIL -gt $(($BKAVAIL + $ALLOWANCE)) ] ; then
1256 echo "**** FAIL: df avail($DFAVAIL) mismatch OST avail($BKAVAIL)"
1261 stop fs3ost -f || return 200
1262 stop fs2ost -f || return 201
1263 stop fs2mds -f || return 202
1264 rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev $fs3ostdev
1265 unload_modules || return 203
1268 run_test 36 "df report consistency on OSTs with different block size"
1271 [ -n "$CLIENTONLY" -o -n "$CLIENTMODSONLY" ] && skip "client only testing" && return 0
1272 LOCAL_MDSDEV="$TMP/mdt.img"
1273 SYM_MDSDEV="$TMP/sym_mdt.img"
1275 echo "MDS : $LOCAL_MDSDEV"
1276 echo "SYMLINK : $SYM_MDSDEV"
1280 mkfs.lustre --reformat --fsname=lustre --mdt --mgs --device-size=9000 $LOCAL_MDSDEV ||
1281 error "mkfs.lustre $LOCAL_MDSDEV failed"
1282 ln -s $LOCAL_MDSDEV $SYM_MDSDEV
1284 echo "mount symlink device - $SYM_MDSDEV"
1286 mount_op=`mount -v -t lustre -o loop $SYM_MDSDEV ${MOUNT%/*}/mds 2>&1 | grep "unable to set tunable"`
1287 umount -d ${MOUNT%/*}/mds
1288 rm -f $LOCAL_MDSDEV $SYM_MDSDEV
1290 if [ -n "$mount_op" ]; then
1291 error "**** FAIL: set tunables failed for symlink device"
1295 run_test 37 "verify set tunables works for symlink device"
1297 test_38() { # bug 14222
1302 FILES=`find $SRC -type f -mtime +1 | head -n $COUNT`
1303 log "copying $(echo $FILES | wc -w) files to $DIR/$tdir"
1305 tar cf - $FILES | tar xf - -C $DIR/$tdir || \
1306 error "copying $SRC to $DIR/$tdir"
1308 umount_client $MOUNT
1310 log "rename lov_objid file on MDS"
1311 rm -f $TMP/lov_objid.orig
1313 local dev=${SINGLEMDS}_dev
1314 local MDSDEV=${!dev}
1315 do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.orig\\\" $MDSDEV"
1316 do_facet $SINGLEMDS "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV"
1318 do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.orig"
1319 # check create in mds_lov_connect
1323 [ $V ] && log "verifying $DIR/$tdir/$f"
1324 diff -q $f $DIR/$tdir/$f || ERROR=y
1326 do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new\\\" $MDSDEV"
1327 do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.new"
1328 [ "$ERROR" = "y" ] && error "old and new files are different after connect" || true
1330 # check it's updates in sync
1331 umount_client $MOUNT
1334 do_facet $SINGLEMDS dd if=/dev/zero of=$TMP/lov_objid.clear bs=4096 count=1
1335 do_facet $SINGLEMDS "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV"
1336 do_facet $SINGLEMDS "debugfs -w -R \\\"write $TMP/lov_objid.clear lov_objid\\\" $MDSDEV "
1341 [ $V ] && log "verifying $DIR/$tdir/$f"
1342 diff -q $f $DIR/$tdir/$f || ERROR=y
1344 do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new1\\\" $MDSDEV"
1345 do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.new1"
1346 umount_client $MOUNT
1348 [ "$ERROR" = "y" ] && error "old and new files are different after sync" || true
1350 log "files compared the same"
1353 run_test 38 "MDS recreates missing lov_objid file from OST data"
1359 perl $SRCDIR/leak_finder.pl $TMP/debug 2>&1 | egrep '*** Leak:' &&
1360 error "memory leak detected" || true
1362 run_test 39 "leak_finder recognizes both LUSTRE and LNET malloc messages"
1364 test_40() { # bug 15759
1366 #define OBD_FAIL_TGT_TOOMANY_THREADS 0x706
1367 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000706"
1371 run_test 40 "race during service thread startup"
1373 test_41() { #bug 14134
1375 local dev=${SINGLEMDS}_dev
1376 local MDSDEV=${!dev}
1378 start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nosvc -n
1379 start ost1 `ostdevname 1` $OST_MOUNT_OPTS
1380 start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nomgs
1382 mount_client $MOUNT || return 1
1385 echo "blah blah" > $MOUNT/$tfile
1388 umount_client $MOUNT
1389 stop ost1 -f || return 201
1390 stop_mds -f || return 202
1391 stop_mds -f || return 203
1392 unload_modules || return 204
1395 run_test 41 "mount mds with --nosvc and --nomgs"
1397 test_42() { #bug 14693
1399 check_mount || return 2
1400 do_facet client lctl conf_param lustre.llite.some_wrong_param=10
1401 umount_client $MOUNT
1402 mount_client $MOUNT || return 1
1406 run_test 42 "invalid config param should not prevent client from mounting"
1408 umount_client $MOUNT
1412 equals_msg `basename $0`: test complete
1413 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true