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 17617
15 HEAD_EXCEPT=" 32a 32b 45"
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(){
132 echo "manual umount lustre on ${MOUNT}...."
133 do_facet client "umount -d $MOUNT"
143 stop_mds || return 201
144 stop_ost || return 202
145 unload_modules || return 203
149 umount_client $MOUNT || return 200
150 cleanup_nocli || return $?
154 do_facet client "cp /etc/passwd $DIR/a" || return 71
155 do_facet client "rm $DIR/a" || return 72
156 # make sure lustre is actually mounted (touch will block,
157 # but grep won't, so do it after)
158 do_facet client "grep $MOUNT' ' /proc/mounts > /dev/null" || return 73
159 echo "setup single mount lustre success"
163 do_facet client "touch $DIR/a" || return 71
164 do_facet client "rm $DIR/a" || return 72
165 do_facet client "touch $DIR2/a" || return 73
166 do_facet client "rm $DIR2/a" || return 74
167 echo "setup double mount lustre success"
172 if [ "$ONLY" == "setup" ]; then
177 if [ "$ONLY" == "cleanup" ]; then
182 #create single point mountpoint
190 check_mount || return 41
193 run_test 0 "single mount setup"
197 echo "start ost second time..."
199 check_mount || return 42
202 run_test 1 "start up ost twice (should return errors)"
207 echo "start mds second time.."
210 check_mount || return 43
213 run_test 2 "start up mds twice (should return err)"
217 #mount.lustre returns an error if already in mtab
218 mount_client $MOUNT && return $?
219 check_mount || return 44
222 run_test 3 "mount client twice (should return err)"
226 touch $DIR/$tfile || return 85
230 # ok for ost to fail shutdown
231 if [ 202 -ne $eno ]; then
236 run_test 4 "force cleanup ost, then cleanup"
238 test_5a() { # was test_5
240 touch $DIR/$tfile || return 1
241 fuser -m -v $MOUNT && echo "$MOUNT is in use by user space process."
243 stop_mds -f || return 2
245 # cleanup may return an error from the failed
246 # disconnects; for now I'll consider this successful
247 # if all the modules have unloaded.
251 echo "killing umount"
252 kill -TERM $UMOUNT_PID
253 echo "waiting for umount to finish"
255 if grep " $MOUNT " /proc/mounts; then
256 echo "test 5: /proc/mounts after failed umount"
260 echo "killing umount"
261 kill -TERM $UMOUNT_PID
262 echo "waiting for umount to finish"
264 grep " $MOUNT " /proc/mounts && echo "test 5: /proc/mounts after second umount" && return 11
268 # stop_mds is a no-op here, and should not fail
269 cleanup_nocli || return $?
270 # df may have lingering entry
272 # mtab may have lingering entry
276 while [ "$WAIT" -ne "$MAX_WAIT" ]; do
278 grep -q $MOUNT" " /etc/mtab || break
279 echo "Waiting /etc/mtab updated ... "
280 WAIT=$(( WAIT + sleep))
282 [ "$WAIT" -eq "$MAX_WAIT" ] && error "/etc/mtab is not updated in $WAIT secs"
283 echo "/etc/mtab updated in $WAIT secs"
285 run_test 5a "force cleanup mds, then cleanup"
289 [ -d $MOUNT ] || mkdir -p $MOUNT
290 grep " $MOUNT " /etc/mtab && echo "test 5b: mtab before mount" && return 10
291 mount_client $MOUNT && return 1
292 grep " $MOUNT " /etc/mtab && echo "test 5b: mtab after failed mount" && return 11
294 # stop_mds is a no-op here, and should not fail
295 cleanup_nocli || return $?
298 run_test 5b "mds down, cleanup after failed mount (bug 2712) (should return errs)"
303 [ -d $MOUNT ] || mkdir -p $MOUNT
304 grep " $MOUNT " /etc/mtab && echo "test 5c: mtab before mount" && return 10
305 local oldfs="${FSNAME}"
306 FSNAME="wrong.${FSNAME}"
307 mount_client $MOUNT || :
309 grep " $MOUNT " /etc/mtab && echo "test 5c: mtab after failed mount" && return 11
311 cleanup_nocli || return $?
313 run_test 5c "cleanup after failed mount (bug 2712) (should return errs)"
319 grep " $MOUNT " /etc/mtab && echo "test 5d: mtab before mount" && return 10
320 mount_client $MOUNT || return 1
322 grep " $MOUNT " /etc/mtab && echo "test 5d: mtab after unmount" && return 11
325 run_test 5d "mount with ost down"
331 #define OBD_FAIL_PTLRPC_DELAY_SEND 0x506
332 do_facet client "lctl set_param fail_loc=0x80000506"
333 grep " $MOUNT " /etc/mtab && echo "test 5e: mtab before mount" && return 10
334 mount_client $MOUNT || echo "mount failed (not fatal)"
336 grep " $MOUNT " /etc/mtab && echo "test 5e: mtab after unmount" && return 11
339 run_test 5e "delayed connect, don't crash (bug 10268)"
344 mount_client ${MOUNT} || return 87
345 touch $DIR/a || return 86
348 run_test 6 "manual umount, then mount again"
353 cleanup_nocli || return $?
355 run_test 7 "manual umount, then cleanup"
360 check_mount2 || return 45
361 umount_client $MOUNT2
364 run_test 8 "double mount setup"
369 do_facet ost1 lctl set_param debug=\'inode trace\' || return 1
370 do_facet ost1 lctl set_param subsystem_debug=\'mds ost\' || return 1
372 CHECK_PTLDEBUG="`do_facet ost1 lctl get_param -n debug`"
373 if [ "$CHECK_PTLDEBUG" ] && [ "$CHECK_PTLDEBUG" = "trace inode" ];then
374 echo "lnet.debug success"
376 echo "lnet.debug: want 'trace inode', have '$CHECK_PTLDEBUG'"
379 CHECK_SUBSYS="`do_facet ost1 lctl get_param -n subsystem_debug`"
380 if [ "$CHECK_SUBSYS" ] && [ "$CHECK_SUBSYS" = "mds ost" ]; then
381 echo "lnet.subsystem_debug success"
383 echo "lnet.subsystem_debug: want 'mds ost', have '$CHECK_SUBSYS'"
386 stop_ost || return $?
389 run_test 9 "test ptldebug and subsystem for mkfs"
391 # LOGS/PENDING do not exist anymore since CMD3
393 local TMPMTPT="${TMP}/conf16"
394 local dev=${SINGLEMDS}_dev
396 if [ ! -e "$MDSDEV" ]; then
397 log "no $MDSDEV existing, so mount Lustre to create one"
399 check_mount || return 41
403 [ -f "$MDSDEV" ] && LOOPOPT="-o loop"
405 log "change the mode of $MDSDEV/OBJECTS to 555"
406 do_facet $SINGLEMDS "mkdir -p $TMPMTPT &&
407 mount $LOOPOPT -t $FSTYPE $MDSDEV $TMPMTPT &&
408 chmod 555 $TMPMTPT/OBJECTS &&
409 umount $TMPMTPT" || return $?
411 log "mount Lustre to change the mode of OBJECTS, then umount Lustre"
413 check_mount || return 41
416 log "read the mode of OBJECTS and check if they has been changed properly"
417 EXPECTEDOBJECTSMODE=`do_facet $SINGLEMDS "debugfs -R 'stat OBJECTS' $MDSDEV 2> /dev/null" | grep 'Mode: ' | sed -e "s/.*Mode: *//" -e "s/ *Flags:.*//"`
419 if [ "$EXPECTEDOBJECTSMODE" = "0777" ]; then
420 log "Success:Lustre change the mode of OBJECTS correctly"
422 error "Lustre does not change mode of OBJECTS properly"
425 run_test 16 "verify that lustre will correct the mode of OBJECTS"
428 local dev=${SINGLEMDS}_dev
431 if [ ! -e "$MDSDEV" ]; then
432 echo "no $MDSDEV existing, so mount Lustre to create one"
434 check_mount || return 41
438 echo "Remove mds config log"
439 do_facet $SINGLEMDS "debugfs -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MDSDEV || return \$?" || return $?
442 start_mds && return 42
445 run_test 17 "Verify failed mds_postsetup won't fail assertion (2936) (should return errs)"
448 [ "$FSTYPE" != "ldiskfs" ] && skip "not needed for FSTYPE=$FSTYPE" && return
450 local dev=${SINGLEMDS}_dev
456 # check if current MDSSIZE is large enough
457 [ $MDSSIZE -ge $MIN ] && OK=1 && myMDSSIZE=$MDSSIZE && \
458 log "use MDSSIZE=$MDSSIZE"
460 # check if the global config has a large enough MDSSIZE
461 [ -z "$OK" -a ! -z "$STORED_MDSSIZE" ] && [ $STORED_MDSSIZE -ge $MIN ] && \
462 OK=1 && myMDSSIZE=$STORED_MDSSIZE && \
463 log "use STORED_MDSSIZE=$STORED_MDSSIZE"
465 # check if the block device is large enough
466 [ -z "$OK" -a -b $MDSDEV ] && \
467 [ "$(dd if=$MDSDEV of=/dev/null bs=1k count=1 skip=$MIN 2>&1 |
468 awk '($3 == "in") { print $1 }')" = "1+0" ] && OK=1 && \
469 myMDSSIZE=$MIN && log "use device $MDSDEV with MIN=$MIN"
471 # check if a loopback device has enough space for fs metadata (5%)
472 [ -z "$OK" ] && [ -f $MDSDEV -o ! -e $MDSDEV ] &&
473 SPACE=$(df -P $(dirname $MDSDEV) |
474 awk '($1 != "Filesystem") {print $4}') &&
475 [ $SPACE -gt $((MIN / 20)) ] && OK=1 && myMDSSIZE=$MIN && \
476 log "use file $MDSDEV with MIN=$MIN"
478 [ -z "$OK" ] && skip "$MDSDEV too small for ${MIN}kB MDS" && return
481 echo "mount mds with large journal..."
482 local OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS
484 MDS_MKFS_OPTS="--mgs --mdt --fsname=$FSNAME --device-size=$myMDSSIZE --param sys.timeout=$TIMEOUT $MDSOPT"
487 echo "mount lustre system..."
489 check_mount || return 41
491 echo "check journal size..."
492 local FOUNDSIZE=`do_facet mds "debugfs -c -R 'stat <8>' $MDSDEV" | awk '/Size: / { print $NF; exit;}'`
493 if [ $FOUNDSIZE -gt $((32 * 1024 * 1024)) ]; then
494 log "Success: mkfs creates large journals. Size: $((FOUNDSIZE >> 20))M"
496 error "expected journal size > 32M, found $((FOUNDSIZE >> 20))M"
501 MDS_MKFS_OPTS=$OLD_MDS_MKFS_OPTS
504 run_test 18 "check mkfs creates large journals"
507 start_mds || return 1
508 stop_mds -f || return 2
510 run_test 19a "start/stop MDS without OSTs"
513 start_ost || return 1
514 stop_ost -f || return 2
516 run_test 19b "start/stop OSTs without MDS"
519 # first format the ost/mdt
523 check_mount || return 43
525 remount_client ro $MOUNT || return 44
526 touch $DIR/$tfile && echo "$DIR/$tfile created incorrectly" && return 45
527 [ -e $DIR/$tfile ] && echo "$DIR/$tfile exists incorrectly" && return 46
528 remount_client rw $MOUNT || return 47
530 [ ! -f $DIR/$tfile ] && echo "$DIR/$tfile missing" && return 48
531 MCNT=`grep -c $MOUNT /etc/mtab`
532 [ "$MCNT" -ne 1 ] && echo "$MOUNT in /etc/mtab $MCNT times" && return 49
537 run_test 20 "remount ro,rw mounts work and doesn't break /etc/mtab"
545 run_test 21a "start mds before ost, stop ost first"
553 run_test 21b "start ost before mds, stop mds first"
563 run_test 21c "start mds between two osts, stop mds last"
566 #reformat to remove all logs
569 echo Client mount before any osts are in the logs
571 check_mount && return 41
574 echo Client mount with ost in logs, but none running
578 # check_mount will block trying to contact ost
582 echo Client mount with a running ost
585 # if gss enabled, wait full time to let connection from
586 # mds to ost be established, due to the mismatch between
587 # initial connect timeout and gss context negotiation timeout.
588 # This perhaps could be remove after AT landed.
589 echo "sleep $((TIMEOUT + TIMEOUT + TIMEOUT))s"
590 sleep $((TIMEOUT + TIMEOUT + TIMEOUT))
593 check_mount || return 41
598 run_test 22 "start a client before osts (should return errs)"
600 test_23a() { # was test_23
604 # force down client so that recovering mds waits for reconnect
605 local running=$(grep -c $MOUNT /proc/mounts) || true
606 if [ $running -ne 0 ]; then
607 echo "Stopping client $MOUNT (opts: -f)"
611 # enter recovery on mds
613 # try to start a new client
614 mount_client $MOUNT &
616 MOUNT_PID=$(ps -ef | grep "t lustre" | grep -v grep | awk '{print $2}')
617 MOUNT_LUSTRE_PID=`ps -ef | grep mount.lustre | grep -v grep | awk '{print $2}'`
618 echo mount pid is ${MOUNT_PID}, mount.lustre pid is ${MOUNT_LUSTRE_PID}
620 ps --ppid $MOUNT_LUSTRE_PID
621 # FIXME why o why can't I kill these? Manual "ctrl-c" works...
622 kill -TERM $MOUNT_LUSTRE_PID
623 echo "waiting for mount to finish"
625 # we can not wait $MOUNT_PID because it is not a child of this shell
631 while [ "$WAIT" -lt "$MAX_WAIT" ]; do
633 PID1=$(ps -ef | awk '{print $2}' | grep -w $MOUNT_PID)
634 PID2=$(ps -ef | awk '{print $2}' | grep -w $MOUNT_LUSTRE_PID)
637 [ -z "$PID1" -a -z "$PID2" ] && break
638 echo "waiting for mount to finish ... "
639 WAIT=$(( WAIT + sleep))
641 [ "$WAIT" -eq "$MAX_WAIT" ] && error "MOUNT_PID $MOUNT_PID and \
642 MOUNT__LUSTRE_PID $MOUNT__LUSTRE_PID still not killed in $WAIT secs"
647 run_test 23a "interrupt client during recovery mount delay"
652 test_23b() { # was test_23
655 # Simulate -EINTR during mount OBD_FAIL_LDLM_CLOSE_THREAD
656 lctl set_param fail_loc=0x80000313
660 run_test 23b "Simulate -EINTR during mount"
662 fs2mds_HOST=$mds_HOST
663 fs2ost_HOST=$ost_HOST
667 echo "umount $MOUNT2 ..."
668 umount $MOUNT2 || true
669 echo "stopping fs2mds ..."
670 stop fs2mds -f || true
671 echo "stopping fs2ost ..."
672 stop fs2ost -f || true
679 [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST
680 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then
681 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
682 skip "mixed loopback and real device not working" && return
685 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
686 local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
688 # test 8-char fsname as well
689 local FSNAME2=test1234
690 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --nomgs --mgsnode=$MGSNID --reformat $fs2mdsdev || exit 10
692 add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2ostdev || exit 10
695 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT
696 start fs2ost $fs2ostdev $OST_MOUNT_OPTS
698 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1
700 check_mount || return 2
701 # files written on 1 should not show up on 2
702 cp /etc/passwd $DIR/$tfile
704 [ -e $MOUNT2/$tfile ] && error "File bleed" && return 7
707 cp /etc/passwd $MOUNT2/b || return 3
708 rm $MOUNT2/b || return 4
709 # 2 is actually mounted
710 grep $MOUNT2' ' /proc/mounts > /dev/null || return 5
712 facet_failover fs2mds
713 facet_failover fs2ost
716 # the MDS must remain up until last MDT
718 MDS=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1)
719 [ -z "$MDS" ] && error "No MDT" && return 8
721 cleanup_nocli || return 6
723 run_test 24a "Multiple MDTs on a single node"
726 if [ -z "$fs2mds_DEV" ]; then
727 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
728 skip "mixed loopback and real device not working" && return
731 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
733 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME}2 --mgs --reformat $fs2mdsdev || exit 10
735 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && return 2
738 run_test 24b "Multiple MGSs on a single node (should return err)"
742 check_mount || return 2
743 local MODULES=$($LCTL modules | awk '{ print $2 }')
744 rmmod $MODULES 2>/dev/null || true
747 run_test 25 "Verify modules are referenced"
751 # we need modules before mount for sysctl, so make sure...
752 do_facet $SINGLEMDS "lsmod | grep -q lustre || modprobe lustre"
753 #define OBD_FAIL_MDS_FS_SETUP 0x135
754 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000135"
755 start_mds && echo MDS started && return 1
756 lctl get_param -n devices
757 DEVS=$(lctl get_param -n devices | wc -l)
758 [ $DEVS -gt 0 ] && return 2
759 unload_modules || return 203
761 run_test 26 "MDT startup failure cleans LOV (should return errs)"
767 local ORIG=$(do_facet $myfacet "$TEST")
768 if [ $# -gt 3 ]; then
774 echo "Setting $PARAM from $ORIG to $FINAL"
775 do_facet $SINGLEMDS "$LCTL conf_param $PARAM=$FINAL" || error conf_param failed
781 RESULT=$(do_facet $myfacet "$TEST")
782 if [ $RESULT -eq $FINAL ]; then
783 echo "Updated config after $WAIT sec (got $RESULT)"
787 if [ $WAIT -eq $MAX ]; then
788 echo "Config update not seen: wanted $FINAL got $RESULT"
791 echo "Waiting $(($MAX - $WAIT)) secs for config update"
796 start_ost || return 1
797 start_mds || return 2
798 echo "Requeue thread should have started: "
799 ps -e | grep ll_cfg_requeue
800 set_and_check ost1 "lctl get_param -n obdfilter.$FSNAME-OST0000.client_cache_seconds" "$FSNAME-OST0000.ost.client_cache_seconds" || return 3
803 run_test 27a "Reacquire MGS lock if OST started first"
808 local device=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }')
810 facet_failover $SINGLEMDS
811 set_and_check $SINGLEMDS "lctl get_param -n mdt.$device.identity_acquire_expire" "$device.mdt.identity_acquire_expire" || return 3
812 set_and_check client "lctl get_param -n mdc.$device-mdc-*.max_rpcs_in_flight" "$device.mdc.max_rpcs_in_flight" || return 4
816 run_test 27b "Reacquire MGS lock after failover"
820 TEST="lctl get_param -n llite.$FSNAME-*.max_read_ahead_whole_mb"
821 PARAM="$FSNAME.llite.max_read_ahead_whole_mb"
824 set_and_check client "$TEST" "$PARAM" $FINAL || return 3
825 FINAL=$(($FINAL + 1))
826 set_and_check client "$TEST" "$PARAM" $FINAL || return 4
827 umount_client $MOUNT || return 200
830 if [ $RESULT -ne $FINAL ]; then
831 echo "New config not seen: wanted $FINAL got $RESULT"
834 echo "New config success: got $RESULT"
836 set_and_check client "$TEST" "$PARAM" $ORIG || return 5
839 run_test 28 "permanent parameter setting"
842 [ "$OSTCOUNT" -lt "2" ] && skip "$OSTCOUNT < 2, skipping" && return
843 setup > /dev/null 2>&1
847 local PARAM="$FSNAME-OST0001.osc.active"
848 local PROC_ACT="osc.$FSNAME-OST0001-osc-[^M]*.active"
849 local PROC_UUID="osc.$FSNAME-OST0001-osc-[^M]*.ost_server_uuid"
851 ACTV=$(lctl get_param -n $PROC_ACT)
853 set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $DEAC || return 2
854 # also check ost_server_uuid status
855 RESULT=$(lctl get_param -n $PROC_UUID | grep DEACTIV)
856 if [ -z "$RESULT" ]; then
857 echo "Live client not deactivated: $(lctl get_param -n $PROC_UUID)"
860 echo "Live client success: got $RESULT"
864 local MPROC="osc.$FSNAME-OST0001-osc-[M]*.active"
869 RESULT=`do_facet $SINGLEMDS " lctl get_param -n $MPROC"`
870 [ ${PIPESTATUS[0]} = 0 ] || error "Can't read $MPROC"
871 if [ $RESULT -eq $DEAC ]; then
872 echo "MDT deactivated also after $WAIT sec (got $RESULT)"
876 if [ $WAIT -eq $MAX ]; then
877 echo "MDT not deactivated: wanted $DEAC got $RESULT"
880 echo "Waiting $(($MAX - $WAIT)) secs for MDT deactivated"
883 # test new client starts deactivated
884 umount_client $MOUNT || return 200
886 RESULT=$(lctl get_param -n $PROC_UUID | grep DEACTIV | grep NEW)
887 if [ -z "$RESULT" ]; then
888 echo "New client not deactivated from start: $(lctl get_param -n $PROC_UUID)"
891 echo "New client success: got $RESULT"
894 # make sure it reactivates
895 set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $ACTV || return 6
900 #writeconf to remove all ost2 traces for subsequent tests
906 run_test 29 "permanently remove an OST"
911 TEST="lctl get_param -n llite.$FSNAME-*.max_read_ahead_whole_mb"
913 LIST=(1 2 3 4 5 4 3 2 1 2 3 4 5 4 3 2 1 2 3 4 5)
914 for i in ${LIST[@]}; do
915 set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $i || return 3
917 # make sure client restart still works
919 mount_client $MOUNT || return 4
920 [ "$($TEST)" -ne "$i" ] && return 5
921 set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $ORIG || return 6
924 run_test 30 "Big config llog"
926 test_31() { # bug 10734
927 # ipaddr must not exist
928 mount -t lustre 4.3.2.1@tcp:/lustre $MOUNT || true
931 run_test 31 "Connect to non-existent node (shouldn't crash)"
934 # XXX - make this test verify 1.8 -> 2.0 upgrade is working
935 # XXX - make this run on client-only systems with real hardware on
937 # there appears to be a lot of assumption here about loopback
939 # or maybe this test is just totally useless on a client-only system
940 [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
941 [ "$mds_HOST" = "`hostname`" ] || { skip "remote MDS" && return 0; }
942 [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
943 { skip "remote OST" && return 0; }
945 [ -z "$TUNEFS" ] && skip "No tunefs" && return
946 local DISK1_8=$LUSTRE/tests/disk1_8.tgz
947 [ ! -r $DISK1_8 ] && skip "Cannot find $DISK1_8" && return 0
950 tar xjvf $DISK1_8 -C $TMP/$tdir || \
951 { skip "Cannot untar $DISK1_8" && return 0; }
954 lctl set_param debug=$PTLDEBUG
956 $TUNEFS $tmpdir/mds || error "tunefs failed"
957 # nids are wrong, so client wont work, but server should start
958 start mds $tmpdir/mds "-o loop,exclude=lustre-OST0000" || return 3
959 local UUID=$(lctl get_param -n mdt.lustre-MDT0000.uuid)
961 [ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID"
963 $TUNEFS --mgsnode=`hostname` $tmpdir/ost1 || error "tunefs failed"
964 start ost1 $tmpdir/ost1 "-o loop" || return 5
965 UUID=$(lctl get_param -n obdfilter.lustre-OST0000.uuid)
967 [ "$UUID" == "ost1_UUID" ] || error "UUID is wrong: $UUID"
969 local NID=$($LCTL list_nids | head -1)
971 echo "OSC changes should return err:"
972 $LCTL conf_param lustre-OST0000.osc.max_dirty_mb=15 && return 7
973 $LCTL conf_param lustre-OST0000.failover.node=$NID && return 8
975 echo "MDC changes should succeed:"
976 $LCTL conf_param lustre-MDT0000.mdc.max_rpcs_in_flight=9 || return 9
977 $LCTL conf_param lustre-MDT0000.failover.node=$NID || return 10
980 # With a new good MDT failover nid, we should be able to mount a client
981 # (but it cant talk to OST)
982 local OLDMOUNTOPT=$MOUNTOPT
983 MOUNTOPT="exclude=lustre-OST0000"
985 MOUNTOPT=$OLDMOUNTOPT
986 set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "lustre-MDT0000.mdc.max_rpcs_in_flight" ||
989 zconf_umount `hostname` $MOUNT -f
993 # mount a second time to make sure we didnt leave upgrade flag on
995 $TUNEFS --dryrun $tmpdir/mds || error "tunefs failed"
997 start mds $tmpdir/mds "-o loop,exclude=lustre-OST0000" || return 12
1000 rm -rf $tmpdir || true # true is only for TMP on NFS
1002 run_test 32a "Upgrade from 1.8 (not live)"
1005 # XXX - make this test verify 1.8 -> 2.0 upgrade is working
1006 # XXX - make this run on client-only systems with real hardware on
1008 # there appears to be a lot of assumption here about loopback
1010 # or maybe this test is just totally useless on a client-only system
1011 [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
1012 [ "$mds_HOST" = "`hostname`" ] || { skip "remote MDS" && return 0; }
1013 [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
1014 { skip "remote OST" && return 0; }
1016 [ -z "$TUNEFS" ] && skip "No tunefs" && return
1017 local DISK1_8=$LUSTRE/tests/disk1_8.tgz
1018 [ ! -r $DISK1_8 ] && skip "Cannot find $DISK1_8" && return 0
1020 tar xjvf $DISK1_8 -C $TMP/$tdir || \
1021 { skip "Cannot untar $DISK1_8" && return ; }
1024 lctl set_param debug=$PTLDEBUG
1027 # writeconf will cause servers to register with their current nids
1028 $TUNEFS --writeconf --fsname=$NEWNAME $tmpdir/mds || error "tunefs failed"
1029 start mds $tmpdir/mds "-o loop" || return 3
1030 local UUID=$(lctl get_param -n mdt.${NEWNAME}-MDT0000.uuid)
1032 [ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID"
1034 $TUNEFS --mgsnode=`hostname` --fsname=$NEWNAME --writeconf $tmpdir/ost1 || error "tunefs failed"
1035 start ost1 $tmpdir/ost1 "-o loop" || return 5
1036 UUID=$(lctl get_param -n obdfilter.${NEWNAME}-OST0000.uuid)
1038 [ "$UUID" == "ost1_UUID" ] || error "UUID is wrong: $UUID"
1040 echo "OSC changes should succeed:"
1041 $LCTL conf_param ${NEWNAME}-OST0000.osc.max_dirty_mb=15 || return 7
1042 $LCTL conf_param ${NEWNAME}-OST0000.failover.node=$NID || return 8
1044 echo "MDC changes should succeed:"
1045 $LCTL conf_param ${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight=9 || return 9
1048 # MDT and OST should have registered with new nids, so we should have
1049 # a fully-functioning client
1050 echo "Check client and old fs contents"
1055 set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight" || return 11
1056 [ "$(cksum $MOUNT/passwd | cut -d' ' -f 1,2)" == "2479747619 779" ] || return 12
1060 rm -rf $tmpdir || true # true is only for TMP on NFS
1062 run_test 32b "Upgrade from 1.8 with writeconf"
1064 test_33a() { # bug 12333, was test_33
1066 local FSNAME2=test-123
1067 [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST
1069 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then
1070 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
1071 skip "mixed loopback and real device not working" && return
1074 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
1075 local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
1076 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10
1077 add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --index=8191 --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10
1079 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT
1080 start fs2ost $fs2ostdev $OST_MOUNT_OPTS
1081 do_facet $SINGLEMDS "$LCTL conf_param $FSNAME2.sys.timeout=200" || rc=1
1083 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || rc=2
1086 cp /etc/hosts $MOUNT2/ || rc=3
1087 $LFS getstripe $MOUNT2/hosts
1092 rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev
1093 cleanup_nocli || rc=6
1096 run_test 33a "Mount ost with a large index number"
1098 test_33b() { # was test_34
1101 do_facet client dd if=/dev/zero of=$MOUNT/24 bs=1024k count=1
1102 # Drop lock cancelation reply during umount
1103 #define OBD_FAIL_LDLM_CANCEL 0x304
1104 do_facet client lctl set_param fail_loc=0x80000304
1105 #lctl set_param debug=-1
1106 umount_client $MOUNT
1109 run_test 33b "Drop cancel during umount"
1113 do_facet client "sh runmultiop_bg_pause $DIR/file O_c"
1114 manual_umount_client
1116 do_facet client killall -USR1 multiop
1117 if [ $rc -eq 0 ]; then
1118 error "umount not fail!"
1123 run_test 34a "umount with opened file should be fail"
1128 touch $DIR/$tfile || return 1
1129 stop_mds --force || return 2
1131 manual_umount_client --force
1133 if [ $rc -ne 0 ]; then
1134 error "mtab after failed umount - rc $rc"
1140 run_test 34b "force umount with failed mds should be normal"
1144 touch $DIR/$tfile || return 1
1145 stop_ost --force || return 2
1147 manual_umount_client --force
1149 if [ $rc -ne 0 ]; then
1150 error "mtab after failed umount - rc $rc"
1156 run_test 34c "force umount with failed ost should be normal"
1158 test_35() { # bug 12459
1161 DBG_SAVE="`lctl get_param -n debug`"
1162 lctl set_param debug="ha"
1164 log "Set up a fake failnode for the MDS"
1166 local device=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1)
1167 do_facet $SINGLEMDS $LCTL conf_param ${device}.failover.node=$FAKENID || return 4
1169 log "Wait for RECONNECT_INTERVAL seconds (10s)"
1172 MSG="conf-sanity.sh test_35 `date +%F%kh%Mm%Ss`"
1175 log "Stopping the MDT:"
1176 stop_mds || return 5
1178 df $MOUNT > /dev/null 2>&1 &
1180 log "Restarting the MDT:"
1181 start_mds || return 6
1182 log "Wait for df ($DFPID) ... "
1185 lctl set_param debug="$DBG_SAVE"
1187 # retrieve from the log the first server that the client tried to
1188 # contact after the connection loss
1189 $LCTL dk $TMP/lustre-log-$TESTNAME.log
1190 NEXTCONN=`awk "/${MSG}/ {start = 1;}
1191 /import_select_connection.$device-mdc.* using connection/ {
1193 if (\\\$NF ~ /$FAKENID/)
1199 }" $TMP/lustre-log-$TESTNAME.log`
1200 [ "$NEXTCONN" != "0" ] && log "The client didn't try to reconnect to the last active server (tried ${NEXTCONN} instead)" && return 7
1203 run_test 35 "Reconnect to the last active server first"
1207 local FSNAME2=test1234
1208 local fs3ost_HOST=$ost_HOST
1210 [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST && fs3ost_HOST=$ost1_HOST
1213 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" -o -z "$fs3ost_DEV" ]; then
1214 do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \
1215 skip "mixed loopback and real device not working" && return
1217 [ $OSTCOUNT -lt 2 ] && skip "skipping test for single OST" && return
1219 [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \
1220 { skip "remote OST" && return 0; }
1222 local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2}
1223 local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2}
1224 local fs3ostdev=${fs3ost_DEV:-$(ostdevname 2)_2}
1225 add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10
1226 # XXX after we support non 4K disk blocksize, change following --mkfsoptions with
1228 add fs2ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10
1229 add fs3ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=$MGSNID --reformat $fs3ostdev || exit 10
1231 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS
1232 start fs2ost $fs2ostdev $OST_MOUNT_OPTS
1233 start fs3ost $fs3ostdev $OST_MOUNT_OPTS
1235 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1
1237 sleep 5 # until 11778 fixed
1239 dd if=/dev/zero of=$MOUNT2/$tfile bs=1M count=7 || return 2
1241 BKTOTAL=`lctl get_param -n obdfilter.*.kbytestotal | awk 'BEGIN{total=0}; {total+=$1}; END{print total}'`
1242 BKFREE=`lctl get_param -n obdfilter.*.kbytesfree | awk 'BEGIN{free=0}; {free+=$1}; END{print free}'`
1243 BKAVAIL=`lctl get_param -n obdfilter.*.kbytesavail | awk 'BEGIN{avail=0}; {avail+=$1}; END{print avail}'`
1244 STRING=`df -P $MOUNT2 | tail -n 1 | awk '{print $2","$3","$4}'`
1245 DFTOTAL=`echo $STRING | cut -d, -f1`
1246 DFUSED=`echo $STRING | cut -d, -f2`
1247 DFAVAIL=`echo $STRING | cut -d, -f3`
1248 DFFREE=$(($DFTOTAL - $DFUSED))
1250 ALLOWANCE=$((64 * $OSTCOUNT))
1252 if [ $DFTOTAL -lt $(($BKTOTAL - $ALLOWANCE)) ] ||
1253 [ $DFTOTAL -gt $(($BKTOTAL + $ALLOWANCE)) ] ; then
1254 echo "**** FAIL: df total($DFTOTAL) mismatch OST total($BKTOTAL)"
1257 if [ $DFFREE -lt $(($BKFREE - $ALLOWANCE)) ] ||
1258 [ $DFFREE -gt $(($BKFREE + $ALLOWANCE)) ] ; then
1259 echo "**** FAIL: df free($DFFREE) mismatch OST free($BKFREE)"
1262 if [ $DFAVAIL -lt $(($BKAVAIL - $ALLOWANCE)) ] ||
1263 [ $DFAVAIL -gt $(($BKAVAIL + $ALLOWANCE)) ] ; then
1264 echo "**** FAIL: df avail($DFAVAIL) mismatch OST avail($BKAVAIL)"
1269 stop fs3ost -f || return 200
1270 stop fs2ost -f || return 201
1271 stop fs2mds -f || return 202
1272 rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev $fs3ostdev
1273 unload_modules || return 203
1276 run_test 36 "df report consistency on OSTs with different block size"
1279 [ -n "$CLIENTONLY" -o -n "$CLIENTMODSONLY" ] && skip "client only testing" && return 0
1280 LOCAL_MDSDEV="$TMP/mdt.img"
1281 SYM_MDSDEV="$TMP/sym_mdt.img"
1283 echo "MDS : $LOCAL_MDSDEV"
1284 echo "SYMLINK : $SYM_MDSDEV"
1288 mkfs.lustre --reformat --fsname=lustre --mdt --mgs --device-size=9000 $LOCAL_MDSDEV ||
1289 error "mkfs.lustre $LOCAL_MDSDEV failed"
1290 ln -s $LOCAL_MDSDEV $SYM_MDSDEV
1292 echo "mount symlink device - $SYM_MDSDEV"
1294 mount_op=`mount -v -t lustre -o loop $SYM_MDSDEV ${MOUNT%/*}/mds 2>&1 | grep "unable to set tunable"`
1295 umount -d ${MOUNT%/*}/mds
1296 rm -f $LOCAL_MDSDEV $SYM_MDSDEV
1298 if [ -n "$mount_op" ]; then
1299 error "**** FAIL: set tunables failed for symlink device"
1303 run_test 37 "verify set tunables works for symlink device"
1305 test_38() { # bug 14222
1310 FILES=`find $SRC -type f -mtime +1 | head -n $COUNT`
1311 log "copying $(echo $FILES | wc -w) files to $DIR/$tdir"
1313 tar cf - $FILES | tar xf - -C $DIR/$tdir || \
1314 error "copying $SRC to $DIR/$tdir"
1316 umount_client $MOUNT
1318 log "rename lov_objid file on MDS"
1319 rm -f $TMP/lov_objid.orig
1321 local dev=${SINGLEMDS}_dev
1322 local MDSDEV=${!dev}
1323 do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.orig\\\" $MDSDEV"
1324 do_facet $SINGLEMDS "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV"
1326 do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.orig"
1327 # check create in mds_lov_connect
1331 [ $V ] && log "verifying $DIR/$tdir/$f"
1332 diff -q $f $DIR/$tdir/$f || ERROR=y
1334 do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new\\\" $MDSDEV"
1335 do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.new"
1336 [ "$ERROR" = "y" ] && error "old and new files are different after connect" || true
1338 # check it's updates in sync
1339 umount_client $MOUNT
1342 do_facet $SINGLEMDS dd if=/dev/zero of=$TMP/lov_objid.clear bs=4096 count=1
1343 do_facet $SINGLEMDS "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV"
1344 do_facet $SINGLEMDS "debugfs -w -R \\\"write $TMP/lov_objid.clear lov_objid\\\" $MDSDEV "
1349 [ $V ] && log "verifying $DIR/$tdir/$f"
1350 diff -q $f $DIR/$tdir/$f || ERROR=y
1352 do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new1\\\" $MDSDEV"
1353 do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.new1"
1354 umount_client $MOUNT
1356 [ "$ERROR" = "y" ] && error "old and new files are different after sync" || true
1358 log "files compared the same"
1361 run_test 38 "MDS recreates missing lov_objid file from OST data"
1367 perl $SRCDIR/leak_finder.pl $TMP/debug 2>&1 | egrep '*** Leak:' &&
1368 error "memory leak detected" || true
1370 run_test 39 "leak_finder recognizes both LUSTRE and LNET malloc messages"
1372 test_40() { # bug 15759
1374 #define OBD_FAIL_TGT_TOOMANY_THREADS 0x706
1375 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000706"
1379 run_test 40 "race during service thread startup"
1381 test_41() { #bug 14134
1383 local dev=${SINGLEMDS}_dev
1384 local MDSDEV=${!dev}
1386 start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nosvc -n
1387 start ost1 `ostdevname 1` $OST_MOUNT_OPTS
1388 start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nomgs
1390 mount_client $MOUNT || return 1
1393 echo "blah blah" > $MOUNT/$tfile
1396 umount_client $MOUNT
1397 stop ost1 -f || return 201
1398 stop_mds -f || return 202
1399 stop_mds -f || return 203
1400 unload_modules || return 204
1403 run_test 41 "mount mds with --nosvc and --nomgs"
1405 test_42() { #bug 14693
1407 check_mount || return 2
1408 do_facet client lctl conf_param lustre.llite.some_wrong_param=10
1409 umount_client $MOUNT
1410 mount_client $MOUNT || return 1
1414 run_test 42 "invalid config param should not prevent client from mounting"
1416 umount_client $MOUNT
1422 check_mount || return 2
1427 #define OBD_FAIL_PTLRPC_LONG_UNLINK 0x50f
1428 do_facet client "lctl set_param fail_loc=0x50f"
1431 manual_umount_client --force || return 3
1432 do_facet client "lctl set_param fail_loc=0x0"
1434 mount_client $MOUNT || return 4
1438 run_test 45 "long unlink handling in ptlrpcd"
1441 equals_msg `basename $0`: test complete
1442 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true