2 # vim:expandtab:shiftwidth=4:softtabstop=4:tabstop=4:
5 trap 'echo "test-framework exiting on error"' ERR
11 export GMNALNID=${GMNALNID:-/usr/sbin/gmlndnid}
12 export CATASTROPHE=${CATASTROPHE:-/proc/sys/lnet/catastrophe}
14 # eg, assert_env LUSTRE MDSNODES OSTNODES CLIENTS
18 if [ -z "${!name}" ]; then
19 echo "$0: $name must be set"
23 [ $failed ] && exit 1 || true
27 echo "usage: $0 [-r] [-f cfgfile]"
34 export LUSTRE=`absolute_path $LUSTRE`
35 export TESTSUITE=`basename $0 .sh`
36 export LTESTDIR=${LTESTDIR:-$LUSTRE/../ltest}
38 [ -d /r ] && export ROOT=${ROOT:-/r}
39 export TMP=${TMP:-$ROOT/tmp}
41 export PATH=:$PATH:$LUSTRE/utils:$LUSTRE/utils/gss:$LUSTRE/tests
42 export LCTL=${LCTL:-"$LUSTRE/utils/lctl"}
43 export MKFS=${MKFS:-"$LUSTRE/utils/mkfs.lustre"}
44 export CHECKSTAT="${CHECKSTAT:-checkstat} "
45 export FSYTPE=${FSTYPE:-"ldiskfs"}
46 export LPROC=/proc/fs/lustre
47 export LGSSD=${LGSSD:-"$LUSTRE/utils/gss/lgssd"}
48 export LSVCGSSD=${LSVCGSSD:-"$LUSTRE/utils/gss/lsvcgssd"}
49 export KRB5DIR=${KRB5DIR:-"/usr/kerberos"}
51 if [ "$ACCEPTOR_PORT" ]; then
52 export PORT_OPT="--port $ACCEPTOR_PORT"
57 echo "Using GSS/krb5 ptlrpc security flavor"
62 # Paths on remote nodes, if different
63 export RLUSTRE=${RLUSTRE:-$LUSTRE}
64 export RPWD=${RPWD:-$PWD}
68 while getopts "rvf:" opt $*; do
71 r) REFORMAT=--reformat;;
85 BASE=`basename $module $EXT`
86 lsmod | grep -q ${BASE} || \
87 if [ -f ${LUSTRE}/${module}${EXT} ]; then
88 insmod ${LUSTRE}/${module}${EXT} $@
90 # must be testing a "make install" or "rpm" installation
91 # note failed to load ptlrpc_gss is considered not fatal
92 if [ "$BASE" == "ptlrpc_gss" ]; then
93 modprobe $BASE $@ || echo "gss/krb5 is not supported"
101 if [ -n "$MODPROBE" ]; then
105 if [ "$HAVE_MODULES" = true ]; then
111 echo Loading modules from $LUSTRE
112 load_module ../lnet/libcfs/libcfs
113 # note that insmod will ignore anything in modprobe.conf
114 load_module ../lnet/lnet/lnet $LNETOPTS
115 LNETLND=${LNETLND:-"socklnd/ksocklnd"}
116 load_module ../lnet/klnds/$LNETLND
117 load_module lvfs/lvfs
118 load_module obdclass/obdclass
119 load_module ptlrpc/ptlrpc
120 load_module ptlrpc/gss/ptlrpc_gss
124 load_module quota/lquota
132 load_module ldiskfs/ldiskfs
133 load_module lvfs/fsfilt_ldiskfs
136 load_module obdfilter/obdfilter
137 load_module llite/lustre
140 load_module quota/lquota
141 load_module quota/quotacheck_test
142 load_module quota/quotactl_test
143 load_module obdclass/llog_test
144 load_module obdecho/obdecho
145 load_module ldiskfs/quotafmt_test
148 rm -f $TMP/ogdb-`hostname`
149 $LCTL modules > $TMP/ogdb-`hostname`
150 # 'mount' doesn't look in $PATH, just sbin
151 [ -f $LUSTRE/utils/mount.lustre ] && cp $LUSTRE/utils/mount.lustre /sbin/. || true
158 MODULES=$($LCTL modules | awk '{ print $2 }')
159 while [ -n "$MODULES" ]; do
161 rmmod $MODULES >/dev/null 2>&1 || true
162 MODULES=$($LCTL modules | awk '{ print $2 }')
163 if [ -z "$MODULES" ]; then
167 echo "waiting, $((MAX - WAIT)) secs left"
169 if [ $WAIT -eq $MAX ]; then
170 echo "LNET modules $MODULES will not unload"
178 lsmod | grep lnet > /dev/null && $LCTL dl && $LCTL dk $TMP/debug
179 local MODULES=$($LCTL modules | awk '{ print $2 }')
180 rmmod ldiskfs/quotafmt_test obdclass/llog_test quota/quotactl_test quota/quotacheck_test 2>&1 || true
181 rmmod $MODULES >/dev/null 2>&1 || true
182 # do it again, in case we tried to unload ksocklnd too early
183 MODULES=$($LCTL modules | awk '{ print $2 }')
184 [ -n "$MODULES" ] && rmmod $MODULES >/dev/null || true
185 MODULES=$($LCTL modules | awk '{ print $2 }')
186 if [ -n "$MODULES" ]; then
187 echo "Modules still loaded: "
189 if [ -e $LPROC ]; then
190 echo "Lustre still loaded"
191 cat $LPROC/devices || true
195 echo "Lustre stopped, but LNET is still loaded"
196 wait_for_lnet || return 3
201 LEAK_LUSTRE=$(dmesg | tail -n 30 | grep "obd mem.*leaked" || true)
202 LEAK_PORTALS=$(dmesg | tail -n 20 | grep "Portals memory leaked" || true)
203 if [ "$LEAK_LUSTRE" -o "$LEAK_PORTALS" ]; then
204 echo "$LEAK_LUSTRE" 1>&2
205 echo "$LEAK_PORTALS" 1>&2
206 mv $TMP/debug $TMP/debug-leak.`date +%s`
207 echo "Memory leaks detected"
210 echo "modules unloaded."
214 check_gss_daemon_facet() {
218 num=`do_facet $facet ps -o cmd -C $dname | grep $dname | wc -l`
219 if [ $num -ne 1 ]; then
220 echo "$num instance of $dname on $facet"
229 do_facet $facet "killall -2 $@ 2>/dev/null || true"
232 start_gss_daemons() {
234 do_facet mds "$LSVCGSSD -v"
235 do_facet mds "$LGSSD -v"
237 for num in `seq $OSTCOUNT`; do
238 do_facet ost$num "$LSVCGSSD -v"
241 # FIXME: is "client" the right facet name?
242 do_facet client "$LGSSD -v"
244 # wait daemons entering "stable" status
248 # check daemons are running
250 check_gss_daemon_facet mds lsvcgssd
251 check_gss_daemon_facet mds lgssd
252 for num in `seq $OSTCOUNT`; do
253 check_gss_daemon_facet ost$num lsvcgssd
255 check_gss_daemon_facet client lgssd
259 send_sigint mds lsvcgssd lgssd
260 for num in `seq $OSTCOUNT`; do
261 send_sigint ost$num lsvcgssd
263 send_sigint client lgssd
267 if [ ! -z $SEC ]; then
268 MDS_MOUNT_OPTS=$MDS_MOUNT_OPTS,sec=$SEC
269 OST_MOUNT_OPTS=$OST_MOUNT_OPTS,sec=$SEC
272 if [ ! -z $USING_KRB5 ]; then
278 if [ ! -z $USING_KRB5 ]; then
280 # maybe cleanup credential cache?
285 # start facet device options
291 echo "Starting ${facet}: $@ ${device} ${MOUNT%/*}/${facet}"
292 do_facet ${facet} mkdir -p ${MOUNT%/*}/${facet}
293 do_facet ${facet} mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}
295 if [ $RC -ne 0 ]; then
296 echo mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}
297 echo Start of ${device} on ${facet} failed ${RC}
299 do_facet ${facet} sync
300 label=`do_facet ${facet} "e2label ${device}" | grep -v "CMD: "`
301 [ -z "$label" ] && echo no label for ${device} && exit 1
302 eval export ${facet}_svc=${label}
303 eval export ${facet}_dev=${device}
304 eval export ${facet}_opt=\"$@\"
305 echo Started ${label}
314 HOST=`facet_active_host $facet`
315 [ -z $HOST ] && echo stop: no host for $facet && return 0
317 running=`do_facet ${facet} "grep -c ${MOUNT%/*}/${facet}' ' /proc/mounts" | grep -v "CMD: "`
318 if [ ${running} -ne 0 ]; then
319 echo "Stopping ${MOUNT%/*}/${facet} (opts:$@)"
320 do_facet ${facet} umount -d $@ ${MOUNT%/*}/${facet}
323 [ -e /proc/fs/lustre ] && grep "ST " $LPROC/devices && echo "service didn't stop" && exit 1
332 # Only supply -o to mount if we have options
333 if [ -n "$MOUNTOPT" ]; then
334 OPTIONS="-o $MOUNTOPT"
336 local device=$MGSNID:/$FSNAME
337 if [ -z "$mnt" -o -z "$FSNAME" ]; then
338 echo Bad zconf mount command: opt=$OPTIONS dev=$device mnt=$mnt
342 echo "Starting client: $OPTIONS $device $mnt"
343 do_node $client mkdir -p $mnt
344 do_node $client mount -t lustre $OPTIONS $device $mnt || return 1
346 do_node $client "sysctl -w lnet.debug=$PTLDEBUG; sysctl -w lnet.subsystem_debug=${SUBSYSTEM# }"
347 [ -d /r ] && $LCTL modules > /r/tmp/ogdb-`hostname`
355 local running=`do_node $client "grep -c $mnt' ' /proc/mounts" | grep -v "CMD: "`
356 if [ $running -ne 0 ]; then
357 echo "Stopping client $mnt (opts:$force)"
358 do_node $client umount $force $mnt
364 if [ "$FAILURE_MODE" = HARD ]; then
365 $POWER_DOWN `facet_active_host $facet`
367 elif [ "$FAILURE_MODE" = SOFT ]; then
374 if [ "$FAILURE_MODE" = HARD ]; then
375 $POWER_UP `facet_active_host $facet`
381 # verify that lustre actually cleaned up properly
383 [ -e $CATASTROPHE -a "`cat $CATASTROPHE`" = "1" ] && echo "LBUG" && exit 206
384 BUSY=`dmesg | grep -i destruct || true`
387 [ -e $TMP/debug ] && mv $TMP/debug $TMP/debug-busy.`date +%s`
390 LEAK_LUSTRE=`dmesg | tail -n 30 | grep "obd mem.*leaked" || true`
391 LEAK_PORTALS=`dmesg | tail -n 20 | grep "Portals memory leaked" || true`
392 if [ "$LEAK_LUSTRE" -o "$LEAK_PORTALS" ]; then
393 echo "$0: $LEAK_LUSTRE" 1>&2
394 echo "$0: $LEAK_PORTALS" 1>&2
395 echo "$0: Memory leak(s) detected..." 1>&2
396 mv $TMP/debug $TMP/debug-leak.`date +%s`
400 [ "`lctl dl 2> /dev/null | wc -l`" -gt 0 ] && lctl dl && \
401 echo "$0: lustre didn't clean up..." 1>&2 && return 202 || true
403 if [ "`/sbin/lsmod 2>&1 | egrep 'lnet|libcfs'`" ]; then
404 echo "$0: modules still loaded..." 1>&2
413 check_network "$HOST" 900
414 while ! do_node $HOST "ls -d $LUSTRE " > /dev/null; do sleep 5; done
419 HOST=`facet_active_host $facet`
424 # not every config has many clients
425 if [ ! -z "$CLIENTS" ]; then
426 $PDSH $CLIENTS "df $MOUNT" > /dev/null
431 uname -n >> $MOUNT/recon
432 if [ ! -z "$CLIENTS" ]; then
433 $PDSH $CLIENTS "df $MOUNT; uname -n >> $MOUNT/recon" > /dev/null
435 echo Connected clients:
437 ls -l $MOUNT/recon > /dev/null
443 echo "Failing $facet on node `facet_active_host $facet`"
444 shutdown_facet $facet
448 echo "df pid is $DFPID"
450 TO=`facet_active_host $facet`
451 echo "Failover $facet to $TO"
453 local dev=${facet}_dev
454 local opt=${facet}_opt
455 start $facet ${!dev} ${!opt}
466 local svc=${facet}_svc
467 do_facet $facet $LCTL --device %${!svc} readonly
468 do_facet $facet $LCTL --device %${!svc} notransno
469 do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
470 $LCTL mark "local REPLAY BARRIER on ${!svc}"
473 replay_barrier_nodf() {
474 local facet=$1 echo running=${running}
476 local svc=${facet}_svc
477 echo Replay barrier on ${!svc}
478 do_facet $facet $LCTL --device %${!svc} readonly
479 do_facet $facet $LCTL --device %${!svc} notransno
480 do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
481 $LCTL mark "local REPLAY BARRIER on ${!svc}"
485 UUID=`cat /proc/fs/lustre/mdc/${mds_svc}-mdc-*/uuid`
486 do_facet mds "echo $UUID > /proc/fs/lustre/mds/${mds_svc}/evict_client"
490 UUID=`cat /proc/fs/lustre/osc/${ost1_svc}-osc-*/uuid`
491 do_facet ost1 "echo $UUID > /proc/fs/lustre/obdfilter/${ost1_svc}/evict_client"
496 df $MOUNT || error "post-failover df: $?"
503 local svc=${facet}_svc
504 local dev=${facet}_dev
505 local opt=${facet}_opt
506 start $facet ${!dev} ${!opt}
507 do_facet $facet lctl --device %${!svc} abort_recovery
508 df $MOUNT || echo "first df failed: $?"
510 df $MOUNT || error "post-failover df: $?"
514 echo There is no lmc. This is mountconf, baby.
519 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
520 ID=`$PDSH $1 $GMNALNID -l | cut -d\ -f2`
526 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
533 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
534 if type __h2elan >/dev/null 2>&1; then
537 ID=`echo $1 | sed 's/[^0-9]*//g'`
545 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
546 ID=`echo $1 | sed 's/[^0-9]*//g'`
554 varname=${facet}_HOST
555 if [ -z "${!varname}" ]; then
556 if [ "${facet:0:3}" == "ost" ]; then
557 eval ${facet}_HOST=${ost_HOST}
565 local activevar=${facet}active
567 if [ -f ./${facet}active ] ; then
568 source ./${facet}active
572 if [ -z "$active" ] ; then
579 facet_active_host() {
581 local active=`facet_active $facet`
582 if [ "$facet" == client ]; then
585 echo `facet_host $active`
591 failover=${facet}failover
592 host=`facet_host $failover`
593 [ -z "$host" ] && return
594 curactive=`facet_active $facet`
595 if [ -z "${curactive}" -o "$curactive" == "$failover" ] ; then
596 eval export ${facet}active=$facet
598 eval export ${facet}active=$failover
600 # save the active host for this facet
601 activevar=${facet}active
602 echo "$activevar=${!activevar}" > ./$activevar
609 if [ "$HOST" = "$(hostname)" ]; then
614 $myPDSH $HOST $LCTL mark "$@" > /dev/null 2>&1 || :
616 $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")"
622 HOST=`facet_active_host $facet`
623 [ -z $HOST ] && echo No host defined for facet ${facet} && exit 1
630 # make sure its not already running
633 do_facet ${facet} $MKFS $*
639 #if $OSTDEVn isn't defined, default is $OSTDEVBASE + num
640 eval DEVPTR=${!DEVNAME:=${OSTDEVBASE}${num}}
648 # make sure we are using the primary server, so test-framework will
649 # be able to clean up properly.
650 activemds=`facet_active mds`
651 if [ $activemds != "mds" ]; then
655 # assume client mount is local
656 grep " $MOUNT " /proc/mounts && zconf_umount `hostname` $MOUNT $*
657 grep " $MOUNT2 " /proc/mounts && zconf_umount `hostname` $MOUNT2 $*
659 for num in `seq $OSTCOUNT`; do
673 # We need ldiskfs here, may as well load them all
675 echo Formatting mds, osts
677 add mds $MDS_MKFS_OPTS --reformat $MDSDEV || exit 10
679 add mds $MDS_MKFS_OPTS --reformat $MDSDEV > /dev/null || exit 10
682 for num in `seq $OSTCOUNT`; do
684 add ost$num $OST_MKFS_OPTS --reformat `ostdevname $num` || exit 10
686 add ost$num $OST_MKFS_OPTS --reformat `ostdevname $num` > /dev/null || exit 10
692 grep " $1 " /proc/mounts || zconf_mount `hostname` $*
699 start mds $MDSDEV $MDS_MOUNT_OPTS
700 for num in `seq $OSTCOUNT`; do
701 DEVNAME=`ostdevname $num`
702 start ost$num $DEVNAME $OST_MOUNT_OPTS
704 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
717 while [ $NETWORK -eq 0 ]; do
718 ping -c 1 -w 3 $1 > /dev/null
719 if [ $? -eq 0 ]; then
723 echo "waiting for $1, $((MAX - WAIT)) secs left"
726 if [ $WAIT -gt $MAX ]; then
727 echo "Network not available"
733 while( !($DSH2 $1 "netstat -tna | grep -q $2") ) ; do
744 # the sed converts spaces to commas, but leaves the last space
745 # alone, so the line doesn't end with a comma.
746 echo "$*" | tr -s " " "\n" | sort -b -u | tr "\n" " " | sed 's/ \([^$]\)/,\1/g'
750 (cd `dirname $1`; echo $PWD/`basename $1`)
753 ##################################
757 # OBD_FAIL_MDS_ALL_REQUEST_NET
759 do_facet mds sysctl -w lustre.fail_loc=0x123
760 do_facet client "$1" || RC=$?
761 do_facet mds sysctl -w lustre.fail_loc=0
766 # OBD_FAIL_MDS_ALL_REPLY_NET
768 do_facet mds sysctl -w lustre.fail_loc=0x122
769 do_facet client "$@" || RC=$?
770 do_facet mds sysctl -w lustre.fail_loc=0
775 # OBD_FAIL_MDS_REINT_NET_REP
777 do_facet mds sysctl -w lustre.fail_loc=0x119
778 do_facet client "$@" || RC=$?
779 do_facet mds sysctl -w lustre.fail_loc=0
784 #define OBD_FAIL_OST_BRW_PAUSE_BULK 0x214
786 do_facet ost1 sysctl -w lustre.fail_loc=0x214
787 do_facet client "$1" || RC=$?
788 do_facet client "sync"
789 do_facet ost1 sysctl -w lustre.fail_loc=0
794 #define OBD_FAIL_LDLM_CANCEL 0x304
796 do_facet client sysctl -w lustre.fail_loc=0x304
797 do_facet client "$@" || RC=$?
798 do_facet client sysctl -w lustre.fail_loc=0
803 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
805 do_facet client sysctl -w lustre.fail_loc=0x305
806 do_facet client "$@" || RC=$?
807 do_facet client sysctl -w lustre.fail_loc=0
815 echo "clearing fail_loc on $facet"
816 do_facet $facet "sysctl -w lustre.fail_loc=0"
820 $LCTL mark "cancel_lru_locks start"
821 for d in $LPROC/ldlm/namespaces/*-$1-*; do
822 if [ -f $d/lru_size ]; then
823 echo clear >> $d/lru_size
826 grep "[0-9]" $LPROC/ldlm/namespaces/*-$1-*/lock_unused_count /dev/null
827 $LCTL mark "cancel_lru_locks stop"
832 for a in /proc/fs/lustre/llite/*/dump_page_cache; do
833 if [ `wc -l $a | awk '{print $1}'` -gt 1 ]; then
834 echo there is still data in page cache $a ?
842 ##################################
845 sysctl -w lustre.fail_loc=0 2> /dev/null || true
846 echo "${TESTSUITE}: **** FAIL:" $@
851 build_test_filter() {
852 [ "$ONLY" ] && log "only running test `echo $ONLY`"
856 [ "$EXCEPT$ALWAYS_EXCEPT" ] && \
857 log "skipping tests: `echo $EXCEPT $ALWAYS_EXCEPT`"
858 for E in $EXCEPT $ALWAYS_EXCEPT; do
859 eval EXCEPT_${E}=true
868 IFS=abcdefghijklmnopqrstuvwxyz _basetest $1
872 export base=`basetest $1`
873 if [ ! -z "$ONLY" ]; then
875 if [ ${!testname}x != x ]; then
880 if [ ${!testname}x != x ]; then
888 if [ ${!testname}x != x ]; then
889 echo "skipping excluded test $1"
892 testname=EXCEPT_$base
893 if [ ${!testname}x != x ]; then
894 echo "skipping excluded test $1 (base $base)"
902 EQUALS="======================================================================"
906 local suffixlen=$((${#EQUALS} - ${#msg}))
907 [ $suffixlen -lt 5 ] && suffixlen=5
908 printf '===== %s %.*s\n' "$msg" $suffixlen $EQUALS
913 lsmod | grep lnet > /dev/null || load_modules
914 $LCTL mark "$*" 2> /dev/null || true
922 FFREE=`cat /proc/fs/lustre/mds/*/filesfree`
923 FTOTAL=`cat /proc/fs/lustre/mds/*/filestotal`
924 [ $FFREE -ge $FTOTAL ] && error "files free $FFREE > total $FTOTAL" || true
933 # Pretty tests run faster.
934 equals_msg $testnum: $message
937 log "== test $testnum: $message ============ `date +%H:%M:%S` ($BEFORE)"
939 test_${testnum} || error "test_$testnum failed with $?"
941 [ -f $CATASTROPHE ] && [ `cat $CATASTROPHE` -ne 0 ] && \
942 error "LBUG/LASSERT detected"
943 pass "($((`date +%s` - $BEFORE))s)"
947 (cd `dirname $1`; echo $PWD/`basename $1`)