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 $ACCEPTOR_PORT"
55 if [ "x$SEC" = "xkrb5i" -o "x$SEC" = "xkrb5p" ]; then
59 # Paths on remote nodes, if different
60 export RLUSTRE=${RLUSTRE:-$LUSTRE}
61 export RPWD=${RPWD:-$PWD}
65 while getopts "rvf:" opt $*; do
68 r) REFORMAT=--reformat;;
82 BASE=`basename $module $EXT`
83 lsmod | grep -q ${BASE} || \
84 if [ -f ${LUSTRE}/${module}${EXT} ]; then
85 insmod ${LUSTRE}/${module}${EXT} $@
87 # must be testing a "make install" or "rpm" installation
88 # note failed to load ptlrpc_gss is considered not fatal
89 if [ "$BASE" == "ptlrpc_gss" ]; then
90 modprobe $BASE $@ || echo "gss/krb5 is not supported"
98 if [ -n "$MODPROBE" ]; then
102 if [ "$HAVE_MODULES" = true ]; then
108 echo Loading modules from $LUSTRE
109 load_module ../lnet/libcfs/libcfs
110 # note that insmod will ignore anything in modprobe.conf
111 load_module ../lnet/lnet/lnet $LNETOPTS
112 LNETLND=${LNETLND:-"socklnd/ksocklnd"}
113 echo -1 > /proc/sys/lnet/debug
114 load_module ../lnet/klnds/$LNETLND
115 load_module lvfs/lvfs
116 load_module obdclass/obdclass
117 load_module ptlrpc/ptlrpc
118 load_module ptlrpc/gss/ptlrpc_gss
122 load_module quota/lquota
130 load_module ldiskfs/ldiskfs
131 load_module lvfs/fsfilt_ldiskfs
134 load_module obdfilter/obdfilter
135 load_module llite/lustre
138 load_module quota/lquota
139 load_module quota/quotacheck_test
140 load_module quota/quotactl_test
141 load_module obdclass/llog_test
142 load_module obdecho/obdecho
143 load_module ldiskfs/quotafmt_test
146 rm -f $TMP/ogdb-`hostname`
147 $LCTL modules > $TMP/ogdb-`hostname`
148 # 'mount' doesn't look in $PATH, just sbin
149 [ -f $LUSTRE/utils/mount.lustre ] && cp $LUSTRE/utils/mount.lustre /sbin/. || true
156 MODULES=$($LCTL modules | awk '{ print $2 }')
157 while [ -n "$MODULES" ]; do
159 rmmod $MODULES >/dev/null 2>&1 || true
160 MODULES=$($LCTL modules | awk '{ print $2 }')
161 if [ -z "$MODULES" ]; then
165 echo "waiting, $((MAX - WAIT)) secs left"
167 if [ $WAIT -eq $MAX ]; then
168 echo "LNET modules $MODULES will not unload"
176 lsmod | grep lnet > /dev/null && $LCTL dl && $LCTL dk $TMP/debug
177 local MODULES=$($LCTL modules | awk '{ print $2 }')
178 rmmod ldiskfs/quotafmt_test obdclass/llog_test quota/quotactl_test quota/quotacheck_test 2>&1 || true
179 rmmod $MODULES >/dev/null 2>&1 || true
180 # do it again, in case we tried to unload ksocklnd too early
181 MODULES=$($LCTL modules | awk '{ print $2 }')
182 [ -n "$MODULES" ] && rmmod $MODULES >/dev/null || true
183 MODULES=$($LCTL modules | awk '{ print $2 }')
184 if [ -n "$MODULES" ]; then
185 echo "Modules still loaded: "
187 if [ -e $LPROC ]; then
188 echo "Lustre still loaded"
189 cat $LPROC/devices || true
193 echo "Lustre stopped, but LNET is still loaded"
194 wait_for_lnet || return 3
199 LEAK_LUSTRE=$(dmesg | tail -n 30 | grep "obd mem.*leaked" || true)
200 LEAK_PORTALS=$(dmesg | tail -n 20 | grep "Portals memory leaked" || true)
201 if [ "$LEAK_LUSTRE" -o "$LEAK_PORTALS" ]; then
202 echo "$LEAK_LUSTRE" 1>&2
203 echo "$LEAK_PORTALS" 1>&2
204 mv $TMP/debug $TMP/debug-leak.`date +%s`
205 echo "Memory leaks detected"
208 echo "modules unloaded."
212 check_gss_daemon_facet() {
216 num=`do_facet $facet ps -o cmd -C $dname | grep $dname | wc -l`
217 if [ $num -ne 1 ]; then
218 echo "$num instance of $dname on $facet"
227 do_facet $facet "killall -2 $@ 2>/dev/null || true"
230 start_gss_daemons() {
232 do_facet mds "$LSVCGSSD -v"
233 do_facet mds "$LGSSD -v"
235 for num in `seq $OSTCOUNT`; do
236 do_facet ost$num "$LSVCGSSD -v"
239 # FIXME: is "client" the right facet name?
240 do_facet client "$LGSSD -v"
242 # wait daemons entering "stable" status
246 # check daemons are running
248 check_gss_daemon_facet mds lsvcgssd
249 check_gss_daemon_facet mds lgssd
250 for num in `seq $OSTCOUNT`; do
251 check_gss_daemon_facet ost$num lsvcgssd
253 check_gss_daemon_facet client lgssd
257 send_sigint mds lsvcgssd lgssd
258 for num in `seq $OSTCOUNT`; do
259 send_sigint ost$num lsvcgssd
261 send_sigint client lgssd
265 if [ ! -z $SEC ]; then
266 MDS_MOUNT_OPTS=$MDS_MOUNT_OPTS,sec=$SEC
267 OST_MOUNT_OPTS=$OST_MOUNT_OPTS,sec=$SEC
270 if [ ! -z $USING_KRB5 ]; then
276 if [ ! -z $USING_KRB5 ]; then
278 # maybe cleanup credential cache?
283 # start facet device options
289 echo "Starting ${facet}: $@ ${device} ${MOUNT%/*}/${facet}"
290 do_facet ${facet} mkdir -p ${MOUNT%/*}/${facet}
291 do_facet ${facet} mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}
293 if [ $RC -ne 0 ]; then
294 echo mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}
295 echo Start of ${device} on ${facet} failed ${RC}
297 do_facet ${facet} sync
298 label=`do_facet ${facet} "e2label ${device}" | grep -v "CMD: "`
299 [ -z "$label" ] && echo no label for ${device} && exit 1
300 eval export ${facet}_svc=${label}
301 eval export ${facet}_dev=${device}
302 eval export ${facet}_opt=\"$@\"
303 echo Started ${label}
312 HOST=`facet_active_host $facet`
313 [ -z $HOST ] && echo stop: no host for $facet && return 0
315 running=`do_facet ${facet} "grep -c ${MOUNT%/*}/${facet}' ' /proc/mounts" | grep -v "CMD: "`
316 if [ ${running} -ne 0 ]; then
317 echo "Stopping ${MOUNT%/*}/${facet} (opts:$@)"
318 do_facet ${facet} umount -d $@ ${MOUNT%/*}/${facet}
321 [ -e /proc/fs/lustre ] && grep "ST " $LPROC/devices && echo "service didn't stop" && exit 1
330 # Only supply -o to mount if we have options
331 if [ -n "$MOUNTOPT" ]; then
332 OPTIONS="-o $MOUNTOPT"
334 local device=$MGSNID:/$FSNAME
335 if [ -z "$mnt" -o -z "$FSNAME" ]; then
336 echo Bad zconf mount command: opt=$OPTIONS dev=$device mnt=$mnt
340 echo "Starting client: $OPTIONS $device $mnt"
341 do_node $client mkdir -p $mnt
342 do_node $client mount -t lustre $OPTIONS $device $mnt || return 1
344 do_node $client "sysctl -w lnet.debug=$PTLDEBUG; sysctl -w lnet.subsystem_debug=${SUBSYSTEM# }"
345 [ -d /r ] && $LCTL modules > /r/tmp/ogdb-`hostname`
353 local running=`do_node $client "grep -c $mnt' ' /proc/mounts" | grep -v "CMD: "`
354 if [ $running -ne 0 ]; then
355 echo "Stopping client $mnt (opts:$force)"
356 do_node $client umount $force $mnt
362 if [ "$FAILURE_MODE" = HARD ]; then
363 $POWER_DOWN `facet_active_host $facet`
365 elif [ "$FAILURE_MODE" = SOFT ]; then
372 if [ "$FAILURE_MODE" = HARD ]; then
373 $POWER_UP `facet_active_host $facet`
379 # verify that lustre actually cleaned up properly
381 [ -e $CATASTROPHE -a "`cat $CATASTROPHE`" = "1" ] && echo "LBUG" && exit 206
382 BUSY=`dmesg | grep -i destruct || true`
385 [ -e $TMP/debug ] && mv $TMP/debug $TMP/debug-busy.`date +%s`
388 LEAK_LUSTRE=`dmesg | tail -n 30 | grep "obd mem.*leaked" || true`
389 LEAK_PORTALS=`dmesg | tail -n 20 | grep "Portals memory leaked" || true`
390 if [ "$LEAK_LUSTRE" -o "$LEAK_PORTALS" ]; then
391 echo "$0: $LEAK_LUSTRE" 1>&2
392 echo "$0: $LEAK_PORTALS" 1>&2
393 echo "$0: Memory leak(s) detected..." 1>&2
394 mv $TMP/debug $TMP/debug-leak.`date +%s`
398 [ "`lctl dl 2> /dev/null | wc -l`" -gt 0 ] && lctl dl && \
399 echo "$0: lustre didn't clean up..." 1>&2 && return 202 || true
401 if [ "`/sbin/lsmod 2>&1 | egrep 'lnet|libcfs'`" ]; then
402 echo "$0: modules still loaded..." 1>&2
411 check_network "$HOST" 900
412 while ! do_node $HOST "ls -d $LUSTRE " > /dev/null; do sleep 5; done
417 HOST=`facet_active_host $facet`
422 # not every config has many clients
423 if [ ! -z "$CLIENTS" ]; then
424 $PDSH $CLIENTS "df $MOUNT" > /dev/null
429 uname -n >> $MOUNT/recon
430 if [ ! -z "$CLIENTS" ]; then
431 $PDSH $CLIENTS "df $MOUNT; uname -n >> $MOUNT/recon" > /dev/null
433 echo Connected clients:
435 ls -l $MOUNT/recon > /dev/null
441 echo "Failing $facet on node `facet_active_host $facet`"
442 shutdown_facet $facet
446 echo "df pid is $DFPID"
448 TO=`facet_active_host $facet`
449 echo "Failover $facet to $TO"
451 local dev=${facet}_dev
452 local opt=${facet}_opt
453 start $facet ${!dev} ${!opt}
464 local svc=${facet}_svc
465 do_facet $facet $LCTL --device %${!svc} readonly
466 do_facet $facet $LCTL --device %${!svc} notransno
467 do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
468 $LCTL mark "local REPLAY BARRIER on ${!svc}"
471 replay_barrier_nodf() {
472 local facet=$1 echo running=${running}
474 local svc=${facet}_svc
475 echo Replay barrier on ${!svc}
476 do_facet $facet $LCTL --device %${!svc} readonly
477 do_facet $facet $LCTL --device %${!svc} notransno
478 do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
479 $LCTL mark "local REPLAY BARRIER on ${!svc}"
483 UUID=`cat /proc/fs/lustre/mdc/${mds_svc}-mdc-*/uuid`
484 do_facet mds "echo $UUID > /proc/fs/lustre/mds/${mds_svc}/evict_client"
488 UUID=`cat /proc/fs/lustre/osc/${ost1_svc}-osc-*/uuid`
489 do_facet ost1 "echo $UUID > /proc/fs/lustre/obdfilter/${ost1_svc}/evict_client"
494 df $MOUNT || error "post-failover df: $?"
501 local svc=${facet}_svc
502 local dev=${facet}_dev
503 local opt=${facet}_opt
504 start $facet ${!dev} ${!opt}
505 do_facet $facet lctl --device %${!svc} abort_recovery
506 df $MOUNT || echo "first df failed: $?"
508 df $MOUNT || error "post-failover df: $?"
512 echo There is no lmc. This is mountconf, baby.
517 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
518 ID=`$PDSH $1 $GMNALNID -l | cut -d\ -f2`
524 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
531 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
532 if type __h2elan >/dev/null 2>&1; then
535 ID=`echo $1 | sed 's/[^0-9]*//g'`
543 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
544 ID=`echo $1 | sed 's/[^0-9]*//g'`
552 varname=${facet}_HOST
553 if [ -z "${!varname}" ]; then
554 if [ "${facet:0:3}" == "ost" ]; then
555 eval ${facet}_HOST=${ost_HOST}
563 local activevar=${facet}active
565 if [ -f ./${facet}active ] ; then
566 source ./${facet}active
570 if [ -z "$active" ] ; then
577 facet_active_host() {
579 local active=`facet_active $facet`
580 if [ "$facet" == client ]; then
583 echo `facet_host $active`
589 failover=${facet}failover
590 host=`facet_host $failover`
591 [ -z "$host" ] && return
592 curactive=`facet_active $facet`
593 if [ -z "${curactive}" -o "$curactive" == "$failover" ] ; then
594 eval export ${facet}active=$facet
596 eval export ${facet}active=$failover
598 # save the active host for this facet
599 activevar=${facet}active
600 echo "$activevar=${!activevar}" > ./$activevar
607 if [ "$HOST" = "$(hostname)" ]; then
612 $myPDSH $HOST $LCTL mark "$@" > /dev/null 2>&1 || :
614 $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")"
620 HOST=`facet_active_host $facet`
621 [ -z $HOST ] && echo No host defined for facet ${facet} && exit 1
628 # make sure its not already running
631 do_facet ${facet} $MKFS $*
637 #if $OSTDEVn isn't defined, default is $OSTDEVBASE + num
638 eval DEVPTR=${!DEVNAME:=${OSTDEVBASE}${num}}
646 # make sure we are using the primary server, so test-framework will
647 # be able to clean up properly.
648 activemds=`facet_active mds`
649 if [ $activemds != "mds" ]; then
653 # assume client mount is local
654 grep " $MOUNT " /proc/mounts && zconf_umount `hostname` $MOUNT $*
655 grep " $MOUNT2 " /proc/mounts && zconf_umount `hostname` $MOUNT2 $*
657 for num in `seq $OSTCOUNT`; do
671 # We need ldiskfs here, may as well load them all
673 echo Formatting mds, osts
675 add mds $MDS_MKFS_OPTS --reformat $MDSDEV || exit 10
677 add mds $MDS_MKFS_OPTS --reformat $MDSDEV > /dev/null || exit 10
680 for num in `seq $OSTCOUNT`; do
682 add ost$num $OST_MKFS_OPTS --reformat `ostdevname $num` || exit 10
684 add ost$num $OST_MKFS_OPTS --reformat `ostdevname $num` > /dev/null || exit 10
690 grep " $1 " /proc/mounts || zconf_mount `hostname` $*
697 start mds $MDSDEV $MDS_MOUNT_OPTS
698 for num in `seq $OSTCOUNT`; do
699 DEVNAME=`ostdevname $num`
700 start ost$num $DEVNAME $OST_MOUNT_OPTS
702 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
715 while [ $NETWORK -eq 0 ]; do
716 ping -c 1 -w 3 $1 > /dev/null
717 if [ $? -eq 0 ]; then
721 echo "waiting for $1, $((MAX - WAIT)) secs left"
724 if [ $WAIT -gt $MAX ]; then
725 echo "Network not available"
731 while( !($DSH2 $1 "netstat -tna | grep -q $2") ) ; do
742 # the sed converts spaces to commas, but leaves the last space
743 # alone, so the line doesn't end with a comma.
744 echo "$*" | tr -s " " "\n" | sort -b -u | tr "\n" " " | sed 's/ \([^$]\)/,\1/g'
748 (cd `dirname $1`; echo $PWD/`basename $1`)
751 ##################################
755 # OBD_FAIL_MDS_ALL_REQUEST_NET
757 do_facet mds sysctl -w lustre.fail_loc=0x123
758 do_facet client "$1" || RC=$?
759 do_facet mds sysctl -w lustre.fail_loc=0
764 # OBD_FAIL_MDS_ALL_REPLY_NET
766 do_facet mds sysctl -w lustre.fail_loc=0x122
767 do_facet client "$@" || RC=$?
768 do_facet mds sysctl -w lustre.fail_loc=0
773 # OBD_FAIL_MDS_REINT_NET_REP
775 do_facet mds sysctl -w lustre.fail_loc=0x119
776 do_facet client "$@" || RC=$?
777 do_facet mds sysctl -w lustre.fail_loc=0
782 #define OBD_FAIL_OST_BRW_PAUSE_BULK 0x214
784 do_facet ost1 sysctl -w lustre.fail_loc=0x214
785 do_facet client "$1" || RC=$?
786 do_facet client "sync"
787 do_facet ost1 sysctl -w lustre.fail_loc=0
792 #define OBD_FAIL_LDLM_CANCEL 0x304
794 do_facet client sysctl -w lustre.fail_loc=0x304
795 do_facet client "$@" || RC=$?
796 do_facet client sysctl -w lustre.fail_loc=0
801 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
803 do_facet client sysctl -w lustre.fail_loc=0x305
804 do_facet client "$@" || RC=$?
805 do_facet client sysctl -w lustre.fail_loc=0
813 echo "clearing fail_loc on $facet"
814 do_facet $facet "sysctl -w lustre.fail_loc=0"
818 $LCTL mark "cancel_lru_locks start"
819 for d in $LPROC/ldlm/namespaces/*-$1-*; do
820 if [ -f $d/lru_size ]; then
821 echo clear >> $d/lru_size
824 grep "[0-9]" $LPROC/ldlm/namespaces/*-$1-*/lock_unused_count /dev/null
825 $LCTL mark "cancel_lru_locks stop"
830 for a in /proc/fs/lustre/llite/*/dump_page_cache; do
831 if [ `wc -l $a | awk '{print $1}'` -gt 1 ]; then
832 echo there is still data in page cache $a ?
840 ##################################
843 sysctl -w lustre.fail_loc=0 2> /dev/null || true
844 echo "${TESTSUITE}: **** FAIL:" $@
849 build_test_filter() {
850 [ "$ONLY" ] && log "only running test `echo $ONLY`"
854 [ "$EXCEPT$ALWAYS_EXCEPT" ] && \
855 log "skipping tests: `echo $EXCEPT $ALWAYS_EXCEPT`"
856 for E in $EXCEPT $ALWAYS_EXCEPT; do
857 eval EXCEPT_${E}=true
866 IFS=abcdefghijklmnopqrstuvwxyz _basetest $1
870 export base=`basetest $1`
871 if [ ! -z "$ONLY" ]; then
873 if [ ${!testname}x != x ]; then
878 if [ ${!testname}x != x ]; then
886 if [ ${!testname}x != x ]; then
887 echo "skipping excluded test $1"
890 testname=EXCEPT_$base
891 if [ ${!testname}x != x ]; then
892 echo "skipping excluded test $1 (base $base)"
900 EQUALS="======================================================================"
904 local suffixlen=$((${#EQUALS} - ${#msg}))
905 [ $suffixlen -lt 5 ] && suffixlen=5
906 printf '===== %s %.*s\n' "$msg" $suffixlen $EQUALS
911 lsmod | grep lnet > /dev/null || load_modules
912 $LCTL mark "$*" 2> /dev/null || true
920 FFREE=`cat /proc/fs/lustre/mds/*/filesfree`
921 FTOTAL=`cat /proc/fs/lustre/mds/*/filestotal`
922 [ $FFREE -ge $FTOTAL ] && error "files free $FFREE > total $FTOTAL" || true
931 # Pretty tests run faster.
932 equals_msg $testnum: $message
935 log "== test $testnum: $message ============ `date +%H:%M:%S` ($BEFORE)"
937 test_${testnum} || error "test_$testnum failed with $?"
939 [ -f $CATASTROPHE ] && [ `cat $CATASTROPHE` -ne 0 ] && \
940 error "LBUG/LASSERT detected"
941 pass "($((`date +%s` - $BEFORE))s)"
945 (cd `dirname $1`; echo $PWD/`basename $1`)