2 # vim:expandtab:shiftwidth=4:softtabstop=4:tabstop=4:
5 trap 'echo "test-framework exiting on error"' ERR
10 export VERBOSE=${VERBOSE:-false}
11 export GMNALNID=${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 obdecho/obdecho
144 rm -f $TMP/ogdb-`hostname`
145 $LCTL modules > $TMP/ogdb-`hostname`
146 # 'mount' doesn't look in $PATH, just sbin
147 [ -f $LUSTRE/utils/mount.lustre ] && cp $LUSTRE/utils/mount.lustre /sbin/. || true
154 MODULES=$($LCTL modules | awk '{ print $2 }')
155 while [ -n "$MODULES" ]; do
157 rmmod $MODULES >/dev/null 2>&1 || true
158 MODULES=$($LCTL modules | awk '{ print $2 }')
159 if [ -z "$MODULES" ]; then
163 echo "waiting, $((MAX - WAIT)) secs left"
165 if [ $WAIT -eq $MAX ]; then
166 echo "LNET modules $MODULES will not unload"
174 lsmod | grep lnet > /dev/null && $LCTL dl && $LCTL dk $TMP/debug
175 local MODULES=$($LCTL modules | awk '{ print $2 }')
176 rmmod ldiskfs/quotafmt_test obdclass/llog_test quota/quotactl_test quota/quotacheck_test 2>&1 || true
177 rmmod $MODULES >/dev/null 2>&1 || true
178 # do it again, in case we tried to unload ksocklnd too early
179 MODULES=$($LCTL modules | awk '{ print $2 }')
180 [ -n "$MODULES" ] && rmmod $MODULES >/dev/null || true
181 MODULES=$($LCTL modules | awk '{ print $2 }')
182 if [ -n "$MODULES" ]; then
183 echo "Modules still loaded: "
185 if [ -e $LPROC ]; then
186 echo "Lustre still loaded"
187 cat $LPROC/devices || true
191 echo "Lustre stopped, but LNET is still loaded"
192 wait_for_lnet || return 3
197 LEAK_LUSTRE=$(dmesg | tail -n 30 | grep "obd mem.*leaked" || true)
198 LEAK_PORTALS=$(dmesg | tail -n 20 | grep "Portals memory leaked" || true)
199 if [ "$LEAK_LUSTRE" -o "$LEAK_PORTALS" ]; then
200 echo "$LEAK_LUSTRE" 1>&2
201 echo "$LEAK_PORTALS" 1>&2
202 mv $TMP/debug $TMP/debug-leak.`date +%s`
203 echo "Memory leaks detected"
206 echo "modules unloaded."
210 check_gss_daemon_facet() {
214 num=`do_facet $facet ps -o cmd -C $dname | grep $dname | wc -l`
215 if [ $num -ne 1 ]; then
216 echo "$num instance of $dname on $facet"
225 do_facet $facet "killall -2 $@ 2>/dev/null || true"
228 start_gss_daemons() {
230 for num in `seq $MDSCOUNT`; do
231 do_facet mds$num "$LSVCGSSD -v"
232 do_facet mds$num "$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 for num in `seq $MDSCOUNT`; do
249 check_gss_daemon_facet mds$num lsvcgssd
250 check_gss_daemon_facet mds$num lgssd
252 for num in `seq $OSTCOUNT`; do
253 check_gss_daemon_facet ost$num lsvcgssd
255 check_gss_daemon_facet client lgssd
259 for num in `seq $MDSCOUNT`; do
260 send_sigint mds$num lsvcgssd lgssd
262 for num in `seq $OSTCOUNT`; do
263 send_sigint ost$num lsvcgssd
265 send_sigint client lgssd
269 if [ ! -z $SEC ]; then
270 MDS_MOUNT_OPTS=$MDS_MOUNT_OPTS,sec=$SEC
271 OST_MOUNT_OPTS=$OST_MOUNT_OPTS,sec=$SEC
274 if [ ! -z $USING_KRB5 ]; then
280 if [ ! -z $USING_KRB5 ]; then
282 # maybe cleanup credential cache?
288 local device=`mdsdevname $num`
289 local label=`do_facet mds$num "e2label ${device}" | grep -v "CMD: "`
295 local device=`ostdevname $num`
296 local label=`do_facet ost$num "e2label ${device}" | grep -v "CMD: "`
301 # start facet device options
307 echo "Starting ${facet}: $@ ${device} ${MOUNT%/*}/${facet}"
308 do_facet ${facet} mkdir -p ${MOUNT%/*}/${facet}
309 do_facet ${facet} mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}
311 if [ $RC -ne 0 ]; then
312 echo mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}
313 echo Start of ${device} on ${facet} failed ${RC}
315 do_facet ${facet} sync
316 label=`do_facet ${facet} "e2label ${device}" | grep -v "CMD: "`
317 [ -z "$label" ] && echo no label for ${device} && exit 1
318 eval export ${facet}_svc=${label}
319 eval export ${facet}_dev=${device}
320 eval export ${facet}_opt=\"$@\"
321 echo Started ${label}
330 HOST=`facet_active_host $facet`
331 [ -z $HOST ] && echo stop: no host for $facet && return 0
333 running=`do_facet ${facet} "grep -c ${MOUNT%/*}/${facet}' ' /proc/mounts" | grep -v "CMD: "`
334 if [ ${running} -ne 0 ]; then
335 echo "Stopping ${MOUNT%/*}/${facet} (opts:$@)"
336 do_facet ${facet} umount -d $@ ${MOUNT%/*}/${facet}
339 [ -e /proc/fs/lustre ] && grep "ST " $LPROC/devices && echo "service didn't stop" && exit 1
348 # Only supply -o to mount if we have options
349 if [ -n "$MOUNTOPT" ]; then
350 OPTIONS="-o $MOUNTOPT"
352 local device=$MGSNID:/$FSNAME
353 if [ -z "$mnt" -o -z "$FSNAME" ]; then
354 echo Bad zconf mount command: opt=$OPTIONS dev=$device mnt=$mnt
358 echo "Starting client: $OPTIONS $device $mnt"
359 do_node $client mkdir -p $mnt
360 do_node $client mount -t lustre $OPTIONS $device $mnt || return 1
362 do_node $client "sysctl -w lnet.debug=$PTLDEBUG; sysctl -w lnet.subsystem_debug=${SUBSYSTEM# }"
363 [ -d /r ] && $LCTL modules > /r/tmp/ogdb-`hostname`
371 local running=`do_node $client "grep -c $mnt' ' /proc/mounts" | grep -v "CMD: "`
372 if [ $running -ne 0 ]; then
373 echo "Stopping client $mnt (opts:$force)"
374 do_node $client umount $force $mnt
380 if [ "$FAILURE_MODE" = HARD ]; then
381 $POWER_DOWN `facet_active_host $facet`
383 elif [ "$FAILURE_MODE" = SOFT ]; then
390 if [ "$FAILURE_MODE" = HARD ]; then
391 $POWER_UP `facet_active_host $facet`
397 # verify that lustre actually cleaned up properly
399 [ -e $CATASTROPHE -a "`cat $CATASTROPHE`" = "1" ] && echo "LBUG" && exit 206
400 BUSY=`dmesg | grep -i destruct || true`
403 [ -e $TMP/debug ] && mv $TMP/debug $TMP/debug-busy.`date +%s`
406 LEAK_LUSTRE=`dmesg | tail -n 30 | grep "obd mem.*leaked" || true`
407 LEAK_PORTALS=`dmesg | tail -n 20 | grep "Portals memory leaked" || true`
408 if [ "$LEAK_LUSTRE" -o "$LEAK_PORTALS" ]; then
409 echo "$0: $LEAK_LUSTRE" 1>&2
410 echo "$0: $LEAK_PORTALS" 1>&2
411 echo "$0: Memory leak(s) detected..." 1>&2
412 mv $TMP/debug $TMP/debug-leak.`date +%s`
416 [ "`lctl dl 2> /dev/null | wc -l`" -gt 0 ] && lctl dl && \
417 echo "$0: lustre didn't clean up..." 1>&2 && return 202 || true
419 if [ "`/sbin/lsmod 2>&1 | egrep 'lnet|libcfs'`" ]; then
420 echo "$0: modules still loaded..." 1>&2
429 check_network "$HOST" 900
430 while ! do_node $HOST "ls -d $LUSTRE " > /dev/null; do sleep 5; done
435 HOST=`facet_active_host $facet`
440 # not every config has many clients
441 if [ ! -z "$CLIENTS" ]; then
442 $PDSH $CLIENTS "df $MOUNT" > /dev/null
447 uname -n >> $MOUNT/recon
448 if [ ! -z "$CLIENTS" ]; then
449 $PDSH $CLIENTS "df $MOUNT; uname -n >> $MOUNT/recon" > /dev/null
451 echo Connected clients:
453 ls -l $MOUNT/recon > /dev/null
459 echo "Failing $facet on node `facet_active_host $facet`"
460 shutdown_facet $facet
464 echo "df pid is $DFPID"
466 TO=`facet_active_host $facet`
467 echo "Failover $facet to $TO"
469 local dev=${facet}_dev
470 local opt=${facet}_opt
471 start $facet ${!dev} ${!opt}
482 local svc=${facet}_svc
483 do_facet $facet $LCTL --device %${!svc} readonly
484 do_facet $facet $LCTL --device %${!svc} notransno
485 do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
486 $LCTL mark "local REPLAY BARRIER on ${!svc}"
489 replay_barrier_nodf() {
490 local facet=$1 echo running=${running}
492 local svc=${facet}_svc
493 echo Replay barrier on ${!svc}
494 do_facet $facet $LCTL --device %${!svc} readonly
495 do_facet $facet $LCTL --device %${!svc} notransno
496 do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
497 $LCTL mark "local REPLAY BARRIER on ${!svc}"
501 UUID=`cat /proc/fs/lustre/mdc/${mds1_svc}-mdc-*/uuid`
502 do_facet mds1 "echo $UUID > /proc/fs/lustre/mdd/${mds1_svc}/evict_client"
506 UUID=`grep ${ost1_svc}-osc- $LPROC/devices | egrep -v 'MDT' | awk '{print $5}'`
507 do_facet ost1 "echo $UUID > /proc/fs/lustre/obdfilter/${ost1_svc}/evict_client"
512 df $MOUNT || error "post-failover df: $?"
519 local svc=${facet}_svc
520 local dev=${facet}_dev
521 local opt=${facet}_opt
522 start $facet ${!dev} ${!opt}
523 do_facet $facet lctl --device %${!svc} abort_recovery
524 df $MOUNT || echo "first df failed: $?"
526 df $MOUNT || error "post-failover df: $?"
530 echo There is no lmc. This is mountconf, baby.
535 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
536 ID=`$PDSH $1 $GMNALNID -l | cut -d\ -f2`
542 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
549 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
550 if type __h2elan >/dev/null 2>&1; then
553 ID=`echo $1 | sed 's/[^0-9]*//g'`
561 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
562 ID=`echo $1 | sed 's/[^0-9]*//g'`
570 varname=${facet}_HOST
571 if [ -z "${!varname}" ]; then
572 if [ "${facet:0:3}" == "ost" ]; then
573 eval ${facet}_HOST=${ost_HOST}
581 local activevar=${facet}active
583 if [ -f ./${facet}active ] ; then
584 source ./${facet}active
588 if [ -z "$active" ] ; then
595 facet_active_host() {
597 local active=`facet_active $facet`
598 if [ "$facet" == client ]; then
601 echo `facet_host $active`
607 failover=${facet}failover
608 host=`facet_host $failover`
609 [ -z "$host" ] && return
610 curactive=`facet_active $facet`
611 if [ -z "${curactive}" -o "$curactive" == "$failover" ] ; then
612 eval export ${facet}active=$facet
614 eval export ${facet}active=$failover
616 # save the active host for this facet
617 activevar=${facet}active
618 echo "$activevar=${!activevar}" > ./$activevar
625 if [ "$HOST" = "$(hostname)" ]; then
630 $myPDSH $HOST $LCTL mark "$@" > /dev/null 2>&1 || :
632 $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")"
638 HOST=`facet_active_host $facet`
639 [ -z $HOST ] && echo No host defined for facet ${facet} && exit 1
646 # make sure its not already running
649 do_facet ${facet} $MKFS $*
655 #if $OSTDEVn isn't defined, default is $OSTDEVBASE + num
656 eval DEVPTR=${!DEVNAME:=${OSTDEVBASE}${num}}
663 #if $MDSDEVn isn't defined, default is $MDSDEVBASE + num
664 eval DEVPTR=${!DEVNAME:=${MDSDEVBASE}${num}}
672 # make sure we are using the primary server, so test-framework will
673 # be able to clean up properly.
674 activemds=`facet_active mds1`
675 if [ $activemds != "mds1" ]; then
679 # assume client mount is local
680 grep " $MOUNT " /proc/mounts && zconf_umount `hostname` $MOUNT $*
681 grep " $MOUNT2 " /proc/mounts && zconf_umount `hostname` $MOUNT2 $*
682 for num in `seq $MDSCOUNT`; do
685 for num in `seq $OSTCOUNT`; do
700 test $nr = 1 && echo -n $MDS_MKFS_OPTS || echo -n $MDSn_MKFS_OPTS
706 # We need ldiskfs here, may as well load them all
708 echo "Formatting mdts, osts"
709 for num in `seq $MDSCOUNT`; do
710 echo "Format mds$num: $(mdsdevname $num)"
712 add mds$num `mdsmkfsopts $num` --reformat `mdsdevname $num` || exit 9
714 add mds$num `mdsmkfsopts $num` --reformat `mdsdevname $num` > /dev/null || exit 9
718 for num in `seq $OSTCOUNT`; do
719 echo "Format ost$num: $(ostdevname $num)"
721 add ost$num $OST_MKFS_OPTS --reformat `ostdevname $num` || exit 10
723 add ost$num $OST_MKFS_OPTS --reformat `ostdevname $num` > /dev/null || exit 10
729 grep " $1 " /proc/mounts || zconf_mount `hostname` $*
735 echo "Setup mdts, osts"
736 for num in `seq $MDSCOUNT`; do
737 DEVNAME=`mdsdevname $num`
738 start mds$num $DEVNAME $MDS_MOUNT_OPTS
740 for num in `seq $OSTCOUNT`; do
741 DEVNAME=`ostdevname $num`
742 start ost$num $DEVNAME $OST_MOUNT_OPTS
744 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
757 while [ $NETWORK -eq 0 ]; do
758 ping -c 1 -w 3 $1 > /dev/null
759 if [ $? -eq 0 ]; then
763 echo "waiting for $1, $((MAX - WAIT)) secs left"
766 if [ $WAIT -gt $MAX ]; then
767 echo "Network not available"
773 while( !($DSH2 $1 "netstat -tna | grep -q $2") ) ; do
784 # the sed converts spaces to commas, but leaves the last space
785 # alone, so the line doesn't end with a comma.
786 echo "$*" | tr -s " " "\n" | sort -b -u | tr "\n" " " | sed 's/ \([^$]\)/,\1/g'
790 (cd `dirname $1`; echo $PWD/`basename $1`)
793 ##################################
797 # OBD_FAIL_MDS_ALL_REQUEST_NET
799 do_facet mds sysctl -w lustre.fail_loc=0x123
800 do_facet client "$1" || RC=$?
801 do_facet mds sysctl -w lustre.fail_loc=0
806 # OBD_FAIL_MDS_ALL_REPLY_NET
808 do_facet mds sysctl -w lustre.fail_loc=0x122
809 do_facet client "$@" || RC=$?
810 do_facet mds sysctl -w lustre.fail_loc=0
815 # OBD_FAIL_MDS_REINT_NET_REP
817 do_facet mds sysctl -w lustre.fail_loc=0x119
818 do_facet client "$@" || RC=$?
819 do_facet mds sysctl -w lustre.fail_loc=0
824 #define OBD_FAIL_OST_BRW_PAUSE_BULK 0x214
826 do_facet ost1 sysctl -w lustre.fail_loc=0x214
827 do_facet client "$1" || RC=$?
828 do_facet client "sync"
829 do_facet ost1 sysctl -w lustre.fail_loc=0
834 #define OBD_FAIL_LDLM_CANCEL 0x304
836 do_facet client sysctl -w lustre.fail_loc=0x304
837 do_facet client "$@" || RC=$?
838 do_facet client sysctl -w lustre.fail_loc=0
843 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
845 do_facet client sysctl -w lustre.fail_loc=0x305
846 do_facet client "$@" || RC=$?
847 do_facet client sysctl -w lustre.fail_loc=0
855 echo "clearing fail_loc on $facet"
856 do_facet $facet "sysctl -w lustre.fail_loc=0"
860 $LCTL mark "cancel_lru_locks start"
861 for d in $LPROC/ldlm/namespaces/*-$1-*; do
862 if [ -f $d/lru_size ]; then
863 echo clear >> $d/lru_size
866 grep "[0-9]" $LPROC/ldlm/namespaces/*-$1-*/lock_unused_count /dev/null
867 $LCTL mark "cancel_lru_locks stop"
872 for a in /proc/fs/lustre/llite/*/dump_page_cache; do
873 if [ `wc -l $a | awk '{print $1}'` -gt 1 ]; then
874 echo there is still data in page cache $a ?
882 ##################################
885 sysctl -w lustre.fail_loc=0 2> /dev/null || true
886 echo "${TESTSUITE}: **** FAIL:" $@
891 build_test_filter() {
892 [ "$ONLY" ] && log "only running test `echo $ONLY`"
896 [ "$EXCEPT$ALWAYS_EXCEPT" ] && \
897 log "skipping tests: `echo $EXCEPT $ALWAYS_EXCEPT`"
898 for E in $EXCEPT $ALWAYS_EXCEPT; do
899 eval EXCEPT_${E}=true
908 IFS=abcdefghijklmnopqrstuvwxyz _basetest $1
912 export base=`basetest $1`
913 if [ ! -z "$ONLY" ]; then
915 if [ ${!testname}x != x ]; then
920 if [ ${!testname}x != x ]; then
928 if [ ${!testname}x != x ]; then
929 echo "skipping excluded test $1"
932 testname=EXCEPT_$base
933 if [ ${!testname}x != x ]; then
934 echo "skipping excluded test $1 (base $base)"
942 EQUALS="======================================================================"
946 local suffixlen=$((${#EQUALS} - ${#msg}))
947 [ $suffixlen -lt 5 ] && suffixlen=5
948 printf '===== %s %.*s\n' "$msg" $suffixlen $EQUALS
953 lsmod | grep lnet > /dev/null || load_modules
954 $LCTL mark "$*" 2> /dev/null || true
962 FFREE=`cat /proc/fs/lustre/mds/*/filesfree`
963 FTOTAL=`cat /proc/fs/lustre/mds/*/filestotal`
964 [ $FFREE -ge $FTOTAL ] && error "files free $FFREE > total $FTOTAL" || true
973 # Pretty tests run faster.
974 equals_msg $testnum: $message
977 log "== test $testnum: $message ============ `date +%H:%M:%S` ($BEFORE)"
979 test_${testnum} || error "test_$testnum failed with $?"
981 [ -f $CATASTROPHE ] && [ `cat $CATASTROPHE` -ne 0 ] && \
982 error "LBUG/LASSERT detected"
983 pass "($((`date +%s` - $BEFORE))s)"
987 (cd `dirname $1`; echo $PWD/`basename $1`)