2 # vim:expandtab:shiftwidth=4:softtabstop=4:tabstop=4:
4 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:$LUSTRE/tests
42 export LCTL=${LCTL:-"$LUSTRE/utils/lctl"}
43 [ ! -f "$LCTL" ] && export LCTL=$(which lctl)
44 export MKFS=${MKFS:-"$LUSTRE/utils/mkfs.lustre"}
45 [ ! -f "$MKFS" ] && export MKFS=$(which mkfs.lustre)
46 export TUNEFS=${TUNEFS:-"$LUSTRE/utils/tunefs.lustre"}
47 [ ! -f "$TUNEFS" ] && export TUNEFS=$(which tunefs.lustre)
48 export CHECKSTAT="${CHECKSTAT:-checkstat} "
49 export FSYTPE=${FSTYPE:-"ext3"}
50 export NAME=${NAME:-local}
51 export LPROC=/proc/fs/lustre
53 if [ "$ACCEPTOR_PORT" ]; then
54 export PORT_OPT="--port $ACCEPTOR_PORT"
57 # Paths on remote nodes, if different
58 export RLUSTRE=${RLUSTRE:-$LUSTRE}
59 export RPWD=${RPWD:-$PWD}
63 while getopts "rvf:" opt $*; do
66 r) REFORMAT=--reformat;;
77 2.4.*) EXT=".o"; USE_QUOTA=no; FSTYPE=ext3 ;;
78 *) EXT=".ko"; USE_QUOTA=yes; FSTYPE=ldiskfs ;;
84 BASE=`basename $module $EXT`
85 lsmod | grep -q ${BASE} || \
86 if [ -f ${LUSTRE}/${module}${EXT} ]; then
87 insmod ${LUSTRE}/${module}${EXT} $@
89 # must be testing a "make install" or "rpm" installation
95 if [ -n "$MODPROBE" ]; then
99 if [ "$HAVE_MODULES" = true ]; then
105 echo Loading modules from $LUSTRE
106 load_module ../lnet/libcfs/libcfs
107 # note that insmod will ignore anything in modprobe.conf
108 load_module ../lnet/lnet/lnet $LNETOPTS
109 LNETLND=${LNETLND:-"socklnd/ksocklnd"}
110 load_module ../lnet/klnds/$LNETLND
111 load_module lvfs/lvfs
112 load_module obdclass/obdclass
113 load_module ptlrpc/ptlrpc
114 [ "$USE_QUOTA" = "yes" ] && load_module quota/lquota
119 [ "$FSTYPE" = "ldiskfs" ] && load_module ldiskfs/ldiskfs
120 load_module lvfs/fsfilt_$FSTYPE
122 load_module obdfilter/obdfilter
123 load_module llite/lustre
126 rm -f $TMP/ogdb-`hostname`
127 $LCTL modules > $TMP/ogdb-`hostname`
128 # 'mount' doesn't look in $PATH, just sbin
129 [ -f $LUSTRE/utils/mount.lustre ] && cp $LUSTRE/utils/mount.lustre /sbin/. || true
133 if [ `uname -r | cut -c 3` -eq 4 ]; then
141 MODULES=$($LCTL modules | awk '{ print $2 }')
142 while [ -n "$MODULES" ]; do
144 $RMMOD $MODULES >/dev/null 2>&1 || true
145 MODULES=$($LCTL modules | awk '{ print $2 }')
146 if [ -z "$MODULES" ]; then
150 echo "waiting, $((MAX - WAIT)) secs left"
152 if [ $WAIT -eq $MAX ]; then
153 echo "LNET modules $MODULES will not unload"
161 lsmod | grep lnet > /dev/null && $LCTL dl && $LCTL dk $TMP/debug
162 local MODULES=$($LCTL modules | awk '{ print $2 }')
163 $RMMOD $MODULES >/dev/null 2>&1 || true
164 # do it again, in case we tried to unload ksocklnd too early
165 MODULES=$($LCTL modules | awk '{ print $2 }')
166 [ -n "$MODULES" ] && $RMMOD $MODULES >/dev/null || true
167 MODULES=$($LCTL modules | awk '{ print $2 }')
168 if [ -n "$MODULES" ]; then
169 echo "Modules still loaded: "
171 if [ -e $LPROC ]; then
172 echo "Lustre still loaded"
173 cat $LPROC/devices || true
177 echo "Lustre stopped, but LNET is still loaded"
178 wait_for_lnet || return 3
183 LEAK_LUSTRE=$(dmesg | tail -n 30 | grep "obd mem.*leaked" || true)
184 LEAK_PORTALS=$(dmesg | tail -n 20 | grep "Portals memory leaked" || true)
185 if [ "$LEAK_LUSTRE" -o "$LEAK_PORTALS" ]; then
186 echo "$LEAK_LUSTRE" 1>&2
187 echo "$LEAK_PORTALS" 1>&2
188 mv $TMP/debug $TMP/debug-leak.`date +%s` || true
189 echo "Memory leaks detected"
190 [ "$LEAK_LUSTRE" -a $(echo $LEAK_LUSTRE | awk 'leaked=$8 {print leaked % 56}') == 0 ] && echo "ignoring known bug 10818" && return 0
193 echo "modules unloaded."
198 # start facet device options
204 echo "Starting ${facet}: $@ ${device} ${MOUNT%/*}/${facet}"
205 do_facet ${facet} mkdir -p ${MOUNT%/*}/${facet}
206 do_facet ${facet} mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}
208 if [ $RC -ne 0 ]; then
209 echo mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}
210 echo Start of ${device} on ${facet} failed ${RC}
212 do_facet ${facet} sync
213 label=`do_facet ${facet} "e2label ${device}" | grep -v "CMD: "`
214 [ -z "$label" ] && echo no label for ${device} && exit 1
215 eval export ${facet}_svc=${label}
216 eval export ${facet}_dev=${device}
217 eval export ${facet}_opt=\"$@\"
218 echo Started ${label}
227 HOST=`facet_active_host $facet`
228 [ -z $HOST ] && echo stop: no host for $facet && return 0
230 running=`do_facet ${facet} "grep -c ${MOUNT%/*}/${facet}' ' /proc/mounts" | grep -v "CMD: "`
231 if [ ${running} -ne 0 ]; then
232 echo "Stopping ${MOUNT%/*}/${facet} (opts:$@)"
233 do_facet ${facet} umount -d $@ ${MOUNT%/*}/${facet}
236 # umount should block, but we should wait for unrelated obd's
237 # like the MGS or MGC to also stop.
240 # conf-sanity 31 takes a long time cleanup
241 while [ $WAIT -lt 300 ]; do
242 running=$(do_facet ${facet} "[ -e $LPROC ] && grep ST' ' $LPROC/devices" | grep -v "CMD: ") || true
243 if [ -z "${running}" ]; then
246 echo "waited $WAIT for${running}"
247 if [ $INTERVAL -lt 64 ]; then
248 INTERVAL=$((INTERVAL + INTERVAL))
251 WAIT=$((WAIT + INTERVAL))
253 echo "service didn't stop after $WAIT seconds. Still running:"
262 # Only supply -o to mount if we have options
263 if [ -n "$MOUNTOPT" ]; then
264 OPTIONS="-o $MOUNTOPT"
266 local device=$MGSNID:/$FSNAME
267 if [ -z "$mnt" -o -z "$FSNAME" ]; then
268 echo Bad zconf mount command: opt=$OPTIONS dev=$device mnt=$mnt
272 echo "Starting client: $OPTIONS $device $mnt"
273 do_node $client mkdir -p $mnt
274 do_node $client mount -t lustre $OPTIONS $device $mnt || return 1
276 do_node $client "sysctl -w lnet.debug=$PTLDEBUG; sysctl -w lnet.subsystem_debug=${SUBSYSTEM# }"
277 [ -d /r ] && $LCTL modules > /r/tmp/ogdb-`hostname`
285 local running=`do_node $client "grep -c $mnt' ' /proc/mounts" | grep -v "CMD: "`
286 if [ $running -ne 0 ]; then
287 echo "Stopping client $mnt (opts:$force)"
288 do_node $client umount $force $mnt
294 if [ "$FAILURE_MODE" = HARD ]; then
295 $POWER_DOWN `facet_active_host $facet`
297 elif [ "$FAILURE_MODE" = SOFT ]; then
304 if [ "$FAILURE_MODE" = HARD ]; then
305 $POWER_UP `facet_active_host $facet`
311 # verify that lustre actually cleaned up properly
313 [ -f $CATASTROPHE ] && [ `cat $CATASTROPHE` -ne 0 ] && \
314 error "LBUG/LASSERT detected"
315 BUSY=`dmesg | grep -i destruct || true`
318 [ -e $TMP/debug ] && mv $TMP/debug $TMP/debug-busy.`date +%s`
321 LEAK_LUSTRE=`dmesg | tail -n 30 | grep "obd mem.*leaked" || true`
322 LEAK_PORTALS=`dmesg | tail -n 20 | grep "Portals memory leaked" || true`
323 if [ "$LEAK_LUSTRE" -o "$LEAK_PORTALS" ]; then
324 echo "$0: $LEAK_LUSTRE" 1>&2
325 echo "$0: $LEAK_PORTALS" 1>&2
326 echo "$0: Memory leak(s) detected..." 1>&2
327 mv $TMP/debug $TMP/debug-leak.`date +%s`
331 [ "`lctl dl 2> /dev/null | wc -l`" -gt 0 ] && lctl dl && \
332 echo "$0: lustre didn't clean up..." 1>&2 && return 202 || true
334 if [ "`/sbin/lsmod 2>&1 | egrep 'lnet|libcfs'`" ]; then
335 echo "$0: modules still loaded..." 1>&2
344 check_network "$HOST" 900
345 while ! do_node $HOST "ls -d $LUSTRE " > /dev/null; do sleep 5; done
350 HOST=`facet_active_host $facet`
355 # not every config has many clients
356 if [ ! -z "$CLIENTS" ]; then
357 $PDSH $CLIENTS "df $MOUNT" > /dev/null
362 uname -n >> $MOUNT/recon
363 if [ ! -z "$CLIENTS" ]; then
364 $PDSH $CLIENTS "df $MOUNT; uname -n >> $MOUNT/recon" > /dev/null
366 echo Connected clients:
368 ls -l $MOUNT/recon > /dev/null
374 echo "Failing $facet on node `facet_active_host $facet`"
375 shutdown_facet $facet
379 echo "df pid is $DFPID"
381 TO=`facet_active_host $facet`
382 echo "Failover $facet to $TO"
384 local dev=${facet}_dev
385 local opt=${facet}_opt
386 start $facet ${!dev} ${!opt}
397 local svc=${facet}_svc
398 do_facet $facet $LCTL --device %${!svc} readonly
399 do_facet $facet $LCTL --device %${!svc} notransno
400 do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
401 $LCTL mark "local REPLAY BARRIER on ${!svc}"
404 replay_barrier_nodf() {
405 local facet=$1 echo running=${running}
407 local svc=${facet}_svc
408 echo Replay barrier on ${!svc}
409 do_facet $facet $LCTL --device %${!svc} readonly
410 do_facet $facet $LCTL --device %${!svc} notransno
411 do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
412 $LCTL mark "local REPLAY BARRIER on ${!svc}"
416 UUID=`cat /proc/fs/lustre/mdc/${mds_svc}-mdc-*/uuid`
417 do_facet mds "echo $UUID > /proc/fs/lustre/mds/${mds_svc}/evict_client"
421 UUID=`cat /proc/fs/lustre/osc/${ost1_svc}-osc-*/uuid`
422 do_facet ost1 "echo $UUID > /proc/fs/lustre/obdfilter/${ost1_svc}/evict_client"
426 facet_failover $* || error "failover: $?"
427 df $MOUNT || error "post-failover df: $?"
434 local svc=${facet}_svc
435 local dev=${facet}_dev
436 local opt=${facet}_opt
437 start $facet ${!dev} ${!opt}
438 do_facet $facet lctl --device %${!svc} abort_recovery
439 df $MOUNT || echo "first df failed: $?"
441 df $MOUNT || error "post-failover df: $?"
445 echo There is no lmc. This is mountconf, baby.
450 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
451 ID=`$PDSH $1 $GMNALNID -l | cut -d\ -f2`
457 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
464 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
465 if type __h2elan >/dev/null 2>&1; then
468 ID=`echo $1 | sed 's/[^0-9]*//g'`
476 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
477 ID=`echo $1 | sed 's/[^0-9]*//g'`
485 varname=${facet}_HOST
486 if [ -z "${!varname}" ]; then
487 if [ "${facet:0:3}" == "ost" ]; then
488 eval ${facet}_HOST=${ost_HOST}
496 local activevar=${facet}active
498 if [ -f ./${facet}active ] ; then
499 source ./${facet}active
503 if [ -z "$active" ] ; then
510 facet_active_host() {
512 local active=`facet_active $facet`
513 if [ "$facet" == client ]; then
516 echo `facet_host $active`
522 failover=${facet}failover
523 host=`facet_host $failover`
524 [ -z "$host" ] && return
525 curactive=`facet_active $facet`
526 if [ -z "${curactive}" -o "$curactive" == "$failover" ] ; then
527 eval export ${facet}active=$facet
529 eval export ${facet}active=$failover
531 # save the active host for this facet
532 activevar=${facet}active
533 echo "$activevar=${!activevar}" > ./$activevar
540 if [ "$HOST" = "$(hostname)" ]; then
545 $myPDSH $HOST $LCTL mark "$@" > /dev/null 2>&1 || :
547 $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")"
553 HOST=`facet_active_host $facet`
554 [ -z $HOST ] && echo No host defined for facet ${facet} && exit 1
561 # make sure its not already running
564 do_facet ${facet} $MKFS $*
570 #if $OSTDEVn isn't defined, default is $OSTDEVBASE + num
571 eval DEVPTR=${!DEVNAME:=${OSTDEVBASE}${num}}
579 # make sure we are using the primary server, so test-framework will
580 # be able to clean up properly.
581 activemds=`facet_active mds`
582 if [ $activemds != "mds" ]; then
586 # assume client mount is local
587 grep " $MOUNT " /proc/mounts && zconf_umount `hostname` $MOUNT $*
588 grep " $MOUNT2 " /proc/mounts && zconf_umount `hostname` $MOUNT2 $*
590 for num in `seq $OSTCOUNT`; do
603 # We need ldiskfs here, may as well load them all
605 echo Formatting mds, osts
607 add mds $MDS_MKFS_OPTS --reformat $MDSDEV || exit 10
609 add mds $MDS_MKFS_OPTS --reformat $MDSDEV > /dev/null || exit 10
612 for num in `seq $OSTCOUNT`; do
614 add ost$num $OST_MKFS_OPTS --reformat `ostdevname $num` || exit 10
616 add ost$num $OST_MKFS_OPTS --reformat `ostdevname $num` > /dev/null || exit 10
622 grep " $1 " /proc/mounts || zconf_mount `hostname` $*
628 start mds $MDSDEV $MDS_MOUNT_OPTS
629 for num in `seq $OSTCOUNT`; do
630 DEVNAME=`ostdevname $num`
631 start ost$num $DEVNAME $OST_MOUNT_OPTS
633 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
646 while [ $NETWORK -eq 0 ]; do
647 ping -c 1 -w 3 $1 > /dev/null
648 if [ $? -eq 0 ]; then
652 echo "waiting for $1, $((MAX - WAIT)) secs left"
655 if [ $WAIT -gt $MAX ]; then
656 echo "Network not available"
662 while( !($DSH2 $1 "netstat -tna | grep -q $2") ) ; do
673 # the sed converts spaces to commas, but leaves the last space
674 # alone, so the line doesn't end with a comma.
675 echo "$*" | tr -s " " "\n" | sort -b -u | tr "\n" " " | sed 's/ \([^$]\)/,\1/g'
679 (cd `dirname $1`; echo $PWD/`basename $1`)
682 ##################################
686 # OBD_FAIL_MDS_ALL_REQUEST_NET
688 do_facet mds sysctl -w lustre.fail_loc=0x123
689 do_facet client "$1" || RC=$?
690 do_facet mds sysctl -w lustre.fail_loc=0
695 # OBD_FAIL_MDS_ALL_REPLY_NET
697 do_facet mds sysctl -w lustre.fail_loc=0x122
698 do_facet client "$@" || RC=$?
699 do_facet mds sysctl -w lustre.fail_loc=0
704 # OBD_FAIL_MDS_REINT_NET_REP
706 do_facet mds sysctl -w lustre.fail_loc=0x119
707 do_facet client "$@" || RC=$?
708 do_facet mds sysctl -w lustre.fail_loc=0
713 #define OBD_FAIL_OST_BRW_PAUSE_BULK 0x214
715 do_facet ost1 sysctl -w lustre.fail_loc=0x214
716 do_facet client "$1" || RC=$?
717 do_facet client "sync"
718 do_facet ost1 sysctl -w lustre.fail_loc=0
723 #define OBD_FAIL_LDLM_CANCEL 0x304
725 do_facet client sysctl -w lustre.fail_loc=0x304
726 do_facet client "$@" || RC=$?
727 do_facet client sysctl -w lustre.fail_loc=0
732 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
734 do_facet client sysctl -w lustre.fail_loc=0x305
735 do_facet client "$@" || RC=$?
736 do_facet client sysctl -w lustre.fail_loc=0
741 #define OBD_FAIL_LDLM_REPLY 0x30c
743 do_facet mds sysctl -w lustre.fail_loc=0x30c
744 do_facet client "$@" || RC=$?
745 do_facet mds sysctl -w lustre.fail_loc=0
753 echo "clearing fail_loc on $facet"
754 do_facet $facet "sysctl -w lustre.fail_loc=0"
758 $LCTL mark "cancel_lru_locks $1 start"
759 for d in `find $LPROC/ldlm/namespaces | egrep -i $1`; do
760 [ -f $d/lru_size ] && echo clear > $d/lru_size
761 [ -f $d/lock_unused_count ] && grep [1-9] $d/lock_unused_count /dev/null
763 $LCTL mark "cancel_lru_locks $1 stop"
768 for a in /proc/fs/lustre/llite/*/dump_page_cache; do
769 if [ `wc -l $a | awk '{print $1}'` -gt 1 ]; then
770 echo there is still data in page cache $a ?
778 ##################################
781 sysctl -w lustre.fail_loc=0 2> /dev/null || true
782 echo "${TESTSUITE}: **** FAIL:" $@
783 log "FAIL: $TESTNAME $@"
787 build_test_filter() {
788 [ "$ONLY" ] && log "only running test `echo $ONLY`"
792 [ "$EXCEPT$ALWAYS_EXCEPT" ] && \
793 log "skipping tests: `echo $EXCEPT $ALWAYS_EXCEPT`"
794 for E in $EXCEPT $ALWAYS_EXCEPT; do
795 eval EXCEPT_${E}=true
804 IFS=abcdefghijklmnopqrstuvwxyz _basetest $1
808 export base=`basetest $1`
809 if [ ! -z "$ONLY" ]; then
811 if [ ${!testname}x != x ]; then
816 if [ ${!testname}x != x ]; then
824 if [ ${!testname}x != x ]; then
825 log "skipping excluded test $1"
828 testname=EXCEPT_$base
829 if [ ${!testname}x != x ]; then
830 log "skipping excluded test $1 (base $base)"
838 EQUALS="======================================================================"
842 local suffixlen=$((${#EQUALS} - ${#msg}))
843 [ $suffixlen -lt 5 ] && suffixlen=5
844 printf '===== %s %.*s\n' "$msg" $suffixlen $EQUALS
849 lsmod | grep lnet > /dev/null || load_modules
850 $LCTL mark "$*" 2> /dev/null || true
858 FFREE=`cat /proc/fs/lustre/mds/*/filesfree`
859 FTOTAL=`cat /proc/fs/lustre/mds/*/filestotal`
860 [ $FFREE -ge $FTOTAL ] && error "files free $FFREE > total $FTOTAL" || true
869 # Pretty tests run faster.
870 equals_msg $testnum: $message
873 log "== test $testnum: $message ============ `date +%H:%M:%S` ($BEFORE)"
875 export TESTNAME=test_$testnum
876 test_${testnum} || error "test_$testnum failed with $?"
878 [ -f $CATASTROPHE ] && [ `cat $CATASTROPHE` -ne 0 ] && \
879 error "LBUG/LASSERT detected"
880 pass "($((`date +%s` - $BEFORE))s)"
885 (cd `dirname $1`; echo $PWD/`basename $1`)