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" ] || 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 [ "$FSTYPE" = "ldiskfs" ] && load_module ldiskfs/ldiskfs
120 [ "$FSTYPE" = "ldiskfs2" ] && load_module ldiskfs/ldiskfs2
121 load_module lvfs/fsfilt_$FSTYPE
123 load_module obdfilter/obdfilter
124 load_module llite/lustre
127 rm -f $TMP/ogdb-`hostname`
128 $LCTL modules > $TMP/ogdb-`hostname`
129 # 'mount' doesn't look in $PATH, just sbin
130 [ -f $LUSTRE/utils/mount.lustre ] && cp $LUSTRE/utils/mount.lustre /sbin/. || true
134 if [ `uname -r | cut -c 3` -eq 4 ]; then
142 MODULES=$($LCTL modules | awk '{ print $2 }')
143 while [ -n "$MODULES" ]; do
145 $RMMOD $MODULES >/dev/null 2>&1 || true
146 MODULES=$($LCTL modules | awk '{ print $2 }')
147 if [ -z "$MODULES" ]; then
151 echo "waiting, $((MAX - WAIT)) secs left"
153 if [ $WAIT -eq $MAX ]; then
154 echo "LNET modules $MODULES will not unload"
162 lsmod | grep lnet > /dev/null && $LCTL dl && $LCTL dk $TMP/debug
163 local MODULES=$($LCTL modules | awk '{ print $2 }')
164 $RMMOD $MODULES >/dev/null 2>&1 || true
165 # do it again, in case we tried to unload ksocklnd too early
166 MODULES=$($LCTL modules | awk '{ print $2 }')
167 [ -n "$MODULES" ] && $RMMOD $MODULES >/dev/null || true
168 MODULES=$($LCTL modules | awk '{ print $2 }')
169 if [ -n "$MODULES" ]; then
170 echo "Modules still loaded: "
172 if [ -e $LPROC ]; then
173 echo "Lustre still loaded"
174 cat $LPROC/devices || true
178 echo "Lustre stopped, but LNET is still loaded"
179 wait_for_lnet || return 3
184 LEAK_LUSTRE=$(dmesg | tail -n 30 | grep "obd mem.*leaked" || true)
185 LEAK_PORTALS=$(dmesg | tail -n 20 | grep "Portals memory leaked" || true)
186 if [ "$LEAK_LUSTRE" -o "$LEAK_PORTALS" ]; then
187 echo "$LEAK_LUSTRE" 1>&2
188 echo "$LEAK_PORTALS" 1>&2
189 mv $TMP/debug $TMP/debug-leak.`date +%s` || true
190 echo "Memory leaks detected"
191 [ "$LEAK_LUSTRE" -a $(echo $LEAK_LUSTRE | awk 'leaked=$8 {print leaked % 56}') == 0 ] && echo "ignoring known bug 10818" && return 0
194 echo "modules unloaded."
199 # start facet device options
205 echo "Starting ${facet}: $@ ${device} ${MOUNT%/*}/${facet}"
206 do_facet ${facet} mkdir -p ${MOUNT%/*}/${facet}
207 do_facet ${facet} mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}
209 if [ $RC -ne 0 ]; then
210 echo mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}
211 echo Start of ${device} on ${facet} failed ${RC}
213 do_facet ${facet} sync
214 label=`do_facet ${facet} "e2label ${device}" | grep -v "CMD: "`
215 [ -z "$label" ] && echo no label for ${device} && exit 1
216 eval export ${facet}_svc=${label}
217 eval export ${facet}_dev=${device}
218 eval export ${facet}_opt=\"$@\"
219 echo Started ${label}
228 HOST=`facet_active_host $facet`
229 [ -z $HOST ] && echo stop: no host for $facet && return 0
231 running=`do_facet ${facet} "grep -c ${MOUNT%/*}/${facet}' ' /proc/mounts" | grep -v "CMD: "`
232 if [ ${running} -ne 0 ]; then
233 echo "Stopping ${MOUNT%/*}/${facet} (opts:$@)"
234 do_facet ${facet} umount -d $@ ${MOUNT%/*}/${facet}
237 # umount should block, but we should wait for unrelated obd's
238 # like the MGS or MGC to also stop.
241 # conf-sanity 31 takes a long time cleanup
242 while [ $WAIT -lt 300 ]; do
243 running=$(do_facet ${facet} "[ -e $LPROC ] && grep ST' ' $LPROC/devices" | grep -v "CMD: ") || true
244 if [ -z "${running}" ]; then
247 echo "waited $WAIT for${running}"
248 if [ $INTERVAL -lt 64 ]; then
249 INTERVAL=$((INTERVAL + INTERVAL))
252 WAIT=$((WAIT + INTERVAL))
254 echo "service didn't stop after $WAIT seconds. Still running:"
263 # Only supply -o to mount if we have options
264 if [ -n "$MOUNTOPT" ]; then
265 OPTIONS="-o $MOUNTOPT"
267 local device=$MGSNID:/$FSNAME
268 if [ -z "$mnt" -o -z "$FSNAME" ]; then
269 echo Bad zconf mount command: opt=$OPTIONS dev=$device mnt=$mnt
273 echo "Starting client: $OPTIONS $device $mnt"
274 do_node $client mkdir -p $mnt
275 do_node $client mount -t lustre $OPTIONS $device $mnt || return 1
277 do_node $client "sysctl -w lnet.debug=$PTLDEBUG; sysctl -w lnet.subsystem_debug=${SUBSYSTEM# }"
278 [ -d /r ] && $LCTL modules > /r/tmp/ogdb-`hostname`
286 local running=`do_node $client "grep -c $mnt' ' /proc/mounts" | grep -v "CMD: "`
287 if [ $running -ne 0 ]; then
288 echo "Stopping client $mnt (opts:$force)"
289 do_node $client umount $force $mnt
295 if [ "$FAILURE_MODE" = HARD ]; then
296 $POWER_DOWN `facet_active_host $facet`
298 elif [ "$FAILURE_MODE" = SOFT ]; then
305 if [ "$FAILURE_MODE" = HARD ]; then
306 $POWER_UP `facet_active_host $facet`
312 # verify that lustre actually cleaned up properly
314 [ -f $CATASTROPHE ] && [ `cat $CATASTROPHE` -ne 0 ] && \
315 error "LBUG/LASSERT detected"
316 BUSY=`dmesg | grep -i destruct || true`
319 [ -e $TMP/debug ] && mv $TMP/debug $TMP/debug-busy.`date +%s`
322 LEAK_LUSTRE=`dmesg | tail -n 30 | grep "obd mem.*leaked" || true`
323 LEAK_PORTALS=`dmesg | tail -n 20 | grep "Portals memory leaked" || true`
324 if [ "$LEAK_LUSTRE" -o "$LEAK_PORTALS" ]; then
325 echo "$0: $LEAK_LUSTRE" 1>&2
326 echo "$0: $LEAK_PORTALS" 1>&2
327 echo "$0: Memory leak(s) detected..." 1>&2
328 mv $TMP/debug $TMP/debug-leak.`date +%s`
332 [ "`lctl dl 2> /dev/null | wc -l`" -gt 0 ] && lctl dl && \
333 echo "$0: lustre didn't clean up..." 1>&2 && return 202 || true
335 if [ "`/sbin/lsmod 2>&1 | egrep 'lnet|libcfs'`" ]; then
336 echo "$0: modules still loaded..." 1>&2
345 check_network "$HOST" 900
346 while ! do_node $HOST "ls -d $LUSTRE " > /dev/null; do sleep 5; done
351 HOST=`facet_active_host $facet`
356 # not every config has many clients
357 if [ ! -z "$CLIENTS" ]; then
358 $PDSH $CLIENTS "df $MOUNT" > /dev/null
363 uname -n >> $MOUNT/recon
364 if [ ! -z "$CLIENTS" ]; then
365 $PDSH $CLIENTS "df $MOUNT; uname -n >> $MOUNT/recon" > /dev/null
367 echo Connected clients:
369 ls -l $MOUNT/recon > /dev/null
375 echo "Failing $facet on node `facet_active_host $facet`"
376 shutdown_facet $facet
380 echo "df pid is $DFPID"
382 TO=`facet_active_host $facet`
383 echo "Failover $facet to $TO"
385 local dev=${facet}_dev
386 local opt=${facet}_opt
387 start $facet ${!dev} ${!opt}
398 local svc=${facet}_svc
399 do_facet $facet $LCTL --device %${!svc} readonly
400 do_facet $facet $LCTL --device %${!svc} notransno
401 do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
402 $LCTL mark "local REPLAY BARRIER on ${!svc}"
405 replay_barrier_nodf() {
406 local facet=$1 echo running=${running}
408 local svc=${facet}_svc
409 echo Replay barrier on ${!svc}
410 do_facet $facet $LCTL --device %${!svc} readonly
411 do_facet $facet $LCTL --device %${!svc} notransno
412 do_facet $facet $LCTL mark "$facet REPLAY BARRIER on ${!svc}"
413 $LCTL mark "local REPLAY BARRIER on ${!svc}"
417 UUID=`cat /proc/fs/lustre/mdc/${mds_svc}-mdc-*/uuid`
418 do_facet mds "echo $UUID > /proc/fs/lustre/mds/${mds_svc}/evict_client"
422 UUID=`cat /proc/fs/lustre/osc/${ost1_svc}-osc-*/uuid`
423 do_facet ost1 "echo $UUID > /proc/fs/lustre/obdfilter/${ost1_svc}/evict_client"
427 facet_failover $* || error "failover: $?"
428 df $MOUNT || error "post-failover df: $?"
435 local svc=${facet}_svc
436 local dev=${facet}_dev
437 local opt=${facet}_opt
438 start $facet ${!dev} ${!opt}
439 do_facet $facet lctl --device %${!svc} abort_recovery
440 df $MOUNT || echo "first df failed: $?"
442 df $MOUNT || error "post-failover df: $?"
446 echo There is no lmc. This is mountconf, baby.
451 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
452 ID=`$PDSH $1 $GMNALNID -l | cut -d\ -f2`
458 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
465 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
466 if type __h2elan >/dev/null 2>&1; then
469 ID=`echo $1 | sed 's/[^0-9]*//g'`
477 if [ "$1" = "client" -o "$1" = "'*'" ]; then echo \'*\'; else
478 ID=`echo $1 | sed 's/[^0-9]*//g'`
486 varname=${facet}_HOST
487 if [ -z "${!varname}" ]; then
488 if [ "${facet:0:3}" == "ost" ]; then
489 eval ${facet}_HOST=${ost_HOST}
497 local activevar=${facet}active
499 if [ -f ./${facet}active ] ; then
500 source ./${facet}active
504 if [ -z "$active" ] ; then
511 facet_active_host() {
513 local active=`facet_active $facet`
514 if [ "$facet" == client ]; then
517 echo `facet_host $active`
523 failover=${facet}failover
524 host=`facet_host $failover`
525 [ -z "$host" ] && return
526 curactive=`facet_active $facet`
527 if [ -z "${curactive}" -o "$curactive" == "$failover" ] ; then
528 eval export ${facet}active=$facet
530 eval export ${facet}active=$failover
532 # save the active host for this facet
533 activevar=${facet}active
534 echo "$activevar=${!activevar}" > ./$activevar
541 if [ "$HOST" = "$(hostname)" ]; then
546 $myPDSH $HOST $LCTL mark "$@" > /dev/null 2>&1 || :
548 $myPDSH $HOST "(PATH=\$PATH:$RLUSTRE/utils:$RLUSTRE/tests:/sbin:/usr/sbin; cd $RPWD; sh -c \"$@\")"
554 HOST=`facet_active_host $facet`
555 [ -z $HOST ] && echo No host defined for facet ${facet} && exit 1
562 # make sure its not already running
565 do_facet ${facet} $MKFS $*
571 #if $OSTDEVn isn't defined, default is $OSTDEVBASE + num
572 eval DEVPTR=${!DEVNAME:=${OSTDEVBASE}${num}}
580 # make sure we are using the primary server, so test-framework will
581 # be able to clean up properly.
582 activemds=`facet_active mds`
583 if [ $activemds != "mds" ]; then
587 # assume client mount is local
588 grep " $MOUNT " /proc/mounts && zconf_umount `hostname` $MOUNT $*
589 grep " $MOUNT2 " /proc/mounts && zconf_umount `hostname` $MOUNT2 $*
591 for num in `seq $OSTCOUNT`; do
603 [ "$FSTYPE" ] && FSTYPE_OPT="--backfstype $FSTYPE"
606 # We need ldiskfs here, may as well load them all
608 echo Formatting mds, osts
610 add mds $MDS_MKFS_OPTS $FSTYPE_OPT --reformat $MDSDEV || exit 10
612 add mds $MDS_MKFS_OPTS $FSTYPE_OPT --reformat $MDSDEV > /dev/null || exit 10
615 for num in `seq $OSTCOUNT`; do
617 add ost$num $OST_MKFS_OPTS $FSTYPE_OPT --reformat `ostdevname $num` || exit 10
619 add ost$num $OST_MKFS_OPTS $FSTYPE_OPT --reformat `ostdevname $num` > /dev/null || exit 10
625 grep " $1 " /proc/mounts || zconf_mount `hostname` $*
631 start mds $MDSDEV $MDS_MOUNT_OPTS
632 for num in `seq $OSTCOUNT`; do
633 DEVNAME=`ostdevname $num`
634 start ost$num $DEVNAME $OST_MOUNT_OPTS
636 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
638 if [ "$MOUNT_2" ]; then
652 while [ $NETWORK -eq 0 ]; do
653 ping -c 1 -w 3 $1 > /dev/null
654 if [ $? -eq 0 ]; then
658 echo "waiting for $1, $((MAX - WAIT)) secs left"
661 if [ $WAIT -gt $MAX ]; then
662 echo "Network not available"
668 while( !($DSH2 $1 "netstat -tna | grep -q $2") ) ; do
679 # the sed converts spaces to commas, but leaves the last space
680 # alone, so the line doesn't end with a comma.
681 echo "$*" | tr -s " " "\n" | sort -b -u | tr "\n" " " | sed 's/ \([^$]\)/,\1/g'
685 (cd `dirname $1`; echo $PWD/`basename $1`)
688 ##################################
692 # OBD_FAIL_MDS_ALL_REQUEST_NET
694 do_facet mds sysctl -w lustre.fail_loc=0x123
695 do_facet client "$1" || RC=$?
696 do_facet mds sysctl -w lustre.fail_loc=0
701 # OBD_FAIL_MDS_ALL_REPLY_NET
703 do_facet mds sysctl -w lustre.fail_loc=0x122
704 do_facet client "$@" || RC=$?
705 do_facet mds sysctl -w lustre.fail_loc=0
710 # OBD_FAIL_MDS_REINT_NET_REP
712 do_facet mds sysctl -w lustre.fail_loc=0x119
713 do_facet client "$@" || RC=$?
714 do_facet mds sysctl -w lustre.fail_loc=0
719 #define OBD_FAIL_OST_BRW_PAUSE_BULK 0x214
721 do_facet ost1 sysctl -w lustre.fail_loc=0x214
722 do_facet client "$1" || RC=$?
723 do_facet client "sync"
724 do_facet ost1 sysctl -w lustre.fail_loc=0
729 #define OBD_FAIL_LDLM_CANCEL 0x304
731 do_facet client sysctl -w lustre.fail_loc=0x304
732 do_facet client "$@" || RC=$?
733 do_facet client sysctl -w lustre.fail_loc=0
738 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
740 do_facet client sysctl -w lustre.fail_loc=0x305
741 do_facet client "$@" || RC=$?
742 do_facet client sysctl -w lustre.fail_loc=0
747 #define OBD_FAIL_LDLM_REPLY 0x30c
749 do_facet mds sysctl -w lustre.fail_loc=0x30c
750 do_facet client "$@" || RC=$?
751 do_facet mds sysctl -w lustre.fail_loc=0
759 echo "clearing fail_loc on $facet"
760 do_facet $facet "sysctl -w lustre.fail_loc=0"
764 $LCTL mark "cancel_lru_locks $1 start"
765 for d in `find $LPROC/ldlm/namespaces | egrep -i $1`; do
766 [ -f $d/lru_size ] && echo clear > $d/lru_size
767 [ -f $d/lock_unused_count ] && grep [1-9] $d/lock_unused_count /dev/null
769 $LCTL mark "cancel_lru_locks $1 stop"
774 for a in /proc/fs/lustre/llite/*/dump_page_cache; do
775 if [ `wc -l $a | awk '{print $1}'` -gt 1 ]; then
776 echo there is still data in page cache $a ?
784 ##################################
787 sysctl -w lustre.fail_loc=0 2> /dev/null || true
788 echo "${TESTSUITE}: **** FAIL:" $@
789 log "FAIL: $TESTNAME $@"
793 build_test_filter() {
794 [ "$ONLY" ] && log "only running test `echo $ONLY`"
798 [ "$EXCEPT$ALWAYS_EXCEPT" ] && \
799 log "skipping tests: `echo $EXCEPT $ALWAYS_EXCEPT`"
800 for E in $EXCEPT $ALWAYS_EXCEPT; do
801 eval EXCEPT_${E}=true
810 IFS=abcdefghijklmnopqrstuvwxyz _basetest $1
814 export base=`basetest $1`
815 if [ ! -z "$ONLY" ]; then
817 if [ ${!testname}x != x ]; then
822 if [ ${!testname}x != x ]; then
830 if [ ${!testname}x != x ]; then
831 log "skipping excluded test $1"
834 testname=EXCEPT_$base
835 if [ ${!testname}x != x ]; then
836 log "skipping excluded test $1 (base $base)"
844 EQUALS="======================================================================"
848 local suffixlen=$((${#EQUALS} - ${#msg}))
849 [ $suffixlen -lt 5 ] && suffixlen=5
850 printf '===== %s %.*s\n' "$msg" $suffixlen $EQUALS
855 lsmod | grep lnet > /dev/null || load_modules
856 $LCTL mark "$*" 2> /dev/null || true
864 FFREE=`cat /proc/fs/lustre/mds/*/filesfree`
865 FTOTAL=`cat /proc/fs/lustre/mds/*/filestotal`
866 [ $FFREE -ge $FTOTAL ] && error "files free $FFREE > total $FTOTAL" || true
875 # Pretty tests run faster.
876 equals_msg $testnum: $message
879 log "== test $testnum: $message ============ `date +%H:%M:%S` ($BEFORE)"
881 export TESTNAME=test_$testnum
882 test_${testnum} || error "test_$testnum failed with $?"
884 [ -f $CATASTROPHE ] && [ `cat $CATASTROPHE` -ne 0 ] && \
885 error "LBUG/LASSERT detected"
886 pass "($((`date +%s` - $BEFORE))s)"
891 (cd `dirname $1`; echo $PWD/`basename $1`)