X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fconf-sanity.sh;h=11b276840f6b97f6489e0dada804098154058648;hp=047331d134b363be923540613ee302215f4c3999;hb=e768adb2fa9b1dca5c45be13d38497547e74060c;hpb=6ed82f007a5369e5a8fdb4930f0c17ca87d97bd0 diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh index 047331d..11b2768 100644 --- a/lustre/tests/conf-sanity.sh +++ b/lustre/tests/conf-sanity.sh @@ -11,11 +11,11 @@ set -e ONLY=${ONLY:-"$*"} -# These tests don't apply to mountconf -MOUNTCONFSKIP="9 10 11 12 13 13b 14 15" +# bug number for skipped test: 13739 +HEAD_EXCEPT=" 32a 32b" -# bug number for skipped test: -ALWAYS_EXCEPT=" $CONF_SANITY_EXCEPT $MOUNTCONFSKIP 16 23 36" +# bug number for skipped test: +ALWAYS_EXCEPT=" $CONF_SANITY_EXCEPT $HEAD_EXCEPT" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! SRCDIR=`dirname $0` @@ -25,28 +25,38 @@ PTLDEBUG=${PTLDEBUG:--1} SAVE_PWD=$PWD LUSTRE=${LUSTRE:-`dirname $0`/..} RLUSTRE=${RLUSTRE:-$LUSTRE} -MOUNTLUSTRE=${MOUNTLUSTRE:-/sbin/mount.lustre} -MKFSLUSTRE=${MKFSLUSTRE:-/usr/sbin/mkfs.lustre} -HOSTNAME=`hostname` . $LUSTRE/tests/test-framework.sh init_test_env $@ +# STORED_MDSSIZE is used in test_18 +if [ -n "$MDSSIZE" ]; then + STORED_MDSSIZE=$MDSSIZE +fi # use small MDS + OST size to speed formatting time MDSSIZE=40000 OSTSIZE=40000 -. ${CONFIG:=$LUSTRE/tests/cfg/local.sh} +. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} + +remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0 +remote_ost_nodsh && skip "remote OST with nodsh" && exit 0 + +# +[ "$SLOW" = "no" ] && EXCEPT_SLOW="0 1 2 3 6 7 15 18 24b 25 30 31 32 33 34a 45" + +assert_DIR reformat() { formatall } writeconf() { - local facet=mds + local facet=$SINGLEMDS + local dev=${facet}_dev shift stop ${facet} -f rm -f ${facet}active # who knows if/where $TUNEFS is installed? Better reformat if it fails... - do_facet ${facet} "$TUNEFS --writeconf $MDSDEV" || echo "tunefs failed, reformatting instead" && reformat + do_facet ${facet} "$TUNEFS --writeconf ${!dev}" || echo "tunefs failed, reformatting instead" && reformat } gen_config() { @@ -61,14 +71,18 @@ gen_config() { } start_mds() { - echo "start mds service on `facet_active_host mds`" - start mds $MDSDEV $MDS_MOUNT_OPTS || return 94 + local facet=$SINGLEMDS + # we can not use MDSDEV1 here because SINGLEMDS could be set not to mds1 only + local num=$(echo $facet | tr -d "mds") + local dev=$(mdsdevname $num) + echo "start mds service on `facet_active_host $facet`" + start $facet ${dev} $MDS_MOUNT_OPTS || return 94 } stop_mds() { - echo "stop mds service on `facet_active_host mds`" + echo "stop mds service on `facet_active_host $SINGLEMDS`" # These tests all use non-failover stop - stop mds -f || return 97 + stop $SINGLEMDS -f || return 97 } start_ost() { @@ -115,8 +129,12 @@ umount_client() { } manual_umount_client(){ + local rc + local FORCE=$1 echo "manual umount lustre on ${MOUNT}...." - do_facet client "umount -d $MOUNT" + do_facet client "umount -d ${FORCE} $MOUNT" + rc=$? + return $rc } setup() { @@ -169,7 +187,7 @@ fi gen_config -init_krb5_env +init_gss test_0() { setup @@ -189,7 +207,7 @@ run_test 1 "start up ost twice (should return errors)" test_2() { start_ost - start_mds + start_mds echo "start mds second time.." start_mds mount_client $MOUNT @@ -221,9 +239,11 @@ test_4() { } run_test 4 "force cleanup ost, then cleanup" -test_5() { +test_5a() { # was test_5 setup touch $DIR/$tfile || return 1 + fuser -m -v $MOUNT && echo "$MOUNT is in use by user space process." + stop_mds -f || return 2 # cleanup may return an error from the failed @@ -236,8 +256,8 @@ test_5() { kill -TERM $UMOUNT_PID echo "waiting for umount to finish" wait $UMOUNT_PID - if grep " $MOUNT " /etc/mtab; then - echo "test 5: mtab after failed umount" + if grep " $MOUNT " /proc/mounts; then + echo "test 5: /proc/mounts after failed umount" umount $MOUNT & UMOUNT_PID=$! sleep 2 @@ -245,7 +265,7 @@ test_5() { kill -TERM $UMOUNT_PID echo "waiting for umount to finish" wait $UMOUNT_PID - grep " $MOUNT " /etc/mtab && echo "test 5: mtab after second umount" && return 11 + grep " $MOUNT " /proc/mounts && echo "test 5: /proc/mounts after second umount" && return 11 fi manual_umount_client @@ -254,10 +274,19 @@ test_5() { # df may have lingering entry manual_umount_client # mtab may have lingering entry - grep -v $MOUNT" " /etc/mtab > $TMP/mtabtemp - mv $TMP/mtabtemp /etc/mtab + local WAIT=0 + local MAX_WAIT=20 + local sleep=1 + while [ "$WAIT" -ne "$MAX_WAIT" ]; do + sleep $sleep + grep -q $MOUNT" " /etc/mtab || break + echo "Waiting /etc/mtab updated ... " + WAIT=$(( WAIT + sleep)) + done + [ "$WAIT" -eq "$MAX_WAIT" ] && error "/etc/mtab is not updated in $WAIT secs" + echo "/etc/mtab updated in $WAIT secs" } -run_test 5 "force cleanup mds, then cleanup" +run_test 5a "force cleanup mds, then cleanup" test_5b() { start_ost @@ -265,7 +294,7 @@ test_5b() { grep " $MOUNT " /etc/mtab && echo "test 5b: mtab before mount" && return 10 mount_client $MOUNT && return 1 grep " $MOUNT " /etc/mtab && echo "test 5b: mtab after failed mount" && return 11 - umount_client $MOUNT + umount_client $MOUNT # stop_mds is a no-op here, and should not fail cleanup_nocli || return $? return 0 @@ -277,7 +306,10 @@ test_5c() { start_mds [ -d $MOUNT ] || mkdir -p $MOUNT grep " $MOUNT " /etc/mtab && echo "test 5c: mtab before mount" && return 10 - mount -t lustre $MGSNID:/wrong.$FSNAME $MOUNT || : + local oldfs="${FSNAME}" + FSNAME="wrong.${FSNAME}" + mount_client $MOUNT || : + FSNAME=${oldfs} grep " $MOUNT " /etc/mtab && echo "test 5c: mtab after failed mount" && return 11 umount_client $MOUNT cleanup_nocli || return $? @@ -299,11 +331,9 @@ run_test 5d "mount with ost down" test_5e() { start_ost start_mds - # give MDS a chance to connect to OSTs (bz 10476) - sleep 5 #define OBD_FAIL_PTLRPC_DELAY_SEND 0x506 - do_facet client "sysctl -w lustre.fail_loc=0x80000506" + do_facet client "lctl set_param fail_loc=0x80000506" grep " $MOUNT " /etc/mtab && echo "test 5e: mtab before mount" && return 10 mount_client $MOUNT || echo "mount failed (not fatal)" cleanup || return $? @@ -340,17 +370,19 @@ run_test 8 "double mount setup" test_9() { start_ost - do_facet ost1 sysctl lnet.debug=\'inode trace\' || return 1 - do_facet ost1 sysctl lnet.subsystem_debug=\'mds ost\' || return 1 + do_facet ost1 lctl set_param debug=\'inode trace\' || return 1 + do_facet ost1 lctl set_param subsystem_debug=\'mds ost\' || return 1 - CHECK_PTLDEBUG="`do_facet ost1 sysctl -n lnet.debug`" - if [ "$CHECK_PTLDEBUG" ] && [ "$CHECK_PTLDEBUG" = "trace inode" ];then + CHECK_PTLDEBUG="`do_facet ost1 lctl get_param -n debug`" + if [ "$CHECK_PTLDEBUG" ] && { \ + [ "$CHECK_PTLDEBUG" = "trace inode warning error emerg console" ] || + [ "$CHECK_PTLDEBUG" = "trace inode" ]; }; then echo "lnet.debug success" else echo "lnet.debug: want 'trace inode', have '$CHECK_PTLDEBUG'" return 1 fi - CHECK_SUBSYS="`do_facet ost1 sysctl -n lnet.subsystem_debug`" + CHECK_SUBSYS="`do_facet ost1 lctl get_param -n subsystem_debug`" if [ "$CHECK_SUBSYS" ] && [ "$CHECK_SUBSYS" = "mds ost" ]; then echo "lnet.subsystem_debug success" else @@ -362,262 +394,11 @@ test_9() { run_test 9 "test ptldebug and subsystem for mkfs" -test_10() { - echo "generate configuration with the same name for node and mds" - OLDXMLCONFIG=$XMLCONFIG - XMLCONFIG="broken.xml" - [ -f "$XMLCONFIG" ] && rm -f $XMLCONFIG - facet="mds" - rm -f ${facet}active - add_facet $facet - echo "the name for node and mds is the same" - do_lmc --add mds --node ${facet}_facet --mds ${facet}_facet \ - --dev $MDSDEV --size $MDSSIZE || return $? - do_lmc --add lov --mds ${facet}_facet --lov lov1 --stripe_sz \ - $STRIPE_BYTES --stripe_cnt $STRIPES_PER_OBJ \ - --stripe_pattern 0 || return $? - add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE - facet="client" - add_facet $facet --lustre_upcall $UPCALL - do_lmc --add mtpt --node ${facet}_facet --mds mds_facet \ - --lov lov1 --path $MOUNT - - echo "mount lustre" - start_ost - start_mds - mount_client $MOUNT - check_mount || return 41 - cleanup || return $? - - echo "Success!" - XMLCONFIG=$OLDXMLCONFIG -} -run_test 10 "mount lustre with the same name for node and mds" - -test_11() { - OLDXMLCONFIG=$XMLCONFIG - XMLCONFIG="conf11.xml" - - [ -f "$XMLCONFIG" ] && rm -f $XMLCONFIG - add_mds mds --dev $MDSDEV --size $MDSSIZE - add_ost ost --dev $OSTDEV --size $OSTSIZE - add_client client mds --path $MOUNT --ost ost_svc || return $? - echo "Default lov config success!" - - [ -f "$XMLCONFIG" ] && rm -f $XMLCONFIG - add_mds mds --dev $MDSDEV --size $MDSSIZE - add_ost ost --dev $OSTDEV --size $OSTSIZE - add_client client mds --path $MOUNT && return $? - echo "--add mtpt with neither --lov nor --ost will return error" - - echo "" - echo "Success!" - XMLCONFIG=$OLDXMLCONFIG -} -run_test 11 "use default lov configuration (should return error)" - -test_12() { - OLDXMLCONFIG=$XMLCONFIG - XMLCONFIG="batch.xml" - BATCHFILE="batchfile" - - # test double quote - [ -f "$XMLCONFIG" ] && rm -f $XMLCONFIG - [ -f "$BATCHFILE" ] && rm -f $BATCHFILE - echo "--add net --node $HOSTNAME --nid $HOSTNAME --nettype tcp" > $BATCHFILE - echo "--add mds --node $HOSTNAME --mds mds1 --mkfsoptions \"-I 128\"" >> $BATCHFILE - # --mkfsoptions "-I 128" - do_lmc -m $XMLCONFIG --batch $BATCHFILE || return $? - if [ `sed -n '/>-I 128 $BATCHFILE - echo "--add mds --node $HOSTNAME --mds mds1 --mkfsoptions \"-I 128" >> $BATCHFILE - # --mkfsoptions "-I 128 - do_lmc -m $XMLCONFIG --batch $BATCHFILE && return $? - echo "unmatched double quote should return error" - - # test single quote - rm -f $BATCHFILE - echo "--add net --node $HOSTNAME --nid $HOSTNAME --nettype tcp" > $BATCHFILE - echo "--add mds --node $HOSTNAME --mds mds1 --mkfsoptions '-I 128'" >> $BATCHFILE - # --mkfsoptions '-I 128' - do_lmc -m $XMLCONFIG --batch $BATCHFILE || return $? - if [ `sed -n '/>-I 128 $BATCHFILE - echo "--add mds --node $HOSTNAME --mds mds1 --mkfsoptions '-I 128" >> $BATCHFILE - # --mkfsoptions '-I 128 - do_lmc -m $XMLCONFIG --batch $BATCHFILE && return $? - echo "unmatched single quote should return error" - - # test backslash - rm -f $BATCHFILE - echo "--add net --node $HOSTNAME --nid $HOSTNAME --nettype tcp" > $BATCHFILE - echo "--add mds --node $HOSTNAME --mds mds1 --mkfsoptions \-\I\ \128" >> $BATCHFILE - # --mkfsoptions \-\I\ \128 - do_lmc -m $XMLCONFIG --batch $BATCHFILE || return $? - if [ `sed -n '/>-I 128 $BATCHFILE - echo "--add mds --node $HOSTNAME --mds mds1 --mkfsoptions -I\ 128\\" >> $BATCHFILE - # --mkfsoptions -I\ 128\ - do_lmc -m $XMLCONFIG --batch $BATCHFILE && return $? - echo "backslash followed by nothing should return error" - - rm -f $BATCHFILE - XMLCONFIG=$OLDXMLCONFIG -} -run_test 12 "lmc --batch, with single/double quote, backslash in batchfile" - -test_13() { - OLDXMLCONFIG=$XMLCONFIG - XMLCONFIG="conf13-1.xml" - - # check long uuid will be truncated properly and uniquely - echo "To generate XML configuration file(with long ost name): $XMLCONFIG" - [ -f "$XMLCONFIG" ] && rm -f $XMLCONFIG - do_lmc --add net --node $HOSTNAME --nid $HOSTNAME --nettype tcp - do_lmc --add mds --node $HOSTNAME --mds mds1_name_longer_than_31characters - do_lmc --add mds --node $HOSTNAME --mds mds2_name_longer_than_31characters - if [ ! -f "$XMLCONFIG" ]; then - echo "Error:no file $XMLCONFIG created!" - return 1 - fi - EXPECTEDMDS1UUID="e_longer_than_31characters_UUID" - EXPECTEDMDS2UUID="longer_than_31characters_UUID_2" - FOUNDMDS1UUID=`awk -F"'" '/ $SECONDXMLCONFIG || return $? - echo "Generate the second XML configuration file" - gen_config - # don't compare .xml mtime, it will always be different - if [ `sed -e "s/mtime[^ ]*//" $XMLCONFIG | diff - $SECONDXMLCONFIG | wc -l` -eq 0 ]; then - echo "Success:multiple invocations for lmc generate same XML file" - else - echo "Error: multiple invocations for lmc generate different XML file" - return 1 - fi - - rm -f $XMLCONFIG $SECONDXMLCONFIG - XMLCONFIG=$OLDXMLCONFIG -} -run_test 13b "check lmc generates consistent .xml file" - -test_14() { - rm -f $XMLCONFIG - - # create xml file with --mkfsoptions for ost - echo "create xml file with --mkfsoptions for ost" - add_mds mds --dev $MDSDEV --size $MDSSIZE - add_lov lov1 mds --stripe_sz $STRIPE_BYTES\ - --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0 - add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE \ - --mkfsoptions "-Llabel_conf_14" - add_client client mds --lov lov1 --path $MOUNT - - FOUNDSTRING=`awk -F"<" '//{print $2}' $XMLCONFIG` - EXPECTEDSTRING="mkfsoptions>-Llabel_conf_14" - if [ "$EXPECTEDSTRING" != "$FOUNDSTRING" ]; then - echo "Error: expected: $EXPECTEDSTRING; found: $FOUNDSTRING" - return 1 - fi - echo "Success:mkfsoptions for ost written to xml file correctly." - - # mount lustre to test lconf mkfsoptions-parsing - echo "mount lustre" - start_ost - start_mds - mount_client $MOUNT || return $? - if [ -z "`do_facet ost1 dumpe2fs -h $OSTDEV | grep label_conf_14`" ]; then - echo "Error: the mkoptions not applied to mke2fs of ost." - return 1 - fi - cleanup - echo "lconf mkfsoptions for ost success" - - gen_config -} -run_test 14 "test mkfsoptions of ost for lmc and lconf" - -cleanup_15() { - trap 0 - [ -f $MOUNTLUSTRE ] && echo "remove $MOUNTLUSTRE" && rm -f $MOUNTLUSTRE - if [ -f $MOUNTLUSTRE.sav ]; then - echo "return original $MOUNTLUSTRE.sav to $MOUNTLUSTRE" - mv $MOUNTLUSTRE.sav $MOUNTLUSTRE - fi -} - -# this only tests the kernel mount command, not anything about lustre. -test_15() { - MOUNTLUSTRE=${MOUNTLUSTRE:-/sbin/mount.lustre} - start_ost - start_mds - - echo "mount lustre on ${MOUNT} without $MOUNTLUSTRE....." - if [ -f "$MOUNTLUSTRE" ]; then - echo "save $MOUNTLUSTRE to $MOUNTLUSTRE.sav" - mv $MOUNTLUSTRE $MOUNTLUSTRE.sav && trap cleanup_15 EXIT INT - if [ -f $MOUNTLUSTRE ]; then - skip "$MOUNTLUSTRE cannot be moved, skipping test" - return 0 - fi - fi - - mount_client $MOUNT && error "mount succeeded" && return 1 - echo "mount lustre on $MOUNT without $MOUNTLUSTRE failed as expected" - cleanup_15 - cleanup || return $? -} -run_test 15 "zconf-mount without /sbin/mount.lustre (should return error)" - +# LOGS/PENDING do not exist anymore since CMD3 test_16() { - TMPMTPT="${MOUNT%/*}/conf16" - + local TMPMTPT="${TMP}/conf16" + local dev=${SINGLEMDS}_dev + local MDSDEV=${!dev} if [ ! -e "$MDSDEV" ]; then log "no $MDSDEV existing, so mount Lustre to create one" setup @@ -627,43 +408,32 @@ test_16() { [ -f "$MDSDEV" ] && LOOPOPT="-o loop" - log "change the mode of $MDSDEV/OBJECTS,LOGS,PENDING to 555" - do_facet mds "mkdir -p $TMPMTPT && + log "change the mode of $MDSDEV/OBJECTS to 555" + do_facet $SINGLEMDS "mkdir -p $TMPMTPT && mount $LOOPOPT -t $FSTYPE $MDSDEV $TMPMTPT && - chmod 555 $TMPMTPT/{OBJECTS,LOGS,PENDING} && + chmod 555 $TMPMTPT/OBJECTS && umount $TMPMTPT" || return $? - log "mount Lustre to change the mode of OBJECTS/LOGS/PENDING, then umount Lustre" + log "mount Lustre to change the mode of OBJECTS, then umount Lustre" setup check_mount || return 41 cleanup || return $? - log "read the mode of OBJECTS/LOGS/PENDING and check if they has been changed properly" - EXPECTEDOBJECTSMODE=`do_facet mds "debugfs -R 'stat OBJECTS' $MDSDEV 2> /dev/null" | grep 'Mode: ' | sed -e "s/.*Mode: *//" -e "s/ *Flags:.*//"` - EXPECTEDLOGSMODE=`do_facet mds "debugfs -R 'stat LOGS' $MDSDEV 2> /dev/null" | grep 'Mode: ' | sed -e "s/.*Mode: *//" -e "s/ *Flags:.*//"` - EXPECTEDPENDINGMODE=`do_facet mds "debugfs -R 'stat PENDING' $MDSDEV 2> /dev/null" | grep 'Mode: ' | sed -e "s/.*Mode: *//" -e "s/ *Flags:.*//"` + log "read the mode of OBJECTS and check if they has been changed properly" + EXPECTEDOBJECTSMODE=`do_facet $SINGLEMDS "debugfs -R 'stat OBJECTS' $MDSDEV 2> /dev/null" | grep 'Mode: ' | sed -e "s/.*Mode: *//" -e "s/ *Flags:.*//"` if [ "$EXPECTEDOBJECTSMODE" = "0777" ]; then log "Success:Lustre change the mode of OBJECTS correctly" else error "Lustre does not change mode of OBJECTS properly" fi - - if [ "$EXPECTEDLOGSMODE" = "0777" ]; then - log "Success:Lustre change the mode of LOGS correctly" - else - error "Lustre does not change mode of LOGS properly" - fi - - if [ "$EXPECTEDPENDINGMODE" = "0777" ]; then - log "Success:Lustre change the mode of PENDING correctly" - else - error "Lustre does not change mode of PENDING properly" - fi } -run_test 16 "verify that lustre will correct the mode of OBJECTS/LOGS/PENDING" +run_test 16 "verify that lustre will correct the mode of OBJECTS" test_17() { + local dev=${SINGLEMDS}_dev + local MDSDEV=${!dev} + if [ ! -e "$MDSDEV" ]; then echo "no $MDSDEV existing, so mount Lustre to create one" setup @@ -672,7 +442,7 @@ test_17() { fi echo "Remove mds config log" - do_facet mds "debugfs -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MDSDEV || return \$?" || return $? + do_facet $SINGLEMDS "debugfs -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MDSDEV || return \$?" || return $? start_ost start_mds && return 42 @@ -681,10 +451,41 @@ test_17() { run_test 17 "Verify failed mds_postsetup won't fail assertion (2936) (should return errs)" test_18() { - [ -f $MDSDEV ] && echo "remove $MDSDEV" && rm -f $MDSDEV + [ "$FSTYPE" != "ldiskfs" ] && skip "not needed for FSTYPE=$FSTYPE" && return + + local dev=${SINGLEMDS}_dev + local MDSDEV=${!dev} + + local MIN=2000000 + + local OK= + # check if current MDSSIZE is large enough + [ $MDSSIZE -ge $MIN ] && OK=1 && myMDSSIZE=$MDSSIZE && \ + log "use MDSSIZE=$MDSSIZE" + + # check if the global config has a large enough MDSSIZE + [ -z "$OK" -a ! -z "$STORED_MDSSIZE" ] && [ $STORED_MDSSIZE -ge $MIN ] && \ + OK=1 && myMDSSIZE=$STORED_MDSSIZE && \ + log "use STORED_MDSSIZE=$STORED_MDSSIZE" + + # check if the block device is large enough + [ -z "$OK" -a -b $MDSDEV ] && \ + [ "$(dd if=$MDSDEV of=/dev/null bs=1k count=1 skip=$MIN 2>&1 | + awk '($3 == "in") { print $1 }')" = "1+0" ] && OK=1 && \ + myMDSSIZE=$MIN && log "use device $MDSDEV with MIN=$MIN" + + # check if a loopback device has enough space for fs metadata (5%) + [ -z "$OK" ] && [ -f $MDSDEV -o ! -e $MDSDEV ] && + SPACE=$(df -P $(dirname $MDSDEV) | + awk '($1 != "Filesystem") {print $4}') && + [ $SPACE -gt $((MIN / 20)) ] && OK=1 && myMDSSIZE=$MIN && \ + log "use file $MDSDEV with MIN=$MIN" + + [ -z "$OK" ] && skip "$MDSDEV too small for ${MIN}kB MDS" && return + + echo "mount mds with large journal..." - local myMDSSIZE=2000000 - OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS + local OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS MDS_MKFS_OPTS="--mgs --mdt --fsname=$FSNAME --device-size=$myMDSSIZE --param sys.timeout=$TIMEOUT $MDSOPT" @@ -694,7 +495,7 @@ test_18() { check_mount || return 41 echo "check journal size..." - FOUNDSIZE=`do_facet mds "debugfs -c -R 'stat <8>' $MDSDEV" | awk '/Size: / { print $NF; exit;}'` + local FOUNDSIZE=`do_facet mds "debugfs -c -R 'stat <8>' $MDSDEV" | awk '/Size: / { print $NF; exit;}'` if [ $FOUNDSIZE -gt $((32 * 1024 * 1024)) ]; then log "Success: mkfs creates large journals. Size: $((FOUNDSIZE >> 20))M" else @@ -764,12 +565,12 @@ test_21c() { stop_ost stop_ost2 stop_mds + #writeconf to remove all ost2 traces for subsequent tests + writeconf } run_test 21c "start mds between two osts, stop mds last" test_22() { - #reformat to remove all logs - reformat start_mds echo Client mount before any osts are in the logs mount_client $MOUNT @@ -786,8 +587,15 @@ test_22() { echo Client mount with a running ost start_ost + if $GSS; then + # if gss enabled, wait full time to let connection from + # mds to ost be established, due to the mismatch between + # initial connect timeout and gss context negotiation timeout. + # This perhaps could be remove after AT landed. + echo "sleep $((TIMEOUT + TIMEOUT + TIMEOUT))s" + sleep $((TIMEOUT + TIMEOUT + TIMEOUT)) + fi mount_client $MOUNT - sleep 5 #bz10476 check_mount || return 41 pass @@ -795,40 +603,90 @@ test_22() { } run_test 22 "start a client before osts (should return errs)" -test_23() { +test_23a() { # was test_23 setup # fail mds - stop mds + stop $SINGLEMDS # force down client so that recovering mds waits for reconnect - zconf_umount `hostname` $MOUNT -f + local running=$(grep -c $MOUNT /proc/mounts) || true + if [ $running -ne 0 ]; then + echo "Stopping client $MOUNT (opts: -f)" + umount -f $MOUNT + fi + # enter recovery on mds start_mds # try to start a new client mount_client $MOUNT & - MOUNT_PID=$! sleep 5 + MOUNT_PID=$(ps -ef | grep "t lustre" | grep -v grep | awk '{print $2}') MOUNT_LUSTRE_PID=`ps -ef | grep mount.lustre | grep -v grep | awk '{print $2}'` echo mount pid is ${MOUNT_PID}, mount.lustre pid is ${MOUNT_LUSTRE_PID} ps --ppid $MOUNT_PID ps --ppid $MOUNT_LUSTRE_PID # FIXME why o why can't I kill these? Manual "ctrl-c" works... - kill -TERM $MOUNT_PID + kill -TERM $MOUNT_LUSTRE_PID echo "waiting for mount to finish" ps -ef | grep mount - wait $MOUNT_PID + # we can not wait $MOUNT_PID because it is not a child of this shell + local PID1 + local PID2 + local WAIT=0 + local MAX_WAIT=20 + local sleep=1 + while [ "$WAIT" -lt "$MAX_WAIT" ]; do + sleep $sleep + PID1=$(ps -ef | awk '{print $2}' | grep -w $MOUNT_PID) + PID2=$(ps -ef | awk '{print $2}' | grep -w $MOUNT_LUSTRE_PID) + echo PID1=$PID1 + echo PID2=$PID2 + [ -z "$PID1" -a -z "$PID2" ] && break + echo "waiting for mount to finish ... " + WAIT=$(( WAIT + sleep)) + done + [ "$WAIT" -eq "$MAX_WAIT" ] && error "MOUNT_PID $MOUNT_PID and \ + MOUNT__LUSTRE_PID $MOUNT__LUSTRE_PID still not killed in $WAIT secs" + ps -ef | grep mount + stop_mds || error + stop_ost || error +} +run_test 23a "interrupt client during recovery mount delay" - stop_mds - stop_ost +umount_client $MOUNT +cleanup_nocli + +test_23b() { # was test_23 + start_ost + start_mds + # Simulate -EINTR during mount OBD_FAIL_LDLM_CLOSE_THREAD + lctl set_param fail_loc=0x80000313 + mount_client $MOUNT + cleanup } -#this test isn't working yet -#run_test 23 "interrupt client during recovery mount delay" +run_test 23b "Simulate -EINTR during mount" -test_24a() { - local fs2mds_HOST=$mds_HOST - local fs2ost_HOST=$ost_HOST +fs2mds_HOST=$mds_HOST +fs2ost_HOST=$ost_HOST + +cleanup_24a() { + trap 0 + echo "umount $MOUNT2 ..." + umount $MOUNT2 || true + echo "stopping fs2mds ..." + stop fs2mds -f || true + echo "stopping fs2ost ..." + stop fs2ost -f || true +} - [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ] && [ -b "$MDSDEV" ] && \ - log "mixed loopback and real device not working" && return +test_24a() { + #set up fs1 + gen_config + #set up fs2 + [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST + if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then + do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + skip "mixed loopback and real device not working" && return + fi local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2} @@ -840,7 +698,7 @@ test_24a() { add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2ostdev || exit 10 setup - start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS + start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT start fs2ost $fs2ostdev $OST_MOUNT_OPTS mkdir -p $MOUNT2 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1 @@ -863,23 +721,22 @@ test_24a() { umount_client $MOUNT # the MDS must remain up until last MDT stop_mds - MDS=$(do_facet mds "cat $LPROC/devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }') + MDS=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1) [ -z "$MDS" ] && error "No MDT" && return 8 - umount $MOUNT2 - stop fs2mds -f - stop fs2ost -f + cleanup_24a cleanup_nocli || return 6 } run_test 24a "Multiple MDTs on a single node" test_24b() { - local fs2mds_HOST=$mds_HOST - [ -z "$fs2mds_DEV" ] && [ -b "$MDSDEV" ] && \ - log "mixed loopback and real device not working" && return + if [ -z "$fs2mds_DEV" ]; then + do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + skip "mixed loopback and real device not working" && return + fi local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} - add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME}2 --mgs --reformat $fs2mdsdev || exit 10 + add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME}2 --mgs --reformat $fs2mdsdev || exit 10 setup start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && return 2 cleanup || return 6 @@ -898,12 +755,12 @@ run_test 25 "Verify modules are referenced" test_26() { load_modules # we need modules before mount for sysctl, so make sure... - do_facet mds "lsmod | grep -q lustre || modprobe lustre" + do_facet $SINGLEMDS "lsmod | grep -q lustre || modprobe lustre" #define OBD_FAIL_MDS_FS_SETUP 0x135 - do_facet mds "sysctl -w lustre.fail_loc=0x80000135" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000135" start_mds && echo MDS started && return 1 - cat $LPROC/devices - DEVS=$(cat $LPROC/devices | wc -l) + lctl get_param -n devices + DEVS=$(lctl get_param -n devices | wc -l) [ $DEVS -gt 0 ] && return 2 unload_modules || return 203 } @@ -921,7 +778,7 @@ set_and_check() { FINAL=$(($ORIG + 5)) fi echo "Setting $PARAM from $ORIG to $FINAL" - do_facet mds "$LCTL conf_param $PARAM=$FINAL" || error conf_param failed + do_facet $SINGLEMDS "$LCTL conf_param $PARAM=$FINAL" || error conf_param failed local RESULT local MAX=90 local WAIT=0 @@ -946,28 +803,33 @@ test_27a() { start_mds || return 2 echo "Requeue thread should have started: " ps -e | grep ll_cfg_requeue - set_and_check ost1 "cat $LPROC/obdfilter/$FSNAME-OST0000/client_cache_seconds" "$FSNAME-OST0000.ost.client_cache_seconds" || return 3 + set_and_check ost1 "lctl get_param -n obdfilter.$FSNAME-OST0000.client_cache_seconds" "$FSNAME-OST0000.ost.client_cache_seconds" || return 3 cleanup_nocli } run_test 27a "Reacquire MGS lock if OST started first" test_27b() { + # FIXME. ~grev setup - facet_failover mds - set_and_check mds "cat $LPROC/mdt/$FSNAME-MDT0000/identity_acquire_expire" "$FSNAME-MDT0000.mdt.identity_acquire_expire" || return 3 - set_and_check client "cat $LPROC/mdc/$FSNAME-MDT0000-mdc-*/max_rpcs_in_flight" "$FSNAME-MDT0000.mdc.max_rpcs_in_flight" || return 4 + local device=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }') + + facet_failover $SINGLEMDS + set_and_check $SINGLEMDS "lctl get_param -n mdt.$device.identity_acquire_expire" "$device.mdt.identity_acquire_expire" || return 3 + set_and_check client "lctl get_param -n mdc.$device-mdc-*.max_rpcs_in_flight" "$device.mdc.max_rpcs_in_flight" || return 4 + check_mount cleanup } run_test 27b "Reacquire MGS lock after failover" test_28() { setup - TEST="cat $LPROC/llite/$FSNAME-*/max_read_ahead_whole_mb" - ORIG=$($TEST) - declare -i FINAL - FINAL=$(($ORIG + 10)) - set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" || return 3 - set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" || return 3 + TEST="lctl get_param -n llite.$FSNAME-*.max_read_ahead_whole_mb" + PARAM="$FSNAME.llite.max_read_ahead_whole_mb" + ORIG=$($TEST) + FINAL=$(($ORIG + 1)) + set_and_check client "$TEST" "$PARAM" $FINAL || return 3 + FINAL=$(($FINAL + 1)) + set_and_check client "$TEST" "$PARAM" $FINAL || return 4 umount_client $MOUNT || return 200 mount_client $MOUNT RESULT=$($TEST) @@ -977,6 +839,7 @@ test_28() { else echo "New config success: got $RESULT" fi + set_and_check client "$TEST" "$PARAM" $ORIG || return 5 cleanup } run_test 28 "permanent parameter setting" @@ -988,32 +851,28 @@ test_29() { sleep 10 local PARAM="$FSNAME-OST0001.osc.active" - local PROC_ACT="$LPROC/osc/$FSNAME-OST0001-osc-[^M]*/active" - local PROC_UUID="$LPROC/osc/$FSNAME-OST0001-osc-[^M]*/ost_server_uuid" - if [ ! -r $PROC_ACT ]; then - echo "Can't read $PROC_ACT" - ls $LPROC/osc/$FSNAME-* - return 1 - fi - ACTV=$(cat $PROC_ACT) + local PROC_ACT="osc.$FSNAME-OST0001-osc-[^M]*.active" + local PROC_UUID="osc.$FSNAME-OST0001-osc-[^M]*.ost_server_uuid" + + ACTV=$(lctl get_param -n $PROC_ACT) DEAC=$((1 - $ACTV)) - set_and_check client "cat $PROC_ACT" "$PARAM" $DEAC || return 2 + set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $DEAC || return 2 # also check ost_server_uuid status - RESULT=$(grep DEACTIV $PROC_UUID) + RESULT=$(lctl get_param -n $PROC_UUID | grep DEACTIV) if [ -z "$RESULT" ]; then - echo "Live client not deactivated: $(cat $PROC_UUID)" + echo "Live client not deactivated: $(lctl get_param -n $PROC_UUID)" return 3 else echo "Live client success: got $RESULT" fi # check MDT too - local MPROC="$LPROC/osc/$FSNAME-OST0001-osc-[M]*/active" + local MPROC="osc.$FSNAME-OST0001-osc-[M]*.active" local MAX=30 local WAIT=0 while [ 1 ]; do sleep 5 - RESULT=`do_facet mds " [ -r $MPROC ] && cat $MPROC"` + RESULT=`do_facet $SINGLEMDS " lctl get_param -n $MPROC"` [ ${PIPESTATUS[0]} = 0 ] || error "Can't read $MPROC" if [ $RESULT -eq $DEAC ]; then echo "MDT deactivated also after $WAIT sec (got $RESULT)" @@ -1030,33 +889,35 @@ test_29() { # test new client starts deactivated umount_client $MOUNT || return 200 mount_client $MOUNT - RESULT=$(grep DEACTIV $PROC_UUID | grep NEW) + RESULT=$(lctl get_param -n $PROC_UUID | grep DEACTIV | grep NEW) if [ -z "$RESULT" ]; then - echo "New client not deactivated from start: $(cat $PROC_UUID)" + echo "New client not deactivated from start: $(lctl get_param -n $PROC_UUID)" return 5 else echo "New client success: got $RESULT" fi # make sure it reactivates - set_and_check client "cat $PROC_ACT" "$PARAM" $ACTV || return 6 + set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $ACTV || return 6 umount_client $MOUNT stop_ost2 cleanup_nocli #writeconf to remove all ost2 traces for subsequent tests writeconf + start_mds + start_ost + cleanup } run_test 29 "permanently remove an OST" test_30() { - # start mds first after writeconf - start_mds - start_ost - mount_client $MOUNT - TEST="cat $LPROC/llite/$FSNAME-*/max_read_ahead_whole_mb" - ORIG=$($TEST) - for i in $(seq 1 20); do + setup + + TEST="lctl get_param -n llite.$FSNAME-*.max_read_ahead_whole_mb" + ORIG=$($TEST) + LIST=(1 2 3 4 5 4 3 2 1 2 3 4 5 4 3 2 1 2 3 4 5) + for i in ${LIST[@]}; do set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $i || return 3 done # make sure client restart still works @@ -1076,28 +937,38 @@ test_31() { # bug 10734 run_test 31 "Connect to non-existent node (shouldn't crash)" test_32a() { - # XXX - make this run on client-only systems with real hardware on - # the OST and MDT - # there appears to be a lot of assumption here about loopback - # devices - # or maybe this test is just totally useless on a client-only system - [ -z "$TUNEFS" ] && skip "No tunefs" && return - local DISK1_4=$LUSTRE/tests/disk1_4.zip - [ ! -r $DISK1_4 ] && skip "Cant find $DISK1_4, skipping" && return - unzip -o -j -d $TMP/$tdir $DISK1_4 || { skip "Cant unzip $DISK1_4, skipping" && return ; } + # XXX - make this test verify 1.8 -> 2.0 upgrade is working + # XXX - make this run on client-only systems with real hardware on + # the OST and MDT + # there appears to be a lot of assumption here about loopback + # devices + # or maybe this test is just totally useless on a client-only system + [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; } + [ "$mds_HOST" = "`hostname`" ] || { skip "remote MDS" && return 0; } + [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \ + { skip "remote OST" && return 0; } + + [ -z "$TUNEFS" ] && skip "No tunefs" && return + local DISK1_8=$LUSTRE/tests/disk1_8.tgz + [ ! -r $DISK1_8 ] && skip "Cannot find $DISK1_8" && return 0 + + mkdir -p $TMP/$tdir + tar xjvf $DISK1_8 -C $TMP/$tdir || \ + { skip "Cannot untar $DISK1_8" && return 0; } + load_modules - sysctl lnet.debug=$PTLDEBUG + lctl set_param debug=$PTLDEBUG - $TUNEFS $TMP/$tdir/mds || error "tunefs failed" + $TUNEFS $tmpdir/mds || error "tunefs failed" # nids are wrong, so client wont work, but server should start - start mds $TMP/$tdir/mds "-o loop,exclude=lustre-OST0000" || return 3 - local UUID=$(cat $LPROC/mds/lustre-MDT0000/uuid) + start mds $tmpdir/mds "-o loop,exclude=lustre-OST0000" || return 3 + local UUID=$(lctl get_param -n mdt.lustre-MDT0000.uuid) echo MDS uuid $UUID [ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID" - $TUNEFS --mgsnode=`hostname` $TMP/$tdir/ost1 || error "tunefs failed" - start ost1 $TMP/$tdir/ost1 "-o loop" || return 5 - UUID=$(cat $LPROC/obdfilter/lustre-OST0000/uuid) + $TUNEFS --mgsnode=`hostname` $tmpdir/ost1 || error "tunefs failed" + start ost1 $tmpdir/ost1 "-o loop" || return 5 + UUID=$(lctl get_param -n obdfilter.lustre-OST0000.uuid) echo OST uuid $UUID [ "$UUID" == "ost1_UUID" ] || error "UUID is wrong: $UUID" @@ -1114,112 +985,198 @@ test_32a() { # With a new good MDT failover nid, we should be able to mount a client # (but it cant talk to OST) - local OLDMOUNTOPT=$MOUNTOPT - MOUNTOPT="exclude=lustre-OST0000" + local OLDMOUNTOPT=$MOUNTOPT + MOUNTOPT="exclude=lustre-OST0000" mount_client $MOUNT MOUNTOPT=$OLDMOUNTOPT - set_and_check client "cat $LPROC/mdc/*/max_rpcs_in_flight" "lustre-MDT0000.mdc.max_rpcs_in_flight" || return 11 + set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "lustre-MDT0000.mdc.max_rpcs_in_flight" || + return 11 zconf_umount `hostname` $MOUNT -f cleanup_nocli + load_modules # mount a second time to make sure we didnt leave upgrade flag on - $TUNEFS --dryrun $TMP/$tdir/mds || error "tunefs failed" - start mds $TMP/$tdir/mds "-o loop,exclude=lustre-OST0000" || return 12 - cleanup_nocli + load_modules + $TUNEFS --dryrun $tmpdir/mds || error "tunefs failed" + load_modules + start mds $tmpdir/mds "-o loop,exclude=lustre-OST0000" || return 12 + cleanup_nocli - [ -d $TMP/$tdir ] && rm -rf $TMP/$tdir + rm -rf $tmpdir || true # true is only for TMP on NFS } -run_test 32a "Upgrade from 1.4 (not live)" +run_test 32a "Upgrade from 1.8 (not live)" test_32b() { - # XXX - make this run on client-only systems with real hardware on - # the OST and MDT - # there appears to be a lot of assumption here about loopback - # devices - # or maybe this test is just totally useless on a client-only system - [ -z "$TUNEFS" ] && skip "No tunefs" && return - local DISK1_4=$LUSTRE/tests/disk1_4.zip - [ ! -r $DISK1_4 ] && skip "Cant find $DISK1_4, skipping" && return - unzip -o -j -d $TMP/$tdir $DISK1_4 || { skip "Cant unzip $DISK1_4, skipping" && return ; } + # XXX - make this test verify 1.8 -> 2.0 upgrade is working + # XXX - make this run on client-only systems with real hardware on + # the OST and MDT + # there appears to be a lot of assumption here about loopback + # devices + # or maybe this test is just totally useless on a client-only system + [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; } + [ "$mds_HOST" = "`hostname`" ] || { skip "remote MDS" && return 0; } + [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \ + { skip "remote OST" && return 0; } + + [ -z "$TUNEFS" ] && skip "No tunefs" && return + local DISK1_8=$LUSTRE/tests/disk1_8.tgz + [ ! -r $DISK1_8 ] && skip "Cannot find $DISK1_8" && return 0 + local tmpdir=$TMP/$tdir + mkdir -p $tmpdir + tar xjvf $DISK1_8 -C $tmpdir || \ + { skip "Cannot untar $DISK1_8" && return ; } + load_modules - sysctl lnet.debug=$PTLDEBUG + lctl set_param debug=$PTLDEBUG + NEWNAME=lustre # writeconf will cause servers to register with their current nids - $TUNEFS --writeconf $TMP/$tdir/mds || error "tunefs failed" - start mds $TMP/$tdir/mds "-o loop" || return 3 - local UUID=$(cat $LPROC/mds/lustre-MDT0000/uuid) + $TUNEFS --writeconf --fsname=$NEWNAME $tmpdir/mds || error "tunefs failed" + start mds1 $tmpdir/mds "-o loop" || return 3 + local UUID=$(lctl get_param -n mdt.${NEWNAME}-MDT0000.uuid) echo MDS uuid $UUID - [ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID" + [ "$UUID" == "${NEWNAME}-MDT0000_UUID" ] || error "UUID is wrong: $UUID" - $TUNEFS --mgsnode=`hostname` $TMP/$tdir/ost1 || error "tunefs failed" - start ost1 $TMP/$tdir/ost1 "-o loop" || return 5 - UUID=$(cat $LPROC/obdfilter/lustre-OST0000/uuid) + $TUNEFS --mgsnode=`hostname` --writeconf --fsname=$NEWNAME $tmpdir/ost1 || error "tunefs failed" + start ost1 $tmpdir/ost1 "-o loop" || return 5 + UUID=$(lctl get_param -n obdfilter.${NEWNAME}-OST0000.uuid) echo OST uuid $UUID - [ "$UUID" == "ost1_UUID" ] || error "UUID is wrong: $UUID" + [ "$UUID" == "${NEWNAME}-OST0000_UUID" ] || error "UUID is wrong: $UUID" echo "OSC changes should succeed:" - $LCTL conf_param lustre-OST0000.osc.max_dirty_mb=15 || return 7 - $LCTL conf_param lustre-OST0000.failover.node=$NID || return 8 + $LCTL conf_param ${NEWNAME}-OST0000.osc.max_dirty_mb=15 || return 7 + $LCTL conf_param ${NEWNAME}-OST0000.failover.node=$NID || return 8 echo "ok." echo "MDC changes should succeed:" - $LCTL conf_param lustre-MDT0000.mdc.max_rpcs_in_flight=9 || return 9 + $LCTL conf_param ${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight=9 || return 9 echo "ok." # MDT and OST should have registered with new nids, so we should have # a fully-functioning client echo "Check client and old fs contents" + OLDFS=$FSNAME + FSNAME=$NEWNAME mount_client $MOUNT - set_and_check client "cat $LPROC/mdc/*/max_rpcs_in_flight" "${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight" || return 11 - [ "$(cksum $MOUNT/passwd | cut -d' ' -f 1,2)" == "2479747619 779" ] || return 12 + FSNAME=$OLDFS + set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight" || return 11 + [ "$(cksum $MOUNT/passwd | cut -d' ' -f 1,2)" == "94306271 1478" ] || return 12 echo "ok." cleanup - [ -d $TMP/$tdir ] && rm -rf $TMP/$tdir + rm -rf $tmpdir || true # true is only for TMP on NFS } -run_test 32b "Upgrade from 1.4 with writeconf" +run_test 32b "Upgrade from 1.8 with writeconf" -test_33() { # bug 12333 - local FSNAME2=test1234 - local fs2mds_HOST=$mds_HOST - local fs2ost_HOST=$ost_HOST +test_33a() { # bug 12333, was test_33 + local rc=0 + local FSNAME2=test-123 + [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST - [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ] && [ -b "$MDSDEV" ] && \ - log "mixed loopback and real device not working" && return + if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then + do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + skip "mixed loopback and real device not working" && return + fi local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2} add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10 - add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --index=8191 --mgsnode=`hostname`@tcp --reformat $fs2ostdev || exit 10 + add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --index=8191 --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10 - start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS + start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT start fs2ost $fs2ostdev $OST_MOUNT_OPTS + do_facet $SINGLEMDS "$LCTL conf_param $FSNAME2.sys.timeout=200" || rc=1 mkdir -p $MOUNT2 - mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1 + mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || rc=2 echo "ok." + cp /etc/hosts $MOUNT2/ || rc=3 + $LFS getstripe $MOUNT2/hosts + umount -d $MOUNT2 stop fs2ost -f stop fs2mds -f rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev - cleanup_nocli || return 6 + cleanup_nocli || rc=6 + return $rc +} +run_test 33a "Mount ost with a large index number" + +test_33b() { # was test_34 + setup + + do_facet client dd if=/dev/zero of=$MOUNT/24 bs=1024k count=1 + # Drop lock cancelation reply during umount + #define OBD_FAIL_LDLM_CANCEL 0x304 + do_facet client lctl set_param fail_loc=0x80000304 + #lctl set_param debug=-1 + umount_client $MOUNT + cleanup +} +run_test 33b "Drop cancel during umount" + +test_34a() { + setup + do_facet client "sh runmultiop_bg_pause $DIR/file O_c" + manual_umount_client + rc=$? + do_facet client killall -USR1 multiop + if [ $rc -eq 0 ]; then + error "umount not fail!" + fi + sleep 1 + cleanup } -run_test 33 "Mount ost with a large index number" +run_test 34a "umount with opened file should be fail" + + +test_34b() { + setup + touch $DIR/$tfile || return 1 + stop_mds --force || return 2 + + manual_umount_client --force + rc=$? + if [ $rc -ne 0 ]; then + error "mtab after failed umount - rc $rc" + fi + + cleanup + return 0 +} +run_test 34b "force umount with failed mds should be normal" + +test_34c() { + setup + touch $DIR/$tfile || return 1 + stop_ost --force || return 2 + + manual_umount_client --force + rc=$? + if [ $rc -ne 0 ]; then + error "mtab after failed umount - rc $rc" + fi + + cleanup + return 0 +} +run_test 34c "force umount with failed ost should be normal" test_35() { # bug 12459 setup - DBG_SAVE="`sysctl -n lnet.debug`" - sysctl -w lnet.debug="ha" + DBG_SAVE="`lctl get_param -n debug`" + lctl set_param debug="ha" log "Set up a fake failnode for the MDS" FAKENID="127.0.0.2" - do_facet mds $LCTL conf_param ${FSNAME}-MDT0000.failover.node=$FAKENID || return 4 + local device=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1) + do_facet $SINGLEMDS $LCTL conf_param ${device}.failover.node=$FAKENID || return 4 log "Wait for RECONNECT_INTERVAL seconds (10s)" sleep 10 - MSG="conf-sanity.sh test_33 `date +%F%kh%Mm%Ss`" + MSG="conf-sanity.sh test_35 `date +%F%kh%Mm%Ss`" $LCTL clear log "$MSG" log "Stopping the MDT:" @@ -1232,13 +1189,13 @@ test_35() { # bug 12459 log "Wait for df ($DFPID) ... " wait $DFPID log "done" - sysctl -w lnet.debug="$DBG_SAVE" + lctl set_param debug="$DBG_SAVE" # retrieve from the log the first server that the client tried to # contact after the connection loss $LCTL dk $TMP/lustre-log-$TESTNAME.log NEXTCONN=`awk "/${MSG}/ {start = 1;} - /import_select_connection.*${FSNAME}-MDT0000-mdc.* using connection/ { + /import_select_connection.$device-mdc.* using connection/ { if (start) { if (\\\$NF ~ /$FAKENID/) print \\\$NF; @@ -1255,22 +1212,28 @@ run_test 35 "Reconnect to the last active server first" test_36() { # 12743 local rc local FSNAME2=test1234 - local fs2mds_HOST=$mds_HOST - local fs2ost_HOST=$ost_HOST local fs3ost_HOST=$ost_HOST - rc=0 - [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ] && [ -b "$MDSDEV" ] && \ - log "mixed loopback and real device not working" && return + [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST && fs3ost_HOST=$ost1_HOST + rc=0 + if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" -o -z "$fs3ost_DEV" ]; then + do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + skip "mixed loopback and real device not working" && return + fi [ $OSTCOUNT -lt 2 ] && skip "skipping test for single OST" && return + [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \ + { skip "remote OST" && return 0; } + local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2} local fs3ostdev=${fs3ost_DEV:-$(ostdevname 2)_2} add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10 - add fs2ost $OST_MKFS_OPTS --mkfsoptions='-b1024' --fsname=${FSNAME2} --mgsnode=`hostname`@tcp --reformat $fs2ostdev || exit 10 - add fs3ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=`hostname`@tcp --reformat $fs3ostdev || exit 10 + # XXX after we support non 4K disk blocksize, change following --mkfsoptions with + # other argument + add fs2ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10 + add fs3ost $OST_MKFS_OPTS --mkfsoptions='-b4096' --fsname=${FSNAME2} --mgsnode=$MGSNID --reformat $fs3ostdev || exit 10 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS start fs2ost $fs2ostdev $OST_MOUNT_OPTS @@ -1278,15 +1241,13 @@ test_36() { # 12743 mkdir -p $MOUNT2 mount -t lustre $MGSNID:/${FSNAME2} $MOUNT2 || return 1 + sleep 5 # until 11778 fixed + dd if=/dev/zero of=$MOUNT2/$tfile bs=1M count=7 || return 2 - [ $(grep -c obdfilter $LPROC/devices) -eq 0 ] && - skip "skipping test for remote OST" && return - BKTOTAL=`awk 'BEGIN{total=0}; {total+=$1}; END{print total}' \ - $LPROC/obdfilter/*/kbytestotal` - BKFREE=`awk 'BEGIN{free=0}; {free+=$1}; END{print free}' \ - $LPROC/obdfilter/*/kbytesfree` - BKAVAIL=`awk 'BEGIN{avail=0}; {avail+=$1}; END{print avail}' \ - $LPROC/obdfilter/*/kbytesavail` + + BKTOTAL=`lctl get_param -n obdfilter.*.kbytestotal | awk 'BEGIN{total=0}; {total+=$1}; END{print total}'` + BKFREE=`lctl get_param -n obdfilter.*.kbytesfree | awk 'BEGIN{free=0}; {free+=$1}; END{print free}'` + BKAVAIL=`lctl get_param -n obdfilter.*.kbytesavail | awk 'BEGIN{avail=0}; {avail+=$1}; END{print avail}'` STRING=`df -P $MOUNT2 | tail -n 1 | awk '{print $2","$3","$4}'` DFTOTAL=`echo $STRING | cut -d, -f1` DFUSED=`echo $STRING | cut -d, -f2` @@ -1321,9 +1282,168 @@ test_36() { # 12743 } run_test 36 "df report consistency on OSTs with different block size" -umount_client $MOUNT +test_37() { + [ -n "$CLIENTONLY" -o -n "$CLIENTMODSONLY" ] && skip "client only testing" && return 0 + LOCAL_MDSDEV="$TMP/mdt.img" + SYM_MDSDEV="$TMP/sym_mdt.img" + + echo "MDS : $LOCAL_MDSDEV" + echo "SYMLINK : $SYM_MDSDEV" + rm -f $LOCAL_MDSDEV + + touch $LOCAL_MDSDEV + mkfs.lustre --reformat --fsname=lustre --mdt --mgs --device-size=9000 $LOCAL_MDSDEV || + error "mkfs.lustre $LOCAL_MDSDEV failed" + ln -s $LOCAL_MDSDEV $SYM_MDSDEV + + echo "mount symlink device - $SYM_MDSDEV" + + mount_op=`mount -v -t lustre -o loop $SYM_MDSDEV ${MOUNT%/*}/mds 2>&1 | grep "unable to set tunable"` + umount -d ${MOUNT%/*}/mds + rm -f $LOCAL_MDSDEV $SYM_MDSDEV + + if [ -n "$mount_op" ]; then + error "**** FAIL: set tunables failed for symlink device" + fi + return 0 +} +run_test 37 "verify set tunables works for symlink device" + +test_38() { # bug 14222 + setup + # like runtests + COUNT=10 + SRC="/etc /bin" + FILES=`find $SRC -type f -mtime +1 | head -n $COUNT` + log "copying $(echo $FILES | wc -w) files to $DIR/$tdir" + mkdir -p $DIR/$tdir + tar cf - $FILES | tar xf - -C $DIR/$tdir || \ + error "copying $SRC to $DIR/$tdir" + sync + umount_client $MOUNT + stop_mds + log "rename lov_objid file on MDS" + rm -f $TMP/lov_objid.orig + + local dev=${SINGLEMDS}_dev + local MDSDEV=${!dev} + do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.orig\\\" $MDSDEV" + do_facet $SINGLEMDS "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV" + + do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.orig" + # check create in mds_lov_connect + start_mds + mount_client $MOUNT + for f in $FILES; do + [ $V ] && log "verifying $DIR/$tdir/$f" + diff -q $f $DIR/$tdir/$f || ERROR=y + done + do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new\\\" $MDSDEV" + do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.new" + [ "$ERROR" = "y" ] && error "old and new files are different after connect" || true + + # check it's updates in sync + umount_client $MOUNT + stop_mds + + do_facet $SINGLEMDS dd if=/dev/zero of=$TMP/lov_objid.clear bs=4096 count=1 + do_facet $SINGLEMDS "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV" + do_facet $SINGLEMDS "debugfs -w -R \\\"write $TMP/lov_objid.clear lov_objid\\\" $MDSDEV " + + start_mds + mount_client $MOUNT + for f in $FILES; do + [ $V ] && log "verifying $DIR/$tdir/$f" + diff -q $f $DIR/$tdir/$f || ERROR=y + done + do_facet $SINGLEMDS "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new1\\\" $MDSDEV" + do_facet $SINGLEMDS "od -Ax -td8 $TMP/lov_objid.new1" + umount_client $MOUNT + stop_mds + [ "$ERROR" = "y" ] && error "old and new files are different after sync" || true + + log "files compared the same" + cleanup +} +run_test 38 "MDS recreates missing lov_objid file from OST data" + +test_39() { + PTLDEBUG=+malloc + setup + cleanup + perl $SRCDIR/leak_finder.pl $TMP/debug 2>&1 | egrep '*** Leak:' && + error "memory leak detected" || true +} +run_test 39 "leak_finder recognizes both LUSTRE and LNET malloc messages" + +test_40() { # bug 15759 + start_ost + #define OBD_FAIL_TGT_TOOMANY_THREADS 0x706 + do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000706" + start_mds + cleanup +} +run_test 40 "race during service thread startup" + +test_41() { #bug 14134 + local rc + local dev=${SINGLEMDS}_dev + local MDSDEV=${!dev} + + start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nosvc -n + start ost1 `ostdevname 1` $OST_MOUNT_OPTS + start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nomgs + mkdir -p $MOUNT + mount_client $MOUNT || return 1 + sleep 5 + + echo "blah blah" > $MOUNT/$tfile + cat $MOUNT/$tfile + + umount_client $MOUNT + stop ost1 -f || return 201 + stop_mds -f || return 202 + stop_mds -f || return 203 + unload_modules || return 204 + return $rc +} +run_test 41 "mount mds with --nosvc and --nomgs" + +test_42() { #bug 14693 + setup + check_mount || return 2 + do_facet client lctl conf_param lustre.llite.some_wrong_param=10 + umount_client $MOUNT + mount_client $MOUNT || return 1 + cleanup + return 0 +} +run_test 42 "invalid config param should not prevent client from mounting" + +umount_client $MOUNT cleanup_nocli -cleanup_krb5_env +cleanup_gss + +test_45() { #17310 + setup + check_mount || return 2 + stop_mds + df -h $MOUNT & + log "sleep 60 sec" + sleep 60 +#define OBD_FAIL_PTLRPC_LONG_UNLINK 0x50f + do_facet client "lctl set_param fail_loc=0x50f" + log "sleep 10 sec" + sleep 10 + manual_umount_client --force || return 3 + do_facet client "lctl set_param fail_loc=0x0" + start_mds + mount_client $MOUNT || return 4 + cleanup + return 0 +} +run_test 45 "long unlink handling in ptlrpcd" + equals_msg `basename $0`: test complete [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true