X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fconf-sanity.sh;h=1281d3ffdd789bd8aeb3e71eb03de55672c0cf96;hb=467cf22b99e0bb90635bc35b42ce7dcb9b051b61;hp=d16558456468a14ec287fc29d7c26de2a821d4c3;hpb=0fe560ea5dd4e582f2393f54de106e34a8b2c6ae;p=fs%2Flustre-release.git diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh index d165584..1281d3f 100644 --- a/lustre/tests/conf-sanity.sh +++ b/lustre/tests/conf-sanity.sh @@ -1,6 +1,6 @@ #!/bin/bash -# requirement: -# add uml1 uml2 uml3 in your /etc/hosts +# -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*- +# vim:autoindent:shiftwidth=4:tabstop=4: # FIXME - there is no reason to use all of these different # return codes, espcially when most of them are mapped to something @@ -11,13 +11,9 @@ set -e ONLY=${ONLY:-"$*"} -# These tests don't apply to mountconf -MOUNTCONFSKIP="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 $HEAD_EXCEPT" +# bug number for skipped test: +# 15977 +ALWAYS_EXCEPT="$CONF_SANITY_EXCEPT" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! SRCDIR=`dirname $0` @@ -27,53 +23,70 @@ PTLDEBUG=${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 $@ +init_logging +# STORED_MDSSIZE is used in test_18 +if [ -n "$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/$NAME.sh} -# -[ "$SLOW" = "no" ] && EXCEPT_SLOW="0 1 2 3 6 7 15 18 24b 25 30 31 32 33 34a " +require_dsh_mds || exit 0 +require_dsh_ost || 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 - 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 $MDSDEV" || echo "tunefs failed, reformatting instead" && reformat + local facet=$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 ${!dev}" || echo "tunefs failed, reformatting instead" && reformat + + gen_config } gen_config() { - reformat - # The MGS must be started before the OSTs for a new fs, so start - # and stop to generate the startup logs. + # The MGS must be started before the OSTs for a new fs, so start + # and stop to generate the startup logs. start_mds start_ost - sleep 5 + wait_osc_import_state mds ost FULL stop_ost stop_mds } +reformat_and_config() { + reformat + 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() { @@ -120,8 +133,12 @@ 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() { @@ -130,6 +147,12 @@ setup() { mount_client $MOUNT } +setup_noconfig() { + start_mds + start_ost + mount_client $MOUNT +} + cleanup_nocli() { stop_mds || return 201 stop_ost || return 202 @@ -144,17 +167,17 @@ cleanup() { check_mount() { do_facet client "cp /etc/passwd $DIR/a" || return 71 do_facet client "rm $DIR/a" || return 72 - # make sure lustre is actually mounted (touch will block, - # but grep won't, so do it after) + # make sure lustre is actually mounted (touch will block, + # but grep won't, so do it after) do_facet client "grep $MOUNT' ' /proc/mounts > /dev/null" || return 73 echo "setup single mount lustre success" } check_mount2() { - do_facet client "touch $DIR/a" || return 71 - do_facet client "rm $DIR/a" || return 72 - do_facet client "touch $DIR2/a" || return 73 - do_facet client "rm $DIR2/a" || return 74 + do_facet client "touch $DIR/a" || return 71 + do_facet client "rm $DIR/a" || return 72 + do_facet client "touch $DIR2/a" || return 73 + do_facet client "rm $DIR2/a" || return 74 echo "setup double mount lustre success" } @@ -170,11 +193,11 @@ if [ "$ONLY" == "cleanup" ]; then exit fi -#create single point mountpoint +init_gss -gen_config +#create single point mountpoint -init_gss +reformat_and_config test_0() { setup @@ -194,7 +217,7 @@ 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 @@ -281,7 +304,7 @@ 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 $? return 0 @@ -361,7 +384,9 @@ test_9() { do_facet ost1 lctl set_param subsystem_debug=\'mds ost\' || return 1 CHECK_PTLDEBUG="`do_facet ost1 lctl get_param -n debug`" - if [ "$CHECK_PTLDEBUG" ] && [ "$CHECK_PTLDEBUG" = "trace inode" ];then + if [ "$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 '$CHECK_PTLDEBUG'" @@ -379,263 +404,11 @@ test_9() { run_test 9 "test ptldebug and subsystem for mkfs" -test_10() { - echo "generate configuration with the same name for node and mds" - OLDXMLCONFIG=$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 $? -} -run_test 15 "zconf-mount without /sbin/mount.lustre (should return error)" - # LOGS/PENDING do not exist anymore since CMD3 test_16() { - TMPMTPT="${TMP}/conf16" - + local TMPMTPT="${TMP}/conf16" + local dev=${SINGLEMDS}_dev + local MDSDEV=${!dev} if [ ! -e "$MDSDEV" ]; then log "no $MDSDEV existing, so mount Lustre to create one" setup @@ -646,7 +419,7 @@ test_16() { [ -f "$MDSDEV" ] && LOOPOPT="-o loop" log "change the mode of $MDSDEV/OBJECTS to 555" - do_facet mds "mkdir -p $TMPMTPT && + do_facet $SINGLEMDS "mkdir -p $TMPMTPT && mount $LOOPOPT -t $FSTYPE $MDSDEV $TMPMTPT && chmod 555 $TMPMTPT/OBJECTS && umount $TMPMTPT" || return $? @@ -657,7 +430,7 @@ test_16() { cleanup || return $? log "read the mode of OBJECTS 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:.*//"` + 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" @@ -668,6 +441,9 @@ test_16() { 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 $MDSDEV existing, so mount Lustre to create one" setup @@ -676,29 +452,60 @@ test_17() { fi echo "Remove mds config log" - do_facet mds "debugfs -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MDSDEV || return \$?" || return $? + do_facet $SINGLEMDS "$DEBUGFS -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MDSDEV || return \$?" || return $? start_ost start_mds && return 42 - gen_config + reformat_and_config } 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 "$OK" ] && skip_env "$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" - gen_config + reformat_and_config echo "mount lustre system..." setup 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 @@ -708,7 +515,7 @@ test_18() { cleanup || return $? MDS_MKFS_OPTS=$OLD_MDS_MKFS_OPTS - gen_config + reformat_and_config } run_test 18 "check mkfs creates large journals" @@ -748,6 +555,7 @@ run_test 20 "remount ro,rw mounts work and doesn't break /etc/mtab" test_21a() { start_mds start_ost + wait_osc_import_state mds ost FULL stop_ost stop_mds } @@ -756,6 +564,7 @@ run_test 21a "start mds before ost, stop ost first" test_21b() { start_ost start_mds + wait_osc_import_state mds ost FULL stop_mds stop_ost } @@ -765,26 +574,27 @@ test_21c() { start_ost start_mds start_ost2 + wait_osc_import_state mds ost2 FULL 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 - check_mount && return 41 - pass echo Client mount with ost in logs, but none running start_ost + # wait until mds connected to ost and open client connection + wait_osc_import_state mds ost FULL stop_ost mount_client $MOUNT # check_mount will block trying to contact ost + mcreate $DIR/$tfile || return 40 + rm -f $DIR/$tfile || return 42 umount_client $MOUNT pass @@ -809,7 +619,7 @@ run_test 22 "start a client before osts (should return errs)" test_23a() { # was test_23 setup # fail mds - stop mds + stop $SINGLEMDS # force down client so that recovering mds waits for reconnect local running=$(grep -c $MOUNT /proc/mounts) || true if [ $running -ne 0 ]; then @@ -847,8 +657,8 @@ test_23a() { # was test_23 echo "waiting for mount to finish ... " WAIT=$(( WAIT + sleep)) done - [ "$WAIT" -eq "$MAX_WAIT" ] && error "MOUNT_PID $MOUNT_PID and \ - MOUNT__LUSTRE_PID $MOUNT__LUSTRE_PID still not killed in $WAIT secs" + [ "$WAIT" -eq "$MAX_WAIT" ] && error "MOUNT_PID $MOUNT_PID and "\ + "MOUNT_LUSTRE_PID $MOUNT_LUSTRE_PID still not killed in $WAIT secs" ps -ef | grep mount stop_mds || error stop_ost || error @@ -882,13 +692,13 @@ cleanup_24a() { } test_24a() { - #set up fs1 + #set up fs1 gen_config #set up fs2 [ -n "$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 + skip_env "mixed loopback and real device not working" && return fi local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} @@ -917,14 +727,14 @@ test_24a() { rm $MOUNT2/b || return 4 # 2 is actually mounted grep $MOUNT2' ' /proc/mounts > /dev/null || return 5 - # failover + # failover facet_failover fs2mds facet_failover fs2ost df - umount_client $MOUNT + umount_client $MOUNT # the MDS must remain up until last MDT stop_mds - MDS=$(do_facet $SINGLEMDS "lctl get_param -n 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 cleanup_24a cleanup_nocli || return 6 @@ -934,12 +744,12 @@ run_test 24a "Multiple MDTs on a single node" test_24b() { if [ -z "$fs2mds_DEV" ]; then do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ - skip "mixed loopback and real device not working" && return + skip_env "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 @@ -958,9 +768,9 @@ 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 "lctl set_param fail_loc=0x80000135" + do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000135" start_mds && echo MDS started && return 1 lctl get_param -n devices DEVS=$(lctl get_param -n devices | wc -l) @@ -973,7 +783,7 @@ set_and_check() { local myfacet=$1 local TEST=$2 local PARAM=$3 - local ORIG=$(do_facet $myfacet "$TEST") + local ORIG=$(do_facet $myfacet "$TEST") if [ $# -gt 3 ]; then local FINAL=$4 else @@ -981,41 +791,30 @@ set_and_check() { FINAL=$(($ORIG + 5)) fi echo "Setting $PARAM from $ORIG to $FINAL" - do_facet mds "$LCTL conf_param $PARAM=$FINAL" || error conf_param failed - local RESULT - local MAX=90 - local WAIT=0 - while [ 1 ]; do - sleep 5 - RESULT=$(do_facet $myfacet "$TEST") - if [ $RESULT -eq $FINAL ]; then - echo "Updated config after $WAIT sec (got $RESULT)" - break - fi - WAIT=$((WAIT + 5)) - if [ $WAIT -eq $MAX ]; then - echo "Config update not seen: wanted $FINAL got $RESULT" - return 3 - fi - echo "Waiting $(($MAX - $WAIT)) secs for config update" - done + do_facet $SINGLEMDS "$LCTL conf_param $PARAM='$FINAL'" || error conf_param failed + + wait_update $(facet_host $myfacet) "$TEST" "$FINAL" || error check failed! } test_27a() { start_ost || return 1 start_mds || return 2 - echo "Requeue thread should have started: " - ps -e | grep ll_cfg_requeue + echo "Requeue thread should have started: " + ps -e | grep ll_cfg_requeue set_and_check ost1 "lctl get_param -n obdfilter.$FSNAME-OST0000.client_cache_seconds" "$FSNAME-OST0000.ost.client_cache_seconds" || return 3 cleanup_nocli } run_test 27a "Reacquire MGS lock if OST started first" test_27b() { + # FIXME. ~grev setup - facet_failover mds - set_and_check mds "lctl get_param -n mdt.$FSNAME-MDT0000.identity_acquire_expire" "$FSNAME-MDT0000.mdt.identity_acquire_expire" || return 3 - set_and_check client "lctl get_param -n mdc.$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" "$device.mdc.max_rpcs_in_flight" || return 4 + check_mount cleanup } run_test 27b "Reacquire MGS lock after failover" @@ -1023,11 +822,12 @@ run_test 27b "Reacquire MGS lock after failover" test_28() { setup TEST="lctl get_param -n llite.$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 + 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) @@ -1037,12 +837,13 @@ test_28() { else echo "New config success: got $RESULT" fi + set_and_check client "$TEST" "$PARAM" $ORIG || return 5 cleanup } run_test 28 "permanent parameter setting" test_29() { - [ "$OSTCOUNT" -lt "2" ] && skip "$OSTCOUNT < 2, skipping" && return + [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2, skipping" && return setup > /dev/null 2>&1 start_ost2 sleep 10 @@ -1063,13 +864,13 @@ test_29() { echo "Live client success: got $RESULT" fi - # check MDT too + # check MDT too local MPROC="osc.$FSNAME-OST0001-osc-[M]*.active" local MAX=30 local WAIT=0 while [ 1 ]; do sleep 5 - RESULT=`do_facet mds " lctl get_param -n $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 $RESULT)" @@ -1102,9 +903,6 @@ test_29() { cleanup_nocli #writeconf to remove all ost2 traces for subsequent tests writeconf - start_mds - start_ost - cleanup } run_test 29 "permanently remove an OST" @@ -1112,14 +910,15 @@ test_30() { setup TEST="lctl get_param -n llite.$FSNAME-*.max_read_ahead_whole_mb" - ORIG=$($TEST) - for i in $(seq 1 20); do + ORIG=$($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 + # make sure client restart still works umount_client $MOUNT mount_client $MOUNT || return 4 - [ "$($TEST)" -ne "$i" ] && return 5 + [ "$($TEST)" -ne "$i" ] && return 5 set_and_check client "$TEST" "$FSNAME.llite.max_read_ahead_whole_mb" $ORIG || return 6 cleanup } @@ -1132,131 +931,181 @@ test_31() { # bug 10734 } run_test 31 "Connect to non-existent node (shouldn't crash)" +# Use these start32/stop32 fn instead of t-f start/stop fn, +# for local devices, to skip global facet vars init +stop32 () { + local facet=$1 + shift + echo "Stopping local ${MOUNT%/*}/${facet} (opts:$@)" + umount -d $@ ${MOUNT%/*}/${facet} + losetup -a +} + +start32 () { + local facet=$1 + shift + local device=$1 + shift + mkdir -p ${MOUNT%/*}/${facet} + + echo "Starting local ${facet}: $@ $device ${MOUNT%/*}/${facet}" + mount -t lustre $@ ${device} ${MOUNT%/*}/${facet} + local RC=$? + if [ $RC -ne 0 ]; then + echo "mount -t lustre $@ ${device} ${MOUNT%/*}/${facet}" + echo "Start of ${device} of local ${facet} failed ${RC}" + fi + losetup -a + return $RC +} + +cleanup_nocli32 () { + stop32 mds1 -f + stop32 ost1 -f + wait_exit_ST client +} + +cleanup_32() { + trap 0 + echo "Cleanup test_32 umount $MOUNT ..." + umount -f $MOUNT || true + echo "Cleanup local mds ost1 ..." + cleanup_nocli32 + unload_modules +} + test_32a() { - # XXX - make this run on client-only systems with real hardware on - # the OST and MDT - # there appears to be a lot of assumption here about loopback - # devices - # or maybe this test is just totally useless on a client-only system + client_only && skip "client only testing" && return 0 [ "$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_4=$LUSTRE/tests/disk1_4.zip - [ ! -r $DISK1_4 ] && skip "Cant find $DISK1_4, skipping" && return + [ -z "$TUNEFS" ] && skip_env "No tunefs" && return 0 + local DISK1_8=$LUSTRE/tests/disk1_8.tar.bz2 + [ ! -r $DISK1_8 ] && skip_env "Cannot find $DISK1_8" && return 0 local tmpdir=$TMP/conf32a - unzip -o -j -d $tmpdir $DISK1_4 || { skip "Cant unzip $DISK1_4, skipping" && return ; } + mkdir -p $tmpdir + tar xjvf $DISK1_8 -C $tmpdir || \ + { skip_env "Cannot untar $DISK1_8" && return 0; } + load_modules - lctl set_param debug=$PTLDEBUG + $LCTL set_param debug=$PTLDEBUG $TUNEFS $tmpdir/mds || error "tunefs failed" + # nids are wrong, so client wont work, but server should start - start mds $tmpdir/mds "-o loop,exclude=lustre-OST0000" || return 3 - local UUID=$(lctl get_param -n mdt.lustre-MDT0000.uuid) + start32 mds1 $tmpdir/mds "-o loop,exclude=lustre-OST0000" && \ + trap cleanup_32 EXIT INT || return 3 + + local UUID=$($LCTL get_param -n mdt.lustre-MDT0000.uuid) echo MDS uuid $UUID - [ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID" + [ "$UUID" == "lustre-MDT0000_UUID" ] || error "UUID is wrong: $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) + $TUNEFS --mgsnode=$HOSTNAME $tmpdir/ost1 || error "tunefs failed" + start32 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: $UUID" + [ "$UUID" == "lustre-OST0000_UUID" ] || error "UUID is wrong: $UUID" local NID=$($LCTL list_nids | head -1) - echo "OSC changes should return err:" - $LCTL conf_param lustre-OST0000.osc.max_dirty_mb=15 && return 7 - $LCTL conf_param lustre-OST0000.failover.node=$NID && return 8 + 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 echo "ok." - echo "MDC changes should succeed:" + + echo "MDC changes should succeed:" $LCTL conf_param lustre-MDT0000.mdc.max_rpcs_in_flight=9 || return 9 $LCTL conf_param lustre-MDT0000.failover.node=$NID || return 10 echo "ok." - # 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" - mount_client $MOUNT - MOUNTOPT=$OLDMOUNTOPT - set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "lustre-MDT0000.mdc.max_rpcs_in_flight" || - return 11 + echo "LOV changes should succeed:" + $LCTL pool_new lustre.interop || return 11 + $LCTL conf_param lustre-MDT0000.lov.stripesize=4M || return 12 + echo "ok." - zconf_umount `hostname` $MOUNT -f - cleanup_nocli - load_modules + cleanup_32 - # mount a second time to make sure we didnt leave upgrade flag on + # mount a second time to make sure we didnt leave upgrade flag on load_modules $TUNEFS --dryrun $tmpdir/mds || error "tunefs failed" - load_modules - start mds $tmpdir/mds "-o loop,exclude=lustre-OST0000" || return 12 - cleanup_nocli + start32 mds1 $tmpdir/mds "-o loop,exclude=lustre-OST0000" && \ + trap cleanup_32 EXIT INT || return 12 + + cleanup_32 - [ -d $tmpdir ] && rm -rf $tmpdir + rm -rf $tmpdir || true # true is only for TMP on NFS } -run_test 32a "Upgrade from 1.4 (not live)" +run_test 32a "Upgrade from 1.8 (not live)" test_32b() { - # XXX - make this run on client-only systems with real hardware on - # the OST and MDT - # there appears to be a lot of assumption here about loopback - # devices - # or maybe this test is just totally useless on a client-only system - [ "$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_4=$LUSTRE/tests/disk1_4.zip - [ ! -r $DISK1_4 ] && skip "Cant find $DISK1_4, skipping" && return + client_only && skip "client only testing" && return 0 + [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; } + [ -z "$TUNEFS" ] && skip_env "No tunefs" && return + local DISK1_8=$LUSTRE/tests/disk1_8.tar.bz2 + [ ! -r $DISK1_8 ] && skip_env "Cannot find $DISK1_8" && return 0 local tmpdir=$TMP/conf32b - unzip -o -j -d $tmpdir $DISK1_4 || { skip "Cant unzip $DISK1_4, skipping" && return ; } + mkdir -p $tmpdir + tar xjvf $DISK1_8 -C $tmpdir || \ + { skip_env "Cannot untar $DISK1_8" && return ; } + load_modules - lctl set_param debug=$PTLDEBUG - NEWNAME=sofia + $LCTL set_param debug="config" + local NEWNAME=lustre # writeconf will cause servers to register with their current nids $TUNEFS --writeconf --fsname=$NEWNAME $tmpdir/mds || error "tunefs failed" - start mds $tmpdir/mds "-o loop" || return 3 - local UUID=$(lctl get_param -n mdt.${NEWNAME}-MDT0000.uuid) + start32 mds1 $tmpdir/mds "-o loop" && \ + trap cleanup_32 EXIT INT || return 3 + + local UUID=$($LCTL get_param -n mdt.${NEWNAME}-MDT0000.uuid) echo MDS uuid $UUID - [ "$UUID" == "mdsA_UUID" ] || error "UUID is wrong: $UUID" + [ "$UUID" == "${NEWNAME}-MDT0000_UUID" ] || error "UUID is wrong: $UUID" - $TUNEFS --mgsnode=`hostname` --fsname=$NEWNAME --writeconf $tmpdir/ost1 || error "tunefs failed" - start ost1 $tmpdir/ost1 "-o loop" || return 5 - UUID=$(lctl get_param -n obdfilter.${NEWNAME}-OST0000.uuid) + $TUNEFS --mgsnode=$HOSTNAME --writeconf --fsname=$NEWNAME $tmpdir/ost1 ||\ + error "tunefs failed" + start32 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: $UUID" + + local NID=$($LCTL list_nids | head -1) - echo "OSC changes should succeed:" + echo "OSC changes should succeed:" $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:" + + echo "MDC changes should succeed:" $LCTL conf_param ${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight=9 || return 9 + $LCTL conf_param ${NEWNAME}-MDT0000.failover.node=$NID || return 10 + echo "ok." + + echo "LOV changes should succeed:" + $LCTL pool_new ${NEWNAME}.interop || return 11 + $LCTL conf_param ${NEWNAME}-MDT0000.lov.stripesize=4M || return 12 echo "ok." # MDT and OST should have registered with new nids, so we should have # a fully-functioning client echo "Check client and old fs contents" - OLDFS=$FSNAME - FSNAME=$NEWNAME - mount_client $MOUNT - FSNAME=$OLDFS - set_and_check client "lctl get_param -n mdc.*.max_rpcs_in_flight" "${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight" || return 11 - [ "$(cksum $MOUNT/passwd | cut -d' ' -f 1,2)" == "2479747619 779" ] || return 12 + + local device=`h2$NETTYPE $HOSTNAME`:/$NEWNAME + echo "Starting local client: $HOSTNAME: $device $MOUNT" + mount -t lustre $device $MOUNT || return 1 + + local old=$($LCTL get_param -n mdc.*.max_rpcs_in_flight) + local new=$((old + 5)) + $LCTL conf_param ${NEWNAME}-MDT0000.mdc.max_rpcs_in_flight=$new + wait_update $HOSTNAME "$LCTL get_param -n mdc.*.max_rpcs_in_flight" $new || return 11 + + [ "$(cksum $MOUNT/passwd | cut -d' ' -f 1,2)" == "94306271 1478" ] || return 12 echo "ok." - cleanup - [ -d $tmpdir ] && rm -rf $tmpdir + cleanup_32 + + rm -rf $tmpdir || true # true is only for TMP on NFS } -run_test 32b "Upgrade from 1.4 with writeconf" +run_test 32b "Upgrade from 1.8 with writeconf" test_33a() { # bug 12333, was test_33 local rc=0 @@ -1265,21 +1114,24 @@ test_33a() { # bug 12333, was test_33 if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ - skip "mixed loopback and real device not working" && return + skip_env "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 fs2mds $MDS_MKFS_OPTS --mkfsoptions='\"-J size=8\"' --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10 add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --index=8191 --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT start fs2ost $fs2ostdev $OST_MOUNT_OPTS - do_facet mds "$LCTL conf_param $FSNAME2.sys.timeout=200" || rc=1 + do_facet $SINGLEMDS "$LCTL conf_param $FSNAME2.sys.timeout=200" || rc=1 mkdir -p $MOUNT2 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 @@ -1329,7 +1181,7 @@ test_34b() { fi cleanup - return 0 + return 0 } run_test 34b "force umount with failed mds should be normal" @@ -1345,11 +1197,11 @@ test_34c() { fi cleanup - return 0 + return 0 } run_test 34c "force umount with failed ost should be normal" -test_35() { # bug 12459 +test_35a() { # bug 12459 setup DBG_SAVE="`lctl get_param -n debug`" @@ -1357,12 +1209,13 @@ test_35() { # bug 12459 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_35 `date +%F%kh%Mm%Ss`" + MSG="conf-sanity.sh test_35a `date +%F%kh%Mm%Ss`" $LCTL clear log "$MSG" log "Stopping the MDT:" @@ -1381,7 +1234,7 @@ test_35() { # bug 12459 # 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; @@ -1393,7 +1246,74 @@ test_35() { # bug 12459 [ "$NEXTCONN" != "0" ] && log "The client didn't try to reconnect to the last active server (tried ${NEXTCONN} instead)" && return 7 cleanup } -run_test 35 "Reconnect to the last active server first" +run_test 35a "Reconnect to the last active server first" + +test_35b() { # bug 18674 + remote_mds || { skip "local MDS" && return 0; } + setup + + debugsave + $LCTL set_param debug="ha" + $LCTL clear + MSG="conf-sanity.sh test_35b `date +%F%kh%Mm%Ss`" + log "$MSG" + + log "Set up a fake failnode for the MDS" + FAKENID="127.0.0.2" + local device=$(do_facet mds "$LCTL get_param -n devices" | \ + awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1) + do_facet mds "$LCTL conf_param ${device}.failover.node=$FAKENID" || \ + return 1 + + local at_max_saved=0 + # adaptive timeouts may prevent seeing the issue + if at_is_enabled; then + at_max_saved=$(at_max_get mds) + at_max_set 0 mds client + fi + + mkdir -p $MOUNT/testdir + touch $MOUNT/testdir/test + + log "Injecting EBUSY on MDS" + # Setting OBD_FAIL_MDS_RESEND=0x136 + do_facet mds "$LCTL set_param fail_loc=0x80000136" || return 2 + + log "Stat on a test file" + stat $MOUNT/testdir/test + + log "Stop injecting EBUSY on MDS" + do_facet mds "$LCTL set_param fail_loc=0" || return 3 + rm -f $MOUNT/testdir/test + + log "done" + # restore adaptive timeout + [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds client + + $LCTL dk $TMP/lustre-log-$TESTNAME.log + + # retrieve from the log if the client has ever tried to + # contact the fake server after the loss of connection + FAILCONN=`awk "BEGIN {ret = 0;} + /import_select_connection.*${FSNAME}-MDT0000-mdc.* using connection/ { + ret = 1; + if (\\\$NF ~ /$FAKENID/) { + ret = 2; + exit; + } + } + END {print ret}" $TMP/lustre-log-$TESTNAME.log` + + [ "$FAILCONN" == "0" ] && \ + log "ERROR: The client reconnection has not been triggered" && \ + return 4 + [ "$FAILCONN" == "2" ] && \ + log "ERROR: The client tried to reconnect to the failover server while the primary was busy" && \ + return 5 + + cleanup +} +run_test 35b "Continue reconnection retries, if the active server is busy" test_36() { # 12743 local rc @@ -1405,9 +1325,9 @@ test_36() { # 12743 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 + skip_env "mixed loopback and real device not working" && return fi - [ $OSTCOUNT -lt 2 ] && skip "skipping test for single OST" && return + [ $OSTCOUNT -lt 2 ] && skip_env "skipping test for single OST" && return [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \ { skip "remote OST" && return 0; } @@ -1442,17 +1362,17 @@ test_36() { # 12743 ALLOWANCE=$((64 * $OSTCOUNT)) - if [ $DFTOTAL -lt $(($BKTOTAL - $ALLOWANCE)) ] || + if [ $DFTOTAL -lt $(($BKTOTAL - $ALLOWANCE)) ] || [ $DFTOTAL -gt $(($BKTOTAL + $ALLOWANCE)) ] ; then echo "**** FAIL: df total($DFTOTAL) mismatch OST total($BKTOTAL)" rc=1 fi - if [ $DFFREE -lt $(($BKFREE - $ALLOWANCE)) ] || + if [ $DFFREE -lt $(($BKFREE - $ALLOWANCE)) ] || [ $DFFREE -gt $(($BKFREE + $ALLOWANCE)) ] ; then echo "**** FAIL: df free($DFFREE) mismatch OST free($BKFREE)" rc=2 fi - if [ $DFAVAIL -lt $(($BKAVAIL - $ALLOWANCE)) ] || + if [ $DFAVAIL -lt $(($BKAVAIL - $ALLOWANCE)) ] || [ $DFAVAIL -gt $(($BKAVAIL + $ALLOWANCE)) ] ; then echo "**** FAIL: df avail($DFAVAIL) mismatch OST avail($BKAVAIL)" rc=3 @@ -1469,7 +1389,7 @@ test_36() { # 12743 run_test 36 "df report consistency on OSTs with different block size" test_37() { - [ -n "$CLIENTONLY" -o -n "$CLIENTMODSONLY" ] && skip "client only testing" && return 0 + client_only && skip "client only testing" && return 0 LOCAL_MDSDEV="$TMP/mdt.img" SYM_MDSDEV="$TMP/sym_mdt.img" @@ -1510,10 +1430,13 @@ test_38() { # bug 14222 stop_mds log "rename lov_objid file on MDS" rm -f $TMP/lov_objid.orig - do_facet mds "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.orig\\\" $MDSDEV" - do_facet mds "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV" - do_facet mds "od -Ax -td8 $TMP/lov_objid.orig" + local dev=${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 @@ -1521,17 +1444,17 @@ test_38() { # bug 14222 [ $V ] && log "verifying $DIR/$tdir/$f" diff -q $f $DIR/$tdir/$f || ERROR=y done - do_facet mds "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new\\\" $MDSDEV" - do_facet mds "od -Ax -td8 $TMP/lov_objid.new" - [ "$ERROR" = "y" ] && error "old and new files are different after connect" || true - + 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 $MOUNT stop_mds - do_facet mds dd if=/dev/zero of=$TMP/lov_objid.clear bs=4096 count=1 - do_facet mds "debugfs -w -R \\\"rm lov_objid\\\" $MDSDEV" - do_facet mds "debugfs -w -R \\\"write $TMP/lov_objid.clear lov_objid\\\" $MDSDEV " + 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 @@ -1539,12 +1462,12 @@ test_38() { # bug 14222 [ $V ] && log "verifying $DIR/$tdir/$f" diff -q $f $DIR/$tdir/$f || ERROR=y done - do_facet mds "debugfs -c -R \\\"dump lov_objid $TMP/lov_objid.new1\\\" $MDSDEV" - do_facet mds "od -Ax -td8 $TMP/lov_objid.new1" + 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 } @@ -1554,14 +1477,837 @@ test_39() { PTLDEBUG=+malloc setup cleanup - perl $SRCDIR/leak_finder.pl $TMP/debug 2>&1 | egrep '*** Leak:' && + 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,force + 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 mgs $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" + +test_43() { + [ $UID -ne 0 -o $RUNAS_ID -eq 0 ] && skip_env "run as root" + setup + chmod ugo+x $DIR || error "chmod 0 failed" + set_and_check mds \ + "lctl get_param -n mdt.$FSNAME-MDT0000.root_squash" \ + "$FSNAME.mdt.root_squash" \ + "0:0" + set_and_check mds \ + "lctl get_param -n mdt.$FSNAME-MDT0000.nosquash_nids" \ + "$FSNAME.mdt.nosquash_nids" \ + "NONE" + + # + # create set of test files + # + echo "111" > $DIR/$tfile-userfile || error "write 1 failed" + chmod go-rw $DIR/$tfile-userfile || error "chmod 1 failed" + chown $RUNAS_ID.$RUNAS_ID $DIR/$tfile-userfile || error "chown failed" + + echo "222" > $DIR/$tfile-rootfile || error "write 2 failed" + chmod go-rw $DIR/$tfile-rootfile || error "chmod 2 faield" + + mkdir $DIR/$tdir-rootdir -p || error "mkdir failed" + chmod go-rwx $DIR/$tdir-rootdir || error "chmod 3 failed" + touch $DIR/$tdir-rootdir/tfile-1 || error "touch failed" + + # + # check root_squash: + # set root squash UID:GID to RUNAS_ID + # root should be able to access only files owned by RUNAS_ID + # + set_and_check mds \ + "lctl get_param -n mdt.$FSNAME-MDT0000.root_squash" \ + "$FSNAME.mdt.root_squash" \ + "$RUNAS_ID:$RUNAS_ID" + + ST=$(stat -c "%n: owner uid %u (%A)" $DIR/$tfile-userfile) + dd if=$DIR/$tfile-userfile 1>/dev/null 2>/dev/null || \ + error "$ST: root read permission is denied" + echo "$ST: root read permission is granted - ok" + + echo "444" | \ + dd conv=notrunc if=$DIR/$tfile-userfile 1>/dev/null 2>/dev/null || \ + error "$ST: root write permission is denied" + echo "$ST: root write permission is granted - ok" + + ST=$(stat -c "%n: owner uid %u (%A)" $DIR/$tfile-rootfile) + dd if=$DIR/$tfile-rootfile 1>/dev/null 2>/dev/null && \ + error "$ST: root read permission is granted" + echo "$ST: root read permission is denied - ok" + + echo "555" | \ + dd conv=notrunc of=$DIR/$tfile-rootfile 1>/dev/null 2>/dev/null && \ + error "$ST: root write permission is granted" + echo "$ST: root write permission is denied - ok" + + ST=$(stat -c "%n: owner uid %u (%A)" $DIR/$tdir-rootdir) + rm $DIR/$tdir-rootdir/tfile-1 1>/dev/null 2>/dev/null && \ + error "$ST: root unlink permission is granted" + echo "$ST: root unlink permission is denied - ok" + + touch $DIR/tdir-rootdir/tfile-2 1>/dev/null 2>/dev/null && \ + error "$ST: root create permission is granted" + echo "$ST: root create permission is denied - ok" + + # + # check nosquash_nids: + # put client's NID into nosquash_nids list, + # root should be able to access root file after that + # + local NIDLIST=$(lctl list_nids all | tr '\n' ' ') + NIDLIST="2@elan $NIDLIST 192.168.0.[2,10]@tcp" + NIDLIST=$(echo $NIDLIST | tr -s ' ' ' ') + set_and_check mds \ + "lctl get_param -n mdt.$FSNAME-MDT0000.nosquash_nids" \ + "$FSNAME-MDTall.mdt.nosquash_nids" \ + "$NIDLIST" + + ST=$(stat -c "%n: owner uid %u (%A)" $DIR/$tfile-rootfile) + dd if=$DIR/$tfile-rootfile 1>/dev/null 2>/dev/null || \ + error "$ST: root read permission is denied" + echo "$ST: root read permission is granted - ok" + + echo "666" | \ + dd conv=notrunc of=$DIR/$tfile-rootfile 1>/dev/null 2>/dev/null || \ + error "$ST: root write permission is denied" + echo "$ST: root write permission is granted - ok" + + ST=$(stat -c "%n: owner uid %u (%A)" $DIR/$tdir-rootdir) + rm $DIR/$tdir-rootdir/tfile-1 || \ + error "$ST: root unlink permission is denied" + echo "$ST: root unlink permission is granted - ok" + touch $DIR/$tdir-rootdir/tfile-2 || \ + error "$ST: root create permission is denied" + echo "$ST: root create permission is granted - ok" + + return 0 +} +run_test 43 "check root_squash and nosquash_nids" + umount_client $MOUNT cleanup_nocli -cleanup_gss +test_44() { # 16317 + setup + check_mount || return 2 + UUID=$($LCTL get_param llite.${FSNAME}*.uuid | cut -d= -f2) + STATS_FOUND=no + UUIDS=$(do_facet mds "$LCTL get_param mdt.${FSNAME}*.exports.*.uuid") + for VAL in $UUIDS; do + NID=$(echo $VAL | cut -d= -f1) + CLUUID=$(echo $VAL | cut -d= -f2) + [ "$UUID" = "$CLUUID" ] && STATS_FOUND=yes && break + done + [ "$STATS_FOUND" = "no" ] && error "stats not found for client" + cleanup + return 0 +} +run_test 44 "mounted client proc entry exists" + +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" + +cleanup_46a() { + trap 0 + local rc=0 + local count=$1 + + umount_client $MOUNT2 || rc=$? + umount_client $MOUNT || rc=$? + while [ $count -gt 0 ]; do + stop ost${count} -f || rc=$? + let count=count-1 + done + stop_mds || rc=$? + # writeconf is needed after the test, otherwise, + # we might end up with extra OSTs + writeconf || rc=$? + cleanup_nocli || rc=$? + return $rc +} + +test_46a() { + echo "Testing with $OSTCOUNT OSTs" + reformat + start_mds || return 1 + #first client should see only one ost + start_ost || return 2 + wait_osc_import_state mds ost FULL + #start_client + mount_client $MOUNT || return 3 + trap "cleanup_46a $OSTCOUNT" EXIT ERR + + local i + for (( i=2; i<=$OSTCOUNT; i++ )); do + start ost$i `ostdevname $i` $OST_MOUNT_OPTS || return $((i+2)) + done + + # wait until osts in sync + for (( i=2; i<=$OSTCOUNT; i++ )); do + wait_osc_import_state mds ost$i FULL + done + + + #second client see all ost's + + mount_client $MOUNT2 || return 8 + $LFS setstripe $MOUNT2 -c -1 || return 9 + $LFS getstripe $MOUNT2 || return 10 + + echo "ok" > $MOUNT2/widestripe + $LFS getstripe $MOUNT2/widestripe || return 11 + # fill acl buffer for avoid expand lsm to them + awk -F : '{if (FNR < 25) { print "u:"$1":rwx" }}' /etc/passwd | while read acl; do + setfacl -m $acl $MOUNT2/widestripe + done + + # will be deadlock + stat $MOUNT/widestripe || return 12 + + cleanup_46a $OSTCOUNT || { echo "cleanup_46a failed!" && return 13; } + return 0 +} +run_test 46a "handle ost additional - wide striped file" + +test_47() { #17674 + reformat + setup_noconfig + check_mount || return 2 + $LCTL set_param ldlm.namespaces.$FSNAME-*-*-*.lru_size=100 + + local lru_size=[] + local count=0 + for ns in $($LCTL get_param ldlm.namespaces.$FSNAME-*-*-*.lru_size); do + if echo $ns | grep "MDT[[:digit:]]*"; then + continue + fi + lrs=$(echo $ns | sed 's/.*lru_size=//') + lru_size[count]=$lrs + let count=count+1 + done + + facet_failover ost1 + facet_failover $SINGLEMDS + client_up || return 3 + + count=0 + for ns in $($LCTL get_param ldlm.namespaces.$FSNAME-*-*-*.lru_size); do + if echo $ns | grep "MDT[[:digit:]]*"; then + continue + fi + lrs=$(echo $ns | sed 's/.*lru_size=//') + if ! test "$lrs" -eq "${lru_size[count]}"; then + n=$(echo $ns | sed -e 's/ldlm.namespaces.//' -e 's/.lru_size=.*//') + error "$n has lost lru_size: $lrs vs. ${lru_size[count]}" + fi + let count=count+1 + done + + cleanup + return 0 +} +run_test 47 "server restart does not make client loss lru_resize settings" + +cleanup_48() { + trap 0 + + # reformat after this test is needed - if test will failed + # we will have unkillable file at FS + reformat_and_config +} + +test_48() { # bug 17636 + reformat + setup_noconfig + check_mount || return 2 + + $LFS setstripe $MOUNT -c -1 || return 9 + $LFS getstripe $MOUNT || return 10 + + echo "ok" > $MOUNT/widestripe + $LFS getstripe $MOUNT/widestripe || return 11 + + trap cleanup_48 EXIT ERR + + # fill acl buffer for avoid expand lsm to them + getent passwd | awk -F : '{ print "u:"$1":rwx" }' | while read acl; do + setfacl -m $acl $MOUNT/widestripe + done + + stat $MOUNT/widestripe || return 12 + + cleanup_48 + return 0 +} +run_test 48 "too many acls on file" + +# check PARAM_SYS_LDLM_TIMEOUT option of MKFS.LUSTRE +test_49() { # bug 17710 + local OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS + local OLD_OST_MKFS_OPTS=$OST_MKFS_OPTS + local LOCAL_TIMEOUT=20 + + + OST_MKFS_OPTS="--ost --fsname=$FSNAME --device-size=$OSTSIZE --mgsnode=$MGSNID --param sys.timeout=$LOCAL_TIMEOUT --param sys.ldlm_timeout=$LOCAL_TIMEOUT $MKFSOPT $OSTOPT" + + reformat + start_mds + start_ost + mount_client $MOUNT + check_mount || return 1 + + echo "check ldlm_timout..." + LDLM_MDS="`do_facet mds lctl get_param -n ldlm_timeout`" + LDLM_OST1="`do_facet ost1 lctl get_param -n ldlm_timeout`" + LDLM_CLIENT="`do_facet client lctl get_param -n ldlm_timeout`" + + if [ $LDLM_MDS -ne $LDLM_OST1 ] || [ $LDLM_MDS -ne $LDLM_CLIENT ]; then + error "Different LDLM_TIMEOUT:$LDLM_MDS $LDLM_OST1 $LDLM_CLIENT" + fi + + if [ $LDLM_MDS -ne $((LOCAL_TIMEOUT / 3)) ]; then + error "LDLM_TIMEOUT($LDLM_MDS) is not correct" + fi + + umount_client $MOUNT + stop_ost || return 2 + stop_mds || return 3 + + OST_MKFS_OPTS="--ost --fsname=$FSNAME --device-size=$OSTSIZE --mgsnode=$MGSNID --param sys.timeout=$LOCAL_TIMEOUT --param sys.ldlm_timeout=$((LOCAL_TIMEOUT - 1)) $MKFSOPT $OSTOPT" + + reformat + start_mds || return 4 + start_ost || return 5 + mount_client $MOUNT || return 6 + check_mount || return 7 + + LDLM_MDS="`do_facet mds lctl get_param -n ldlm_timeout`" + LDLM_OST1="`do_facet ost1 lctl get_param -n ldlm_timeout`" + LDLM_CLIENT="`do_facet client lctl get_param -n ldlm_timeout`" + + if [ $LDLM_MDS -ne $LDLM_OST1 ] || [ $LDLM_MDS -ne $LDLM_CLIENT ]; then + error "Different LDLM_TIMEOUT:$LDLM_MDS $LDLM_OST1 $LDLM_CLIENT" + fi + + if [ $LDLM_MDS -ne $((LOCAL_TIMEOUT - 1)) ]; then + error "LDLM_TIMEOUT($LDLM_MDS) is not correct" + fi + + cleanup || return $? + + MDS_MKFS_OPTS=$OLD_MDS_MKFS_OPTS + OST_MKFS_OPTS=$OLD_OST_MKFS_OPTS +} +run_test 49 "check PARAM_SYS_LDLM_TIMEOUT option of MKFS.LUSTRE" + +lazystatfs() { + # Test both statfs and lfs df and fail if either one fails + multiop_bg_pause $1 f_ + RC1=$? + PID=$! + killall -USR1 multiop + [ $RC1 -ne 0 ] && log "lazystatfs multiop failed" + wait $PID || { RC1=$?; log "multiop return error "; } + + $LFS df & + PID=$! + sleep 5 + kill -s 0 $PID + RC2=$? + if [ $RC2 -eq 0 ]; then + kill -s 9 $PID + log "lazystatfs df failed" + fi + + RC=0 + [[ $RC1 -ne 0 || $RC2 -eq 0 ]] && RC=1 + return $RC +} + +test_50a() { + setup + lctl set_param llite.$FSNAME-*.lazystatfs=1 + touch $DIR/$tfile + + lazystatfs $MOUNT || error "lazystatfs failed but no down servers" + + cleanup || return $? +} +run_test 50a "lazystatfs all servers available ==========================" + +test_50b() { + setup + lctl set_param llite.$FSNAME-*.lazystatfs=1 + touch $DIR/$tfile + + # Wait for client to detect down OST + stop_ost || error "Unable to stop OST1" + wait_osc_import_state mds ost DISCONN + + lazystatfs $MOUNT || error "lazystatfs should don't have returned EIO" + + umount_client $MOUNT || error "Unable to unmount client" + stop_mds || error "Unable to stop MDS" +} +run_test 50b "lazystatfs all servers down ==========================" + +test_50c() { + start_mds || error "Unable to start MDS" + start_ost || error "Unable to start OST1" + start_ost2 || error "Unable to start OST2" + mount_client $MOUNT || error "Unable to mount client" + lctl set_param llite.$FSNAME-*.lazystatfs=1 + touch $DIR/$tfile + + # Wait for client to detect down OST + stop_ost || error "Unable to stop OST1" + wait_osc_import_state mds ost DISCONN + lazystatfs $MOUNT || error "lazystatfs failed with one down server" + + umount_client $MOUNT || error "Unable to unmount client" + stop_ost2 || error "Unable to stop OST2" + stop_mds || error "Unable to stop MDS" +} +run_test 50c "lazystatfs one server down ==========================" + +test_50d() { + start_mds || error "Unable to start MDS" + start_ost || error "Unable to start OST1" + start_ost2 || error "Unable to start OST2" + mount_client $MOUNT || error "Unable to mount client" + lctl set_param llite.$FSNAME-*.lazystatfs=1 + touch $DIR/$tfile + + # Issue the statfs during the window where the client still + # belives the OST to be available but it is in fact down. + # No failure just a statfs which hangs for a timeout interval. + stop_ost || error "Unable to stop OST1" + lazystatfs $MOUNT || error "lazystatfs failed with one down server" + + umount_client $MOUNT || error "Unable to unmount client" + stop_ost2 || error "Unable to stop OST2" + stop_mds || error "Unable to stop MDS" +} +run_test 50d "lazystatfs client/server conn race ==========================" + +test_50e() { + local RC1 + local pid + + reformat_and_config + start_mds || return 1 + #first client should see only one ost + start_ost || return 2 + wait_osc_import_state mds ost FULL + + # Wait for client to detect down OST + stop_ost || error "Unable to stop OST1" + wait_osc_import_state mds ost DISCONN + + mount_client $MOUNT || error "Unable to mount client" + lctl set_param llite.$FSNAME-*.lazystatfs=0 + + multiop_bg_pause $MOUNT _f + RC1=$? + pid=$! + + if [ $RC1 -ne 0 ]; then + log "multiop failed $RC1" + else + kill -USR1 $pid + sleep $(( $TIMEOUT+1 )) + kill -0 $pid + [ $? -ne 0 ] && error "process isn't sleep" + start_ost || error "Unable to start OST1" + wait $pid || error "statfs failed" + fi + + umount_client $MOUNT || error "Unable to unmount client" + stop_ost || error "Unable to stop OST1" + stop_mds || error "Unable to stop MDS" +} +run_test 50e "normal statfs all servers down ==========================" + +test_50f() { + local RC1 + local pid + CONN_PROC="osc.$FSNAME-OST0001-osc-[M]*.ost_server_uuid" + + start_mds || error "Unable to start mds" + #first client should see only one ost + start_ost || error "Unable to start OST1" + wait_osc_import_state mds ost FULL + + start_ost2 || error "Unable to start OST2" + wait_osc_import_state mds ost2 FULL + + # Wait for client to detect down OST + stop_ost2 || error "Unable to stop OST2" + + wait_osc_import_state mds ost2 DISCONN + mount_client $MOUNT || error "Unable to mount client" + lctl set_param llite.$FSNAME-*.lazystatfs=0 + + multiop_bg_pause $MOUNT _f + RC1=$? + pid=$! + + if [ $RC1 -ne 0 ]; then + log "lazystatfs multiop failed $RC1" + else + kill -USR1 $pid + sleep $(( $TIMEOUT+1 )) + kill -0 $pid + [ $? -ne 0 ] && error "process isn't sleep" + start_ost2 || error "Unable to start OST2" + wait $pid || error "statfs failed" + stop_ost2 || error "Unable to stop OST2" + fi + + umount_client $MOUNT || error "Unable to unmount client" + stop_ost || error "Unable to stop OST1" + stop_mds || error "Unable to stop MDS" + writeconf +} +run_test 50f "normal statfs one server in down ==========================" + +test_50g() { + [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2, skipping" && return + setup + start_ost2 || error "Unable to start OST2" + + local PARAM="${FSNAME}-OST0001.osc.active" + + $LFS setstripe -c -1 $DIR/$tfile || error "Unable to lfs setstripe" + do_facet mgs $LCTL conf_param $PARAM=0 || error "Unable to deactivate OST" + + umount_client $MOUNT || error "Unable to unmount client" + mount_client $MOUNT || error "Unable to mount client" + # This df should not cause a panic + df -k $MOUNT + + do_facet mgs $LCTL conf_param $PARAM=1 || error "Unable to activate OST" + rm -f $DIR/$tfile + umount_client $MOUNT || error "Unable to unmount client" + stop_ost2 || error "Unable to stop OST2" + stop_ost || error "Unable to stop OST1" + stop_mds || error "Unable to stop MDS" + writeconf +} +run_test 50g "deactivated OST should not cause panic=====================" + +test_51() { + local LOCAL_TIMEOUT=20 + + reformat + start_mds + start_ost + mount_client $MOUNT + check_mount || return 1 + + mkdir $MOUNT/d1 + $LFS setstripe -c -1 $MOUNT/d1 + #define OBD_FAIL_MDS_REINT_DELAY 0x142 + do_facet $SINGLEMDS "lctl set_param fail_loc=0x142" + touch $MOUNT/d1/f1 & + local pid=$! + sleep 2 + start_ost2 || return 2 + wait $pid + stop_ost2 || return 3 + cleanup +} +run_test 51 "Verify that mdt_reint handles RMF_MDT_MD correctly when an OST is added" + +copy_files_xattrs() +{ + local node=$1 + local dest=$2 + local xattrs=$3 + shift 3 + + do_node $node mkdir -p $dest + [ $? -eq 0 ] || { error "Unable to create directory"; return 1; } + + do_node $node 'tar cf - '$@' | tar xf - -C '$dest'; + [ \"\${PIPESTATUS[*]}\" = \"0 0\" ] || exit 1' + [ $? -eq 0 ] || { error "Unable to tar files"; return 2; } + + do_node $node 'getfattr -d -m "[a-z]*\\." '$@' > '$xattrs + [ $? -eq 0 ] || { error "Unable to read xattrs"; return 3; } +} + +diff_files_xattrs() +{ + local node=$1 + local backup=$2 + local xattrs=$3 + shift 3 + + local backup2=${TMP}/backup2 + + do_node $node mkdir -p $backup2 + [ $? -eq 0 ] || { error "Unable to create directory"; return 1; } + + do_node $node 'tar cf - '$@' | tar xf - -C '$backup2'; + [ \"\${PIPESTATUS[*]}\" = \"0 0\" ] || exit 1' + [ $? -eq 0 ] || { error "Unable to tar files to diff"; return 2; } + + do_node $node "diff -rq $backup $backup2" + [ $? -eq 0 ] || { error "contents differ"; return 3; } + + local xattrs2=${TMP}/xattrs2 + do_node $node 'getfattr -d -m "[a-z]*\\." '$@' > '$xattrs2 + [ $? -eq 0 ] || { error "Unable to read xattrs to diff"; return 4; } + + do_node $node "diff $xattrs $xattrs2" + [ $? -eq 0 ] || { error "xattrs differ"; return 5; } + + do_node $node "rm -rf $backup2 $xattrs2" + [ $? -eq 0 ] || { error "Unable to delete temporary files"; return 6; } +} + +test_52() { + start_mds + [ $? -eq 0 ] || { error "Unable to start MDS"; return 1; } + start_ost + [ $? -eq 0 ] || { error "Unable to start OST1"; return 2; } + mount_client $MOUNT + [ $? -eq 0 ] || { error "Unable to mount client"; return 3; } + + local nrfiles=8 + local ost1mnt=${MOUNT%/*}/ost1 + local ost1node=$(facet_active_host ost1) + local ost1tmp=$TMP/conf52 + + mkdir -p $DIR/$tdir + [ $? -eq 0 ] || { error "Unable to create tdir"; return 4; } + touch $TMP/modified_first + [ $? -eq 0 ] || { error "Unable to create temporary file"; return 5; } + do_node $ost1node "mkdir -p $ost1tmp && touch $ost1tmp/modified_first" + [ $? -eq 0 ] || { error "Unable to create temporary file"; return 6; } + sleep 1 + + $LFS setstripe $DIR/$tdir -c -1 -s 1M + [ $? -eq 0 ] || { error "lfs setstripe failed"; return 7; } + + for (( i=0; i < nrfiles; i++ )); do + multiop $DIR/$tdir/$tfile-$i Ow1048576w1048576w524288c + [ $? -eq 0 ] || { error "multiop failed"; return 8; } + echo -n . + done + echo + + # backup files + echo backup files to $TMP/files + local files=$(find $DIR/$tdir -type f -newer $TMP/modified_first) + copy_files_xattrs `hostname` $TMP/files $TMP/file_xattrs $files + [ $? -eq 0 ] || { error "Unable to copy files"; return 9; } + + umount_client $MOUNT + [ $? -eq 0 ] || { error "Unable to umount client"; return 10; } + stop_ost + [ $? -eq 0 ] || { error "Unable to stop ost1"; return 11; } + + echo mount ost1 as ldiskfs + do_node $ost1node mount -t $FSTYPE $ost1_dev $ost1mnt $OST_MOUNT_OPTS + [ $? -eq 0 ] || { error "Unable to mount ost1 as ldiskfs"; return 12; } + + # backup objects + echo backup objects to $ost1tmp/objects + local objects=$(do_node $ost1node 'find '$ost1mnt'/O/0 -type f -size +0'\ + '-newer '$ost1tmp'/modified_first -regex ".*\/[0-9]+"') + copy_files_xattrs $ost1node $ost1tmp/objects $ost1tmp/object_xattrs $objects + [ $? -eq 0 ] || { error "Unable to copy objects"; return 13; } + + # move objects to lost+found + do_node $ost1node 'mv '$objects' '${ost1mnt}'/lost+found' + [ $? -eq 0 ] || { error "Unable to move objects"; return 14; } + + # recover objects + do_node $ost1node "ll_recover_lost_found_objs -d $ost1mnt/lost+found" + [ $? -eq 0 ] || { error "ll_recover_lost_found_objs failed"; return 15; } + + # compare restored objects against saved ones + diff_files_xattrs $ost1node $ost1tmp/objects $ost1tmp/object_xattrs $objects + [ $? -eq 0 ] || { error "Unable to diff objects"; return 16; } + + do_node $ost1node "umount $ost1_dev" + [ $? -eq 0 ] || { error "Unable to umount ost1 as ldiskfs"; return 17; } + + start_ost + [ $? -eq 0 ] || { error "Unable to start ost1"; return 18; } + mount_client $MOUNT + [ $? -eq 0 ] || { error "Unable to mount client"; return 19; } + + # compare files + diff_files_xattrs `hostname` $TMP/files $TMP/file_xattrs $files + [ $? -eq 0 ] || { error "Unable to diff files"; return 20; } + + rm -rf $TMP/files $TMP/file_xattrs + [ $? -eq 0 ] || { error "Unable to delete temporary files"; return 21; } + do_node $ost1node "rm -rf $ost1tmp" + [ $? -eq 0 ] || { error "Unable to delete temporary files"; return 22; } + cleanup +} +run_test 52 "check recovering objects from lost+found" + +# Checks threads_min/max/started for some service +# +# Arguments: service name (OST or MDT), facet (e.g., ost1, $SINGLEMDS), and a +# parameter pattern prefix like 'ost.*.ost'. +thread_sanity() { + local modname=$1 + local facet=$2 + local parampat=$3 + local opts=$4 + local tmin + local tmin2 + local tmax + local tmax2 + local tstarted + local paramp + local msg="Insane $modname thread counts" + shift 4 + + setup + check_mount || return 41 + + # We need to expand $parampat, but it may match multiple parameters, so + # we'll pick the first one + if ! paramp=$(do_facet $facet "lctl get_param -N ${parampat}.threads_min"|head -1); then + error "Couldn't expand ${parampat}.threads_min parameter name" + return 22 + fi + + # Remove the .threads_min part + paramp=${paramp%.threads_min} + + # Check for sanity in defaults + tmin=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0) + tmax=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0) + tstarted=$(do_facet $facet "lctl get_param -n ${paramp}.threads_started" || echo 0) + lassert 23 "$msg (PDSH problems?)" '(($tstarted && $tmin && $tmax))' || return $? + lassert 24 "$msg" '(($tstarted >= $tmin && $tstarted <= tmax ))' || return $? + + # Check that we can lower min/max + do_facet $facet "lctl set_param ${paramp}.threads_min=$((tmin - 1))" + do_facet $facet "lctl set_param ${paramp}.threads_max=$((tmax - 10))" + tmin2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0) + tmax2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0) + lassert 25 "$msg" '(($tmin2 == ($tmin - 1) && $tmax2 == ($tmax -10)))' || return $? + + # Check that we can set min/max to the same value + do_facet $facet "lctl set_param ${paramp}.threads_min=$tmin" + do_facet $facet "lctl set_param ${paramp}.threads_max=$tmin" + tmin2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0) + tmax2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0) + lassert 26 "$msg" '(($tmin2 == $tmin && $tmax2 == $tmin))' || return $? + + # Check that we can't set max < min + do_facet $facet "lctl set_param ${paramp}.threads_max=$((tmin - 1))" + tmin2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0) + tmax2=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max" || echo 0) + lassert 27 "$msg" '(($tmin <= $tmax2))' || return $? + + # We need to ensure that we get the module options desired; to do this + # we set LOAD_MODULES_REMOTE=true and we call setmodopts below. + LOAD_MODULES_REMOTE=true + cleanup + local oldvalue + setmodopts -a $modname "$opts" oldvalue + + load_modules + setup + check_mount || return 41 + + # Restore previous setting of MODOPTS_* + setmodopts $modname "$oldvalue" + + # Check that $opts took + tmin=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min") + tmax=$(do_facet $facet "lctl get_param -n ${paramp}.threads_max") + tstarted=$(do_facet $facet "lctl get_param -n ${paramp}.threads_started") + lassert 28 "$msg" '(($tstarted == $tmin && $tstarted == $tmax ))' || return $? + cleanup + + # Workaround a YALA bug where YALA expects that modules will remain + # loaded on the servers + LOAD_MODULES_REMOTE=false + setup + cleanup +} + +test_53a() { + thread_sanity OST ost1 'ost.*.ost' 'oss_num_threads=64' +} +run_test 53a "check OSS thread count params" + +test_53b() { + thread_sanity MDT $SINGLEMDS 'mdt.*.*.' 'mdt_num_threads=64' +} +run_test 53b "check MDT thread count params" + +cleanup_gss equals_msg `basename $0`: test complete -[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true +[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true