X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fconf-sanity.sh;h=b55c9d3583c382082a17e01881c764e17cfa7813;hp=055302dda5b48abee8f971e3bfc94a359ce2d766;hb=468de179b3a2c33df62454216de13bdb6f14614e;hpb=fbe66c3a22bfa136c077cbd46c4186ecbf4946d2 diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh index 055302d..b55c9d3 100644 --- a/lustre/tests/conf-sanity.sh +++ b/lustre/tests/conf-sanity.sh @@ -1,6 +1,4 @@ #!/bin/bash -# -*- 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 @@ -12,16 +10,38 @@ set -e ONLY=${ONLY:-"$*"} # bug number for skipped test: -# 15977 ALWAYS_EXCEPT="$CONF_SANITY_EXCEPT" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! +is_sles11() # LU-2181 +{ + if [ -r /etc/SuSE-release ] + then + local vers=`grep VERSION /etc/SuSE-release | awk '{print $3}'` + local patchlev=`grep PATCHLEVEL /etc/SuSE-release \ + | awk '{print $3}'` + if [ $vers -eq 11 ] && [ $patchlev -eq 2 ] + then + return 0 + fi + fi + return 1 +} + +if is_sles11; then # LU-2181 + ALWAYS_EXCEPT="$ALWAYS_EXCEPT 23a 34b" +fi + if [ "$FAILURE_MODE" = "HARD" ]; then CONFIG_EXCEPTIONS="24a " && \ echo "Except the tests: $CONFIG_EXCEPTIONS for FAILURE_MODE=$FAILURE_MODE, bug 23573" && \ ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS" fi +# bug number for skipped test: +# a tool to create lustre filesystem images +ALWAYS_EXCEPT="32newtarball $ALWAYS_EXCEPT" + SRCDIR=`dirname $0` PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH @@ -29,6 +49,7 @@ PTLDEBUG=${PTLDEBUG:--1} SAVE_PWD=$PWD LUSTRE=${LUSTRE:-`dirname $0`/..} RLUSTRE=${RLUSTRE:-$LUSTRE} +export MULTIOP=${MULTIOP:-multiop} . $LUSTRE/tests/test-framework.sh init_test_env $@ @@ -39,11 +60,21 @@ MDSSIZE=200000 OSTSIZE=200000 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} +if ! combined_mgs_mds; then + # bug number for skipped test: 23954 + ALWAYS_EXCEPT="$ALWAYS_EXCEPT 24b" +fi + # STORED_MDSSIZE is used in test_18 if [ -n "$MDSSIZE" ]; then STORED_MDSSIZE=$MDSSIZE fi +# pass "-E lazy_itable_init" to mke2fs to speed up the formatting time +if [[ "$LDISKFS_MKFS_OPTS" != *lazy_itable_init* ]]; then + LDISKFS_MKFS_OPTS=$(csa_add "$LDISKFS_MKFS_OPTS" -E lazy_itable_init) +fi + init_logging # @@ -125,6 +156,12 @@ stop_mds() { stop $SINGLEMDS -f || return 97 } +stop_mgs() { + echo "stop mgs service on `facet_active_host mgs`" + # These tests all use non-failover stop + stop mgs -f || return 97 +} + start_ost() { echo "start ost1 service on `facet_active_host ost1`" start ost1 `ostdevname 1` $OST_MOUNT_OPTS $@ || return 95 @@ -154,12 +191,10 @@ mount_client() { } remount_client() { - local SAVEMOUNTOPT=$MOUNTOPT - MOUNTOPT="remount,$1" + local mountopt="-o remount,$1" local MOUNTPATH=$2 echo "remount '$1' lustre on ${MOUNTPATH}....." - zconf_mount `hostname` $MOUNTPATH || return 96 - MOUNTOPT=$SAVEMOUNTOPT + zconf_mount `hostname` $MOUNTPATH "$mountopt" || return 96 } umount_client() { @@ -181,6 +216,7 @@ setup() { start_mds || error "MDT start failed" start_ost || error "OST start failed" mount_client $MOUNT || error "client start failed" + client_up || error "client_up failed" } setup_noconfig() { @@ -193,14 +229,20 @@ setup_noconfig() { mount_client $MOUNT } +unload_modules_conf () { + if combined_mgs_mds || ! local_mode; then + unload_modules || return 1 + fi +} + cleanup_nocli() { stop_ost || return 202 stop_mds || return 201 - unload_modules || return 203 + unload_modules_conf || return 203 } cleanup() { - umount_client $MOUNT || return 200 + umount_client $MOUNT || return 200 cleanup_nocli || return $? } @@ -301,7 +343,7 @@ test_5a() { # was test_5 # cleanup may return an error from the failed # disconnects; for now I'll consider this successful # if all the modules have unloaded. - umount -d $MOUNT & + umount -d $MOUNT & UMOUNT_PID=$! sleep 6 echo "killing umount" @@ -332,7 +374,7 @@ test_5a() { # was test_5 while [ "$WAIT" -ne "$MAX_WAIT" ]; do sleep $sleep grep -q $MOUNT" " /etc/mtab || break - echo "Waiting /etc/mtab updated ... " + echo "Waiting /etc/mtab updated ... " WAIT=$(( WAIT + sleep)) done [ "$WAIT" -eq "$MAX_WAIT" ] && error "/etc/mtab is not updated in $WAIT secs" @@ -340,60 +382,135 @@ test_5a() { # was test_5 } run_test 5a "force cleanup mds, then cleanup" +cleanup_5b () { + trap 0 + start_mgs +} + test_5b() { + grep " $MOUNT " /etc/mtab && \ + error false "unexpected entry in mtab before mount" && return 10 + + local rc=0 start_ost + if ! combined_mgs_mds ; then + trap cleanup_5b EXIT ERR + start_mds + stop mgs + fi + [ -d $MOUNT ] || mkdir -p $MOUNT - 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 + mount_client $MOUNT && rc=1 + grep " $MOUNT " /etc/mtab && \ + error "$MOUNT entry in mtab after failed mount" && rc=11 umount_client $MOUNT # stop_mds is a no-op here, and should not fail - cleanup_nocli || return $? - return 0 + cleanup_nocli || rc=$? + if ! combined_mgs_mds ; then + cleanup_5b + fi + return $rc } -run_test 5b "mds down, cleanup after failed mount (bug 2712) (should return errs)" +run_test 5b "Try to start a client with no MGS (should return errs)" test_5c() { + grep " $MOUNT " /etc/mtab && \ + error false "unexpected entry in mtab before mount" && return 10 + + local rc=0 start_mds start_ost [ -d $MOUNT ] || mkdir -p $MOUNT - grep " $MOUNT " /etc/mtab && echo "test 5c: mtab before mount" && return 10 local oldfs="${FSNAME}" FSNAME="wrong.${FSNAME}" mount_client $MOUNT || : FSNAME=${oldfs} - grep " $MOUNT " /etc/mtab && echo "test 5c: mtab after failed mount" && return 11 + grep " $MOUNT " /etc/mtab && \ + error "$MOUNT entry in mtab after failed mount" && rc=11 umount_client $MOUNT - cleanup_nocli || return $? + cleanup_nocli || rc=$? + return $rc } run_test 5c "cleanup after failed mount (bug 2712) (should return errs)" test_5d() { + grep " $MOUNT " /etc/mtab && \ + error false "unexpected entry in mtab before mount" && return 10 + + [ "$(facet_fstype ost1)" = "zfs" ] && + skip "LU-2059: no local config for ZFS OSTs" && return + + local rc=0 start_ost start_mds stop_ost -f - grep " $MOUNT " /etc/mtab && echo "test 5d: mtab before mount" && return 10 - mount_client $MOUNT || return 1 - cleanup || return $? - grep " $MOUNT " /etc/mtab && echo "test 5d: mtab after unmount" && return 11 - return 0 + mount_client $MOUNT || rc=1 + cleanup || rc=$? + grep " $MOUNT " /etc/mtab && \ + error "$MOUNT entry in mtab after unmount" && rc=11 + return $rc } run_test 5d "mount with ost down" test_5e() { + grep " $MOUNT " /etc/mtab && \ + error false "unexpected entry in mtab before mount" && return 10 + + local rc=0 start_mds start_ost #define OBD_FAIL_PTLRPC_DELAY_SEND 0x506 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 $? - grep " $MOUNT " /etc/mtab && echo "test 5e: mtab after unmount" && return 11 - return 0 + cleanup || rc=$? + grep " $MOUNT " /etc/mtab && \ + error "$MOUNT entry in mtab after unmount" && rc=11 + return $rc } run_test 5e "delayed connect, don't crash (bug 10268)" +test_5f() { + if combined_mgs_mds ; then + skip "combined mgs and mds" + return 0 + fi + + grep " $MOUNT " /etc/mtab && \ + error false "unexpected entry in mtab before mount" && return 10 + + local rc=0 + start_ost + [ -d $MOUNT ] || mkdir -p $MOUNT + mount_client $MOUNT & + local pid=$! + echo client_mount pid is $pid + + sleep 5 + + if ! ps -f -p $pid >/dev/null; then + wait $pid + rc=$? + grep " $MOUNT " /etc/mtab && echo "test 5f: mtab after mount" + error "mount returns $rc, expected to hang" + rc=11 + cleanup || rc=$? + return $rc + fi + + # start mds + start_mds + + # mount should succeed after start mds + wait $pid + rc=$? + [ $rc -eq 0 ] || error "mount returned $rc" + grep " $MOUNT " /etc/mtab && echo "test 5f: mtab after mount" + cleanup || return $? + return $rc +} +run_test 5f "mds down, cleanup after failed mount (bug 2712)" + test_6() { setup manual_umount_client @@ -445,6 +562,21 @@ test_9() { } run_test 9 "test ptldebug and subsystem for mkfs" +is_blkdev () { + local facet=$1 + local dev=$2 + local size=${3:-""} + + local rc=0 + do_facet $facet "test -b $dev" || rc=1 + if [[ "$size" ]]; then + local in=$(do_facet $facet "dd if=$dev of=/dev/null bs=1k count=1 skip=$size 2>&1" |\ + awk '($3 == "in") { print $1 }') + [[ $in = "1+0" ]] || rc=1 + fi + return $rc +} + # # Test 16 was to "verify that lustre will correct the mode of OBJECTS". # But with new MDS stack we don't care about the mode of local objects @@ -452,26 +584,37 @@ run_test 9 "test ptldebug and subsystem for mkfs" # test_17() { - local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi - if [ ! -e "$MDSDEV" ]; then - echo "no $MDSDEV existing, so mount Lustre to create one" - setup - check_mount || return 41 - cleanup || return $? - fi + setup + check_mount || return 41 + cleanup || return $? echo "Remove mds config log" - do_facet $SINGLEMDS "$DEBUGFS -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MDSDEV || return \$?" || return $? + if ! combined_mgs_mds ; then + stop mgs + fi + + do_facet mgs "$DEBUGFS -w -R 'unlink CONFIGS/$FSNAME-MDT0000' $MGSDEV || return \$?" || return $? + + if ! combined_mgs_mds ; then + start_mgs + fi start_ost - start_mds && return 42 - reformat_and_config + start_mds && return 42 + reformat_and_config } run_test 17 "Verify failed mds_postsetup won't fail assertion (2936) (should return errs)" test_18() { - [ "$FSTYPE" != "ldiskfs" ] && skip "not needed for FSTYPE=$FSTYPE" && return + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) @@ -488,33 +631,34 @@ test_18() { log "use STORED_MDSSIZE=$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 $1 }')" = "1+0" ] && OK=1 && \ + [ -z "$OK" ] && $(is_blkdev $SINGLEMDS $MDSDEV $MIN) && OK=1 && myMDSSIZE=$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 && \ + + if [ -z "$OK" ]; then + local SPACE=$(do_facet $SINGLEMDS "[ -f $MDSDEV -o ! -e $MDSDEV ] && df -P \\\$(dirname $MDSDEV)" | + awk '($1 != "Filesystem") {print $4}') + ! [ -z "$SPACE" ] && [ $SPACE -gt $((MIN / 20)) ] && \ + OK=1 && myMDSSIZE=$MIN && \ log "use file $MDSDEV with MIN=$MIN" + fi [ -z "$OK" ] && skip_env "$MDSDEV too small for ${MIN}kB MDS" && return echo "mount mds with large journal..." - local OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS - MDS_MKFS_OPTS="--mgs --mdt --fsname=$FSNAME --device-size=$myMDSSIZE --param sys.timeout=$TIMEOUT $MDSOPT" + local OLD_MDSSIZE=$MDSSIZE + MDSSIZE=$myMDSSIZE reformat_and_config echo "mount lustre system..." - setup + setup check_mount || return 41 echo "check journal size..." - local FOUNDSIZE=`do_facet mds "$DEBUGFS -c -R 'stat <8>' $MDSDEV" | awk '/Size: / { print $NF; exit;}'` + local FOUNDSIZE=$(do_facet $SINGLEMDS "$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 @@ -523,8 +667,8 @@ test_18() { cleanup || return $? - MDS_MKFS_OPTS=$OLD_MDS_MKFS_OPTS - reformat_and_config + MDSSIZE=$OLD_MDSSIZE + reformat_and_config } run_test 18 "check mkfs creates large journals" @@ -535,6 +679,9 @@ test_19a() { run_test 19a "start/stop MDS without OSTs" test_19b() { + [ "$(facet_fstype ost1)" = "zfs" ] && + skip "LU-2059: no local config for ZFS OSTs" && return + start_ost || return 1 stop_ost -f || return 2 } @@ -571,6 +718,9 @@ test_21a() { run_test 21a "start mds before ost, stop ost first" test_21b() { + [ "$(facet_fstype ost1)" = "zfs" ] && + skip "LU-2059: no local config for ZFS OSTs" && return + start_ost start_mds wait_osc_import_state mds ost FULL @@ -592,6 +742,30 @@ test_21c() { } run_test 21c "start mds between two osts, stop mds last" +test_21d() { + if combined_mgs_mds ; then + skip "need separate mgs device" && return 0 + fi + stopall + + reformat + + start_mgs + start_ost + start_ost2 + start_mds + wait_osc_import_state mds ost2 FULL + + stop_ost + stop_ost2 + stop_mds + stop_mgs + #writeconf to remove all ost2 traces for subsequent tests + writeconf + start_mgs +} +run_test 21d "start mgs then ost and then mds" + test_22() { start_mds @@ -618,6 +792,8 @@ test_22() { sleep $((TIMEOUT + TIMEOUT + TIMEOUT)) fi mount_client $MOUNT + wait_osc_import_state mds ost FULL + wait_osc_import_state client ost FULL check_mount || return 41 pass @@ -631,10 +807,10 @@ test_23a() { # was test_23 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 - echo "Stopping client $MOUNT (opts: -f)" - umount -f $MOUNT - fi + if [ $running -ne 0 ]; then + echo "Stopping client $MOUNT (opts: -f)" + umount -f $MOUNT + fi # enter recovery on mds start_mds @@ -646,10 +822,12 @@ test_23a() { # was test_23 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_LUSTRE_PID echo "waiting for mount to finish" ps -ef | grep mount + # "ctrl-c" sends SIGINT but it usually (in script) does not work on child process + # SIGTERM works but it does not spread to offspring processses + kill -s TERM $MOUNT_PID + kill -s TERM $MOUNT_LUSTRE_PID # we can not wait $MOUNT_PID because it is not a child of this shell local PID1 local PID2 @@ -692,6 +870,10 @@ run_test 23b "Simulate -EINTR during mount" fs2mds_HOST=$mds_HOST fs2ost_HOST=$ost_HOST +MDSDEV1_2=$fs2mds_DEV +OSTDEV1_2=$fs2ost_DEV +OSTDEV2_2=$fs3ost_DEV + cleanup_24a() { trap 0 echo "umount $MOUNT2 ..." @@ -703,26 +885,27 @@ cleanup_24a() { } test_24a() { - #set up fs1 - gen_config - - #set up fs2 local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) - [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then - do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + is_blkdev $SINGLEMDS $MDSDEV && \ skip_env "mixed loopback and real device not working" && return fi - local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} - local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2} + [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST + + local fs2mdsdev=$(mdsdevname 1_2) + local fs2ostdev=$(ostdevname 1_2) + local fs2mdsvdev=$(mdsvdevname 1_2) + local fs2ostvdev=$(ostvdevname 1_2) # test 8-char fsname as well local FSNAME2=test1234 - add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME2} --nomgs --mgsnode=$MGSNID --reformat $fs2mdsdev || exit 10 + add fs2mds $(mkfs_opts mds1) --nomgs --mgsnode=$MGSNID \ + --fsname=${FSNAME2} --reformat $fs2mdsdev $fs2mdsvdev || exit 10 - add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --reformat $fs2ostdev || exit 10 + add fs2ost $(mkfs_opts ost1) --fsname=${FSNAME2} --reformat \ + $fs2ostdev $fs2ostvdev || exit 10 setup start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT @@ -745,7 +928,7 @@ test_24a() { 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 $4 }' | head -1) @@ -759,13 +942,17 @@ test_24b() { local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) if [ -z "$fs2mds_DEV" ]; then - do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + local dev=${SINGLEMDS}_dev + local MDSDEV=${!dev} + is_blkdev $SINGLEMDS $MDSDEV && \ skip_env "mixed loopback and real device not working" && return fi - local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} + local fs2mdsdev=$(mdsdevname 1_2) + local fs2mdsvdev=$(mdsvdevname 1_2) - add fs2mds $MDS_MKFS_OPTS --fsname=${FSNAME}2 --mgs --reformat $fs2mdsdev || exit 10 + add fs2mds $(mkfs_opts mds1) --mgs --fsname=${FSNAME}2 --reformat \ + $fs2mdsdev $fs2mdsvdev || exit 10 setup start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && return 2 cleanup || return 6 @@ -789,35 +976,26 @@ test_26() { 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) + DEVS=$(lctl get_param -n devices | egrep -v MG | wc -l) [ $DEVS -gt 0 ] && return 2 - unload_modules || return 203 + # start mds to drop writeconf setting + start_mds || return 3 + stop_mds || return 4 + unload_modules_conf || return $? } run_test 26 "MDT startup failure cleans LOV (should return errs)" -set_and_check() { - local myfacet=$1 - local TEST=$2 - local PARAM=$3 - local ORIG=$(do_facet $myfacet "$TEST") - if [ $# -gt 3 ]; then - local FINAL=$4 - else - local -i FINAL - FINAL=$(($ORIG + 5)) - fi - echo "Setting $PARAM from $ORIG to $FINAL" - do_facet $SINGLEMDS "$LCTL conf_param $PARAM='$FINAL'" || error conf_param failed - - wait_update $(facet_host $myfacet) "$TEST" "$FINAL" || error check failed! -} - test_27a() { + [ "$(facet_fstype ost1)" = "zfs" ] && + skip "LU-2059: no local config for ZFS OSTs" && return + start_ost || return 1 start_mds || return 2 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 + set_conf_param_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" @@ -828,23 +1006,27 @@ test_27b() { 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 + set_conf_param_and_check $SINGLEMDS \ + "lctl get_param -n mdt.$device.identity_acquire_expire" \ + "$device.mdt.identity_acquire_expire" || return 3 + set_conf_param_and_check client \ + "lctl get_param -n mdc.$device-mdc-*.max_rpcs_in_flight"\ + "$device.mdc.max_rpcs_in_flight" || return 4 check_mount cleanup } run_test 27b "Reacquire MGS lock after failover" -test_28a() { +test_28() { setup TEST="lctl get_param -n llite.$FSNAME-*.max_read_ahead_whole_mb" - PARAM="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" $FINAL || return 3 + set_conf_param_and_check client "$TEST" "$PARAM" $FINAL || return 3 FINAL=$(($FINAL + 1)) - set_and_check client "$TEST" "$PARAM" $FINAL || return 4 - umount_client $MOUNT || return 200 + set_conf_param_and_check client "$TEST" "$PARAM" $FINAL || return 4 + umount_client $MOUNT || return 200 mount_client $MOUNT RESULT=$($TEST) if [ $RESULT -ne $FINAL ]; then @@ -853,61 +1035,26 @@ test_28a() { else echo "New config success: got $RESULT" fi - set_and_check client "$TEST" "$PARAM" $ORIG || return 5 + set_conf_param_and_check client "$TEST" "$PARAM" $ORIG || return 5 cleanup } -run_test 28a "permanent parameter setting" - -check_28b() { - local NODE=$1 - shift - set_and_check $NODE "$LCTL get_param -n $1*.$2 | head -1" "$1.$2" "$3" || \ - error "conf_param $1.$2 failed" -} - -test_28b() { - setup > /dev/null - # should error - do_facet mgs "$LCTL conf_param foo=1 2>/dev/null" && \ - error "Bad format should fail" - do_facet mgs "$LCTL conf_param osc.notanfs-OST0000.active=0 2>/dev/null" && \ - error "Setting on unknown fs should fail" - do_facet mgs "$LCTL conf_param osc.$FSNAME-OST00000.active=0 2>/dev/null" && \ - error "Bad target name should fail" - # should succeed - check_28b mds mdt.$FSNAME-MDT0000 capa_timeout 1500 - check_28b mds mdt.$FSNAME-MDT* identity_expire 150 - check_28b mds mdd.$FSNAME-MDT0000 atime_diff 15 - check_28b mds mdd.$FSNAME-MDT* sync_permission 0 - check_28b ost1 obdfilter.$FSNAME-OST0000 client_cache_seconds 15 - check_28b ost1 obdfilter.$FSNAME-OST* client_cache_count 15 - check_28b mds lov.$FSNAME-MDT0000 qos_maxage "15 Sec" - check_28b mds lov.$FSNAME-MDT0000 qos_prio_free "15%" - check_28b client mdc.$FSNAME-MDT0000 max_rpcs_in_flight 15 - check_28b client osc.$FSNAME-OST0000 active 0 - check_28b client osc.$FSNAME-OST0000 active 1 - check_28b client osc.$FSNAME-OST0000 max_dirty_mb 15 - check_28b client llite.$FSNAME max_read_ahead_mb 15 - set_and_check client "$LCTL get_param -n at_max" "sys.$FSNAME.at_max" 1500 || \ - error "conf_param sys.fsname.at_max failed" - cleanup > /dev/null -} -run_test 28b "permanent parameter setting, set_param syntax" +run_test 28 "permanent parameter setting" test_29() { [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2, skipping" && return - setup > /dev/null 2>&1 + setup > /dev/null 2>&1 start_ost2 sleep 10 - local PARAM="osc.$FSNAME-OST0001.active" - local PROC_ACT="osc.$FSNAME-OST0001-osc-[^M]*.active" - local PROC_UUID="osc.$FSNAME-OST0001-osc-[^M]*.ost_server_uuid" + local PARAM="$FSNAME-OST0001.osc.active" + 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 $PROC_ACT) + ACTV=$(lctl get_param -n $PROC_ACT) DEAC=$((1 - $ACTV)) - set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $DEAC || return 2 - # also check ost_server_uuid status + set_conf_param_and_check client \ + "lctl get_param -n $PROC_ACT" "$PARAM" $DEAC || return 2 + # also check ost_server_uuid status RESULT=$(lctl get_param -n $PROC_UUID | grep DEACTIV) if [ -z "$RESULT" ]; then echo "Live client not deactivated: $(lctl get_param -n $PROC_UUID)" @@ -939,7 +1086,7 @@ test_29() { done # test new client starts deactivated - umount_client $MOUNT || return 200 + umount_client $MOUNT || return 200 mount_client $MOUNT RESULT=$(lctl get_param -n $PROC_UUID | grep DEACTIV | grep NEW) if [ -z "$RESULT" ]; then @@ -950,9 +1097,10 @@ test_29() { fi # make sure it reactivates - set_and_check client "lctl get_param -n $PROC_ACT" "$PARAM" $ACTV || return 6 + set_conf_param_and_check client \ + "lctl get_param -n $PROC_ACT" "$PARAM" $ACTV || return 6 - umount_client $MOUNT + umount_client $MOUNT stop_ost2 cleanup_nocli #writeconf to remove all ost2 traces for subsequent tests @@ -968,16 +1116,17 @@ test_30a() { ORIG=$($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" "llite.$FSNAME.max_read_ahead_whole_mb" $i || return 3 + set_conf_param_and_check client "$TEST" \ + "$FSNAME.llite.max_read_ahead_whole_mb" $i || return 3 done # make sure client restart still works - umount_client $MOUNT + umount_client $MOUNT mount_client $MOUNT || return 4 [ "$($TEST)" -ne "$i" ] && error "Param didn't stick across restart $($TEST) != $i" pass echo Erase parameter setting - do_facet mgs "$LCTL conf_param -d llite.$FSNAME.max_read_ahead_whole_mb" || return 6 + do_facet mgs "$LCTL conf_param -d $FSNAME.llite.max_read_ahead_whole_mb" || return 6 umount_client $MOUNT mount_client $MOUNT || return 6 FINAL=$($TEST) @@ -1002,13 +1151,15 @@ test_30b() { echo "Using fake nid $NEW" TEST="$LCTL get_param -n osc.$FSNAME-OST0000-osc-[^M]*.import | grep failover_nids | sed -n 's/.*\($NEW\).*/\1/p'" - set_and_check client "$TEST" "osc.$FSNAME-OST0000.failover.node" $NEW || error "didn't add failover nid $NEW" + set_conf_param_and_check client "$TEST" \ + "$FSNAME-OST0000.failover.node" $NEW || + error "didn't add failover nid $NEW" NIDS=$($LCTL get_param -n osc.$FSNAME-OST0000-osc-[^M]*.import | grep failover_nids) echo $NIDS NIDCOUNT=$(($(echo "$NIDS" | wc -w) - 1)) echo "should have 2 failover nids: $NIDCOUNT" [ $NIDCOUNT -eq 2 ] || error "Failover nid not added" - do_facet mgs "$LCTL conf_param -d osc.$FSNAME-OST0000.failover.node" || error "conf_param delete failed" + do_facet mgs "$LCTL conf_param -d $FSNAME-OST0000.failover.node" || error "conf_param delete failed" umount_client $MOUNT mount_client $MOUNT || return 3 @@ -1029,202 +1180,472 @@ 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 -} +# +# This is not really a test but a tool to create new disk +# image tarballs for the upgrade tests. +# +# Disk image tarballs should be created on single-node +# clusters by running this test with default configurations +# plus a few mandatory environment settings that are verified +# at the beginning of the test. +# +test_32newtarball() { + local version + local dst=. + local src=/etc/rc.d + local tmp=$TMP/t32_image_create + + if [ $FSNAME != t32fs -o $MDSCOUNT -ne 1 -o \ + \( -z "$MDSDEV" -a -z "$MDSDEV1" \) -o $OSTCOUNT -ne 1 -o \ + -z "$OSTDEV1" ]; then + error "Needs FSNAME=t32fs MDSCOUNT=1 MDSDEV1=" \ + "(or MDSDEV, in the case of b1_8) OSTCOUNT=1" \ + "OSTDEV1=" + fi -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 -} + mkdir $tmp || { + echo "Found stale $tmp" + return 1 + } -cleanup_nocli32 () { - stop32 mds1 -f - stop32 ost1 -f - wait_exit_ST client -} + mkdir $tmp/src + tar cf - -C $src . | tar xf - -C $tmp/src -cleanup_32() { - trap 0 - echo "Cleanup test_32 umount $MOUNT ..." - umount -f $MOUNT || true - echo "Cleanup local mds ost1 ..." - cleanup_nocli32 - unload_modules -} + formatall -test_32a() { - client_only && skip "client only testing" && return 0 - [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; } - [ -z "$TUNEFS" ] && skip_env "No tunefs" && return 0 + setupall + tar cf - -C $tmp/src . | tar xf - -C /mnt/$FSNAME + stopall - local DISK1_8=$LUSTRE/tests/disk1_8.tar.bz2 - [ ! -r $DISK1_8 ] && skip_env "Cannot find $DISK1_8" && return 0 - local tmpdir=$TMP/conf32a - mkdir -p $tmpdir - tar xjvf $DISK1_8 -C $tmpdir || \ - { skip_env "Cannot untar $DISK1_8" && return 0; } + mkdir $tmp/img - load_modules - $LCTL set_param debug=$PTLDEBUG + setupall + pushd /mnt/$FSNAME + ls -Rni --time-style=+%s >$tmp/img/list + find . ! -name .lustre -type f -exec sha1sum {} \; | + sort -k 2 >$tmp/img/sha1sums + popd + $LCTL get_param -n version | head -n 1 | + sed -e 's/^lustre: *//' >$tmp/img/commit + stopall - $TUNEFS $tmpdir/mds || error "tunefs failed" + pushd $tmp/src + find -type f -exec sha1sum {} \; | sort -k 2 >$tmp/sha1sums.src + popd - # nids are wrong, so client wont work, but server should start - start32 mds1 $tmpdir/mds "-o loop,exclude=lustre-OST0000" && \ - trap cleanup_32 EXIT INT || return 3 + if ! diff -u $tmp/sha1sums.src $tmp/img/sha1sums; then + echo "Data verification failed" + fi + + uname -r >$tmp/img/kernel + uname -m >$tmp/img/arch - local UUID=$($LCTL get_param -n mdt.lustre-MDT0000.uuid) - echo MDS uuid $UUID - [ "$UUID" == "lustre-MDT0000_UUID" ] || error "UUID is wrong: $UUID" + mv ${MDSDEV1:-$MDSDEV} $tmp/img + mv $OSTDEV1 $tmp/img - $TUNEFS --mgsnode=$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" == "lustre-OST0000_UUID" ] || error "UUID is wrong: $UUID" + version=$(sed -e 's/\(^[0-9]\+\.[0-9]\+\)\(.*$\)/\1/' $tmp/img/commit | + sed -e 's/\./_/g') # E.g., "1.8.7" -> "1_8" + dst=$(cd $dst; pwd) + pushd $tmp/img + tar cjvf $dst/disk$version-$(facet_fstype $SINGLEMDS).tar.bz2 -S * + popd - local NID=$($LCTL list_nids | head -1) + rm -r $tmp +} +#run_test 32newtarball "Create a new test_32 disk image tarball for this version" - 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." +# +# The list of applicable tarballs is returned via the caller's +# variable "tarballs". +# +t32_check() { + local node=$(facet_active_host $SINGLEMDS) + local r="do_node $node" - 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." + if [ "$CLIENTONLY" ]; then + skip "Client-only testing" + exit 0 + fi - echo "LOV changes should succeed:" - $LCTL pool_new lustre.interop || return 11 - $LCTL conf_param lustre-MDT0000.lov.stripesize=4M || return 12 - echo "ok." + if ! $r which $TUNEFS; then + skip_env "tunefs.lustre required on $node" + exit 0 + fi - cleanup_32 + if [ -n "$($LCTL list_nids | grep -v '\(tcp\|lo\)[[:digit:]]*$')" ]; then + skip "LU-2200: Test cannot run over Infiniband" + exit 0 + fi - # mount a second time to make sure we didnt leave upgrade flag on - load_modules - $TUNEFS --dryrun $tmpdir/mds || error "tunefs failed" - start32 mds1 $tmpdir/mds "-o loop,exclude=lustre-OST0000" && \ - trap cleanup_32 EXIT INT || return 12 + local IMGTYPE=$(facet_fstype $SINGLEMDS) - cleanup_32 + tarballs=$($r find $RLUSTRE/tests -maxdepth 1 -name \'disk*-$IMGTYPE.tar.bz2\') - rm -rf $tmpdir || true # true is only for TMP on NFS + if [ -z "$tarballs" ]; then + skip "No applicable tarballs found" + exit 0 + fi } -run_test 32a "Upgrade from 1.8 (not live)" -test_32b() { - client_only && skip "client only testing" && return 0 - [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; } - [ -z "$TUNEFS" ] && skip_env "No tunefs" && return +t32_test_cleanup() { + local node=$(facet_active_host $SINGLEMDS) + local r="do_node $node" + local tmp=$TMP/t32 + local rc=$? - local DISK1_8=$LUSTRE/tests/disk1_8.tar.bz2 - [ ! -r $DISK1_8 ] && skip_env "Cannot find $DISK1_8" && return 0 - local tmpdir=$TMP/conf32b - mkdir -p $tmpdir - tar xjvf $DISK1_8 -C $tmpdir || \ - { skip_env "Cannot untar $DISK1_8" && return ; } + if $shall_cleanup_lustre; then + umount $tmp/mnt/lustre || rc=$? + fi + if $shall_cleanup_mdt; then + $r umount -d $tmp/mnt/mdt || rc=$? + fi + if $shall_cleanup_ost; then + $r umount -d $tmp/mnt/ost || rc=$? + fi + $r rm -rf $tmp || rc=$? + rm -rf $tmp || rc=$? + return $rc +} - load_modules - $LCTL set_param debug="config" - local NEWNAME=lustre +t32_bits_per_long() { + # + # Yes, this is not meant to be perfect. + # + case $1 in + ppc64|x86_64) + echo -n 64;; + i*86) + echo -n 32;; + esac +} - # writeconf will cause servers to register with their current nids - $TUNEFS --writeconf --fsname=$NEWNAME $tmpdir/mds || error "tunefs failed" - start32 mds1 $tmpdir/mds "-o loop" && \ - trap cleanup_32 EXIT INT || return 3 +t32_reload_modules() { + local node=$1 + local all_removed=false + local i=0 + + while ((i < 20)); do + echo "Unloading modules on $node: Attempt $i" + do_rpc_nodes $node $LUSTRE_RMMOD $(facet_fstype $SINGLEMDS) && + all_removed=true + do_rpc_nodes $node check_mem_leak || return 1 + if $all_removed; then + load_modules + return 0 + fi + sleep 5 + i=$((i + 1)) + done + echo "Unloading modules on $node: Given up" + return 1 +} - local UUID=$($LCTL get_param -n mdt.${NEWNAME}-MDT0000.uuid) - echo MDS uuid $UUID - [ "$UUID" == "${NEWNAME}-MDT0000_UUID" ] || error "UUID is wrong: $UUID" +t32_wait_til_devices_gone() { + local node=$1 + local devices + local i=0 + + echo wait for devices to go + while ((i < 20)); do + devices=$(do_rpc_nodes $node $LCTL device_list | wc -l) + echo $device + ((devices == 0)) && return 1 + sleep 5 + i=$((i + 1)) + done + echo "waiting for devices on $node: Given up" + return 1 +} + +t32_test() { + local tarball=$1 + local writeconf=$2 + local shall_cleanup_mdt=false + local shall_cleanup_ost=false + local shall_cleanup_lustre=false + local node=$(facet_active_host $SINGLEMDS) + local r="do_node $node" + local tmp=$TMP/t32 + local img_commit + local img_kernel + local img_arch + local fsname=t32fs + local nid=$($r $LCTL list_nids | head -1) + local mopts + local uuid + local nrpcs_orig + local nrpcs + local list + + trap 'trap - RETURN; t32_test_cleanup' RETURN + + mkdir -p $tmp/mnt/lustre + $r mkdir -p $tmp/mnt/{mdt,ost} + $r tar xjvf $tarball -S -C $tmp || { + error_noexit "Unpacking the disk image tarball" + return 1 + } + img_commit=$($r cat $tmp/commit) + img_kernel=$($r cat $tmp/kernel) + img_arch=$($r cat $tmp/arch) + echo "Upgrading from $(basename $tarball), created with:" + echo " Commit: $img_commit" + echo " Kernel: $img_kernel" + echo " Arch: $img_arch" + + $r $LCTL set_param debug="$PTLDEBUG" + + $r $TUNEFS --dryrun $tmp/mdt || { + error_noexit "tunefs.lustre before mounting the MDT" + return 1 + } + if [ "$writeconf" ]; then + mopts=loop,writeconf + else + mopts=loop,exclude=$fsname-OST0000 + fi - $TUNEFS --mgsnode=$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" == "${NEWNAME}-OST0000_UUID" ] || error "UUID is wrong: $UUID" + t32_wait_til_devices_gone $node - local NID=$($LCTL list_nids | head -1) + $r mount -t lustre -o $mopts $tmp/mdt $tmp/mnt/mdt || { + error_noexit "Mounting the MDT" + return 1 + } + shall_cleanup_mdt=true - 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." + uuid=$($r $LCTL get_param -n mdt.$fsname-MDT0000.uuid) || { + error_noexit "Getting MDT UUID" + return 1 + } + if [ "$uuid" != $fsname-MDT0000_UUID ]; then + error_noexit "Unexpected MDT UUID: \"$uuid\"" + return 1 + fi - 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." + $r $TUNEFS --dryrun $tmp/ost || { + error_noexit "tunefs.lustre before mounting the OST" + return 1 + } + if [ "$writeconf" ]; then + mopts=loop,mgsnode=$nid,$writeconf + else + mopts=loop,mgsnode=$nid + fi + $r mount -t lustre -o $mopts $tmp/ost $tmp/mnt/ost || { + error_noexit "Mounting the OST" + return 1 + } + shall_cleanup_ost=true - echo "LOV changes should succeed:" - $LCTL pool_new ${NEWNAME}.interop || return 11 - $LCTL conf_param ${NEWNAME}-MDT0000.lov.stripesize=4M || return 12 - echo "ok." + uuid=$($r $LCTL get_param -n obdfilter.$fsname-OST0000.uuid) || { + error_noexit "Getting OST UUID" + return 1 + } + if [ "$uuid" != $fsname-OST0000_UUID ]; then + error_noexit "Unexpected OST UUID: \"$uuid\"" + return 1 + fi - # MDT and OST should have registered with new nids, so we should have - # a fully-functioning client - echo "Check client and old fs contents" + $r $LCTL conf_param $fsname-OST0000.osc.max_dirty_mb=15 || { + error_noexit "Setting \"max_dirty_mb\"" + return 1 + } + $r $LCTL conf_param $fsname-OST0000.failover.node=$nid || { + error_noexit "Setting OST \"failover.node\"" + return 1 + } + $r $LCTL conf_param $fsname-MDT0000.mdc.max_rpcs_in_flight=9 || { + error_noexit "Setting \"max_rpcs_in_flight\"" + return 1 + } + $r $LCTL conf_param $fsname-MDT0000.failover.node=$nid || { + error_noexit "Setting MDT \"failover.node\"" + return 1 + } + $r $LCTL pool_new $fsname.interop || { + error_noexit "Setting \"interop\"" + return 1 + } + $r $LCTL conf_param $fsname-MDT0000.lov.stripesize=4M || { + error_noexit "Setting \"lov.stripesize\"" + return 1 + } + + if [ "$writeconf" ]; then + mount -t lustre $nid:/$fsname $tmp/mnt/lustre || { + error_noexit "Mounting the client" + return 1 + } + shall_cleanup_lustre=true + $LCTL set_param debug="$PTLDEBUG" + + if $r test -f $tmp/sha1sums; then + $r sort -k 2 $tmp/sha1sums >$tmp/sha1sums.orig + pushd $tmp/mnt/lustre + find ! -name .lustre -type f -exec sha1sum {} \; | + sort -k 2 >$tmp/sha1sums || { + error_noexit "sha1sum" + return 1 + } + popd + if ! diff -ub $tmp/sha1sums.orig $tmp/sha1sums; then + error_noexit "sha1sum verification failed" + return 1 + fi + else + echo "sha1sum verification skipped" + fi - local device=`h2$NETTYPE $HOSTNAME`:/$NEWNAME - echo "Starting local client: $HOSTNAME: $device $MOUNT" - mount -t lustre $device $MOUNT || return 1 + if $r test -f $tmp/list; then + # + # There is not a Test Framework API to copy files to or + # from a remote node. + # + $r sort -k 6 $tmp/list >$tmp/list.orig + pushd $tmp/mnt/lustre + ls -Rni --time-style=+%s | sort -k 6 >$tmp/list || { + error_noexit "ls" + return 1 + } + popd + # + # 32-bit and 64-bit clients use different algorithms to + # convert FIDs into inode numbers. Hence, remove the inode + # numbers from the lists, if the original list was created + # on an architecture with different number of bits per + # "long". + # + if [ $(t32_bits_per_long $(uname -m)) != \ + $(t32_bits_per_long $img_arch) ]; then + echo "Different number of bits per \"long\" from the disk image" + for list in list.orig list; do + sed -i -e 's/^[0-9]\+[ \t]\+//' $tmp/$list + done + fi + if ! diff -ub $tmp/list.orig $tmp/list; then + error_noexit "list verification failed" + return 1 + fi + else + echo "list verification skipped" + fi - local old=$($LCTL get_param -n mdc.*.max_rpcs_in_flight) - local new=$((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" $new || return 11 + # + # When adding new data verification tests, please check for + # the presence of the required reference files first, like + # the "sha1sums" and "list" tests above, to avoid the need to + # regenerate every image for each test addition. + # + + nrpcs_orig=$($LCTL get_param -n mdc.*.max_rpcs_in_flight) || { + error_noexit "Getting \"max_rpcs_in_flight\"" + return 1 + } + nrpcs=$((nrpcs_orig + 5)) + $r $LCTL conf_param $fsname-MDT0000.mdc.max_rpcs_in_flight=$nrpcs || { + error_noexit "Changing \"max_rpcs_in_flight\"" + return 1 + } + wait_update $HOSTNAME "$LCTL get_param -n mdc.*.max_rpcs_in_flight" \ + $nrpcs || { + error_noexit "Verifying \"max_rpcs_in_flight\"" + return 1 + } + + umount $tmp/mnt/lustre || { + error_noexit "Unmounting the client" + return 1 + } + shall_cleanup_lustre=false + else + $r umount -d $tmp/mnt/mdt || { + error_noexit "Unmounting the MDT" + return 1 + } + shall_cleanup_mdt=false + $r umount -d $tmp/mnt/ost || { + error_noexit "Unmounting the OST" + return 1 + } + shall_cleanup_ost=false + + t32_reload_modules $node || { + error_noexit "Reloading modules" + return 1 + } + + # mount a second time to make sure we didnt leave upgrade flag on + $r $TUNEFS --dryrun $tmp/mdt || { + error_noexit "tunefs.lustre before remounting the MDT" + return 1 + } + $r mount -t lustre -o loop,exclude=$fsname-OST0000 $tmp/mdt \ + $tmp/mnt/mdt || { + error_noexit "Remounting the MDT" + return 1 + } + shall_cleanup_mdt=true + fi +} - [ "$(cksum $MOUNT/passwd | cut -d' ' -f 1,2)" == "94306271 1478" ] || return 12 - echo "ok." +test_32a() { + local tarballs + local tarball + local rc=0 - cleanup_32 + t32_check + for tarball in $tarballs; do + t32_test $tarball || rc=$? + done + return $rc +} +run_test 32a "Upgrade (not live)" - rm -rf $tmpdir || true # true is only for TMP on NFS +test_32b() { + local tarballs + local tarball + local rc=0 + + t32_check + for tarball in $tarballs; do + t32_test $tarball writeconf || rc=$? + done + return $rc } -run_test 32b "Upgrade from 1.8 with writeconf" +run_test 32b "Upgrade with writeconf" test_33a() { # bug 12333, was test_33 local rc=0 local FSNAME2=test-123 local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) + local mkfsoptions [ -n "$ost1_HOST" ] && fs2ost_HOST=$ost1_HOST if [ -z "$fs2ost_DEV" -o -z "$fs2mds_DEV" ]; then - do_facet $SINGLEMDS [ -b "$MDSDEV" ] && \ + local dev=${SINGLEMDS}_dev + local MDSDEV=${!dev} + is_blkdev $SINGLEMDS $MDSDEV && \ skip_env "mixed loopback and real device not working" && return fi - local fs2mdsdev=${fs2mds_DEV:-${MDSDEV}_2} - local fs2ostdev=${fs2ost_DEV:-$(ostdevname 1)_2} - add fs2mds $MDS_MKFS_OPTS --mkfsoptions='\"-J size=8\"' --fsname=${FSNAME2} --reformat $fs2mdsdev || exit 10 - add fs2ost $OST_MKFS_OPTS --fsname=${FSNAME2} --index=8191 --mgsnode=$MGSNID --reformat $fs2ostdev || exit 10 + local fs2mdsdev=$(mdsdevname 1_2) + local fs2ostdev=$(ostdevname 1_2) + local fs2mdsvdev=$(mdsvdevname 1_2) + local fs2ostvdev=$(ostvdevname 1_2) + + if [ $(facet_fstype mds1) == ldiskfs ]; then + mkfsoptions="--mkfsoptions=\\\"-J size=8\\\"" # See bug 17931. + fi + + add fs2mds $(mkfs_opts mds1) --fsname=${FSNAME2} --reformat \ + $mkfsoptions $fs2mdsdev $fs2mdsvdev || exit 10 + add fs2ost $(mkfs_opts ost1) --mgsnode=$MGSNID --fsname=${FSNAME2} \ + --index=8191 --reformat $fs2ostdev $fs2ostvdev || exit 10 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS && trap cleanup_24a EXIT INT start fs2ost $fs2ostdev $OST_MOUNT_OPTS - do_facet mgs "$LCTL conf_param sys.$FSNAME2.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." @@ -1235,7 +1656,6 @@ test_33a() { # bug 12333, was test_33 umount -d $MOUNT2 stop fs2ost -f stop fs2mds -f - rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev cleanup_nocli || rc=6 return $rc } @@ -1246,7 +1666,7 @@ test_33b() { # was test_34 do_facet client dd if=/dev/zero of=$MOUNT/24 bs=1024k count=1 # Drop lock cancelation reply during umount - #define OBD_FAIL_LDLM_CANCEL 0x304 + #define OBD_FAIL_LDLM_CANCEL_NET 0x304 do_facet client lctl set_param fail_loc=0x80000304 #lctl set_param debug=-1 umount_client $MOUNT @@ -1266,7 +1686,7 @@ test_34a() { sleep 1 cleanup } -run_test 34a "umount with opened file should fail" +run_test 34a "umount with opened file should be fail" test_34b() { @@ -1274,7 +1694,7 @@ test_34b() { touch $DIR/$tfile || return 1 stop_mds --force || return 2 - manual_umount_client --force + manual_umount_client --force rc=$? if [ $rc -ne 0 ]; then error "mtab after failed umount - rc $rc" @@ -1290,7 +1710,7 @@ test_34c() { touch $DIR/$tfile || return 1 stop_ost --force || return 2 - manual_umount_client --force + manual_umount_client --force rc=$? if [ $rc -ne 0 ]; then error "mtab after failed umount - rc $rc" @@ -1310,7 +1730,8 @@ test_35a() { # bug 12459 log "Set up a fake failnode for the MDS" FAKENID="127.0.0.2" local device=$(do_facet $SINGLEMDS "lctl get_param -n devices" | awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1) - do_facet $SINGLEMDS $LCTL conf_param mdc.${device}.failover.node=$FAKENID || return 4 + do_facet mgs "$LCTL conf_param ${device}.failover.node=" \ + "$(h2$NETTYPE $FAKENID)" || return 4 log "Wait for RECONNECT_INTERVAL seconds (10s)" sleep 10 @@ -1362,31 +1783,33 @@ test_35b() { # bug 18674 log "Set up a fake failnode for the MDS" FAKENID="127.0.0.2" - local device=$(do_facet mds "$LCTL get_param -n devices" | \ + local device=$(do_facet $SINGLEMDS "$LCTL get_param -n devices" | \ awk '($3 ~ "mdt" && $4 ~ "MDT") { print $4 }' | head -1) - do_facet mds "$LCTL conf_param mdc.${device}.failover.node=$FAKENID" || \ - return 1 + do_facet mgs "$LCTL conf_param ${device}.failover.node=" \ + "$(h2$NETTYPE $FAKENID)" || return 1 local at_max_saved=0 - # adaptive timeouts may prevent seeing the issue + # adaptive timeouts may prevent seeing the issue if at_is_enabled; then at_max_saved=$(at_max_get mds) at_max_set 0 mds client fi - mkdir -p $MOUNT/testdir - touch $MOUNT/testdir/test + mkdir -p $MOUNT/$tdir log "Injecting EBUSY on MDS" # Setting OBD_FAIL_MDS_RESEND=0x136 - do_facet mds "$LCTL set_param fail_loc=0x80000136" || return 2 + do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x80000136" || return 2 - log "Stat on a test file" - stat $MOUNT/testdir/test + $LCTL set_param mdc.${FSNAME}*.stats=clear + + log "Creating a test file and stat it" + touch $MOUNT/$tdir/$tfile + stat $MOUNT/$tdir/$tfile log "Stop injecting EBUSY on MDS" - do_facet mds "$LCTL set_param fail_loc=0" || return 3 - rm -f $MOUNT/testdir/test + do_facet $SINGLEMDS "$LCTL set_param fail_loc=0" || return 3 + rm -f $MOUNT/$tdir/$tfile log "done" # restore adaptive timeout @@ -1394,6 +1817,8 @@ test_35b() { # bug 18674 $LCTL dk $TMP/lustre-log-$TESTNAME.log + CONNCNT=`$LCTL get_param mdc.${FSNAME}*.stats | awk '/mds_connect/{print $2}'` + # 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;} @@ -1413,38 +1838,52 @@ test_35b() { # bug 18674 log "ERROR: The client tried to reconnect to the failover server while the primary was busy" && \ return 5 - cleanup + # LU-290 + # When OBD_FAIL_MDS_RESEND is hit, we sleep for 2 * obd_timeout + # Reconnects are supposed to be rate limited to one every 5s + [ $CONNCNT -gt $((2 * $TIMEOUT / 5 + 1)) ] && \ + log "ERROR: Too many reconnects $CONNCNT" && \ + return 6 + + cleanup # remove nid settings writeconf } run_test 35b "Continue reconnection retries, if the active server is busy" test_36() { # 12743 - local rc + [ $OSTCOUNT -lt 2 ] && skip_env "skipping test for single OST" && return + + [ "$ost_HOST" = "`hostname`" -o "$ost1_HOST" = "`hostname`" ] || \ + { skip "remote OST" && return 0; } + + local rc=0 local FSNAME2=test1234 local fs3ost_HOST=$ost_HOST local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) [ -n "$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" ] && \ + is_blkdev $SINGLEMDS $MDSDEV && \ skip_env "mixed loopback and real device not working" && return fi - [ $OSTCOUNT -lt 2 ] && skip_env "skipping test for single OST" && return - - [ "$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 - # 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 $fs3ostdev || exit 10 + local fs2mdsdev=$(mdsdevname 1_2) + local fs2ostdev=$(ostdevname 1_2) + local fs3ostdev=$(ostdevname 2_2) + local fs2mdsvdev=$(mdsvdevname 1_2) + local fs2ostvdev=$(ostvdevname 1_2) + local fs3ostvdev=$(ostvdevname 2_2) + + add fs2mds $(mkfs_opts mds1) --fsname=${FSNAME2} --reformat \ + $fs2mdsdev $fs2mdsvdev || exit 10 + # XXX after we support non 4K disk blocksize in ldiskfs, specify a + # different one than the default value here. + add fs2ost $(mkfs_opts ost1) --mgsnode=$MGSNID --fsname=${FSNAME2} \ + --reformat $fs2ostdev $fs2ostvdev || exit 10 + add fs3ost $(mkfs_opts ost1) --mgsnode=$MGSNID --fsname=${FSNAME2} \ + --reformat $fs3ostdev $fs3ostvdev || exit 10 start fs2mds $fs2mdsdev $MDS_MOUNT_OPTS start fs2ost $fs2ostdev $OST_MOUNT_OPTS @@ -1487,40 +1926,58 @@ test_36() { # 12743 stop fs3ost -f || return 200 stop fs2ost -f || return 201 stop fs2mds -f || return 202 - rm -rf $MOUNT2 $fs2mdsdev $fs2ostdev $fs3ostdev - unload_modules || return 203 + unload_modules_conf || return 203 return $rc } run_test 36 "df report consistency on OSTs with different block size" test_37() { - client_only && skip "client only testing" && return 0 - LOCAL_MDSDEV="$TMP/mdt.img" - SYM_MDSDEV="$TMP/sym_mdt.img" + local mntpt=$(facet_mntpt $SINGLEMDS) + local mdsdev=$(mdsdevname ${SINGLEMDS//mds/}) + local mdsdev_sym="$TMP/sym_mdt.img" + local opts=$MDS_MOUNT_OPTS + local rc=0 + + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Currently only applicable to ldiskfs-based MDTs" + return + fi - echo "MDS : $LOCAL_MDSDEV" - echo "SYMLINK : $SYM_MDSDEV" - rm -f $LOCAL_MDSDEV + echo "MDS : $mdsdev" + echo "SYMLINK : $mdsdev_sym" + do_facet $SINGLEMDS rm -f $mdsdev_sym - 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 + do_facet $SINGLEMDS ln -s $mdsdev $mdsdev_sym + + echo "mount symlink device - $mdsdev_sym" + + if ! do_facet $SINGLEMDS test -b $mdsdev; then + opts=$(csa_add "$opts" -o loop) + fi + mount_op=$(do_facet $SINGLEMDS mount -v -t lustre $opts \ + $mdsdev_sym $mntpt 2>&1) + rc=${PIPESTATUS[0]} - echo "mount symlink device - $SYM_MDSDEV" + echo mount_op=$mount_op - 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 + do_facet $SINGLEMDS "umount -d $mntpt && rm -f $mdsdev_sym" - if [ -n "$mount_op" ]; then - error "**** FAIL: set tunables failed for symlink device" + if $(echo $mount_op | grep -q "unable to set tunable"); then + error "set tunables failed for symlink device" fi + + [ $rc -eq 0 ] || error "mount symlink $mdsdev_sym failed! rc=$rc" + return 0 } run_test 37 "verify set tunables works for symlink device" test_38() { # bug 14222 + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi + setup # like runtests COUNT=10 @@ -1589,13 +2046,19 @@ 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" + do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x80000706" start_mds cleanup } run_test 40 "race during service thread startup" -test_41() { #bug 14134 +test_41a() { #bug 14134 + if [ $(facet_fstype $SINGLEMDS) == ldiskfs ] && + ! do_facet $SINGLEMDS test -b $(mdsdevname 1); then + skip "Loop devices does not work with nosvc option" + return + fi + local rc local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) @@ -1613,15 +2076,46 @@ test_41() { #bug 14134 stop ost1 -f || return 201 stop_mds -f || return 202 stop_mds -f || return 203 - unload_modules || return 204 + unload_modules_conf || return 204 return $rc } -run_test 41 "mount mds with --nosvc and --nomgs" +run_test 41a "mount mds with --nosvc and --nomgs" + +test_41b() { + if [ $(facet_fstype $SINGLEMDS) == ldiskfs ] && + ! do_facet $SINGLEMDS test -b $(mdsdevname 1); then + skip "Loop devices does not work with nosvc option" + return + fi + + ! combined_mgs_mds && skip "needs combined mgs device" && return 0 + + stopall + reformat + local MDSDEV=$(mdsdevname ${SINGLEMDS//mds/}) + + start $SINGLEMDS $MDSDEV $MDS_MOUNT_OPTS -o nosvc -n + start_ost + 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 || return 200 + + umount_client $MOUNT + stop_ost || return 201 + stop_mds -f || return 202 + stop_mds -f || return 203 + +} +run_test 41b "mount mds with --nosvc and --nomgs on first mount" test_42() { #bug 14693 setup check_mount || return 2 - do_facet mgs $LCTL conf_param llite.$FSNAME.some_wrong_param=10 + do_facet mgs $LCTL conf_param lustre.llite.some_wrong_param=10 umount_client $MOUNT mount_client $MOUNT || return 1 cleanup @@ -1630,17 +2124,17 @@ test_42() { #bug 14693 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" \ - "mdt.$FSNAME-MDT*.root_squash" \ - "0:0" - set_and_check mds \ - "lctl get_param -n mdt.$FSNAME-MDT0000.nosquash_nids" \ - "mdt.$FSNAME-MDT*.nosquash_nids" \ - "NONE" + [ $UID -ne 0 -o $RUNAS_ID -eq 0 ] && skip_env "run as root" + setup + chmod ugo+x $DIR || error "chmod 0 failed" + set_conf_param_and_check mds \ + "lctl get_param -n mdt.$FSNAME-MDT0000.root_squash" \ + "$FSNAME.mdt.root_squash" \ + "0:0" + set_conf_param_and_check mds \ + "lctl get_param -n mdt.$FSNAME-MDT0000.nosquash_nids" \ + "$FSNAME.mdt.nosquash_nids" \ + "NONE" # # create set of test files @@ -1656,15 +2150,15 @@ test_43() { 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.$FSNAME-MDT0000.root_squash" \ - "mdt.$FSNAME-MDT*.root_squash" \ - "$RUNAS_ID:$RUNAS_ID" + # + # check root_squash: + # set root squash UID:GID to RUNAS_ID + # root should be able to access only files owned by RUNAS_ID + # + set_conf_param_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 || \ @@ -1695,18 +2189,18 @@ test_43() { 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" \ - "mdt.$FSNAME-MDT*.nosquash_nids" \ - "$NIDLIST" + # + # 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 $NIDLIST | tr -s ' ' ' ') + set_conf_param_and_check mds \ + "lctl get_param -n mdt.$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 || \ @@ -1738,7 +2232,7 @@ test_44() { # 16317 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") + UUIDS=$(do_facet $SINGLEMDS "$LCTL get_param mdt.${FSNAME}*.exports.*.uuid") for VAL in $UUIDS; do NID=$(echo $VAL | cut -d= -f1) CLUUID=$(echo $VAL | cut -d= -f2) @@ -1781,7 +2275,7 @@ cleanup_46a() { stop ost${count} -f || rc=$? let count=count-1 done - stop_mds || rc=$? + stop_mds || rc=$? cleanup_nocli || rc=$? #writeconf to remove all ost2 traces for subsequent tests writeconf @@ -1799,7 +2293,7 @@ test_46a() { mount_client $MOUNT || return 3 trap "cleanup_46a $OSTCOUNT" EXIT ERR - local i + local i for (( i=2; i<=$OSTCOUNT; i++ )); do start ost$i `ostdevname $i` $OST_MOUNT_OPTS || return $((i+2)) done @@ -1807,13 +2301,13 @@ test_46a() { # wait until osts in sync for (( i=2; i<=$OSTCOUNT; i++ )); do wait_osc_import_state mds ost$i FULL + wait_osc_import_state client ost$i FULL done - #second client see all ost's mount_client $MOUNT2 || return 8 - $LFS setstripe $MOUNT2 -c -1 || return 9 + $LFS setstripe -c -1 $MOUNT2 || return 9 $LFS getstripe $MOUNT2 || return 10 echo "ok" > $MOUNT2/widestripe @@ -1881,9 +2375,9 @@ cleanup_48() { test_48() { # bug 17636 reformat setup_noconfig - check_mount || return 2 + check_mount || return 2 - $LFS setstripe $MOUNT -c -1 || return 9 + $LFS setstripe -c -1 $MOUNT || return 9 $LFS getstripe $MOUNT || return 10 echo "ok" > $MOUNT/widestripe @@ -1905,19 +2399,19 @@ run_test 48 "too many acls on file" # check PARAM_SYS_LDLM_TIMEOUT option of MKFS.LUSTRE test_49() { # bug 17710 - local OLD_MDS_MKFS_OPTS=$MDS_MKFS_OPTS - local OLD_OST_MKFS_OPTS=$OST_MKFS_OPTS + local timeout_orig=$TIMEOUT + local ldlm_timeout_orig=$LDLM_TIMEOUT local LOCAL_TIMEOUT=20 - - OST_MKFS_OPTS="--ost --fsname=$FSNAME --device-size=$OSTSIZE --mgsnode=$MGSNID --param sys.timeout=$LOCAL_TIMEOUT --param sys.ldlm_timeout=$LOCAL_TIMEOUT $MKFSOPT $OSTOPT" + LDLM_TIMEOUT=$LOCAL_TIMEOUT + TIMEOUT=$LOCAL_TIMEOUT reformat setup_noconfig check_mount || return 1 echo "check ldlm_timout..." - LDLM_MDS="`do_facet mds lctl get_param -n ldlm_timeout`" + LDLM_MDS="`do_facet $SINGLEMDS lctl get_param -n ldlm_timeout`" LDLM_OST1="`do_facet ost1 lctl get_param -n ldlm_timeout`" LDLM_CLIENT="`do_facet client lctl get_param -n ldlm_timeout`" @@ -1933,13 +2427,13 @@ test_49() { # bug 17710 stop_ost || return 2 stop_mds || return 3 - OST_MKFS_OPTS="--ost --fsname=$FSNAME --device-size=$OSTSIZE --mgsnode=$MGSNID --param sys.timeout=$LOCAL_TIMEOUT --param sys.ldlm_timeout=$((LOCAL_TIMEOUT - 1)) $MKFSOPT $OSTOPT" + LDLM_TIMEOUT=$((LOCAL_TIMEOUT - 1)) reformat setup_noconfig check_mount || return 7 - LDLM_MDS="`do_facet mds lctl get_param -n ldlm_timeout`" + LDLM_MDS="`do_facet $SINGLEMDS 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`" @@ -1953,8 +2447,8 @@ test_49() { # bug 17710 cleanup || return $? - MDS_MKFS_OPTS=$OLD_MDS_MKFS_OPTS - OST_MKFS_OPTS=$OLD_OST_MKFS_OPTS + LDLM_TIMEOUT=$ldlm_timeout_orig + TIMEOUT=$timeout_orig } run_test 49 "check PARAM_SYS_LDLM_TIMEOUT option of MKFS.LUSTRE" @@ -2022,9 +2516,11 @@ test_50c() { wait_osc_import_state mds ost DISCONN lazystatfs $MOUNT || error "lazystatfs failed with one down server" - umount_client $MOUNT || error "Unable to unmount client" + umount_client $MOUNT || error "Unable to unmount client" stop_ost2 || error "Unable to stop OST2" stop_mds || error "Unable to stop MDS" + #writeconf to remove all ost2 traces for subsequent tests + writeconf } run_test 50c "lazystatfs one server down ==========================" @@ -2042,9 +2538,11 @@ test_50d() { stop_ost || error "Unable to stop OST1" lazystatfs $MOUNT || error "lazystatfs failed with one down server" - umount_client $MOUNT || error "Unable to unmount client" + umount_client $MOUNT || error "Unable to unmount client" stop_ost2 || error "Unable to stop OST2" stop_mds || error "Unable to stop MDS" + #writeconf to remove all ost2 traces for subsequent tests + writeconf } run_test 50d "lazystatfs client/server conn race ==========================" @@ -2134,6 +2632,8 @@ test_50g() { [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2, skipping" && return setup start_ost2 || error "Unable to start OST2" + wait_osc_import_state mds ost2 FULL + wait_osc_import_state client ost2 FULL local PARAM="${FSNAME}-OST0001.osc.active" @@ -2174,6 +2674,8 @@ test_51() { wait $pid stop_ost2 || return 3 cleanup + #writeconf to remove all ost2 traces for subsequent tests + writeconf } run_test 51 "Verify that mdt_reint handles RMF_MDT_MD correctly when an OST is added" @@ -2226,6 +2728,11 @@ diff_files_xattrs() } test_52() { + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi + start_mds [ $? -eq 0 ] || { error "Unable to start MDS"; return 1; } start_ost @@ -2234,19 +2741,22 @@ test_52() { [ $? -eq 0 ] || { error "Unable to mount client"; return 3; } local nrfiles=8 - local ost1mnt=${MOUNT%/*}/ost1 + local ost1mnt=$(facet_mntpt ost1) local ost1node=$(facet_active_host ost1) local ost1tmp=$TMP/conf52 + local loop 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 $ost1tmp/modified_first" + local mtime=$(stat -c %Y $TMP/modified_first) + do_node $ost1node "mkdir -p $ost1tmp && touch -m -d @$mtime $ost1tmp/modified_first" + [ $? -eq 0 ] || { error "Unable to create temporary file"; return 6; } sleep 1 - $LFS setstripe $DIR/$tdir -c -1 -s 1M + $LFS setstripe -c -1 -S 1M $DIR/$tdir [ $? -eq 0 ] || { error "lfs setstripe failed"; return 7; } for (( i=0; i < nrfiles; i++ )); do @@ -2268,7 +2778,13 @@ test_52() { [ $? -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 + do_node $ost1node mkdir -p $ost1mnt + [ $? -eq 0 ] || { error "Unable to create $ost1mnt"; return 23; } + if ! do_node $ost1node test -b $ost1_dev; then + loop="-o loop" + fi + do_node $ost1node mount -t $(facet_fstype ost1) $loop $ost1_dev \ + $ost1mnt [ $? -eq 0 ] || { error "Unable to mount ost1 as ldiskfs"; return 12; } # backup objects @@ -2290,7 +2806,7 @@ test_52() { 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" + do_node $ost1node "umount $ost1mnt" [ $? -eq 0 ] || { error "Unable to umount ost1 as ldiskfs"; return 17; } start_ost @@ -2319,6 +2835,7 @@ thread_sanity() { local facet=$2 local parampat=$3 local opts=$4 + local basethr=$5 local tmin local tmin2 local tmax @@ -2326,6 +2843,8 @@ thread_sanity() { local tstarted local paramp local msg="Insane $modname thread counts" + local ncpts=$(check_cpt_number $facet) + local nthrs shift 4 setup @@ -2333,8 +2852,7 @@ thread_sanity() { # We need to expand $parampat, but it may match multiple parameters, so # we'll pick the first one - paramp=$(do_facet $facet "lctl get_param -N ${parampat}.threads_min"|head -1) - if [ -z "$paramp" ]; then + if ! paramp=$(do_facet $facet "lctl get_param -N ${parampat}.threads_min"|head -1); then error "Couldn't expand ${parampat}.threads_min parameter name" return 22 fi @@ -2347,17 +2865,27 @@ thread_sanity() { tmax=$(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 $? + lassert 24 "$msg" '(($tstarted >= $tmin && $tstarted <= $tmax ))' || return $? + nthrs=$(expr $tmax - $tmin) + if [ $nthrs -lt $ncpts ]; then + nthrs=0 + else + nthrs=$ncpts + fi - # 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 $? + [ $tmin -eq $tmax -a $tmin -eq $tstarted ] && + skip_env "module parameter forced $facet thread count" && + tmin=3 && tmax=$((3 * tmax)) + + # Check that we can change min/max + do_facet $facet "lctl set_param ${paramp}.threads_min=$((tmin + nthrs))" + do_facet $facet "lctl set_param ${paramp}.threads_max=$((tmax - nthrs))" + 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 + $nthrs) && $tmax2 == ($tmax - $nthrs)))' || return $? # Check that we can set min/max to the same value - do_facet $facet "lctl set_param ${paramp}.threads_min=$tmin" + tmin=$(do_facet $facet "lctl get_param -n ${paramp}.threads_min" || echo 0) 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) @@ -2367,14 +2895,15 @@ thread_sanity() { 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 $? + lassert 27 "$msg" '(($tmin2 <= $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 + local newvalue="${opts}=$(expr $basethr \* $ncpts)" + setmodopts -a $modname "$newvalue" oldvalue load_modules setup @@ -2390,51 +2919,45 @@ thread_sanity() { 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 load_modules setup cleanup } test_53a() { - thread_sanity OST ost1 'ost.*.ost' 'oss_num_threads=64' + thread_sanity OST ost1 'ost.*.ost' 'oss_num_threads' '16' } run_test 53a "check OSS thread count params" test_53b() { - thread_sanity MDT $SINGLEMDS 'mdt.*.*.' 'mdt_num_threads=64' + thread_sanity MDT $SINGLEMDS 'mdt.*.*.' 'mdt_num_threads' '16' } run_test 53b "check MDT thread count params" -run_llverfs() -{ - local dir=$1 - local partial_arg="" - local size=$(df -B G $dir | tail -1 | awk '{print $2}' | sed 's/G//') # Gb - - # Run in partial (fast) mode if the size - # of a partition > 10 GB - [ $size -gt 10 ] && partial_arg="-p" - - llverfs $partial_arg $dir -} - test_54a() { - do_rpc_nodes $(facet_host ost1) run_llverdev $(ostdevname 1) + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi + + do_rpc_nodes $(facet_host ost1) run_llverdev $(ostdevname 1) -p [ $? -eq 0 ] || error "llverdev failed!" reformat_and_config } -run_test 54a "llverdev" +run_test 54a "test llverdev and partial verify of device" test_54b() { + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi + setup - run_llverfs $MOUNT + run_llverfs $MOUNT -p [ $? -eq 0 ] || error "llverfs failed!" cleanup } -run_test 54b "llverfs" +run_test 54b "test llverfs and partial verify of filesystem" lov_objid_size() { @@ -2443,20 +2966,25 @@ lov_objid_size() } test_55() { + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi + local mdsdev=$(mdsdevname 1) - local ostdev=$(ostdevname 1) - local saved_opts=$OST_MKFS_OPTS + local mdsvdev=$(mdsvdevname 1) - for i in 0 1023 2048 + for i in 1023 2048 do - OST_MKFS_OPTS="$saved_opts --index $i" - reformat - + add mds1 $(mkfs_opts mds1) --reformat $mdsdev $mdsvdev || + exit 10 + add ost1 $(mkfs_opts ost1) --index=$i --reformat \ + $(ostdevname 1) $(ostvdevname 1) setup_noconfig stopall - - setup + setup_noconfig sync + echo checking size of lov_objid for ost index $i LOV_OBJID_SIZE=$(do_facet mds1 "$DEBUGFS -R 'stat lov_objid' $mdsdev 2>/dev/null" | grep ^User | awk '{print $6}') if [ "$LOV_OBJID_SIZE" != $(lov_objid_size $i) ]; then @@ -2467,42 +2995,81 @@ test_55() { stopall done - OST_MKFS_OPTS=$saved_opts reformat } run_test 55 "check lov_objid size" test_56() { - add mds1 $MDS_MKFS_OPTS --mkfsoptions='\"-J size=16\"' --reformat $(mdsdevname 1) - add ost1 $OST_MKFS_OPTS --index=1000 --reformat $(ostdevname 1) - add ost2 $OST_MKFS_OPTS --index=10000 --reformat $(ostdevname 2) + local mds_journal_size_orig=$MDSJOURNALSIZE - start_mds + MDSJOURNALSIZE=16 + add mds1 $(mkfs_opts mds1) --reformat $(mdsdevname 1) $(mdsvdevname 1) + add ost1 $(mkfs_opts ost1) --index=1000 --reformat \ + $(ostdevname 1) $(ostvdevname 1) + add ost2 $(mkfs_opts ost2) --index=10000 --reformat \ + $(ostdevname 2) $(ostvdevname 2) + + start_mgsmds start_ost start_ost2 || error "Unable to start second ost" mount_client $MOUNT || error "Unable to mount client" echo ok $LFS osts stopall + MDSJOURNALSIZE=$mds_journal_size_orig reformat } run_test 56 "check big indexes" -test_57() { # bug 22656 +test_57a() { # bug 22656 local NID=$(do_facet ost1 "$LCTL get_param nis" | tail -1 | awk '{print $1}') writeconf do_facet ost1 "$TUNEFS --failnode=$NID `ostdevname 1`" || error "tunefs failed" start_mgsmds start_ost && error "OST registration from failnode should fail" - stop_mds reformat } -run_test 57 "initial registration from failnode should fail (should return errs)" +run_test 57a "initial registration from failnode should fail (should return errs)" + +test_57b() { + local NID=$(do_facet ost1 "$LCTL get_param nis" | tail -1 | awk '{print $1}') + writeconf + do_facet ost1 "$TUNEFS --servicenode=$NID `ostdevname 1`" || error "tunefs failed" + start_mgsmds + start_ost || error "OST registration from servicenode should not fail" + reformat +} +run_test 57b "initial registration from servicenode should not fail" count_osts() { do_facet mgs $LCTL get_param mgs.MGS.live.$FSNAME | grep OST | wc -l } +test_58() { # bug 22658 + if [ $(facet_fstype mds) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi + setup_noconfig + mkdir -p $DIR/$tdir + createmany -o $DIR/$tdir/$tfile-%d 100 + # make sure that OSTs do not cancel llog cookies before we unmount the MDS +#define OBD_FAIL_OBD_LOG_CANCEL_NET 0x601 + do_facet mds "lctl set_param fail_loc=0x601" + unlinkmany $DIR/$tdir/$tfile-%d 100 + stop mds + local MNTDIR=$(facet_mntpt mds) + # remove all files from the OBJECTS dir + do_facet mds "mount -t ldiskfs $MDSDEV $MNTDIR" + do_facet mds "find $MNTDIR/OBJECTS -type f -delete" + do_facet mds "umount $MNTDIR" + # restart MDS with missing llog files + start_mds + do_facet mds "lctl set_param fail_loc=0" + reformat +} +run_test 58 "missing llog files must not prevent MDT from mounting" + test_59() { start_mgsmds >> /dev/null local C1=$(count_osts) @@ -2530,13 +3097,163 @@ test_59() { [ $C4 -eq 2 ] || error "OST2 writeconf should add log" stop_ost2 >> /dev/null cleanup_nocli >> /dev/null + #writeconf to remove all ost2 traces for subsequent tests + writeconf } run_test 59 "writeconf mount option" +test_60() { # LU-471 + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi + + add mds1 $(mkfs_opts mds1) \ + --mkfsoptions='\" -E stride=64 -O ^uninit_bg\"' --reformat \ + $(mdsdevname 1) $(mdsvdevname 1) || exit 10 + + dump=$(do_facet $SINGLEMDS dumpe2fs $(mdsdevname 1)) + rc=${PIPESTATUS[0]} + [ $rc -eq 0 ] || error "dumpe2fs $(mdsdevname 1) failed" + + # MDT default has dirdata feature + echo $dump | grep dirdata > /dev/null || error "dirdata is not set" + # we disable uninit_bg feature + echo $dump | grep uninit_bg > /dev/null && error "uninit_bg is set" + # we set stride extended options + echo $dump | grep stride > /dev/null || error "stride is not set" + reformat +} +run_test 60 "check mkfs.lustre --mkfsoptions -E -O options setting" + +test_61() { # LU-80 + local reformat=false + + [ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.1.53) ] || + { skip "Need MDS version at least 2.1.53"; return 0; } + + if [ $(facet_fstype $SINGLEMDS) == ldiskfs ] && + ! large_xattr_enabled; then + reformat=true + local mds_dev=$(mdsdevname ${SINGLEMDS//mds/}) + LDISKFS_MKFS_OPTS+=" -O large_xattr" + add $SINGLEMDS $(mkfs_opts $SINGLEMDS) --reformat $mds_dev || + error "reformatting $mds_dev failed" + fi + + setup_noconfig || error "setting up the filesystem failed" + client_up || error "starting client failed" + + local file=$DIR/$tfile + touch $file + + local large_value="$(generate_string $(max_xattr_size))" + local small_value="bar" + + local name="trusted.big" + log "save large xattr $name on $file" + setfattr -n $name -v $large_value $file || + error "saving $name on $file failed" + + local new_value=$(get_xattr_value $name $file) + [[ "$new_value" != "$large_value" ]] && + error "$name different after saving" + + log "shrink value of $name on $file" + setfattr -n $name -v $small_value $file || + error "shrinking value of $name on $file failed" + + new_value=$(get_xattr_value $name $file) + [[ "$new_value" != "$small_value" ]] && + error "$name different after shrinking" + + log "grow value of $name on $file" + setfattr -n $name -v $large_value $file || + error "growing value of $name on $file failed" + + new_value=$(get_xattr_value $name $file) + [[ "$new_value" != "$large_value" ]] && + error "$name different after growing" + + log "check value of $name on $file after remounting MDS" + fail $SINGLEMDS + new_value=$(get_xattr_value $name $file) + [[ "$new_value" != "$large_value" ]] && + error "$name different after remounting MDS" + + log "remove large xattr $name from $file" + setfattr -x $name $file || error "removing $name from $file failed" + + rm -f $file + stopall + if $reformat; then + LDISKFS_MKFS_OPTS=${LDISKFS_MKFS_OPTS% -O large_xattr} + reformat + fi +} +run_test 61 "large xattr" + +test_62() { + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi + + # MRP-118 + local mdsdev=$(mdsdevname 1) + local ostdev=$(ostdevname 1) + + [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.2.51) ]] || + { skip "Need MDS version at least 2.2.51"; return 0; } + + echo "disable journal for mds" + do_facet mds tune2fs -O ^has_journal $mdsdev || error "tune2fs failed" + start_mds && error "MDT start should fail" + echo "disable journal for ost" + do_facet ost1 tune2fs -O ^has_journal $ostdev || error "tune2fs failed" + start_ost && error "OST start should fail" + cleanup || return $? + reformat_and_config +} +run_test 62 "start with disabled journal" + +test_63() { + if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then + skip "Only applicable to ldiskfs-based MDTs" + return + fi + + local inode_slab=$(do_facet $SINGLEMDS \ + "awk '/ldiskfs_inode_cache/ { print \\\$5 }' /proc/slabinfo") + if [ -z "$inode_slab" ]; then + skip "ldiskfs module has not been loaded" + return + fi + + echo "$inode_slab ldisk inodes per page" + [ "$inode_slab" -ge "3" ] || + error "ldisk inode size is too big, $inode_slab objs per page" + return +} +run_test 63 "Verify each page can at least hold 3 ldisk inodes" + +test_64() { + start_mds + start_ost + start_ost2 || error "Unable to start second ost" + mount_client $MOUNT || error "Unable to mount client" + stop_ost2 || error "Unable to stop second ost" + echo "$LFS df" + $LFS df --lazy || error "lfs df failed" + cleanup || return $? +} +run_test 64 "check lfs df --lazy " if ! combined_mgs_mds ; then stop mgs fi + cleanup_gss -equals_msg `basename $0`: test complete -[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true + +complete $SECONDS +exit_status