X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity-quota.sh;h=fb50169f7e9830f22895dffd6690db09108bca3f;hp=0c35e4135793c532dd7739a30b0717cab6d04e6f;hb=03402f45086f79b5f003ad76e93263cead84d9ea;hpb=cb5a84b6f2805c285b5572172034f32502698ca2 diff --git a/lustre/tests/sanity-quota.sh b/lustre/tests/sanity-quota.sh index 0c35e41..fb50169 100644 --- a/lustre/tests/sanity-quota.sh +++ b/lustre/tests/sanity-quota.sh @@ -19,7 +19,9 @@ export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/../utils:$PATH:/sbin ONLY=${ONLY:-"$*"} # test_11 has been used to protect a kernel bug(bz10912), now it isn't # useful any more. Then add it to ALWAYS_EXCEPT. b=19835 -ALWAYS_EXCEPT="10 $SANITY_QUOTA_EXCEPT" +# We have changed the mechanism of quota, test_12 is meanless now. +# b=20877 +ALWAYS_EXCEPT="10 12 $SANITY_QUOTA_EXCEPT" # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT! case `uname -r` in @@ -42,19 +44,24 @@ BUNIT_SZ=${BUNIT_SZ:-1024} # min block quota unit(kB) IUNIT_SZ=${IUNIT_SZ:-10} # min inode quota unit MAX_DQ_TIME=604800 MAX_IQ_TIME=604800 - -unset ENABLE_QUOTA +SANITY_QUOTA_USERS="quota15_1 quota15_2 quota15_3 quota15_4 quota15_5 quota15_6 \ + quota15_7 quota15_8 quota15_9 quota15_10 quota15_11 quota15_12 \ + quota15_13 quota15_14 quota15_15 quota15_16 quota15_17 quota15_18 \ + quota15_19 quota15_20 quota15_21 quota15_22 quota15_23 quota15_24 \ + quota15_25 quota15_26 quota15_27 quota15_28 quota15_29 quota15_30" TRACE=${TRACE:-""} LUSTRE=${LUSTRE:-`dirname $0`/..} . $LUSTRE/tests/test-framework.sh init_test_env $@ . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh} +init_logging DIRECTIO=${DIRECTIO:-$LUSTRE/tests/directio} [ $MDSCOUNT -gt 1 ] && skip "CMD case" && exit 0 -remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0 -remote_ost_nodsh && skip "remote OST with nodsh" && exit 0 + +require_dsh_mds || exit 0 +require_dsh_ost || exit 0 [ "$SLOW" = "no" ] && EXCEPT_SLOW="9 10 11 18b 21" @@ -65,20 +72,19 @@ QUOTALOG=${TESTSUITELOG:-$TMP/$(basename $0 .sh).log} DIR=${DIR:-$MOUNT} DIR2=${DIR2:-$MOUNT2} -check_and_setup_lustre +QUOTA_AUTO_OLD=$QUOTA_AUTO +export QUOTA_AUTO=0 -if [ x"$(som_check)" = x"enabled" ]; then - echo "Som is enabled, Quota is temporary conflicts with it" - check_and_cleanup_lustre - exit 0 -fi +check_and_setup_lustre LOVNAME=`lctl get_param -n llite.*.lov.common_name | tail -n 1` OSTCOUNT=`lctl get_param -n lov.$LOVNAME.numobd` SHOW_QUOTA_USER="$LFS quota -v -u $TSTUSR $DIR" +SHOW_QUOTA_USERID="$LFS quota -v -u $TSTID $DIR" SHOW_QUOTA_USER2="$LFS quota -v -u $TSTUSR2 $DIR" SHOW_QUOTA_GROUP="$LFS quota -v -g $TSTUSR $DIR" +SHOW_QUOTA_GROUPID="$LFS quota -v -g $TSTID $DIR" SHOW_QUOTA_GROUP2="$LFS quota -v -g $TSTUSR2 $DIR" SHOW_QUOTA_INFO_USER="$LFS quota -t -u $DIR" SHOW_QUOTA_INFO_GROUP="$LFS quota -t -g $DIR" @@ -190,15 +196,15 @@ resetquota() { fi while [ $((count++)) -lt $timeout ]; do - $LFS setquota "$1" "$2" -b 0 -B 0 -i 0 -I 0 $MOUNT - RC=$? + local RC=0 + OUTPUT=`$LFS setquota "$1" "$2" -b 0 -B 0 -i 0 -I 0 $MOUNT 2>&1` || RC=${PIPESTATUS[0]} if [ $RC -ne 0 ]; then - if [ $RC -eq 240 ]; then # 240 means -EBUSY - log "resetquota is blocked for quota master recovery, retry after 1 sec" - sleep 1 + if echo "$OUTPUT" | grep -q busy; then + log "resetquota is blocked for quota master recovery, retry after $((count * 3)) sec" + sleep 3 continue else - error "resetquota failed: $RC" + error "resetquota failed" fi fi break @@ -234,6 +240,36 @@ quota_log() { log "$*" } +# +# get quota info for a user or a group +# usage: getquota -u|-g | global| bhardlimit|bsoftlimit|bgrace|ihardlimit|isoftlimit|igrace +# +getquota() { + local spec + local uuid + + [ "$#" != 4 ] && error "getquota: wrong number of arguments: $#" + [ "$1" != "-u" -a "$1" != "-g" ] && error "getquota: wrong u/g specifier $1 passed" + + uuid="$3" + + case "$4" in + curspace) spec=1;; + bsoftlimit) spec=2;; + bhardlimit) spec=3;; + bgrace) spec=4;; + curinodes) spec=5;; + isoftlimit) spec=6;; + ihardlimit) spec=7;; + igrace) spec=8;; + *) error "unknown quota parameter $4";; + esac + + [ "$uuid" = "global" ] && uuid=$DIR + + $LFS quota -v "$1" "$2" $DIR | awk 'BEGIN { num='$spec' } { if ($1 == "'$uuid'") { if (NF == 1) { getline } else { num++ } ; print $num;} }' | tr -d "*" +} + quota_show_check() { LOCAL_BF=$1 LOCAL_UG=$2 @@ -243,7 +279,7 @@ quota_show_check() { $LFS quota -v -$LOCAL_UG $LOCAL_ID $DIR if [ "$LOCAL_BF" == "a" -o "$LOCAL_BF" == "b" ]; then - USAGE="`$LFS quota -$LOCAL_UG $LOCAL_ID $DIR | awk '/^.*'$PATTERN'.*[[:digit:]+][[:space:]+]/ { print $2 }'`" + USAGE=`getquota -$LOCAL_UG $LOCAL_ID global curspace` if [ -z $USAGE ]; then quota_error $LOCAL_UG $LOCAL_ID "System is error when query quota for block ($LOCAL_UG:$LOCAL_ID)." else @@ -252,7 +288,7 @@ quota_show_check() { fi if [ "$LOCAL_BF" == "a" -o "$LOCAL_BF" == "f" ]; then - USAGE="`$LFS quota -$LOCAL_UG $LOCAL_ID $DIR | awk '/^.*'$PATTERN'.*[[:digit:]+][[:space:]+]/ { print $5 }'`" + USAGE=`getquota -$LOCAL_UG $LOCAL_ID global curinodes` if [ -z $USAGE ]; then quota_error $LOCAL_UG $LOCAL_ID "System is error when query quota for file ($LOCAL_UG:$LOCAL_ID)." else @@ -272,6 +308,36 @@ quota_init() { } quota_init +test_quota_performance() { + TESTFILE="$DIR/$tdir/$tfile-0" + local size=$1 + local stime=`date +%s` + $RUNAS dd if=/dev/zero of=$TESTFILE bs=1M count=$size || quota_error u $TSTUSR "write failure" + local etime=`date +%s` + delta=$((etime - stime)) + if [ $delta -gt 0 ]; then + rate=$((size * 1024 / delta)) + [ $rate -gt 1024 ] || error "SLOW IO for $TSTUSR (user): $rate KB/sec" + fi + rm -f $TESTFILE +} + +# test basic quota performance b=21696 +test_0() { + mkdir -p $DIR/$tdir + chmod 0777 $DIR/$tdir + MB=100 + [ "$SLOW" = "no" ] && MB=10 + + test_quota_performance $MB + + $LFS setquota -u $TSTUSR -b 0 -B $((1024*1024)) -i 0 -I 0 $DIR + test_quota_performance $MB + + resetquota -u $TSTUSR +} +run_test_with_stat 0 "Test basic quota performance ===" + # test for specific quota limitation, qunit, qtune $1=block_quota_limit test_1_sub() { LIMIT=$1 @@ -291,14 +357,7 @@ test_1_sub() { chown $TSTUSR.$TSTUSR $TESTFILE log " Write ..." - stime=`date +%s` $RUNAS dd if=/dev/zero of=$TESTFILE bs=$BLK_SZ count=$(($LIMIT/2)) || quota_error u $TSTUSR "(usr) write failure, but expect success" - etime=`date +%s` - delta=$((etime - stime)) - if [ $delta -gt 0 ]; then - rate=$((BLK_SZ * LIMIT / 2 / delta / 1024)) - [ $rate -gt 1024 ] || error "SLOW IO for $TSTUSR (user): $rate KB/sec" - fi log " Done" log " Write out of block quota ..." # this time maybe cache write, ignore it's failure @@ -310,7 +369,7 @@ test_1_sub() { rm -f $TESTFILE sync; sleep 1; sync; OST0_UUID=`do_facet ost1 $LCTL dl | grep -m1 obdfilter | awk '{print $((NF-1))}'` - OST0_QUOTA_USED=`$LFS quota -o $OST0_UUID -u $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $1 }'` + OST0_QUOTA_USED=`getquota -u $TSTUSR $OST0_UUID curspace` echo $OST0_QUOTA_USED [ $OST0_QUOTA_USED -ne 0 ] && \ ($SHOW_QUOTA_USER; quota_error u $TSTUSR "(usr) quota deleted isn't released") @@ -329,12 +388,7 @@ test_1_sub() { chown $TSTUSR.$TSTUSR $TESTFILE log " Write ..." - stime=`date +%s` $RUNAS dd if=/dev/zero of=$TESTFILE bs=$BLK_SZ count=$(($LIMIT/2)) || quota_error g $TSTUSR "(grp) write failure, but expect success" - etime=`date +%s` - delta=$((etime - stime)) - rate=$((BLK_SZ * LIMIT / 2 / delta / 1024)) - [ $rate -gt 1024 ] || error "SLOW IO for $TSTUSR (group): $rate KB/sec" log " Done" log " Write out of block quota ..." # this time maybe cache write, ignore it's failure @@ -346,7 +400,7 @@ test_1_sub() { rm -f $TESTFILE sync; sleep 1; sync; OST0_UUID=`do_facet ost1 $LCTL dl | grep -m1 obdfilter | awk '{print $((NF-1))}'` - OST0_QUOTA_USED=`$LFS quota -o $OST0_UUID -g $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $1 }'` + OST0_QUOTA_USED=`getquota -g $TSTUSR $OST0_UUID curspace` echo $OST0_QUOTA_USED [ $OST0_QUOTA_USED -ne 0 ] && \ ($SHOW_QUOTA_GROUP; quota_error g $TSTUSR "(grp) quota deleted isn't released") @@ -401,7 +455,7 @@ test_2_sub() { sync; sleep 1; sync; MDS_UUID=`do_facet $SINGLEMDS $LCTL dl | grep -m1 " mdt " | awk '{print $((NF-1))}'` - MDS_QUOTA_USED=`$LFS quota -o $MDS_UUID -u $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $4 }'` + MDS_QUOTA_USED=`getquota -u $TSTUSR $MDS_UUID curinodes` echo $MDS_QUOTA_USED [ $MDS_QUOTA_USED -ne 0 ] && \ ($SHOW_QUOTA_USER; quota_error u $TSTUSR "(usr) quota deleted isn't released") @@ -429,7 +483,7 @@ test_2_sub() { sync; sleep 1; sync; MDS_UUID=`do_facet $SINGLEMDS $LCTL dl | grep -m1 " mdt " | awk '{print $((NF-1))}'` - MDS_QUOTA_USED=`$LFS quota -o $MDS_UUID -g $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $4 }'` + MDS_QUOTA_USED=`getquota -g $TSTUSR $MDS_UUID curinodes` echo $MDS_QUOTA_USED [ $MDS_QUOTA_USED -ne 0 ] && \ ($SHOW_QUOTA_GROUP; quota_error g $TSTUSR "(grp) quota deleted isn't released") @@ -717,7 +771,7 @@ run_test_with_stat 5 "Chown & chgrp successfully even out of block/file quota == # block quota acquire & release test_6() { if [ $OSTCOUNT -lt 2 ]; then - skip "$OSTCOUNT < 2, too few osts" + skip_env "$OSTCOUNT < 2, too few osts" return 0; fi @@ -777,7 +831,7 @@ test_6() { while [ $((count--)) -gt 0 ]; do sync && sleep 5 - OST0_QUOTA_HOLD=`$LFS quota -o $OST0_UUID -u $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $3 }'` + OST0_QUOTA_HOLD=`getquota -u $TSTUSR $OST0_UUID bhardlimit` if [ -z $OST0_QUOTA_HOLD ]; then error "System is error when query quota for block (U:$TSTUSR)." else @@ -793,7 +847,7 @@ test_6() { while [ $((count--)) -gt 0 ]; do sync && sleep 5 - OST0_QUOTA_HOLD=`$LFS quota -o $OST0_UUID -g $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $3 }'` + OST0_QUOTA_HOLD=`getquota -g $TSTUSR $OST0_UUID bhardlimit` if [ -z $OST0_QUOTA_HOLD ]; then error "System is error when query quota for block (G:$TSTUSR)." else @@ -858,13 +912,13 @@ test_7() # check limits PATTERN="`echo $DIR | sed 's/\//\\\\\//g'`" - TOTAL_LIMIT="`$LFS quota -v -u $TSTUSR $DIR | awk '/^.*'$PATTERN'.*[[:digit:]+][[:space:]+]/ { print $4 }'`" + TOTAL_LIMIT=`getquota -u $TSTUSR global bhardlimit` [ $TOTAL_LIMIT -eq $LIMIT ] || error "total limits not recovery!" echo " total limits = $TOTAL_LIMIT" OST0_UUID=`do_facet ost1 "$LCTL dl | grep -m1 obdfilter" | awk '{print $((NF-1))}'` [ -z "$OST0_UUID" ] && OST0_UUID=`do_facet ost1 "$LCTL dl | grep -m1 obdfilter" | awk '{print $((NF-1))}'` - OST0_LIMIT="`$LFS quota -o $OST0_UUID -u $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $3 }'`" + OST0_LIMIT=`getquota -u $TSTUSR $OST0_UUID bhardlimit` [ $OST0_LIMIT -eq $BUNIT_SZ ] || error "high limits not released!" echo " limits on $OST0_UUID = $OST0_LIMIT" @@ -970,64 +1024,8 @@ test_9() { } run_test_with_stat 9 "run for fixing bug10707(64bit) ===========" -# run for fixing bug10707, it need a big room. test for 32bit -# 2.0 version does not support 32 bit qd_count, so such test is obsolete. -test_10() { - mkdir -p $DIR/$tdir - chmod 0777 $DIR/$tdir - check_whether_skip && return 0 - - wait_delete_completed - - set_blk_tunesz 512 - set_blk_unitsz 1024 - - # make qd_count 32 bit - lustre_fail mds_ost 0xA00 - - TESTFILE="$DIR/$tdir/$tfile-0" - - BLK_LIMIT=$((100 * KB * KB)) # 100G - FILE_LIMIT=1000000 - - log " Set enough high limit(block:$BLK_LIMIT; file: $FILE_LIMIT) for user: $TSTUSR" - $LFS setquota -u $TSTUSR -b 0 -B $BLK_LIMIT -i 0 -I $FILE_LIMIT $DIR - log " Set enough high limit(block:$BLK_LIMIT; file: $FILE_LIMIT) for group: $TSTUSR" - $LFS setquota -g $TSTUSR -b 0 -B $BLK_LIMIT -i 0 -I $FILE_LIMIT $DIR - - quota_show_check a u $TSTUSR - quota_show_check a g $TSTUSR - - echo " Set stripe" - $LFS setstripe $TESTFILE -c 1 - touch $TESTFILE - chown $TSTUSR.$TSTUSR $TESTFILE - - log " Write the big file of 4.5 G ..." - $RUNAS dd if=/dev/zero of=$TESTFILE bs=$blksize count=$((size_file / blksize)) || \ - quota_error a $TSTUSR "(usr) write 4.5 G file failure, but expect success" - - $SHOW_QUOTA_USER - $SHOW_QUOTA_GROUP - - log " delete the big file of 4.5 G..." - $RUNAS rm -f $TESTFILE - sync; sleep 3; sync; - - $SHOW_QUOTA_USER - $SHOW_QUOTA_GROUP - - RC=$? - - # make qd_count 64 bit - lustre_fail mds_ost 0 - - set_blk_unitsz $((128 * 1024)) - set_blk_tunesz $((128 * 1024 / 2)) - - return $RC -} -#run_test_with_stat 10 "run for fixing bug10707(32bit) ===========" +# 2.0 version does not support 32 bit qd_count, +# test_10 "run for fixing bug10707(32bit) " is obsolete # test a deadlock between quota and journal b=11693 test_12() { @@ -1035,7 +1033,7 @@ test_12() { chmod 0777 $DIR/$tdir [ "$(grep $DIR2 /proc/mounts)" ] || mount_client $DIR2 || \ - { skip "Need lustre mounted on $MOUNT2 " && retutn 0; } + { skip_env "Need lustre mounted on $MOUNT2 " && retutn 0; } LIMIT=$(( $BUNIT_SZ * $(($OSTCOUNT + 1)) * 10)) # 10 bunits each sever TESTFILE="$DIR/$tdir/$tfile-0" @@ -1062,17 +1060,24 @@ test_12() { DDPID=$! echo " step2: testing ......" - count=0 - while [ true ]; do - if ! ps -p ${DDPID1} > /dev/null 2>&1; then break; fi - count=$[count+1] - if [ $count -gt 64 ]; then + local last_size=$(stat -c %s $TESTFILE2) + local stall_secs=0 + local start_secs=$SECONDS + while [ -d /proc/${DDPID1} ]; do + local size=$(stat -c %s $TESTFILE2) + if [ $size -eq $last_size ]; then + stall_secs=$[stall_secs+1] + else + stall_secs=0 + fi + if [ $stall_secs -gt 30 ]; then lustre_fail ost 0 - quota_error u $TSTUSR2 "dd should be finished!" + quota_error u $TSTUSR2 "giving up: dd stalled (i.e. made no progress) for 30 seconds!" fi + last_size=$size sleep 1 done - echo "(dd_pid=$DDPID1, time=$count)successful" + echo "(dd_pid=$DDPID1, time=$((SECONDS-start_secs)))successful" #Recover fail_loc and dd will finish soon lustre_fail ost 0 @@ -1164,9 +1169,9 @@ check_if_quota_zero(){ line=`$LFS quota -v -$1 $2 $DIR | wc -l` for i in `seq 3 $line`; do if [ $i -eq 3 ]; then - field="3 4 6 7" + field="3 4 7 8" else - field="3 5" + field="3 6" fi for j in $field; do tmp=`$LFS quota -v -$1 $2 $DIR | sed -n ${i}p | @@ -1223,14 +1228,14 @@ test_15(){ # test for user $LFS setquota -u $TSTUSR -b 0 -B $LIMIT -i 0 -I 0 $DIR - TOTAL_LIMIT="`$LFS quota -v -u $TSTUSR $DIR | awk '/^.*'$PATTERN'.*[[:digit:]+][[:space:]+]/ { print $4 }'`" + TOTAL_LIMIT=`getquota -u $TSTUSR global bhardlimit` [ $TOTAL_LIMIT -eq $LIMIT ] || error " (user)total limits = $TOTAL_LIMIT; limit = $LIMIT, failed!" echo " (user)total limits = $TOTAL_LIMIT; limit = $LIMIT, successful!" resetquota -u $TSTUSR # test for group $LFS setquota -g $TSTUSR -b 0 -B $LIMIT -i 0 -I 0 $DIR - TOTAL_LIMIT="`$LFS quota -v -g $TSTUSR $DIR | awk '/^.*'$PATTERN'.*[[:digit:]+][[:space:]+]/ { print $4 }'`" + TOTAL_LIMIT=`getquota -g $TSTUSR global bhardlimit` [ $TOTAL_LIMIT -eq $LIMIT ] || error " (group)total limits = $TOTAL_LIMIT; limit = $LIMIT, failed!" echo " (group)total limits = $TOTAL_LIMIT; limit = $LIMIT, successful!" resetquota -g $TSTUSR @@ -1249,67 +1254,8 @@ test_15(){ } run_test_with_stat 15 "set block quota more than 4T ===" -# $1=u/g $2=with qunit adjust or not -test_16_tub() { - LIMIT=$(( $BUNIT_SZ * $(($OSTCOUNT + 1)) * 4)) - TESTFILE="$DIR/$tdir/$tfile" - mkdir -p $DIR/$tdir - - wait_delete_completed - - echo " User quota (limit: $LIMIT kbytes)" - if [ $1 == "u" ]; then - $LFS setquota -u $TSTUSR -b 0 -B $LIMIT -i 0 -I 0 $DIR - quota_show_check b u $TSTUSR - else - $LFS setquota -g $TSTUSR -b 0 -B $LIMIT -i 0 -I 0 $DIR - quota_show_check b g $TSTUSR - fi - - $LFS setstripe $TESTFILE -c 1 - chown $TSTUSR.$TSTUSR $TESTFILE - - echo " Write ..." - $RUNAS dd if=/dev/zero of=$TESTFILE bs=$BLK_SZ count=$((BUNIT_SZ * 4)) || \ - quota_error a $TSTUSR "(usr) write failure, but expect success" - echo " Done" - echo " Write out of block quota ..." - # this time maybe cache write, ignore it's failure - $RUNAS dd if=/dev/zero of=$TESTFILE bs=$BLK_SZ count=$BUNIT_SZ seek=$((BUNIT_SZ * 4)) || true - # flush cache, ensure noquota flag is setted on client - cancel_lru_locks osc - if [ $2 -eq 1 ]; then - $RUNAS dd if=/dev/zero of=$TESTFILE bs=$BLK_SZ count=$BUNIT_SZ seek=$((BUNIT_SZ * 4)) || \ - quota_error a $TSTUSR "(write failure, but expect success" - else - $RUNAS dd if=/dev/zero of=$TESTFILE bs=$BLK_SZ count=$BUNIT_SZ seek=$((BUNIT_SZ * 4)) && \ - quota_error a $TSTUSR "(write success, but expect EDQUOT" - fi - - rm -f $TESTFILE - sync; sleep 3; sync; - resetquota -$1 $TSTUSR -} - -# test without adjusting qunit -# 2.0 version does not support WITHOUT_CHANGE_QS, so such test is obsolete -test_16 () { - set_blk_tunesz $((BUNIT_SZ * 2)) - set_blk_unitsz $((BUNIT_SZ * 4)) - for i in u g; do - for j in 0 1; do - # define OBD_FAIL_QUOTA_WITHOUT_CHANGE_QS 0xA01 - echo " grp/usr: $i, adjust qunit: $j" - echo "-------------------------------" - [ $j -eq 1 ] && lustre_fail mds_ost 0 - [ $j -eq 0 ] && lustre_fail mds_ost 0xA01 - test_16_tub $i $j - done - done - set_blk_unitsz $((128 * 1024)) - set_blk_tunesz $((128 * 1024 / 2)) -} -#run_test_with_stat 16 "test without adjusting qunit" +# 2.0 version does not support WITHOUT_CHANGE_QS, +# test_16 "test without adjusting qunit" is obsolete # run for fixing bug14526, failed returned quota reqs shouldn't ruin lustre. test_17() { @@ -1518,7 +1464,7 @@ test_18bc_sub() { fi DDPID=$! - do_facet $SINGLEMDS "$LCTL conf_param ${FSNAME}-MDT*.mdd.quota_type=ug" + do_facet $SINGLEMDS "$LCTL conf_param mdd.${FSNAME}-MDT*.quota_type=ug" log "failing mds for $((2 * timeout)) seconds" fail $SINGLEMDS $((2 * timeout)) @@ -1569,7 +1515,7 @@ test_18b() { # check if watchdog is triggered do_facet ost1 dmesg > $TMP/lustre-log-${TESTNAME}.log watchdog=`awk '/test 18b/ {start = 1;} - /Watchdog triggered/ { + /Service thread pid/ && /was inactive/ { if (start) { print; } @@ -1650,9 +1596,10 @@ test_20() --inode-hardlimit ${LSTR[3]} \ $MOUNT || error "could not set quota limits" - ($LFS quota -v -u $TSTUSR $MOUNT | \ - grep -E '^ *'$MOUNT' *[0-9]+\** *'${LVAL[0]}' *'${LVAL[1]}' *[0-9]+\** *'${LVAL[2]}' *'${LVAL[3]}) \ - || error "lfs quota output is unexpected" + [ "`getquota -u $TSTUSR global bsoftlimit`" = "${LVAL[0]}" ] || error "bsoftlimit was not set properly" + [ "`getquota -u $TSTUSR global bhardlimit`" = "${LVAL[1]}" ] || error "bhardlimit was not set properly" + [ "`getquota -u $TSTUSR global isoftlimit`" = "${LVAL[2]}" ] || error "isoftlimit was not set properly" + [ "`getquota -u $TSTUSR global ihardlimit`" = "${LVAL[3]}" ] || error "ihardlimit was not set properly" resetquota -u $TSTUSR } @@ -1788,7 +1735,7 @@ test_23_sub() { rm -f $TESTFILE wait_delete_completed OST0_UUID=`do_facet ost1 $LCTL dl | grep -m1 obdfilter | awk '{print $((NF-1))}'` - OST0_QUOTA_USED=`$LFS quota -o $OST0_UUID -u $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $1 }'` + OST0_QUOTA_USED=`getquota -u $TSTUSR $OST0_UUID curspace` echo $OST0_QUOTA_USED [ $OST0_QUOTA_USED -ne 0 ] && \ ($SHOW_QUOTA_USER; quota_error u $TSTUSR "quota deleted isn't released") @@ -1874,10 +1821,10 @@ test_25_sub() { $LFS setstripe $DIR/$tdir -c 1 -i 0 MDS_UUID=`do_facet $SINGLEMDS $LCTL dl | grep -m1 " mdt " | awk '{print $((NF-1))}'` OST0_UUID=`do_facet ost1 $LCTL dl | grep -m1 obdfilter | awk '{print $((NF-1))}'` - MDS_QUOTA_USED_OLD=`$LFS quota -o $MDS_UUID $1 $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $4 }'` - OST0_QUOTA_USED_OLD=`$LFS quota -o $OST0_UUID $1 $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $1 }'` - MDS_QUOTA_USED2_OLD=`$LFS quota -o $MDS_UUID $1 $TSTUSR2 $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $4 }'` - OST0_QUOTA_USED2_OLD=`$LFS quota -o $OST0_UUID $1 $TSTUSR2 $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $1 }'` + MDS_QUOTA_USED_OLD=`getquota $1 $TSTUSR $MDS_UUID curinodes` + OST0_QUOTA_USED_OLD=`getquota $1 $TSTUSR $OST0_UUID curspace` + MDS_QUOTA_USED2_OLD=`getquota $1 $TSTUSR2 $MDS_UUID curinodes` + OST0_QUOTA_USED2_OLD=`getquota $1 $TSTUSR2 $OST0_UUID curspace` # TSTUSR write 4M log "$TSTUSR write 4M to $TESTFILE" @@ -1885,10 +1832,10 @@ test_25_sub() { sync show_quota $1 $TSTUSR show_quota $1 $TSTUSR2 - MDS_QUOTA_USED_NEW=`$LFS quota -o $MDS_UUID $1 $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $4 }'` + MDS_QUOTA_USED_NEW=`getquota $1 $TSTUSR $MDS_UUID curinodes` [ $MDS_QUOTA_USED_NEW -ne $((MDS_QUOTA_USED_OLD + 1)) ] && \ quota_error a $TSTUSR "$TSTUSR inode quota usage error: [$MDS_QUOTA_USED_OLD|$MDS_QUOTA_USED_NEW]" - OST0_QUOTA_USED_NEW=`$LFS quota -o $OST0_UUID $1 $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $1 }'` + OST0_QUOTA_USED_NEW=`getquota $1 $TSTUSR $OST0_UUID curspace` OST0_QUOTA_USED_DELTA=$((OST0_QUOTA_USED_NEW - OST0_QUOTA_USED_OLD)) [ $OST0_QUOTA_USED_DELTA -lt 4096 ] && \ quota_error a $TSTUSR "$TSTUSR block quota usage error: [$OST0_QUOTA_USED_OLD|$OST0_QUOTA_USED_NEW]" @@ -1904,19 +1851,19 @@ test_25_sub() { sync show_quota $1 $TSTUSR show_quota $1 $TSTUSR2 - MDS_QUOTA_USED2_NEW=`$LFS quota -o $MDS_UUID $1 $TSTUSR2 $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $4 }'` + MDS_QUOTA_USED2_NEW=`getquota $1 $TSTUSR2 $MDS_UUID curinodes` [ $MDS_QUOTA_USED2_NEW -ne $((MDS_QUOTA_USED2_OLD + 1)) ] && \ quota_error a $TSTUSR2 "$TSTUSR2 inode quota usage transfer from $TSTUSR to $TSTUSR2 failed: [$MDS_QUOTA_USED2_OLD|$MDS_QUOTA_USED2_NEW]" - OST0_QUOTA_USED2_NEW=`$LFS quota -o $OST0_UUID $1 $TSTUSR2 $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $1 }'` + OST0_QUOTA_USED2_NEW=`getquota $1 $TSTUSR2 $OST0_UUID curspace` # when chown, the quota on ost could be displayed out of quota temporarily. Delete the '*' in this situation. b=20433 OST0_QUOTA_USED2_NEW=${OST0_QUOTA_USED2_NEW%\*} OST0_QUOTA_USED2_DELTA=$((OST0_QUOTA_USED2_NEW - OST0_QUOTA_USED2_OLD)) [ $OST0_QUOTA_USED2_DELTA -ne $OST0_QUOTA_USED_DELTA ] && \ quota_error a $TSTUSR2 "$TSTUSR2 block quota usage transfer from $TSTUSR to $TSTUSR2 failed: [$OST0_QUOTA_USED2_OLD|$OST0_QUOTA_USED2_NEW]" - MDS_QUOTA_USED_NEW=`$LFS quota -o $MDS_UUID $1 $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $4 }'` + MDS_QUOTA_USED_NEW=`getquota $1 $TSTUSR $MDS_UUID curinodes` [ $MDS_QUOTA_USED_NEW -ne $MDS_QUOTA_USED_OLD ] && \ quota_error a $TSTUSR "$TSTUSR inode quota usage transfer from $TSTUSR to $TSTUSR2 failed: [$MDS_QUOTA_USED_OLD|$MDS_QUOTA_USED_NEW]" - OST0_QUOTA_USED_NEW=`$LFS quota -o $OST0_UUID $1 $TSTUSR $DIR | awk '/^.*[[:digit:]+][[:space:]+]/ { print $1 }'` + OST0_QUOTA_USED_NEW=`getquota $1 $TSTUSR $OST0_UUID curspace` [ $OST0_QUOTA_USED_NEW -ne $OST0_QUOTA_USED_OLD ] && \ quota_error a $TSTUSR "$TSTUSR block quota usage transfer from $TSTUSR to $TSTUSR2 failed: [$OST0_QUOTA_USED_OLD|$OST0_QUOTA_USED_NEW]" @@ -1980,12 +1927,25 @@ test_26() { } run_test_with_stat 26 "test for false quota error(bz18491) ======================================" -test_27() { +test_27a() { $LFS quota $TSTUSR $DIR && error "lfs succeeded with no type, but should have failed" $LFS setquota $TSTUSR $DIR && error "lfs succeeded with no type, but should have failed" return 0 } -run_test_with_stat 27 "lfs quota/setquota should handle wrong arguments (19612) =================" +run_test_with_stat 27a "lfs quota/setquota should handle wrong arguments (19612) =================" + +test_27b() { + $LFS setquota -u $TSTID -b 1000 -B 1000 -i 1000 -I 1000 $DIR || \ + error "lfs setquota failed with uid argument" + $LFS setquota -g $TSTID -b 1000 -B 1000 -i 1000 -I 1000 $DIR || \ + error "lfs stequota failed with gid argument" + $SHOW_QUOTA_USERID || error "lfs quota failed with uid argument" + $SHOW_QUOTA_GROUPID || error "lfs quota failed with gid argument" + resetquota -u $TSTUSR + resetquota -g $TSTUSR + return 0 +} +run_test 27b "lfs quota/setquota should handle user/group ID (20200) =================" test_28() { BLK_LIMIT=$((100 * 1024 * 1024)) # 100G @@ -2035,6 +1995,42 @@ test_28() { } run_test_with_stat 28 "test for consistency for qunit when setquota (18574) ===========" +test_29() +{ + local BLK_LIMIT=$((100 * 1024 * 1024)) # 100G + local timeout + local pid + + if at_is_enabled; then + timeout=$(at_max_get client) + at_max_set 10 client + else + timeout=$(lctl get_param -n timeout) + lctl set_param timeout=10 + fi + + #define OBD_FAIL_MDS_QUOTACTL_NET 0x12e + lustre_fail mds 0x12e + + $LFS setquota -u $TSTUSR -b 0 -B $BLK_LIMIT -i 0 -I 0 $DIR & pid=$! + + echo "sleeping for $((10 * 2)) seconds" + sleep $((10 * 2)) + ps -p $pid && error "lfs hadn't finished by timeout" + wait $pid && error "succeeded, but should have failed" + + lustre_fail mds 0 + + if at_is_enabled; then + at_max_set $timeout client + else + lctl set_param timeout=$timeout + fi + + resetquota -u $TSTUSR +} +run_test_with_stat 29 "unhandled quotactls must not hang lustre client (19778) ========" + test_30() { local output @@ -2042,23 +2038,160 @@ test_30() local TESTFILE="$DIR/$tdir/$tfile" local GRACE=10 + set_blk_tunesz 512 + set_blk_unitsz 1024 + mkdir -p $DIR/$tdir chmod 0777 $DIR/$tdir + $LFS setstripe $TESTFILE -i 0 -c 1 + chown $TSTUSR.$TSTUSR $TESTFILE + $LFS setquota -t -u --block-grace $GRACE --inode-grace $MAX_IQ_TIME $DIR $LFS setquota -u $TSTUSR -b $LIMIT -B 0 -i 0 -I 0 $DIR $RUNAS dd if=/dev/zero of=$TESTFILE bs=1024 count=$((LIMIT * 2)) || true cancel_lru_locks osc - sleep 5 + sleep $GRACE $LFS setquota -u $TSTUSR -B 0 $DIR - output=`$SHOW_QUOTA_USER | grep $MOUNT | awk '{ print $5 }' | tr -d s` - [ "$output" -le "$((GRACE - 5))" ] || error "grace times were reset or unexpectedly high latency" + # over-quota flag has not yet settled since we do not trigger async events + # based on grace time period expiration + $SHOW_QUOTA_USER + $RUNAS dd if=/dev/zero of=$TESTFILE conv=notrunc oflag=append bs=1048576 count=1 || true + cancel_lru_locks osc + # now over-quota flag should be settled and further writes should fail + $SHOW_QUOTA_USER + $RUNAS dd if=/dev/zero of=$TESTFILE conv=notrunc oflag=append bs=1048576 count=1 && error "grace times were reset" rm -f $TESTFILE resetquota -u $TSTUSR $LFS setquota -t -u --block-grace $MAX_DQ_TIME --inode-grace $MAX_IQ_TIME $DIR + + set_blk_unitsz $((128 * 1024)) + set_blk_tunesz $((128 * 1024 / 2)) } run_test_with_stat 30 "hard limit updates should not reset grace times ================" +# test duplicate quota releases b=18630 +test_31() { + mkdir -p $DIR/$tdir + chmod 0777 $DIR/$tdir + + LIMIT=$(( $BUNIT_SZ * $(($OSTCOUNT + 1)) * 10)) # 10 bunits each sever + TESTFILE="$DIR/$tdir/$tfile-0" + TESTFILE2="$DIR/$tdir/$tfile-1" + + wait_delete_completed + + log " User quota (limit: $LIMIT kbytes)" + $LFS setquota -u $TSTUSR -b 0 -B $LIMIT -i 0 -I 0 $DIR + + $LFS setstripe $TESTFILE -i 0 -c 1 + chown $TSTUSR.$TSTUSR $TESTFILE + $LFS setstripe $TESTFILE2 -i 0 -c 1 + chown $TSTUSR.$TSTUSR $TESTFILE2 + + log " step1: write out of block quota ..." + $RUNAS dd if=/dev/zero of=$TESTFILE bs=$BLK_SZ count=5120 + $RUNAS dd if=/dev/zero of=$TESTFILE2 bs=$BLK_SZ count=5120 + + #define OBD_FAIL_QUOTA_DELAY_SD 0xA04 + #define OBD_FAIL_SOME 0x10000000 /* fail N times */ + lustre_fail ost $((0x00000A04 | 0x10000000)) 1 + + log " step2: delete two files so that triggering duplicate quota release ..." + rm -f $TESTFILE $TESTFILE2 + sync; sleep 5; sync # OBD_FAIL_QUOTA_DELAY_SD will delay for 5 seconds + wait_delete_completed + + log " step3: verify if the ost failed" + do_facet ost1 dmesg > $TMP/lustre-log-${TESTNAME}.log + watchdog=`awk '/test 31/ {start = 1;} + /release quota error/ { + if (start) { + print; + } + }' $TMP/lustre-log-${TESTNAME}.log` + [ "$watchdog" ] && error "$watchdog" + rm -f $TMP/lustre-log-${TESTNAME}.log + + lustre_fail ost 0 + resetquota -u $TSTUSR +} +run_test_with_stat 31 "test duplicate quota releases ===" + +# check hash_cur_bits +check_quota_hash_cur_bits() { + local bits=$1 + + # check quota_hash_cur_bits on all obdfilters + for num in `seq $OSTCOUNT`; do + cb=`do_facet ost$num "cat /sys/module/lquota/parameters/hash_lqs_cur_bits"` + if [ $cb -gt $bits ]; then + echo "hash_lqs_cur_bits of ost$num is too large(cur_bits=$cb)" + return 1; + fi + done + # check quota_hash_cur_bits on mds + cb=`do_facet $SINGLEMDS "cat /sys/module/lquota/parameters/hash_lqs_cur_bits"` + if [ $cb -gt $bits ]; then + echo "hash_lqs_cur_bits of mds is too large(cur_bits=$cb)" + return 1; + fi + return 0; +} + +# check lqs hash +check_lqs_hash() { + # check distribution of all obdfilters + for num in `seq $OSTCOUNT`; do + do_facet ost$num "lctl get_param obdfilter.${FSNAME}-OST*.hash_stats | grep LQS_HASH" | while read line; do + rehash_count=`echo $line | awk '{print $9}'` + if [ $rehash_count -eq 0 ]; then + echo -e "ost$num:\n $line" + error "Rehearsh didn't happen" + fi + done + done + # check distribution of mds + do_facet $SINGLEMDS "lctl get_param mdt.${FSNAME}-MDT*.hash_stats | grep LQS_HASH" | while read line; do + rehash_count=`echo $line | awk '{print $9}'` + if [ $rehash_count -eq 0 ]; then + echo -e "mdt:\n $line" + error "Rehearsh didn't happen" + fi + done +} + +test_32() +{ + # reset system so that quota_hash_cur_bits==3 + echo "Reset system ..." + local LMR_orig=$LOAD_MODULES_REMOTE + LOAD_MODULES_REMOTE=true + cleanup_and_setup_lustre + LOAD_MODULES_REMOTE=$LMR_orig + + for user in $SANITY_QUOTA_USERS; do + check_runas_id_ret $user quota_usr "runas -u $user -g quota_usr" >/dev/null 2>/dev/null || \ + missing_users="$missing_users $user" + done + [ -n "$missing_users" ] && { skip_env "the following users are missing: $missing_users" ; return 0 ; } + check_quota_hash_cur_bits 3 || { skip_env "hash_lqs_cur_bits isn't set properly"; return 0;} + + $LFS quotaoff -ug $DIR + $LFS quotacheck -ug $DIR + + for user in $SANITY_QUOTA_USERS; do + $LFS setquota -u $user --block-hardlimit 1048576 $DIR + done + + check_lqs_hash + + for user in $SANITY_QUOTA_USERS; do + resetquota -u $user + done +} +run_test 32 "check lqs hash(bug 21846) ==========================================" + # turn off quota quota_fini() { @@ -2071,5 +2204,6 @@ log "cleanup: ======================================================" cd $ORIG_PWD check_and_cleanup_lustre echo '=========================== finished ===============================' +export QUOTA_AUTO=$QUOTA_AUTO_OLD [ -f "$QUOTALOG" ] && cat $QUOTALOG && grep -q FAIL $QUOTALOG && exit 1 || true echo "$0: completed"