X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity.sh;h=414d95e8238e3efc4d020c88c2cc76ae1498f895;hp=3f5767f7dfb44d30855e609f30e4727c67056576;hb=3eff6bada2bc789fe6089e792e4043ef7ec8396f;hpb=7a7309fa849577ddd5a4f6bb5bfb69e84a7fec89 diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh index 3f5767f..414d95e8 100755 --- a/lustre/tests/sanity.sh +++ b/lustre/tests/sanity.sh @@ -1635,6 +1635,28 @@ test_27ce() { } run_test 27ce "more stripes than OSTs with -o" +test_27cf() { + local osp_proc="osp.$FSNAME-OST0000-osc-MDT000*.active" + local pid=0 + + test_mkdir -p $DIR/$tdir || error "failed to mkdir $DIR/$tdir" + do_facet $SINGLEMDS "$LCTL set_param -n $osp_proc=0" + stack_trap "do_facet $SINGLEMDS $LCTL set_param -n $osp_proc=1" EXIT + wait_update_facet $SINGLEMDS "$LCTL get_param -n $osp_proc | grep 1" || + error "failed to set $osp_proc=0" + + $LFS setstripe -o 0 $DIR/$tdir/$tfile & + pid=$! + sleep 1 + do_facet $SINGLEMDS "$LCTL set_param -n $osp_proc=1" + wait_update_facet $SINGLEMDS "$LCTL get_param -n $osp_proc | grep 0" || + error "failed to set $osp_proc=1" + wait $pid + [[ $pid -ne 0 ]] || + error "should return error due to $osp_proc=0" +} +run_test 27cf "'setstripe -o' on inactive OSTs should return error" + test_27d() { test_mkdir $DIR/$tdir $LFS setstripe -c 0 -i -1 -S 0 $DIR/$tdir/$tfile || @@ -1758,17 +1780,24 @@ run_test 27m "create file while OST0 was full" # OSCs keep a NOSPC flag that will be reset after ~5s (qos_maxage) # if the OST isn't full anymore. reset_enospc() { - local OSTIDX=${1:-""} + local ostidx=${1:-""} + local delay + local ready + local get_prealloc local list=$(comma_list $(osts_nodes)) - [ "$OSTIDX" ] && list=$(facet_host ost$((OSTIDX + 1))) + [ "$ostidx" ] && list=$(facet_host ost$((ostidx + 1))) do_nodes $list lctl set_param fail_loc=0 - sync # initiate all OST_DESTROYs from MDS to OST - sleep_maxage + wait_delete_completed # initiate all OST_DESTROYs from MDS to OST + delay=$(do_facet $SINGLEMDS lctl get_param -n lov.*.qos_maxage | + awk '{print $1 * 2;exit;}') + get_prealloc="$LCTL get_param -n osc.*MDT*.prealloc_status | + grep -v \"^0$\"" + wait_update_facet $SINGLEMDS "$get_prealloc" "" $delay } -exhaust_precreations() { +__exhaust_precreations() { local OSTIDX=$1 local FAILLOC=$2 local FAILIDX=${3:-$OSTIDX} @@ -1799,14 +1828,19 @@ exhaust_precreations() { createmany -o $DIR/$tdir/${OST}/f $next_id $((last_id - next_id + 2)) do_facet $mfacet lctl get_param osp.$mdtosc_proc2.prealloc* do_facet $ofacet lctl set_param fail_loc=$FAILLOC +} + +exhaust_precreations() { + __exhaust_precreations $1 $2 $3 sleep_maxage } exhaust_all_precreations() { local i for (( i=0; i < OSTCOUNT; i++ )) ; do - exhaust_precreations $i $1 -1 + __exhaust_precreations $i $1 -1 done + sleep_maxage } test_27n() { @@ -3053,7 +3087,7 @@ run_test 30b "execute binary from Lustre as non-root ===========" test_30c() { # b=22376 [ $PARALLEL == "yes" ] && skip "skip parallel run" - cp `which ls` $DIR || cp /bin/ls $DIR + cp $(which ls) $DIR || cp /bin/ls $DIR chmod a-rw $DIR/ls cancel_lru_locks mdc cancel_lru_locks osc @@ -3062,6 +3096,22 @@ test_30c() { # b=22376 } run_test 30c "execute binary from Lustre without read perms ====" +test_30d() { + cp $(which dd) $DIR || error "failed to copy dd to $DIR/dd" + + for i in {1..10}; do + $DIR/dd bs=1M count=128 if=/dev/zero of=$DIR/$tfile & + local PID=$! + sleep 1 + $LCTL set_param ldlm.namespaces.*MDT*.lru_size=clear + wait $PID || error "executing dd from Lustre failed" + rm -f $DIR/$tfile + done + + rm -f $DIR/dd +} +run_test 30d "execute binary from Lustre while clear locks" + test_31a() { $OPENUNLINK $DIR/f31 $DIR/f31 || error "openunlink failed" $CHECKSTAT -a $DIR/f31 || error "$DIR/f31 exists" @@ -3749,6 +3799,46 @@ test_33g() { } run_test 33g "nonroot user create already existing root created file" +test_33h() { + [ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs" + [ $MDS1_VERSION -lt $(version_code 2.13.50) ] && + skip "Need MDS version at least 2.13.50" + + test_mkdir -c $MDSCOUNT -H crush $DIR/$tdir || + error "mkdir $tdir failed" + touch $DIR/$tdir/$tfile || error "touch $tfile failed" + + local index=$($LFS getstripe -m $DIR/$tdir/$tfile) + local index2 + + for fname in $DIR/$tdir/$tfile.bak \ + $DIR/$tdir/$tfile.SAV \ + $DIR/$tdir/$tfile.orig \ + $DIR/$tdir/$tfile~; do + touch $fname || error "touch $fname failed" + index2=$($LFS getstripe -m $fname) + [ $index -eq $index2 ] || + error "$fname MDT index mismatch $index != $index2" + done + + local failed=0 + for i in {1..50}; do + for fname in $(mktemp -u $DIR/$tdir/.$tfile.XXXXXX) \ + $(mktemp $DIR/$tdir/$tfile.XXXXXXXX); do + touch $fname || error "touch $fname failed" + index2=$($LFS getstripe -m $fname) + if [[ $index != $index2 ]]; then + failed=$((failed + 1)) + echo "$fname MDT index mismatch $index != $index2" + fi + done + done + echo "$failed MDT index mismatches" + (( failed < 4 )) || error "MDT index mismatch $failed times" + +} +run_test 33h "temp file is located on the same MDT as target" + TEST_34_SIZE=${TEST_34_SIZE:-2000000000000} test_34a() { rm -f $DIR/f34 @@ -4468,6 +4558,49 @@ test_39p() { } run_test 39p "remote directory cached attributes updated after create ========" +test_39r() { + [ $OST1_VERSION -ge $(version_code 2.13.52) ] || + skip "no atime update on old OST" + if [ "$ost1_FSTYPE" != ldiskfs ]; then + skip_env "ldiskfs only test" + fi + + local saved_adiff + saved_adiff=$(do_facet ost1 \ + lctl get_param -n obdfilter.*OST0000.atime_diff) + stack_trap "do_facet ost1 \ + lctl set_param obdfilter.*.atime_diff=$saved_adiff" + + do_facet ost1 "lctl set_param obdfilter.*.atime_diff=5" + + $LFS setstripe -i 0 $DIR/$tfile + dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || + error "can't write initial file" + cancel_lru_locks osc + + # exceed atime_diff and access file + sleep 6 + dd if=$DIR/$tfile of=/dev/null || error "can't udpate atime" + + local atime_cli=$(stat -c %X $DIR/$tfile) + echo "client atime: $atime_cli" + # allow atime update to be written to device + do_facet ost1 "$LCTL set_param -n osd*.*OST*.force_sync 1" + sleep 5 + + local ostdev=$(ostdevname 1) + local fid=($(lfs getstripe -y $DIR/$tfile | + awk '/l_fid:/ { print $2 }' | tr ':' ' ')) + local objpath="O/0/d$((${fid[1]} % 32))/$((${fid[1]}))" + local cmd="debugfs -c -R \\\"stat $objpath\\\" $ostdev" + + echo "OST atime: $(do_facet ost1 "$cmd" |& grep atime)" + local atime_ost=$(do_facet ost1 "$cmd" |& + awk -F'[: ]' '/atime:/ { print $4 }') + (( atime_cli == atime_ost )) || + error "atime on client $atime_cli != ost $atime_ost" +} +run_test 39r "lazy atime update on OST" test_39q() { # LU-8041 local testdir=$DIR/$tdir @@ -5503,16 +5636,24 @@ test_56c() { local ost_idx=0 local ost_name=$(ostname_from_index $ost_idx) local old_status=$(ost_dev_status $ost_idx) + local p="$TMP/$TESTSUITE-$TESTNAME.parameters" [[ -z "$old_status" ]] || skip_env "OST $ost_name is in $old_status status" do_facet ost1 $LCTL set_param -n obdfilter.$ost_name.degraded=1 - [[ $OST1_VERSION -ge $(version_code 2.12.55) ]] && do_facet ost1 \ + [[ $OST1_VERSION -lt $(version_code 2.12.55) ]] || do_facet ost1 \ $LCTL set_param -n obdfilter.$ost_name.no_precreate=1 + if [[ $OST1_VERSION -ge $(version_code 2.12.57) ]]; then + save_lustre_params ost1 osd-*.$ost_name.nonrotational > $p + do_facet ost1 $LCTL set_param -n osd-*.$ost_name.nonrotational=1 + fi + + [[ $($LFS df -v $MOUNT |& grep -c "inactive device") -eq 0 ]] || + error "$LFS df -v showing inactive devices" sleep_maxage - local new_status=$(ost_dev_status $ost_idx) + local new_status=$(ost_dev_status $ost_idx $MOUNT -v) [[ "$new_status" =~ "D" ]] || error "$ost_name status is '$new_status', missing 'D'" @@ -5520,18 +5661,37 @@ test_56c() { [[ "$new_status" =~ "N" ]] || error "$ost_name status is '$new_status', missing 'N'" fi + if [[ $OST1_VERSION -ge $(version_code 2.12.57) ]]; then + [[ "$new_status" =~ "f" ]] || + error "$ost_name status is '$new_status', missing 'f'" + fi do_facet ost1 $LCTL set_param -n obdfilter.$ost_name.degraded=0 - [[ $OST1_VERSION -ge $(version_code 2.12.55) ]] && do_facet ost1 \ + [[ $OST1_VERSION -lt $(version_code 2.12.55) ]] || do_facet ost1 \ $LCTL set_param -n obdfilter.$ost_name.no_precreate=0 + [[ -z "$p" ]] && restore_lustre_params < $p || true sleep_maxage new_status=$(ost_dev_status $ost_idx) [[ ! "$new_status" =~ "D" && ! "$new_status" =~ "N" ]] || error "$ost_name status is '$new_status', has 'D' and/or 'N'" + # can't check 'f' as devices may actually be on flash } run_test 56c "check 'lfs df' showing device status" +test_56d() { + local mdts=$($LFS df -v $MOUNT | grep -c MDT) + local osts=$($LFS df -v $MOUNT | grep -c OST) + + $LFS df $MOUNT + + (( mdts == MDSCOUNT )) || + error "lfs df -v showed $mdts MDTs, not $MDSCOUNT" + (( osts == OSTCOUNT )) || + error "lfs df -v showed $osts OSTs, not $OSTCOUNT" +} +run_test 56d "'lfs df -v' prints only configured devices" + NUMFILES=3 NUMDIRS=3 setup_56() { @@ -5759,18 +5919,26 @@ test_newerXY_base() { local negref if [ $y == "t" ]; then - ref="\"$(date +"%Y-%m-%d %H:%M:%S")\"" + if [ $x == "b" ]; then + ref="\"$(do_facet mds1 date +"%Y-%m-%d\ %H:%M:%S")\"" + else + ref="\"$(date +"%Y-%m-%d %H:%M:%S")\"" + fi else - ref=$DIR/$tfile.newer + ref=$DIR/$tfile.newer.$x$y touch $ref || error "touch $ref failed" fi sleep 2 setup_56 $dir $NUMFILES $NUMDIRS "-i0 -c1" "-i0 -c1" sleep 2 if [ $y == "t" ]; then - negref="\"$(date +"%Y-%m-%d %H:%M:%S")\"" + if [ $x == "b" ]; then + negref="\"$(do_facet mds1 date +"%Y-%m-%d\ %H:%M:%S")\"" + else + negref="\"$(date +"%Y-%m-%d %H:%M:%S")\"" + fi else - negref=$DIR/$tfile.newerneg + negref=$DIR/$tfile.negnewer.$x$y touch $negref || error "touch $negref failed" fi @@ -5795,6 +5963,7 @@ test_newerXY_base() { } test_56oc() { + test_newerXY_base "b" "t" test_newerXY_base "a" "a" test_newerXY_base "a" "m" test_newerXY_base "a" "c" @@ -5804,12 +5973,96 @@ test_56oc() { test_newerXY_base "c" "a" test_newerXY_base "c" "m" test_newerXY_base "c" "c" + test_newerXY_base "b" "b" test_newerXY_base "a" "t" test_newerXY_base "m" "t" test_newerXY_base "c" "t" + test_newerXY_base "b" "t" } run_test 56oc "check lfs find -newerXY work" +btime_supported() { + local dir=$DIR/$tdir + local rc + + mkdir -p $dir + touch $dir/$tfile + $LFS find $dir -btime -1d -type f + rc=$? + rm -rf $dir + return $rc +} + +test_56od() { + [ $MDS1_VERSION -lt $(version_code 2.13.53) ] && + ! btime_supported && skip "btime unsupported on MDS" + + [ $CLIENT_VERSION -lt $(version_code 2.13.53) ] && + ! btime_supported && skip "btime unsupported on clients" + + local dir=$DIR/$tdir + local ref=$DIR/$tfile.ref + local negref=$DIR/$tfile.negref + + mkdir $dir || error "mkdir $dir failed" + touch $dir/$tfile.n1 || error "touch $dir/$tfile.n1 failed" + touch $dir/$tfile.n2 || error "touch $dir/$tfile.n2 failed" + mkdir $dir/$tdir.n1 || error "mkdir $dir/$tdir.n1 failed" + mkdir $dir/$tdir.n2 || error "mkdir $dir/$tdir.n2 failed" + touch $ref || error "touch $ref failed" + # sleep 3 seconds at least + sleep 3 + + local before=$(do_facet mds1 date +%s) + local skew=$(($(date +%s) - before + 1)) + + if (( skew < 0 && skew > -5 )); then + sleep $((0 - skew + 1)) + skew=0 + fi + + # Set the dir stripe params to limit files all on MDT0, + # otherwise we need to calc the max clock skew between + # the client and MDTs. + setup_56 $dir/d.btime $NUMFILES $NUMDIRS "-i0 -c1" "-i0 -c1" + sleep 2 + touch $negref || error "touch $negref failed" + + local cmd="$LFS find $dir -newerbb $ref ! -newerbb $negref -type f" + local nums=$($cmd | wc -l) + local expected=$(((NUMFILES + 1) * NUMDIRS)) + + [ $nums -eq $expected ] || + error "'$cmd' wrong: found $nums, expected $expected" + + cmd="$LFS find $dir -newerbb $ref ! -newerbb $negref -type d" + nums=$($cmd | wc -l) + expected=$((NUMFILES + 1)) + [ $nums -eq $expected ] || + error "'$cmd' wrong: found $nums, expected $expected" + + [ $skew -lt 0 ] && return + + local after=$(do_facet mds1 date +%s) + local age=$((after - before + 1 + skew)) + + cmd="$LFS find $dir -btime -${age}s -type f" + nums=$($cmd | wc -l) + expected=$(((NUMFILES + 1) * NUMDIRS)) + + echo "Clock skew between client and server: $skew, age:$age" + [ $nums -eq $expected ] || + error "'$cmd' wrong: found $nums, expected $expected" + + expected=$(($NUMDIRS + 1)) + cmd="$LFS find $dir -btime -${age}s -type d" + nums=$($cmd | wc -l) + [ $nums -eq $expected ] || + error "'$cmd' wrong: found $nums, expected $expected" + rm -f $ref $negref || error "Failed to remove $ref $negref" +} +run_test 56od "check lfs find -btime with units" + test_56p() { [ $RUNAS_ID -eq $UID ] && skip_env "RUNAS_ID = UID = $UID -- skipping" @@ -5938,156 +6191,89 @@ test_56r() { } run_test 56r "check lfs find -size works" -test_56ra() { - [[ $MDS1_VERSION -gt $(version_code 2.12.58) ]] || - skip "MDS < 2.12.58 doesn't return LSOM data" - local dir=$DIR/$tdir - - [[ $OSC == "mdc" ]] && skip "DoM files" && return - - setup_56 $dir $NUMFILES $NUMDIRS "-c 1" +test_56ra_sub() { + local expected=$1 + local glimpses=$2 + local cmd="$3" cancel_lru_locks $OSC local rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - local expected=12 - local cmd="$LFS find -size 0 -type f -lazy $dir" local nums=$($cmd | wc -l) [ $nums -eq $expected ] || error "'$cmd' wrong: found $nums, expected $expected" local rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - [ $rpcs_before -eq $rpcs_after ] || - error "'$cmd' should not send glimpse RPCs to OST" - cmd="$LFS find -size 0 -type f $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - echo "Before: $rpcs_before After: $rpcs_after $NUMFILES" - $LCTL get_param osc.*.stats - [ $rpcs_after -eq $((rpcs_before + 12)) ] || - error "'$cmd' should send 12 glimpse RPCs to OST" + if (( rpcs_before + glimpses != rpcs_after )); then + echo "Before: $rpcs_before After: $rpcs_after $NUMFILES" + $LCTL get_param osc.*.stats | grep ldlm_glimpse_enqueue + + if [[ $glimpses == 0 ]]; then + error "'$cmd' should not send glimpse RPCs to OST" + else + error "'$cmd' should send $glimpses glimpse RPCs to OST" + fi + fi +} + +test_56ra() { + [[ $MDS1_VERSION -gt $(version_code 2.12.58) ]] || + skip "MDS < 2.12.58 doesn't return LSOM data" + local dir=$DIR/$tdir + + [[ $OSC == "mdc" ]] && skip "DoM files" && return + + setup_56 $dir $NUMFILES $NUMDIRS "-c 1" + # open and close all files to ensure LSOM is updated cancel_lru_locks $OSC - rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - expected=0 - cmd="$LFS find ! -size 0 -type f -lazy $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - $LCTL get_param mdc.*.stats - [ $rpcs_before -eq $rpcs_after ] || - error "'$cmd' should not send glimpse RPCs to OST" - cmd="$LFS find ! -size 0 -type f $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - echo "Before: $rpcs_before After: $rpcs_after $NUMFILES" - [ $rpcs_after -eq $((rpcs_before + 12)) ] || - error "'$cmd' should send 12 glimpse RPCs to OST" + find $dir -type f | xargs cat > /dev/null + + # expect_found glimpse_rpcs command_to_run + test_56ra_sub 12 0 "$LFS find -size 0 -type f -lazy $dir" + test_56ra_sub 12 12 "$LFS find -size 0 -type f $dir" + test_56ra_sub 0 0 "$LFS find ! -size 0 -type f -lazy $dir" + test_56ra_sub 0 12 "$LFS find ! -size 0 -type f $dir" echo "test" > $dir/$tfile echo "test2" > $dir/$tfile.2 && sync cancel_lru_locks $OSC - rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - expected=1 - cmd="$LFS find -size 5 -type f -lazy $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - [ $rpcs_before -eq $rpcs_after ] || - error "'$cmd' should not send glimpse RPCs to OST" - cmd="$LFS find -size 5 -type f $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - echo "Before: $rpcs_before After: $rpcs_after $NUMFILES" - [ $rpcs_after -eq $((rpcs_before + 14)) ] || - error "'$cmd' should send 14 glimpse RPCs to OST" + cat $dir/$tfile $dir/$tfile.2 > /dev/null - cancel_lru_locks $OSC - rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - expected=1 - cmd="$LFS find -size +5 -type f -lazy $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - [ $rpcs_before -eq $rpcs_after ] || - error "'$cmd' should not send glimpse RPCs to OST" - cmd="$LFS find -size +5 -type f $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - echo "Before: $rpcs_before After: $rpcs_after $NUMFILES" - [ $rpcs_after -eq $((rpcs_before + 14)) ] || - error "'$cmd' should send 14 glimpse RPCs to OST" + test_56ra_sub 1 0 "$LFS find -size 5 -type f -lazy $dir" + test_56ra_sub 1 14 "$LFS find -size 5 -type f $dir" + test_56ra_sub 1 0 "$LFS find -size +5 -type f -lazy $dir" + test_56ra_sub 1 14 "$LFS find -size +5 -type f $dir" - cancel_lru_locks $OSC - rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - expected=2 - cmd="$LFS find -size +0 -type f -lazy $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - [ $rpcs_before -eq $rpcs_after ] || - error "'$cmd' should not send glimpse RPCs to OST" - cmd="$LFS find -size +0 -type f $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - echo "Before: $rpcs_before After: $rpcs_after $NUMFILES" - [ $rpcs_after -eq $((rpcs_before + 14)) ] || - error "'$cmd' should send 14 glimpse RPCs to OST" + test_56ra_sub 2 0 "$LFS find -size +0 -type f -lazy $dir" + test_56ra_sub 2 14 "$LFS find -size +0 -type f $dir" + test_56ra_sub 2 0 "$LFS find ! -size -5 -type f -lazy $dir" + test_56ra_sub 2 14 "$LFS find ! -size -5 -type f $dir" + test_56ra_sub 12 0 "$LFS find -size -5 -type f -lazy $dir" + test_56ra_sub 12 14 "$LFS find -size -5 -type f $dir" +} +run_test 56ra "check lfs find -size -lazy works for data on OSTs" - cancel_lru_locks $OSC - rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - expected=2 - cmd="$LFS find ! -size -5 -type f -lazy $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - [ $rpcs_before -eq $rpcs_after ] || - error "'$cmd' should not send glimpse RPCs to OST" - cmd="$LFS find ! -size -5 -type f $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - echo "Before: $rpcs_before After: $rpcs_after $NUMFILES" - [ $rpcs_after -eq $((rpcs_before + 14)) ] || - error "'$cmd' should send 14 glimpse RPCs to OST" +test_56rb() { + local dir=$DIR/$tdir + local tmp=$TMP/$tfile.log + local mdt_idx; - cancel_lru_locks $OSC - rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - expected=12 - cmd="$LFS find -size -5 -type f -lazy $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - [ $rpcs_before -eq $rpcs_after ] || - error "'$cmd' should not send glimpse RPCs to OST" - cmd="$LFS find -size -5 -type f $dir" - nums=$($cmd | wc -l) - [ $nums -eq $expected ] || - error "'$cmd' wrong: found $nums, expected $expected" - rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) - echo "Before: $rpcs_before After: $rpcs_after $NUMFILES" - [ $rpcs_after -eq $((rpcs_before + 14)) ] || - error "'$cmd' should send 14 glimpse RPCs to OST" + test_mkdir -p $dir || error "failed to mkdir $dir" + $LFS setstripe -c 1 -i 0 $dir/$tfile || + error "failed to setstripe $dir/$tfile" + dd if=/dev/zero of=$dir/$tfile bs=1M count=1 + + stack_trap "rm -f $tmp" EXIT + $LFS find --size +100K --ost 0 $dir 2>&1 | tee $tmp + [ -z "$(cat $tmp | grep "obd_uuid: ")" ] || + error "failed to find --size +100K --ost 0 $dir" + $LFS find --size +100K --mdt $mdt_idx $dir 2>&1 | tee $tmp + [ -z "$(cat $tmp | grep "obd_uuid: ")" ] || + error "failed to find --size +100K --mdt $mdt_idx $dir" } -run_test 56ra "check lfs find -size -lazy works for data on OSTs" +run_test 56rb "check lfs find --size --ost/--mdt works" test_56s() { # LU-611 #LU-9369 [[ $OSTCOUNT -lt 2 ]] && skip_env "need at least 2 OSTs" @@ -6610,6 +6796,24 @@ test_56wd() { } run_test 56wd "check lfs_migrate --rsync and --no-rsync work" +test_56we() { + local td=$DIR/$tdir + local tf=$td/$tfile + + test_mkdir $td || error "cannot create $td" + touch $tf || error "cannot touch $tf" + + echo -n "Make sure --non-direct|-D works..." + $LFS_MIGRATE -y --non-direct -v $tf 2>&1 | + grep -q "lfs migrate --non-direct" || + error "--non-direct option cannot work correctly" + $LFS_MIGRATE -y -D -v $tf 2>&1 | + grep -q "lfs migrate -D" || + error "-D option cannot work correctly" + echo "done." +} +run_test 56we "check lfs_migrate --non-direct|-D support" + test_56x() { [[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs" check_swap_layouts_support @@ -6914,6 +7118,46 @@ test_56xe() { } run_test 56xe "migrate a composite layout file" +test_56xf() { + [[ $OSTCOUNT -ge 2 ]] || skip_env "needs >= 2 OSTs" + + [[ $MDS1_VERSION -ge $(version_code 2.13.53) ]] || + skip "Need server version at least 2.13.53" + + local dir=$DIR/$tdir + local f_comp=$dir/$tfile + local layout="-E 1M -c1 -E -1 -c2" + local fid_before="" + local fid_after="" + + test_mkdir "$dir" || error "cannot create dir $dir" + $LFS setstripe $layout $f_comp || + error "cannot setstripe $f_comp with layout $layout" + fid_before=$($LFS getstripe --fid $f_comp) + dd if=/dev/zero of=$f_comp bs=1M count=4 + + # 1. migrate a comp layout file to a comp layout + $LFS migrate $f_comp || error "cannot migrate $f_comp by lfs migrate" + fid_after=$($LFS getstripe --fid $f_comp) + [ "$fid_before" == "$fid_after" ] || + error "comp-to-comp migrate: $fid_before != $fid_after" + + # 2. migrate a comp layout file to a plain layout + $LFS migrate -c2 $f_comp || + error "cannot migrate $f_comp by lfs migrate" + fid_after=$($LFS getstripe --fid $f_comp) + [ "$fid_before" == "$fid_after" ] || + error "comp-to-plain migrate: $fid_before != $fid_after" + + # 3. migrate a plain layout file to a comp layout + $LFS migrate $layout $f_comp || + error "cannot migrate $f_comp by lfs migrate" + fid_after=$($LFS getstripe --fid $f_comp) + [ "$fid_before" == "$fid_after" ] || + error "plain-to-comp migrate: $fid_before != $fid_after" +} +run_test 56xf "FID is not lost during migration of a composite layout file" + test_56y() { [ $MDS1_VERSION -lt $(version_code 2.4.53) ] && skip "No HSM $(lustre_build_version $SINGLEMDS) MDS < 2.4.53" @@ -7573,8 +7817,8 @@ run_test 63b "async write errors should be returned to fsync ===" test_64a () { [ $PARALLEL == "yes" ] && skip "skip parallel run" - df $DIR - lctl get_param -n osc.*[oO][sS][cC][_-]*.cur* | grep "[0-9]" + lfs df $DIR + lctl get_param osc.*[oO][sS][cC][_-]*.cur* | grep "=[1-9]" } run_test 64a "verify filter grant calculations (in kernel) =====" @@ -9016,7 +9260,7 @@ test_101d() { [ $PARALLEL == "yes" ] && skip "skip parallel run" local file=$DIR/$tfile - local sz_MB=${FILESIZE_101d:-500} + local sz_MB=${FILESIZE_101d:-80} local ra_MB=${READAHEAD_MB:-40} local free_MB=$(($(df -P $DIR | tail -n 1 | awk '{ print $4 }') / 1024)) @@ -9036,24 +9280,30 @@ test_101d() { stack_trap "$LCTL set_param -n llite.*.max_read_ahead_mb $old_RA" EXIT $LCTL get_param -n llite.*.max_read_ahead_mb - echo Reading the test file $file with read-ahead disabled - local raOFF=$(do_and_time "dd if=$file of=/dev/null bs=1M count=$sz_MB") + echo "Reading the test file $file with read-ahead disabled" + local sz_KB=$((sz_MB * 1024 / 4)) + # 10485760 bytes transferred in 0.000938 secs (11179579337 bytes/sec) + # 104857600 bytes (105 MB) copied, 0.00876352 s, 12.0 GB/s + local raOFF=$(LANG=C dd if=$file of=/dev/null bs=4k count=$sz_KB |& + sed -e '/records/d' -e 's/.* \([0-9]*\.[0-9]*\) *s.*/\1/') - echo Cancel LRU locks on lustre client to flush the client cache + echo "Cancel LRU locks on lustre client to flush the client cache" cancel_lru_locks osc echo Enable read-ahead with ${ra_MB}MB $LCTL set_param -n llite.*.max_read_ahead_mb=$ra_MB - echo Reading the test file $file with read-ahead enabled - local raON=$(do_and_time "dd if=$file of=/dev/null bs=1M count=$sz_MB") + echo "Reading the test file $file with read-ahead enabled" + local raON=$(LANG=C dd if=$file of=/dev/null bs=4k count=$sz_KB |& + sed -e '/records/d' -e 's/.* \([0-9]*\.[0-9]*\) *s.*/\1/') echo "read-ahead disabled time read $raOFF" - echo "read-ahead enabled time read $raON" + echo "read-ahead enabled time read $raON" rm -f $file wait_delete_completed - [ $raOFF -le 1 ] || [ $raON -lt $raOFF ] || + # use awk for this check instead of bash because it handles decimals + awk "{ exit !($raOFF < 1.0 || $raOFF > $raON) }" <<<"ignore_me" || error "readahead ${raON}s > no-readahead ${raOFF}s ${sz_MB}M" } run_test 101d "file read with and without read-ahead enabled" @@ -9233,6 +9483,58 @@ test_101h() { } run_test 101h "Readahead should cover current read window" +test_101i() { + dd if=/dev/zero of=$DIR/$tfile bs=1M count=10 || + error "dd 10M file failed" + + local max_per_file_mb=$($LCTL get_param -n \ + llite.*.max_read_ahead_per_file_mb 2>/dev/null) + cancel_lru_locks osc + stack_trap "$LCTL set_param llite.*.max_read_ahead_per_file_mb=$max_per_file_mb" + $LCTL set_param llite.*.max_read_ahead_per_file_mb=1 || + error "set max_read_ahead_per_file_mb to 1 failed" + + echo "Reset readahead stats" + $LCTL set_param llite.*.read_ahead_stats=0 + + dd if=$DIR/$tfile of=/dev/null bs=2M + + $LCTL get_param llite.*.read_ahead_stats + local miss=$($LCTL get_param -n llite.*.read_ahead_stats | + awk '/misses/ { print $2 }') + [ $miss -eq 5 ] || error "expected misses 5 but got $miss" + rm -f $DIR/$tfile +} +run_test 101i "allow current readahead to exceed reservation" + +test_101j() { + $LFS setstripe -i 0 -c 1 $DIR/$tfile || + error "setstripe $DIR/$tfile failed" + local file_size=$((1048576 * 16)) + local old_ra=$($LCTL get_param -n llite.*.max_read_ahead_mb | head -n 1) + stack_trap "$LCTL set_param -n llite.*.max_read_ahead_mb $old_ra" EXIT + + echo Disable read-ahead + $LCTL set_param -n llite.*.max_read_ahead_mb=0 + + dd if=/dev/zero of=$DIR/$tfile bs=1M count=$(($file_size / 1048576)) + for blk in $PAGE_SIZE 1048576 $file_size; do + cancel_lru_locks osc + echo "Reset readahead stats" + $LCTL set_param -n llite.*.read_ahead_stats=0 + local count=$(($file_size / $blk)) + dd if=$DIR/$tfile bs=$blk count=$count of=/dev/null + local miss=$($LCTL get_param -n llite.*.read_ahead_stats | + get_named_value 'failed to fast read' | + cut -d" " -f1 | calc_total) + $LCTL get_param -n llite.*.read_ahead_stats + [ $miss -eq $count ] || error "expected $count got $miss" + done + + rm -f $p $DIR/$tfile +} +run_test 101j "A complete read block should be submitted when no RA" + setup_test102() { test_mkdir $DIR/$tdir chown $RUNAS_ID $DIR/$tdir @@ -9326,36 +9628,54 @@ test_102a() { } run_test 102a "user xattr test ==================================" +check_102b_layout() { + local layout="$*" + local testfile=$DIR/$tfile + + echo "test layout '$layout'" + $LFS setstripe $layout $testfile || error "setstripe failed" + $LFS getstripe -y $testfile + + echo "get/set/list trusted.lov xattr ..." # b=10930 + local value=$(getfattr -n trusted.lov -e hex $testfile | grep trusted) + [[ "$value" =~ "trusted.lov" ]] || + error "can't get trusted.lov from $testfile" + local stripe_count_orig=$($LFS getstripe -c $testfile) || + error "getstripe failed" + + $MCREATE $testfile.2 || error "mcreate $testfile.2 failed" + + value=$(cut -d= -f2 <<<$value) + # LU-13168: truncated xattr should fail if short lov_user_md header + [ $CLIENT_VERSION -lt $(version_code 2.13.53) ] && + lens="${#value}" || lens="$(seq 4 2 ${#value})" + for len in $lens; do + echo "setfattr $len $testfile.2" + setfattr -n trusted.lov -v ${value:0:$len} $testfile.2 && + [ $len -lt 66 ] && error "short xattr len=$len worked" + done + local stripe_size=$($LFS getstripe -S $testfile.2) + local stripe_count=$($LFS getstripe -c $testfile.2) + [[ $stripe_size -eq 65536 ]] || + error "stripe size $stripe_size != 65536" + [[ $stripe_count -eq $stripe_count_orig ]] || + error "stripe count $stripe_count != $stripe_count_orig" + rm $testfile $testfile.2 +} + test_102b() { [ -z "$(which setfattr 2>/dev/null)" ] && skip_env "could not find setfattr" [[ $OSTCOUNT -lt 2 ]] && skip_env "needs >= 2 OSTs" - # b10930: get/set/list trusted.lov xattr - echo "get/set/list trusted.lov xattr ..." - local testfile=$DIR/$tfile - $LFS setstripe -S 65536 -i 1 -c $OSTCOUNT $testfile || - error "setstripe failed" - local STRIPECOUNT=$($LFS getstripe -c $testfile) || - error "getstripe failed" - getfattr -d -m "^trusted" $testfile 2>/dev/null | grep "trusted.lov" || - error "can't get trusted.lov from $testfile" + # check plain layout + check_102b_layout -S 65536 -i 1 -c $OSTCOUNT - local testfile2=${testfile}2 - local value=$(getfattr -n trusted.lov $testfile 2>/dev/null | - grep "trusted.lov" | sed -e 's/[^=]\+=//') + # and also check composite layout + check_102b_layout -E 1M -S 65536 -i 1 -c $OSTCOUNT -Eeof -S4M - $MCREATE $testfile2 - setfattr -n trusted.lov -v $value $testfile2 - local stripe_size=$($LFS getstripe -S $testfile2) - local stripe_count=$($LFS getstripe -c $testfile2) - [[ $stripe_size -eq 65536 ]] || - error "stripe size $stripe_size != 65536" - [[ $stripe_count -eq $STRIPECOUNT ]] || - error "stripe count $stripe_count != $STRIPECOUNT" - rm -f $DIR/$tfile } -run_test 102b "getfattr/setfattr for trusted.lov EAs ============" +run_test 102b "getfattr/setfattr for trusted.lov EAs" test_102c() { [ -z "$(which setfattr 2>/dev/null)" ] && @@ -11135,8 +11455,8 @@ test_121() { #bug #10589 } run_test 121 "read cancel race =========" -test_123a() { # was test 123, statahead(bug 11401) - [ $PARALLEL == "yes" ] && skip "skip parallel run" +test_123a_base() { # was test 123, statahead(bug 11401) + local lsx="$1" SLOWOK=0 if ! grep -q "processor.*: 1" /proc/cpuinfo; then @@ -11148,76 +11468,120 @@ test_123a() { # was test 123, statahead(bug 11401) test_mkdir $DIR/$tdir NUMFREE=$(df -i -P $DIR | tail -n 1 | awk '{ print $4 }') [[ $NUMFREE -gt 100000 ]] && NUMFREE=100000 || NUMFREE=$((NUMFREE-1000)) - MULT=10 - for ((i=100, j=0; i<=$NUMFREE; j=$i, i=$((i * MULT)) )); do - createmany -o $DIR/$tdir/$tfile $j $((i - j)) - - max=`lctl get_param -n llite.*.statahead_max | head -n 1` - lctl set_param -n llite.*.statahead_max 0 - lctl get_param llite.*.statahead_max - cancel_lru_locks mdc - cancel_lru_locks osc - stime=`date +%s` - time ls -l $DIR/$tdir | wc -l - etime=`date +%s` - delta=$((etime - stime)) - log "ls $i files without statahead: $delta sec" - lctl set_param llite.*.statahead_max=$max - - swrong=`lctl get_param -n llite.*.statahead_stats | grep "statahead wrong:" | awk '{print $3}'` - lctl get_param -n llite.*.statahead_max | grep '[0-9]' - cancel_lru_locks mdc - cancel_lru_locks osc - stime=`date +%s` - time ls -l $DIR/$tdir | wc -l - etime=`date +%s` - delta_sa=$((etime - stime)) - log "ls $i files with statahead: $delta_sa sec" - lctl get_param -n llite.*.statahead_stats - ewrong=`lctl get_param -n llite.*.statahead_stats | grep "statahead wrong:" | awk '{print $3}'` + MULT=10 + for ((i=100, j=0; i<=$NUMFREE; j=$i, i=$((i * MULT)) )); do + createmany -o $DIR/$tdir/$tfile $j $((i - j)) + + max=$(lctl get_param -n llite.*.statahead_max | head -n 1) + lctl set_param -n llite.*.statahead_max 0 + lctl get_param llite.*.statahead_max + cancel_lru_locks mdc + cancel_lru_locks osc + stime=$(date +%s) + time $lsx $DIR/$tdir | wc -l + etime=$(date +%s) + delta=$((etime - stime)) + log "$lsx $i files without statahead: $delta sec" + lctl set_param llite.*.statahead_max=$max + + swrong=$(lctl get_param -n llite.*.statahead_stats | + grep "statahead wrong:" | awk '{print $3}') + lctl get_param -n llite.*.statahead_max | grep '[0-9]' + cancel_lru_locks mdc + cancel_lru_locks osc + stime=$(date +%s) + time $lsx $DIR/$tdir | wc -l + etime=$(date +%s) + delta_sa=$((etime - stime)) + log "$lsx $i files with statahead: $delta_sa sec" + lctl get_param -n llite.*.statahead_stats + ewrong=$(lctl get_param -n llite.*.statahead_stats | + grep "statahead wrong:" | awk '{print $3}') [[ $swrong -lt $ewrong ]] && log "statahead was stopped, maybe too many locks held!" [[ $delta -eq 0 || $delta_sa -eq 0 ]] && continue - if [ $((delta_sa * 100)) -gt $((delta * 105)) -a $delta_sa -gt $((delta + 2)) ]; then - max=`lctl get_param -n llite.*.statahead_max | head -n 1` - lctl set_param -n llite.*.statahead_max 0 - lctl get_param llite.*.statahead_max - cancel_lru_locks mdc - cancel_lru_locks osc - stime=`date +%s` - time ls -l $DIR/$tdir | wc -l - etime=`date +%s` - delta=$((etime - stime)) - log "ls $i files again without statahead: $delta sec" - lctl set_param llite.*.statahead_max=$max - if [ $((delta_sa * 100)) -gt $((delta * 105)) -a $delta_sa -gt $((delta + 2)) ]; then - if [ $SLOWOK -eq 0 ]; then - error "ls $i files is slower with statahead!" - else - log "ls $i files is slower with statahead!" - fi - break - fi - fi + if [ $((delta_sa * 100)) -gt $((delta * 105)) -a $delta_sa -gt $((delta + 2)) ]; then + max=$(lctl get_param -n llite.*.statahead_max | + head -n 1) + lctl set_param -n llite.*.statahead_max 0 + lctl get_param llite.*.statahead_max + cancel_lru_locks mdc + cancel_lru_locks osc + stime=$(date +%s) + time $lsx $DIR/$tdir | wc -l + etime=$(date +%s) + delta=$((etime - stime)) + log "$lsx $i files again without statahead: $delta sec" + lctl set_param llite.*.statahead_max=$max + if [ $((delta_sa * 100 > delta * 105 && delta_sa > delta + 2)) ]; then + if [ $SLOWOK -eq 0 ]; then + error "$lsx $i files is slower with statahead!" + else + log "$lsx $i files is slower with statahead!" + fi + break + fi + fi - [ $delta -gt 20 ] && break - [ $delta -gt 8 ] && MULT=$((50 / delta)) - [ "$SLOW" = "no" -a $delta -gt 5 ] && break - done - log "ls done" + [ $delta -gt 20 ] && break + [ $delta -gt 8 ] && MULT=$((50 / delta)) + [ "$SLOW" = "no" -a $delta -gt 5 ] && break + done + log "$lsx done" - stime=`date +%s` - rm -r $DIR/$tdir - sync - etime=`date +%s` - delta=$((etime - stime)) - log "rm -r $DIR/$tdir/: $delta seconds" - log "rm done" - lctl get_param -n llite.*.statahead_stats + stime=$(date +%s) + rm -r $DIR/$tdir + sync + etime=$(date +%s) + delta=$((etime - stime)) + log "rm -r $DIR/$tdir/: $delta seconds" + log "rm done" + lctl get_param -n llite.*.statahead_stats +} + +test_123aa() { + [ $PARALLEL == "yes" ] && skip "skip parallel run" + + test_123a_base "ls -l" } -run_test 123a "verify statahead work" +run_test 123aa "verify statahead work" + +test_123ab() { + [ $PARALLEL == "yes" ] && skip "skip parallel run" + + statx_supported || skip_env "Test must be statx() syscall supported" + + test_123a_base "$STATX -l" +} +run_test 123ab "verify statahead work by using statx" + +test_123ac() { + [ $PARALLEL == "yes" ] && skip "skip parallel run" + + statx_supported || skip_env "Test must be statx() syscall supported" + + local rpcs_before + local rpcs_after + local agl_before + local agl_after + + cancel_lru_locks $OSC + rpcs_before=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + agl_before=$($LCTL get_param -n llite.*.statahead_stats | + awk '/agl.total:/ {print $3}') + test_123a_base "$STATX -c \"%n %i %A %h %u %g %W %X %Z\" -D" + test_123a_base "$STATX --cached=always -D" + agl_after=$($LCTL get_param -n llite.*.statahead_stats | + awk '/agl.total:/ {print $3}') + [ $agl_before -eq $agl_after ] || + error "Should not trigger AGL thread - $agl_before:$agl_after" + rpcs_after=$(calc_stats $OSC.*$OSC*.stats ldlm_glimpse_enqueue) + [ $rpcs_after -eq $rpcs_before ] || + error "$STATX should not send glimpse RPCs to $OSC" +} +run_test 123ac "verify statahead work by using statx without glimpse RPCs" test_123b () { # statahead(bug 15027) [ $PARALLEL == "yes" ] && skip "skip parallel run" @@ -11742,7 +12106,7 @@ set_dir_limits () { check_mds_dmesg() { local facets=$(get_facets MDS) for facet in ${facets//,/ }; do - do_facet $facet "dmesg | tail -3 | grep -q $1" && return 0 + do_facet $facet "dmesg | tail -3 | grep $1" && return 0 done return 1 } @@ -11757,55 +12121,49 @@ test_129() { remote_mds_nodsh && skip "remote MDS with nodsh" local ENOSPC=28 - local EFBIG=27 local has_warning=false rm -rf $DIR/$tdir mkdir -p $DIR/$tdir # block size of mds1 - local maxsize=$(($($LCTL get_param -n mdc.*MDT0000*.blocksize) * 5)) - set_dir_limits $maxsize $maxsize + local maxsize=$(($($LCTL get_param -n mdc.*MDT0000*.blocksize) * 8)) + set_dir_limits $maxsize $((maxsize * 6 / 8)) + stack_trap "set_dir_limits 0 0" + stack_trap "unlinkmany $DIR/$tdir/file_base_ 2000 || true" local dirsize=$(stat -c%s "$DIR/$tdir") local nfiles=0 - while [[ $dirsize -le $maxsize ]]; do - $MULTIOP $DIR/$tdir/file_base_$nfiles Oc + while (( $dirsize <= $maxsize )); do + $MCREATE $DIR/$tdir/file_base_$nfiles rc=$? - if ! $has_warning; then - check_mds_dmesg '"is approaching"' && has_warning=true - fi # check two errors: - # ENOSPC for new ext4 max_dir_size (kernel commit df981d03ee) - # EFBIG for previous versions included in ldiskfs series - if [ $rc -eq $EFBIG ] || [ $rc -eq $ENOSPC ]; then + # ENOSPC for ext4 max_dir_size, which has been used since + # kernel v3.6-rc1-8-gdf981d03ee, lustre v2_4_50_0-79-gaed82035c0 + if (( rc == ENOSPC )); then set_dir_limits 0 0 - echo "return code $rc received as expected" + echo "rc=$rc returned as expected after $nfiles files" createmany -o $DIR/$tdir/file_extra_$nfiles. 5 || - error_exit "create failed w/o dir size limit" - - check_mds_dmesg '"has reached"' || - error_exit "reached message should be output" + error "create failed w/o dir size limit" - [ $has_warning = "false" ] && - error_exit "warning message should be output" + # messages may be rate limited if test is run repeatedly + check_mds_dmesg '"is approaching max"' || + echo "warning message should be output" + check_mds_dmesg '"has reached max"' || + echo "reached message should be output" dirsize=$(stat -c%s "$DIR/$tdir") [[ $dirsize -ge $maxsize ]] && return 0 - error_exit "current dir size $dirsize, " \ - "previous limit $maxsize" - elif [ $rc -ne 0 ]; then - set_dir_limits 0 0 - error_exit "return $rc received instead of expected " \ - "$EFBIG or $ENOSPC, files in dir $dirsize" + error "dirsize $dirsize < $maxsize after $nfiles files" + elif (( rc != 0 )); then + break fi nfiles=$((nfiles + 1)) dirsize=$(stat -c%s "$DIR/$tdir") done - set_dir_limits 0 0 - error "exceeded dir size limit $maxsize($MDSCOUNT) : $dirsize bytes" + error "rc=$rc, size=$dirsize/$maxsize, mdt=$MDSCOUNT, nfiles=$nfiles" } run_test 129 "test directory size limit ========================" @@ -12532,38 +12890,21 @@ test_133e() { } run_test 133e "Verifying OST {read,write}_bytes nid stats =================" -proc_regexp="/{proc,sys}/{fs,sys,kernel/debug}/{lustre,lnet}/" - -# Some versions of find (4.5.11, 4.5.14) included in CentOS 7.3-7.5 do -# not honor the -ignore_readdir_race option correctly. So we call -# error_ignore() rather than error() in these cases. See LU-11152. -error_133() { - if (find --version; do_facet mds1 find --version) | - grep -q '\b4\.5\.1[1-4]\b'; then - error_ignore LU-11152 "$@" - else - error "$@" - fi -} - test_133f() { - # First without trusting modes. - local proc_dirs=$(eval \ls -d $proc_regexp 2>/dev/null) - echo "proc_dirs='$proc_dirs'" - [ -n "$proc_dirs" ] || error "no proc_dirs on $HOSTNAME" - find $proc_dirs -exec cat '{}' \; &> /dev/null + [[ $(lustre_version_code $facet) -ge $(version_code 2.7.65) ]] || + skip "too old lustre for get_param -R ($facet_ver)" - # Second verifying readability. + # verifying readability. $LCTL get_param -R '*' &> /dev/null # Verifing writability with badarea_io. - find $proc_dirs \ - -ignore_readdir_race \ - -type f \ - -not -name force_lbug \ - -not -name changelog_mask \ - -exec badarea_io '{}' \; || - error_133 "find $proc_dirs failed" + $LCTL list_param -FR '*' | grep '=' | tr -d = | + egrep -v 'force_lbug|changelog_mask' | xargs badarea_io || + error "client badarea_io failed" + + # remount the FS in case writes/reads /proc break the FS + cleanup || error "failed to unmount" + setup || error "failed to setup" } run_test 133f "Check reads/writes of client lustre proc files with bad area io" @@ -12580,10 +12921,10 @@ test_133g() { log "$facet: too old lustre for get_param -R" fi if [ $facet_ver -ge $(version_code 2.5.54) ]; then - do_facet $facet "$LCTL list_param -R '*' | grep '=' | - tr -d= | egrep -v 'force_lbug|changelog_mask' | + do_facet $facet "$LCTL list_param -FR '*' | grep '=' | + tr -d = | egrep -v 'force_lbug|changelog_mask' | xargs badarea_io" || - error_133 "$facet badarea_io failed" + error "$facet badarea_io failed" else skip_noexit "$facet: too old lustre for get_param -R" fi @@ -12592,7 +12933,6 @@ test_133g() { # remount the FS in case writes/reads /proc break the FS cleanup || error "failed to unmount" setup || error "failed to setup" - true } run_test 133g "Check reads/writes of server lustre proc files with bad area io" @@ -12603,24 +12943,21 @@ test_133h() { skip "Need MDS version at least 2.9.54" local facet - for facet in client mds1 ost1; do - local facet_proc_dirs=$(do_facet $facet \ - \\\ls -d $proc_regexp 2> /dev/null) - [ -z "$facet_proc_dirs" ] && error "no proc_dirs on $facet" - echo "${facet}_proc_dirs='$facet_proc_dirs'" # Get the list of files that are missing the terminating newline - local missing=($(do_facet $facet \ - find ${facet_proc_dirs} -type f \| \ - while read F\; do \ - awk -v FS='\v' -v RS='\v\v' \ - "'END { if(NR>0 && \ - \\\$NF !~ /.*\\\n\$/) \ - print FILENAME}'" \ - '\$F'\; \ - done 2>/dev/null)) - [ ${#missing[*]} -eq 0 ] || - error "files do not end with newline: ${missing[*]}" + local plist=$(do_facet $facet + $LCTL list_param -FR '*' | grep '=' | tr -d =) + local ent + for ent in $plist; do + local missing=$(do_facet $facet $LCTL get_param $ent \|\ + awk -v FS='\v' -v RS='\v\v' \ + "'END { if(NR>0 && \\\$NF !~ /.*\\\n\$/) \ + print FILENAME}'" 2>/dev/null) + [ -z $missing ] || { + do_facet $facet $LCTL get_param $ent | od -An -tx1 + error "file does not end with newline: $facet-$ent" + } + done done } run_test 133h "Proc files should end with newlines" @@ -12809,7 +13146,7 @@ test_140() { #bug-17379 } run_test 140 "Check reasonable stack depth (shouldn't LBUG) ====" -test_150() { +test_150a() { [ $PARALLEL == "yes" ] && skip "skip parallel run" local TF="$TMP/$tfile" @@ -12827,20 +13164,73 @@ test_150() { cancel_lru_locks $OSC cmp $TF $DIR/$tfile || error "$TF $DIR/$tfile differ (truncate1)" - echo "12345" >>$TF - echo "12345" >>$DIR/$tfile + echo "12345" >>$TF + echo "12345" >>$DIR/$tfile + cancel_lru_locks $OSC + cmp $TF $DIR/$tfile || error "$TF $DIR/$tfile differ (append1)" + + echo "12345" >>$TF + echo "12345" >>$DIR/$tfile + cancel_lru_locks $OSC + cmp $TF $DIR/$tfile || error "$TF $DIR/$tfile differ (append2)" + + rm -f $TF + true +} +run_test 150a "truncate/append tests" + +test_150b() { + [ "$ost1_FSTYPE" != ldiskfs ] && skip "non-ldiskfs backend" + [ $OST1_VERSION -lt $(version_code 2.13.50) ] && + skip "Need OST version at least 2.13.53" + touch $DIR/$tfile + check_fallocate $DIR/$tfile || error "fallocate failed" +} +run_test 150b "Verify fallocate (prealloc) functionality" + +test_150c() { + local bytes + local want + + [ "$ost1_FSTYPE" != ldiskfs ] && skip "non-ldiskfs backend" + [ $OST1_VERSION -lt $(version_code 2.13.50) ] && + skip "Need OST version at least 2.13.53" + + $LFS setstripe -c $OSTCOUNT -S1M $DIR/$tdir || error "setstripe failed" + fallocate -l ${OSTCOUNT}m $DIR/$tdir || error "fallocate failed" + sync; sync_all_data cancel_lru_locks $OSC - cmp $TF $DIR/$tfile || error "$TF $DIR/$tfile differ (append1)" + sleep 5 + bytes=$(($(stat -c '%b * %B' $DIR/$tdir))) + want=$((OSTCOUNT * 1048576)) - echo "12345" >>$TF - echo "12345" >>$DIR/$tfile + # Must allocate all requested space, not more than 5% extra + (( $bytes >= $want && $bytes < $want * 105 / 100 )) || + error "bytes $bytes is not $want" +} +run_test 150c "Verify fallocate Size and Blocks" + +test_150d() { + local bytes + local want + + [ "$ost1_FSTYPE" != ldiskfs ] && skip "non-ldiskfs backend" + [ $OST1_VERSION -lt $(version_code 2.13.50) ] && + skip "Need OST version at least 2.13.53" + + $LFS setstripe -c $OSTCOUNT -S1M $DIR/$tdir || error "setstripe failed" + fallocate -o 1G -l ${OSTCOUNT}m $DIR/$tdir || error "fallocate failed" + sync; sync_all_data cancel_lru_locks $OSC - cmp $TF $DIR/$tfile || error "$TF $DIR/$tfile differ (append2)" + sleep 5 + bytes=$(($(stat -c '%b * %B' $DIR/$tdir))) + want=$((OSTCOUNT * 1048576)) - rm -f $TF - true + # Must allocate all requested space, not more than 5% extra + (( $bytes >= $want && $bytes < $want * 105 / 100 )) || + error "bytes $bytes is not $want" } -run_test 150 "truncate/append tests" +run_test 150d "Verify fallocate Size and Blocks - Non zero start" #LU-2902 roc_hit was not able to read all values from lproc function roc_hit_init() { @@ -13952,7 +14342,9 @@ test_160f() { local i # generate some changelog records to accumulate on each MDT - test_mkdir -c $MDSCOUNT $DIR/$tdir || error "test_mkdir $tdir failed" + # use fnv1a because created files should be evenly distributed + test_mkdir -c $MDSCOUNT -H fnv_1a_64 $DIR/$tdir || + error "test_mkdir $tdir failed" log "$(date +%s): creating first files" createmany -m $DIR/$tdir/$tfile $((MDSCOUNT * 2)) || error "create $DIR/$tdir/$tfile failed" @@ -14078,7 +14470,9 @@ test_160g() { local i # generate some changelog records to accumulate on each MDT - test_mkdir -c $MDSCOUNT $DIR/$tdir || error "mkdir $tdir failed" + # use fnv1a because created files should be evenly distributed + test_mkdir -c $MDSCOUNT -H fnv_1a_64 $DIR/$tdir || + error "mkdir $tdir failed" createmany -m $DIR/$tdir/$tfile $((MDSCOUNT * 2)) || error "create $DIR/$tdir/$tfile failed" @@ -14192,7 +14586,9 @@ test_160h() { local i # generate some changelog records to accumulate on each MDT - test_mkdir -c $MDSCOUNT $DIR/$tdir || error "test_mkdir $tdir failed" + # use fnv1a because created files should be evenly distributed + test_mkdir -c $MDSCOUNT -H fnv_1a_64 $DIR/$tdir || + error "test_mkdir $tdir failed" createmany -m $DIR/$tdir/$tfile $((MDSCOUNT * 2)) || error "create $DIR/$tdir/$tfile failed" @@ -14341,7 +14737,9 @@ test_160i() { changelog_register || error "first changelog_register failed" # generate some changelog records to accumulate on each MDT - test_mkdir -c $MDSCOUNT $DIR/$tdir || error "mkdir $tdir failed" + # use fnv1a because created files should be evenly distributed + test_mkdir -c $MDSCOUNT -H fnv_1a_64 $DIR/$tdir || + error "mkdir $tdir failed" createmany -m $DIR/$tdir/$tfile $((MDSCOUNT * 2)) || error "create $DIR/$tdir/$tfile failed" @@ -14406,7 +14804,9 @@ test_160j() { stack_trap "changelog_deregister" EXIT # generate some changelog - test_mkdir -c $MDSCOUNT $DIR/$tdir || error "mkdir $tdir failed" + # use fnv1a because created files should be evenly distributed + test_mkdir -c $MDSCOUNT -H fnv_1a_64 $DIR/$tdir || + error "mkdir $tdir failed" createmany -m $DIR/$tdir/${tfile}bis $((MDSCOUNT * 2)) || error "create $DIR/$tdir/${tfile}bis failed" @@ -14429,7 +14829,6 @@ test_160j() { error "User $cl_user not found in changelog_users" printf 'clear:'$cl_user':0' >&3 - } run_test 160j "client can be umounted while its chanangelog is being used" @@ -14788,6 +15187,233 @@ test_162c() { } run_test 162c "fid2path works with paths 100 or more directories deep" +oalr_event_count() { + local event="${1}" + local trace="${2}" + + awk -v name="${FSNAME}-OST0000" \ + -v event="${event}" \ + '$1 == "TRACE" && $2 == event && $3 == name' \ + "${trace}" | + wc -l +} + +oalr_expect_event_count() { + local event="${1}" + local trace="${2}" + local expect="${3}" + local count + + count=$(oalr_event_count "${event}" "${trace}") + if ((count == expect)); then + return 0 + fi + + error_noexit "${event} event count was '${count}', expected ${expect}" + cat "${trace}" >&2 + exit 1 +} + +cleanup_165() { + do_facet ost1 killall --quiet -KILL ofd_access_log_reader || true + stop ost1 + start ost1 "$(ostdevname 1)" $OST_MOUNT_OPTS +} + +setup_165() { + sync # Flush previous IOs so we can count log entries. + do_facet ost1 $LCTL set_param "obdfilter.${FSNAME}-OST0000.access_log_size=4096" + stack_trap cleanup_165 EXIT +} + +test_165a() { + local trace="/tmp/${tfile}.trace" + local rc + local count + + do_facet ost1 ofd_access_log_reader --debug=- --trace=- > "${trace}" & + setup_165 + sleep 5 + + do_facet ost1 ofd_access_log_reader --list + stop ost1 + + do_facet ost1 killall -TERM ofd_access_log_reader + wait + rc=$? + + if ((rc != 0)); then + error "ofd_access_log_reader exited with rc = '${rc}'" + fi + + # Parse trace file for discovery events: + oalr_expect_event_count alr_log_add "${trace}" 1 + oalr_expect_event_count alr_log_eof "${trace}" 1 + oalr_expect_event_count alr_log_free "${trace}" 1 +} +run_test 165a "ofd access log discovery" + +test_165b() { + local trace="/tmp/${tfile}.trace" + local file="${DIR}/${tfile}" + local pfid1 + local pfid2 + local -a entry + local rc + local count + local size + local flags + + setup_165 + + lfs setstripe -c 1 -i 0 "${file}" + $MULTIOP "${file}" oO_CREAT:O_DIRECT:O_WRONLY:w1048576c || error "cannot create '${file}'" + do_facet ost1 ofd_access_log_reader --list + + do_facet ost1 ofd_access_log_reader --debug=- --trace=- > "${trace}" & + sleep 5 + do_facet ost1 killall -TERM ofd_access_log_reader + wait + rc=$? + + if ((rc != 0)); then + error "ofd_access_log_reader exited with rc = '${rc}'" + fi + + oalr_expect_event_count alr_log_entry "${trace}" 1 + + pfid1=$($LFS path2fid "${file}") + + # 1 2 3 4 5 6 7 8 9 10 + # TRACE alr_log_entry OST PFID BEGIN END TIME SIZE COUNT FLAGS + entry=( - $(awk -v pfid="${pfid}" '$1 == "TRACE" && $2 == "alr_log_entry"' "${trace}" ) ) + + echo "entry = '${entry[*]}'" >&2 + + pfid2=${entry[4]} + if [[ "${pfid1}" != "${pfid2}" ]]; then + error "entry '${entry[*]}' has invalid PFID '${pfid2}', expected ${pfid1}" + fi + + size=${entry[8]} + if ((size != 1048576)); then + error "entry '${entry[*]}' has invalid io size '${size}', expected 1048576" + fi + + flags=${entry[10]} + if [[ "${flags}" != "w" ]]; then + error "entry '${entry[*]}' has invalid io flags '${flags}', expected 'w'" + fi + + do_facet ost1 ofd_access_log_reader --debug=- --trace=- > "${trace}" & + $MULTIOP "${file}" oO_CREAT:O_DIRECT:O_RDONLY:r524288c || error "cannot read '${file}'" + sleep 5 + do_facet ost1 killall -TERM ofd_access_log_reader + wait + rc=$? + + if ((rc != 0)); then + error "ofd_access_log_reader exited with rc = '${rc}'" + fi + + oalr_expect_event_count alr_log_entry "${trace}" 1 + + entry=( - $(awk -v pfid="${pfid}" '$1 == "TRACE" && $2 == "alr_log_entry"' "${trace}" ) ) + echo "entry = '${entry[*]}'" >&2 + + pfid2=${entry[4]} + if [[ "${pfid1}" != "${pfid2}" ]]; then + error "entry '${entry[*]}' has invalid PFID '${pfid2}', expected ${pfid1}" + fi + + size=${entry[8]} + if ((size != 524288)); then + error "entry '${entry[*]}' has invalid io size '${size}', 524288" + fi + + flags=${entry[10]} + if [[ "${flags}" != "r" ]]; then + error "entry '${entry[*]}' has invalid io flags '${flags}', expected 'r'" + fi +} +run_test 165b "ofd access log entries are produced and consumed" + +test_165c() { + local file="${DIR}/${tdir}/${tfile}" + test_mkdir "${DIR}/${tdir}" + + setup_165 + + lfs setstripe -c 1 -i 0 "${DIR}/${tdir}" + + # 4096 / 64 = 64. Create twice as many entries. + for ((i = 0; i < 128; i++)); do + $MULTIOP "${file}-${i}" oO_CREAT:O_WRONLY:w512c || error "cannot create file" + done + + sync + do_facet ost1 ofd_access_log_reader --list + unlinkmany "${file}-%d" 128 +} +run_test 165c "full ofd access logs do not block IOs" + +oal_peek_entry_count() { + do_facet ost1 ofd_access_log_reader --list | awk '$1 == "_entry_count:" { print $2; }' +} + +oal_expect_entry_count() { + local entry_count=$(oal_peek_entry_count) + local expect="$1" + + if ((entry_count == expect)); then + return 0 + fi + + error_noexit "bad entry count, got ${entry_count}, expected ${expect}" + do_facet ost1 ofd_access_log_reader --list >&2 + exit 1 +} + +test_165d() { + local trace="/tmp/${tfile}.trace" + local file="${DIR}/${tdir}/${tfile}" + local param="obdfilter.${FSNAME}-OST0000.access_log_mask" + local entry_count + test_mkdir "${DIR}/${tdir}" + + setup_165 + lfs setstripe -c 1 -i 0 "${file}" + + do_facet ost1 lctl set_param "${param}=rw" + $MULTIOP "${file}" oO_CREAT:O_DIRECT:O_WRONLY:w1048576c || error "cannot create '${file}'" + oal_expect_entry_count 1 + + $MULTIOP "${file}" oO_CREAT:O_DIRECT:O_RDONLY:r1048576c || error "cannot read '${file}'" + oal_expect_entry_count 2 + + do_facet ost1 lctl set_param "${param}=r" + $MULTIOP "${file}" oO_CREAT:O_DIRECT:O_WRONLY:w1048576c || error "cannot create '${file}'" + oal_expect_entry_count 2 + + $MULTIOP "${file}" oO_CREAT:O_DIRECT:O_RDONLY:r1048576c || error "cannot read '${file}'" + oal_expect_entry_count 3 + + do_facet ost1 lctl set_param "${param}=w" + $MULTIOP "${file}" oO_CREAT:O_DIRECT:O_WRONLY:w1048576c || error "cannot create '${file}'" + oal_expect_entry_count 4 + + $MULTIOP "${file}" oO_CREAT:O_DIRECT:O_RDONLY:r1048576c || error "cannot read '${file}'" + oal_expect_entry_count 4 + + do_facet ost1 lctl set_param "${param}=0" + $MULTIOP "${file}" oO_CREAT:O_DIRECT:O_WRONLY:w1048576c || error "cannot create '${file}'" + oal_expect_entry_count 4 + + $MULTIOP "${file}" oO_CREAT:O_DIRECT:O_RDONLY:r1048576c || error "cannot read '${file}'" + oal_expect_entry_count 4 +} +run_test 165d "ofd_access_log mask works" + test_169() { # do directio so as not to populate the page cache log "creating a 10 Mb file" @@ -15727,16 +16353,20 @@ test_205a() { # Job stats } run_test 205a "Verify job stats" -# LU-13117 +# LU-13117, LU-13597 test_205b() { + job_stats="mdt.*.job_stats" + $LCTL set_param $job_stats=clear $LCTL set_param jobid_var=USER jobid_name="%e.%u" env -i USERTESTJOBSTATS=foolish touch $DIR/$tfile.1 - do_facet $SINGLEMDS $LCTL get_param mdt.*.job_stats | - grep job_id: | grep foolish && + do_facet $SINGLEMDS $LCTL get_param $job_stats | + grep "job_id:.*foolish" && error "Unexpected jobid found" - true + do_facet $SINGLEMDS $LCTL get_param $job_stats | + grep "open:.*min.*max.*sum" || + error "wrong job_stats format found" } -run_test 205b "Verify job stats jobid parsing" +run_test 205b "Verify job stats jobid and output format" # LU-1480, LU-1773 and LU-1657 test_206() { @@ -15861,6 +16491,27 @@ test_209() { } run_test 209 "read-only open/close requests should be freed promptly" +test_210() { + local pid + + $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:eW_E+eUc & + pid=$! + sleep 1 + + $LFS getstripe $DIR/$tfile + kill -USR1 $pid + wait $pid || error "multiop failed" + + $MULTIOP $DIR/$tfile oO_RDONLY:eR_E+eUc & + pid=$! + sleep 1 + + $LFS getstripe $DIR/$tfile + kill -USR1 $pid + wait $pid || error "multiop failed" +} +run_test 210 "lfs getstripe does not break leases" + test_212() { size=`date +%s` size=$((size % 8192 + 1)) @@ -16682,6 +17333,27 @@ test_230b() { ln -s $migrate_dir/$tfile $migrate_dir/${tfile}_ln ln -s $other_dir/$tfile $migrate_dir/${tfile}_ln_other + local len + local lnktgt + + # inline symlink + for len in 58 59 60; do + lnktgt=$(str_repeat 'l' $len) + touch $migrate_dir/$lnktgt + ln -s $lnktgt $migrate_dir/${len}char_ln + done + + # PATH_MAX + for len in 4094 4095; do + lnktgt=$(str_repeat 'l' $len) + ln -s $lnktgt $migrate_dir/${len}char_ln + done + + # NAME_MAX + for len in 254 255; do + touch $migrate_dir/$(str_repeat 'l' $len) + done + $LFS migrate -m $MDTIDX $migrate_dir || error "fails on migrating remote dir to MDT1" @@ -16689,7 +17361,8 @@ test_230b() { for ((i = 0; i < 10; i++)); do for file in $(find $migrate_dir/dir_${i}); do mdt_index=$($LFS getstripe -m $file) - [ $mdt_index == $MDTIDX ] || + # broken symlink getstripe will fail + [ $mdt_index -ne $MDTIDX ] && stat -L $file && error "$file is not on MDT${MDTIDX}" done done @@ -16753,7 +17426,7 @@ test_230b() { echo "migrate back to MDT0, checking.." for file in $(find $migrate_dir); do mdt_index=$($LFS getstripe -m $file) - [ $mdt_index == $MDTIDX ] || + [ $mdt_index -ne $MDTIDX ] && stat -L $file && error "$file is not on MDT${MDTIDX}" done @@ -16907,7 +17580,7 @@ test_230d() { old_count=$((MDSCOUNT - old_index)) new_index=$((RANDOM % MDSCOUNT)) new_count=$((MDSCOUNT - new_index)) - new_hash="all_char" + new_hash=1 # for all_char [ $old_count -gt 1 ] && old_count=$((old_count - RANDOM % old_count)) [ $new_count -gt 1 ] && new_count=$((new_count - RANDOM % new_count)) @@ -17099,7 +17772,7 @@ run_test 230i "lfs migrate -m tolerates trailing slashes" test_230j() { [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" - [ $MDS1_VERSION -lt $(version_code 2.11.52) ] && + [ $MDS1_VERSION -lt $(version_code 2.13.52) ] && skip "Need MDS version at least 2.11.52" $LFS mkdir -m 0 -c 1 $DIR/$tdir || error "mkdir $tdir failed" @@ -17107,7 +17780,7 @@ test_230j() { error "create $tfile failed" cat /etc/passwd > $DIR/$tdir/$tfile - $LFS migrate -m 1 $DIR/$tdir + $LFS migrate -m 1 $DIR/$tdir || error "migrate failed" cmp /etc/passwd $DIR/$tdir/$tfile || error "DoM file mismatch after migration" @@ -17231,6 +17904,171 @@ test_230m() { } run_test 230m "xattrs not changed after dir migration" +test_230n() { + [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" + [[ $MDS1_VERSION -ge $(version_code 2.13.53) ]] || + skip "Need MDS version at least 2.13.53" + + $LFS mkdir -i 0 $DIR/$tdir || error "mkdir $tdir failed" + cat /etc/hosts > $DIR/$tdir/$tfile + $LFS mirror extend -N1 $DIR/$tdir/$tfile || error "Mirroring failed" + $LFS migrate -m 1 $DIR/$tdir || error "Migration failed" + + cmp /etc/hosts $DIR/$tdir/$tfile || + error "File data mismatch after migration" +} +run_test 230n "Dir migration with mirrored file" + +test_230o() { + [ $MDSCOUNT -ge 2 ] || skip "needs >= 2 MDTs" + [ $MDS1_VERSION -ge $(version_code 2.13.52) ] || + skip "Need MDS version at least 2.13.52" + + local mdts=$(comma_list $(mdts_nodes)) + + local restripe_status + local delta + local i + local j + + # in case "crush" hash type is not set + do_nodes $mdts "$LCTL set_param lod.*.mdt_hash=crush" + + restripe_status=$(do_facet mds1 $LCTL get_param -n \ + mdt.*MDT0000.enable_dir_restripe) + do_nodes $mdts "$LCTL set_param mdt.*.enable_dir_restripe=1" + stack_trap "do_nodes $mdts $LCTL set_param \ + mdt.*.enable_dir_restripe=$restripe_status" + + mkdir $DIR/$tdir + createmany -m $DIR/$tdir/f 100 || + error "create files under remote dir failed $i" + createmany -d $DIR/$tdir/d 100 || + error "create dirs under remote dir failed $i" + + for i in $(seq 2 $MDSCOUNT); do + do_nodes $mdts "$LCTL set_param mdt.*.md_stats=clear > /dev/null" + $LFS setdirstripe -c $i $DIR/$tdir || + error "split -c $i $tdir failed" + wait_update $HOSTNAME \ + "$LFS getdirstripe -H $DIR/$tdir" "crush" 100 || + error "dir split not finished" + delta=$(do_nodes $mdts "lctl get_param -n mdt.*MDT*.md_stats" | + awk '/migrate/ {sum += $2} END { print sum }') + echo "$delta files migrated when dir split from $((i - 1)) to $i stripes" + # delta is around total_files/stripe_count + [ $delta -lt $((200 /(i - 1))) ] || + error "$delta files migrated" + done +} +run_test 230o "dir split" + +test_230p() { + [ $MDSCOUNT -ge 2 ] || skip "needs >= 2 MDTs" + [ $MDS1_VERSION -ge $(version_code 2.13.52) ] || + skip "Need MDS version at least 2.13.52" + + local mdts=$(comma_list $(mdts_nodes)) + + local restripe_status + local delta + local i + local j + + do_nodes $mdts "$LCTL set_param lod.*.mdt_hash=crush" + + restripe_status=$(do_facet mds1 $LCTL get_param -n \ + mdt.*MDT0000.enable_dir_restripe) + do_nodes $mdts "$LCTL set_param mdt.*.enable_dir_restripe=1" + stack_trap "do_nodes $mdts $LCTL set_param \ + mdt.*.enable_dir_restripe=$restripe_status" + + test_mkdir -c $MDSCOUNT -H crush $DIR/$tdir + createmany -m $DIR/$tdir/f 100 || + error "create files under remote dir failed $i" + createmany -d $DIR/$tdir/d 100 || + error "create dirs under remote dir failed $i" + + for i in $(seq $((MDSCOUNT - 1)) -1 1); do + local mdt_hash="crush" + + do_nodes $mdts "$LCTL set_param mdt.*.md_stats=clear > /dev/null" + $LFS setdirstripe -c $i $DIR/$tdir || + error "split -c $i $tdir failed" + [ $i -eq 1 ] && mdt_hash="none" + wait_update $HOSTNAME \ + "$LFS getdirstripe -H $DIR/$tdir" $mdt_hash 100 || + error "dir merge not finished" + delta=$(do_nodes $mdts "lctl get_param -n mdt.*MDT*.md_stats" | + awk '/migrate/ {sum += $2} END { print sum }') + echo "$delta files migrated when dir merge from $((i + 1)) to $i stripes" + # delta is around total_files/stripe_count + [ $delta -lt $((200 / i)) ] || + error "$delta files migrated" + done +} +run_test 230p "dir merge" + +test_230q() { + [ $MDSCOUNT -ge 2 ] || skip "needs >= 2 MDTs" + [ $MDS1_VERSION -ge $(version_code 2.13.52) ] || + skip "Need MDS version at least 2.13.52" + + local mdts=$(comma_list $(mdts_nodes)) + local saved_threshold=$(do_facet mds1 \ + $LCTL get_param -n mdt.*-MDT0000.dir_split_count) + local saved_delta=$(do_facet mds1 \ + $LCTL get_param -n mdt.*-MDT0000.dir_split_delta) + local threshold=100 + local delta=2 + local total=0 + local stripe_count=0 + local stripe_index + local nr_files + + stack_trap "do_nodes $mdts $LCTL set_param \ + mdt.*.dir_split_count=$saved_threshold" + stack_trap "do_nodes $mdts $LCTL set_param \ + mdt.*.dir_split_delta=$saved_delta" + stack_trap "do_nodes $mdts $LCTL set_param mdt.*.dir_restripe_nsonly=1" + do_nodes $mdts "$LCTL set_param mdt.*.enable_dir_auto_split=1" + do_nodes $mdts "$LCTL set_param mdt.*.dir_split_count=$threshold" + do_nodes $mdts "$LCTL set_param mdt.*.dir_split_delta=$delta" + do_nodes $mdts "$LCTL set_param mdt.*.dir_restripe_nsonly=0" + do_nodes $mdts "$LCTL set_param lod.*.mdt_hash=crush" + + $LFS mkdir -i -1 -c 1 $DIR/$tdir || error "mkdir $tdir failed" + stripe_index=$($LFS getdirstripe -i $DIR/$tdir) + + while [ $stripe_count -lt $MDSCOUNT ]; do + createmany -m $DIR/$tdir/f $total $((threshold * 3 / 2)) || + error "create sub files failed" + stat $DIR/$tdir > /dev/null + total=$((total + threshold * 3 / 2)) + stripe_count=$((stripe_count + delta)) + [ $stripe_count -gt $MDSCOUNT ] && stripe_count=$MDSCOUNT + + wait_update $HOSTNAME \ + "$LFS getdirstripe -c $DIR/$tdir" "$stripe_count" 40 || + error "stripe count $($LFS getdirstripe -c $DIR/$tdir) != $stripe_count" + + wait_update $HOSTNAME \ + "$LFS getdirstripe -H $DIR/$tdir" "crush" 200 || + error "stripe hash $($LFS getdirstripe -H $DIR/$tdir) != crush" + + nr_files=$($LFS getstripe -m $DIR/$tdir/* | + grep -w $stripe_index | wc -l) + echo "$nr_files files on MDT$stripe_index after split" + [ $nr_files -lt $((total / (stripe_count - 1))) ] || + error "$nr_files files on MDT$stripe_index after split" + + nr_files=$(ls $DIR/$tdir | wc -w) + [ $nr_files -eq $total ] || + error "total sub files $nr_files != $total" + done +} +run_test 230q "dir auto split" + test_231a() { # For simplicity this test assumes that max_pages_per_rpc @@ -17727,6 +18565,38 @@ test_247e() { } run_test 247e "mount .. as fileset" +test_247f() { + [ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs" + [ $MDS1_VERSION -lt $(version_code 2.13.52) ] && + skip "Need at least version 2.13.52" + lctl get_param -n mdc.$FSNAME-MDT0000*.import | + grep -q subtree || + skip "Fileset feature is not supported" + + mkdir $DIR/$tdir || error "mkdir $tdir failed" + $LFS mkdir -i $((MDSCOUNT - 1)) $DIR/$tdir/remote || + error "mkdir remote failed" + mkdir $DIR/$tdir/remote/subdir || error "mkdir remote/subdir failed" + $LFS mkdir -c $MDSCOUNT $DIR/$tdir/striped || + error "mkdir striped failed" + mkdir $DIR/$tdir/striped/subdir || error "mkdir striped/subdir failed" + + local submount=${MOUNT}_$tdir + + mkdir -p $submount || error "mkdir $submount failed" + + local dir + local fileset=$FILESET + + for dir in $tdir/remote $tdir/remote/subdir \ + $tdir/striped $tdir/striped/subdir $tdir/striped/. ; do + FILESET="$fileset/$dir" mount_client $submount || + error "mount $dir failed" + umount_client $submount + done +} +run_test 247f "mount striped or remote directory as fileset" + test_248a() { local fast_read_sav=$($LCTL get_param -n llite.*.fast_read 2>/dev/null) [ -z "$fast_read_sav" ] && skip "no fast read support" @@ -18881,6 +19751,101 @@ test_270f() { } run_test 270f "DoM: maximum DoM stripe size checks" +test_270g() { + [ $MDS1_VERSION -ge $(version_code 2.13.52) ] || + skip "Need MDS version at least 2.13.52" + local dom=$DIR/$tdir/$tfile + + $LFS mkdir -i 0 -c 1 $DIR/$tdir + local lodname=${FSNAME}-MDT0000-mdtlov + + local save="$TMP/$TESTSUITE-$TESTNAME.parameters" + save_lustre_params mds1 "lod.${lodname}.dom_stripesize_max_kb" > $save + save_lustre_params mds1 "lod.${lodname}.dom_threshold_free_mb" >> $save + stack_trap "restore_lustre_params < $save; rm -f $save" EXIT + + local dom_limit=1024 + local dom_threshold="50%" + + $LFS setstripe -d $DIR/$tdir + $LFS setstripe -E ${dom_limit}K -L mdt $DIR/$tdir || + error "Can't set directory default striping" + + do_facet mds1 $LCTL set_param -n \ + lod.${lodname}.dom_stripesize_max_kb=$dom_limit + # set 0 threshold and create DOM file to change tunable stripesize + do_facet mds1 $LCTL set_param -n lod.${lodname}.dom_threshold_free_mb=0 + $LFS setstripe -E ${dom_limit}K -L mdt -E -1 $dom || + error "Failed to create $dom file" + # now tunable dom_cur_stripesize should reach maximum + local dom_current=$(do_facet mds1 $LCTL get_param -n \ + lod.${lodname}.dom_stripesize_cur_kb) + [[ $dom_current == $dom_limit ]] || + error "Current DOM stripesize is not maximum" + rm $dom + + # set threshold for further tests + do_facet mds1 $LCTL set_param -n \ + lod.${lodname}.dom_threshold_free_mb=$dom_threshold + echo "DOM threshold is $dom_threshold free space" + local dom_def + local dom_set + # Spoof bfree to exceed threshold + #define OBD_FAIL_MDS_STATFS_SPOOF 0x168 + do_facet mds1 $LCTL set_param -n fail_loc=0x0168 + for spfree in 40 20 0 15 30 55; do + do_facet mds1 $LCTL set_param -n fail_val=$spfree + $LFS setstripe -E ${dom_limit}K -L mdt -E -1 $dom || + error "Failed to create $dom file" + dom_def=$(do_facet mds1 $LCTL get_param -n \ + lod.${lodname}.dom_stripesize_cur_kb) + echo "Free space: ${spfree}%, default DOM stripe: ${dom_def}K" + [[ $dom_def != $dom_current ]] || + error "Default stripe size was not changed" + if [[ $spfree > 0 ]] ; then + dom_set=$($LFS getstripe -S $dom) + [[ $dom_set == $((dom_def * 1024)) ]] || + error "DOM component size is still old" + else + [[ $($LFS getstripe -L $dom) != "mdt" ]] || + error "DoM component is set with no free space" + fi + rm $dom + dom_current=$dom_def + done +} +run_test 270g "DoM: default DoM stripe size depends on free space" + +test_270h() { + [[ $MDS1_VERSION -ge $(version_code 2.13.53) ]] || + skip "Need MDS version at least 2.13.53" + + local mdtname=${FSNAME}-MDT0000-mdtlov + local dom=$DIR/$tdir/$tfile + local save="$TMP/$TESTSUITE-$TESTNAME.parameters" + + save_lustre_params mds1 "lod.*.dom_stripesize" > $save + stack_trap "restore_lustre_params < $save; rm -f $save" EXIT + + $LFS mkdir -i 0 -c 1 $DIR/$tdir + $LFS setstripe -E 1M -c1 -E -1 -c2 ${dom}_1 || + error "can't create OST file" + # mirrored file with DOM entry in the second mirror + $LFS mirror extend -N -E 1M -L mdt -E eof -c2 ${dom}_1 || + error "can't create mirror with DoM component" + + do_facet mds1 $LCTL set_param -n lod.$mdtname.dom_stripesize=0 + + # DOM component in the middle and has other enries in the same mirror, + # should succeed but lost DoM component + $LFS setstripe --copy=${dom}_1 $dom || + error "Can't create file from OST|DOM mirror layout" + # check new file has no DoM layout after all + [[ $($LFS getstripe -L $dom) != "mdt" ]] || + error "File has DoM component while DoM is disabled" +} +run_test 270h "DoM: DoM stripe removal when disabled on server" + test_271a() { [ $MDS1_VERSION -lt $(version_code 2.10.55) ] && skip "Need MDS version at least 2.10.55" @@ -19466,6 +20431,25 @@ test_278() { } run_test 278 "Race starting MDS between MDTs stop/start" +test_280() { + [ $MGS_VERSION -lt $(version_code 2.13.52) ] && + skip "Need MGS version at least 2.13.52" + [ $PARALLEL == "yes" ] && skip "skip parallel run" + combined_mgs_mds || skip "needs combined MGS/MDT" + + umount_client $MOUNT +#define OBD_FAIL_MDS_LLOG_UMOUNT_RACE 0x15e + do_facet mgs $LCTL set_param fail_loc=0x8000015e fail_val=0 + + mount_client $MOUNT & + sleep 1 + stop mgs || error "stop mgs failed" + #for a race mgs would crash + start mgs $(mgsdevname) $MGS_MOUNT_OPTS || error "start mgs failed" + mount_client $MOUNT || error "mount client failed" +} +run_test 280 "Race between MGS umount and client llog processing" + cleanup_test_300() { trap 0 umask $SAVE_UMASK @@ -19619,12 +20603,22 @@ test_300d() { #local striped directory $LFS setdirstripe -i 0 -c 2 -H all_char $DIR/$tdir/striped_dir || error "set striped dir error" + #look at the directories for debug purposes + ls -l $DIR/$tdir + $LFS getdirstripe $DIR/$tdir + ls -l $DIR/$tdir/striped_dir + $LFS getdirstripe $DIR/$tdir/striped_dir createmany -o $DIR/$tdir/striped_dir/f 10 || error "create 10 files failed" #remote striped directory $LFS setdirstripe -i 1 -c 2 $DIR/$tdir/remote_striped_dir || error "set striped dir error" + #look at the directories for debug purposes + ls -l $DIR/$tdir + $LFS getdirstripe $DIR/$tdir + ls -l $DIR/$tdir/remote_striped_dir + $LFS getdirstripe $DIR/$tdir/remote_striped_dir createmany -o $DIR/$tdir/remote_striped_dir/f 10 || error "create 10 files failed" @@ -20473,7 +21467,7 @@ test_316() { chown nobody $DIR/$tdir/d touch $DIR/$tdir/d/file - $LFS mv -M1 $DIR/$tdir/d || error "lfs mv failed" + $LFS mv -m1 $DIR/$tdir/d || error "lfs mv failed" } run_test 316 "lfs mv" @@ -20555,9 +21549,8 @@ test_318() { llite.*.max_read_ahead_async_active 2>/dev/null) [ $max_active -ne 256 ] && error "expected 256 but got $max_active" - # currently reset to 0 is unsupported, leave it 512 for now. - $LCTL set_param llite.*.max_read_ahead_async_active=0 && - error "set max_read_ahead_async_active should fail" + $LCTL set_param llite.*.max_read_ahead_async_active=0 || + error "set max_read_ahead_async_active should succeed" $LCTL set_param llite.*.max_read_ahead_async_active=512 max_active=$($LCTL get_param -n \ @@ -20857,6 +21850,7 @@ test_401a() { #LU-7437 #count the number of parameters by "list_param -R" local params=$($LCTL list_param -R '*' 2>/dev/null | wc -l) #count the number of parameters by listing proc files + local proc_regexp="/{proc,sys}/{fs,sys,kernel/debug}/{lustre,lnet}/" local proc_dirs=$(eval \ls -d $proc_regexp 2>/dev/null) echo "proc_dirs='$proc_dirs'" [ -n "$proc_dirs" ] || error "no proc_dirs on $HOSTNAME" @@ -22017,6 +23011,14 @@ test_423() { } run_test 423 "statfs should return a right data" +test_424() { +#define OBD_FAIL_PTLRPC_BULK_REPLY_ATTACH 0x522 | OBD_FAIL_ONCE + $LCTL set_param fail_loc=0x80000522 + dd if=/dev/zero of=$DIR/$tfile bs=2M count=1 oflag=sync + rm -f $DIR/$tfile +} +run_test 424 "simulate ENOMEM in ptl_send_rpc bulk reply ME attach" + prep_801() { [[ $MDS1_VERSION -lt $(version_code 2.9.55) ]] || [[ $OST1_VERSION -lt $(version_code 2.9.55) ]] && @@ -22494,7 +23496,8 @@ test_805() { do_facet $SINGLEMDS zfs set quota=$(((usedkb+freekb)*1024)) $fsset trap cleanup_805 EXIT mkdir $DIR/$tdir - $LFS setstripe -E 1M -L mdt $DIR/$tdir || error "DoM not working" + $LFS setstripe -E 1M -c2 -E 4M -c2 -E -1 -c2 $DIR/$tdir || + error "Can't set PFL layout" createmany -m $DIR/$tdir/f- 1000000 && error "ENOSPC wasn't met" rm -rf $DIR/$tdir || error "not able to remove" do_facet $SINGLEMDS zfs set quota=$old $fsset @@ -22646,7 +23649,7 @@ test_807() { error "truncate $tdir/trunc failed" local bs=1048576 - dd if=/dev/zero of=$DIR/$tdir/single_dd bs=$bs count=1 || + dd if=/dev/zero of=$DIR/$tdir/single_dd bs=$bs count=1 conv=fsync || error "write $tfile failed" # multi-client wirtes @@ -22667,7 +23670,8 @@ test_807() { wait ${pids[$i]} done - sleep 5 + do_rpc_nodes "$CLIENTS" cancel_lru_locks osc + do_nodes "$CLIENTS" "sync ; sleep 5 ; sync" $LSOM_SYNC -u $cl_user -m $FSNAME-MDT0000 $MOUNT check_lsom_data $DIR/$tdir/trunc check_lsom_data $DIR/$tdir/single_dd @@ -22767,22 +23771,6 @@ test_810() { } run_test 810 "partial page writes on ZFS (LU-11663)" -test_811() { - [ $MDS1_VERSION -lt $(version_code 2.11.56) ] && - skip "Need MDS version at least 2.11.56" - - #define OBD_FAIL_MDS_ORPHAN_DELETE 0x165 - do_facet mds1 $LCTL set_param fail_loc=0x165 - $MULTIOP $DIR/$tfile Ouc || error "multiop failed" - - stop mds1 - start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS - - wait_update_facet mds1 "pgrep orph_.*-MDD | wc -l" "0" || - error "MDD orphan cleanup thread not quit" -} -run_test 811 "orphan name stub can be cleaned up in startup" - test_812a() { [ $OST1_VERSION -lt $(version_code 2.12.51) ] && skip "OST < 2.12.51 doesn't support this fail_loc" @@ -23177,6 +24165,18 @@ test_901() { } run_test 901 "don't leak a mgc lock on client umount" +# LU-13377 +test_902() { + [ $CLIENT_VERSION -lt $(version_code 2.13.52) ] && + skip "client does not have LU-13377 fix" + #define OBD_FAIL_LLITE_SHORT_COMMIT 0x1415 + $LCTL set_param fail_loc=0x1415 + dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 + cancel_lru_locks osc + rm -f $DIR/$tfile +} +run_test 902 "test short write doesn't hang lustre" + complete $SECONDS [ -f $EXT2_DEV ] && rm $EXT2_DEV || true check_and_cleanup_lustre