X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fsanity.sh;h=39ab34d34731453fe63286aa2839c80f97bb0d8d;hb=4894683342d77964daeded9fbc608fc46aa479ee;hp=ea94563e795f2d257dcd497d512179cfc92e3a38;hpb=c7d8fe31064990d7053436ccd720c531bc78a2dc;p=fs%2Flustre-release.git diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh index ea94563..39ab34d 100755 --- a/lustre/tests/sanity.sh +++ b/lustre/tests/sanity.sh @@ -77,8 +77,8 @@ if (( $LINUX_VERSION_CODE >= $(version_code 4.18.0) && ALWAYS_EXCEPT+=" 411" fi -# 5 12 8 12 (min)" -[ "$SLOW" = "no" ] && EXCEPT_SLOW="27m 60i 64b 68 71 115 135 136 300o" +# 5 12 8 12 15 (min)" +[ "$SLOW" = "no" ] && EXCEPT_SLOW="27m 60i 64b 68 71 115 135 136 230d 300o" if [ "$mds1_FSTYPE" = "zfs" ]; then # bug number for skipped test: @@ -92,6 +92,8 @@ if [ "$ost1_FSTYPE" = "zfs" ]; then ALWAYS_EXCEPT+=" 130a 130b 130c 130d 130e 130f 130g" fi +proc_regexp="/{proc,sys}/{fs,sys,kernel/debug}/{lustre,lnet}/" + # Get the SLES distro version # # Returns a version string that should only be used in comparing @@ -162,7 +164,7 @@ check_and_setup_lustre DIR=${DIR:-$MOUNT} assert_DIR -MAXFREE=${MAXFREE:-$((200000 * $OSTCOUNT))} +MAXFREE=${MAXFREE:-$((300000 * $OSTCOUNT))} [ -f $DIR/d52a/foo ] && chattr -a $DIR/d52a/foo [ -f $DIR/d52b/foo ] && chattr -i $DIR/d52b/foo @@ -251,6 +253,35 @@ test_0d() { # LU-3397 } run_test 0d "check export proc =============================" +test_0e() { # LU-13417 + (( $MDSCOUNT > 1 )) || + skip "We need at least 2 MDTs for this test" + + (( $MDS1_VERSION >= $(version_code 2.14.51) )) || + skip "Need server version at least 2.14.51" + + local default_lmv_count=$($LFS getdirstripe -D -c $MOUNT) + local default_lmv_index=$($LFS getdirstripe -D -i $MOUNT) + + [ $default_lmv_count -eq 1 ] || + error "$MOUNT default stripe count $default_lmv_count" + + [ $default_lmv_index -eq -1 ] || + error "$MOUNT default stripe index $default_lmv_index" + + mkdir $MOUNT/$tdir.1 || error "mkdir $MOUNT/$tdir.1 failed" + mkdir $MOUNT/$tdir.2 || error "mkdir $MOUNT/$tdir.2 failed" + + local mdt_index1=$($LFS getdirstripe -i $MOUNT/$tdir.1) + local mdt_index2=$($LFS getdirstripe -i $MOUNT/$tdir.2) + + [ $mdt_index1 -eq $mdt_index2 ] && + error "directories are on the same MDT $mdt_index1=$mdt_index2" + + rmdir $MOUNT/$tdir.1 $MOUNT/$tdir.2 +} +run_test 0e "Enable DNE MDT balancing for mkdir in the ROOT" + test_1() { test_mkdir $DIR/$tdir test_mkdir $DIR/$tdir/d2 @@ -1835,7 +1866,7 @@ __exhaust_precreations() { local FAILIDX=${3:-$OSTIDX} local ofacet=ost$((OSTIDX + 1)) - test_mkdir -p -c1 $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir local mdtidx=$($LFS getstripe -m $DIR/$tdir) local mfacet=mds$((mdtidx + 1)) echo OSTIDX=$OSTIDX MDTIDX=$mdtidx @@ -1991,7 +2022,7 @@ test_27q() { reset_enospc rm -f $DIR/$tdir/$tfile - test_mkdir $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir $MCREATE $DIR/$tdir/$tfile || error "mcreate $DIR/$tdir/$tfile failed" $TRUNCATE $DIR/$tdir/$tfile 80000000 || error "truncate $DIR/$tdir/$tfile failed" @@ -2755,7 +2786,8 @@ test_27I() { save_layout_restore_at_exit $MOUNT $LFS setstripe -c 2 -i 0 $MOUNT pool_add $pool || error "pool_add failed" - pool_add_targets $pool $ostrange || "pool_add_targets failed" + pool_add_targets $pool $ostrange || + error "pool_add_targets failed" test_mkdir $DIR/$tdir $LFS setstripe -p $pool $DIR/$tdir $MULTIOP $DIR/$tdir/$tfile Oc || error "multiop failed" @@ -2984,11 +3016,23 @@ test_27M() { test_mkdir $DIR/$tdir # Set default striping on directory - $LFS setstripe -C 4 $DIR/$tdir + local setcount=4 + local stripe_opt + + # if we run against a 2.12 server which lacks overstring support + # then the connect_flag will not report overstriping, even if client + # is 2.14+ + if [[ $($LCTL get_param mdc.*.connect_flags) =~ overstriping ]]; then + stripe_opt="-C $setcount" + elif (( $OSTCOUNT >= $setcount )); then + stripe_opt="-c $setcount" + else + skip "server does not support overstriping" + fi + $LFS setstripe $stripe_opt $DIR/$tdir echo 1 > $DIR/$tdir/${tfile}.1 local count=$($LFS getstripe -c $DIR/$tdir/${tfile}.1) - local setcount=4 [ $count -eq $setcount ] || error "(1) stripe count $count, should be $setcount" @@ -3054,11 +3098,11 @@ test_27M() { # Clean up DOM layout $LFS setstripe -d $DIR/$tdir + save_layout_restore_at_exit $MOUNT # Now test that append striping works when layout is from root $LFS setstripe -c 2 $MOUNT # Make a special directory for this mkdir $DIR/${tdir}/${tdir}.2 - stack_trap "$LFS setstripe -d $MOUNT" EXIT # Verify for normal file setcount=2 @@ -3677,6 +3721,28 @@ test_31q() { } run_test 31q "create striped directory on specific MDTs" +#LU-14949 +test_31r() { + touch $DIR/$tfile.target + touch $DIR/$tfile.source + + #OBD_FAIL_LLITE_OPEN_DELAY 0x1419 + $LCTL set_param fail_loc=0x1419 fail_val=3 + cat $DIR/$tfile.target & + CATPID=$! + + # Guarantee open is waiting before we get here + sleep 1 + mv $DIR/$tfile.source $DIR/$tfile.target + + wait $CATPID + RC=$? + if [[ $RC -ne 0 ]]; then + error "open with cat failed, rc=$RC" + fi +} +run_test 31r "open-rename(replace) race" + cleanup_test32_mount() { local rc=0 trap 0 @@ -4766,7 +4832,7 @@ test_39l() { local atime_diff=$(do_facet $SINGLEMDS \ lctl get_param -n mdd.*MDT0000*.atime_diff) rm -rf $DIR/$tdir - mkdir -p $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir # test setting directory atime to future touch -a -d @$TEST_39_ATIME $DIR/$tdir @@ -5558,7 +5624,7 @@ test_51b() { # cleanup the directory rm -fr $dir - test_mkdir -c1 $dir + mkdir_on_mdt -i $((RANDOM % MDSCOUNT)) $dir $LFS df $LFS df -i @@ -5616,38 +5682,51 @@ run_test 51b "exceed 64k subdirectory nlink limit on create, verify unlink" test_51d() { [ $PARALLEL == "yes" ] && skip "skip parallel run" [[ $OSTCOUNT -lt 3 ]] && skip_env "needs >= 3 OSTs" + local qos_old test_mkdir $DIR/$tdir + $LFS setstripe -c $OSTCOUNT $DIR/$tdir + + qos_old=$(do_facet mds1 \ + "$LCTL get_param -n lod.$FSNAME-*.qos_threshold_rr" | head -n 1) + do_nodes $(comma_list $(mdts_nodes)) \ + "$LCTL set_param lod.$FSNAME-*.qos_threshold_rr=100" + stack_trap "do_nodes $(comma_list $(mdts_nodes)) \ + '$LCTL set_param lod.$FSNAME-*.qos_threshold_rr=${qos_old%%%}'" + createmany -o $DIR/$tdir/t- 1000 $LFS getstripe $DIR/$tdir > $TMP/$tfile - for N in $(seq 0 $((OSTCOUNT - 1))); do - OBJS[$N]=$(awk -vobjs=0 '($1 == '$N') { objs += 1 } \ - END { printf("%0.0f", objs) }' $TMP/$tfile) - OBJS0[$N]=$(grep -A 1 idx $TMP/$tfile | awk -vobjs=0 \ - '($1 == '$N') { objs += 1 } \ - END { printf("%0.0f", objs) }') - log "OST$N has ${OBJS[$N]} objects, ${OBJS0[$N]} are index 0" + for ((n = 0; n < $OSTCOUNT; n++)); do + objs[$n]=$(awk -vobjs=0 '($1 == '$n') { objs += 1 } \ + END { printf("%0.0f", objs) }' $TMP/$tfile) + objs0[$n]=$(grep -A 1 idx $TMP/$tfile | awk -vobjs=0 \ + '($1 == '$n') { objs += 1 } \ + END { printf("%0.0f", objs) }') + log "OST$n has ${objs[$n]} objects, ${objs0[$n]} are index 0" done unlinkmany $DIR/$tdir/t- 1000 - NLAST=0 - for N in $(seq 1 $((OSTCOUNT - 1))); do - [[ ${OBJS[$N]} -lt $((${OBJS[$NLAST]} - 20)) ]] && - error "OST $N has less objects vs OST $NLAST" \ - " (${OBJS[$N]} < ${OBJS[$NLAST]}" - [[ ${OBJS[$N]} -gt $((${OBJS[$NLAST]} + 20)) ]] && - error "OST $N has less objects vs OST $NLAST" \ - " (${OBJS[$N]} < ${OBJS[$NLAST]}" - - [[ ${OBJS0[$N]} -lt $((${OBJS0[$NLAST]} - 20)) ]] && - error "OST $N has less #0 objects vs OST $NLAST" \ - " (${OBJS0[$N]} < ${OBJS0[$NLAST]}" - [[ ${OBJS0[$N]} -gt $((${OBJS0[$NLAST]} + 20)) ]] && - error "OST $N has less #0 objects vs OST $NLAST" \ - " (${OBJS0[$N]} < ${OBJS0[$NLAST]}" - NLAST=$N + nlast=0 + for ((n = 0; n < $OSTCOUNT; n++)); do + (( ${objs[$n]} > ${objs[$nlast]} * 4 / 5 )) || + { $LFS df && $LFS df -i && + error "OST $n has fewer objects vs. OST $nlast" \ + " (${objs[$n]} < ${objs[$nlast]}"; } + (( ${objs[$n]} < ${objs[$nlast]} * 5 / 4 )) || + { $LFS df && $LFS df -i && + error "OST $n has fewer objects vs. OST $nlast" \ + " (${objs[$n]} < ${objs[$nlast]}"; } + + (( ${objs0[$n]} > ${objs0[$nlast]} * 4 / 5 )) || + { $LFS df && $LFS df -i && + error "OST $n has fewer #0 objects vs. OST $nlast" \ + " (${objs0[$n]} < ${objs0[$nlast]}"; } + (( ${objs0[$n]} < ${objs0[$nlast]} * 5 / 4 )) || + { $LFS df && $LFS df -i && + error "OST $n has fewer #0 objects vs. OST $nlast" \ + " (${objs0[$n]} < ${objs0[$nlast]}"; } + nlast=$n done - rm -f $TMP/$tfile } run_test 51d "check object distribution" @@ -6065,6 +6144,47 @@ test_56d() { } run_test 56d "'lfs df -v' prints only configured devices" +test_56e() { + err_enoent=2 # No such file or directory + err_eopnotsupp=95 # Operation not supported + + enoent_mnt=/pmt1 # Invalid dentry. Path not present + notsup_mnt=/tmp # Valid dentry, but Not a lustreFS + + # Check for handling of path not exists + output=$($LFS df $enoent_mnt 2>&1) + ret=$? + + fs=$(echo $output | awk -F: '{print $2}' | awk '{print $3}' | tr -d \') + [[ $fs = $enoent_mnt && $ret -eq $err_enoent ]] || + error "expect failure $err_enoent, not $ret" + + # Check for handling of non-Lustre FS + output=$($LFS df $notsup_mnt) + ret=$? + + fs=$(echo $output | awk '{print $1}' | awk -F: '{print $2}') + [[ $fs = $notsup_mnt && $ret -eq $err_eopnotsupp ]] || + error "expect success $err_eopnotsupp, not $ret" + + # Check for multiple LustreFS argument + output=$($LFS df $MOUNT $MOUNT $MOUNT | grep -c "filesystem_summary:") + ret=$? + + [[ $output -eq 3 && $ret -eq 0 ]] || + error "expect success 3, not $output, rc = $ret" + + # Check for correct non-Lustre FS handling among multiple + # LustreFS argument + output=$($LFS df $MOUNT $notsup_mnt $MOUNT | + grep -c "filesystem_summary:"; exit ${PIPESTATUS[0]}) + ret=$? + + [[ $output -eq 2 && $ret -eq $err_eopnotsupp ]] || + error "expect success 2, not $output, rc = $ret" +} +run_test 56e "'lfs df' Handle non LustreFS & multiple LustreFS" + NUMFILES=3 NUMDIRS=3 setup_56() { @@ -7578,6 +7698,85 @@ test_56xf() { } run_test 56xf "FID is not lost during migration of a composite layout file" +check_file_ost_range() { + local file="$1" + shift + local range="$*" + local -a file_range + local idx + + file_range=($($LFS getstripe -y "$file" | + awk '/l_ost_idx:/ { print $NF }')) + + if [[ "${#file_range[@]}" = 0 ]]; then + echo "No osts found for $file" + return 1 + fi + + for idx in "${file_range[@]}"; do + [[ " $range " =~ " $idx " ]] || + return 1 + done + + return 0 +} + +sub_test_56xg() { + local stripe_opt="$1" + local pool="$2" + shift 2 + local pool_ostidx="$(seq $* | tr '\n' ' ')" + + $LFS migrate $stripe_opt -p $pool $DIR/$tfile || + error "Fail to migrate $tfile on $pool" + [[ "$($LFS getstripe -p $DIR/$tfile)" = "$pool" ]] || + error "$tfile is not in pool $pool" + check_file_ost_range "$DIR/$tfile" $pool_ostidx || + error "$tfile osts mismatch with pool $pool (osts $pool_ostidx)" +} + +test_56xg() { + [[ $PARALLEL != "yes" ]] || skip "skip parallel run" + [[ $OSTCOUNT -ge 2 ]] || skip "needs >= 2 OSTs" + [[ $MDS1_VERSION -gt $(version_code 2.14.52) ]] || + skip "Need MDS version newer than 2.14.52" + + local -a pool_names=("${TESTNAME}_0" "${TESTNAME}_1" "${TESTNAME}_2") + local -a pool_ranges=("0 0" "1 1" "0 1") + + # init pools + for i in "${!pool_names[@]}"; do + pool_add ${pool_names[$i]} || + error "pool_add failed (pool: ${pool_names[$i]})" + pool_add_targets ${pool_names[$i]} ${pool_ranges[$i]} || + error "pool_add_targets failed (pool: ${pool_names[$i]})" + done + + # init the file to migrate + $LFS setstripe -c1 -i1 $DIR/$tfile || + error "Unable to create $tfile on OST1" + dd if=/dev/urandom of=$DIR/$tfile bs=1M count=4 status=none || + error "Unable to write on $tfile" + + echo "1. migrate $tfile on pool ${pool_names[0]}" + sub_test_56xg "-c-1" "${pool_names[0]}" ${pool_ranges[0]} + + echo "2. migrate $tfile on pool ${pool_names[2]}" + sub_test_56xg "-c-1 -S2M" "${pool_names[2]}" ${pool_ranges[2]} + + echo "3. migrate $tfile on pool ${pool_names[1]}" + sub_test_56xg "-n -c-1" "${pool_names[1]}" ${pool_ranges[1]} + + echo "4. migrate $tfile on pool ${pool_names[2]} with default stripe parameters" + sub_test_56xg "" "${pool_names[2]}" ${pool_ranges[2]} + echo + + # Clean pools + destroy_test_pools || + error "pool_destroy failed" +} +run_test 56xg "lfs migrate pool support" + test_56y() { [ $MDS1_VERSION -lt $(version_code 2.4.53) ] && skip "No HSM $(lustre_build_version $SINGLEMDS) MDS < 2.4.53" @@ -8630,6 +8829,104 @@ test_64f() { } run_test 64f "check grant consumption (with grant allocation)" +test_64g() { + #[ $MDS1_VERSION -lt $(version_code 2.14.54) ] && + # skip "Need MDS version at least 2.14.54" + + local mdts=$(comma_list $(mdts_nodes)) + + local old=$($LCTL get_param mdc.$FSNAME-*.grant_shrink_interval | + tr '\n' ' ') + stack_trap "$LCTL set_param $old" + + # generate dirty pages and increase dirty granted on MDT + stack_trap "rm -f $DIR/$tfile-*" + for (( i = 0; i < 10; i++)); do + $LFS setstripe -E 1M -L mdt $DIR/$tfile-$i || + error "can't set stripe" + dd if=/dev/zero of=$DIR/$tfile-$i bs=128k count=1 || + error "can't dd" + $LFS getstripe $DIR/$tfile-$i | grep -q pattern.*mdt || { + $LFS getstripe $DIR/$tfile-$i + error "not DoM file" + } + done + + # flush dirty pages + sync + + # wait until grant shrink reset grant dirty on MDTs + for ((i = 0; i < 120; i++)); do + grant_dirty=$(do_nodes $mdts $LCTL get_param -n mdt.*.tot_dirty | + awk '{sum=sum+$1} END {print sum}') + vm_dirty=$(awk '/Dirty:/{print $2}' /proc/meminfo) + echo "$grant_dirty grants, $vm_dirty pages" + (( grant_dirty + vm_dirty == 0 )) && break + (( i == 3 )) && sync && + $LCTL set_param mdc.$FSNAME-*.grant_shrink_interval=5 + sleep 1 + done + + grant_dirty=$(do_nodes $mdts $LCTL get_param -n mdt.*.tot_dirty | + awk '{sum=sum+$1} END {print sum}') + (( grant_dirty == 0 )) || error "$grant_dirty on MDT" +} +run_test 64g "grant shrink on MDT" + +test_64h() { + local instance=$($LFS getname -i $DIR) + local osc_tgt="$FSNAME-OST0000-osc-$instance" + local num_exps=$(do_facet ost1 \ + $LCTL get_param -n obdfilter.*OST0000*.num_exports) + local max_brw_size=$(import_param $osc_tgt max_brw_size) + local avail=$($LCTL get_param -n osc.*OST0000-osc-$instance.kbytesavail) + local p="$TMP/$TESTSUITE-$TESTNAME.parameters" + + # 10MiB is for file to be written, max_brw_size * 16 * + # num_exps is space reserve so that tgt_grant_shrink() decided + # to not shrink + local expect=$((max_brw_size * 16 * num_exps + 10 * 1048576)) + (( avail * 1024 < expect )) && + skip "need $expect bytes on ost1, have $(( avail * 1024 )) only" + + save_lustre_params client "osc.*OST0000*.grant_shrink" > $p + save_lustre_params client "osc.*OST0000*.grant_shrink_interval" >> $p + stack_trap "restore_lustre_params < $p; rm -f $save" EXIT + $LCTL set_param osc.*OST0000*.grant_shrink=1 + $LCTL set_param osc.*OST0000*.grant_shrink_interval=10 + + $LFS setstripe -c 1 -i 0 $DIR/$tfile + dd if=/dev/zero of=$DIR/$tfile bs=1M count=10 oflag=sync + + # drop cache so that coming read would do rpc + cancel_lru_locks osc + + # shrink interval is set to 10, pause for 7 seconds so that + # grant thread did not wake up yet but coming read entered + # shrink mode for rpc (osc_should_shrink_grant()) + sleep 7 + + declare -a cur_grant_bytes + declare -a tot_granted + cur_grant_bytes[0]=$($LCTL get_param -n osc.*OST0000*.cur_grant_bytes) + tot_granted[0]=$(do_facet ost1 \ + $LCTL get_param -n obdfilter.*OST0000*.tot_granted) + + dd if=$DIR/$tfile bs=4K count=1 of=/dev/null + + cur_grant_bytes[1]=$($LCTL get_param -n osc.*OST0000*.cur_grant_bytes) + tot_granted[1]=$(do_facet ost1 \ + $LCTL get_param -n obdfilter.*OST0000*.tot_granted) + + # grant change should be equal on both sides + (( cur_grant_bytes[0] - cur_grant_bytes[1] == + tot_granted[0] - tot_granted[1])) || + error "grant change mismatch, " \ + "server: ${tot_granted[0]} to ${tot_granted[1]}, " \ + "client: ${cur_grant_bytes[0]} to ${cur_grant_bytes[1]}" +} +run_test 64h "grant shrink on read" + # bug 1414 - set/get directories' stripe info test_65a() { [ $PARALLEL == "yes" ] && skip "skip parallel run" @@ -9448,6 +9745,7 @@ test_77d() { # bug 10889 [ $PARALLEL == "yes" ] && skip "skip parallel run" $GSS && skip_env "could not run with gss" + stack_trap "rm -f $DIR/$tfile" #define OBD_FAIL_OSC_CHECKSUM_SEND 0x409 $LCTL set_param fail_loc=0x80000409 set_checksums 1 @@ -9472,6 +9770,7 @@ test_77f() { # bug 10889 $GSS && skip_env "could not run with gss" set_checksums 1 + stack_trap "rm -f $DIR/$tfile" for algo in $CKSUM_TYPES; do cancel_lru_locks osc set_checksum_type $algo @@ -9593,6 +9892,95 @@ run_test 77l "preferred checksum type is remembered after reconnected" rm -f $F77_TMP unset F77_TMP +test_77m() { + (( $CLIENT_VERSION >= $(version_code 2.14.52) )) || + skip "Need at least version 2.14.52" + local param=checksum_speed + + $LCTL get_param $param || error "reading $param failed" + + csum_speeds=$($LCTL get_param -n $param) + + [[ "$csum_speeds" =~ "adler32" && "$csum_speeds" =~ "crc32" ]] || + error "known checksum types are missing" +} +run_test 77m "Verify checksum_speed is correctly read" + +check_filefrag_77n() { + local nr_ext=0 + local starts=() + local ends=() + + while read extidx a b start end rest; do + if [[ "${extidx}" =~ ^[0-9]+: ]]; then + nr_ext=$(( $nr_ext + 1 )) + starts+=( ${start%..} ) + ends+=( ${end%:} ) + fi + done < <( filefrag -sv $1 ) + + [[ $nr_ext -eq 2 ]] && [[ "${starts[-1]}" == $(( ${ends[0]} + 1 )) ]] && return 0 + return 1 +} + +test_77n() { + [[ "$CKSUM_TYPES" =~ t10 ]] || skip "no T10 checksum support on osc" + + touch $DIR/$tfile + $TRUNCATE $DIR/$tfile 0 + dd if=/dev/urandom of=$DIR/$tfile bs=4k conv=notrunc count=1 seek=0 + dd if=/dev/urandom of=$DIR/$tfile bs=4k conv=notrunc count=1 seek=2 + check_filefrag_77n $DIR/$tfile || + skip "$tfile blocks not contiguous around hole" + + set_checksums 1 + stack_trap "set_checksums $ORIG_CSUM" EXIT + stack_trap "set_checksum_type $ORIG_CSUM_TYPE" EXIT + stack_trap "rm -f $DIR/$tfile" + + for algo in $CKSUM_TYPES; do + if [[ "$algo" =~ ^t10 ]]; then + set_checksum_type $algo || + error "fail to set checksum type $algo" + dd if=$DIR/$tfile of=/dev/null bs=12k count=1 iflag=direct || + error "fail to read $tfile with $algo" + fi + done + rm -f $DIR/$tfile + return 0 +} +run_test 77n "Verify read from a hole inside contiguous blocks with T10PI" + +test_77o() { + (( $CLIENT_VERSION >= $(version_code 2.14.54) )) || + skip "Need at least version 2.14.54" + local ofd=obdfilter + local mdt=mdt + + # print OST checksum_type + echo "$ofd.$FSNAME-*.checksum_type:" + do_nodes $(comma_list $(osts_nodes)) \ + $LCTL get_param -n $ofd.$FSNAME-*.checksum_type + + # print MDT checksum_type + echo "$mdt.$FSNAME-*.checksum_type:" + do_nodes $(comma_list $(mdts_nodes)) \ + $LCTL get_param -n $mdt.$FSNAME-*.checksum_type + + local o_count=$(do_nodes $(comma_list $(osts_nodes)) \ + $LCTL get_param -n $ofd.$FSNAME-*.checksum_type | wc -l) + + (( $o_count == $OSTCOUNT )) || + error "found $o_count checksums, not \$MDSCOUNT=$OSTCOUNT" + + local m_count=$(do_nodes $(comma_list $(mdts_nodes)) \ + $LCTL get_param -n $mdt.$FSNAME-*.checksum_type | wc -l) + + (( $m_count == $MDSCOUNT )) || + error "found $m_count checksums, not \$MDSCOUNT=$MDSCOUNT" +} +run_test 77o "Verify checksum_type for server (mdt and ofd(obdfilter))" + cleanup_test_78() { trap 0 rm -f $DIR/$tfile @@ -13815,8 +14203,12 @@ test_133f() { $LCTL get_param -R '*' &> /dev/null # Verifing writability with badarea_io. + local proc_dirs=$(eval \ls -d $proc_regexp 2>/dev/null) + local skipped_params='force_lbug|changelog_mask|daemon_file' $LCTL list_param -FR '*' | grep '=' | tr -d = | - egrep -v 'force_lbug|changelog_mask' | xargs badarea_io || + egrep -v "$skipped_params" | + xargs -n 1 find $proc_dirs -name | + xargs -n 1 badarea_io || error "client badarea_io failed" # remount the FS in case writes/reads /proc break the FS @@ -13829,6 +14221,8 @@ test_133g() { remote_mds_nodsh && skip "remote MDS with nodsh" remote_ost_nodsh && skip "remote OST with nodsh" + local proc_dirs=$(eval \ls -d $proc_regexp 2>/dev/null) + local skipped_params="'force_lbug|changelog_mask|daemon_file'" local facet for facet in mds1 ost1; do local facet_ver=$(lustre_version_code $facet) @@ -13839,8 +14233,9 @@ test_133g() { fi if [ $facet_ver -ge $(version_code 2.5.54) ]; then do_facet $facet "$LCTL list_param -FR '*' | grep '=' | - tr -d = | egrep -v 'force_lbug|changelog_mask' | - xargs badarea_io" || + tr -d = | egrep -v $skipped_params | + xargs -n 1 find $proc_dirs -name | + xargs -n 1 badarea_io" || error "$facet badarea_io failed" else skip_noexit "$facet: too old lustre for get_param -R" @@ -13884,7 +14279,7 @@ test_134a() { [[ $MDS1_VERSION -lt $(version_code 2.7.54) ]] && skip "Need MDS version at least 2.7.54" - mkdir -p $DIR/$tdir || error "failed to create $DIR/$tdir" + mkdir_on_mdt0 $DIR/$tdir || error "failed to create $DIR/$tdir" cancel_lru_locks mdc local nsdir="ldlm.namespaces.*-MDT0000-mdc-*" @@ -13920,7 +14315,7 @@ test_134b() { [[ $MDS1_VERSION -lt $(version_code 2.7.54) ]] && skip "Need MDS version at least 2.7.54" - mkdir -p $DIR/$tdir || error "failed to create $DIR/$tdir" + mkdir_on_mdt0 $DIR/$tdir || error "failed to create $DIR/$tdir" cancel_lru_locks mdc local low_wm=$(do_facet mds1 $LCTL get_param -n \ @@ -14129,31 +14524,30 @@ run_test 150bb "Verify fallocate modes both zero space" test_150c() { check_set_fallocate_or_skip + local striping="-c2" stack_trap "rm -f $DIR/$tfile; wait_delete_completed" $LFS setstripe -c $OSTCOUNT -S1M $DIR/$tfile || error "setstripe failed" fallocate -l ${OSTCOUNT}m $DIR/$tfile || error "fallocate failed" - sync; sync_all_data - cancel_lru_locks $OSC - sleep 5 - bytes=$(($(stat -c '%b * %B' $DIR/$tfile))) - want=$((OSTCOUNT * 1048576)) + local bytes=$(($(stat -c '%b * %B' $DIR/$tfile))) + local want=$((OSTCOUNT * 1048576)) # Must allocate all requested space, not more than 5% extra (( $bytes >= $want && $bytes < $want * 105 / 100 )) || error "bytes $bytes is not $want" rm -f $DIR/$tfile - # verify fallocate on PFL file - $LFS setstripe -E1M -c1 -E16M -c3 -Eeof -c 4 $DIR/$tfile || + + echo "verify fallocate on PFL file" + + [[ "x$DOM" == "xyes" ]] && striping="-L mdt" + + $LFS setstripe -E1M $striping -E16M -c3 -Eeof -c 4 $DIR/$tfile || error "Create $DIR/$tfile failed" - fallocate -l $((1048576 * 1024)) $DIR/$tfile || + fallocate -l $((1048576 * 512)) $DIR/$tfile || error "fallocate failed" - sync; sync_all_data - cancel_lru_locks $OSC - sleep 5 - local bytes=$(($(stat -c '%b * %B' $DIR/$tfile))) - local want=$((1024 * 1048576)) + bytes=$(($(stat -c '%b * %B' $DIR/$tfile))) + want=$((512 * 1048576)) # Must allocate all requested space, not more than 5% extra (( $bytes >= $want && $bytes < $want * 105 / 100 )) || @@ -14163,13 +14557,14 @@ run_test 150c "Verify fallocate Size and Blocks" test_150d() { check_set_fallocate_or_skip + local striping="-c2" + + [[ "x$DOM" == "xyes" ]] && striping="-L mdt" stack_trap "rm -f $DIR/$tfile; wait_delete_completed" - $LFS setstripe -c $OSTCOUNT -S1M $DIR/$tdir || error "setstripe failed" + $LFS setstripe -E1M $striping -E eof -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 $OSC - sleep 5 local bytes=$(($(stat -c '%b * %B' $DIR/$tdir))) local want=$((OSTCOUNT * 1048576)) @@ -14246,6 +14641,9 @@ test_150f() { check_set_fallocate_or_skip stack_trap "rm -f $DIR/$tfile; wait_delete_completed" + [[ "x$DOM" == "xyes" ]] && + $LFS setstripe -E1M -L mdt -E eof $DIR/$tfile + echo "Verify fallocate punch: Range within the file range" yes 'A' | dd of=$DIR/$tfile bs=4096 count=5 || error "dd failed for bs 4096 and count 5" @@ -14321,8 +14719,13 @@ test_150g() { check_set_fallocate_or_skip stack_trap "rm -f $DIR/$tfile; wait_delete_completed" - $LFS setstripe -c${OSTCOUNT} $DIR/$tfile || - error "$LFS setstripe -c${OSTCOUNT} $DIR/$tfile failed" + if [[ "x$DOM" == "xyes" ]]; then + $LFS setstripe -E2M -L mdt -E eof -c${OSTCOUNT} $DIR/$tfile || + error "$LFS setstripe DoM + ${OSTCOUNT} OST failed" + else + $LFS setstripe -c${OSTCOUNT} $DIR/$tfile || + error "$LFS setstripe -c${OSTCOUNT} $DIR/$tfile failed" + fi # Get 100MB per OST of the available space to reduce run time # else 60% of the available space if we are running SLOW tests @@ -14848,35 +15251,35 @@ test_154f() { [ -n "$FILESET" ] && skip "SKIP due to FILESET set" # create parent directory on a single MDT to avoid cross-MDT hardlinks - test_mkdir -p -c1 $DIR/$tdir/d + mkdir_on_mdt0 $DIR/$tdir # test dirs inherit from its stripe - mkdir -p $DIR/$tdir/d/foo1 || error "mkdir error" - mkdir -p $DIR/$tdir/d/foo2 || error "mkdir error" - cp /etc/hosts $DIR/$tdir/d/foo1/$tfile - ln $DIR/$tdir/d/foo1/$tfile $DIR/$tdir/d/foo2/link + mkdir -p $DIR/$tdir/foo1 || error "mkdir error" + mkdir -p $DIR/$tdir/foo2 || error "mkdir error" + cp /etc/hosts $DIR/$tdir/foo1/$tfile + ln $DIR/$tdir/foo1/$tfile $DIR/$tdir/foo2/link touch $DIR/f # get fid of parents - local FID0=$($LFS path2fid $DIR/$tdir/d) - local FID1=$($LFS path2fid $DIR/$tdir/d/foo1) - local FID2=$($LFS path2fid $DIR/$tdir/d/foo2) + local FID0=$($LFS path2fid $DIR/$tdir) + local FID1=$($LFS path2fid $DIR/$tdir/foo1) + local FID2=$($LFS path2fid $DIR/$tdir/foo2) local FID3=$($LFS path2fid $DIR) # check that path2fid --parents returns expected /name # 1) test for a directory (single parent) - local parent=$($LFS path2fid --parents $DIR/$tdir/d/foo1) + local parent=$($LFS path2fid --parents $DIR/$tdir/foo1) [ "$parent" == "$FID0/foo1" ] || error "expected parent: $FID0/foo1, got: $parent" # 2) test for a file with nlink > 1 (multiple parents) - parent=$($LFS path2fid --parents $DIR/$tdir/d/foo1/$tfile) + parent=$($LFS path2fid --parents $DIR/$tdir/foo1/$tfile) echo "$parent" | grep -F "$FID1/$tfile" || error "$FID1/$tfile not returned in parent list" echo "$parent" | grep -F "$FID2/link" || error "$FID2/link not returned in parent list" # 3) get parent by fid - local file_fid=$($LFS path2fid $DIR/$tdir/d/foo1/$tfile) + local file_fid=$($LFS path2fid $DIR/$tdir/foo1/$tfile) parent=$($LFS path2fid --parents $MOUNT/.lustre/fid/$file_fid) echo "$parent" | grep -F "$FID1/$tfile" || error "$FID1/$tfile not returned in parent list (by fid)" @@ -14898,7 +15301,7 @@ test_154f() { lctl set_param llite.*.xattr_cache 1 # 6.1) linkea update on rename - mv $DIR/$tdir/d/foo1/$tfile $DIR/$tdir/d/foo2/$tfile.moved + mv $DIR/$tdir/foo1/$tfile $DIR/$tdir/foo2/$tfile.moved # get parents by fid parent=$($LFS path2fid --parents $MOUNT/.lustre/fid/$file_fid) @@ -14910,7 +15313,7 @@ test_154f() { error "$FID2/$tfile.moved is not in parent list" # 6.2) linkea update on unlink - rm -f $DIR/$tdir/d/foo2/link + rm -f $DIR/$tdir/foo2/link parent=$($LFS path2fid --parents $MOUNT/.lustre/fid/$file_fid) # foo2/link should no longer be returned in parent list echo "$parent" | grep -F "$FID2/link" && @@ -14930,7 +15333,7 @@ test_154g() $CLIENT_VERSION -gt $(version_code 2.6.99) ]] || skip "Need MDS version at least 2.6.92" - mkdir -p $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir llapi_fid_test -d $DIR/$tdir } run_test 154g "various llapi FID tests" @@ -15281,6 +15684,8 @@ test_160a() { changelog_users $SINGLEMDS | grep -q $cl_user || error "User $cl_user not found in changelog_users" + mkdir_on_mdt0 $DIR/$tdir + # change something test_mkdir -p $DIR/$tdir/pics/2008/zachy changelog_clear 0 || error "changelog_clear failed" @@ -15291,8 +15696,6 @@ test_160a() { ln -s $DIR/$tdir/pics/2008/portland.jpg $DIR/$tdir/pics/desktop.jpg rm $DIR/$tdir/pics/desktop.jpg - changelog_dump | tail -10 - echo "verifying changelog mask" changelog_chmask "-MKDIR" changelog_chmask "-CLOSE" @@ -15306,7 +15709,6 @@ test_160a() { test_mkdir -p $DIR/$tdir/pics/2008/sofia # mkdir 1 echo "zzzzzz" > $DIR/$tdir/pics/zach/file # open 3 - changelog_dump | tail -10 MKDIRS=$(changelog_dump | grep -c "MKDIR") CLOSES=$(changelog_dump | grep -c "CLOSE") [ $MKDIRS -eq 1 ] || error "MKDIR changelog mask count $MKDIRS != 1" @@ -15369,10 +15771,10 @@ test_160a() { error "User '$cl_user' still in changelog_users" # lctl get_param -n mdd.*.changelog_users - # current index: 144 + # current_index: 144 # ID index (idle seconds) - # cl3 144 (2) - if ! changelog_users $SINGLEMDS | grep "^cl"; then + # cl3 144 (2) mask= + if [ -z "$(changelog_users $SINGLEMDS | grep -v current.index)" ]; then # this is the normal case where all users were deregistered # make sure no new records are added when no users are present local last_rec1=$(changelog_users $SINGLEMDS | @@ -15468,11 +15870,21 @@ test_160e() { # Create a user changelog_register || error "changelog_register failed" - # Delete a future user (expect fail) local MDT0=$(facet_svc $SINGLEMDS) - do_facet $SINGLEMDS $LCTL --device $MDT0 changelog_deregister "cl77" - local rc=$? + local rc + + # No user (expect fail) + do_facet $SINGLEMDS $LCTL --device $MDT0 changelog_deregister + rc=$? + if [ $rc -eq 0 ]; then + error "Should fail without user" + elif [ $rc -ne 4 ]; then + error "changelog_deregister failed with $rc, expect 4(CMD_HELP)" + fi + # Delete a future user (expect fail) + do_facet $SINGLEMDS $LCTL --device $MDT0 changelog_deregister "cl77" + rc=$? if [ $rc -eq 0 ]; then error "Deleted non-existant user cl77" elif [ $rc -ne 2 ]; then @@ -16191,6 +16603,72 @@ test_160n() { } run_test 160n "Changelog destroy race" +test_160o() { + local mdt="$(facet_svc $SINGLEMDS)" + + [[ $PARALLEL != "yes" ]] || skip "skip parallel run" + remote_mds_nodsh && skip "remote MDS with nodsh" + [ $MDS1_VERSION -ge $(version_code 2.14.52) ] || + skip "Need MDS version at least 2.14.52" + + changelog_register --user test_160o -m unlnk+close+open || + error "changelog_register failed" + + do_facet $SINGLEMDS $LCTL --device $mdt \ + changelog_register -u "Tt3_-#" && + error "bad symbols in name should fail" + + do_facet $SINGLEMDS $LCTL --device $mdt \ + changelog_register -u test_160o && + error "the same name registration should fail" + + do_facet $SINGLEMDS $LCTL --device $mdt \ + changelog_register -u test_160toolongname && + error "too long name registration should fail" + + changelog_chmask "MARK+HSM" + lctl get_param mdd.*.changelog*mask + local cl_user="${CL_USERS[$SINGLEMDS]%% *}" + changelog_users $SINGLEMDS | grep -q $cl_user || + error "User $cl_user not found in changelog_users" + #verify username + echo $cl_user | grep -q test_160o || + error "User $cl_user has no specific name 'test160o'" + + # change something + changelog_clear 0 || error "changelog_clear failed" + # generate some changelog records to accumulate on MDT0 + test_mkdir -p -i0 -c1 $DIR/$tdir || error "test_mkdir $tdir failed" + touch $DIR/$tdir/$tfile # open 1 + + OPENS=$(changelog_dump | grep -c "OPEN") + [[ $OPENS -eq 1 ]] || error "OPEN changelog mask count $OPENS != 1" + + # must be no MKDIR it wasn't set as user mask + MKDIR=$(changelog_dump | grep -c "MKDIR") + [[ $MKDIR -eq 0 ]] || error "MKDIR changelog mask found $MKDIR > 0" + + oldmask=$(do_facet $SINGLEMDS $LCTL get_param \ + mdd.$mdt.changelog_current_mask -n) + # register maskless user + changelog_register || error "changelog_register failed" + # effective mask should be not changed because it is not minimal + mask=$(do_facet $SINGLEMDS $LCTL get_param \ + mdd.$mdt.changelog_current_mask -n) + [[ $mask == $oldmask ]] || error "mask was changed: $mask vs $oldmask" + # set server mask to minimal value + changelog_chmask "MARK" + # check effective mask again, should be treated as DEFMASK now + mask=$(do_facet $SINGLEMDS $LCTL get_param \ + mdd.$mdt.changelog_current_mask -n) + [[ $mask == *"HLINK"* ]] || error "mask is not DEFMASK as expected" + + do_facet $SINGLEMDS $LCTL --device $mdt \ + changelog_deregister -u test_160o || + error "cannot deregister by name" +} +run_test 160o "changelog user name and mask" + test_160p() { remote_mds_nodsh && skip "remote MDS with nodsh" && return [[ $MDS1_VERSION -ge $(version_code 2.14.51) ]] || @@ -16226,6 +16704,28 @@ test_160p() { } run_test 160p "Changelog orphan cleanup with no users" +test_160q() { + local mdt="$(facet_svc $SINGLEMDS)" + local clu + + [[ $PARALLEL != "yes" ]] || skip "skip parallel run" + remote_mds_nodsh && skip "remote MDS with nodsh" + [ $MDS1_VERSION -ge $(version_code 2.14.54) ] || + skip "Need MDS version at least 2.14.54" + + # set server mask to minimal value like server init does + changelog_chmask "MARK" + clu=$(do_facet $SINGLEMDS $LCTL --device $mdt changelog_register -n) || + error "changelog_register failed" + # check effective mask again, should be treated as DEFMASK now + mask=$(do_facet $SINGLEMDS $LCTL get_param \ + mdd.$mdt.changelog_current_mask -n) + do_facet $SINGLEMDS $LCTL --device $mdt changelog_deregister $clu || + error "changelog_deregister failed" + [[ $mask == *"HLINK"* ]] || error "mask is not DEFMASK as expected" +} +run_test 160q "changelog effective mask is DEFMASK if not set" + test_161a() { [ $PARALLEL == "yes" ] && skip "skip parallel run" @@ -17187,7 +17687,7 @@ test_183() { # LU-2275 [[ $MDS1_VERSION -lt $(version_code 2.3.56) ]] && skip "Need MDS version at least 2.3.56" - mkdir -p $DIR/$tdir || error "creating dir $DIR/$tdir" + mkdir_on_mdt0 $DIR/$tdir || error "creating dir $DIR/$tdir" echo aaa > $DIR/$tdir/$tfile #define OBD_FAIL_MDS_NEGATIVE_POSITIVE 0x148 @@ -17779,7 +18279,7 @@ test_205a() { # Job stats local cmd # mkdir - cmd="mkdir $DIR/$tdir" + cmd="$LFS mkdir -i 0 -c 1 $DIR/$tdir" verify_jobstats "$cmd" "$SINGLEMDS" # rmdir cmd="rmdir $DIR/$tdir" @@ -17820,7 +18320,7 @@ test_205a() { # Job stats [ $left -ge 0 ] && wait_update_facet $SINGLEMDS \ "lctl get_param *.*.job_stats | grep -c 'job_id.*mkdir'" \ "0" $left - cmd="mkdir $DIR/$tdir.expire" + cmd="$LFS mkdir -i 0 -c 1 $DIR/$tdir.expire" verify_jobstats "$cmd" "$SINGLEMDS" [ $(do_facet $SINGLEMDS lctl get_param *.*.job_stats | grep -c "job_id.*mkdir") -gt 1 ] && error "old jobstats not expired" @@ -17873,6 +18373,9 @@ run_test 205a "Verify job stats" # LU-13117, LU-13597 test_205b() { + (( $MDS1_VERSION >= $(version_code 2.13.54.91) )) || + skip "Need MDS version at least 2.13.54.91" + job_stats="mdt.*.job_stats" $LCTL set_param $job_stats=clear # Setting jobid_var to USER might not be supported @@ -17958,23 +18461,23 @@ test_208() { $MULTIOP $DIR/$tfile oO_CREAT:O_RDWR:eRE+eU || error "get lease error" echo "==== test 2: verify lease can be broken by upcoming open" - $MULTIOP $DIR/$tfile oO_RDONLY:eR_E-eUc & + $MULTIOP $DIR/$tfile oO_RDWR:eR_E-eUc & local PID=$! sleep 1 - $MULTIOP $DIR/$tfile oO_RDONLY:c + $MULTIOP $DIR/$tfile oO_RDWR:c kill -USR1 $PID && wait $PID || error "break lease error" echo "==== test 3: verify lease can't be granted if an open already exists" - $MULTIOP $DIR/$tfile oO_RDONLY:_c & + $MULTIOP $DIR/$tfile oO_RDWR:_c & local PID=$! sleep 1 - $MULTIOP $DIR/$tfile oO_RDONLY:eReUc && error "apply lease should fail" + $MULTIOP $DIR/$tfile oO_RDWR:eReUc && error "apply lease should fail" kill -USR1 $PID && wait $PID || error "open file error" echo "==== test 4: lease can sustain over recovery" - $MULTIOP $DIR/$tfile oO_RDONLY:eR_E+eUc & + $MULTIOP $DIR/$tfile oO_RDWR:eR_E+eUc & PID=$! sleep 1 @@ -17983,7 +18486,7 @@ test_208() { kill -USR1 $PID && wait $PID || error "lease broken over recovery" echo "==== test 5: lease broken can't be regained by replay" - $MULTIOP $DIR/$tfile oO_RDONLY:eR_E-eUc & + $MULTIOP $DIR/$tfile oO_RDWR:eR_E-eUc & PID=$! sleep 1 @@ -18311,7 +18814,7 @@ test_220() { #LU-325 local OSTIDX=0 # create on MDT0000 so the last_id and next_id are correct - mkdir $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir local OST=$($LFS df $DIR | awk '/OST:'$OSTIDX'/ { print $1 }') OST=${OST%_UUID} @@ -18429,26 +18932,48 @@ run_test 223 "osc reenqueue if without AGL lock granted =======================" test_224a() { # LU-1039, MRP-303 [ $PARALLEL == "yes" ] && skip "skip parallel run" - #define OBD_FAIL_PTLRPC_CLIENT_BULK_CB 0x508 $LCTL set_param fail_loc=0x508 - dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=fsync + dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=fsync $LCTL set_param fail_loc=0 df $DIR } run_test 224a "Don't panic on bulk IO failure" -test_224b() { # LU-1039, MRP-303 +test_224bd_sub() { # LU-1039, MRP-303 [ $PARALLEL == "yes" ] && skip "skip parallel run" + local timeout=$1 - dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 + shift + dd if=/dev/urandom of=$TMP/$tfile bs=1M count=1 + + $LFS setstripe -c 1 -i 0 $DIR/$tfile + + dd if=$TMP/$tfile of=$DIR/$tfile bs=1M count=1 cancel_lru_locks osc + set_checksums 0 + stack_trap "set_checksums $ORIG_CSUM" EXIT + local at_max_saved=0 + + # adaptive timeouts may prevent seeing the issue + if at_is_enabled; then + at_max_saved=$(at_max_get mds) + at_max_set 0 mds client + stack_trap "at_max_set $at_max_saved mds client" EXIT + fi + #define OBD_FAIL_PTLRPC_CLIENT_BULK_CB2 0x515 - $LCTL set_param fail_loc=0x515 - dd of=/dev/null if=$DIR/$tfile bs=4096 count=1 - $LCTL set_param fail_loc=0 + do_facet ost1 $LCTL set_param fail_val=$timeout fail_loc=0x80000515 + dd of=$TMP/$tfile.new if=$DIR/$tfile bs=1M count=1 || "$@" + + do_facet ost1 $LCTL set_param fail_loc=0 + cmp $TMP/$tfile $TMP/$tfile.new || error "file contents wrong" df $DIR } + +test_224b() { + test_224bd_sub 3 error "dd failed" +} run_test 224b "Don't panic on bulk IO failure" test_224c() { # LU-6441 @@ -18489,6 +19014,11 @@ test_224c() { # LU-6441 } run_test 224c "Don't hang if one of md lost during large bulk RPC" +test_224d() { # LU-11169 + test_224bd_sub $((TIMEOUT + 2)) error "dd failed" +} +run_test 224d "Don't corrupt data on bulk IO timeout" + MDSSURVEY=${MDSSURVEY:-$(which mds-survey 2>/dev/null || true)} test_225a () { [ $PARALLEL == "yes" ] && skip "skip parallel run" @@ -19168,7 +19698,7 @@ test_230d() { error "migrate remote dir error" echo "Finish migration, then checking.." - for file in $(find $migrate_dir); do + for file in $(find $migrate_dir -maxdepth 1); do mdt_index=$($LFS getstripe -m $file) if [ $mdt_index -lt $new_index ] || [ $mdt_index -gt $((new_index + new_count - 1)) ]; then @@ -19191,7 +19721,7 @@ test_230e() { local a_fid local b_fid - mkdir -p $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir mkdir $DIR/$tdir/migrate_dir mkdir $DIR/$tdir/other_dir touch $DIR/$tdir/migrate_dir/a @@ -19528,15 +20058,15 @@ test_230o() { run_test 230o "dir split" test_230p() { - [ $MDSCOUNT -ge 2 ] || skip "needs >= 2 MDTs" - [ $MDS1_VERSION -ge $(version_code 2.13.52) ] || + (( MDSCOUNT > 1 )) || skip "needs >= 2 MDTs" + (( MDS1_VERSION >= $(version_code 2.13.52) )) || skip "Need MDS version at least 2.13.52" local mdts=$(comma_list $(mdts_nodes)) local timeout=100 local restripe_status local delta - local i + local c [[ $mds1_FSTYPE == zfs ]] && timeout=300 @@ -19550,33 +20080,37 @@ test_230p() { test_mkdir -c $MDSCOUNT -H crush $DIR/$tdir createmany -m $DIR/$tdir/f 100 || - error "create files under remote dir failed $i" + error "create files under remote dir failed" createmany -d $DIR/$tdir/d 100 || - error "create dirs under remote dir failed $i" + error "create dirs under remote dir failed" - for i in $(seq $((MDSCOUNT - 1)) -1 1); do + for c 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" + $LFS setdirstripe -c $c $DIR/$tdir || + error "split -c $c $tdir failed" + if (( MDS1_VERSION >= $(version_code 2.14.51) )); then + mdt_hash="$mdt_hash,fixed" + elif [ $c -eq 1 ]; then + mdt_hash="none" + fi wait_update $HOSTNAME \ "$LFS getdirstripe -H $DIR/$tdir" $mdt_hash $timeout || error "dir merge not finished" delta=$(do_nodes $mdts "lctl get_param -n mdt.*MDT*.md_stats" | awk '/migrate/ {sum += $2} END { print sum }') - echo "$delta migrated when dir merge $((i + 1)) to $i stripes" + echo "$delta migrated when dir merge $((c + 1)) to $c stripes" # delta is around total_files/stripe_count - (( $delta < 200 / i + 4 )) || - error "$delta files migrated >= $((200 / i + 4))" + (( delta < 200 / c + 4 )) || + error "$delta files migrated >= $((200 / c + 4))" done } run_test 230p "dir merge" test_230q() { - [ $MDSCOUNT -ge 2 ] || skip "needs >= 2 MDTs" - [ $MDS1_VERSION -ge $(version_code 2.13.52) ] || + (( MDSCOUNT > 1)) || skip "needs >= 2 MDTs" + (( MDS1_VERSION >= $(version_code 2.13.52) )) || skip "Need MDS version at least 2.13.52" local mdts=$(comma_list $(mdts_nodes)) @@ -19636,6 +20170,15 @@ test_230q() { [ $nr_files -eq $total ] || error "total sub files $nr_files != $total" done + + (( MDS1_VERSION >= $(version_code 2.14.51) )) || return 0 + + echo "fixed layout directory won't auto split" + $LFS migrate -m 0 $DIR/$tdir || error "migrate $tdir failed" + wait_update $HOSTNAME "$LFS getdirstripe -H $DIR/$tdir" "crush,fixed" \ + 10 || error "stripe hash $($LFS getdirstripe -H $DIR/$tdir)" + wait_update $HOSTNAME "$LFS getdirstripe -c $DIR/$tdir" 1 10 || + error "stripe count $($LFS getdirstripe -c $DIR/$tdir)" } run_test 230q "dir auto split" @@ -19712,18 +20255,60 @@ test_230t() } run_test 230t "migrate directory with project ID set" -test_231a() +test_230u() { - # For simplicity this test assumes that max_pages_per_rpc - # is the same across all OSCs - local max_pages=$($LCTL get_param -n osc.*.max_pages_per_rpc | head -n1) - local bulk_size=$((max_pages * PAGE_SIZE)) - local brw_size=$(do_facet ost1 $LCTL get_param -n obdfilter.*.brw_size | - head -n 1) + (( MDSCOUNT > 3 )) || skip_env "needs >= 4 MDTs" + (( MDS1_VERSION >= $(version_code 2.14.53) )) || + skip "Need MDS version at least 2.14.53" - mkdir -p $DIR/$tdir - $LFS setstripe -S ${brw_size}M $DIR/$tdir || - error "failed to set stripe with -S ${brw_size}M option" + local count + + mkdir_on_mdt0 $DIR/$tdir || error "mkdir $tdir failed" + mkdir $DIR/$tdir/sub{0..99} || error "mkdir sub failed" + $LFS migrate -m -1 $DIR/$tdir/sub{0..99} || error "migrate sub failed" + for i in $(seq 0 $((MDSCOUNT - 1))); do + count=$($LFS getstripe -m $DIR/$tdir/sub* | grep -c ^$i) + echo "$count dirs migrated to MDT$i" + done + count=$($LFS getstripe -m $DIR/$tdir/sub* | sort -u | wc -l) + (( count >= MDSCOUNT - 1 )) || error "dirs migrated to $count MDTs" +} +run_test 230u "migrate directory by QOS" + +test_230v() +{ + (( MDSCOUNT > 3 )) || skip_env "needs >= 4 MDTs" + (( MDS1_VERSION >= $(version_code 2.14.53) )) || + skip "Need MDS version at least 2.14.53" + + local count + + mkdir $DIR/$tdir || error "mkdir $tdir failed" + mkdir $DIR/$tdir/sub{0..99} || error "mkdir sub failed" + $LFS migrate -m 0,2,1 $DIR/$tdir || error "migrate $tdir failed" + for i in $(seq 0 $((MDSCOUNT - 1))); do + count=$($LFS getstripe -m $DIR/$tdir/sub* | grep -c ^$i) + echo "$count subdirs migrated to MDT$i" + (( i == 3 )) && (( count > 0 )) && + error "subdir shouldn't be migrated to MDT3" + done + count=$($LFS getstripe -m $DIR/$tdir/sub* | sort -u | wc -l) + (( count == 3 )) || error "dirs migrated to $count MDTs" +} +run_test 230v "subdir migrated to the MDT where its parent is located" + +test_231a() +{ + # For simplicity this test assumes that max_pages_per_rpc + # is the same across all OSCs + local max_pages=$($LCTL get_param -n osc.*.max_pages_per_rpc | head -n1) + local bulk_size=$((max_pages * PAGE_SIZE)) + local brw_size=$(do_facet ost1 $LCTL get_param -n obdfilter.*.brw_size | + head -n 1) + + mkdir -p $DIR/$tdir + $LFS setstripe -S ${brw_size}M $DIR/$tdir || + error "failed to set stripe with -S ${brw_size}M option" # clear the OSC stats $LCTL set_param osc.*.stats=0 &>/dev/null @@ -20043,7 +20628,7 @@ run_test 241b "dio vs dio" test_242() { remote_mds_nodsh && skip "remote MDS with nodsh" - mkdir -p $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir touch $DIR/$tdir/$tfile #define OBD_FAIL_MDS_READPAGE_PACK 0x105 @@ -20235,10 +20820,11 @@ test_247f() { grep -q subtree || skip "Fileset feature is not supported" - mkdir $DIR/$tdir || error "mkdir $tdir failed" + mkdir_on_mdt0 $DIR/$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 -i $((MDSCOUNT - 1)) $DIR/$tdir/remote/subdir || + error "mkdir remote/subdir failed" $LFS mkdir -i 0 -c $MDSCOUNT $DIR/$tdir/striped || error "mkdir striped failed" mkdir $DIR/$tdir/striped/subdir || error "mkdir striped/subdir failed" @@ -20762,6 +21348,7 @@ test_255a() { skip "lustre < 2.8.54 does not support ladvise " remote_ost_nodsh && skip "remote OST with nodsh" + stack_trap "rm -f $DIR/$tfile" lfs setstripe -c -1 -i 0 $DIR/$tfile || error "$tfile failed" ladvise_no_type willread $DIR/$tfile && @@ -20849,6 +21436,7 @@ test_255b() { skip "lustre < 2.8.54 does not support ladvise " remote_ost_nodsh && skip "remote OST with nodsh" + stack_trap "rm -f $DIR/$tfile" lfs setstripe -c 1 -i 0 $DIR/$tfile ladvise_no_type dontneed $DIR/$tfile && @@ -21011,7 +21599,7 @@ test_256() { changelog_register || error "changelog_register failed" rm -rf $DIR/$tdir - mkdir -p $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir changelog_clear 0 || error "changelog_clear failed" @@ -21162,7 +21750,7 @@ test_270a() { local dom=$DIR/$tdir/dom_file local tmp=$DIR/$tdir/tmp_file - mkdir -p $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir # basic checks for DoM component creation $LFS setstripe -E 1024K -E 2048K -L mdt $dom 2>/dev/null && @@ -21563,6 +22151,16 @@ test_270h() { } run_test 270h "DoM: DoM stripe removal when disabled on server" +test_270i() { + (( $MDS1_VERSION >= $(version_code 2.14.54) )) || + skip "Need MDS version at least 2.14.54" + + mkdir $DIR/$tdir + $LFS setstripe -L mdt -S 128k -c -1 $DIR/$tdir && + error "setstripe should fail" || true +} +run_test 270i "DoM: setting invalid DoM striping should fail" + test_271a() { [ $MDS1_VERSION -lt $(version_code 2.10.55) ] && skip "Need MDS version at least 2.10.55" @@ -22301,7 +22899,7 @@ test_300c() { local file_count - mkdir -p $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir $LFS setdirstripe -i 0 -c 2 $DIR/$tdir/striped_dir || error "set striped dir error" @@ -22511,7 +23109,7 @@ test_300g() { local stripe_count local stripe_index - mkdir $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir mkdir $DIR/$tdir/normal_dir #Checking when client cache stripe index @@ -22547,9 +23145,6 @@ test_300g() { stripe_count=$($LFS getdirstripe -c $dir) [ $stripe_count -eq 0 ] || error "expect 1 get $stripe_count for $dir" - stripe_index=$($LFS getdirstripe -i $dir) - [ $stripe_index -eq 0 ] || - error "expect 0 get $stripe_index for $dir" done } run_test 300g "check default striped directory for normal directory" @@ -22851,7 +23446,7 @@ test_300p() { [ $MDSCOUNT -lt 2 ] && skip_env "needs >= 2 MDTs" remote_mds_nodsh && skip "remote MDS with nodsh" - mkdir -p $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir #define OBD_FAIL_OUT_ENOSPC 0x1704 do_facet mds2 lctl set_param fail_loc=0x80001704 @@ -23556,6 +24151,275 @@ test_398f() { # LU-14687 } run_test 398f "verify aio handles ll_direct_rw_pages errors correctly" +# NB: To get the parallel DIO behavior in LU-13798, there must be > 1 +# stripe and i/o size must be > stripe size +# Old style synchronous DIO waits after submitting each chunk, resulting in a +# single RPC in flight. This test shows async DIO submission is working by +# showing multiple RPCs in flight. +test_398g() { # LU-13798 + $LFS setstripe -o 0,0 -S 1M $DIR/$tfile + + # We need to do some i/o first to acquire enough grant to put our RPCs + # in flight; otherwise a new connection may not have enough grant + # available + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=1 oflag=direct || + error "parallel dio failed" + stack_trap "rm -f $DIR/$tfile" + + # Reduce RPC size to 1M to avoid combination in to larger RPCs + local pages_per_rpc=$($LCTL get_param osc.*-OST0000-*.max_pages_per_rpc) + $LCTL set_param osc.*-OST0000-*.max_pages_per_rpc=1M + stack_trap "$LCTL set_param -n $pages_per_rpc" + + # Recreate file so it's empty + rm -f $DIR/$tfile + $LFS setstripe -o 0,0 -S 1M $DIR/$tfile + #Pause rpc completion to guarantee we see multiple rpcs in flight + #define OBD_FAIL_OST_BRW_PAUSE_BULK + do_facet ost1 $LCTL set_param fail_loc=0x214 fail_val=2 + stack_trap "do_facet ost1 $LCTL set_param fail_loc=0" + + # Clear rpc stats + $LCTL set_param osc.*.rpc_stats=c + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=1 oflag=direct || + error "parallel dio failed" + stack_trap "rm -f $DIR/$tfile" + + $LCTL get_param osc.*-OST0000-*.rpc_stats + pct=$($LCTL get_param osc.*-OST0000-*.rpc_stats | + grep -A 8 'rpcs in flight' | grep -v 'rpcs in flight' | + grep "8:" | awk '{print $8}') + # We look at the "8 rpcs in flight" field, and verify A) it is present + # and B) it includes all RPCs. This proves we had 8 RPCs in flight, + # as expected for an 8M DIO to a file with 1M stripes. + [ $pct -eq 100 ] || error "we should see 8 RPCs in flight" + + # Verify turning off parallel dio works as expected + # Clear rpc stats + $LCTL set_param osc.*.rpc_stats=c + $LCTL set_param llite.*.parallel_dio=0 + stack_trap '$LCTL set_param llite.*.parallel_dio=1' + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=1 oflag=direct || + error "dio with parallel dio disabled failed" + + # Ideally, we would see only one RPC in flight here, but there is an + # unavoidable race between i/o completion and RPC in flight counting, + # so while only 1 i/o is in flight at a time, the RPC in flight counter + # will sometimes exceed 1 (3 or 4 is not rare on VM testing). + # So instead we just verify it's always < 8. + $LCTL get_param osc.*-OST0000-*.rpc_stats + ret=$($LCTL get_param osc.*-OST0000-*.rpc_stats | + grep -A 8 'rpcs in flight' | grep -v 'rpcs in flight' | + grep '^$' -B1 | grep . | awk '{print $1}') + [ $ret != "8:" ] || + error "we should see fewer than 8 RPCs in flight (saw $ret)" +} +run_test 398g "verify parallel dio async RPC submission" + +test_398h() { # LU-13798 + local dio_file=$DIR/$tfile.dio + + $LFS setstripe -C 2 -S 1M $DIR/$tfile $dio_file + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct + stack_trap "rm -f $DIR/$tfile $dio_file" + + dd if=$DIR/$tfile of=$dio_file bs=8M count=8 iflag=direct oflag=direct || + error "parallel dio failed" + diff $DIR/$tfile $dio_file + [[ $? == 0 ]] || error "file diff after aiocp" +} +run_test 398h "verify correctness of read & write with i/o size >> stripe size" + +test_398i() { # LU-13798 + local dio_file=$DIR/$tfile.dio + + $LFS setstripe -C 2 -S 1M $DIR/$tfile $dio_file + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct + stack_trap "rm -f $DIR/$tfile $dio_file" + + #define OBD_FAIL_LLITE_PAGE_ALLOC 0x1418 + $LCTL set_param fail_loc=0x1418 + # make sure we don't crash and fail properly + dd if=$DIR/$tfile of=$dio_file bs=8M count=8 iflag=direct oflag=direct && + error "parallel dio page allocation failure succeeded" + diff $DIR/$tfile $dio_file + [[ $? != 0 ]] || error "no diff after failed aiocp" +} +run_test 398i "verify parallel dio handles ll_direct_rw_pages errors correctly" + +test_398j() { # LU-13798 + # Stripe size > RPC size but less than i/o size tests split across + # stripes and RPCs for individual i/o op + $LFS setstripe -o 0,0 -S 4M $DIR/$tfile $DIR/$tfile.2 + + # Reduce RPC size to 1M to guarantee split to multiple RPCs per stripe + local pages_per_rpc=$($LCTL get_param osc.*-OST0000-*.max_pages_per_rpc) + $LCTL set_param osc.*-OST0000-*.max_pages_per_rpc=1M + stack_trap "$LCTL set_param -n $pages_per_rpc" + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct || + error "parallel dio write failed" + stack_trap "rm -f $DIR/$tfile $DIR/$tfile.2" + + dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 iflag=direct || + error "parallel dio read failed" + diff $DIR/$tfile $DIR/$tfile.2 + [[ $? == 0 ]] || error "file diff after parallel dio read" +} +run_test 398j "test parallel dio where stripe size > rpc_size" + +test_398k() { # LU-13798 + wait_delete_completed + wait_mds_ost_sync + + # 4 stripe file; we will cause out of space on OST0 + $LFS setstripe -o 0,1,0,1 -S 1M $DIR/$tfile + + # Fill OST0 (if it's not too large) + ORIGFREE=$($LCTL get_param -n lov.$FSNAME-clilov-*.kbytesavail | + head -n1) + if [[ $ORIGFREE -gt $MAXFREE ]]; then + skip "$ORIGFREE > $MAXFREE skipping out-of-space test on OST0" + fi + $LFS setstripe -i 0 -c 1 $DIR/$tfile.1 + dd if=/dev/zero of=$DIR/$tfile.1 bs=1024 count=$MAXFREE && + error "dd should fill OST0" + stack_trap "rm -f $DIR/$tfile.1" + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct + err=$? + + ls -la $DIR/$tfile + $CHECKSTAT -t file -s 0 $DIR/$tfile || + error "file is not 0 bytes in size" + + # dd above should not succeed, but don't error until here so we can + # get debug info above + [[ $err != 0 ]] || + error "parallel dio write with enospc succeeded" + stack_trap "rm -f $DIR/$tfile" +} +run_test 398k "test enospc on first stripe" + +test_398l() { # LU-13798 + wait_delete_completed + wait_mds_ost_sync + + # 4 stripe file; we will cause out of space on OST0 + # Note the 1M stripe size and the > 1M i/o size mean this ENOSPC + # happens on the second i/o chunk we issue + $LFS setstripe -o 1,0,1,0 -S 1M $DIR/$tfile $DIR/$tfile.2 + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=2 oflag=direct + stack_trap "rm -f $DIR/$tfile" + + # Fill OST0 (if it's not too large) + ORIGFREE=$($LCTL get_param -n lov.$FSNAME-clilov-*.kbytesavail | + head -n1) + if [[ $ORIGFREE -gt $MAXFREE ]]; then + skip "$ORIGFREE > $MAXFREE skipping out-of-space test on OST0" + fi + $LFS setstripe -i 0 -c 1 $DIR/$tfile.1 + dd if=/dev/zero of=$DIR/$tfile.1 bs=1024 count=$MAXFREE && + error "dd should fill OST0" + stack_trap "rm -f $DIR/$tfile.1" + + dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 oflag=direct + err=$? + stack_trap "rm -f $DIR/$tfile.2" + + # Check that short write completed as expected + ls -la $DIR/$tfile.2 + $CHECKSTAT -t file -s 1048576 $DIR/$tfile.2 || + error "file is not 1M in size" + + # dd above should not succeed, but don't error until here so we can + # get debug info above + [[ $err != 0 ]] || + error "parallel dio write with enospc succeeded" + + # Truncate source file to same length as output file and diff them + $TRUNCATE $DIR/$tfile 1048576 + diff $DIR/$tfile $DIR/$tfile.2 + [[ $? == 0 ]] || error "data incorrect after short write" +} +run_test 398l "test enospc on intermediate stripe/RPC" + +test_398m() { # LU-13798 + $LFS setstripe -o 0,1,0,1 -S 1M $DIR/$tfile + + # Set up failure on OST0, the first stripe: + #define OBD_FAIL_OST_BRW_WRITE_BULK 0x20e + #NB: Fail val is ost # + 1, because we cannot use cfs_fail_val = 0 + # So this fail_val specifies OST0 + do_facet ost1 $LCTL set_param fail_loc=0x20e fail_val=1 + stack_trap "do_facet ost1 $LCTL set_param fail_loc=0" + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct && + error "parallel dio write with failure on first stripe succeeded" + stack_trap "rm -f $DIR/$tfile" + do_facet ost1 $LCTL set_param fail_loc=0 fail_val=0 + + # Place data in file for read + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct || + error "parallel dio write failed" + + # Fail read on OST0, first stripe + #define OBD_FAIL_OST_BRW_READ_BULK 0x20f + do_facet ost1 $LCTL set_param fail_loc=0x20f fail_val=1 + dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 iflag=direct && + error "parallel dio read with error on first stripe succeeded" + rm -f $DIR/$tfile.2 + do_facet ost1 $LCTL set_param fail_loc=0 fail_val=0 + + # Switch to testing on OST1, second stripe + # Clear file contents, maintain striping + echo > $DIR/$tfile + # Set up failure on OST1, second stripe: + do_facet ost1 $LCTL set_param fail_loc=0x20e fail_val=2 + stack_trap "do_facet ost1 $LCTL set_param fail_loc=0" + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct && + error "parallel dio write with failure on first stripe succeeded" + stack_trap "rm -f $DIR/$tfile" + do_facet ost1 $LCTL set_param fail_loc=0 fail_val=0 + + # Place data in file for read + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct || + error "parallel dio write failed" + + # Fail read on OST1, second stripe + #define OBD_FAIL_OST_BRW_READ_BULK 0x20f + do_facet ost2 $LCTL set_param fail_loc=0x20f fail_val=2 + dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 iflag=direct && + error "parallel dio read with error on first stripe succeeded" + rm -f $DIR/$tfile.2 + do_facet ost2 $LCTL set_param fail_loc=0 fail_val=0 +} +run_test 398m "test RPC failures with parallel dio" + +# Parallel submission of DIO should not cause problems for append, but it's +# important to verify. +test_398n() { # LU-13798 + $LFS setstripe -C 2 -S 1M $DIR/$tfile + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 || + error "dd to create source file failed" + stack_trap "rm -f $DIR/$tfile" + + dd if=$DIR/$tfile of=$DIR/$tfile.1 bs=8M count=8 oflag=direct oflag=append || + error "parallel dio write with failure on second stripe succeeded" + stack_trap "rm -f $DIR/$tfile $DIR/$tfile.1" + diff $DIR/$tfile $DIR/$tfile.1 + [[ $? == 0 ]] || error "data incorrect after append" + +} +run_test 398n "test append with parallel DIO" + test_fake_rw() { local read_write=$1 if [ "$read_write" = "write" ]; then @@ -23698,7 +24562,6 @@ 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" @@ -23809,6 +24672,15 @@ test_401d() { } run_test 401d "Verify 'lctl set_param' accepts values containing '='" +test_401e() { # LU-14779 + $LCTL list_param -R "ldlm.namespaces.MGC*" || + error "lctl list_param MGC* failed" + $LCTL get_param "ldlm.namespaces.MGC*" || error "lctl get_param failed" + $LCTL get_param "ldlm.namespaces.MGC*.lru_size" || + error "lctl get_param lru_size failed" +} +run_test 401e "verify 'lctl get_param' works with NID in parameter" + test_402() { [[ $MDS1_VERSION -ge $(version_code 2.7.66) ]] || [[ $MDS1_VERSION -ge $(version_code 2.7.18.4) && @@ -24115,6 +24987,99 @@ test_412() { } run_test 412 "mkdir on specific MDTs" +generate_uneven_mdts() { + local threshold=$1 + local lmv_qos_maxage + local lod_qos_maxage + local ffree + local bavail + local max + local min + local max_index + local min_index + local tmp + local i + + lmv_qos_maxage=$($LCTL get_param -n lmv.*.qos_maxage) + $LCTL set_param lmv.*.qos_maxage=1 + stack_trap "$LCTL set_param \ + lmv.*.qos_maxage=$lmv_qos_maxage > /dev/null" RETURN + lod_qos_maxage=$(do_facet mds1 $LCTL get_param -n \ + lod.$FSNAME-MDT0000-mdtlov.qos_maxage | awk '{ print $1 }') + do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param \ + lod.*.mdt_qos_maxage=1 + stack_trap "do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param \ + lod.*.mdt_qos_maxage=$lod_qos_maxage > /dev/null" RETURN + + echo + echo "Check for uneven MDTs: " + + ffree=($(lctl get_param -n mdc.*[mM][dD][cC]-[^M]*.filesfree)) + bavail=($(lctl get_param -n mdc.*[mM][dD][cC]-[^M]*.kbytesavail)) + bsize=$(lctl get_param -n mdc.*MDT0000*.blocksize) + + max=$(((${ffree[0]} >> 8) * (${bavail[0]} * bsize >> 16))) + min=$(((${ffree[0]} >> 8) * (${bavail[0]} * bsize >> 16))) + max_index=0 + min_index=0 + for ((i = 1; i < ${#ffree[@]}; i++)); do + tmp=$(((${ffree[i]} >> 8) * (${bavail[i]} * bsize >> 16))) + if [ $tmp -gt $max ]; then + max=$tmp + max_index=$i + fi + if [ $tmp -lt $min ]; then + min=$tmp + min_index=$i + fi + done + + (( ${ffree[min_index]} > 0 )) || + skip "no free files in MDT$min_index" + (( ${ffree[min_index]} < 10000000 )) || + skip "too many free files in MDT$min_index" + + # Check if we need to generate uneven MDTs + local diff=$(((max - min) * 100 / min)) + local testdir=$DIR/$tdir-fillmdt + local start + + mkdir -p $testdir + + i=0 + while (( diff < threshold )); do + # generate uneven MDTs, create till $threshold% diff + echo -n "weight diff=$diff% must be > $threshold% ..." + echo "Fill MDT$min_index with 1000 files: loop $i" + testdir=$DIR/$tdir-fillmdt/$i + [ -d $testdir ] || $LFS mkdir -i $min_index $testdir || + error "mkdir $testdir failed" + $LFS setstripe -E 1M -L mdt $testdir || + error "setstripe $testdir failed" + start=$SECONDS + for F in f.{0..999}; do + dd if=/dev/zero of=$testdir/$F bs=64K count=1 > \ + /dev/null 2>&1 || error "dd $F failed" + done + + # wait for QOS to update + (( SECONDS < start + 1 )) && sleep $((start + 1 - SECONDS)) + + ffree=($(lctl get_param -n mdc.*[mM][dD][cC]-*.filesfree)) + bavail=($(lctl get_param -n mdc.*[mM][dD][cC]-*.kbytesavail)) + max=$(((${ffree[max_index]} >> 8) * \ + (${bavail[max_index]} * bsize >> 16))) + min=$(((${ffree[min_index]} >> 8) * \ + (${bavail[min_index]} * bsize >> 16))) + diff=$(((max - min) * 100 / min)) + i=$((i + 1)) + done + + echo "MDT filesfree available: ${ffree[@]}" + echo "MDT blocks available: ${bavail[@]}" + echo "weight diff=$diff%" +} + test_qos_mkdir() { local mkdir_cmd=$1 local stripe_count=$2 @@ -24137,11 +25102,11 @@ test_qos_mkdir() { lmv_qos_threshold_rr=${lmv_qos_threshold_rr%%%} lmv_qos_maxage=$($LCTL get_param -n lmv.*.qos_maxage) stack_trap "$LCTL set_param \ - lmv.*.qos_prio_free=$lmv_qos_prio_free > /dev/null" EXIT + lmv.*.qos_prio_free=$lmv_qos_prio_free > /dev/null" stack_trap "$LCTL set_param \ - lmv.*.qos_threshold_rr=$lmv_qos_threshold_rr > /dev/null" EXIT + lmv.*.qos_threshold_rr=$lmv_qos_threshold_rr > /dev/null" stack_trap "$LCTL set_param \ - lmv.*.qos_maxage=$lmv_qos_maxage > /dev/null" EXIT + lmv.*.qos_maxage=$lmv_qos_maxage > /dev/null" lod_qos_prio_free=$(do_facet mds1 $LCTL get_param -n \ lod.$FSNAME-MDT0000-mdtlov.mdt_qos_prio_free | head -n1) @@ -24152,12 +25117,11 @@ test_qos_mkdir() { lod_qos_maxage=$(do_facet mds1 $LCTL get_param -n \ lod.$FSNAME-MDT0000-mdtlov.qos_maxage | awk '{ print $1 }') stack_trap "do_nodes $mdts $LCTL set_param \ - lod.*.mdt_qos_prio_free=$lod_qos_prio_free > /dev/null" EXIT + lod.*.mdt_qos_prio_free=$lod_qos_prio_free > /dev/null" stack_trap "do_nodes $mdts $LCTL set_param \ - lod.*.mdt_qos_threshold_rr=$lod_qos_threshold_rr > /dev/null" \ - EXIT + lod.*.mdt_qos_threshold_rr=$lod_qos_threshold_rr > /dev/null" stack_trap "do_nodes $mdts $LCTL set_param \ - lod.*.mdt_qos_maxage=$lod_qos_maxage > /dev/null" EXIT + lod.*.mdt_qos_maxage=$lod_qos_maxage > /dev/null" $LCTL set_param lmv.*.qos_threshold_rr=100 > /dev/null do_nodes $mdts $LCTL set_param lod.*.mdt_qos_threshold_rr=100 > /dev/null @@ -24167,11 +25131,11 @@ test_qos_mkdir() { local stripe_index=$($LFS getstripe -m $testdir) local test_mkdir_rr=true - getfattr -d -m dmv $testdir | grep dmv - if [ $? -eq 0 ] && [ $MDS1_VERSION -ge $(version_code 2.14.51) ]; then - local inherit_rr=$($LFS getdirstripe -D --max-inherit-rr $testdir) - - (( $inherit_rr == 0 )) && test_mkdir_rr=false + getfattr -d -m dmv -e hex $testdir | grep dmv + if (( $? == 0 && $MDS1_VERSION >= $(version_code 2.14.51) )); then + echo "defstripe: '$($LFS getdirstripe -D $testdir)'" + (( $($LFS getdirstripe -D --max-inherit-rr $testdir) == 0 )) && + test_mkdir_rr=false fi echo @@ -24179,41 +25143,37 @@ test_qos_mkdir() { echo "Mkdir (stripe_count $stripe_count) roundrobin:" || echo "Mkdir (stripe_count $stripe_count) on stripe $stripe_index" - for i in $(seq $((100 * MDSCOUNT))); do + stack_trap "unlinkmany -d $testdir/subdir $((100 * MDSCOUNT))" + for (( i = 0; i < 100 * MDSCOUNT; i++ )); do eval $mkdir_cmd $testdir/subdir$i || error "$mkdir_cmd subdir$i failed" done - for i in $(seq $MDSCOUNT); do - count=$($LFS getdirstripe -i $testdir/* | - grep ^$((i - 1))$ | wc -l) - echo "$count directories created on MDT$((i - 1))" + for (( i = 0; i < $MDSCOUNT; i++ )); do + count=$($LFS getdirstripe -i $testdir/* | grep -c "^$i$") + echo "$count directories created on MDT$i" if $test_mkdir_rr; then (( $count == 100 )) || error "subdirs are not evenly distributed" - elif [ $((i - 1)) -eq $stripe_index ]; then + elif (( $i == $stripe_index )); then (( $count == 100 * MDSCOUNT )) || - error "$count subdirs created on MDT$((i - 1))" + error "$count subdirs created on MDT$i" else (( $count == 0 )) || - error "$count subdirs created on MDT$((i - 1))" + error "$count subdirs created on MDT$i" fi if $test_mkdir_rr && [ $stripe_count -gt 1 ]; then count=$($LFS getdirstripe $testdir/* | - grep -P "^\s+$((i - 1))\t" | wc -l) - echo "$count stripes created on MDT$((i - 1))" + grep -c -P "^\s+$i\t") + echo "$count stripes created on MDT$i" # deviation should < 5% of average - (( $count < 95 * stripe_count )) || - (( $count > 105 * stripe_count)) && + (( $count >= 95 * stripe_count && + $count <= 105 * stripe_count)) || error "stripes are not evenly distributed" fi done - $LCTL set_param lmv.*.qos_threshold_rr=$lmv_qos_threshold_rr > /dev/null - do_nodes $mdts $LCTL set_param \ - lod.*.mdt_qos_threshold_rr=$lod_qos_threshold_rr > /dev/null - echo echo "Check for uneven MDTs: " @@ -24245,47 +25205,20 @@ test_qos_mkdir() { fi done - (( ${ffree[min_index]} == 0 )) && + (( ${ffree[min_index]} > 0 )) || skip "no free files in MDT$min_index" - (( ${ffree[min_index]} > 100000000 )) && + (( ${ffree[min_index]} < 10000000 )) || skip "too many free files in MDT$min_index" - # Check if we need to generate uneven MDTs - local threshold=50 - local diff=$(((max - min) * 100 / min)) - local value="$(generate_string 1024)" - - while [ $diff -lt $threshold ]; do - # generate uneven MDTs, create till $threshold% diff - echo -n "weight diff=$diff% must be > $threshold% ..." - count=$((${ffree[min_index]} / 10)) - # 50 sec per 10000 files in vm - (( $count < 100000 )) || [ "$SLOW" != "no" ] || - skip "$count files to create" - echo "Fill MDT$min_index with $count files" - [ -d $DIR/$tdir-MDT$min_index ] || - $LFS mkdir -i $min_index $DIR/$tdir-MDT$min_index || - error "mkdir $tdir-MDT$min_index failed" - createmany -d $DIR/$tdir-MDT$min_index/d $count || - error "create d$count failed" - - ffree=($(lctl get_param -n mdc.*[mM][dD][cC]-*.filesfree)) - bavail=($(lctl get_param -n mdc.*[mM][dD][cC]-*.kbytesavail)) - max=$(((${ffree[max_index]} >> 8) * \ - (${bavail[max_index]} * bsize >> 16))) - min=$(((${ffree[min_index]} >> 8) * \ - (${bavail[min_index]} * bsize >> 16))) - diff=$(((max - min) * 100 / min)) - done - echo "MDT filesfree available: ${ffree[@]}" echo "MDT blocks available: ${bavail[@]}" - echo "weight diff=$diff%" - + echo "weight diff=$(((max - min) * 100 / min))%" echo echo "Mkdir (stripe_count $stripe_count) with balanced space usage:" + $LCTL set_param lmv.*.qos_threshold_rr=0 > /dev/null $LCTL set_param lmv.*.qos_prio_free=100 > /dev/null + do_nodes $mdts $LCTL set_param lod.*.mdt_qos_threshold_rr=0 > /dev/null do_nodes $mdts $LCTL set_param lod.*.mdt_qos_prio_free=100 > /dev/null # decrease statfs age, so that it can be updated in time $LCTL set_param lmv.*.qos_maxage=1 > /dev/null @@ -24294,42 +25227,66 @@ test_qos_mkdir() { sleep 1 testdir=$DIR/$tdir-s$stripe_count/qos + local num=200 - for i in $(seq $((100 * MDSCOUNT))); do + stack_trap "unlinkmany -d $testdir/subdir $((num * MDSCOUNT))" + for (( i = 0; i < num * MDSCOUNT; i++ )); do eval $mkdir_cmd $testdir/subdir$i || error "$mkdir_cmd subdir$i failed" done - for i in $(seq $MDSCOUNT); do - count=$($LFS getdirstripe -i $testdir/* | grep ^$((i - 1))$ | - wc -l) - echo "$count directories created on MDT$((i - 1))" - - if [ $stripe_count -gt 1 ]; then - count=$($LFS getdirstripe $testdir/* | - grep -P "^\s+$((i - 1))\t" | wc -l) - echo "$count stripes created on MDT$((i - 1))" - fi + max=0 + for (( i = 0; i < $MDSCOUNT; i++ )); do + count=$($LFS getdirstripe -i $testdir/* | grep -c "^$i$") + (( count > max )) && max=$count + echo "$count directories created on MDT$i" done - max=$($LFS getdirstripe -i $testdir/* | grep ^$max_index$ | wc -l) - min=$($LFS getdirstripe -i $testdir/* | grep ^$min_index$ | wc -l) + min=$($LFS getdirstripe -i $testdir/* | grep -c "^$min_index$") # D-value should > 10% of averge - (( $max - $min < 10 )) && - error "subdirs shouldn't be evenly distributed" + (( max - min > num / 10 )) || + error "subdirs shouldn't be evenly distributed: $max - $min < $((num / 10))" + + # ditto for stripes + if (( stripe_count > 1 )); then + max=0 + for (( i = 0; i < $MDSCOUNT; i++ )); do + count=$($LFS getdirstripe $testdir/* | + grep -c -P "^\s+$i\t") + (( count > max )) && max=$count + echo "$count stripes created on MDT$i" + done - # ditto - if [ $stripe_count -gt 1 ]; then - max=$($LFS getdirstripe $testdir/* | - grep -P "^\s+$max_index\t" | wc -l) min=$($LFS getdirstripe $testdir/* | - grep -P "^\s+$min_index\t" | wc -l) - (( $max - $min < 10 * $stripe_count )) && - error "stripes shouldn't be evenly distributed"|| true + grep -c -P "^\s+$min_index\t") + (( max - min > num * stripe_count / 10 )) || + error "stripes shouldn't be evenly distributed: $max - $min < $((num / 10)) * $stripe_count" fi } +most_full_mdt() { + local ffree + local bavail + local bsize + local min + local min_index + local tmp + + ffree=($(lctl get_param -n mdc.*[mM][dD][cC]-[^M]*.filesfree)) + bavail=($(lctl get_param -n mdc.*[mM][dD][cC]-[^M]*.kbytesavail)) + bsize=$(lctl get_param -n mdc.*MDT0000*.blocksize) + + min=$(((${ffree[0]} >> 8) * (${bavail[0]} * bsize >> 16))) + min_index=0 + for ((i = 1; i < ${#ffree[@]}; i++)); do + tmp=$(((${ffree[i]} >> 8) * (${bavail[i]} * bsize >> 16))) + (( tmp < min )) && min=$tmp && min_index=$i + done + + echo -n $min_index +} + test_413a() { [ $MDSCOUNT -lt 2 ] && skip "We need at least 2 MDTs for this test" @@ -24339,11 +25296,13 @@ test_413a() { local stripe_count + generate_uneven_mdts 100 for stripe_count in $(seq 1 $((MDSCOUNT - 1))); do mkdir $DIR/$tdir-s$stripe_count || error "mkdir failed" mkdir $DIR/$tdir-s$stripe_count/rr || error "mkdir failed" - mkdir $DIR/$tdir-s$stripe_count/qos || error "mkdir failed" - test_qos_mkdir "$LFS mkdir -c $stripe_count" $stripe_count + $LFS mkdir -i $(most_full_mdt) $DIR/$tdir-s$stripe_count/qos || + error "mkdir failed" + test_qos_mkdir "$LFS mkdir -i -1 -c $stripe_count" $stripe_count done } run_test 413a "QoS mkdir with 'lfs mkdir -i -1'" @@ -24358,11 +25317,13 @@ test_413b() { local testdir local stripe_count + generate_uneven_mdts 100 for stripe_count in $(seq 1 $((MDSCOUNT - 1))); do testdir=$DIR/$tdir-s$stripe_count mkdir $testdir || error "mkdir $testdir failed" mkdir $testdir/rr || error "mkdir rr failed" - mkdir $testdir/qos || error "mkdir qos failed" + $LFS mkdir -i $(most_full_mdt) $testdir/qos || + error "mkdir qos failed" $LFS setdirstripe -D -c $stripe_count --max-inherit-rr 2 \ $testdir/rr || error "setdirstripe rr failed" $LFS setdirstripe -D -c $stripe_count $testdir/qos || @@ -24373,11 +25334,11 @@ test_413b() { run_test 413b "QoS mkdir under dir whose default LMV starting MDT offset is -1" test_413c() { - [ $MDSCOUNT -ge 2 ] || + (( $MDSCOUNT >= 2 )) || skip "We need at least 2 MDTs for this test" - [ $MDS1_VERSION -ge $(version_code 2.14.51) ] || - skip "Need server version at least 2.14.50" + (( $MDS1_VERSION >= $(version_code 2.14.51) )) || + skip "Need server version at least 2.14.51" local testdir local inherit @@ -24386,11 +25347,11 @@ test_413c() { testdir=$DIR/${tdir}-s1 mkdir $testdir || error "mkdir $testdir failed" mkdir $testdir/rr || error "mkdir rr failed" - mkdir $testdir/qos || error "mkdir qos failed" + $LFS mkdir -i $(most_full_mdt) $testdir/qos || error "mkdir qos failed" # default max_inherit is -1, default max_inherit_rr is 0 $LFS setdirstripe -D -c 1 $testdir/rr || error "setdirstripe rr failed" - $LFS setdirstripe -D -c 1 -X 2 --max-inherit-rr 1 $testdir/qos || + $LFS setdirstripe -D -c 1 -i -1 -X 2 --max-inherit-rr 1 $testdir/qos || error "setdirstripe qos failed" test_qos_mkdir "mkdir" 1 @@ -24398,21 +25359,62 @@ test_413c() { inherit=$($LFS getdirstripe -D -X $testdir/rr/level1) (( $inherit == -1 )) || error "rr/level1 inherit $inherit != -1" inherit_rr=$($LFS getdirstripe -D --max-inherit-rr $testdir/rr/level1) - (( $inherit_rr == 0 )) || - error "rr/level1 inherit-rr $inherit_rr != 0" + (( $inherit_rr == 0 )) || error "rr/level1 inherit-rr $inherit_rr != 0" mkdir $testdir/qos/level1 || error "mkdir qos/level1 failed" inherit=$($LFS getdirstripe -D -X $testdir/qos/level1) (( $inherit == 1 )) || error "qos/level1 inherit $inherit != 1" inherit_rr=$($LFS getdirstripe -D --max-inherit-rr $testdir/qos/level1) - (( $inherit_rr == 0 )) || - error "qos/level1 inherit-rr $inherit_rr !=0" + (( $inherit_rr == 0 )) || error "qos/level1 inherit-rr $inherit_rr != 0" mkdir $testdir/qos/level1/level2 || error "mkdir level2 failed" - getfattr -d -m dmv $testdir/qos/level1/level2 | grep dmv && + getfattr -d -m dmv -e hex $testdir/qos/level1/level2 | grep dmv && error "level2 shouldn't have default LMV" || true } run_test 413c "mkdir with default LMV max inherit rr" +test_413d() { + (( MDSCOUNT >= 2 )) || + skip "We need at least 2 MDTs for this test" + + (( MDS1_VERSION >= $(version_code 2.14.51) )) || + skip "Need server version at least 2.14.51" + + local lmv_qos_threshold_rr + + lmv_qos_threshold_rr=$($LCTL get_param -n lmv.*.qos_threshold_rr | + head -n1) + stack_trap "$LCTL set_param \ + lmv.*.qos_threshold_rr=$lmv_qos_threshold_rr > /dev/null" EXIT + + $LCTL set_param lmv.*.qos_threshold_rr=100 > /dev/null + mkdir -p $DIR/$tdir || error "mkdir $tdir failed" + getfattr -d -m dmv -e hex $DIR/$tdir | grep dmv && + error "$tdir shouldn't have default LMV" + createmany -d $DIR/$tdir/sub $((100 * MDSCOUNT)) || + error "mkdir sub failed" + + local count=$($LFS getstripe -m $DIR/$tdir/* | grep -c ^0) + + (( count == 100 )) || error "$count subdirs on MDT0" +} +run_test 413d "inherit ROOT default LMV" + +test_413z() { + local pids="" + local subdir + local pid + + for subdir in $(\ls -1 -d $DIR/d413*-fillmdt/*); do + unlinkmany $subdir/f. 1000 & + pids="$pids $!" + done + + for pid in $pids; do + wait $pid + done +} +run_test 413z "413 test cleanup" + test_414() { #define OBD_FAIL_PTLRPC_BULK_ATTACH 0x521 $LCTL set_param fail_loc=0x80000521 @@ -24528,7 +25530,8 @@ check_lfs_df() { [ "$1" == "blocks" ] && inodes= || inodes="-i" for count in {1..100}; do - cancel_lru_locks + do_nodes "$CLIENTS" \ + $LCTL set_param ldlm.namespaces.*.lru_size=clear sync; sleep 0.2 # read the lines of interest @@ -24541,7 +25544,9 @@ check_lfs_df() { # ":/" for df, "filesystem_summary:" for lfs df # compare the two outputs passed=true - for i in {1..5}; do + # skip "available" on MDT until LU-13997 is fixed. + #for i in {1..5}; do + for i in 1 2 4 5; do [ "${df_out[i]}" != "${lfs_df_out[i]}" ] && passed=false done $passed && break @@ -25340,6 +26345,38 @@ test_431() { # LU-14187 } run_test 431 "Restart transaction for IO" +cleanup_test_432() { + do_facet mgs $LCTL nodemap_activate 0 + wait_nm_sync active +} + +test_432() { + local tmpdir=$TMP/dir432 + + (( $MDS1_VERSION >= $(version_code 2.14.52) )) || + skip "Need MDS version at least 2.14.52" + + stack_trap cleanup_test_432 EXIT + mkdir $DIR/$tdir + mkdir $tmpdir + + do_facet mgs $LCTL nodemap_activate 1 + wait_nm_sync active + do_facet mgs $LCTL nodemap_modify --name default \ + --property admin --value 1 + do_facet mgs $LCTL nodemap_modify --name default \ + --property trusted --value 1 + cancel_lru_locks mdc + wait_nm_sync default admin_nodemap + wait_nm_sync default trusted_nodemap + + if [ $(mv $tmpdir $DIR/$tdir/ 2>&1 | + grep -ci "Operation not permitted") -ne 0 ]; then + error "mv $tmpdir $DIR/$tdir/ hits 'Operation not permitted'" + fi +} +run_test 432 "mv dir from outside Lustre" + prep_801() { [[ $MDS1_VERSION -lt $(version_code 2.9.55) ]] || [[ $OST1_VERSION -lt $(version_code 2.9.55) ]] && @@ -25653,7 +26690,7 @@ test_803a() { [ $MDS1_VERSION -lt $(version_code 2.10.54) ] && skip "MDS needs to be newer than 2.10.54" - mkdir -p $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir # Create some objects on all MDTs to trigger related logs objects for idx in $(seq $MDSCOUNT); do $LFS mkdir -c $MDSCOUNT -i $((idx % $MDSCOUNT)) \ @@ -25848,7 +26885,7 @@ test_805() { fi do_facet $SINGLEMDS zfs set quota=$(((usedkb+freekb)*1024)) $fsset trap cleanup_805 EXIT - mkdir $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir $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" @@ -25996,7 +27033,7 @@ test_807() { stack_trap "restore_lustre_params < $save; rm -f $save" EXIT rm -rf $DIR/$tdir || error "rm $tdir failed" - mkdir -p $DIR/$tdir || error "mkdir $tdir failed" + mkdir_on_mdt0 $DIR/$tdir || error "mkdir $tdir failed" touch $DIR/$tdir/trunc || error "touch $tdir/trunc failed" $TRUNCATE $DIR/$tdir/trunc 1024 || error "truncate $tdir/trunc failed" $TRUNCATE $DIR/$tdir/trunc 1048576 || @@ -26534,6 +27571,36 @@ test_822() { } run_test 822 "test precreate failure" +test_823() { + local p="$TMP/$TESTSUITE-$TESTNAME.parameters" + local OST_MAX_PRECREATE=20000 + + save_lustre_params mds1 \ + "osp.$FSNAME-OST*-osc-MDT0000.max_create_count" > $p + do_facet $SINGLEMDS "$LCTL set_param -n \ + osp.$FSNAME-OST*MDT0000.max_create_count=0" + do_facet $SINGLEMDS "$LCTL set_param -n \ + osp.$FSNAME-OST0000*MDT0000.max_create_count=$OST_MAX_PRECREATE" + + stack_trap "restore_lustre_params < $p; rm $p" + + do_facet $SINGLEMDS "$LCTL set_param -n \ + osp.$FSNAME-OST*-osc-MDT*.create_count=100200" + + local count=$(do_facet $SINGLEMDS "$LCTL get_param -n \ + osp.$FSNAME-OST0000*MDT0000.create_count") + local max=$(do_facet $SINGLEMDS "$LCTL get_param -n \ + osp.$FSNAME-OST0000*MDT0000.max_create_count") + local expect_count=$(((($max/2)/256) * 256)) + + log "setting create_count to 100200:" + log " -result- count: $count with max: $max, expecting: $expect_count" + + [[ $count -eq expect_count ]] || + error "Create count not set to max precreate." +} +run_test 823 "Setting create_count > OST_MAX_PRECREATE is lowered to maximum" + # # tests that do cleanup/setup should be run at the end # @@ -26589,6 +27656,21 @@ test_902() { } run_test 902 "test short write doesn't hang lustre" +# LU-14711 +test_903() { + $LFS setstripe -i 0 -c 1 $DIR/$tfile $DIR/${tfile}-2 + echo "blah" > $DIR/${tfile}-2 + dd if=/dev/zero of=$DIR/$tfile bs=1M count=6 conv=fsync + #define OBD_FAIL_OSC_SLOW_PAGE_EVICT 0x417 + $LCTL set_param fail_loc=0x417 fail_val=20 + + mv $DIR/${tfile}-2 $DIR/$tfile # Destroys the big object + sleep 1 # To start the destroy + wait_destroy_complete 150 || error "Destroy taking too long" + cat $DIR/$tfile > /dev/null || error "Evicted" +} +run_test 903 "Test long page discard does not cause evictions" + complete $SECONDS [ -f $EXT2_DEV ] && rm $EXT2_DEV || true check_and_cleanup_lustre