X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftests%2Fsanity.sh;h=16d3553021d7323c6316f5bb83b662610e9342ba;hb=51a483375eed8882cfe0d707158ed0eafdaf76c1;hp=986228d0f04920608faa6cf2aa5e635d7b8503a9;hpb=d2cb789537485eec9bfc5bc7d4c85025b0488478;p=fs%2Flustre-release.git diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh index 986228d..16d3553 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 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 @@ -1586,7 +1617,7 @@ test_27cb() { # Strip getstripe output to a space separated list of OSTs local getstripe_osts=$(echo "$getstripe" | sed -e '1,/obdidx/d' |\ - awk '{print $1}' | tr '\n' '\ ' | sed -e 's/[[:space:]]*$//') + awk '{print $1}' | tr '\n' ' ' | sed -e 's/[[:space:]]*$//') [ "$getstripe_osts" = "${osts//,/ }" ] || error "stripes not on specified OSTs" @@ -1607,7 +1638,7 @@ test_27cc() { # Strip getstripe output to a space separated list of OSTs local getstripe_osts=$(echo "$getstripe" | sed -e '1,/obdidx/d' |\ - awk '{print $1}' | tr '\n' '\ ' | sed -e 's/[[:space:]]*$//') + awk '{print $1}' | tr '\n' ' ' | sed -e 's/[[:space:]]*$//') [ "$getstripe_osts" = "${osts//,/ }" ] || error "stripes not on specified OSTs" @@ -1627,7 +1658,7 @@ test_27cd() { # Strip getstripe output to a space separated list of OSTs local getstripe_osts=$(echo "$getstripe" | sed -e '1,/obdidx/d' |\ - awk '{print $1}' | tr '\n' '\ ' | sed -e 's/[[:space:]]*$//') + awk '{print $1}' | tr '\n' ' ' | sed -e 's/[[:space:]]*$//') [ "$getstripe_osts" = "${osts//,/ }" ] || error "stripes not on specified OSTs" @@ -1641,7 +1672,7 @@ test_27ce() { [[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] || skip "server does not support overstriping" # We do one more stripe than we have OSTs - [ $OSTCOUNT -ge 159 ] || large_xattr_enabled || + [ $OSTCOUNT -lt 159 ] || large_xattr_enabled || skip_env "ea_inode feature disabled" test_mkdir -p $DIR/$tdir @@ -1659,7 +1690,7 @@ test_27ce() { # Strip getstripe output to a space separated list of OSTs local getstripe_osts=$(echo "$getstripe" | sed -e '1,/obdidx/d' |\ - awk '{print $1}' | tr '\n' '\ ' | sed -e 's/[[:space:]]*$//') + awk '{print $1}' | tr '\n' ' ' | sed -e 's/[[:space:]]*$//') [ "$getstripe_osts" = "${osts//,/ }" ] || error "stripes not on specified OSTs" @@ -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" @@ -2448,7 +2479,7 @@ test_27Cb() { test_mkdir -p $DIR/$tdir local setcount=$(($OSTCOUNT * 2)) - [ $setcount -ge 160 ] || large_xattr_enabled || + [ $setcount -lt 160 ] || large_xattr_enabled || skip_env "ea_inode feature disabled" $LFS setstripe -C $setcount $DIR/$tdir/$tfile || @@ -2474,7 +2505,7 @@ test_27Cc() { test_mkdir -p $DIR/$tdir local setcount=$(($OSTCOUNT - 1)) - [ $setcount -ge 160 ] || large_xattr_enabled || + [ $setcount -lt 160 ] || large_xattr_enabled || skip_env "ea_inode feature disabled" $LFS setstripe -C $setcount $DIR/$tdir/$tfile || @@ -2554,7 +2585,7 @@ test_27Cf() { test_mkdir -p $DIR/$tdir local setcount=$(($OSTCOUNT * 2)) - [ $setcount -ge 160 ] || large_xattr_enabled || + [ $setcount -lt 160 ] || large_xattr_enabled || skip_env "ea_inode feature disabled" $LFS setstripe -C $setcount $DIR/$tdir/ || @@ -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 @@ -3319,6 +3363,38 @@ test_27P() { } run_test 27P "basic ops on foreign dir of foreign_symlink type" +test_27Q() { + rm -f $TMP/$tfile $TMP/$tfile.loop $TMP/$tfile.none $TMP/$tfile.broken + stack_trap "rm -f $TMP/$tfile*" + + test_mkdir $DIR/$tdir-1 + test_mkdir $DIR/$tdir-2 + + echo 'It is what it is' > $DIR/$tdir-1/$tfile + lov_getstripe_old $DIR/$tdir-1/$tfile || error "$DIR/$tdir-1/$tfile: rc = $?" + + ln -s $DIR/$tdir-1/$tfile $DIR/$tdir-2/$tfile + lov_getstripe_old $DIR/$tdir-2/$tfile || error "$DIR/$tdir-2/$tfile: rc = $?" + + ln -s $DIR/$tdir-1/$tfile $TMP/$tfile + lov_getstripe_old $TMP/$tfile || error "$TMP/$tfile: rc = $?" + + # Create some bad symlinks and ensure that we don't loop + # forever or something. These should return ELOOP (40) and + # ENOENT (2) but I don't want to test for that because there's + # always some weirdo architecture that needs to ruin + # everything by defining these error numbers differently. + + ln -s $TMP/$tfile.loop $TMP/$tfile.loop + lov_getstripe_old $TMP/$tfile.loop && error "$TMP/$tfile.loop: rc = $?" + + ln -s $TMP/$tfile.none $TMP/$tfile.broken + lov_getstripe_old $TMP/$tfile.broken && error "$TMP/$tfile.broken: rc = $?" + + return 0 +} +run_test 27Q "llapi_file_get_stripe() works on symlinks" + # createtest also checks that device nodes are created and # then visible correctly (#2091) test_28() { # bug 2091 @@ -3645,6 +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 @@ -3954,63 +4052,57 @@ test_33c() { local write_bytes local all_zeros - all_zeros=: - rm -fr $DIR/$tdir + all_zeros=true test_mkdir $DIR/$tdir # Read: 0, Write: 4, create/destroy: 2/0, stat: 1, punch: 0 - sync - for ostnum in $(seq $OSTCOUNT); do - # test-framework's OST numbering is one-based, while Lustre's - # is zero-based - ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1))) - # Parsing llobdstat's output sucks; we could grep the /proc - # path, but that's likely to not be as portable as using the - # llobdstat utility. So we parse lctl output instead. - write_bytes=$(do_facet ost$ostnum lctl get_param -n \ - obdfilter/$ostname/stats | - awk '/^write_bytes/ {print $7}' ) - echo "baseline_write_bytes@$OSTnum/$ostname=$write_bytes" - if (( ${write_bytes:-0} > 0 )) - then - all_zeros=false - break; - fi - done + sync + for ostnum in $(seq $OSTCOUNT); do + # test-framework's OST numbering is one-based, while Lustre's + # is zero-based + ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1))) + # check if at least some write_bytes stats are counted + write_bytes=$(do_facet ost$ostnum lctl get_param -n \ + obdfilter.$ostname.stats | + awk '/^write_bytes/ {print $7}' ) + echo "baseline_write_bytes@ost$ostnum/$ostname=$write_bytes" + if (( ${write_bytes:-0} > 0 )); then + all_zeros=false + break + fi + done - $all_zeros || return 0 + $all_zeros || return 0 # Write four bytes echo foo > $DIR/$tdir/bar # Really write them sync - # Total up write_bytes after writing. We'd better find non-zeros. - for ostnum in $(seq $OSTCOUNT); do - ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1))) - write_bytes=$(do_facet ost$ostnum lctl get_param -n \ - obdfilter/$ostname/stats | - awk '/^write_bytes/ {print $7}' ) - echo "write_bytes@$OSTnum/$ostname=$write_bytes" - if (( ${write_bytes:-0} > 0 )) - then - all_zeros=false - break; - fi - done + # Total up write_bytes after writing. We'd better find non-zeros. + for ostnum in $(seq $OSTCOUNT); do + ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1))) + write_bytes=$(do_facet ost$ostnum lctl get_param -n \ + obdfilter/$ostname/stats | + awk '/^write_bytes/ {print $7}' ) + echo "write_bytes@ost$ostnum/$ostname=$write_bytes" + if (( ${write_bytes:-0} > 0 )); then + all_zeros=false + break + fi + done - if $all_zeros - then - for ostnum in $(seq $OSTCOUNT); do - ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1))) - echo "Check that write_bytes is present in obdfilter/*/stats:" - do_facet ost$ostnum lctl get_param -n \ - obdfilter/$ostname/stats - done - error "OST not keeping write_bytes stats (b22312)" - fi + if $all_zeros; then + for ostnum in $(seq $OSTCOUNT); do + ostname=$(printf "$FSNAME-OST%.4x" $((ostnum - 1))) + echo "Check write_bytes is in obdfilter.*.stats:" + do_facet ost$ostnum lctl get_param -n \ + obdfilter.$ostname.stats + done + error "OST not keeping write_bytes stats (b=22312)" + fi } -run_test 33c "test llobdstat and write_bytes" +run_test 33c "test write_bytes stats" test_33d() { [[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs" @@ -4740,7 +4832,7 @@ test_39l() { local atime_diff=$(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 @@ -5532,7 +5624,7 @@ test_51b() { # cleanup the directory rm -fr $dir - test_mkdir -c1 $dir + mkdir_on_mdt -i $((RANDOM % MDSCOUNT)) $dir $LFS df $LFS df -i @@ -5868,6 +5960,7 @@ run_test 54e "console/tty device works in lustre ======================" test_56a() { local numfiles=3 + local numdirs=2 local dir=$DIR/$tdir rm -rf $dir @@ -5907,9 +6000,10 @@ test_56a() { #test lfs getstripe with -v prints lmm_fid filenum=$($LFS getstripe -v $dir | grep -c lmm_fid) - [[ $filenum -eq $((numfiles * numcomp)) ]] || + local countfids=$((numdirs + numfiles * numcomp)) + [[ $filenum -eq $countfids ]] || error "$LFS getstripe -v $dir: "\ - "got $filenum want $((numfiles * numcomp)) lmm_fid" + "got $filenum want $countfids lmm_fid" [[ $($LFS getstripe $dir | grep -c lmm_fid) -eq 0 ]] || error "$LFS getstripe $dir: showed lmm_fid by default" echo "$LFS getstripe --verbose passed" @@ -6037,6 +6131,47 @@ 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() { @@ -6265,9 +6400,9 @@ test_newerXY_base() { if [ $y == "t" ]; then if [ $x == "b" ]; then - ref="\"$(do_facet mds1 date +"%Y-%m-%d\ %H:%M:%S")\"" + ref=\"$(do_facet mds1 date +"%Y-%m-%d\ %H:%M:%S")\" else - ref="\"$(date +"%Y-%m-%d %H:%M:%S")\"" + ref=\"$(date +"%Y-%m-%d %H:%M:%S")\" fi else ref=$DIR/$tfile.newer.$x$y @@ -6280,9 +6415,9 @@ test_newerXY_base() { sleep 2 if [ $y == "t" ]; then if [ $x == "b" ]; then - negref="\"$(do_facet mds1 date +"%Y-%m-%d\ %H:%M:%S")\"" + negref=\"$(do_facet mds1 date +"%Y-%m-%d\ %H:%M:%S")\" else - negref="\"$(date +"%Y-%m-%d %H:%M:%S")\"" + negref=\"$(date +"%Y-%m-%d %H:%M:%S")\" fi else negref=$DIR/$tfile.negnewer.$x$y @@ -6637,6 +6772,36 @@ test_56rb() { } run_test 56rb "check lfs find --size --ost/--mdt works" +test_56rc() { + (( MDSCOUNT >= 2 )) || skip "needs at least 2 MDTs" + local dir=$DIR/$tdir + local found + + test_mkdir -c 2 -H all_char $dir || error "failed to mkdir $dir" + $LFS mkdir -c 2 --mdt-hash all_char $dir/$tdir-all{1..10} + (( $MDSCOUNT > 2 )) && + $LFS mkdir -c 3 --mdt-hash fnv_1a_64 $dir/$tdir-fnv{1..10} + mkdir $dir/$tdir-{1..10} + touch $dir/$tfile-{1..10} + + found=$($LFS find $dir --mdt-count 2 | wc -l) + expect=11 + (( $found == $expect )) || error "found $found 2-stripe, expect $expect" + + found=$($LFS find $dir -T +1 | wc -l) + (( $MDSCOUNT > 2 )) && expect=$((expect + 10)) + (( $found == $expect )) || error "found $found 2+stripe, expect $expect" + + found=$($LFS find $dir --mdt-hash all_char | wc -l) + expect=11 + (( $found == $expect )) || error "found $found all_char, expect $expect" + + found=$($LFS find $dir --mdt-hash fnv_1a_64 | wc -l) + (( $MDSCOUNT > 2 )) && expect=10 || expect=0 + (( $found == $expect )) || error "found $found all_char, expect $expect" +} +run_test 56rc "check lfs find --mdt-count/--mdt-hash works" + test_56s() { # LU-611 #LU-9369 [[ $OSTCOUNT -lt 2 ]] && skip_env "need at least 2 OSTs" @@ -7520,6 +7685,85 @@ 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 $SINGLEMDS) MDS < 2.4.53" @@ -7617,6 +7861,94 @@ test_56ab() { # LU-10705 } run_test 56ab "lfs find --blocks" +# LU-11188 +test_56aca() { + local dir="$DIR/$tdir" + local perms=(001 002 003 004 005 006 007 + 010 020 030 040 050 060 070 + 100 200 300 400 500 600 700 + 111 222 333 444 555 666 777) + local perm_minus=(8 8 4 8 4 4 2 + 8 8 4 8 4 4 2 + 8 8 4 8 4 4 2 + 4 4 2 4 2 2 1) + local perm_slash=(8 8 12 8 12 12 14 + 8 8 12 8 12 12 14 + 8 8 12 8 12 12 14 + 16 16 24 16 24 24 28) + + test_mkdir "$dir" + for perm in ${perms[*]}; do + touch "$dir/$tfile.$perm" + chmod $perm "$dir/$tfile.$perm" + done + + for ((i = 0; i < ${#perms[*]}; i++)); do + local num=$($LFS find $dir -perm ${perms[i]} | wc -l) + (( $num == 1 )) || + error "lfs find -perm ${perms[i]}:"\ + "$num != 1" + + num=$($LFS find $dir -perm -${perms[i]} -type f| wc -l) + (( $num == ${perm_minus[i]} )) || + error "lfs find -perm -${perms[i]}:"\ + "$num != ${perm_minus[i]}" + + num=$($LFS find $dir -perm /${perms[i]} -type f| wc -l) + (( $num == ${perm_slash[i]} )) || + error "lfs find -perm /${perms[i]}:"\ + "$num != ${perm_slash[i]}" + done +} +run_test 56aca "check lfs find -perm with octal representation" + +test_56acb() { + local dir=$DIR/$tdir + # p is the permission of write and execute for user, group and other + # without the umask. It is used to test +wx. + local p=$(printf "%o" "$((0333 & ~$(umask)))") + local perms=(1000 000 2000 4000 $p 644 111 110 100 004) + local symbolic=(+t a+t u+t g+t o+t + g+s u+s o+s +s o+sr + o=r,ug+o,u+w + u+ g+ o+ a+ ugo+ + u- g- o- a- ugo- + u= g= o= a= ugo= + o=r,ug+o,u+w u=r,a+u,u+w + g=r,ugo=g,u+w u+x,+X +X + u+x,u+X u+X u+x,g+X o+r,+X + u+x,go+X +wx +rwx) + + test_mkdir $dir + for perm in ${perms[*]}; do + touch "$dir/$tfile.$perm" + chmod $perm "$dir/$tfile.$perm" + done + + for (( i = 0; i < ${#symbolic[*]}; i++ )); do + local num=$($LFS find $dir -perm ${symbolic[i]} | wc -l) + + (( $num == 1 )) || + error "lfs find $dir -perm ${symbolic[i]}: $num != 1" + done +} +run_test 56acb "check lfs find -perm with symbolic representation" + +test_56acc() { + local dir=$DIR/$tdir + local tests="17777 787 789 abcd + ug=uu ug=a ug=gu uo=ou urw + u+xg+x a=r,u+x," + + test_mkdir $dir + for err in $tests; do + if $LFS find $dir -perm $err 2>/dev/null; then + error "lfs find -perm $err: parsing should have failed" + fi + done +} +run_test 56acc "check parsing error for lfs find -perm" + test_56ba() { [ $MDS1_VERSION -lt $(version_code 2.10.50) ] && skip "Need MDS version at least 2.10.50" @@ -7756,6 +8088,38 @@ test_56ca() { } run_test 56ca "check lfs find --mirror-count|-N and --mirror-state" +test_56da() { # LU-14179 + local path=$DIR/$tdir + + test_mkdir $path + cd $path + + local longdir=$(str_repeat 'a' 255) + + for i in {1..15}; do + path=$path/$longdir + test_mkdir $longdir + cd $longdir + done + + local len=${#path} + local lastdir=$(str_repeat 'a' $((4096 - 1 - $len - 1))) + + test_mkdir $lastdir + cd $lastdir + # PATH_MAX-1 + (( ${#PWD} == 4095 )) || error "bad PWD length ${#PWD}, expect 4095" + + # NAME_MAX + touch $(str_repeat 'f' 255) + + $LFS find $DIR/$tdir --type d |& grep "lfs find: error" && + error "lfs find reported an error" + + rm -rf $DIR/$tdir +} +run_test 56da "test lfs find with long paths" + test_57a() { [ $PARALLEL == "yes" ] && skip "skip parallel run" # note test will not do anything if MDS is not local @@ -8017,6 +8381,25 @@ test_60e() { } run_test 60e "no space while new llog is being created" +test_60f() { + local old_path=$($LCTL get_param -n debug_path) + + stack_trap "$LCTL set_param debug_path=$old_path" + stack_trap "rm -f $TMP/$tfile*" + rm -f $TMP/$tfile* 2> /dev/null + #define OBD_FAIL_PTLRPC_DUMP_LOG 0x50e + $LCTL set_param debug_path=$TMP/$tfile fail_loc=0x8000050e + test_mkdir $DIR/$tdir + # retry in case the open is cached and not released + for (( i = 0; i < 100 && $(ls $TMP/$tfile* | wc -l) == 0; i++ )); do + echo $i > $DIR/$tdir/$tfile.$i && cat $DIR/$tdir/$tfile.$i + sleep 0.1 + done + ls $TMP/$tfile* + (( $(ls $TMP/$tfile* | wc -l) > 0 )) || error "$TMP/$tfile not dumped" +} +run_test 60f "change debug_path works" + test_60g() { local pid local i @@ -8101,6 +8484,34 @@ test_60h() { } run_test 60h "striped directory with missing stripes can be accessed" +function t60i_load() { + mkdir $DIR/$tdir + #define OBD_FAIL_LLOG_PAUSE_AFTER_PAD 0x131c + $LCTL set_param fail_loc=0x131c fail_val=1 + for ((i=0; i<5000; i++)); do + touch $DIR/$tdir/f$i + done +} + +test_60i() { + changelog_register || error "changelog_register failed" + local cl_user="${CL_USERS[$SINGLEMDS]%% *}" + changelog_users $SINGLEMDS | grep -q $cl_user || + error "User $cl_user not found in changelog_users" + changelog_chmask "ALL" + t60i_load & + local PID=$! + for((i=0; i<100; i++)); do + changelog_dump >/dev/null || + error "can't read changelog" + done + kill $PID + wait $PID + changelog_deregister || error "changelog_deregister failed" + $LCTL set_param fail_loc=0 +} +run_test 60i "llog: new record vs reader race" + test_61a() { [ $PARALLEL == "yes" ] && skip "skip parallel run" @@ -8654,8 +9065,7 @@ test_65n() { which getfattr > /dev/null 2>&1 || skip_env "no getfattr command" which setfattr > /dev/null 2>&1 || skip_env "no setfattr command" - local root_layout=$(save_layout $MOUNT) - stack_trap "restore_layout $MOUNT $root_layout" EXIT + save_layout_restore_at_exit $MOUNT # new subdirectory under root directory should not inherit # the default layout from root @@ -9224,6 +9634,7 @@ 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 @@ -9248,6 +9659,7 @@ 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 @@ -9369,6 +9781,65 @@ 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" + cleanup_test_78() { trap 0 rm -f $DIR/$tfile @@ -9584,18 +10055,9 @@ run_test 100 "check local port using privileged port ===========" function get_named_value() { - local tag - - tag=$1 - while read ;do - line=$REPLY - case $line in - $tag*) - echo $line | sed "s/^$tag[ ]*//" - break - ;; - esac - done + local tag=$1 + + grep -w "$tag" | sed "s/^$tag *\([0-9]*\) *.*/\1/" } export CACHE_MAX=$($LCTL get_param -n llite.*.max_cached_mb | @@ -9614,10 +10076,10 @@ test_101a() { local nreads=10000 local cache_limit=32 - $LCTL set_param -n osc.*-osc*.rpc_stats 0 + $LCTL set_param -n osc.*-osc*.rpc_stats=0 trap cleanup_101a EXIT - $LCTL set_param -n llite.*.read_ahead_stats 0 - $LCTL set_param -n llite.*.max_cached_mb $cache_limit + $LCTL set_param -n llite.*.read_ahead_stats=0 + $LCTL set_param -n llite.*.max_cached_mb=$cache_limit # # randomly read 10000 of 64K chunks from file 3x 32MB in size @@ -9627,7 +10089,7 @@ test_101a() { discard=0 for s in $($LCTL get_param -n llite.*.read_ahead_stats | - get_named_value 'read but discarded' | cut -d" " -f1); do + get_named_value 'read.but.discarded'); do discard=$(($discard + $s)) done cleanup_101a @@ -9688,8 +10150,7 @@ ra_check_101() { local discard_limit=$((((STRIDE_LENGTH - 1)*3/(STRIDE_LENGTH*OSTCOUNT))* \ (STRIDE_LENGTH*OSTCOUNT - STRIDE_LENGTH))) DISCARD=$($LCTL get_param -n llite.*.read_ahead_stats | - get_named_value 'read but discarded' | - cut -d" " -f1 | calc_total) + get_named_value 'read.but.discarded' | calc_total) if [[ $DISCARD -gt $discard_limit ]]; then $LCTL get_param llite.*.read_ahead_stats error "Too many ($DISCARD) discarded pages with size (${READ_SIZE})" @@ -9722,7 +10183,7 @@ test_101b() { local READ_COUNT=$((STRIPE_SIZE/BSIZE)) local STRIDE_LENGTH=$((STRIDE_SIZE/BSIZE)) local OFFSET=$((STRIPE_SIZE/BSIZE*(OSTCOUNT - 1))) - $LCTL set_param -n llite.*.read_ahead_stats 0 + $LCTL set_param -n llite.*.read_ahead_stats=0 $READS -f $DIR/$tfile -l $STRIDE_LENGTH -o $OFFSET \ -s $FILE_LENGTH -b $STRIPE_SIZE -a $READ_COUNT -n $ITERATION cancel_lru_locks osc @@ -9745,7 +10206,7 @@ test_101c() { setup_test101bc $STRIPE_SIZE $FILE_LENGTH cancel_lru_locks osc - $LCTL set_param osc.*.rpc_stats 0 + $LCTL set_param osc.*.rpc_stats=0 $READS -f $DIR/$tfile -s$FILE_LENGTH -b$rsize -n$nreads -t 180 $LCTL get_param osc.*.rpc_stats for osc_rpc_stats in $($LCTL get_param -N osc.*.rpc_stats); do @@ -9768,7 +10229,7 @@ test_101c() { cleanup_test101bc true } -run_test 101c "check stripe_size aligned read-ahead =================" +run_test 101c "check stripe_size aligned read-ahead" test_101d() { [ $PARALLEL == "yes" ] && skip "skip parallel run" @@ -9791,7 +10252,7 @@ test_101d() { echo Disable read-ahead local old_RA=$($LCTL get_param -n llite.*.max_read_ahead_mb | head -n 1) $LCTL set_param -n llite.*.max_read_ahead_mb=0 - stack_trap "$LCTL set_param -n llite.*.max_read_ahead_mb $old_RA" EXIT + stack_trap "$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" @@ -9844,14 +10305,16 @@ test_101e() { cancel_lru_locks $OSC echo "Reset readahead stats" - $LCTL set_param -n llite.*.read_ahead_stats 0 + $LCTL set_param -n llite.*.read_ahead_stats=0 for ((i = 0; i < $count; i++)); do dd if=$file.$i of=/dev/null bs=$bsize count=$size_KB 2>/dev/null done + $LCTL get_param llite.*.max_cached_mb + $LCTL get_param llite.*.read_ahead_stats local miss=$($LCTL get_param -n llite.*.read_ahead_stats | - get_named_value 'misses' | cut -d" " -f1 | calc_total) + get_named_value 'misses' | calc_total) for ((i = 0; i < $count; i++)); do rm -rf $file.$i 2>/dev/null @@ -9876,7 +10339,7 @@ test_101f() { cancel_lru_locks osc echo Reset readahead stats - $LCTL set_param -n llite.*.read_ahead_stats 0 + $LCTL set_param -n llite.*.read_ahead_stats=0 echo mmap read the file with small block size iozone -i 1 -u 1 -l 1 -+n -r 32k -s 128m -B -f $DIR/$tfile \ @@ -9885,7 +10348,7 @@ test_101f() { echo checking missing pages $LCTL get_param llite.*.read_ahead_stats local miss=$($LCTL get_param -n llite.*.read_ahead_stats | - get_named_value 'misses' | cut -d" " -f1 | calc_total) + get_named_value 'misses' | calc_total) $LCTL set_param debug="$old_debug" [ $miss -lt 3 ] || error "misses too much pages ('$miss')!" @@ -9990,7 +10453,7 @@ test_101h() { echo "Read 10M of data but cross 64M bundary" dd if=$DIR/$tfile of=/dev/null bs=10M skip=6 count=1 local miss=$($LCTL get_param -n llite.*.read_ahead_stats | - get_named_value 'misses' | cut -d" " -f1 | calc_total) + get_named_value 'misses' | calc_total) [ $miss -eq 1 ] || error "expected miss 1 but got $miss" rm -f $p $DIR/$tfile } @@ -10038,8 +10501,7 @@ test_101j() { 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) + get_named_value 'failed.to.fast.read' | calc_total) $LCTL get_param -n llite.*.read_ahead_stats [ $miss -eq $count ] || error "expected $count got $miss" done @@ -10771,20 +11233,82 @@ test_103c() { run_test 103c "'cp -rp' won't set empty acl" test_103e() { - (( $MDS1_VERSION >= $(version_code 2.13.59) )) || - skip "MDS needs to be at least 2.13.59" + local numacl + local fileacl + local saved_debug=$($LCTL get_param -n debug) + + (( $MDS1_VERSION >= $(version_code 2.14.0) )) || + skip "MDS needs to be at least 2.14.0" + + large_xattr_enabled || skip_env "ea_inode feature disabled" mkdir -p $DIR/$tdir - # one default ACL will be created for the file owner - for U in {2..256}; do - setfacl -m default:user:$U:rwx $DIR/$tdir - numacl=$(getfacl $DIR/$tdir |& grep -c "default:user") - touch $DIR/$tdir/$tfile.$U || - error "failed to create $tfile.$U with $numacl ACLs" + # add big LOV EA to cause reply buffer overflow earlier + $LFS setstripe -C 1000 $DIR/$tdir + lctl set_param mdc.*-mdc*.stats=clear + + $LCTL set_param debug=0 + stack_trap "$LCTL set_param debug=\"$saved_debug\"" EXIT + stack_trap "$LCTL get_param mdc.*-mdc*.stats" EXIT + + # add a large number of default ACLs (expect 8000+ for 2.13+) + for U in {2..7000}; do + setfacl -d -m user:$U:rwx $DIR/$tdir || + error "Able to add just $U default ACLs" done + numacl=$(getfacl $DIR/$tdir |& grep -c "default:user") + echo "$numacl default ACLs created" + + stat $DIR/$tdir || error "Cannot stat directory" + # check file creation + touch $DIR/$tdir/$tfile || + error "failed to create $tfile with $numacl default ACLs" + stat $DIR/$tdir/$tfile || error "Cannot stat file" + fileacl=$(getfacl $DIR/$tdir/$tfile |& grep -c "user:") + echo "$fileacl ACLs were inherited" + (( $fileacl == $numacl )) || + error "Not all default ACLs were inherited: $numacl != $fileacl" + # check that new ACLs creation adds new ACLs to inherited ACLs + setfacl -m user:19000:rwx $DIR/$tdir/$tfile || + error "Cannot set new ACL" + numacl=$((numacl + 1)) + fileacl=$(getfacl $DIR/$tdir/$tfile |& grep -c "user:") + (( $fileacl == $numacl )) || + error "failed to add new ACL: $fileacl != $numacl as expected" + # adds more ACLs to a file to reach their maximum at 8000+ + numacl=0 + for U in {20000..25000}; do + setfacl -m user:$U:rwx $DIR/$tdir/$tfile || break + numacl=$((numacl + 1)) + done + echo "Added $numacl more ACLs to the file" + fileacl=$(getfacl $DIR/$tdir/$tfile |& grep -c "user:") + echo "Total $fileacl ACLs in file" + stat $DIR/$tdir/$tfile > /dev/null || error "Cannot stat file" + rm -f $DIR/$tdir/$tfile || error "Cannot remove file" + rmdir $DIR/$tdir || error "Cannot remove directory" } run_test 103e "inheritance of big amount of default ACLs" +test_103f() { + (( $MDS1_VERSION >= $(version_code 2.14.51) )) || + skip "MDS needs to be at least 2.14.51" + + large_xattr_enabled || skip_env "ea_inode feature disabled" + + # enable changelog to consume more internal MDD buffers + changelog_register + + mkdir -p $DIR/$tdir + # add big LOV EA + $LFS setstripe -C 1000 $DIR/$tdir + setfacl -d -m user:$U:rwx $DIR/$tdir || error "Cannot add default ACLs" + mkdir $DIR/$tdir/inherited || error "failed to create subdirectory" + rmdir $DIR/$tdir/inherited || error "Cannot remove subdirectory" + rmdir $DIR/$tdir || error "Cannot remove directory" +} +run_test 103f "changelog doesn't interfere with default ACLs buffers" + test_104a() { [ $PARALLEL == "yes" ] && skip "skip parallel run" @@ -10821,6 +11345,106 @@ test_104b() { } run_test 104b "$RUNAS lfs check servers test ====================" +# +# Verify $1 is within range of $2. +# Success when $1 is within range. That is, when $1 is >= 2% of $2 and +# $1 is <= 2% of $2. Else Fail. +# +value_in_range() { + # Strip all units (M, G, T) + actual=$(echo $1 | tr -d A-Z) + expect=$(echo $2 | tr -d A-Z) + + expect_lo=$(($expect * 98 / 100)) # 2% below + expect_hi=$(($expect * 102 / 100)) # 2% above + + # permit 2% drift above and below + (( $actual >= $expect_lo && $actual <= $expect_hi )) +} + +test_104c() { + [ $PARALLEL == "yes" ] && skip "skip parallel run" + [ "$ost1_FSTYPE" == "zfs" ] || skip "zfs only test" + + local ost_param="osd-zfs.$FSNAME-OST0000." + local mdt_param="osd-zfs.$FSNAME-MDT0000." + local ofacets=$(get_facets OST) + local mfacets=$(get_facets MDS) + local saved_ost_blocks= + local saved_mdt_blocks= + + echo "Before recordsize change" + lfs_df=($($LFS df -h | grep "filesystem_summary:")) + df=($(df -h | grep "/mnt/lustre"$)) + + # For checking. + echo "lfs output : ${lfs_df[*]}" + echo "df output : ${df[*]}" + + for facet in ${ofacets//,/ }; do + if [ -z $saved_ost_blocks ]; then + saved_ost_blocks=$(do_facet $facet \ + lctl get_param -n $ost_param.blocksize) + echo "OST Blocksize: $saved_ost_blocks" + fi + ost=$(do_facet $facet lctl get_param -n $ost_param.mntdev) + do_facet $facet zfs set recordsize=32768 $ost + done + + # BS too small. Sufficient for functional testing. + for facet in ${mfacets//,/ }; do + if [ -z $saved_mdt_blocks ]; then + saved_mdt_blocks=$(do_facet $facet \ + lctl get_param -n $mdt_param.blocksize) + echo "MDT Blocksize: $saved_mdt_blocks" + fi + mdt=$(do_facet $facet lctl get_param -n $mdt_param.mntdev) + do_facet $facet zfs set recordsize=32768 $mdt + done + + # Give new values chance to reflect change + sleep 2 + + echo "After recordsize change" + lfs_df_after=($($LFS df -h | grep "filesystem_summary:")) + df_after=($(df -h | grep "/mnt/lustre"$)) + + # For checking. + echo "lfs output : ${lfs_df_after[*]}" + echo "df output : ${df_after[*]}" + + # Verify lfs df + value_in_range ${lfs_df_after[1]%.*} ${lfs_df[1]%.*} || + error "lfs_df bytes: ${lfs_df_after[1]%.*} != ${lfs_df[1]%.*}" + value_in_range ${lfs_df_after[2]%.*} ${lfs_df[2]%.*} || + error "lfs_df used: ${lfs_df_after[2]%.*} != ${lfs_df[2]%.*}" + value_in_range ${lfs_df_after[3]%.*} ${lfs_df[3]%.*} || + error "lfs_df avail: ${lfs_df_after[3]%.*} != ${lfs_df[3]%.*}" + + # Verify df + value_in_range ${df_after[1]%.*} ${df[1]%.*} || + error "df bytes: ${df_after[1]%.*} != ${df[1]%.*}" + value_in_range ${df_after[2]%.*} ${df[2]%.*} || + error "df used: ${df_after[2]%.*} != ${df[2]%.*}" + value_in_range ${df_after[3]%.*} ${df[3]%.*} || + error "df avail: ${df_after[3]%.*} != ${df[3]%.*}" + + # Restore MDT recordize back to original + for facet in ${mfacets//,/ }; do + mdt=$(do_facet $facet lctl get_param -n $mdt_param.mntdev) + do_facet $facet zfs set recordsize=$saved_mdt_blocks $mdt + done + + # Restore OST recordize back to original + for facet in ${ofacets//,/ }; do + ost=$(do_facet $facet lctl get_param -n $ost_param.mntdev) + do_facet $facet zfs set recordsize=$saved_ost_blocks $ost + done + + return 0 +} +run_test 104c "Verify df vs lfs_df stays same after recordsize change" + test_105a() { # doesn't work on 2.4 kernels touch $DIR/$tfile @@ -13438,8 +14062,12 @@ 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 $proc_dirs -name | + xargs -n 1 badarea_io || error "client badarea_io failed" # remount the FS in case writes/reads /proc break the FS @@ -13452,6 +14080,8 @@ test_133g() { remote_mds_nodsh && skip "remote MDS with nodsh" remote_ost_nodsh && skip "remote OST with nodsh" + local proc_dirs=$(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) @@ -13462,8 +14092,9 @@ 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" @@ -13507,7 +14138,7 @@ 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-*" @@ -13543,7 +14174,7 @@ 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 \ @@ -13752,31 +14383,30 @@ 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 < $want * 105 / 100 )) || @@ -13786,13 +14416,14 @@ run_test 150c "Verify fallocate Size and Blocks" test_150d() { check_set_fallocate_or_skip + local striping="-c2" + + [[ "x$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=$((OSTCOUNT * 1048576)) @@ -13851,6 +14482,176 @@ test_150e() { } run_test 150e "Verify 60% of available OST space consumed by fallocate" +test_150f() { + local size + local blocks + local want_size_before=20480 # in bytes + local want_blocks_before=40 # 512 sized blocks + local want_blocks_after=24 # 512 sized blocks + local length=$(((want_blocks_before - want_blocks_after) * 512)) + + [[ $OST1_VERSION -ge $(version_code 2.14.0) ]] || + skip "need at least 2.14.0 for fallocate punch" + + if [ "$ost1_FSTYPE" = "zfs" ] || [ "$mds1_FSTYPE" = "zfs" ]; then + skip "LU-14160: punch mode is not implemented on OSD ZFS" + fi + + 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" + + # Call fallocate with punch range which is within the file range + fallocate -p --offset 4096 -l $length $DIR/$tfile || + error "fallocate failed: offset 4096 and length $length" + # client must see changes immediately after fallocate + size=$(stat -c '%s' $DIR/$tfile) + blocks=$(stat -c '%b' $DIR/$tfile) + + # Verify punch worked. + (( blocks == want_blocks_after )) || + error "punch failed: blocks $blocks != $want_blocks_after" + + (( size == want_size_before )) || + error "punch failed: size $size != $want_size_before" + + # Verify there is hole in file + local data_off=$(lseek_test -d 4096 $DIR/$tfile) + # precomputed md5sum + local expect="4a9a834a2db02452929c0a348273b4aa" + + cksum=($(md5sum $DIR/$tfile)) + [[ "${cksum[0]}" == "$expect" ]] || + error "unexpected MD5SUM after punch: ${cksum[0]}" + + # Start second sub-case for fallocate punch. + echo "Verify fallocate punch: Range overlapping and less than blocksize" + yes 'A' | dd of=$DIR/$tfile bs=4096 count=5 || + error "dd failed for bs 4096 and count 5" + + # Punch range less than block size will have no change in block count + want_blocks_after=40 # 512 sized blocks + + # Punch overlaps two blocks and less than blocksize + fallocate -p --offset 4000 -l 3000 $DIR/$tfile || + error "fallocate failed: offset 4000 length 3000" + size=$(stat -c '%s' $DIR/$tfile) + blocks=$(stat -c '%b' $DIR/$tfile) + + # Verify punch worked. + (( blocks == want_blocks_after )) || + error "punch failed: blocks $blocks != $want_blocks_after" + + (( size == want_size_before )) || + error "punch failed: size $size != $want_size_before" + + # Verify if range is really zero'ed out. We expect Zeros. + # precomputed md5sum + expect="c57ec5d769c3dbe3426edc3f7d7e11d3" + cksum=($(md5sum $DIR/$tfile)) + [[ "${cksum[0]}" == "$expect" ]] || + error "unexpected MD5SUM after punch: ${cksum[0]}" +} +run_test 150f "Verify fallocate punch functionality" + +test_150g() { + local space + local size + local blocks + local blocks_after + local size_after + local BS=4096 # Block size in bytes + + [[ $OST1_VERSION -ge $(version_code 2.14.0) ]] || + skip "need at least 2.14.0 for fallocate punch" + + if [ "$ost1_FSTYPE" = "zfs" ] || [ "$mds1_FSTYPE" = "zfs" ]; then + skip "LU-14160: punch mode is not implemented on OSD ZFS" + fi + + check_set_fallocate_or_skip + stack_trap "rm -f $DIR/$tfile; wait_delete_completed" + + 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 + if [ $SLOW == "no" ]; then + space=$((1024 * 100 * OSTCOUNT)) + else + # Find OST with Minimum Size + space=$($LFS df | awk "/$FSNAME-OST/ { print \$4 }" | + sort -un | head -1) + echo "min size OST: $space" + space=$(((space * 60)/100 * OSTCOUNT)) + fi + # space in 1k units, round to 4k blocks + local blkcount=$((space * 1024 / $BS)) + + echo "Verify fallocate punch: Very large Range" + fallocate -l${space}k $DIR/$tfile || + error "fallocate ${space}k $DIR/$tfile failed" + # write 1M at the end, start and in the middle + yes 'A' | dd of=$DIR/$tfile bs=$BS count=256 || + error "dd failed: bs $BS count 256" + yes 'A' | dd of=$DIR/$tfile bs=$BS seek=$((blkcount - 256)) count=256 || + error "dd failed: bs $BS count 256 seek $((blkcount - 256))" + yes 'A' | dd of=$DIR/$tfile bs=$BS seek=$((blkcount / 2)) count=1024 || + error "dd failed: bs $BS count 256 seek $((blkcount / 2))" + + # Gather stats. + size=$(stat -c '%s' $DIR/$tfile) + + # gather punch length. + local punch_size=$((size - (BS * 2))) + + echo "punch_size = $punch_size" + echo "size - punch_size: $((size - punch_size))" + echo "size - punch_size in blocks: $(((size - punch_size)/BS))" + + # Call fallocate to punch all except 2 blocks. We leave the + # first and the last block + echo "fallocate -p --offset $BS -l $punch_size $DIR/$tfile" + fallocate -p --offset $BS -l $punch_size $DIR/$tfile || + error "fallocate failed: offset $BS length $punch_size" + + size_after=$(stat -c '%s' $DIR/$tfile) + blocks_after=$(stat -c '%b' $DIR/$tfile) + + # Verify punch worked. + # Size should be kept + (( size == size_after )) || + error "punch failed: size $size != $size_after" + + # two 4k data blocks to remain plus possible 1 extra extent block + (( blocks_after <= ((BS / 512) * 3) )) || + error "too many blocks remains: $blocks_after" + + # Verify that file has hole between the first and the last blocks + local hole_start=$(lseek_test -l 0 $DIR/$tfile) + local hole_end=$(lseek_test -d $BS $DIR/$tfile) + + echo "Hole at [$hole_start, $hole_end)" + (( hole_start == BS )) || + error "no hole at offset $BS after punch" + + (( hole_end == BS + punch_size )) || + error "data at offset $hole_end < $((BS + punch_size))" +} +run_test 150g "Verify fallocate punch on large range" + #LU-2902 roc_hit was not able to read all values from lproc function roc_hit_init() { local list=$(comma_list $(osts_nodes)) @@ -14309,35 +15110,35 @@ test_154f() { [ -n "$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)" @@ -14359,7 +15160,7 @@ 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) @@ -14371,7 +15172,7 @@ 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" && @@ -14391,7 +15192,7 @@ 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" @@ -14742,6 +15543,8 @@ 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" @@ -14752,8 +15555,6 @@ 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" @@ -14767,7 +15568,6 @@ 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 $MKDIRS != 1" @@ -14830,10 +15630,10 @@ test_160a() { error "User '$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 $SINGLEMDS | @@ -14929,11 +15729,21 @@ test_160e() { # Create a user changelog_register || error "changelog_register failed" - # Delete a future user (expect fail) local MDT0=$(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 @@ -15560,6 +16370,221 @@ test_160l() { } run_test 160l "Verify that MTIME changelog records contain the parent FID" +test_160m() { + remote_mds_nodsh && skip "remote MDS with nodsh" && return + [[ $MDS1_VERSION -ge $(version_code 2.14.51) ]] || + skip "Need MDS version at least 2.14.51" + local cl_users + local cl_user1 + local cl_user2 + local pid1 + + # Create a user + changelog_register || error "first changelog_register failed" + changelog_register || error "second changelog_register failed" + + cl_users=(${CL_USERS[mds1]}) + cl_user1="${cl_users[0]}" + cl_user2="${cl_users[1]}" + # generate some changelog records to accumulate on MDT0 + test_mkdir -p -i0 -c1 $DIR/$tdir || error "test_mkdir $tdir failed" + createmany -m $DIR/$tdir/$tfile 50 || + error "create $DIR/$tdir/$tfile failed" + unlinkmany $DIR/$tdir/$tfile 50 || error "unlinkmany failed" + rm -f $DIR/$tdir + + # check changelogs have been generated + local nbcl=$(changelog_dump | wc -l) + [[ $nbcl -eq 0 ]] && error "no changelogs found" + +#define OBD_FAIL_MDS_CHANGELOG_RACE 0x15f + do_facet mds1 $LCTL set_param fail_loc=0x8000015f fail_val=0 + + __changelog_clear mds1 $cl_user1 +10 + __changelog_clear mds1 $cl_user2 0 & + pid1=$! + sleep 2 + __changelog_clear mds1 $cl_user1 0 || + error "fail to cancel record for $cl_user1" + wait $pid1 + [[ $? -eq 0 ]] || error "fail to cancel record for $cl_user2" +} +run_test 160m "Changelog clear race" + +test_160n() { + remote_mds_nodsh && skip "remote MDS with nodsh" && return + [[ $MDS1_VERSION -ge $(version_code 2.14.51) ]] || + skip "Need MDS version at least 2.14.51" + local cl_users + local cl_user1 + local cl_user2 + local pid1 + local first_rec + local last_rec=0 + + # Create a user + changelog_register || error "first changelog_register failed" + + cl_users=(${CL_USERS[mds1]}) + cl_user1="${cl_users[0]}" + + # generate some changelog records to accumulate on MDT0 + test_mkdir -i0 -c1 $DIR/$tdir || error "test_mkdir $tdir failed" + first_rec=$(changelog_users $SINGLEMDS | + awk '/^current.index:/ { print $NF }') + while (( last_rec < (( first_rec + 65000)) )); do + createmany -m $DIR/$tdir/$tfile 10000 || + error "create $DIR/$tdir/$tfile failed" + + for i in $(seq 0 10000); do + mrename $DIR/$tdir/$tfile$i $DIR/$tdir/$tfile-new$i \ + > /dev/null + done + + unlinkmany $DIR/$tdir/$tfile-new 10000 || + error "unlinkmany failed unlink" + last_rec=$(changelog_users $SINGLEMDS | + awk '/^current.index:/ { print $NF }') + echo last record $last_rec + (( last_rec == 0 )) && error "no changelog found" + done + +#define OBD_FAIL_MDS_CHANGELOG_DEL 0x16c + do_facet mds1 $LCTL set_param fail_loc=0x8000016c fail_val=0 + + __changelog_clear mds1 $cl_user1 0 & + pid1=$! + sleep 2 + __changelog_clear mds1 $cl_user1 0 || + error "fail to cancel record for $cl_user1" + wait $pid1 + [[ $? -eq 0 ]] || error "fail to cancel record for $cl_user2" +} +run_test 160n "Changelog destroy race" + +test_160o() { + local mdt="$(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 $(version_code 2.14.51) ]] || + skip "Need MDS version at least 2.14.51" + [[ "$mds1_FSTYPE" == "ldiskfs" ]] || skip "ldiskfs only test" + local cl_users + local cl_user1 + local entry_count + + # Create a user + changelog_register || error "first changelog_register failed" + + cl_users=(${CL_USERS[mds1]}) + cl_user1="${cl_users[0]}" + + test_mkdir -p -i0 -c1 $DIR/$tdir || error "test_mkdir $tdir failed" + createmany -m $DIR/$tdir/$tfile 50 || + error "create $DIR/$tdir/$tfile failed" + unlinkmany $DIR/$tdir/$tfile 50 || error "unlinkmany failed" + rm -rf $DIR/$tdir + + # check changelogs have been generated + entry_count=$(changelog_dump | wc -l) + ((entry_count != 0)) || error "no changelog entries found" + + # remove changelog_users and check that orphan entries are removed + stop mds1 + do_facet mds1 "$DEBUGFS -w -R 'rm changelog_users' $(mdsdevname 1)" + start mds1 || error "cannot start mdt" + entry_count=$(changelog_dump | wc -l) + ((entry_count == 0)) || + error "found $entry_count changelog entries, expected none" +} +run_test 160p "Changelog orphan cleanup with no users" + +test_160q() { + local mdt="$(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" @@ -16521,7 +17546,7 @@ 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 @@ -17113,7 +18138,7 @@ 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" @@ -17154,7 +18179,7 @@ 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" @@ -17207,6 +18232,9 @@ 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 @@ -17292,23 +18320,23 @@ 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 @@ -17317,7 +18345,7 @@ 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 @@ -17645,7 +18673,7 @@ 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} @@ -18525,7 +19553,7 @@ 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 @@ -18862,15 +19890,15 @@ 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 @@ -18884,33 +19912,37 @@ 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)) @@ -18970,6 +20002,15 @@ 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" @@ -19008,6 +20049,44 @@ test_230r() { } run_test 230r "migrate with too many local locks" +test_230s() { + [ $MDS1_VERSION -ge $(version_code 2.13.57) ] || + skip "Need MDS version at least 2.13.57" + + local mdts=$(comma_list $(mdts_nodes)) + local restripe_status=$(do_facet mds1 $LCTL get_param -n \ + mdt.*MDT0000.enable_dir_restripe) + + stack_trap "do_nodes $mdts $LCTL set_param \ + mdt.*.enable_dir_restripe=$restripe_status" + + local st + for st in 0 1; do + do_nodes $mdts "$LCTL set_param mdt.*.enable_dir_restripe=$st" + test_mkdir $DIR/$tdir + $LFS mkdir $DIR/$tdir |& grep "File exists" || + error "$LFS mkdir doesn't return -EEXIST if target exists" + rmdir $DIR/$tdir + done +} +run_test 230s "lfs mkdir should return -EEXIST if target exists" + +test_230t() +{ + [[ $MDSCOUNT -ge 2 ]] || skip_env "needs >= 2 MDTs" + [[ $MDS1_VERSION -ge $(version_code 2.14.50) ]] || + skip "Need MDS version at least 2.14.50" + + test_mkdir $DIR/$tdir || error "mkdir $tdir failed" + test_mkdir $DIR/$tdir/subdir || error "mkdir subdir failed" + $LFS project -p 1 -s $DIR/$tdir || + error "set $tdir project id failed" + $LFS project -p 2 -s $DIR/$tdir/subdir || + error "set subdir project id failed" + $LFS migrate -m 1 -c $MDSCOUNT $DIR/$tdir || error "migrate failed" +} +run_test 230t "migrate directory with project ID set" + test_231a() { # For simplicity this test assumes that max_pages_per_rpc @@ -19339,7 +20418,7 @@ 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 @@ -19531,10 +20610,11 @@ 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" @@ -20058,6 +21138,7 @@ 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 && @@ -20145,6 +21226,7 @@ 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 && @@ -20307,7 +21389,7 @@ 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" @@ -20458,7 +21540,7 @@ 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 && @@ -21350,6 +22432,17 @@ test_273a() { } run_test 273a "DoM: layout swapping should fail with DOM" +test_273b() { + mkdir -p $DIR/$tdir + $LFS setstripe -E 1M -L mdt -E -1 -c -1 $DIR/$tdir + +#define OBD_FAIL_MDS_COMMITRW_DELAY 0x16b + do_facet mds1 $LCTL set_param fail_loc=0x8000016b fail_val=2 + + $MULTIOP $DIR/$tdir/$tfile Ouw2097152c +} +run_test 273b "DoM: race writeback and object destroy" + test_275() { remote_ost_nodsh && skip "remote OST with nodsh" [ $OST1_VERSION -lt $(version_code 2.10.57) ] && @@ -21459,6 +22552,9 @@ test_280() { stop mgs || error "stop mgs failed" #for a race mgs would crash start mgs $(mgsdevname) $MGS_MOUNT_OPTS || error "start mgs failed" + # make sure we unmount client before remounting + wait + umount_client $MOUNT mount_client $MOUNT || error "mount client failed" } run_test 280 "Race between MGS umount and client llog processing" @@ -21583,7 +22679,7 @@ 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" @@ -21745,7 +22841,7 @@ test_300_check_default_striped_dir() echo "checking $dirname $default_count $default_index" $LFS setdirstripe -D -c $default_count -i $default_index \ - -t all_char $DIR/$tdir/$dirname || + -H all_char $DIR/$tdir/$dirname || error "set default stripe on striped dir error" stripe_count=$($LFS getdirstripe -D -c $DIR/$tdir/$dirname) [ $stripe_count -eq $default_count ] || @@ -21762,8 +22858,9 @@ test_300_check_default_striped_dir() unlinkmany $DIR/$tdir/$dirname/f- 10 || error "unlink files failed" for dir in $(find $DIR/$tdir/$dirname/*); do stripe_count=$($LFS getdirstripe -c $dir) - [ $stripe_count -eq $default_count ] || - [ $stripe_count -eq 0 ] || [ $default_count -eq 1 ] || + (( $stripe_count == $default_count )) || + (( $stripe_count == $MDSCOUNT && $default_count == -1 )) || + (( $stripe_count == 0 )) || (( $default_count == 1 )) || error "stripe count $default_count != $stripe_count for $dir" stripe_index=$($LFS getdirstripe -i $dir) @@ -21792,7 +22889,7 @@ 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 @@ -21815,7 +22912,7 @@ test_300g() { #check default stripe count/stripe index test_300_check_default_striped_dir normal_dir $MDSCOUNT 1 test_300_check_default_striped_dir normal_dir 1 0 - test_300_check_default_striped_dir normal_dir 2 1 + test_300_check_default_striped_dir normal_dir -1 1 test_300_check_default_striped_dir normal_dir 2 -1 #delete default stripe information @@ -21828,9 +22925,6 @@ test_300g() { stripe_count=$($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" @@ -21849,7 +22943,7 @@ test_300h() { test_300_check_default_striped_dir striped_dir $MDSCOUNT 1 test_300_check_default_striped_dir striped_dir 1 0 - test_300_check_default_striped_dir striped_dir 2 1 + test_300_check_default_striped_dir striped_dir -1 1 test_300_check_default_striped_dir striped_dir 2 -1 #delete default stripe information @@ -21889,15 +22983,17 @@ test_300i() { error "create dir0 under hash dir failed" $LFS setdirstripe -i0 -c$MDSCOUNT -H fnv_1a_64 $DIR/$tdir/hashdir/d1 || error "create dir1 under hash dir failed" + $LFS setdirstripe -i0 -c$MDSCOUNT -H crush $DIR/$tdir/hashdir/d2 || + error "create dir2 under hash dir failed" # unfortunately, we need to umount to clear dir layout cache for now # once we fully implement dir layout, we can drop this umount_client $MOUNT || error "umount failed" mount_client $MOUNT || error "mount failed" - $LFS find -H fnv_1a_64 $DIR/$tdir/hashdir - local dircnt=$($LFS find -H fnv_1a_64 $DIR/$tdir/hashdir | wc -l) - [ $dircnt -eq 1 ] || error "lfs find striped dir got:$dircnt,except:1" + $LFS find -H fnv_1a_64,crush $DIR/$tdir/hashdir + local dircnt=$($LFS find -H fnv_1a_64,crush $DIR/$tdir/hashdir | wc -l) + [ $dircnt -eq 2 ] || error "lfs find striped dir got:$dircnt,except:1" #set the stripe to be unknown hash type #define OBD_FAIL_UNKNOWN_LMV_STRIPE 0x1901 @@ -22130,7 +23226,7 @@ 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 @@ -22326,7 +23422,7 @@ test_311() { skip "lustre < 2.8.54 does not contain LU-4825 fix" remote_mds_nodsh && skip "remote MDS with nodsh" - local old_iused=$($LFS df -i | grep OST0000 | awk '{ print $3 }') + local old_iused=$($LFS df -i | awk '/OST0000/ { print $3; exit; }') local mdts=$(comma_list $(mdts_nodes)) mkdir -p $DIR/$tdir @@ -22358,7 +23454,7 @@ test_311() { local new_iused for i in $(seq 120); do - new_iused=$($LFS df -i | grep OST0000 | awk '{ print $3 }') + new_iused=$($LFS df -i | awk '/OST0000/ { print $3; exit; }') # system may be too busy to destroy all objs in time, use # a somewhat small value to not fail autotest [ $((old_iused - new_iused)) -gt 400 ] && break @@ -22604,12 +23700,15 @@ test_317() { run_test 317 "Verify blocks get correctly update after truncate" test_318() { + local llite_name="llite.$($LFS getname $MOUNT | awk '{print $1}')" local old_max_active=$($LCTL get_param -n \ - llite.*.max_read_ahead_async_active 2>/dev/null) + ${llite_name}.max_read_ahead_async_active \ + 2>/dev/null) $LCTL set_param llite.*.max_read_ahead_async_active=256 local max_active=$($LCTL get_param -n \ - llite.*.max_read_ahead_async_active 2>/dev/null) + ${llite_name}.max_read_ahead_async_active \ + 2>/dev/null) [ $max_active -ne 256 ] && error "expected 256 but got $max_active" $LCTL set_param llite.*.max_read_ahead_async_active=0 || @@ -22617,7 +23716,7 @@ test_318() { $LCTL set_param llite.*.max_read_ahead_async_active=512 max_active=$($LCTL get_param -n \ - llite.*.max_read_ahead_async_active 2>/dev/null) + ${llite_name}.max_read_ahead_async_active 2>/dev/null) [ $max_active -eq 512 ] || error "expected 512 but got $max_active" # restore @max_active @@ -22625,9 +23724,9 @@ test_318() { llite.*.max_read_ahead_async_active=$old_max_active local old_threshold=$($LCTL get_param -n \ - llite.*.read_ahead_async_file_threshold_mb 2>/dev/null) + ${llite_name}.read_ahead_async_file_threshold_mb 2>/dev/null) local max_per_file_mb=$($LCTL get_param -n \ - llite.*.max_read_ahead_per_file_mb 2>/dev/null) + ${llite_name}.max_read_ahead_per_file_mb 2>/dev/null) local invalid=$(($max_per_file_mb + 1)) $LCTL set_param \ @@ -22639,7 +23738,7 @@ test_318() { llite.*.read_ahead_async_file_threshold_mb=$valid || error "set $valid should succeed" local threshold=$($LCTL get_param -n \ - llite.*.read_ahead_async_file_threshold_mb 2>/dev/null) + ${llite_name}.read_ahead_async_file_threshold_mb 2>/dev/null) [ $threshold -eq $valid ] || error \ "expect threshold $valid got $threshold" $LCTL set_param \ @@ -22720,7 +23819,7 @@ test_398b() { # LU-4198 --filename=$DIR/$tfile || true wait $bg_pid - rm -rf $DIR/$tfile + rm -f $DIR/$tfile } run_test 398b "DIO and buffer IO race" @@ -22761,7 +23860,7 @@ test_398c() { # LU-4198 error "Locks were requested while doing AIO" # get the percentage of 1-page I/O - pct=$($LCTL get_param osc.${FSNAME}-OST0000-osc-ffff*.rpc_stats | + pct=$($LCTL get_param osc.${imp_name}.rpc_stats | grep -A 1 'pages per rpc' | grep -v 'pages per rpc' | awk '{print $7}') [ $pct -le 50 ] || error "$pct% of I/O are 1-page" @@ -22780,27 +23879,28 @@ test_398c() { # LU-4198 --filename=$DIR/$tfile [ $? -eq 0 ] || error "fio large block size failed" - rm -rf $DIR/$tfile + rm -f $DIR/$tfile $LCTL set_param debug="$saved_debug" } run_test 398c "run fio to test AIO" test_398d() { # LU-13846 - test -f aiocp || skip_env "no aiocp installed" - local aio_file=$DIR/aio_file + which aiocp || skip_env "no aiocp installed" + local aio_file=$DIR/$tfile.aio $LFS setstripe -c -1 -S 1M $DIR/$tfile $aio_file dd if=/dev/urandom of=$DIR/$tfile bs=1M count=64 aiocp -a $PAGE_SIZE -b 64M -s 64M -f O_DIRECT $DIR/$tfile $aio_file + stack_trap "rm -f $DIR/$tfile $aio_file" - diff $DIR/$tfile $aio_file || "file diff after aiocp" + diff $DIR/$tfile $aio_file || error "file diff after aiocp" # make sure we don't crash and fail properly aiocp -a 512 -b 64M -s 64M -f O_DIRECT $DIR/$tfile $aio_file && error "aio not aligned with PAGE SIZE should fail" - rm -rf $DIR/$tfile $aio_file + rm -f $DIR/$tfile $aio_file } run_test 398d "run aiocp to verify block size > stripe size" @@ -22811,6 +23911,295 @@ test_398e() { } run_test 398e "O_Direct open cleared by fcntl doesn't cause hang" +test_398f() { # LU-14687 + which aiocp || skip_env "no aiocp installed" + local aio_file=$DIR/$tfile.aio + + $LFS setstripe -c -1 -S 1M $DIR/$tfile $aio_file + + dd if=/dev/zero of=$DIR/$tfile bs=1M count=64 + stack_trap "rm -f $DIR/$tfile $aio_file" + + #define OBD_FAIL_LLITE_PAGE_ALLOC 0x1418 + $LCTL set_param fail_loc=0x1418 + # make sure we don't crash and fail properly + aiocp -b 64M -s 64M -f O_DIRECT $DIR/$tfile $aio_file && + error "aio with page allocation failure succeeded" + $LCTL set_param fail_loc=0 + diff $DIR/$tfile $aio_file + [[ $? != 0 ]] || error "no diff after failed aiocp" +} +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 @@ -22828,7 +24217,7 @@ test_fake_rw() { $LFS setstripe -c 1 -i 0 $DIR/$tfile # get ost1 size - $FSNAME-OST0000 - local ost1_avail_size=$($LFS df | awk /${ost1_svc}/'{ print $4 }') + local ost1_avail_size=$($LFS df $DIR | awk /${ost1_svc}/'{ print $4 }') local blocks=$((ost1_avail_size/2/1024)) # half avail space by megabytes [ $blocks -gt 1000 ] && blocks=1000 # 1G in maximum @@ -22953,7 +24342,6 @@ test_401a() { #LU-7437 #count the number of parameters by "list_param -R" local params=$($LCTL list_param -R '*' 2>/dev/null | wc -l) #count the number of parameters by listing proc files - local proc_regexp="/{proc,sys}/{fs,sys,kernel/debug}/{lustre,lnet}/" local proc_dirs=$(eval \ls -d $proc_regexp 2>/dev/null) echo "proc_dirs='$proc_dirs'" [ -n "$proc_dirs" ] || error "no proc_dirs on $HOSTNAME" @@ -23064,6 +24452,15 @@ 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) && @@ -23370,6 +24767,76 @@ test_412() { } run_test 412 "mkdir on specific MDTs" +generate_uneven_mdts() { + local threshold=$1 + local ffree + local bavail + local max + local min + local max_index + local min_index + local tmp + local i + + echo + echo "Check for uneven MDTs: " + + ffree=($(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=$i + fi + done + + # Check if we need to generate uneven MDTs + local diff=$(((max - min) * 100 / min)) + local testdir=$DIR/$tdir-fillmdt + + 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 100 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 $testdir failed" + for F in f.{0..99}; do + dd if=/dev/zero of=$testdir/$F bs=1M count=1 > \ + /dev/null 2>&1 || error "dd $F failed" + done + + ffree=($(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=$2 @@ -23392,11 +24859,11 @@ test_qos_mkdir() { lmv_qos_threshold_rr=${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) @@ -23407,47 +24874,64 @@ 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 - - echo - echo "Mkdir (stripe_count $stripe_count) roundrobin:" + lod.*.mdt_qos_maxage=$lod_qos_maxage > /dev/null" $LCTL set_param lmv.*.qos_threshold_rr=100 > /dev/null do_nodes $mdts $LCTL set_param lod.*.mdt_qos_threshold_rr=100 > /dev/null testdir=$DIR/$tdir-s$stripe_count/rr - for i in $(seq $((100 * MDSCOUNT))); do + local stripe_index=$($LFS getstripe -m $testdir) + local test_mkdir_rr=true + + echo "dirstripe: '$($LFS getdirstripe $testdir)'" + 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 + $test_mkdir_rr && + echo "Mkdir (stripe_count $stripe_count) roundrobin:" || + echo "Mkdir (stripe_count $stripe_count) on stripe $stripe_index" + + 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))" - [ $count -eq 100 ] || error "subdirs are not evenly distributed" + 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 == $stripe_index )); then + (( $count == 100 * MDSCOUNT )) || + error "$count subdirs created on MDT$i" + else + (( $count == 0 )) || + error "$count subdirs created on MDT$i" + fi - if [ $stripe_count -gt 1 ]; then + 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 -lt $((95 * stripe_count)) ] || - [ $count -gt $((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: " @@ -23479,53 +24963,20 @@ test_qos_mkdir() { fi done - [ ${ffree[min_index]} -eq 0 ] && + (( ${ffree[min_index]} > 0 )) || skip "no free files in MDT$min_index" - [ ${ffree[min_index]} -gt 100000000 ] && - skip "too much free files in MDT$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 - [ $count -gt 40000 ] && [ "$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" - for i in $(seq $count); do - $OPENFILE -f O_CREAT:O_LOV_DELAY_CREATE \ - $DIR/$tdir-MDT$min_index/f$j_$i > /dev/null || - error "create f$j_$i failed" - setfattr -n user.413b -v $value \ - $DIR/$tdir-MDT$min_index/f$j_$i || - error "setfattr f$j_$i failed" - done - - ffree=($(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 + (( ${ffree[min_index]} < 100000000 )) || + skip "too many free files in MDT$min_index" 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 $LCTL set_param lmv.*.qos_maxage=1 > /dev/null @@ -23534,42 +24985,65 @@ test_qos_mkdir() { sleep 1 testdir=$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))" + for (( i = 0; i < $MDSCOUNT; i++ )); do + count=$($LFS getdirstripe -i $testdir/* | grep -c "^$i$") + echo "$count directories created on MDT$i" 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))" + grep -c -P "^\s+$i\t") + echo "$count stripes created on MDT$i" fi done - max=$($LFS getdirstripe -i $testdir/* | grep ^$max_index$ | wc -l) - min=$($LFS getdirstripe -i $testdir/* | grep ^$min_index$ | wc -l) + max=$($LFS getdirstripe -i $testdir/* | grep -c "^$max_index$") + min=$($LFS getdirstripe -i $testdir/* | grep -c "^$min_index$") # D-value should > 10% of averge - [ $((max - min)) -lt 10 ] && - error "subdirs shouldn't be evenly distributed" + (( max - min >= num / 10 )) || + error "subdirs shouldn't be evenly distributed: $max - $min < $((num / 10))" - # ditto - if [ $stripe_count -gt 1 ]; then + # 5% for stripes + if (( stripe_count > 1 )); then max=$($LFS getdirstripe $testdir/* | - grep -P "^\s+$max_index\t" | wc -l) + grep -c -P "^\s+$max_index\t") min=$($LFS getdirstripe $testdir/* | - grep -P "^\s+$min_index\t" | wc -l) - [ $((max - min)) -le $((10 * stripe_count)) ] && - error "stripes shouldn't be evenly distributed"|| true + grep -c -P "^\s+$min_index\t") + (( max - min >= num * stripe_count / 20 )) || + error "stripes shouldn't be evenly distributed: $max - $min < $((num / 20)) * $stripe_count" fi } +most_full_mdt() { + local ffree + local bavail + local bsize + local min + local min_index + local tmp + + ffree=($(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" @@ -23579,11 +25053,13 @@ 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'" @@ -23595,23 +25071,107 @@ test_413b() { [ $MDS1_VERSION -lt $(version_code 2.12.52) ] && skip "Need server version at least 2.12.52" + local testdir 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" - $LFS setdirstripe -D -c $stripe_count \ - $DIR/$tdir-s$stripe_count/rr || - error "setdirstripe failed" - $LFS setdirstripe -D -c $stripe_count \ - $DIR/$tdir-s$stripe_count/qos || + testdir=$DIR/$tdir-s$stripe_count + mkdir $testdir || error "mkdir $testdir failed" + mkdir $testdir/rr || error "mkdir rr 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 || error "setdirstripe failed" test_qos_mkdir "mkdir" $stripe_count done } run_test 413b "QoS mkdir under dir whose default LMV starting MDT offset is -1" +test_413c() { + (( $MDSCOUNT >= 2 )) || + skip "We need at least 2 MDTs for this test" + + (( $MDS1_VERSION >= $(version_code 2.14.51) )) || + skip "Need server version at least 2.14.51" + + local testdir + local inherit + local inherit_rr + + testdir=$DIR/${tdir}-s1 + mkdir $testdir || error "mkdir $testdir failed" + mkdir $testdir/rr || error "mkdir rr 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 -i -1 -X 2 --max-inherit-rr 1 $testdir/qos || + error "setdirstripe qos failed" + test_qos_mkdir "mkdir" 1 + + mkdir $testdir/rr/level1 || error "mkdir rr/level1 failed" + inherit=$($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" + + 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" + mkdir $testdir/qos/level1/level2 || error "mkdir level2 failed" + 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 $subdir/f. 100 & + pids="$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 @@ -23727,7 +25287,7 @@ check_lfs_df() { [ "$1" == "blocks" ] && inodes= || inodes="-i" for count in {1..100}; do - cancel_lru_locks + do_rpc_nodes "$CLIENTS" cancel_lru_locks sync; sleep 0.2 # read the lines of interest @@ -24019,26 +25579,24 @@ test_421f() { cnt=$(ls -1 $DIR/$tdir | wc -l) [ $cnt == 1 ] || error "unexpected #files after (5): $cnt" - umount_client $MOUNT || error "failed to umount client" - mount_client $MOUNT "$MOUNT_OPTS,user_fid2path" || + tmpdir=$(mktemp -d /tmp/lustre-XXXXXX) + stack_trap "rmdir $tmpdir" + mount_client $tmpdir "$MOUNT_OPTS,user_fid2path" || error "failed to mount client'" + stack_trap "umount_client $tmpdir" - $RUNAS $LFS rmfid $DIR $FID || error "rmfid failed" + $RUNAS $LFS rmfid $tmpdir $FID || error "rmfid failed" # rmfid should succeed - cnt=$(ls -1 $DIR/$tdir | wc -l) + cnt=$(ls -1 $tmpdir/$tdir | wc -l) [ $cnt == 0 ] || error "unexpected #files after (6): $cnt" # rmfid shouldn't allow to remove files due to dir's permission - chmod a+rwx $DIR/$tdir - touch $DIR/$tdir/f - ls -la $DIR/$tdir - FID=$(lfs path2fid $DIR/$tdir/f) - $RUNAS $LFS rmfid $DIR $FID && error "rmfid didn't fail" - - umount_client $MOUNT || error "failed to umount client" - mount_client $MOUNT "$MOUNT_OPTS" || - error "failed to mount client'" - + chmod a+rwx $tmpdir/$tdir + touch $tmpdir/$tdir/f + ls -la $tmpdir/$tdir + FID=$(lfs path2fid $tmpdir/$tdir/f) + $RUNAS $LFS rmfid $tmpdir $FID && error "rmfid didn't fail" + return 0 } run_test 421f "rmfid checks permissions" @@ -24222,6 +25780,78 @@ test_427() { } run_test 427 "Failed DNE2 update request shouldn't corrupt updatelog" +test_428() { + [ $PARALLEL == "yes" ] && skip "skip parallel run" + local cache_limit=$CACHE_MAX + + stack_trap "$LCTL set_param -n llite.*.max_cached_mb=$cache_limit" + $LCTL set_param -n llite.*.max_cached_mb=64 + + mkdir $DIR/$tdir + $LFS setstripe -c 1 $DIR/$tdir + eval touch $DIR/$tdir/$tfile.{1..$OSTCOUNT} + stack_trap "rm -f $DIR/$tdir/$tfile.*" + #test write + for f in $(seq 4); do + dd if=/dev/zero of=$DIR/$tdir/$tfile.$f bs=128M count=1 & + done + wait + + cancel_lru_locks osc + # Test read + for f in $(seq 4); do + dd if=$DIR/$tdir/$tfile.$f of=/dev/null bs=128M count=1 & + done + wait +} +run_test 428 "large block size IO should not hang" + +test_429() { # LU-7915 / LU-10948 + local ll_opencache_threshold_count="llite.*.opencache_threshold_count" + local testfile=$DIR/$tfile + local mdc_rpcstats="mdc.$FSNAME-MDT0000-*.stats" + local new_flag=1 + local first_rpc + local second_rpc + local third_rpc + + $LCTL get_param $ll_opencache_threshold_count || + skip "client does not have opencache parameter" + + set_opencache $new_flag + stack_trap "restore_opencache" + [ $($LCTL get_param -n $ll_opencache_threshold_count) == $new_flag ] || + error "enable opencache failed" + touch $testfile + # drop MDC DLM locks + cancel_lru_locks mdc + # clear MDC RPC stats counters + $LCTL set_param $mdc_rpcstats=clear + + # According to the current implementation, we need to run 3 times + # open & close file to verify if opencache is enabled correctly. + # 1st, RPCs are sent for lookup/open and open handle is released on + # close finally. + # 2nd, RPC is sent for open, MDS_OPEN_LOCK is fetched automatically, + # so open handle won't be released thereafter. + # 3rd, No RPC is sent out. + $MULTIOP $testfile oc || error "multiop failed" + first_rpc=$(calc_stats $mdc_rpcstats ldlm_ibits_enqueue) + echo "1st: $first_rpc RPCs in flight" + + $MULTIOP $testfile oc || error "multiop failed" + second_rpc=$(calc_stats $mdc_rpcstats ldlm_ibits_enqueue) + echo "2nd: $second_rpc RPCs in flight" + + $MULTIOP $testfile oc || error "multiop failed" + third_rpc=$(calc_stats $mdc_rpcstats ldlm_ibits_enqueue) + echo "3rd: $third_rpc RPCs in flight" + + #verify no MDC RPC is sent + [[ $second_rpc == $third_rpc ]] || error "MDC RPC is still sent" +} +run_test 429 "verify if opencache flag on client side does work" + lseek_test_430() { local offset local file=$1 @@ -24469,6 +26099,38 @@ 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) ]] && @@ -24782,7 +26444,7 @@ 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)) \ @@ -24977,7 +26639,7 @@ 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" @@ -24991,11 +26653,9 @@ run_test 805 "ZFS can remove from full fs" check_lsom_data() { local file=$1 - local size=$($LFS getsom -s $file) local expect=$(stat -c %s $file) - [[ $size == $expect ]] || - error "$file expected size: $expect, got: $size" + check_lsom_size $1 $expect local blocks=$($LFS getsom -b $file) expect=$(stat -c %b $file) @@ -25005,9 +26665,12 @@ check_lsom_data() check_lsom_size() { - local size=$($LFS getsom -s $1) + local size local expect=$2 + cancel_lru_locks mdc + + size=$($LFS getsom -s $1) [[ $size == $expect ]] || error "$file expected size: $expect, got: $size" } @@ -25124,7 +26787,7 @@ 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 || @@ -25296,6 +26959,28 @@ test_812b() { # LU-12378 } run_test 812b "do not drop no resend request for idle connect" +test_812c() { + local old + + old=$($LCTL get_param -n osc.*.idle_timeout | head -n 1) + + $LFS setstripe -c 1 -o 0 $DIR/$tfile + $LFS getstripe $DIR/$tfile + $LCTL set_param osc.*.idle_timeout=10 + stack_trap "$LCTL set_param osc.*.idle_timeout=$old" EXIT + # ensure ost1 is connected + stat $DIR/$tfile >/dev/null || error "can't stat" + wait_osc_import_state client ost1 FULL + # no locks, no reqs to let the connection idle + cancel_lru_locks osc + +#define OBD_FAIL_PTLRPC_IDLE_RACE 0x533 + $LCTL set_param fail_loc=0x80000533 + sleep 15 + dd if=/dev/zero of=$DIR/$tfile count=1 conv=sync || error "dd failed" +} +run_test 812c "idle import vs lock enqueue race" + test_813() { local file_heat_sav=$($LCTL get_param -n llite.*.file_heat 2>/dev/null) [ -z "$file_heat_sav" ] && skip "no file heat support" @@ -25695,6 +27380,21 @@ 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