X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftests%2Fsanity.sh;h=49afa477de8b896c6035ab2705c3c7bbcf888b43;hp=f1732e581b5e8116e7b49564c06080a8ba6349e5;hb=c0a7f78529e21c9cafa986abea255925b4b41244;hpb=b57c88fe4b70392ca915c18e51d50ce90256ba69 diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh index f1732e5..49afa47 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 (min)" +[ "$SLOW" = "no" ] && EXCEPT_SLOW="27m 60i 64b 68 71 115 135 136 300o" if [ "$mds1_FSTYPE" = "zfs" ]; then # bug number for skipped test: @@ -92,6 +92,8 @@ if [ "$ost1_FSTYPE" = "zfs" ]; then ALWAYS_EXCEPT+=" 130a 130b 130c 130d 130e 130f 130g" fi +proc_regexp="/{proc,sys}/{fs,sys,kernel/debug}/{lustre,lnet}/" + # Get the SLES distro version # # Returns a version string that should only be used in comparing @@ -162,7 +164,7 @@ check_and_setup_lustre DIR=${DIR:-$MOUNT} assert_DIR -MAXFREE=${MAXFREE:-$((200000 * $OSTCOUNT))} +MAXFREE=${MAXFREE:-$((300000 * $OSTCOUNT))} [ -f $DIR/d52a/foo ] && chattr -a $DIR/d52a/foo [ -f $DIR/d52b/foo ] && chattr -i $DIR/d52b/foo @@ -251,6 +253,35 @@ test_0d() { # LU-3397 } run_test 0d "check export proc =============================" +test_0e() { # LU-13417 + (( $MDSCOUNT > 1 )) || + skip "We need at least 2 MDTs for this test" + + (( $MDS1_VERSION >= $(version_code 2.14.51) )) || + skip "Need server version at least 2.14.51" + + local default_lmv_count=$($LFS getdirstripe -D -c $MOUNT) + local default_lmv_index=$($LFS getdirstripe -D -i $MOUNT) + + [ $default_lmv_count -eq 1 ] || + error "$MOUNT default stripe count $default_lmv_count" + + [ $default_lmv_index -eq -1 ] || + error "$MOUNT default stripe index $default_lmv_index" + + mkdir $MOUNT/$tdir.1 || error "mkdir $MOUNT/$tdir.1 failed" + mkdir $MOUNT/$tdir.2 || error "mkdir $MOUNT/$tdir.2 failed" + + local mdt_index1=$($LFS getdirstripe -i $MOUNT/$tdir.1) + local mdt_index2=$($LFS getdirstripe -i $MOUNT/$tdir.2) + + [ $mdt_index1 -eq $mdt_index2 ] && + error "directories are on the same MDT $mdt_index1=$mdt_index2" + + rmdir $MOUNT/$tdir.1 $MOUNT/$tdir.2 +} +run_test 0e "Enable DNE MDT balancing for mkdir in the ROOT" + test_1() { test_mkdir $DIR/$tdir test_mkdir $DIR/$tdir/d2 @@ -1835,7 +1866,7 @@ __exhaust_precreations() { local FAILIDX=${3:-$OSTIDX} local ofacet=ost$((OSTIDX + 1)) - test_mkdir -p -c1 $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir local mdtidx=$($LFS getstripe -m $DIR/$tdir) local mfacet=mds$((mdtidx + 1)) echo OSTIDX=$OSTIDX MDTIDX=$mdtidx @@ -1991,7 +2022,7 @@ test_27q() { reset_enospc rm -f $DIR/$tdir/$tfile - test_mkdir $DIR/$tdir + mkdir_on_mdt0 $DIR/$tdir $MCREATE $DIR/$tdir/$tfile || error "mcreate $DIR/$tdir/$tfile failed" $TRUNCATE $DIR/$tdir/$tfile 80000000 || error "truncate $DIR/$tdir/$tfile failed" @@ -2755,7 +2786,8 @@ test_27I() { save_layout_restore_at_exit $MOUNT $LFS setstripe -c 2 -i 0 $MOUNT pool_add $pool || error "pool_add failed" - pool_add_targets $pool $ostrange || "pool_add_targets failed" + pool_add_targets $pool $ostrange || + error "pool_add_targets failed" test_mkdir $DIR/$tdir $LFS setstripe -p $pool $DIR/$tdir $MULTIOP $DIR/$tdir/$tfile Oc || error "multiop failed" @@ -2984,11 +3016,23 @@ test_27M() { test_mkdir $DIR/$tdir # Set default striping on directory - $LFS setstripe -C 4 $DIR/$tdir + local setcount=4 + local stripe_opt + + # if we run against a 2.12 server which lacks overstring support + # then the connect_flag will not report overstriping, even if client + # is 2.14+ + if [[ $($LCTL get_param mdc.*.connect_flags) =~ overstriping ]]; then + stripe_opt="-C $setcount" + elif (( $OSTCOUNT >= $setcount )); then + stripe_opt="-c $setcount" + else + skip "server does not support overstriping" + fi + $LFS setstripe $stripe_opt $DIR/$tdir echo 1 > $DIR/$tdir/${tfile}.1 local count=$($LFS getstripe -c $DIR/$tdir/${tfile}.1) - local setcount=4 [ $count -eq $setcount ] || error "(1) stripe count $count, should be $setcount" @@ -3054,11 +3098,11 @@ test_27M() { # Clean up DOM layout $LFS setstripe -d $DIR/$tdir + save_layout_restore_at_exit $MOUNT # Now test that append striping works when layout is from root $LFS setstripe -c 2 $MOUNT # Make a special directory for this mkdir $DIR/${tdir}/${tdir}.2 - stack_trap "$LFS setstripe -d $MOUNT" EXIT # Verify for normal file setcount=2 @@ -3321,6 +3365,7 @@ 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 @@ -3676,6 +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 @@ -4765,7 +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 @@ -5557,7 +5624,7 @@ test_51b() { # cleanup the directory rm -fr $dir - test_mkdir -c1 $dir + mkdir_on_mdt -i $((RANDOM % MDSCOUNT)) $dir $LFS df $LFS df -i @@ -6664,6 +6731,36 @@ test_56rb() { } run_test 56rb "check lfs find --size --ost/--mdt works" +test_56rc() { + (( MDSCOUNT >= 2 )) || skip "needs at least 2 MDTs" + local dir=$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 [[ $OSTCOUNT -lt 2 ]] && skip_env "need at least 2 OSTs" @@ -7644,6 +7741,94 @@ test_56ab() { # LU-10705 } run_test 56ab "lfs find --blocks" +# LU-11188 +test_56aca() { + local dir="$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" @@ -8179,6 +8364,34 @@ 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() { [ $PARALLEL == "yes" ] && skip "skip parallel run" @@ -8732,8 +8945,7 @@ test_65n() { which getfattr > /dev/null 2>&1 || skip_env "no getfattr command" which setfattr > /dev/null 2>&1 || skip_env "no setfattr command" - local root_layout=$(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 @@ -9302,6 +9514,7 @@ 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 @@ -9326,6 +9539,7 @@ 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 @@ -9447,6 +9661,65 @@ 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 @@ -13669,8 +13942,12 @@ 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 $proc_dirs -name | + xargs -n 1 badarea_io || error "client badarea_io failed" # remount the FS in case writes/reads /proc break the FS @@ -13683,6 +13960,8 @@ test_133g() { remote_mds_nodsh && skip "remote MDS with nodsh" remote_ost_nodsh && skip "remote OST with nodsh" + local proc_dirs=$(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) @@ -13693,8 +13972,9 @@ 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" @@ -13738,7 +14018,7 @@ 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-*" @@ -13774,7 +14054,7 @@ 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 \ @@ -13983,31 +14263,30 @@ 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 < $want * 105 / 100 )) || @@ -14017,13 +14296,14 @@ run_test 150c "Verify fallocate Size and Blocks" test_150d() { check_set_fallocate_or_skip + local striping="-c2" + + [[ "x$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)) @@ -14100,6 +14380,9 @@ test_150f() { check_set_fallocate_or_skip stack_trap "rm -f $DIR/$tfile; wait_delete_completed" + [[ "x$DOM" == "xyes" ]] && + $LFS setstripe -E1M -L mdt -E eof $DIR/$tfile + echo "Verify fallocate punch: Range within the file range" yes 'A' | dd of=$DIR/$tfile bs=4096 count=5 || error "dd failed for bs 4096 and count 5" @@ -14175,8 +14458,13 @@ test_150g() { check_set_fallocate_or_skip stack_trap "rm -f $DIR/$tfile; wait_delete_completed" - $LFS setstripe -c${OSTCOUNT} $DIR/$tfile || - error "$LFS setstripe -c${OSTCOUNT} $DIR/$tfile failed" + if [[ "x$DOM" == "xyes" ]]; then + $LFS setstripe -E2M -L mdt -E eof -c${OSTCOUNT} $DIR/$tfile || + error "$LFS setstripe DoM + ${OSTCOUNT} OST failed" + else + $LFS setstripe -c${OSTCOUNT} $DIR/$tfile || + error "$LFS setstripe -c${OSTCOUNT} $DIR/$tfile failed" + fi # Get 100MB per OST of the available space to reduce run time # else 60% of the available space if we are running SLOW tests @@ -14702,35 +14990,35 @@ 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)" @@ -14752,7 +15040,7 @@ 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) @@ -14764,7 +15052,7 @@ 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" && @@ -14784,7 +15072,7 @@ 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" @@ -15135,6 +15423,8 @@ 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" @@ -15145,8 +15435,6 @@ 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" @@ -15160,7 +15448,6 @@ 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 $MKDIRS != 1" @@ -15223,10 +15510,10 @@ test_160a() { error "User '$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 $SINGLEMDS | @@ -15322,11 +15609,21 @@ test_160e() { # Create a user changelog_register || error "changelog_register failed" - # Delete a future user (expect fail) local MDT0=$(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 @@ -15994,6 +16291,157 @@ test_160m() { } 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" + # drop server mask so it doesn't interfere + 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_161a() { [ $PARALLEL == "yes" ] && skip "skip parallel run" @@ -16956,7 +17404,7 @@ 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 @@ -17548,7 +17996,7 @@ 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" @@ -17589,7 +18037,7 @@ 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" @@ -17642,6 +18090,9 @@ 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 @@ -17727,23 +18178,23 @@ 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 @@ -17752,7 +18203,7 @@ 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 @@ -18080,7 +18531,7 @@ 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} @@ -18960,7 +19411,7 @@ 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 @@ -19297,15 +19748,15 @@ 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 @@ -19319,33 +19770,37 @@ 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)) @@ -19405,6 +19860,15 @@ 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" @@ -19812,7 +20276,7 @@ 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 @@ -20004,10 +20468,11 @@ 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" @@ -20531,6 +20996,7 @@ 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 && @@ -20618,6 +21084,7 @@ 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 && @@ -20780,7 +21247,7 @@ 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" @@ -20931,7 +21398,7 @@ 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 && @@ -22070,7 +22537,7 @@ 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" @@ -22280,7 +22747,7 @@ 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 @@ -22316,9 +22783,6 @@ 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" @@ -22620,7 +23084,7 @@ 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 @@ -23325,6 +23789,275 @@ test_398f() { # LU-14687 } run_test 398f "verify aio handles ll_direct_rw_pages errors correctly" +# NB: To get the parallel DIO behavior in LU-13798, there must be > 1 +# stripe and i/o size must be > stripe size +# Old style synchronous DIO waits after submitting each chunk, resulting in a +# single RPC in flight. This test shows async DIO submission is working by +# showing multiple RPCs in flight. +test_398g() { # LU-13798 + $LFS setstripe -o 0,0 -S 1M $DIR/$tfile + + # We need to do some i/o first to acquire enough grant to put our RPCs + # in flight; otherwise a new connection may not have enough grant + # available + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=1 oflag=direct || + error "parallel dio failed" + stack_trap "rm -f $DIR/$tfile" + + # Reduce RPC size to 1M to avoid combination in to larger RPCs + local pages_per_rpc=$($LCTL get_param osc.*-OST0000-*.max_pages_per_rpc) + $LCTL set_param osc.*-OST0000-*.max_pages_per_rpc=1M + stack_trap "$LCTL set_param -n $pages_per_rpc" + + # Recreate file so it's empty + rm -f $DIR/$tfile + $LFS setstripe -o 0,0 -S 1M $DIR/$tfile + #Pause rpc completion to guarantee we see multiple rpcs in flight + #define OBD_FAIL_OST_BRW_PAUSE_BULK + do_facet ost1 $LCTL set_param fail_loc=0x214 fail_val=2 + stack_trap "do_facet ost1 $LCTL set_param fail_loc=0" + + # Clear rpc stats + $LCTL set_param osc.*.rpc_stats=c + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=1 oflag=direct || + error "parallel dio failed" + stack_trap "rm -f $DIR/$tfile" + + $LCTL get_param osc.*-OST0000-*.rpc_stats + pct=$($LCTL get_param osc.*-OST0000-*.rpc_stats | + grep -A 8 'rpcs in flight' | grep -v 'rpcs in flight' | + grep "8:" | awk '{print $8}') + # We look at the "8 rpcs in flight" field, and verify A) it is present + # and B) it includes all RPCs. This proves we had 8 RPCs in flight, + # as expected for an 8M DIO to a file with 1M stripes. + [ $pct -eq 100 ] || error "we should see 8 RPCs in flight" + + # Verify turning off parallel dio works as expected + # Clear rpc stats + $LCTL set_param osc.*.rpc_stats=c + $LCTL set_param llite.*.parallel_dio=0 + stack_trap '$LCTL set_param llite.*.parallel_dio=1' + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=1 oflag=direct || + error "dio with parallel dio disabled failed" + + # Ideally, we would see only one RPC in flight here, but there is an + # unavoidable race between i/o completion and RPC in flight counting, + # so while only 1 i/o is in flight at a time, the RPC in flight counter + # will sometimes exceed 1 (3 or 4 is not rare on VM testing). + # So instead we just verify it's always < 8. + $LCTL get_param osc.*-OST0000-*.rpc_stats + ret=$($LCTL get_param osc.*-OST0000-*.rpc_stats | + grep -A 8 'rpcs in flight' | grep -v 'rpcs in flight' | + grep '^$' -B1 | grep . | awk '{print $1}') + [ $ret != "8:" ] || + error "we should see fewer than 8 RPCs in flight (saw $ret)" +} +run_test 398g "verify parallel dio async RPC submission" + +test_398h() { # LU-13798 + local dio_file=$DIR/$tfile.dio + + $LFS setstripe -C 2 -S 1M $DIR/$tfile $dio_file + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct + stack_trap "rm -f $DIR/$tfile $dio_file" + + dd if=$DIR/$tfile of=$dio_file bs=8M count=8 iflag=direct oflag=direct || + error "parallel dio failed" + diff $DIR/$tfile $dio_file + [[ $? == 0 ]] || error "file diff after aiocp" +} +run_test 398h "verify correctness of read & write with i/o size >> stripe size" + +test_398i() { # LU-13798 + local dio_file=$DIR/$tfile.dio + + $LFS setstripe -C 2 -S 1M $DIR/$tfile $dio_file + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct + stack_trap "rm -f $DIR/$tfile $dio_file" + + #define OBD_FAIL_LLITE_PAGE_ALLOC 0x1418 + $LCTL set_param fail_loc=0x1418 + # make sure we don't crash and fail properly + dd if=$DIR/$tfile of=$dio_file bs=8M count=8 iflag=direct oflag=direct && + error "parallel dio page allocation failure succeeded" + diff $DIR/$tfile $dio_file + [[ $? != 0 ]] || error "no diff after failed aiocp" +} +run_test 398i "verify parallel dio handles ll_direct_rw_pages errors correctly" + +test_398j() { # LU-13798 + # Stripe size > RPC size but less than i/o size tests split across + # stripes and RPCs for individual i/o op + $LFS setstripe -o 0,0 -S 4M $DIR/$tfile $DIR/$tfile.2 + + # Reduce RPC size to 1M to guarantee split to multiple RPCs per stripe + local pages_per_rpc=$($LCTL get_param osc.*-OST0000-*.max_pages_per_rpc) + $LCTL set_param osc.*-OST0000-*.max_pages_per_rpc=1M + stack_trap "$LCTL set_param -n $pages_per_rpc" + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct || + error "parallel dio write failed" + stack_trap "rm -f $DIR/$tfile $DIR/$tfile.2" + + dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 iflag=direct || + error "parallel dio read failed" + diff $DIR/$tfile $DIR/$tfile.2 + [[ $? == 0 ]] || error "file diff after parallel dio read" +} +run_test 398j "test parallel dio where stripe size > rpc_size" + +test_398k() { # LU-13798 + wait_delete_completed + wait_mds_ost_sync + + # 4 stripe file; we will cause out of space on OST0 + $LFS setstripe -o 0,1,0,1 -S 1M $DIR/$tfile + + # Fill OST0 (if it's not too large) + ORIGFREE=$($LCTL get_param -n lov.$FSNAME-clilov-*.kbytesavail | + head -n1) + if [[ $ORIGFREE -gt $MAXFREE ]]; then + skip "$ORIGFREE > $MAXFREE skipping out-of-space test on OST0" + fi + $LFS setstripe -i 0 -c 1 $DIR/$tfile.1 + dd if=/dev/zero of=$DIR/$tfile.1 bs=1024 count=$MAXFREE && + error "dd should fill OST0" + stack_trap "rm -f $DIR/$tfile.1" + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct + err=$? + + ls -la $DIR/$tfile + $CHECKSTAT -t file -s 0 $DIR/$tfile || + error "file is not 0 bytes in size" + + # dd above should not succeed, but don't error until here so we can + # get debug info above + [[ $err != 0 ]] || + error "parallel dio write with enospc succeeded" + stack_trap "rm -f $DIR/$tfile" +} +run_test 398k "test enospc on first stripe" + +test_398l() { # LU-13798 + wait_delete_completed + wait_mds_ost_sync + + # 4 stripe file; we will cause out of space on OST0 + # Note the 1M stripe size and the > 1M i/o size mean this ENOSPC + # happens on the second i/o chunk we issue + $LFS setstripe -o 1,0,1,0 -S 1M $DIR/$tfile $DIR/$tfile.2 + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=2 oflag=direct + stack_trap "rm -f $DIR/$tfile" + + # Fill OST0 (if it's not too large) + ORIGFREE=$($LCTL get_param -n lov.$FSNAME-clilov-*.kbytesavail | + head -n1) + if [[ $ORIGFREE -gt $MAXFREE ]]; then + skip "$ORIGFREE > $MAXFREE skipping out-of-space test on OST0" + fi + $LFS setstripe -i 0 -c 1 $DIR/$tfile.1 + dd if=/dev/zero of=$DIR/$tfile.1 bs=1024 count=$MAXFREE && + error "dd should fill OST0" + stack_trap "rm -f $DIR/$tfile.1" + + dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 oflag=direct + err=$? + stack_trap "rm -f $DIR/$tfile.2" + + # Check that short write completed as expected + ls -la $DIR/$tfile.2 + $CHECKSTAT -t file -s 1048576 $DIR/$tfile.2 || + error "file is not 1M in size" + + # dd above should not succeed, but don't error until here so we can + # get debug info above + [[ $err != 0 ]] || + error "parallel dio write with enospc succeeded" + + # Truncate source file to same length as output file and diff them + $TRUNCATE $DIR/$tfile 1048576 + diff $DIR/$tfile $DIR/$tfile.2 + [[ $? == 0 ]] || error "data incorrect after short write" +} +run_test 398l "test enospc on intermediate stripe/RPC" + +test_398m() { # LU-13798 + $LFS setstripe -o 0,1,0,1 -S 1M $DIR/$tfile + + # Set up failure on OST0, the first stripe: + #define OBD_FAIL_OST_BRW_WRITE_BULK 0x20e + #NB: Fail val is ost # + 1, because we cannot use cfs_fail_val = 0 + # So this fail_val specifies OST0 + do_facet ost1 $LCTL set_param fail_loc=0x20e fail_val=1 + stack_trap "do_facet ost1 $LCTL set_param fail_loc=0" + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct && + error "parallel dio write with failure on first stripe succeeded" + stack_trap "rm -f $DIR/$tfile" + do_facet ost1 $LCTL set_param fail_loc=0 fail_val=0 + + # Place data in file for read + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct || + error "parallel dio write failed" + + # Fail read on OST0, first stripe + #define OBD_FAIL_OST_BRW_READ_BULK 0x20f + do_facet ost1 $LCTL set_param fail_loc=0x20f fail_val=1 + dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 iflag=direct && + error "parallel dio read with error on first stripe succeeded" + rm -f $DIR/$tfile.2 + do_facet ost1 $LCTL set_param fail_loc=0 fail_val=0 + + # Switch to testing on OST1, second stripe + # Clear file contents, maintain striping + echo > $DIR/$tfile + # Set up failure on OST1, second stripe: + do_facet ost1 $LCTL set_param fail_loc=0x20e fail_val=2 + stack_trap "do_facet ost1 $LCTL set_param fail_loc=0" + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct && + error "parallel dio write with failure on first stripe succeeded" + stack_trap "rm -f $DIR/$tfile" + do_facet ost1 $LCTL set_param fail_loc=0 fail_val=0 + + # Place data in file for read + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 oflag=direct || + error "parallel dio write failed" + + # Fail read on OST1, second stripe + #define OBD_FAIL_OST_BRW_READ_BULK 0x20f + do_facet ost2 $LCTL set_param fail_loc=0x20f fail_val=2 + dd if=$DIR/$tfile of=$DIR/$tfile.2 bs=8M count=8 iflag=direct && + error "parallel dio read with error on first stripe succeeded" + rm -f $DIR/$tfile.2 + do_facet ost2 $LCTL set_param fail_loc=0 fail_val=0 +} +run_test 398m "test RPC failures with parallel dio" + +# Parallel submission of DIO should not cause problems for append, but it's +# important to verify. +test_398n() { # LU-13798 + $LFS setstripe -C 2 -S 1M $DIR/$tfile + + dd if=/dev/urandom of=$DIR/$tfile bs=8M count=8 || + error "dd to create source file failed" + stack_trap "rm -f $DIR/$tfile" + + dd if=$DIR/$tfile of=$DIR/$tfile.1 bs=8M count=8 oflag=direct oflag=append || + error "parallel dio write with failure on second stripe succeeded" + stack_trap "rm -f $DIR/$tfile $DIR/$tfile.1" + diff $DIR/$tfile $DIR/$tfile.1 + [[ $? == 0 ]] || error "data incorrect after append" + +} +run_test 398n "test append with parallel DIO" + test_fake_rw() { local read_write=$1 if [ "$read_write" = "write" ]; then @@ -23467,7 +24200,6 @@ 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" @@ -23578,6 +24310,15 @@ 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) && @@ -23884,6 +24625,76 @@ 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=$2 @@ -23906,11 +24717,11 @@ test_qos_mkdir() { lmv_qos_threshold_rr=${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) @@ -23921,12 +24732,11 @@ test_qos_mkdir() { lod_qos_maxage=$(do_facet mds1 $LCTL get_param -n \ lod.$FSNAME-MDT0000-mdtlov.qos_maxage | awk '{ print $1 }') stack_trap "do_nodes $mdts $LCTL set_param \ - lod.*.mdt_qos_prio_free=$lod_qos_prio_free > /dev/null" EXIT + lod.*.mdt_qos_prio_free=$lod_qos_prio_free > /dev/null" stack_trap "do_nodes $mdts $LCTL set_param \ - lod.*.mdt_qos_threshold_rr=$lod_qos_threshold_rr > /dev/null" \ - EXIT + lod.*.mdt_qos_threshold_rr=$lod_qos_threshold_rr > /dev/null" stack_trap "do_nodes $mdts $LCTL set_param \ - lod.*.mdt_qos_maxage=$lod_qos_maxage > /dev/null" EXIT + lod.*.mdt_qos_maxage=$lod_qos_maxage > /dev/null" $LCTL set_param lmv.*.qos_threshold_rr=100 > /dev/null do_nodes $mdts $LCTL set_param lod.*.mdt_qos_threshold_rr=100 > /dev/null @@ -23936,11 +24746,12 @@ test_qos_mkdir() { local stripe_index=$($LFS getstripe -m $testdir) local test_mkdir_rr=true - getfattr -d -m dmv $testdir | grep dmv - if [ $? -eq 0 ] && [ $MDS1_VERSION -ge $(version_code 2.14.51) ]; then - local inherit_rr=$($LFS getdirstripe -D --max-inherit-rr $testdir) - - (( $inherit_rr == 0 )) && test_mkdir_rr=false + echo "dirstripe: '$($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 @@ -23948,41 +24759,37 @@ test_qos_mkdir() { echo "Mkdir (stripe_count $stripe_count) roundrobin:" || echo "Mkdir (stripe_count $stripe_count) on stripe $stripe_index" - for i in $(seq $((100 * MDSCOUNT))); do + stack_trap "unlinkmany -d $testdir/subdir $((100 * MDSCOUNT))" + for (( i = 0; i < 100 * MDSCOUNT; i++ )); do eval $mkdir_cmd $testdir/subdir$i || error "$mkdir_cmd subdir$i failed" done - for i in $(seq $MDSCOUNT); do - count=$($LFS getdirstripe -i $testdir/* | - grep ^$((i - 1))$ | wc -l) - echo "$count directories created on MDT$((i - 1))" + for (( i = 0; i < $MDSCOUNT; i++ )); do + count=$($LFS getdirstripe -i $testdir/* | grep -c "^$i$") + echo "$count directories created on MDT$i" if $test_mkdir_rr; then (( $count == 100 )) || error "subdirs are not evenly distributed" - elif [ $((i - 1)) -eq $stripe_index ]; then + elif (( $i == $stripe_index )); then (( $count == 100 * MDSCOUNT )) || - error "$count subdirs created on MDT$((i - 1))" + error "$count subdirs created on MDT$i" else (( $count == 0 )) || - error "$count subdirs created on MDT$((i - 1))" + error "$count subdirs created on MDT$i" fi if $test_mkdir_rr && [ $stripe_count -gt 1 ]; then count=$($LFS getdirstripe $testdir/* | - grep -P "^\s+$((i - 1))\t" | wc -l) - echo "$count stripes created on MDT$((i - 1))" + grep -c -P "^\s+$i\t") + echo "$count stripes created on MDT$i" # deviation should < 5% of average - (( $count < 95 * stripe_count )) || - (( $count > 105 * stripe_count)) && + (( $count >= 95 * stripe_count && + $count <= 105 * stripe_count)) || error "stripes are not evenly distributed" fi done - $LCTL set_param lmv.*.qos_threshold_rr=$lmv_qos_threshold_rr > /dev/null - do_nodes $mdts $LCTL set_param \ - lod.*.mdt_qos_threshold_rr=$lod_qos_threshold_rr > /dev/null - echo echo "Check for uneven MDTs: " @@ -24014,47 +24821,20 @@ test_qos_mkdir() { fi done - (( ${ffree[min_index]} == 0 )) && + (( ${ffree[min_index]} > 0 )) || skip "no free files in MDT$min_index" - (( ${ffree[min_index]} > 100000000 )) && + (( ${ffree[min_index]} < 100000000 )) || skip "too many free files in MDT$min_index" - # Check if we need to generate uneven MDTs - local threshold=50 - local diff=$(((max - min) * 100 / min)) - local value="$(generate_string 1024)" - - while [ $diff -lt $threshold ]; do - # generate uneven MDTs, create till $threshold% diff - echo -n "weight diff=$diff% must be > $threshold% ..." - count=$((${ffree[min_index]} / 10)) - # 50 sec per 10000 files in vm - (( $count < 100000 )) || [ "$SLOW" != "no" ] || - skip "$count files to create" - echo "Fill MDT$min_index with $count files" - [ -d $DIR/$tdir-MDT$min_index ] || - $LFS mkdir -i $min_index $DIR/$tdir-MDT$min_index || - error "mkdir $tdir-MDT$min_index failed" - createmany -d $DIR/$tdir-MDT$min_index/d $count || - error "create d$count failed" - - ffree=($(lctl get_param -n mdc.*[mM][dD][cC]-*.filesfree)) - bavail=($(lctl get_param -n mdc.*[mM][dD][cC]-*.kbytesavail)) - max=$(((${ffree[max_index]} >> 8) * \ - (${bavail[max_index]} * bsize >> 16))) - min=$(((${ffree[min_index]} >> 8) * \ - (${bavail[min_index]} * bsize >> 16))) - diff=$(((max - min) * 100 / min)) - done - echo "MDT filesfree available: ${ffree[@]}" echo "MDT blocks available: ${bavail[@]}" - echo "weight diff=$diff%" - + echo "weight diff=$(((max - min) * 100 / min))%" echo echo "Mkdir (stripe_count $stripe_count) with balanced space usage:" + $LCTL set_param lmv.*.qos_threshold_rr=0 > /dev/null $LCTL set_param lmv.*.qos_prio_free=100 > /dev/null + do_nodes $mdts $LCTL set_param lod.*.mdt_qos_threshold_rr=0 > /dev/null do_nodes $mdts $LCTL set_param lod.*.mdt_qos_prio_free=100 > /dev/null # decrease statfs age, so that it can be updated in time $LCTL set_param lmv.*.qos_maxage=1 > /dev/null @@ -24063,42 +24843,65 @@ test_qos_mkdir() { sleep 1 testdir=$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 < 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 < 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" @@ -24108,11 +24911,13 @@ 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'" @@ -24127,11 +24932,13 @@ test_413b() { local testdir local stripe_count + generate_uneven_mdts 100 for stripe_count in $(seq 1 $((MDSCOUNT - 1))); do testdir=$DIR/$tdir-s$stripe_count mkdir $testdir || error "mkdir $testdir failed" mkdir $testdir/rr || error "mkdir rr failed" - mkdir $testdir/qos || error "mkdir qos failed" + $LFS mkdir -i $(most_full_mdt) $testdir/qos || + error "mkdir qos failed" $LFS setdirstripe -D -c $stripe_count --max-inherit-rr 2 \ $testdir/rr || error "setdirstripe rr failed" $LFS setdirstripe -D -c $stripe_count $testdir/qos || @@ -24142,11 +24949,11 @@ test_413b() { run_test 413b "QoS mkdir under dir whose default LMV starting MDT offset is -1" test_413c() { - [ $MDSCOUNT -ge 2 ] || + (( $MDSCOUNT >= 2 )) || skip "We need at least 2 MDTs for this test" - [ $MDS1_VERSION -ge $(version_code 2.14.51) ] || - skip "Need server version at least 2.14.50" + (( $MDS1_VERSION >= $(version_code 2.14.51) )) || + skip "Need server version at least 2.14.51" local testdir local inherit @@ -24155,11 +24962,11 @@ test_413c() { testdir=$DIR/${tdir}-s1 mkdir $testdir || error "mkdir $testdir failed" mkdir $testdir/rr || error "mkdir rr failed" - mkdir $testdir/qos || error "mkdir qos failed" + $LFS mkdir -i $(most_full_mdt) $testdir/qos || error "mkdir qos failed" # default max_inherit is -1, default max_inherit_rr is 0 $LFS setdirstripe -D -c 1 $testdir/rr || error "setdirstripe rr failed" - $LFS setdirstripe -D -c 1 -X 2 --max-inherit-rr 1 $testdir/qos || + $LFS setdirstripe -D -c 1 -i -1 -X 2 --max-inherit-rr 1 $testdir/qos || error "setdirstripe qos failed" test_qos_mkdir "mkdir" 1 @@ -24167,21 +24974,62 @@ test_413c() { inherit=$($LFS getdirstripe -D -X $testdir/rr/level1) (( $inherit == -1 )) || error "rr/level1 inherit $inherit != -1" inherit_rr=$($LFS getdirstripe -D --max-inherit-rr $testdir/rr/level1) - (( $inherit_rr == 0 )) || - error "rr/level1 inherit-rr $inherit_rr != 0" + (( $inherit_rr == 0 )) || error "rr/level1 inherit-rr $inherit_rr != 0" mkdir $testdir/qos/level1 || error "mkdir qos/level1 failed" inherit=$($LFS getdirstripe -D -X $testdir/qos/level1) (( $inherit == 1 )) || error "qos/level1 inherit $inherit != 1" inherit_rr=$($LFS getdirstripe -D --max-inherit-rr $testdir/qos/level1) - (( $inherit_rr == 0 )) || - error "qos/level1 inherit-rr $inherit_rr !=0" + (( $inherit_rr == 0 )) || error "qos/level1 inherit-rr $inherit_rr != 0" mkdir $testdir/qos/level1/level2 || error "mkdir level2 failed" - getfattr -d -m dmv $testdir/qos/level1/level2 | grep dmv && + getfattr -d -m dmv -e hex $testdir/qos/level1/level2 | grep dmv && error "level2 shouldn't have default LMV" || true } run_test 413c "mkdir with default LMV max inherit rr" +test_413d() { + (( MDSCOUNT >= 2 )) || + skip "We need at least 2 MDTs for this test" + + (( MDS1_VERSION >= $(version_code 2.14.51) )) || + skip "Need server version at least 2.14.51" + + local lmv_qos_threshold_rr + + lmv_qos_threshold_rr=$($LCTL get_param -n lmv.*.qos_threshold_rr | + head -n1) + stack_trap "$LCTL set_param \ + lmv.*.qos_threshold_rr=$lmv_qos_threshold_rr > /dev/null" EXIT + + $LCTL set_param lmv.*.qos_threshold_rr=100 > /dev/null + mkdir -p $DIR/$tdir || error "mkdir $tdir failed" + getfattr -d -m dmv -e hex $DIR/$tdir | grep dmv && + error "$tdir shouldn't have default LMV" + createmany -d $DIR/$tdir/sub $((100 * MDSCOUNT)) || + error "mkdir sub failed" + + local count=$($LFS getstripe -m $DIR/$tdir/* | grep -c ^0) + + (( count == 100 )) || error "$count subdirs on MDT0" +} +run_test 413d "inherit ROOT default LMV" + +test_413z() { + local pids="" + local subdir + local pid + + for subdir in $(\ls -1 -d $DIR/d413*-fillmdt/*); do + unlinkmany $subdir/f. 100 & + pids="$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 @@ -25109,6 +25957,38 @@ 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) ]] && @@ -25422,7 +26302,7 @@ 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)) \ @@ -25617,7 +26497,7 @@ 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" @@ -25765,7 +26645,7 @@ 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 || @@ -26358,6 +27238,21 @@ 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