3 # Run select tests by setting ONLY, or as arguments to the script.
4 # Skip specific tests by setting EXCEPT.
12 LUSTRE=${LUSTRE:-$(dirname $0)/..}
13 . $LUSTRE/tests/test-framework.sh
17 ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT "
19 # bug number for skipped tests: LU-9795
20 ALWAYS_EXCEPT+=" 402b "
21 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
24 # Skip tests for PPC that fail frequently
25 if [[ $(uname -m) = ppc64 ]]; then
26 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
27 ALWAYS_EXCEPT+=" 1a 1b 1d 1e 12c 12f "
28 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
29 ALWAYS_EXCEPT+=" 12g 12h 12m 12n 12o 12p "
30 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
31 ALWAYS_EXCEPT+=" 12q 21 22 23 24a 24b "
32 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
33 ALWAYS_EXCEPT+=" 24d 24e 24f 25b 30c 37 "
34 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
35 ALWAYS_EXCEPT+=" 57 58 90 110b 111b 113 "
36 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
37 ALWAYS_EXCEPT+=" 222b 222d 228 260a 260b 260c "
38 # bug number: LU-12252 LU-12252 LU-12252 LU-12252 LU-12252 LU-12252
39 ALWAYS_EXCEPT+=" 220A 220a 221 222a 222c 223a "
40 # bug number: LU-12252 LU-12252 LU-12252 LU-12252 LU-12252 LU-12252
41 ALWAYS_EXCEPT+=" 223b 224A 224a 226 227 600"
42 # bug number: LU-12252 LU-12252 LU-12252 LU-12252 LU-12252 LU-12252
43 ALWAYS_EXCEPT+=" 601 602 603 604 605 "
48 [ -n "$FILESET" ] && skip "Not functional for FILESET set"
50 OPENFILE=${OPENFILE:-openfile}
51 MOUNT_2=${MOUNT_2:-"yes"}
54 # script only handles up to 10 MDTs (because of MDT_PREFIX)
55 [ $MDSCOUNT -gt 9 ] &&
56 error "script cannot handle more than 9 MDTs, please fix"
58 check_and_setup_lustre
60 if [[ $MDS1_VERSION -lt $(version_code 2.4.53) ]]; then
61 skip_env "Need MDS version at least 2.4.53"
64 # $RUNAS_ID may get set incorrectly somewhere else
65 if [[ $UID -eq 0 && $RUNAS_ID -eq 0 ]]; then
66 skip_env "\$RUNAS_ID set to 0, but \$UID is also 0!"
68 check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS
69 if getent group nobody; then
71 elif getent group nogroup; then
74 error "No generic nobody group"
77 # if there is no CLIENT1 defined, some tests can be ran on localhost
78 CLIENT1=${CLIENT1:-$HOSTNAME}
79 # if CLIENT2 doesn't exist then use CLIENT1 instead
80 # All tests should use CLIENT2 with MOUNT2 only therefore it will work if
82 # Exception is the test which need two separate nodes
83 CLIENT2=${CLIENT2:-$CLIENT1}
86 local hosts=${1:-$(facet_active_host $SINGLEAGT)}
87 do_nodesv $hosts "pgrep --pidfile=$HSMTOOL_PID_FILE hsmtool"
91 local hosts=${1:-$(facet_active_host $SINGLEAGT)}
92 local wait_timeout=200
93 local wait_start=$SECONDS
94 local wait_end=$((wait_start + wait_timeout))
97 while ((SECONDS < wait_end)); do
98 if ! search_copytools $hosts; then
99 echo "copytools stopped in $((SECONDS - wait_start))s"
103 echo "copytools still running on $hosts"
105 [ $sleep_time -lt 5 ] && sleep_time=$((sleep_time + 1))
108 # try to dump Copytool's stack
109 do_nodesv $hosts "echo 1 >/proc/sys/kernel/sysrq ; " \
110 "echo t >/proc/sysrq-trigger"
112 echo "copytools failed to stop in ${wait_timeout}s"
117 copytool_monitor_setup() {
118 local facet=${1:-$SINGLEAGT}
119 local agent=$(facet_active_host $facet)
121 local cmd="mktemp --tmpdir=/tmp -d ${TESTSUITE}.${TESTNAME}.XXXX"
122 local test_dir=$(do_node $agent "$cmd") ||
123 error "Failed to create tempdir on $agent"
124 export HSMTOOL_MONITOR_DIR=$test_dir
126 # Create the fifo and a monitor (cat dies when copytool dies)
127 do_node $agent "mkfifo -m 0644 $test_dir/fifo" ||
128 error "failed to create copytool fifo on $agent"
129 cmd="cat $test_dir/fifo > $test_dir/events &"
130 cmd+=" echo \\\$! > $test_dir/monitor_pid"
132 # This background subshell nonsense is required when pdsh/ssh decides
133 # to wait for the cat process to exit on the remote client
134 (do_node $agent "$cmd") &
135 export HSMTOOL_MONITOR_PDSH=$!
137 # Slightly racy, but just making a best-effort to catch obvious
140 do_node $agent "stat $HSMTOOL_MONITOR_DIR/monitor_pid 2>&1 > /dev/null"
142 error "Failed to start copytool monitor on $agent"
150 case "$HSMTOOL_ARCHIVE_FORMAT" in
152 printf "%s" "$(hsm_root)/*/*/*/*/*/*/$fid"
155 printf "%s" "$(hsm_root)/*/$fid"
160 get_copytool_event_log() {
161 local facet=${1:-$SINGLEAGT}
162 local agent=$(facet_active_host $facet)
164 [ -z "$HSMTOOL_MONITOR_DIR" ] &&
165 error "Can't get event log: No monitor directory!"
167 do_node $agent "cat $HSMTOOL_MONITOR_DIR/events" ||
168 error "Could not collect event log from $agent"
172 local agents=${1:-$(facet_active_host $SINGLEAGT)}
174 stack_trap "pkill_copytools $agents CONT || true" EXIT
175 pkill_copytools $agents STOP || return 0
176 echo "Copytool is suspended on $agents"
179 copytool_remove_backend() {
181 local be=$(do_facet $SINGLEAGT find "$(hsm_root)" -name $fid)
182 echo "Remove from backend: $fid = $be"
183 do_facet $SINGLEAGT rm -f $be
186 file_creation_failure() {
193 df $MOUNT $MOUNT2 >&2
194 error "Not enough space to create $file with $cmd"
197 error "cannot create $file with $cmd, status=$err"
202 # Creates a file using dd
208 local source=${5:-/dev/zero}
212 if [ -n "$conv" ]; then
216 # Create the directory in case it does not exist
217 mkdir -p "$(dirname "$file")"
218 # Delete the file in case it already exist
221 if dd if="$source" of="$file" count="$count" bs="$bs" $args; then
222 path2fid "$file" || error "cannot get FID of '$file'"
225 echo "cannot create file '$file'" >&2;
226 # Let the caller decide what to do on error
231 create_empty_file() {
232 create_file "${1/$DIR/$DIR2}" 1M 0 ||
233 file_creation_failure dd "${1/$DIR/$DIR2}" $?
236 create_small_file() {
237 local source_file=/dev/urandom
240 local conv=${2:-fsync}
242 create_file "${1/$DIR/$DIR2}" $bs $count $conv $source_file ||
243 file_creation_failure dd "${1/$DIR/$DIR2}" $?
246 create_small_sync_file() {
247 create_small_file "$1" sync
250 create_archive_file() {
251 local file="$(hsm_root)/$1"
253 local source=/dev/urandom
255 # Create the counterpart directory of the archive
256 do_facet "$SINGLEAGT" mkdir -p "$(dirname "$file")" ||
257 error "cannot create archive directory '$(dirname "$file")'"
259 do_facet "$SINGLEAGT" dd if=$source of="$file" bs=1M count=$count ||
260 error "cannot create archive file '$file'"
264 local hsm_root="$(hsm_root)"
265 local file="$hsm_root/$2"
267 stack_trap "do_facet $SINGLEAGT rm -rf '$hsm_root'" EXIT
268 do_facet $SINGLEAGT mkdir -p "$(dirname "$file")" ||
269 error "mkdir '$(dirname "$file")' failed"
270 do_facet $SINGLEAGT cp -p "$1" "$file" ||
271 error "cannot copy '$1' to '$file'"
276 local val=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.$param)
283 mdts_set_param "" hsm_control "$cmd"
284 mdts_check_param hsm_control "$target" 10
289 mdts_set_param "" hsm.policy "+NRA"
290 CDT_POLICY_HAD_CHANGED=true
293 cdt_clear_no_retry() {
294 mdts_set_param "" hsm.policy "-NRA"
295 CDT_POLICY_HAD_CHANGED=true
298 cdt_set_non_blocking_restore() {
299 mdts_set_param "" hsm.policy "+NBR"
300 CDT_POLICY_HAD_CHANGED=true
303 cdt_clear_non_blocking_restore() {
304 mdts_set_param "" hsm.policy "-NBR"
305 CDT_POLICY_HAD_CHANGED=true
308 cdt_clear_mount_state() {
309 mdts_set_param "-P -d" hsm_control ""
313 set_test_state disabled disabled
317 set_test_state enabled enabled
321 set_test_state shutdown stopped
325 set_test_state purge enabled
331 cdt_set_sanity_policy
335 get_hsm_archive_id() {
338 st=$($LFS hsm_state $f)
339 [[ $? == 0 ]] || error "$LFS hsm_state $f failed"
341 local ar=$(echo $st | grep -oP '(?<=archive_id:).*')
345 check_hsm_flags_user() {
349 local st=$(get_hsm_flags $f user)
350 [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
362 if [[ "$3" != 1 ]]; then
366 cp $1 $f || file_creation_failure cp $f $?
368 path2fid $f || error "cannot get fid on $f"
371 # Delete any file bigger than 10M under $MOUNT and wait for deletes to complete
373 # Note that this might lead to surprising behaviours such as deleting an
374 # important file for the currently running test
375 delete_large_files() {
376 printf "Deleting large files...\n" >&2
377 find $MOUNT -size +10M -delete
378 wait_delete_completed
381 get_request_state() {
385 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
386 "awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
389 get_request_count() {
393 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
394 "awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
397 # Ensure the number of HSM request for a given FID is correct
398 # assert_request_count FID REQUEST_TYPE COUNT [ERROR_MSG]
399 assert_request_count() {
400 local request_count=$(get_request_count $1 $2)
401 local default_error_msg=("expected $3 '$2' request(s) for '$1', found "
403 [ $request_count -eq $3 ] || error "${4:-"${default_error_msg[@]}"}"
410 local cmd="$LCTL get_param -n $HSM_PARAM.actions"
411 [[ -n $fid ]] && cmd+=" | grep '$fid'"
412 cmd+=" | egrep 'WAITING|STARTED'"
414 wait_update_facet --verbose mds1 "$cmd" "" $timeout ||
415 error "requests did not complete"
418 wait_for_grace_delay() {
419 local val=$(get_hsm_param grace_delay)
423 wait_for_loop_period() {
424 local val=$(get_hsm_param loop_period)
431 # python2.6 in EL6 includes an internal json module
432 local PYTHON='python'
433 local json_parser='import json; import fileinput;'
434 json_parser+=' print("\n".join(["local %s=\"%s\"" % tuple for tuple in '
435 json_parser+='json.loads([line for line in '
436 json_parser+='fileinput.input()][0]).items()]))'
438 # check if python/python2/python3 is available
439 if ! which $PYTHON > /dev/null 2>&1 ; then
441 if ! which $PYTHON > /dev/null 2>&1 ; then
445 echo $raw_event | $PYTHON -c "$json_parser"
448 get_agent_by_uuid_mdt() {
451 local mds=mds$(($mdtidx + 1))
452 do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
456 check_agent_registered_by_mdt() {
459 local mds=mds$(($mdtidx + 1))
460 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
461 if [[ ! -z "$agent" ]]; then
462 echo "found agent $agent on $mds"
464 error "uuid $uuid not found in agent list on $mds"
468 check_agent_unregistered_by_mdt() {
471 local mds=mds$(($mdtidx + 1))
472 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
473 if [[ -z "$agent" ]]; then
474 echo "uuid not found in agent list on $mds"
476 error "uuid found in agent list on $mds: $agent"
480 check_agent_registered() {
483 for mdsno in $(seq 1 $MDSCOUNT); do
484 check_agent_registered_by_mdt $uuid $((mdsno - 1))
488 check_agent_unregistered() {
491 for mdsno in $(seq 1 $MDSCOUNT); do
492 check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
497 local agent=${1:-$(facet_active_host $SINGLEAGT)}
499 # Lustre mount-point is mandatory and last parameter on
501 local mntpnt=$(do_rpc_nodes $agent \
502 pgrep --pidfile=$HSMTOOL_PID_FILE --list-full hsmtool |
504 [ -n "$mntpnt" ] || error "Found no Agent or with no mount-point "\
506 do_rpc_nodes $agent get_client_uuid $mntpnt | cut -d' ' -f2
512 # populate MDT device array
515 # cleanup from previous bad setup
518 # for recovery tests, coordinator needs to be started at mount
520 # the lustre conf must be without hsm on (like for sanity.sh)
521 echo "Set HSM on and start"
522 cdt_set_mount_state enabled
523 cdt_check_state enabled
525 echo "Set sanity-hsm HSM policy"
526 cdt_set_sanity_policy
528 # finished requests are quickly removed from list
529 set_hsm_param grace_delay 10
531 CLIENT_NIDS=( $($LCTL list_nids all) )
533 test_1A() { # was test_1
537 local f=$DIR/$tdir/$tfile
541 check_hsm_flags_user $f "0x00000000"
543 $RUNAS $LFS hsm_set --norelease $f ||
544 error "user could not change hsm flags"
545 check_hsm_flags_user $f "0x00000010"
547 $RUNAS $LFS hsm_clear --norelease $f ||
548 error "user could not clear hsm flags"
549 check_hsm_flags_user $f "0x00000000"
551 # User could not change those flags...
552 $RUNAS $LFS hsm_set --exists $f &&
553 error "user should not set this flag"
554 check_hsm_flags_user $f "0x00000000"
557 $LFS hsm_set --exists $f ||
558 error "root could not change hsm flags"
559 check_hsm_flags_user $f "0x00000001"
561 $LFS hsm_clear --exists $f ||
562 error "root could not clear hsm state"
563 check_hsm_flags_user $f "0x00000000"
566 run_test 1A "lfs hsm flags root/non-root access"
569 mkdir_on_mdt0 $DIR/$tdir
571 local f=$DIR/$tdir/$tfile
572 local fid=$(create_small_file $f)
576 $LFS hsm_archive $f || error "could not archive file"
577 wait_request_state $fid ARCHIVE SUCCEED
579 # Release and check states
580 $LFS hsm_release $f || error "could not release file"
581 echo -n "Verifying released state: "
582 check_hsm_flags $f "0x0000000d"
584 $MMAP_CAT $f > /dev/null || error "failed mmap & cat release file"
586 run_test 1a "mmap & cat a HSM released file"
592 dd if=/dev/urandom of=$f bs=1M count=1 conv=sync ||
593 error "failed to create file"
594 local fid=$(path2fid $f)
599 $LFS hsm_archive $f || error "could not archive file"
600 wait_request_state $fid ARCHIVE SUCCEED
603 $LFS hsm_release $f || error "could not release file"
604 echo "verify released state: "
605 check_hsm_flags $f "0x0000000d" && echo "pass"
608 $LFS hsm_restore $f || error "could not restore file"
609 wait_request_state $fid RESTORE SUCCEED
610 echo "verify restored state: "
611 check_hsm_flags $f "0x00000009" && echo "pass"
615 mkdir_on_mdt0 $DIR/$tdir
616 $LFS setstripe -E 1M -S 1M -E 64M -c 2 -E -1 -c 4 $DIR/$tdir ||
617 error "failed to set default stripe"
618 local f=$DIR/$tdir/$tfile
622 run_test 1b "Archive, Release and Restore composite file"
628 local f=$DIR/$tdir/$tfile
631 # Test whether we can set the maximum archive number.
632 local LOCAL_HSM_ARCHIVE_NUMBER=32
633 $LFS hsm_set --exists --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f ||
634 error "root could not change hsm flags"
635 check_hsm_flags_user $f "0x00000001"
636 echo "verifying archive number is $LOCAL_HSM_ARCHIVE_NUMBER"
637 local st=$(get_hsm_archive_id $f)
638 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
639 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
641 # Test whether setting archive number 0 results in no change.
642 $LFS hsm_set --exists --archive-id 0 $f ||
643 error "root could not change hsm flags"
644 check_hsm_flags_user $f "0x00000001"
645 echo "verifying archive number is still $LOCAL_HSM_ARCHIVE_NUMBER"
646 st=$(get_hsm_archive_id $f)
647 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
648 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
650 LOCAL_HSM_ARCHIVE_NUMBER=33
651 if [ "$CLIENT_VERSION" -ge $(version_code 2.11.56) ] &&
652 [ "$MDS1_VERSION" -ge $(version_code 2.11.56) ]; then
653 # lustre in the new version supports unlimited archiveID.
654 # Test whether setting archive number > 32 is supported
655 $LFS hsm_set --exists --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f ||
656 error "archive ID $LOCAL_HSM_ARCHIVE_NUMBER too large?"
657 check_hsm_flags_user $f "0x00000001"
659 echo "verifying archive number is $LOCAL_HSM_ARCHIVE_NUMBER"
660 st=$(get_hsm_archive_id $f)
661 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
662 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
664 # old client or old mds can only support at most 32 archiveID
665 # test whether setting archive number > 32 results in error.
666 $LFS hsm_set --exists --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f &&
667 error "bitmap archive number is larger than 32"
668 check_hsm_flags_user $f "0x00000001"
671 # Test whether setting archive number 16 and archived flag.
672 LOCAL_HSM_ARCHIVE_NUMBER=16
673 $LFS hsm_set --exists --archived \
674 --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f ||
675 error "root could not change hsm flags"
676 check_hsm_flags_user $f "0x00000009"
677 echo "verifying archive number is $LOCAL_HSM_ARCHIVE_NUMBER"
678 st=$(get_hsm_archive_id $f)
679 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
680 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
682 run_test 1c "Check setting archive-id in lfs hsm_set"
685 [ $MDS1_VERSION -lt $(version_code 2.10.59) ] &&
686 skip "need MDS version at least 2.10.59"
688 mkdir_on_mdt0 $DIR/$tdir
689 $LFS setstripe -E 1M -L mdt -E -1 -c 2 $DIR/$tdir ||
690 error "failed to set default stripe"
691 local f=$DIR/$tdir/$tfile
695 run_test 1d "Archive, Release and Restore DoM file"
698 [ "$MDS1_VERSION" -lt $(version_code $SEL_VER) ] &&
699 skip "skipped for lustre < $SEL_VER"
701 mkdir_on_mdt0 $DIR/$tdir
702 $LFS setstripe -E 1G -z 64M -E 10G -z 512M -E -1 -z 1G $DIR/$tdir ||
703 error "failed to set default stripe"
704 local comp_file=$DIR/$tdir/$tfile
706 test_1bde_base $comp_file
708 local flg_opts="--comp-start 0 -E 64M --comp-flags init"
709 local found=$($LFS find $flg_opts $comp_file | wc -l)
710 [ $found -eq 1 ] || error "1st component not found"
712 flg_opts="--comp-start 64M -E 1G --comp-flags extension"
713 found=$($LFS find $flg_opts $comp_file | wc -l)
714 [ $found -eq 1 ] || error "2nd component not found"
716 flg_opts="--comp-start 1G -E 1G --comp-flags ^init"
717 found=$($LFS find $flg_opts $comp_file | wc -l)
718 [ $found -eq 1 ] || error "3rd component not found"
720 flg_opts="--comp-start 1G -E 10G --comp-flags extension"
721 found=$($LFS find $flg_opts $comp_file | wc -l)
722 [ $found -eq 1 ] || error "4th component not found"
724 flg_opts="--comp-start 10G -E 10G --comp-flags ^init"
725 found=$($LFS find $flg_opts $comp_file | wc -l)
726 [ $found -eq 1 ] || error "5th component not found"
728 flg_opts="--comp-start 10G -E EOF --comp-flags extension"
729 found=$($LFS find $flg_opts $comp_file | wc -l)
730 [ $found -eq 1 ] || error "6th component not found"
732 sel_layout_sanity $comp_file 6
734 run_test 1e "Archive, Release and Restore SEL file"
737 (( $MDS1_VERSION >= $(version_code 2.15.55.203) )) ||
738 skip "need MDS version at least 2.15.55.203"
739 local dom=$DIR/$tdir/$tfile
741 mkdir_on_mdt0 $DIR/$tdir
742 $LFS setstripe -E 512K -L mdt -E -1 -c 2 $DIR/$tdir ||
743 error "failed to set default stripe"
747 [[ $($LFS getstripe --component-start=0 -L $dom) == 'mdt' ]] ||
748 error "MDT stripe isn't set"
750 # check that metadata change doesn't prevent second release
751 chmod 600 $dom || error "chmod failed"
753 echo "release again $dom"
754 $LFS hsm_release $dom || error "second release failed"
756 echo "verify released state: "
757 check_hsm_flags $dom "0x0000000d" && echo "pass"
759 run_test 1f "DoM file release after restore"
762 local f=$DIR/$tdir/$tfile
764 mkdir_on_mdt0 $DIR/$tdir
765 create_empty_file "$f"
766 # New files are not dirty
767 check_hsm_flags $f "0x00000000"
769 # For test, we simulate an archived file.
770 $LFS hsm_set --exists $f || error "user could not change hsm flags"
771 check_hsm_flags $f "0x00000001"
773 # chmod do not put the file dirty
774 chmod 600 $f || error "could not chmod test file"
775 check_hsm_flags $f "0x00000001"
777 # chown do not put the file dirty
778 chown $RUNAS_ID $f || error "could not chown test file"
779 check_hsm_flags $f "0x00000001"
781 # truncate put the file dirty
782 $TRUNCATE $f 1 || error "could not truncate test file"
783 check_hsm_flags $f "0x00000003"
785 $LFS hsm_clear --dirty $f || error "could not clear hsm flags"
786 check_hsm_flags $f "0x00000001"
788 run_test 2 "Check file dirtyness when doing setattr"
791 mkdir_on_mdt0 $DIR/$tdir
794 # New files are not dirty
796 check_hsm_flags $f "0x00000000"
798 # For test, we simulate an archived file.
799 $LFS hsm_set --exists $f ||
800 error "user could not change hsm flags"
801 check_hsm_flags $f "0x00000001"
803 # Reading a file, does not set dirty
804 cat $f > /dev/null || error "could not read file"
805 check_hsm_flags $f "0x00000001"
807 # Open for write without modifying data, does not set dirty
808 openfile -f O_WRONLY $f || error "could not open test file"
809 check_hsm_flags $f "0x00000001"
811 # Append to a file sets it dirty
812 cp -p /etc/passwd $f.append || error "could not create file"
813 $LFS hsm_set --exists $f.append ||
814 error "user could not change hsm flags"
815 dd if=/etc/passwd of=$f.append bs=1 count=3\
816 conv=notrunc oflag=append status=noxfer ||
817 file_creation_failure dd $f.append $?
818 check_hsm_flags $f.append "0x00000003"
820 # Modify a file sets it dirty
821 cp -p /etc/passwd $f.modify || error "could not create file"
822 $LFS hsm_set --exists $f.modify ||
823 error "user could not change hsm flags"
824 dd if=/dev/zero of=$f.modify bs=1 count=3\
825 conv=notrunc status=noxfer ||
826 file_creation_failure dd $f.modify $?
827 check_hsm_flags $f.modify "0x00000003"
829 # Open O_TRUNC sets dirty
830 cp -p /etc/passwd $f.trunc || error "could not create file"
831 $LFS hsm_set --exists $f.trunc ||
832 error "user could not change hsm flags"
833 cp /etc/group $f.trunc || error "could not override a file"
834 check_hsm_flags $f.trunc "0x00000003"
836 # Mmapped a file sets dirty
837 cp -p /etc/passwd $f.mmap || error "could not create file"
838 $LFS hsm_set --exists $f.mmap ||
839 error "user could not change hsm flags"
840 multiop $f.mmap OSMWUc || error "could not mmap a file"
841 check_hsm_flags $f.mmap "0x00000003"
843 run_test 3 "Check file dirtyness when opening for write"
846 local f=$DIR/$tdir/$tfile
847 local fid=$(create_small_file $f)
850 local st=$(get_request_state $fid CANCEL)
851 [[ -z "$st" ]] || error "hsm_cancel must not be registered (state=$st)"
853 run_test 4 "Useless cancel must not be registered"
856 # test needs a running copytool
859 mkdir_on_mdt0 $DIR/$tdir
860 local f=$DIR/$tdir/$tfile
861 local fid=$(copy_file /etc/passwd $f)
863 wait_request_state $fid ARCHIVE SUCCEED
865 check_hsm_flags $f "0x00000009"
867 run_test 8 "Test default archive number"
869 test_9A() { # was test_9
870 # we do not use the default one to be sure
871 local archive_id=$((HSM_ARCHIVE_NUMBER + 1))
872 copytool setup --archive-id $archive_id
874 # give time for CT to register with MDTs
875 sleep $(($MDSCOUNT*2))
876 local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
877 check_agent_registered $uuid
879 mkdir_on_mdt0 $DIR/$tdir
881 local f=$DIR/$tdir/$tfile
882 local fid=$(copy_file /etc/passwd $f)
883 $LFS hsm_archive --archive $archive_id $f
884 wait_request_state $fid ARCHIVE SUCCEED
886 check_hsm_flags $f "0x00000009"
888 run_test 9A "Use of explicit archive number, with dedicated copytool"
891 needclients 3 || return 0
897 # start all of the copytools
898 for n in $(seq $AGTCOUNT); do
899 copytool setup --facet agt$n
902 mkdir_on_mdt0 $DIR/$tdir
905 for n in $(seq $AGTCOUNT); do
906 file=$DIR/$tdir/$tfile.$n
907 fid=$(create_small_file $file)
909 $LFS hsm_archive $file || error "could not archive file $file"
910 wait_request_state $fid ARCHIVE SUCCEED
911 check_hsm_flags $file "0x00000009"
914 run_test 9a "Multiple remote agents"
917 # test needs a running copytool
920 mkdir_on_mdt0 $DIR/$tdir
921 mkdir -p $DIR/$tdir/d1
922 local f=$DIR/$tdir/$tfile
923 local fid=$(copy_file /etc/hosts $f)
924 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
925 error "hsm_archive failed"
926 wait_request_state $fid ARCHIVE SUCCEED
928 local hsm_root="$(copytool_device $SINGLEAGT)"
929 local archive="$(do_facet $SINGLEAGT \
930 find "$hsm_root" -name "$fid" -print0)"
931 [ -n "$archive" ] || error "fid '$fid' not in archive '$hsm_root'"
933 echo "Verifying content"
934 do_facet $SINGLEAGT diff $f $archive || error "archived file differs"
935 echo "Verifying hsm state "
936 check_hsm_flags $f "0x00000009"
938 echo "Verifying archive number is $HSM_ARCHIVE_NUMBER"
939 local st=$(get_hsm_archive_id $f)
940 [[ $st == $HSM_ARCHIVE_NUMBER ]] ||
941 error "Wrong archive number, $st != $HSM_ARCHIVE_NUMBER"
943 run_test 10a "Archive a file"
946 # test needs a running copytool
949 mkdir_on_mdt0 $DIR/$tdir
950 local f=$DIR/$tdir/$tfile
951 local fid=$(copy_file /etc/hosts $f)
952 $LFS hsm_archive $f || error "archive request failed"
953 wait_request_state $fid ARCHIVE SUCCEED
955 $LFS hsm_archive $f || error "archive of non dirty file failed"
956 local cnt=$(get_request_count $fid ARCHIVE)
957 [[ "$cnt" == "1" ]] ||
958 error "archive of non dirty file must not make a request"
960 run_test 10b "Archive of non dirty file must work without doing request"
963 # test needs a running copytool
967 local f=$DIR/$tdir/$tfile
968 local fid=$(copy_file /etc/hosts $f)
969 $LFS hsm_set --noarchive $f
970 $LFS hsm_archive $f && error "archive a noarchive file must fail"
973 run_test 10c "Check forbidden archive"
976 # test needs a running copytool
979 mkdir_on_mdt0 $DIR/$tdir
980 local f=$DIR/$tdir/$tfile
981 local fid=$(copy_file /etc/hosts $f)
982 $LFS hsm_archive $f || error "cannot archive $f"
983 wait_request_state $fid ARCHIVE SUCCEED
985 local ar=$(get_hsm_archive_id $f)
986 local dflt=$(get_hsm_param default_archive_id)
987 [[ $ar == $dflt ]] ||
988 error "archived file is not on default archive: $ar != $dflt"
990 run_test 10d "Archive a file on the default archive id"
993 mkdir_on_mdt0 $DIR/$tdir
994 copy2archive /etc/hosts $tdir/$tfile
995 local f=$DIR/$tdir/$tfile
997 copytool import $tdir/$tfile $f
998 echo -n "Verifying released state: "
999 check_hsm_flags $f "0x0000000d"
1001 local LSZ=$(stat -c "%s" $f)
1002 local ASZ=$(do_facet $SINGLEAGT stat -c "%s" "$(hsm_root)/$tdir/$tfile")
1004 echo "Verifying imported size $LSZ=$ASZ"
1005 [[ $LSZ -eq $ASZ ]] || error "Incorrect size $LSZ != $ASZ"
1006 echo -n "Verifying released pattern: "
1007 local PTRN=$($LFS getstripe -L $f)
1009 [[ $PTRN == released ]] || error "Is not released"
1010 local fid=$(path2fid $f)
1011 echo "Verifying new fid $fid in archive"
1013 do_facet $SINGLEAGT "[ -f \"$(fid2archive "$fid")\" ]" ||
1014 error "No archive for fid $fid"
1016 run_test 11a "Import a file"
1019 # test needs a running copytool
1022 mkdir_on_mdt0 $DIR/$tdir
1023 local f=$DIR/$tdir/$tfile
1024 local fid=$(copy_file /etc/hosts $f)
1025 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
1026 error "hsm_archive failed"
1027 wait_request_state $fid ARCHIVE SUCCEED
1029 local FILE_HASH=$(md5sum $f)
1032 copytool import $fid $f
1034 echo "$FILE_HASH" | md5sum -c
1036 [[ $? -eq 0 ]] || error "Restored file differs"
1038 run_test 11b "Import a deleted file using its FID"
1041 pool_add $TESTNAME || error "Pool creation failed"
1042 pool_add_targets $TESTNAME 1 1 || error "pool_add_targets failed"
1045 $LFS setstripe -p "$TESTNAME" $DIR/$tdir
1047 copy2archive /etc/hosts $tdir/$tfile
1048 copytool import $tdir/$tfile $DIR/$tdir/$tfile
1050 run_test 11c "Import a file to a directory with a pool"
1053 # test needs a running copytool
1056 mkdir_on_mdt0 $DIR/$tdir
1057 copy2archive /etc/hosts $tdir/$tfile
1059 local f=$DIR/$tdir/$tfile
1060 copytool import $tdir/$tfile $f
1061 local f2=$DIR2/$tdir/$tfile
1062 echo "Verifying released state: "
1063 check_hsm_flags $f2 "0x0000000d"
1065 local fid=$(path2fid $f2)
1066 $LFS hsm_restore $f2
1067 wait_request_state $fid RESTORE SUCCEED
1069 echo "Verifying file state: "
1070 check_hsm_flags $f2 "0x00000009"
1072 do_facet $SINGLEAGT diff -q $(hsm_root)/$tdir/$tfile $f
1074 [[ $? -eq 0 ]] || error "Restored file differs"
1076 run_test 12a "Restore an imported file explicitly"
1079 # test needs a running copytool
1082 mkdir_on_mdt0 $DIR/$tdir
1083 copy2archive /etc/hosts $tdir/$tfile
1085 local f=$DIR/$tdir/$tfile
1086 copytool import $tdir/$tfile $f
1087 echo "Verifying released state: "
1088 check_hsm_flags $f "0x0000000d"
1090 cat $f > /dev/null || error "File read failed"
1092 echo "Verifying file state after restore: "
1093 check_hsm_flags $f "0x00000009"
1095 do_facet $SINGLEAGT diff -q $(hsm_root)/$tdir/$tfile $f
1097 [[ $? -eq 0 ]] || error "Restored file differs"
1099 run_test 12b "Restore an imported file implicitly"
1102 [ "$OSTCOUNT" -lt "2" ] && skip_env "needs >= 2 OSTs" && return
1104 # test needs a running copytool
1107 local f=$DIR/$tdir/$tfile
1108 mkdir_on_mdt0 $DIR/$tdir
1109 $LFS setstripe -c 2 "$f"
1110 local fid=$(create_file "$f" 1M 5)
1112 local FILE_CRC=$(md5sum $f)
1114 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1115 wait_request_state $fid ARCHIVE SUCCEED
1116 $LFS hsm_release $f || error "release $f failed"
1118 echo "$FILE_CRC" | md5sum -c
1120 [[ $? -eq 0 ]] || error "Restored file differs"
1122 run_test 12c "Restore a file with stripe of 2"
1125 # test needs a running copytool
1128 mkdir_on_mdt0 $DIR/$tdir
1130 local f=$DIR/$tdir/$tfile
1131 local fid=$(copy_file /etc/hosts $f)
1132 $LFS hsm_restore $f || error "restore of non archived file failed"
1133 local cnt=$(get_request_count $fid RESTORE)
1134 [[ "$cnt" == "0" ]] ||
1135 error "restore non archived must not make a request"
1136 $LFS hsm_archive $f ||
1137 error "archive request failed"
1138 wait_request_state $fid ARCHIVE SUCCEED
1139 $LFS hsm_restore $f ||
1140 error "restore of non released file failed"
1141 local cnt=$(get_request_count $fid RESTORE)
1142 [[ "$cnt" == "0" ]] ||
1143 error "restore a non dirty file must not make a request"
1145 run_test 12d "Restore of a non archived, non released file must work"\
1146 " without doing request"
1149 # test needs a running copytool
1152 mkdir_on_mdt0 $DIR/$tdir
1153 local f=$DIR/$tdir/$tfile
1154 local fid=$(copy_file /etc/hosts $f)
1155 $LFS hsm_archive $f || error "archive request failed"
1156 wait_request_state $fid ARCHIVE SUCCEED
1159 cat /etc/hosts >> $f
1163 $LFS hsm_restore $f && error "restore a dirty file must fail"
1166 run_test 12e "Check forbidden restore"
1169 # test needs a running copytool
1172 mkdir_on_mdt0 $DIR/$tdir
1173 local f=$DIR/$tdir/$tfile
1174 local fid=$(copy_file /etc/hosts $f)
1176 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1177 wait_request_state $fid ARCHIVE SUCCEED
1178 $LFS hsm_release $f || error "release of $f failed"
1180 wait_request_state $fid RESTORE SUCCEED
1182 echo -n "Verifying file state: "
1183 check_hsm_flags $f "0x00000009"
1185 diff -q /etc/hosts $f
1187 [[ $? -eq 0 ]] || error "Restored file differs"
1189 run_test 12f "Restore a released file explicitly"
1192 # test needs a running copytool
1195 mkdir_on_mdt0 $DIR/$tdir
1196 local f=$DIR/$tdir/$tfile
1197 local fid=$(copy_file /etc/hosts $f)
1199 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1200 wait_request_state $fid ARCHIVE SUCCEED
1201 $LFS hsm_release $f || error "release of $f failed"
1203 diff -q /etc/hosts $f
1206 # we check we had a restore done
1207 wait_request_state $fid RESTORE SUCCEED
1209 [[ $st -eq 0 ]] || error "Restored file differs"
1211 run_test 12g "Restore a released file implicitly"
1214 needclients 2 || return 0
1216 # test needs a running copytool
1219 mkdir_on_mdt0 $DIR/$tdir
1220 local f=$DIR/$tdir/$tfile
1221 local fid=$(copy_file /etc/hosts $f)
1223 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1224 wait_request_state $fid ARCHIVE SUCCEED
1225 $LFS hsm_release $f || error "release of $f failed"
1227 do_node $CLIENT2 diff -q /etc/hosts $f
1230 # we check we had a restore done
1231 wait_request_state $fid RESTORE SUCCEED
1233 [[ $st -eq 0 ]] || error "Restored file differs"
1235 run_test 12h "Restore a released file implicitly from a second node"
1238 # test needs a running copytool
1241 mkdir_on_mdt0 $DIR/$tdir
1242 local f=$DIR/$tdir/$tfile
1243 local fid=$(copy_file /etc/passwd $f)
1244 $LFS hsm_archive $f || error "archive of $f failed"
1245 wait_request_state $fid ARCHIVE SUCCEED
1247 $LFS hsm_release $f || error "release of $f failed"
1251 [[ $? -eq 0 ]] || error "Restored file differs"
1253 run_test 12m "Archive/release/implicit restore"
1256 # test needs a running copytool
1260 copy2archive /etc/hosts $tdir/$tfile
1262 local f=$DIR/$tdir/$tfile
1263 copytool import $tdir/$tfile $f
1265 do_facet $SINGLEAGT cmp /etc/hosts $f ||
1266 error "Restored file differs"
1268 $LFS hsm_release $f || error "release of $f failed"
1270 run_test 12n "Import/implicit restore/release"
1273 # test needs a running copytool
1276 mkdir_on_mdt0 $DIR/$tdir
1277 local f=$DIR/$tdir/$tfile
1278 local fid=$(copy_file /etc/hosts $f)
1280 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1281 wait_request_state $fid ARCHIVE SUCCEED
1282 $LFS hsm_release $f || error "release of $f failed"
1284 #define OBD_FAIL_MDS_HSM_SWAP_LAYOUTS 0x152
1285 do_facet $SINGLEMDS lctl set_param fail_loc=0x152
1287 # set no retry action mode
1290 diff -q /etc/hosts $f
1293 # we check we had a restore failure
1294 wait_request_state $fid RESTORE FAILED
1296 [[ $st -eq 0 ]] && error "Restore must fail"
1298 # remove no retry action mode
1301 # check file is still released
1302 check_hsm_flags $f "0x0000000d"
1304 # retry w/o failure injection
1305 do_facet $SINGLEMDS lctl set_param fail_loc=0
1307 # to be sure previous RESTORE result is gone
1309 wait_for_grace_delay
1311 diff -q /etc/hosts $f
1314 # we check we had a restore done
1315 wait_request_state $fid RESTORE SUCCEED
1317 [[ $st -eq 0 ]] || error "Restored file differs"
1319 run_test 12o "Layout-swap failure during Restore leaves file released"
1322 # test needs a running copytool
1325 mkdir_on_mdt0 $DIR/$tdir
1326 local f=$DIR/$tdir/$tfile
1327 local fid=$(copy_file /etc/hosts $f)
1329 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1330 wait_request_state $fid ARCHIVE SUCCEED
1331 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1332 $LFS hsm_release $f || error "cannot release $f"
1333 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1334 $LFS hsm_release $f || error "cannot release $f"
1335 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1337 run_test 12p "implicit restore of a file on copytool mount point"
1340 [ $MDS1_VERSION -lt $(version_code 2.7.58) ] &&
1341 skip "need MDS version at least 2.7.58"
1343 stack_trap "zconf_umount \"$(facet_host $SINGLEAGT)\" \"$MOUNT3\"" EXIT
1344 zconf_mount $(facet_host $SINGLEAGT) $MOUNT3 ||
1345 error "cannot mount $MOUNT3 on $SINGLEAGT"
1347 # test needs a running copytool
1348 copytool setup -m "$MOUNT3"
1350 mkdir_on_mdt0 $DIR/$tdir
1352 local f=$DIR/$tdir/$tfile
1353 local f2=$DIR2/$tdir/$tfile
1354 local fid=$(create_small_file $f)
1355 local orig_size=$(stat -c "%s" $f)
1357 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1358 wait_request_state $fid ARCHIVE SUCCEED
1360 $LFS hsm_release $f || error "could not release file"
1361 check_hsm_flags $f "0x0000000d"
1364 wait_copytools || error "copytool failed to stop"
1366 cat $f > /dev/null &
1368 # wait a bit to allow implicit restore request to be handled.
1369 # if not, next stat would also block on layout-lock.
1372 local size=$(stat -c "%s" $f2)
1373 [ $size -eq $orig_size ] ||
1374 error "$f2: wrong size after archive: $size != $orig_size"
1376 copytool setup -m "$MOUNT3"
1380 size=$(stat -c "%s" $f)
1381 [ $size -eq $orig_size ] ||
1382 error "$f: wrong size after restore: $size != $orig_size"
1384 size=$(stat -c "%s" $f2)
1385 [ $size -eq $orig_size ] ||
1386 error "$f2: wrong size after restore: $size != $orig_size"
1390 size=$(stat -c "%s" $f)
1392 error "$f: wrong size after overwrite: $size != 0"
1394 size=$(stat -c "%s" $f2)
1396 error "$f2: wrong size after overwrite: $size != 0"
1398 run_test 12q "file attributes are refreshed after restore"
1401 # test needs a running copytool
1404 mkdir_on_mdt0 $DIR/$tdir
1405 local f=$DIR/$tdir/$tfile
1406 local fid=$(copy_file /etc/hosts $f)
1408 $LFS hsm_archive $f || error "archive of $f failed"
1409 wait_request_state $fid ARCHIVE SUCCEED
1410 $LFS hsm_release $f || error "release of $f failed"
1412 offset=$(lseek_test -d 7 $f)
1414 # we check we had a restore done
1415 wait_request_state $fid RESTORE SUCCEED
1416 [[ $offset == 7 ]] || error "offset $offset != 7"
1418 run_test 12r "lseek restores released file"
1421 local f=$DIR/$tdir/$tfile
1425 (( MDS1_VERSION >= $(version_code 2.15.50) )) ||
1426 skip "Need MDS version newer than 2.15.50"
1428 # test needs a running copytool
1431 mkdir_on_mdt0 $DIR/$tdir
1432 fid=$(copy_file /etc/hosts $f)
1434 $LFS hsm_archive $f || error "archive of $f failed"
1435 wait_request_state $fid ARCHIVE SUCCEED
1436 $LFS hsm_release $f || error "release of $f failed"
1438 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_HSM_RESTORE_RACE 0x8000018b
1439 do_facet mds1 $LCTL set_param fail_loc=0x8000018b
1440 cat $f > /dev/null & pid1=$!
1441 cat $f > /dev/null & pid2=$!
1443 wait $pid1 || error "cat process 1 fail (pid: $pid1)"
1444 wait $pid2 || error "cat process 2 fail (pid: $pid2)"
1446 # Race exists if more than 1 restore requests is registered
1447 assert_request_count $fid RESTORE 1
1449 run_test 12s "race between restore requests"
1454 mkdir_on_mdt0 $DIR/$tdir
1456 local file=$DIR/$tdir/$tfile
1457 local fid=$(copy_file /etc/hosts $file)
1460 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $file ||
1461 error "failed to HSM archive $file"
1462 wait_request_state $fid ARCHIVE SUCCEED
1463 rm -f $file || error "failed to rm $file"
1465 copytool import $fid $file
1466 check_hsm_flags $file "0x0000000d"
1470 for ((i=0; i < $n; i++)); do
1471 cat $file > /dev/null &
1475 for ((i=0; i < $n; i++));do
1476 wait ${pids[$i]} || error "$?: failed to read pid=${pids[$i]}"
1479 run_test 12t "Multiple parallel reads for a HSM imported file"
1482 local dir=$DIR/$tdir
1487 mkdir_on_mdt0 $dir || error "failed to mkdir $dir"
1492 for ((i=0; i<$n; i++)); do
1493 fids[$i]=$(copy_file /etc/hosts $dir/$tfile.$i)
1494 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $dir/$tfile.$i ||
1495 error "failed to HSM archive $dir/$tfile.$i"
1498 for ((i=0; i<$n; i++)); do
1499 wait_request_state ${fids[$i]} ARCHIVE SUCCEED
1500 rm $dir/$tfile.$i || error "failed to rm $dir/$tfile.$i"
1503 for ((i=0; i<$n; i++)); do
1504 copytool import ${fids[$i]} $dir/$tfile.$i
1505 check_hsm_flags $dir/$tfile.$i "0x0000000d"
1508 for ((i=0; i<$n; i++)); do
1509 for ((j=0; j<$t; j++)); do
1510 cat $dir/$tfile.$i > /dev/null &
1511 pids[$((i * t + j))]=$!
1517 for pid in "${pids[@]}"; do
1518 wait $pid || error "$pid: failed to cat file: $?"
1521 run_test 12u "Multiple reads on multiple HSM imported files in parallel"
1525 for i in {1..10}; do
1526 local archive_dir="$(hsm_root)"/subdir/dir.$i
1528 do_facet $SINGLEAGT mkdir -p "$archive_dir"
1529 for j in {1..10}; do
1530 local archive_file="$archive_dir"/file.$j
1532 do_facet $SINGLEAGT "echo $k > \"$archive_dir\"/file.$j"
1538 copytool import "subdir" "$DIR/$tdir"
1540 # To check the import, the test uses diff with the -r flag
1541 # This is nice, but diff only checks files one by one, and triggering
1542 # an implicit restore for one file at a time will consume as many
1543 # seconds as there are files to compare. To speed this up, a restore
1544 # operation is triggered manually first.
1546 find "$DIR/$tdir"/subdir -type f -exec $LFS hsm_restore {} \;
1548 # Compare the imported data
1549 do_facet $SINGLEAGT \
1550 diff -r "$(hsm_root)"/subdir "$DIR/$tdir"/subdir ||
1551 error "imported files differ from archived data"
1553 run_test 13 "Recursively import and restore a directory"
1556 # test needs a running copytool
1559 mkdir_on_mdt0 $DIR/$tdir
1562 local f=$DIR/$tdir/$tfile
1563 local fid=$(create_small_file $f)
1564 local sum=$(md5sum $f | awk '{print $1}')
1565 $LFS hsm_archive $f || error "could not archive file"
1566 wait_request_state $fid ARCHIVE SUCCEED
1568 # create released file (simulate llapi_hsm_import call)
1569 local fid2=$(create_empty_file "$f")
1570 $LFS hsm_set --archived --exists $f || error "could not force hsm flags"
1571 $LFS hsm_release $f || error "could not release file"
1573 # rebind the archive to the newly created file
1574 echo "rebind $fid to $fid2"
1576 copytool rebind $fid $fid2
1578 # restore file and compare md5sum
1579 local sum2=$(md5sum $f | awk '{print $1}')
1581 [[ $sum == $sum2 ]] || error "md5sum mismatch after restore"
1583 run_test 14 "Rebind archived file to a new fid"
1586 # test needs a running copytool
1589 mkdir_on_mdt0 $DIR/$tdir
1592 local f=$DIR/$tdir/$tfile
1594 local tmpfile=$SHARED_DIRECTORY/tmp.$$
1598 for i in $(seq 1 $count); do
1599 fids[$i]=$(create_small_file $f.$i)
1600 sums[$i]=$(md5sum $f.$i | awk '{print $1}')
1601 $LFS hsm_archive $f.$i || error "could not archive file"
1603 wait_all_done $(($count*60))
1605 stack_trap "rm -f $tmpfile" EXIT
1608 for i in $(seq 1 $count); do
1609 local fid2=$(create_empty_file "${f}.${i}")
1610 # add the rebind operation to the list
1611 echo ${fids[$i]} $fid2 >> $tmpfile
1613 # set it released (simulate llapi_hsm_import call)
1614 $LFS hsm_set --archived --exists $f.$i ||
1615 error "could not force hsm flags"
1616 $LFS hsm_release $f.$i || error "could not release file"
1618 nl=$(wc -l < $tmpfile)
1619 [[ $nl == $count ]] || error "$nl files in list, $count expected"
1621 echo "rebind list of files"
1622 copytool rebind "$tmpfile"
1624 # restore files and compare md5sum
1625 for i in $(seq 1 $count); do
1626 local sum2=$(md5sum $f.$i | awk '{print $1}')
1627 [[ $sum2 == ${sums[$i]} ]] ||
1628 error "md5sum mismatch after restore ($sum2 != ${sums[$i]})"
1631 run_test 15 "Rebind a list of files"
1634 # test needs a running copytool
1638 # create a known size file so we can verify transfer speed
1641 dd if=/dev/zero of=$ref bs=1M count=20
1643 mkdir_on_mdt0 $DIR/$tdir
1644 local f=$DIR/$tdir/$tfile
1645 local fid=$(copy_file $ref $f)
1647 local start=$(date +%s)
1649 wait_request_state $fid ARCHIVE SUCCEED
1650 local end=$(date +%s)
1651 # Add 1 to account for rounding errors between start and end (LU-8155)
1652 local duration=$((end - start + 1))
1654 [[ $duration -ge $((goal - 1)) ]] ||
1655 error "Transfer is too fast $duration < $goal"
1657 run_test 16 "Test CT bandwith control option"
1660 local f=$DIR/$tdir/$tfile
1661 create_empty_file "$f"
1663 # Could not release a non-archived file
1664 $LFS hsm_release $f && error "release should not succeed"
1666 # For following tests, we must test them with HS_ARCHIVED set
1667 $LFS hsm_set --exists --archived $f || error "could not add flag"
1669 # Could not release a file if no-release is set
1670 $LFS hsm_set --norelease $f || error "could not add flag"
1671 $LFS hsm_release $f && error "release should not succeed"
1672 $LFS hsm_clear --norelease $f || error "could not remove flag"
1674 # Could not release a file if lost
1675 $LFS hsm_set --lost $f || error "could not add flag"
1676 $LFS hsm_release $f && error "release should not succeed"
1677 $LFS hsm_clear --lost $f || error "could not remove flag"
1679 # Could not release a file if dirty
1680 $LFS hsm_set --dirty $f || error "could not add flag"
1681 $LFS hsm_release $f && error "release should not succeed"
1682 $LFS hsm_clear --dirty $f || error "could not remove flag"
1684 run_test 20 "Release is not permitted"
1687 # test needs a running copytool
1689 mkdir_on_mdt0 $DIR/$tdir
1691 local f=$DIR/$tdir/test_release
1693 # Create a file and check its states
1694 local fid=$(create_small_file $f)
1695 check_hsm_flags $f "0x00000000"
1697 # LU-4388/LU-4389 - ZFS does not report full number of blocks
1698 # used until file is flushed to disk
1699 if [ "$ost1_FSTYPE" == "zfs" ]; then
1700 # this causes an OST_SYNC rpc to be sent
1701 dd if=/dev/zero of=$f bs=512 count=1 oflag=sync conv=notrunc,fsync
1702 # clear locks to reread file data
1703 cancel_lru_locks osc
1706 local orig_size=$(stat -c "%s" $f)
1707 local orig_blocks=$(stat -c "%b" $f)
1709 start_full_debug_logging
1711 $LFS hsm_archive $f || error "could not archive file"
1712 wait_request_state $fid ARCHIVE SUCCEED
1714 local blocks=$(stat -c "%b" $f)
1715 [ $blocks -eq $orig_blocks ] ||
1716 error "$f: wrong block number after archive: " \
1717 "$blocks != $orig_blocks"
1718 local size=$(stat -c "%s" $f)
1719 [ $size -eq $orig_size ] ||
1720 error "$f: wrong size after archive: $size != $orig_size"
1722 # Release and check states
1723 $LFS hsm_release $f || error "could not release file"
1724 check_hsm_flags $f "0x0000000d"
1726 blocks=$(stat -c "%b" $f)
1727 [ $blocks -gt 5 ] &&
1728 error "$f: too many blocks after release: $blocks > 5"
1729 size=$(stat -c "%s" $f)
1730 [ $size -ne $orig_size ] &&
1731 error "$f: wrong size after release: $size != $orig_size"
1733 # Check we can release an file without stripe info
1737 check_hsm_flags $f "0x00000000"
1738 $LFS hsm_archive $f || error "could not archive file"
1739 wait_request_state $fid ARCHIVE SUCCEED
1741 # Release and check states
1742 $LFS hsm_release $f || error "could not release file"
1743 check_hsm_flags $f "0x0000000d"
1745 # Release again a file that is already released is OK
1746 $LFS hsm_release $f || fail "second release should succeed"
1747 check_hsm_flags $f "0x0000000d"
1749 stop_full_debug_logging
1751 run_test 21 "Simple release tests"
1754 # test needs a running copytool
1756 mkdir_on_mdt0 $DIR/$tdir
1758 local f=$DIR/$tdir/test_release
1759 local swap=$DIR/$tdir/test_swap
1761 # Create a file and check its states
1762 local fid=$(create_small_file $f)
1763 check_hsm_flags $f "0x00000000"
1765 $LFS hsm_archive $f || error "could not archive file"
1766 wait_request_state $fid ARCHIVE SUCCEED
1768 # Release and check states
1769 $LFS hsm_release $f || error "could not release file"
1770 check_hsm_flags $f "0x0000000d"
1772 create_small_file $swap
1773 $LFS swap_layouts $swap $f && error "swap_layouts should failed"
1777 run_test 22 "Could not swap a release file"
1780 # test needs a running copytool
1782 mkdir_on_mdt0 $DIR/$tdir
1784 local f=$DIR/$tdir/test_mtime
1786 # Create a file and check its states
1787 local fid=$(create_small_file $f)
1788 check_hsm_flags $f "0x00000000"
1790 $LFS hsm_archive $f || error "could not archive file"
1791 wait_request_state $fid ARCHIVE SUCCEED
1793 # Set modification time in the past
1794 touch -m -a -d @978261179 $f
1796 # Release and check states
1797 $LFS hsm_release $f || error "could not release file"
1798 check_hsm_flags $f "0x0000000d"
1800 local MTIME=$(stat -c "%Y" $f)
1801 local ATIME=$(stat -c "%X" $f)
1802 [ $MTIME -eq "978261179" ] || fail "bad mtime: $MTIME"
1803 [ $ATIME -eq "978261179" ] || fail "bad atime: $ATIME"
1805 run_test 23 "Release does not change a/mtime (utime)"
1808 local file=$DIR/$tdir/$tfile
1817 # test needs a running copytool
1819 mkdir_on_mdt0 $DIR/$tdir
1821 fid=$(create_small_file $file)
1823 # Create a file and check its states
1824 check_hsm_flags $file "0x00000000"
1826 # Ensure atime is less than mtime and ctime.
1830 atime0=$(stat -c "%X" $file)
1831 mtime0=$(stat -c "%Y" $file)
1832 ctime0=$(stat -c "%Z" $file)
1834 [ $atime0 -lt $mtime0 ] ||
1835 error "atime $atime0 is not less than mtime $mtime0"
1837 [ $atime0 -lt $ctime0 ] ||
1838 error "atime $atime0 is not less than ctime $ctime0"
1840 # Archive should not change any timestamps.
1841 $LFS hsm_archive $file || error "cannot archive '$file'"
1842 wait_request_state $fid ARCHIVE SUCCEED
1844 atime1=$(stat -c "%X" $file)
1845 mtime1=$(stat -c "%Y" $file)
1846 ctime1=$(stat -c "%Z" $file)
1848 [ $atime0 -eq $atime1 ] ||
1849 error "archive changed atime from $atime0 to $atime1"
1851 [ $mtime0 -eq $mtime1 ] ||
1852 error "archive changed mtime from $mtime0 to $mtime1"
1854 [ $ctime0 -eq $ctime1 ] ||
1855 error "archive changed ctime from $ctime0 to $ctime1"
1857 # Release should not change any timestamps.
1858 $LFS hsm_release $file || error "cannot release '$file'"
1859 check_hsm_flags $file "0x0000000d"
1861 atime1=$(stat -c "%X" $file)
1862 mtime1=$(stat -c "%Y" $file)
1863 ctime1=$(stat -c "%Z" $file)
1865 [ $atime0 -eq $atime1 ] ||
1866 error "release changed atime from $atime0 to $atime1"
1868 [ $mtime0 -eq $mtime1 ] ||
1869 error "release changed mtime from $mtime0 to $mtime1"
1871 [ $ctime0 -eq $ctime1 ] ||
1872 error "release changed ctime from $ctime0 to $ctime1"
1874 # Restore should not change any timestamps.
1875 $LFS hsm_restore $file
1876 wait_request_state $fid RESTORE SUCCEED
1878 atime1=$(stat -c "%X" $file)
1879 mtime1=$(stat -c "%Y" $file)
1880 ctime1=$(stat -c "%Z" $file)
1882 [ $atime0 -eq $atime1 ] ||
1883 error "restore changed atime from $atime0 to $atime1"
1885 [ $mtime0 -eq $mtime1 ] ||
1886 error "restore changed mtime from $mtime0 to $mtime1"
1888 [ $ctime0 -eq $ctime1 ] ||
1889 error "restore changed ctime from $ctime0 to $ctime1"
1892 wait_copytools || error "Copytools failed to stop"
1894 # Once more, after unmount and mount.
1895 umount_client $MOUNT || error "cannot unmount '$MOUNT'"
1896 mount_client $MOUNT || error "cannot mount '$MOUNT'"
1898 atime1=$(stat -c "%X" $file)
1899 mtime1=$(stat -c "%Y" $file)
1900 ctime1=$(stat -c "%Z" $file)
1902 [ $atime0 -eq $atime1 ] ||
1903 error "remount changed atime from $atime0 to $atime1"
1905 [ $mtime0 -eq $mtime1 ] ||
1906 error "remount changed mtime from $mtime0 to $mtime1"
1908 [ $ctime0 -eq $ctime1 ] ||
1909 error "remount changed ctime from $ctime0 to $ctime1"
1911 run_test 24a "Archive, release, and restore does not change a/mtime (i/o)"
1914 local file=$DIR/$tdir/$tfile
1920 # Test needs a running copytool.
1922 mkdir_on_mdt0 $DIR/$tdir
1924 # Check that root can do HSM actions on a regular user's file.
1925 fid=$(create_small_file $file)
1926 sum0=$(md5sum $file)
1928 chown $RUNAS_ID:$RUNAS_GID $file ||
1929 error "cannot chown '$file' to '$RUNAS_ID'"
1931 chmod ugo-w $DIR/$tdir ||
1932 error "cannot chmod '$DIR/$tdir'"
1934 $LFS hsm_archive $file
1935 wait_request_state $fid ARCHIVE SUCCEED
1937 $LFS hsm_release $file
1938 check_hsm_flags $file "0x0000000d"
1940 $LFS hsm_restore $file
1941 wait_request_state $fid RESTORE SUCCEED
1943 # Check that ordinary user can get HSM state.
1944 $RUNAS $LFS hsm_state $file ||
1945 error "user '$RUNAS_ID' cannot get HSM state of '$file'"
1947 $LFS hsm_release $file
1948 check_hsm_flags $file "0x0000000d"
1950 # Check that ordinary user can accessed released file.
1951 sum1=$($RUNAS md5sum $file) ||
1952 error "user '$RUNAS_ID' cannot read '$file'"
1954 [ "$sum0" == "$sum1" ] ||
1955 error "md5sum mismatch for '$file'"
1957 run_test 24b "root can archive, release, and restore user files"
1960 local file=$DIR/$tdir/$tfile
1961 local action=archive
1966 # test needs a running copytool
1968 mkdir_on_mdt0 $DIR/$tdir
1970 # Save the default masks and check that cleanup_24c will
1971 # restore the request masks correctly.
1972 user_save=$(get_hsm_param user_request_mask)
1973 stack_trap "set_hsm_param user_request_mask '$user_save'" EXIT
1974 group_save=$(get_hsm_param group_request_mask)
1975 stack_trap "set_hsm_param group_request_mask '$group_save'" EXIT
1976 other_save=$(get_hsm_param other_request_mask)
1977 stack_trap "set_hsm_param other_request_mask '$other_save'" EXIT
1979 [ "$user_save" == RESTORE ] ||
1980 error "user_request_mask is '$user_save' expected 'RESTORE'"
1981 [ "$group_save" == RESTORE ] ||
1982 error "group_request_mask is '$group_save' expected 'RESTORE'"
1983 [ "$other_save" == RESTORE ] ||
1984 error "other_request_mask is '$other_save' expected 'RESTORE'"
1987 create_small_file $file
1988 chown $RUNAS_ID:$GROUP $file ||
1989 error "cannot chown '$file' to '$RUNAS_ID:$GROUP'"
1991 $RUNAS $LFS hsm_$action $file &&
1992 error "$action by user should fail"
1994 set_hsm_param user_request_mask $action
1995 $RUNAS $LFS hsm_$action $file ||
1996 error "$action by user should succeed"
1999 create_small_file $file
2000 chown nobody:$RUNAS_GID $file ||
2001 error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
2003 $RUNAS $LFS hsm_$action $file &&
2004 error "$action by group should fail"
2006 set_hsm_param group_request_mask $action
2007 $RUNAS $LFS hsm_$action $file ||
2008 error "$action by group should succeed"
2011 create_small_file $file
2012 chown nobody:$GROUP $file ||
2013 error "cannot chown '$file' to 'nobody:$GROUP'"
2015 $RUNAS $LFS hsm_$action $file &&
2016 error "$action by other should fail"
2018 set_hsm_param other_request_mask $action
2019 $RUNAS $LFS hsm_$action $file ||
2020 error "$action by other should succeed"
2022 run_test 24c "check that user,group,other request masks work"
2025 local file1=$DIR/$tdir/$tfile
2026 local file2=$DIR2/$tdir/$tfile
2030 mkdir_on_mdt0 $DIR/$tdir
2032 fid1=$(create_small_file $file1)
2035 $LFS getstripe $file1
2037 stack_trap "zconf_umount \"$(facet_host $SINGLEAGT)\" \"$MOUNT3\"" EXIT
2038 zconf_mount "$(facet_host $SINGLEAGT)" "$MOUNT3" ||
2039 error "cannot mount '$MOUNT3' on '$SINGLEAGT'"
2041 copytool setup -m "$MOUNT3"
2043 stack_trap "mount -o remount,rw \"$MOUNT2\"" EXIT
2044 mount -o remount,ro $MOUNT2
2046 do_nodes $(comma_list $(nodes_list)) $LCTL clear
2048 fid2=$(path2fid $file2)
2049 [ "$fid1" == "$fid2" ] ||
2050 error "FID mismatch '$fid1' != '$fid2'"
2052 $LFS hsm_archive $file2 &&
2053 error "archive should fail on read-only mount"
2054 check_hsm_flags $file1 "0x00000000"
2056 $LFS hsm_archive $file1 || error "Fail to archive $file1"
2057 wait_request_state $fid1 ARCHIVE SUCCEED
2059 $LFS hsm_release $file1
2060 $LFS hsm_restore $file2
2061 wait_request_state $fid1 RESTORE SUCCEED
2063 $LFS hsm_release $file1 || error "cannot release '$file1'"
2064 dd if=$file2 of=/dev/null bs=1M || error "cannot read '$file2'"
2066 $LFS hsm_release $file2 &&
2067 error "release should fail on read-only mount"
2071 run_test 24d "check that read-only mounts are respected"
2075 mkdir_on_mdt0 $DIR/$tdir
2077 local f=$DIR/$tdir/$tfile
2080 fid=$(create_small_file $f) || error "cannot create $f"
2081 $LFS hsm_archive $f || error "cannot archive $f"
2082 wait_request_state $fid ARCHIVE SUCCEED
2083 $LFS hsm_release $f || error "cannot release $f"
2084 while ! $LFS hsm_state $f | grep released; do
2088 tar -cf $TMP/$tfile.tar $DIR/$tdir || error "cannot tar $DIR/$tdir"
2090 run_test 24e "tar succeeds on HSM released files" # LU-6213
2093 # test needs a running copytool
2095 mkdir_on_mdt0 $DIR/$tdir
2097 mkdir -p $DIR/$tdir/d1
2098 local f=$DIR/$tdir/$tfile
2099 local fid=$(copy_file /etc/hosts $f)
2102 $LFS hsm_archive $f ||
2103 error "hsm_archive failed"
2104 wait_request_state $fid ARCHIVE SUCCEED
2105 $LFS hsm_release $f || error "cannot release $f"
2106 tar --xattrs -cvf $f.tar -C $DIR/$tdir $tfile
2109 tar --xattrs -xvf $f.tar -C $DIR/$tdir ||
2110 error "Can not recover the tar contents"
2112 echo "Sum0 = $sum0, sum1 = $sum1"
2113 [ "$sum0" == "$sum1" ] || error "md5sum mismatch for '$tfile'"
2115 run_test 24f "root can archive, release, and restore tar files"
2118 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
2119 skip "need MDS version 2.11.56 or later"
2121 local file=$DIR/$tdir/$tfile
2124 echo "RUNAS = '$RUNAS'"
2128 mkdir_on_mdt0 $DIR/$tdir
2129 chmod ugo+rwx $DIR/$tdir
2131 echo "Please listen carefully as our options have changed." | tee $file
2132 fid=$(path2fid $file)
2135 $LFS hsm_archive $file
2136 wait_request_state $fid ARCHIVE SUCCEED
2137 check_hsm_flags $file 0x00000009 # exists archived
2139 echo "To be electrocuted by your telephone, press #." | $RUNAS tee $file
2140 check_hsm_flags $file 0x0000000b # exists dirty archived
2142 run_test 24g "write by non-owner still sets dirty" # LU-11369
2145 # test needs a running copytool
2149 copy2archive /etc/hosts $tdir/$tfile
2151 local f=$DIR/$tdir/$tfile
2153 copytool import $tdir/$tfile $f
2155 $LFS hsm_set --lost $f
2160 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2162 run_test 25a "Restore lost file (HS_LOST flag) from import"\
2163 " (Operation not permitted)"
2166 # test needs a running copytool
2169 mkdir_on_mdt0 $DIR/$tdir
2171 local f=$DIR/$tdir/$tfile
2172 local fid=$(copy_file /etc/passwd $f)
2174 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2175 wait_request_state $fid ARCHIVE SUCCEED
2178 $LFS hsm_set --lost $f
2182 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2184 run_test 25b "Restore lost file (HS_LOST flag) after release"\
2185 " (Operation not permitted)"
2187 test_26A() { # was test_26
2188 # test needs a running copytool
2190 mkdir_on_mdt0 $DIR/$tdir
2192 local f=$DIR/$tdir/$tfile
2193 local fid=$(create_empty_file "$f")
2195 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2196 wait_request_state $fid ARCHIVE SUCCEED
2199 wait_request_state $fid REMOVE SUCCEED
2201 check_hsm_flags $f "0x00000000"
2203 run_test 26A "Remove the archive of a valid file"
2206 local raolu=$(get_hsm_param remove_archive_on_last_unlink)
2207 [[ $raolu -eq 0 ]] || error "RAoLU policy should be off"
2209 # test needs a running copytool
2211 mkdir_on_mdt0 $DIR/$tdir
2213 local f=$DIR/$tdir/$tfile
2214 local fid=$(copy_file /etc/passwd $f)
2216 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2217 wait_request_state $fid ARCHIVE SUCCEED
2219 local f2=$DIR/$tdir/${tfile}_2
2220 local fid2=$(copy_file /etc/passwd $f2)
2222 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
2223 wait_request_state $fid2 ARCHIVE SUCCEED
2225 local f3=$DIR/$tdir/${tfile}_3
2226 local fid3=$(copy_file /etc/passwd $f3)
2228 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f3
2229 wait_request_state $fid3 ARCHIVE SUCCEED
2231 # set a long grace_delay vs short loop_period
2232 local orig_loop_period=$(get_hsm_param loop_period)
2233 local orig_grace_delay=$(get_hsm_param grace_delay)
2234 stack_trap "set_hsm_param loop_period $orig_loop_period" EXIT
2235 set_hsm_param loop_period 10
2236 stack_trap "set_hsm_param grace_delay $orig_grace_delay" EXIT
2237 set_hsm_param grace_delay 100
2241 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2242 set_hsm_param remove_archive_on_last_unlink 1
2244 ln "$f3" "$f3"_bis || error "Unable to create hard-link"
2249 wait_request_state $fid2 REMOVE SUCCEED
2251 assert_request_count $fid REMOVE 0 \
2252 "Unexpected archived data remove request for $f"
2253 assert_request_count $fid3 REMOVE 0 \
2254 "Unexpected archived data remove request for $f3"
2256 run_test 26a "Remove Archive On Last Unlink (RAoLU) policy"
2259 # test needs a running copytool
2261 mkdir_on_mdt0 $DIR/$tdir
2263 local f=$DIR/$tdir/$tfile
2264 local fid=$(copy_file /etc/passwd $f)
2266 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2267 wait_request_state $fid ARCHIVE SUCCEED
2269 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2270 set_hsm_param remove_archive_on_last_unlink 1
2273 cdt_check_state stopped
2277 wait_request_state $fid REMOVE WAITING
2281 # copytool must re-register
2283 wait_copytools || error "copytool failed to stop"
2286 wait_request_state $fid REMOVE SUCCEED
2288 run_test 26b "RAoLU policy when CDT off"
2291 # test needs a running copytool
2293 mkdir_on_mdt0 $DIR/$tdir
2295 local f=$DIR/$tdir/$tfile
2296 local fid=$(copy_file /etc/passwd $f)
2298 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2299 wait_request_state $fid ARCHIVE SUCCEED
2301 local f2=$DIR/$tdir/${tfile}_2
2302 local fid2=$(copy_file /etc/passwd $f2)
2304 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
2305 wait_request_state $fid2 ARCHIVE SUCCEED
2307 # set a long grace_delay vs short loop_period
2308 local orig_loop_period=$(get_hsm_param loop_period)
2309 local orig_grace_delay=$(get_hsm_param grace_delay)
2310 stack_trap "set_hsm_param loop_period $orig_loop_period" EXIT
2311 set_hsm_param loop_period 10
2312 stack_trap "set_hsm_param grace_delay $orig_grace_delay" EXIT
2313 set_hsm_param grace_delay 100
2315 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2316 set_hsm_param remove_archive_on_last_unlink 1
2318 multiop_bg_pause $f O_c || error "open $f failed"
2324 wait_request_state $fid2 REMOVE SUCCEED
2325 assert_request_count $fid REMOVE 0 \
2326 "Unexpected archived data remove request for $f"
2328 kill -USR1 $pid || error "multiop early exit"
2329 # should reach autotest timeout if multiop fails to trap
2330 # signal, close file, and exit ...
2331 wait $pid || error "wait PID $PID failed"
2333 wait_request_state $fid REMOVE SUCCEED
2335 run_test 26c "RAoLU effective when file closed"
2338 # test needs a running copytool
2340 mkdir_on_mdt0 $DIR/$tdir
2342 local f=$DIR/$tdir/$tfile
2343 local fid=$(create_small_file $f)
2345 $LFS hsm_archive $f || error "could not archive file"
2346 wait_request_state $fid ARCHIVE SUCCEED
2348 # set a long grace_delay vs short loop_period
2349 local orig_loop_period=$(get_hsm_param loop_period)
2350 local orig_grace_delay=$(get_hsm_param grace_delay)
2351 stack_trap "set_hsm_param loop_period $orig_loop_period" EXIT
2352 set_hsm_param loop_period 10
2353 stack_trap "set_hsm_param grace_delay $orig_grace_delay" EXIT
2354 set_hsm_param grace_delay 100
2356 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2357 set_hsm_param remove_archive_on_last_unlink 1
2359 multiop_bg_pause $f O_c || error "multiop failed"
2366 wait_request_state $fid REMOVE SUCCEED
2368 client_up || client_up || true
2370 kill -USR1 $MULTIPID
2371 wait $MULTIPID || error "multiop close failed"
2373 run_test 26d "RAoLU when Client eviction"
2376 # test needs a running copytool
2379 create_archive_file $tdir/$tfile
2380 local f=$DIR/$tdir/$tfile
2381 copytool import $tdir/$tfile $f
2382 local fid=$(path2fid $f)
2386 [[ $? != 0 ]] || error "Remove of a released file should fail"
2388 run_test 27a "Remove the archive of an imported file (Operation not permitted)"
2391 # test needs a running copytool
2393 mkdir_on_mdt0 $DIR/$tdir
2395 local f=$DIR/$tdir/$tfile
2396 local fid=$(create_empty_file "$f")
2398 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2399 wait_request_state $fid ARCHIVE SUCCEED
2404 [[ $? != 0 ]] || error "Remove of a released file should fail"
2406 run_test 27b "Remove the archive of a relased file (Operation not permitted)"
2409 # test needs a running copytool
2411 mkdir_on_mdt0 $DIR/$tdir
2413 local f=$DIR/$tdir/$tfile
2414 local fid=$(create_empty_file "$f")
2416 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2417 wait_request_state $fid ARCHIVE SUCCEED
2426 wait_request_state $fid REMOVE SUCCEED
2428 run_test 28 "Concurrent archive/file remove"
2431 # Tests --mntpath and --archive options
2434 copytool setup -m "$MOUNT" -a $archive_id
2436 # Bad archive number
2437 $LFS hsm_remove -m "$MOUNT" -a 33 0x857765760:0x8:0x2 2>&1 |
2438 grep "Invalid argument" ||
2439 error "unexpected hsm_remove failure (1)"
2441 # mntpath is present but file is given
2442 $LFS hsm_remove --mntpath "$MOUNT" --archive 30 /qwerty/uyt 2>&1 |
2443 grep "hsm: '/qwerty/uyt' is not a valid FID" ||
2444 error "unexpected hsm_remove failure (2)"
2446 run_test 29a "Tests --mntpath and --archive options"
2449 # test needs a running copytool
2451 mkdir_on_mdt0 $DIR/$tdir
2453 local f=$DIR/$tdir/$tfile
2454 local fid=$(create_small_file $f)
2456 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2457 wait_request_state $fid ARCHIVE SUCCEED
2461 $LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER $fid
2462 wait_request_state $fid REMOVE SUCCEED
2464 run_test 29b "Archive/delete/remove by FID from the archive."
2467 # test needs a running copytool
2469 mkdir_on_mdt0 $DIR/$tdir
2471 local fid1=$(create_small_file $DIR/$tdir/$tfile-1)
2472 local fid2=$(create_small_file $DIR/$tdir/$tfile-2)
2473 local fid3=$(create_small_file $DIR/$tdir/$tfile-3)
2475 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/$tfile-[1-3]
2476 wait_request_state $fid1 ARCHIVE SUCCEED
2477 wait_request_state $fid2 ARCHIVE SUCCEED
2478 wait_request_state $fid3 ARCHIVE SUCCEED
2480 rm -f $DIR/$tdir/$tfile-[1-3]
2482 echo $fid1 > $DIR/$tdir/list
2483 echo $fid2 >> $DIR/$tdir/list
2484 echo $fid3 >> $DIR/$tdir/list
2486 $LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER \
2487 --filelist $DIR/$tdir/list
2488 wait_request_state $fid1 REMOVE SUCCEED
2489 wait_request_state $fid2 REMOVE SUCCEED
2490 wait_request_state $fid3 REMOVE SUCCEED
2492 run_test 29c "Archive/delete/remove by FID, using a file list."
2495 # test needs more than one CT
2496 needclients 3 || return 0
2502 # start all of the copytools
2503 for n in $(seq $AGTCOUNT); do
2504 copytool setup -f agt$n -a $n
2507 mkdir_on_mdt0 $DIR/$tdir
2510 file=$DIR/$tdir/$tfile
2511 fid=$(create_small_file $file)
2513 $LFS hsm_archive $file
2514 wait_request_state $fid ARCHIVE SUCCEED
2515 check_hsm_flags $file "0x00000009"
2519 $LFS hsm_remove --mntpath "$MOUNT" -a 0 $fid ||
2520 error "cannot hsm_remove '$fid'"
2522 # give time for CDT to handle remove request and create broadcasted
2525 # remove request has been broadcasted ?
2526 local cnt=$(get_request_count $fid REMOVE)
2527 # broadcasted requests + original
2528 [[ $cnt -eq $((AGTCOUNT + 1)) ]] ||
2529 error "remove not broadcasted to all CTs"
2531 # give time for CDT and CTs to handle broadcasted
2532 wait_for_loop_period
2534 # each agent serves one different archive_id, so broadcasted
2535 # hsm_remove request should only succeed once and fail at all others
2539 for n in $(seq $AGTCOUNT); do
2540 res=$(do_facet $SINGLEMDS "$LCTL get_param -n \
2541 $HSM_PARAM.actions | awk \
2542 '/'$fid'.*action=REMOVE archive#='$n'/ \
2543 {print \\\$13}' | cut -f2 -d=")
2544 if [[ "$res" == "SUCCEED" ]]; then
2546 elif [[ "$res" == "FAILED" ]]; then
2551 [[ $scnt -eq 1 ]] ||
2552 error "one and only CT should have removed successfully"
2554 [[ $AGTCOUNT -eq $((scnt + fcnt)) ]] ||
2555 error "all but one CT should have failed to remove"
2557 run_test 29d "hsm_remove by FID with archive_id 0 for unlinked file cause "\
2558 "request to be sent once for each registered archive_id"
2561 # restore at exec cannot work on agent node (because of Linux kernel
2562 # protection of executables)
2563 needclients 2 || return 0
2565 # test needs a running copytool
2569 copy2archive /bin/true $tdir/$tfile
2571 local f=$DIR/$tdir/true
2572 copytool import $tdir/$tfile $f
2574 local fid=$(path2fid $f)
2576 stack_trap "cdt_clear_no_retry" EXIT
2577 # set no retry action mode
2584 [[ $st == 0 ]] || error "Failed to exec a released file"
2586 run_test 30a "Restore at exec (import case)"
2589 # restore at exec cannot work on agent node (because of Linux kernel
2590 # protection of executables)
2591 needclients 2 || return 0
2593 # test needs a running copytool
2596 mkdir_on_mdt0 $DIR/$tdir
2597 local f=$DIR/$tdir/true
2598 local fid=$(copy_file /bin/true $f)
2600 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2601 wait_request_state $fid ARCHIVE SUCCEED
2605 stack_trap cdt_clear_no_retry EXIT
2606 # set no retry action mode
2614 [[ $st == 0 ]] || error "Failed to exec a released file"
2616 run_test 30b "Restore at exec (release case)"
2619 needclients 2 || return 0
2621 # test needs a running copytool
2624 mkdir_on_mdt0 $DIR/$tdir
2625 local f=$DIR/$tdir/SLEEP
2626 local slp_sum1=$(md5sum /bin/sleep)
2627 local fid=$(copy_file /bin/sleep $f)
2629 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2630 wait_request_state $fid ARCHIVE SUCCEED
2632 check_hsm_flags $f "0x0000000d"
2634 stack_trap cdt_clear_no_retry EXIT
2635 # set no retry action mode
2638 do_node $CLIENT2 "$f 10" &
2642 [[ $? == 0 ]] && error "Update during exec of released file must fail"
2644 [[ $? == 0 ]] || error "Execution failed during run"
2646 if [[ $? != 0 ]]; then
2647 local slp_sum2=$(md5sum /bin/sleep)
2648 # in case sleep file is modified during the test
2649 [[ $slp_sum1 == $slp_sum2 ]] &&
2650 error "Binary overwritten during exec"
2653 check_hsm_flags $f "0x00000009"
2655 run_test 30c "Update during exec of released file must fail"
2657 restore_and_check_size() {
2660 local s=$(stat -c "%s" $f)
2662 local st=$(get_hsm_flags $f)
2666 while [[ "$st" != "0x00000009" && $cpt -le 10 ]]
2668 n=$(stat -c "%s" $f)
2669 # we echo in both cases to show stat is not hang
2670 if [[ $n != $s ]]; then
2671 echo "size seen is $n != $s"
2674 echo "size seen is right: $n == $s"
2678 st=$(get_hsm_flags $f)
2680 if [[ "$st" = "0x00000009" ]]; then
2683 echo " restore is too long"
2684 wait_request_state $fid RESTORE SUCCEED
2690 # test needs a running copytool
2692 mkdir_on_mdt0 $DIR/$tdir
2694 create_archive_file $tdir/$tfile
2695 local f=$DIR/$tdir/$tfile
2696 copytool import $tdir/$tfile $f
2697 local fid=$($LFS path2fid $f)
2700 restore_and_check_size $f $fid
2703 [[ $err -eq 0 ]] || error "File size changed during restore"
2705 run_test 31a "Import a large file and check size during restore"
2709 # test needs a running copytool
2711 mkdir_on_mdt0 $DIR/$tdir
2713 local f=$DIR/$tdir/$tfile
2714 local fid=$(create_file "$f" 1MB 39)
2716 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2717 wait_request_state $fid ARCHIVE SUCCEED
2720 restore_and_check_size $f $fid
2723 [[ $err -eq 0 ]] || error "File size changed during restore"
2725 run_test 31b "Restore a large unaligned file and check size during restore"
2728 # test needs a running copytool
2730 mkdir_on_mdt0 $DIR/$tdir
2732 local f=$DIR/$tdir/$tfile
2733 local fid=$(create_file "$f" 1M 39)
2735 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2736 wait_request_state $fid ARCHIVE SUCCEED
2739 restore_and_check_size $f $fid
2742 [[ $err -eq 0 ]] || error "File size changed during restore"
2744 run_test 31c "Restore a large aligned file and check size during restore"
2747 local f=$DIR/$tdir/$tfile
2749 mkdir_on_mdt0 $DIR/$tdir
2750 local fid=$(create_empty_file "$f")
2754 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2755 wait_request_state $fid ARCHIVE SUCCEED
2758 # Prevent restore from completing
2762 md5sum $f >/dev/null &
2765 wait_request_state $fid RESTORE STARTED
2770 # Check restore trigger process was killed
2772 [ $? -eq 143 ] || error "md5sum was not 'Terminated'"
2774 run_test 33 "Kill a restore waiting process"
2777 # test needs a running copytool
2779 mkdir_on_mdt0 $DIR/$tdir
2781 local f=$DIR/$tdir/$tfile
2782 local fid=$(create_empty_file "$f")
2784 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2785 wait_request_state $fid ARCHIVE SUCCEED
2788 # Prevent restore from completing
2791 md5sum $f >/dev/null &
2794 wait_request_state $fid RESTORE STARTED
2796 # rm must not block during restore
2797 timeout --signal=KILL 1 rm "$f" || error "rm $f failed"
2800 wait_request_state $fid RESTORE SUCCEED
2802 # Check md5sum pgm finished
2803 kill -0 $pid && error "Restore initiatior still running"
2804 wait $pid || error "Restore initiator failed with $?"
2806 # Check the file was actually deleted
2807 [ ! -f "$f" ] || error "$f was not deleted"
2809 run_test 34 "Remove file during restore"
2812 # test needs a running copytool
2814 mkdir_on_mdt0 $DIR/$tdir
2816 local f=$DIR/$tdir/$tfile
2817 local f1=$DIR/$tdir/$tfile-1
2818 local fid=$(create_empty_file "$f")
2819 local fid1=$(copy_file /etc/passwd $f1)
2821 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2822 wait_request_state $fid ARCHIVE SUCCEED
2825 # Prevent restore from completing
2828 md5sum $f >/dev/null &
2831 wait_request_state $fid RESTORE STARTED
2833 # mv must not block during restore
2834 timeout --signal=KILL 1 mv "$f1" "$f" || error "mv $f1 $f failed"
2837 wait_request_state $fid RESTORE SUCCEED
2839 # Check md5sum pgm finished
2840 kill -0 $pid && error "Restore initiatior still running"
2841 wait $pid || error "Restore initiator failed with $?"
2843 local fid2=$(path2fid $f)
2844 [[ $fid2 == $fid1 ]] || error "Wrong fid after mv $fid2 != $fid1"
2846 run_test 35 "Overwrite file during restore"
2849 # test needs a running copytool
2851 mkdir_on_mdt0 $DIR/$tdir
2853 local f=$DIR/$tdir/$tfile
2854 local fid=$(create_empty_file "$f")
2856 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2857 wait_request_state $fid ARCHIVE SUCCEED
2860 # Prevent restore from completing
2863 md5sum $f >/dev/null &
2866 wait_request_state $fid RESTORE STARTED
2868 # mv must not block during restore
2869 timeout --signal=KILL 10 mv "$f" "$f.new" ||
2870 error "mv '$f' '$f.new' failed with rc=$?"
2873 wait_request_state $fid RESTORE SUCCEED
2875 # Check md5sum pgm finished
2876 kill -0 $pid && error "Restore initiator is still running"
2877 wait $pid || error "Restore initiator failed with $?"
2879 run_test 36 "Move file during restore"
2882 # LU-5683: check that an archived dirty file can be rearchived.
2884 mkdir_on_mdt0 $DIR/$tdir
2886 local f=$DIR/$tdir/$tfile
2889 fid=$(create_small_file $f) || error "cannot create small file"
2891 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2892 wait_request_state $fid ARCHIVE SUCCEED
2893 $LFS hsm_release $f || error "cannot release $f"
2895 # Allow previous archive request to expire from the actions log.
2896 wait_for_grace_delay
2899 dd if=/dev/urandom of=$f bs=1M count=1 || error "cannot dirty file"
2901 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2902 wait_request_state $fid ARCHIVE SUCCEED
2904 run_test 37 "re-archive a dirty file"
2911 for n in $(seq 1 $count); do
2912 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $prefix.$n
2914 echo "$count archive requests submitted"
2918 local stream_count=4
2919 local file_count=100
2921 local f=$DIR/$tdir/$tfile
2925 local max_requests=$(get_hsm_param max_requests)
2927 stack_trap "set_hsm_param max_requests $max_requests" EXIT
2928 # Increase the number of HSM request that can be performed in
2929 # parallel. With the coordinator running once per second, this
2930 # also limits the number of requests per seconds that can be
2931 # performed, so we pick a decent number. But we also need to keep
2932 # that number low because the copytool has no rate limit and will
2933 # fail some requests if if gets too many at once.
2934 set_hsm_param max_requests 300
2936 for i in $(seq 1 $file_count); do
2937 for p in $(seq 1 $stream_count); do
2938 fid=$(copy_file /etc/hosts $f.$p.$i)
2944 # to be sure wait_all_done will not be mislead by previous tests
2946 wait_for_grace_delay
2948 # start archive streams in background (archive files in parallel)
2949 for p in $(seq 1 $stream_count); do
2950 multi_archive $f.$p $file_count &
2953 echo -n "Wait for all requests being enqueued..."
2958 run_test 40 "Parallel archive requests"
2960 hsm_archive_batch() {
2968 while [ $i -lt $files_num ]; do
2969 if [ $batch_num -eq $batch_max ]; then
2970 $LFS hsm_archive $fileset || error "HSM archive failed"
2971 # Reset the batch container.
2976 fileset+="${filebase}$i "
2977 batch_num=$(( batch_num + 1 ))
2981 if [ $batch_num -ne 0 ]; then
2982 $LFS hsm_archive $fileset || error "HSM archive failed"
2989 local dir=$DIR/$tdir
2992 set_hsm_param max_requests 1000000
2993 mkdir $dir || error "mkdir $dir failed"
3002 filebase="$dir/$tfile.start."
3003 createmany -m $filebase $files_num ||
3004 error "createmany -m $filebase failed: $?"
3007 hsm_archive_batch $files_num $batch_max "$filebase"
3008 elapsed=$((SECONDS - start))
3009 do_facet $SINGLEMDS "$LCTL get_param -n \
3010 $HSM_PARAM.actions | grep WAITING | wc -l"
3011 unlinkmany $filebase $files_num || error "unlinkmany $filabase failed"
3012 echo "Start Phase files_num: $files_num time: $elapsed"
3015 filebase="$dir/$tfile.in."
3016 createmany -m $filebase $files_num ||
3017 error "createmany -m $filebase failed: $?"
3019 hsm_archive_batch $files_num $batch_max "$filebase"
3020 elapsed=$((SECONDS - start))
3021 unlinkmany $filebase $files_num || error "unlinkmany $filabase failed"
3022 echo "Middle Phase files_num: $files_num time: $elapsed"
3025 filebase="$dir/$tfile.end."
3026 createmany -m $filebase $files_num ||
3027 error "createmany -m $filebase failed: $?"
3030 hsm_archive_batch $files_num $batch_max "$filebase"
3031 elapsed=$((SECONDS - start))
3032 do_facet $SINGLEMDS "$LCTL get_param -n \
3033 $HSM_PARAM.actions | grep WAITING | wc -l"
3035 unlinkmany $filebase $files_num || error "unlinkmany $filebase failed"
3036 echo "End Phase files_num: $files_num time: $elapsed"
3038 do_facet $SINGLEMDS "$LCTL get_param -n \
3039 $HSM_PARAM.actions | grep WAITING | wc -l"
3043 run_test 50 "Archive with large number of pending HSM actions"
3046 # test needs a running copytool
3049 mkdir_on_mdt0 $DIR/$tdir
3050 local f=$DIR/$tdir/$tfile
3051 local fid=$(create_small_file $f)
3053 $LFS hsm_archive $f || error "could not archive file"
3054 wait_request_state $fid ARCHIVE SUCCEED
3055 check_hsm_flags $f "0x00000009"
3057 multiop_bg_pause $f O_c || error "multiop failed"
3061 client_up || client_up || true
3063 kill -USR1 $MULTIPID
3064 wait $MULTIPID || error "multiop close failed"
3066 check_hsm_flags $f "0x0000000b"
3068 run_test 52 "Opened for write file on an evicted client should be set dirty"
3071 # test needs a running copytool
3074 mkdir_on_mdt0 $DIR/$tdir
3075 local f=$DIR/$tdir/$tfile
3076 local fid=$(create_small_file $f)
3078 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3079 error "could not archive file"
3080 wait_request_state $fid ARCHIVE SUCCEED
3081 check_hsm_flags $f "0x00000009"
3083 multiop_bg_pause $f o_c || error "multiop failed"
3087 client_up || client_up || true
3089 kill -USR1 $MULTIPID
3090 wait $MULTIPID || error "multiop close failed"
3092 check_hsm_flags $f "0x00000009"
3094 run_test 53 "Opened for read file on an evicted client should not be set dirty"
3097 mkdir_on_mdt0 $DIR/$tdir
3099 local f=$DIR/$tdir/$tfile
3100 local fid=$(create_file "$f" 1MB 39)
3104 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3105 error "could not archive file"
3106 wait_request_state $fid ARCHIVE STARTED
3108 check_hsm_flags $f "0x00000001"
3110 stack_trap "cdt_clear_no_retry" EXIT
3111 # Avoid coordinator resending this request as soon it has failed.
3116 wait_request_state $fid ARCHIVE FAILED
3118 check_hsm_flags $f "0x00000003"
3120 run_test 54 "Write during an archive cancels it"
3123 mkdir_on_mdt0 $DIR/$tdir
3125 local f=$DIR/$tdir/$tfile
3126 local fid=$(create_file "$f" 1MB 39)
3130 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3131 error "could not archive file"
3132 wait_request_state $fid ARCHIVE STARTED
3134 check_hsm_flags $f "0x00000001"
3136 stack_trap "cdt_clear_no_retry" EXIT
3137 # Avoid coordinator resending this request as soon it has failed.
3140 $TRUNCATE $f 1024 || error "truncate failed"
3142 wait_request_state $fid ARCHIVE FAILED
3144 check_hsm_flags $f "0x00000003"
3146 run_test 55 "Truncate during an archive cancels it"
3149 mkdir_on_mdt0 $DIR/$tdir
3151 local f=$DIR/$tdir/$tfile
3152 local fid=$(create_file "$f" 1MB 39)
3156 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3157 error "could not archive file"
3158 wait_request_state $fid ARCHIVE STARTED
3160 check_hsm_flags $f "0x00000001"
3162 # Change metadata and sync to be sure we are not changing only
3167 wait_request_state $fid ARCHIVE SUCCEED
3169 check_hsm_flags $f "0x00000009"
3171 run_test 56 "Setattr during an archive is ok"
3174 # Need one client for I/O, one for request
3175 needclients 2 || return 0
3177 # test needs a running copytool
3180 mkdir_on_mdt0 $DIR/$tdir
3181 local f=$DIR/$tdir/test_archive_remote
3182 # Create a file on a remote node
3183 do_node $CLIENT2 "dd if=/dev/urandom of=$f bs=1M "\
3184 "count=2 conv=fsync"
3187 do_node $CLIENT2 "$LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f" ||
3188 error "hsm_archive failed"
3189 local fid=$(path2fid $f)
3190 wait_request_state $fid ARCHIVE SUCCEED
3192 # Release and implicit restore it
3193 do_node $CLIENT2 "$LFS hsm_release $f" ||
3194 error "hsm_release failed"
3195 do_node $CLIENT2 "md5sum $f" ||
3196 error "hsm_restore failed"
3198 wait_request_state $fid RESTORE SUCCEED
3200 run_test 57 "Archive a file with dirty cache on another node"
3202 truncate_released_file() {
3206 local sz=$(stat -c %s $src_file)
3207 local f=$DIR/$tdir/$tfile
3208 local fid=$(copy_file $1 $f)
3212 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3213 error "could not archive file"
3214 wait_request_state $fid ARCHIVE SUCCEED
3216 $LFS hsm_release $f || error "could not release file"
3218 $TRUNCATE $f $trunc_to || error "truncate failed"
3221 local sz1=$(stat -c %s $f)
3222 [[ $sz1 == $trunc_to ]] ||
3223 error "size after trunc: $sz1 expect $trunc_to, original $sz"
3226 check_hsm_flags $f "0x0000000b"
3228 local state=$(get_request_state $fid RESTORE)
3229 [[ "$state" == "SUCCEED" ]] ||
3230 error "truncate $sz does not trig restore, state = $state"
3232 $TRUNCATE $ref $trunc_to
3233 cmp $ref $f || error "file data wrong after truncate"
3239 # test needs a running copytool
3242 mkdir_on_mdt0 $DIR/$tdir
3244 local sz=$(stat -c %s /etc/passwd)
3246 echo "truncate up from $sz to $((sz*2))"
3247 truncate_released_file /etc/passwd $((sz*2))
3249 echo "truncate down from $sz to $((sz/2))"
3250 truncate_released_file /etc/passwd $((sz/2))
3252 echo "truncate to 0"
3253 truncate_released_file /etc/passwd 0
3255 run_test 58 "Truncate a released file will trigger restore"
3259 [[ $MDS1_VERSION -lt $(version_code 2.7.63) ]] &&
3260 skip "Need MDS version at least 2.7.63"
3263 $MCREATE $DIR/$tfile || error "mcreate failed"
3264 $TRUNCATE $DIR/$tfile 42 || error "truncate failed"
3265 $LFS hsm_archive $DIR/$tfile || error "archive request failed"
3266 fid=$(path2fid $DIR/$tfile)
3267 wait_request_state $fid ARCHIVE SUCCEED
3268 $LFS hsm_release $DIR/$tfile || error "release failed"
3270 run_test 59 "Release stripeless file with non-zero size"
3273 # This test validates the fix for LU-4512. Ensure that the -u
3274 # option changes the progress reporting interval from the
3275 # default (30 seconds) to the user-specified interval.
3276 mkdir_on_mdt0 $DIR/$tdir
3278 local f=$DIR/$tdir/$tfile
3279 local fid=$(create_file "$f" 1M 10)
3282 local progress_timeout=$((interval * 4))
3283 copytool setup -b 1 --update-interval $interval
3286 local mdt=${MDT_PREFIX}${mdtidx}
3287 local mds=mds$((mdtidx + 1))
3289 # Wait for copytool to register
3290 wait_update_facet $mds \
3291 "$LCTL get_param -n ${mdt}.hsm.agents | grep -o ^uuid" \
3292 uuid 100 || error "coyptool failed to register with $mdt"
3294 local start_at=$(date +%s)
3295 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3296 error "could not archive file"
3298 local agent=$(facet_active_host $SINGLEAGT)
3299 local logfile=$(copytool_logfile $SINGLEAGT)
3301 wait_update $agent \
3302 "grep -o start.copy \"$logfile\"" "start copy" 100 ||
3303 error "copytool failed to start"
3305 local cmd="$LCTL get_param -n ${mdt}.hsm.active_requests"
3306 cmd+=" | awk '/'$fid'.*action=ARCHIVE/ {print \\\$12}' | cut -f2 -d="
3312 echo -n "Expecting a progress update within $progress_timeout seconds... "
3314 RESULT=$(do_node $(facet_active_host $mds) "$cmd")
3315 if [ -n "$RESULT" ] && [ "$RESULT" -gt 0 ]; then
3316 echo "$RESULT bytes copied in $WAIT seconds."
3318 elif [ $WAIT -ge $progress_timeout ]; then
3319 error "Timed out waiting for progress update!"
3322 WAIT=$((WAIT + sleep))
3326 local finish_at=$(date +%s)
3327 local elapsed=$((finish_at - start_at))
3329 # Ensure that the progress update occurred within the expected window.
3330 if [ $elapsed -lt $((interval - 1)) ]; then
3331 error "Expected progress update after at least $interval seconds"
3334 echo "Wait for on going archive hsm action to complete"
3335 wait_update $agent "grep -o copied \"$logfile\"" "copied" 10 ||
3336 echo "File archiving not completed even after 10 secs"
3338 run_test 60 "Changing progress update interval from default"
3341 # test needs a running copytool
3344 mkdir_on_mdt0 $DIR/$tdir
3345 local f=$DIR/$tdir/$tfile
3346 local fid=$(copy_file /etc/passwd $f)
3348 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3351 wait_request_state $fid ARCHIVE FAILED
3353 run_test 61 "Waiting archive of a removed file should fail"
3356 # test needs a new running copytool
3357 stack_trap copytool_monitor_cleanup EXIT
3358 copytool_monitor_setup
3359 copytool setup --event-fifo "$HSMTOOL_MONITOR_DIR/fifo"
3361 # Wait for the copytool to register.
3362 wait_update --verbose $(facet_active_host mds1) \
3363 "$LCTL get_param -n ${MDT_PREFIX}0.hsm.agents | grep -o ^uuid" \
3365 error "copytool failed to register with MDT0000"
3368 wait_copytools || error "Copytools failed to stop"
3370 local REGISTER_EVENT
3371 local UNREGISTER_EVENT
3372 while read event; do
3373 local parsed=$(parse_json_event "$event")
3374 if [ -z "$parsed" ]; then
3375 error "Copytool sent malformed event: $event"
3379 if [ $event_type == "REGISTER" ]; then
3380 REGISTER_EVENT=$event
3381 elif [ $event_type == "UNREGISTER" ]; then
3382 UNREGISTER_EVENT=$event
3384 done < <(echo $"$(get_copytool_event_log)")
3386 if [ -z "$REGISTER_EVENT" ]; then
3387 error "Copytool failed to send register event to FIFO"
3390 if [ -z "$UNREGISTER_EVENT" ]; then
3391 error "Copytool failed to send unregister event to FIFO"
3394 echo "Register/Unregister events look OK."
3396 run_test 70 "Copytool logs JSON register/unregister events to FIFO"
3399 # Bump progress interval for livelier events.
3402 # test needs a new running copytool
3403 stack_trap copytool_monitor_cleanup EXIT
3404 copytool_monitor_setup
3405 copytool setup --update-interval $interval --event-fifo \
3406 "$HSMTOOL_MONITOR_DIR/fifo"
3408 stack_trap "cdt_clear_no_retry" EXIT
3409 # Just start and stop the copytool to generate events.
3412 mkdir_on_mdt0 $DIR/$tdir
3414 local f=$DIR/$tdir/$tfile
3415 local fid=$(create_small_file "$f")
3417 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3418 error "could not archive file"
3419 wait_request_state $fid ARCHIVE SUCCEED
3421 local expected_fields="event_time data_fid source_fid"
3422 expected_fields+=" total_bytes current_bytes"
3425 [ARCHIVE_START]=false
3426 [ARCHIVE_FINISH]=false
3427 [ARCHIVE_RUNNING]=false
3429 while read event; do
3430 # Make sure we're not getting anything from previous events.
3431 for field in $expected_fields; do
3435 local parsed=$(parse_json_event "$event")
3436 if [ -z "$parsed" ]; then
3437 error "Copytool sent malformed event: $event"
3441 events["$event_type"]=true
3443 [ "$event_type" != ARCHIVE_RUNNING ] && continue
3445 # Do some simple checking of the progress update events.
3446 for expected_field in $expected_fields; do
3447 if [ -z ${!expected_field+x} ]; then
3448 error "Missing $expected_field field in event"
3452 [ $total_bytes -gt 0 ] || error "Expected total_bytes to be > 0"
3454 # These should be identical throughout an archive operation
3455 [ $source_fid == $data_fid ] ||
3456 error "Expected source_fid to equal data_fid"
3457 done < <(echo $"$(get_copytool_event_log)")
3459 # Check we received every type of events we were expecting
3460 for event in "${!events[@]}"; do
3461 ${events["$event"]} ||
3462 error "Copytool failed to send '$event' event to FIFO"
3465 echo "Archive events look OK."
3467 run_test 71 "Copytool logs JSON archive events to FIFO"
3470 # Bump progress interval for livelier events.
3473 # test needs a new running copytool
3474 stack_trap copytool_monitor_cleanup EXIT
3475 copytool_monitor_setup
3476 copytool setup --update-interval $interval --event-fifo \
3477 "$HSMTOOL_MONITOR_DIR/fifo"
3478 local test_file=$HSMTOOL_MONITOR_DIR/file
3480 local cmd="dd if=/dev/urandom of=$test_file count=16 bs=1000000 "
3482 do_facet $SINGLEAGT "$cmd" ||
3483 error "cannot create $test_file on $SINGLEAGT"
3484 copy2archive $test_file $tdir/$tfile
3486 mkdir_on_mdt0 $DIR/$tdir
3487 local f=$DIR/$tdir/$tfile
3488 copytool import $tdir/$tfile $f
3489 f=$DIR2/$tdir/$tfile
3490 echo "Verifying released state: "
3491 check_hsm_flags $f "0x0000000d"
3493 local fid=$(path2fid $f)
3495 wait_request_state $fid RESTORE SUCCEED
3497 local expected_fields="event_time data_fid source_fid"
3498 expected_fields+=" total_bytes current_bytes"
3502 while read event; do
3503 # Make sure we're not getting anything from previous events.
3504 for field in $expected_fields; do
3508 local parsed=$(parse_json_event "$event")
3509 if [ -z "$parsed" ]; then
3510 error "Copytool sent malformed event: $event"
3514 if [ $event_type == "RESTORE_START" ]; then
3516 if [ $source_fid != $data_fid ]; then
3517 error "source_fid should == data_fid at start"
3520 elif [ $event_type == "RESTORE_FINISH" ]; then
3522 if [ $source_fid != $data_fid ]; then
3523 error "source_fid should == data_fid at finish"
3526 elif [ $event_type != "RESTORE_RUNNING" ]; then
3530 # Do some simple checking of the progress update events.
3531 for expected_field in $expected_fields; do
3532 if [ -z ${!expected_field+x} ]; then
3533 error "Missing $expected_field field in event"
3537 if [ $total_bytes -eq 0 ]; then
3538 error "Expected total_bytes to be > 0"
3541 # When a restore starts out, the data fid is the same as the
3542 # source fid. After the restore has gotten going, we learn
3543 # the new data fid. Once the restore has finished, the source
3544 # fid is set to the new data fid.
3546 # We test this because some monitoring software may depend on
3547 # this behavior. If it changes, then the consumers of these
3548 # events may need to be modified.
3549 if [ $source_fid == $data_fid ]; then
3550 error "source_fid should != data_fid during restore"
3552 done < <(echo $"$(get_copytool_event_log)")
3554 if [ -z "$START_EVENT" ]; then
3555 error "Copytool failed to send restore start event to FIFO"
3558 if [ -z "$FINISH_EVENT" ]; then
3559 error "Copytool failed to send restore finish event to FIFO"
3562 echo "Restore events look OK."
3564 run_test 72 "Copytool logs JSON restore events to FIFO"
3567 file_count=51 # Max number of files constrained by LNET message size
3568 mkdir_on_mdt0 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3569 local f=$DIR/$tdir/$tfile
3570 local FILELIST=/tmp/filelist.txt
3574 for i in $(seq 1 $file_count); do
3575 fid=$(copy_file /etc/hosts $f.$i)
3576 echo $f.$i >> $FILELIST
3580 # to be sure wait_all_done will not be mislead by previous tests
3582 wait_for_grace_delay
3583 $LFS hsm_archive --filelist $FILELIST ||
3584 error "cannot archive a file list"
3586 $LFS hsm_release --filelist $FILELIST ||
3587 error "cannot release a file list"
3588 $LFS hsm_restore --filelist $FILELIST ||
3589 error "cannot restore a file list"
3592 run_test 90 "Archive/restore a file list"
3594 double_verify_reset_hsm_param() {
3596 echo "Testing $HSM_PARAM.$p"
3597 local val=$(get_hsm_param $p)
3599 local val2=$(($val * 2))
3600 set_hsm_param $p $val2
3601 val=$(get_hsm_param $p)
3602 [[ $val == $val2 ]] ||
3603 error "$HSM_PARAM.$p: $val != $val2 should be (2 * $save)"
3604 echo "Set $p to 0 must failed"
3608 set_hsm_param $p $save
3610 if [[ $rc == 0 ]]; then
3611 error "we must not be able to set $HSM_PARAM.$p to 0"
3616 double_verify_reset_hsm_param loop_period
3617 double_verify_reset_hsm_param grace_delay
3618 double_verify_reset_hsm_param active_request_timeout
3619 double_verify_reset_hsm_param max_requests
3620 double_verify_reset_hsm_param default_archive_id
3622 run_test 100 "Set coordinator /proc tunables"
3629 run_test 102 "Verify coordinator control"
3632 # test needs a running copytool
3639 for i in $(seq 1 20); do
3640 fid=$(copy_file /etc/passwd $DIR/$tdir/$i)
3642 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/*
3646 echo "Current requests"
3647 local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3648 $HSM_PARAM.actions |\
3649 grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
3651 [[ -z "$res" ]] || error "Some request have not been canceled"
3653 run_test 103 "Purge all requests"
3656 (( MDS1_VERSION >= $(version_code 2.14.56) )) ||
3657 skip "Need MDS version at least 2.14.56"
3659 cdt_clear_non_blocking_restore
3661 # test needs a running copytool
3666 local rpcs_inflight=$($LCTL get_param -n \
3667 "mdc.$(facet_svc mds1)*.max_rpcs_in_flight" |
3670 mkdir_on_mdt0 $DIR/$tdir
3671 for ((i=0; i < rpcs_inflight; i++)); do
3672 fids+=( $(copy_file /etc/passwd $DIR/$tdir/${tfile}_$i) )
3674 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/*
3678 local grep_regex="($(tr ' ' '|' <<< "${fids[*]}")).*action=ARCHIVE.*status=SUCCEED"
3680 while [[ $time -lt 5 ]] && [[ $cnt -ne ${#fids[@]} ]]; do
3681 cnt=$(do_facet mds1 "$LCTL get_param -n $HSM_PARAM.actions |
3682 grep -c -E '$grep_regex'")
3686 [[ $cnt -eq ${#fids[@]} ]] || error "Fail to archive files $cnt/${#fids[@]}"
3688 $LFS hsm_release $DIR/$tdir/*
3691 wait_copytools || error "Copytool failed to stop"
3694 for i in "${fids[@]}"; do
3695 cat $DIR/.lustre/fid/$i > /dev/null & pids+=($!)
3699 grep_regex="($(tr ' ' '|' <<< "${fids[*]}")).*action=RESTORE.*status=CANCELED"
3700 cnt=$(do_facet mds1 "$LCTL get_param -n $HSM_PARAM.actions |
3701 grep -cE '$grep_regex'")
3703 [[ "$cnt" -eq ${#fids[@]} ]] ||
3704 error "Some request have not been canceled ($cnt/${#fids[@]} canceled)"
3706 # cat cmds should not hang and should fail
3707 for i in "${!pids[@]}"; do
3709 error "Restore for ${tfile}_$i (${pids[$i]}) should fail" ||
3713 run_test 103a "Purge pending restore requests"
3718 mkdir_on_mdt0 $DIR/$tdir
3720 local f=$DIR/$tdir/$tfile
3721 local fid=$(create_empty_file "$f")
3723 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
3724 local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3725 $HSM_PARAM.actions |\
3726 grep $fid | cut -f16 -d=")
3728 [[ "$data1" == "$DATAHEX" ]] ||
3729 error "Data field in records is ($data1) and not ($DATAHEX)"
3733 run_test 104 "Copy tool data field"
3736 local max_requests=$(get_hsm_param max_requests)
3737 mkdir_on_mdt0 $DIR/$tdir
3740 stack_trap "set_hsm_param max_requests $max_requests" EXIT
3741 set_hsm_param max_requests 300
3744 for i in $(seq -w 1 10); do
3745 cp /etc/passwd $DIR/$tdir/$i
3746 $LFS hsm_archive $DIR/$tdir/$i
3748 local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3749 $HSM_PARAM.actions |\
3750 grep WAITING | wc -l")
3754 local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3755 $HSM_PARAM.actions |\
3756 grep WAITING | wc -l")
3759 [[ "$reqcnt1" == "$reqcnt2" ]] ||
3760 error "Requests count after shutdown $reqcnt2 != "\
3761 "before shutdown $reqcnt1"
3763 run_test 105 "Restart of coordinator"
3766 # test needs a running copytool
3769 local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
3771 check_agent_registered $uuid
3773 search_copytools || error "No copytool found"
3776 wait_copytools || error "Copytool failed to stop"
3778 check_agent_unregistered $uuid
3781 uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
3782 check_agent_registered $uuid
3784 run_test 106 "Copytool register/unregister"
3787 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
3789 # test needs a running copytool
3791 # create and archive file
3792 mkdir_on_mdt0 $DIR/$tdir
3793 local f1=$DIR/$tdir/$tfile
3794 local fid=$(copy_file /etc/passwd $f1)
3795 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
3796 wait_request_state $fid ARCHIVE SUCCEED
3797 # shutdown and restart MDS
3799 # check the copytool still gets messages from MDT
3800 local f2=$DIR/$tdir/2
3801 local fid=$(copy_file /etc/passwd $f2)
3802 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
3803 # main check of this sanity: this request MUST succeed
3804 wait_request_state $fid ARCHIVE SUCCEED
3806 run_test 107 "Copytool re-register after MDS restart"
3808 policy_set_and_test()
3812 do_facet $SINGLEMDS $LCTL set_param "$HSM_PARAM.policy=\\\"$change\\\""
3813 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3814 [[ "$policy" == "$target" ]] ||
3815 error "Wrong policy after '$change': '$policy' != '$target'"
3819 # to force default policy setting if error
3820 CDT_POLICY_HAD_CHANGED=true
3822 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3823 local default="NonBlockingRestore [NoRetryAction]"
3824 [[ "$policy" == "$default" ]] ||
3825 error "default policy has changed,"\
3826 " '$policy' != '$default' update the test"
3827 policy_set_and_test "+NBR" "[NonBlockingRestore] [NoRetryAction]"
3828 policy_set_and_test "+NRA" "[NonBlockingRestore] [NoRetryAction]"
3829 policy_set_and_test "-NBR" "NonBlockingRestore [NoRetryAction]"
3830 policy_set_and_test "-NRA" "NonBlockingRestore NoRetryAction"
3831 policy_set_and_test "NRA NBR" "[NonBlockingRestore] [NoRetryAction]"
3832 # useless bacause we know but safer for futur changes to use real value
3833 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3834 echo "Next set_param must failed"
3835 policy_set_and_test "wrong" "$policy"
3838 echo "Back to default policy"
3839 cdt_set_sanity_policy
3841 run_test 109 "Policy display/change"
3844 # test needs a running copytool
3847 mkdir_on_mdt0 $DIR/$tdir
3849 copy2archive /etc/passwd $tdir/$tfile
3851 local f=$DIR/$tdir/$tfile
3852 copytool import $tdir/$tfile $f
3853 local fid=$(path2fid $f)
3855 cdt_set_non_blocking_restore
3860 wait_request_state $fid RESTORE SUCCEED
3861 cdt_clear_non_blocking_restore
3865 error "md5sum returns $st != 1, "\
3866 "should also perror ENODATA (No data available)"
3868 run_test 110a "Non blocking restore policy (import case)"
3871 # test needs a running copytool
3874 mkdir_on_mdt0 $DIR/$tdir
3875 local f=$DIR/$tdir/$tfile
3876 local fid=$(copy_file /etc/passwd $f)
3877 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3878 wait_request_state $fid ARCHIVE SUCCEED
3881 cdt_set_non_blocking_restore
3886 wait_request_state $fid RESTORE SUCCEED
3887 cdt_clear_non_blocking_restore
3891 error "md5sum returns $st != 1, "\
3892 "should also perror ENODATA (No data available)"
3894 run_test 110b "Non blocking restore policy (release case)"
3897 # test needs a running copytool
3900 mkdir_on_mdt0 $DIR/$tdir
3901 copy2archive /etc/passwd $tdir/$tfile
3903 local f=$DIR/$tdir/$tfile
3905 copytool import $tdir/$tfile $f
3906 local fid=$(path2fid $f)
3910 copytool_remove_backend $fid
3913 wait_request_state $fid RESTORE FAILED
3920 [[ $st == 0 ]] || error "Restore does not failed"
3922 run_test 111a "No retry policy (import case), restore will error"\
3923 " (No such file or directory)"
3926 # test needs a running copytool
3929 mkdir_on_mdt0 $DIR/$tdir
3930 local f=$DIR/$tdir/$tfile
3931 local fid=$(copy_file /etc/passwd $f)
3932 stack_trap cdt_clear_no_retry EXIT
3934 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3935 wait_request_state $fid ARCHIVE SUCCEED
3938 copytool_remove_backend $fid
3941 wait_request_state $fid RESTORE FAILED
3945 [[ $st == 0 ]] || error "Restore does not failed"
3947 run_test 111b "No retry policy (release case), restore will error"\
3948 " (No such file or directory)"
3951 # test needs a running copytool
3954 mkdir_on_mdt0 $DIR/$tdir
3955 local f=$DIR/$tdir/$tfile
3956 local fid=$(copy_file /etc/passwd $f)
3958 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3959 local l=$($LFS hsm_action $f)
3961 local res=$(echo $l | cut -f 2- -d" " | grep ARCHIVE)
3964 wait_request_state $fid ARCHIVE SUCCEED
3967 [[ ! -z "$res" ]] || error "action is $l which is not an ARCHIVE"
3969 run_test 112 "State of recorded request"
3972 mkdir_on_mdt0 $DIR/$tdir
3974 local file1=$DIR/$tdir/$tfile
3975 local file2=$DIR2/$tdir/$tfile
3977 local fid=$(create_small_sync_file $file1)
3979 stack_trap "zconf_umount \"$(facet_host $SINGLEAGT)\" \"$MOUNT3\"" EXIT
3980 zconf_mount "$(facet_host $SINGLEAGT)" "$MOUNT3" ||
3981 error "cannot mount '$MOUNT3' on '$SINGLEAGT'"
3983 copytool setup -m "$MOUNT3"
3985 do_nodes $(comma_list $(nodes_list)) $LCTL clear
3987 $LFS hsm_archive $file1 || error "Fail to archive $file1"
3988 wait_request_state $fid ARCHIVE SUCCEED
3990 $LFS hsm_release $file1
3991 echo "Verifying released state: "
3992 check_hsm_flags $file1 "0x0000000d"
3994 multiop_bg_pause $file1 oO_WRONLY:O_APPEND:_w4c || error "multiop failed"
3997 kill -USR1 $MULTIPID
4002 local size1=$(stat -c "%s" $file1)
4003 local size2=$(stat -c "%s" $file2)
4005 [ $size1 -eq $size2 ] || error "sizes are different $size1 $size2"
4007 run_test 113 "wrong stat after restore"
4010 (( MDS1_VERSION >= $(version_code 2.15.54) )) ||
4011 skip "need MDS version at least 2.15.54"
4013 mkdir_on_mdt0 $DIR/$tdir
4015 local f1=$DIR/$tdir/${tfile}1
4016 local f2=$DIR/$tdir/${tfile}2
4017 local fid1=$(create_empty_file "$f1")
4018 local fid2=$(create_empty_file "$f2")
4022 # Prevent archive from completing
4025 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1 $f2
4026 # wait archive to register at CDT
4027 wait_request_state "$fid1" ARCHIVE WAITING
4029 # Set f2 in an incompatible state
4030 $LFS hsm_set --noarchive $f2
4034 wait_request_state "$fid1" ARCHIVE SUCCEED
4035 wait_request_state "$fid2" ARCHIVE FAILED
4037 run_test 114 "Incompatible request does not set other requests as STARTED"
4040 mkdir_on_mdt0 $DIR/$tdir
4042 local f=$DIR/$tdir/$tfile
4043 local fid=$(create_empty_file "$f")
4047 # Prevent archive from completing
4050 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4051 # wait archive to register at CDT
4052 wait_request_state $fid ARCHIVE STARTED
4054 # Cancel the archive
4055 $LFS hsm_cancel "$f"
4057 wait_request_state $fid ARCHIVE CANCELED
4060 wait_request_state $fid CANCEL SUCCEED
4062 run_test 200 "Register/Cancel archive"
4065 # test needs a running copytool
4067 mkdir_on_mdt0 $DIR/$tdir
4069 local f=$DIR/$tdir/$tfile
4070 create_archive_file $tdir/$tfile
4071 copytool import $tdir/$tfile $f
4072 local fid=$(path2fid $f)
4074 # test with cdt on is made in test_222
4077 # wait restore to register at CDT
4078 wait_request_state $fid RESTORE WAITING
4081 wait_request_state $fid RESTORE CANCELED
4082 wait_request_state $fid CANCEL SUCCEED
4084 run_test 201 "Register/Cancel restore"
4087 mkdir_on_mdt0 $DIR/$tdir
4089 local f=$DIR/$tdir/$tfile
4090 local fid=$(create_empty_file "$f")
4092 # test needs a running copytool
4095 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4096 wait_request_state $fid ARCHIVE SUCCEED
4100 # wait remove to register at CDT
4101 wait_request_state $fid REMOVE STARTED
4104 wait_request_state $fid REMOVE CANCELED
4106 run_test 202 "Register/Cancel remove"
4108 test_220A() { # was test_220
4109 # test needs a running copytool
4112 mkdir_on_mdt0 $DIR/$tdir
4114 local f=$DIR/$tdir/$tfile
4115 local fid=$(copy_file /etc/passwd $f)
4119 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4120 wait_request_state $fid ARCHIVE SUCCEED
4122 changelog_find -type HSM -target-fid $fid -flags 0x0 ||
4123 error "The expected changelog was not emitted"
4125 run_test 220A "Changelog for archive"
4128 # test needs a running copytool
4131 mkdir_on_mdt0 $DIR/$tdir
4133 local f=$DIR/$tdir/$tfile
4134 local fid=$(copy_file /etc/passwd $f)
4138 # block copytool operations to allow for HSM request to be
4139 # submitted and file be unlinked (CDT will find object removed)
4142 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4144 # wait request to reach CT
4145 wait_request_state $fid ARCHIVE STARTED
4151 wait_request_state $fid ARCHIVE FAILED
4154 changelog_find -type HSM -target-fid $fid -flags 0x2 ||
4155 error "The expected changelog was not emitted"
4157 run_test 220a "Changelog for failed archive"
4160 mkdir_on_mdt0 $DIR/$tdir
4162 local f=$DIR/$tdir/$tfile
4163 local fid=$(create_empty_file "$f")
4168 # Prevent archive from completing
4170 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4171 wait_request_state $fid ARCHIVE STARTED
4174 wait_request_state $fid ARCHIVE CANCELED
4177 wait_request_state $fid CANCEL SUCCEED
4179 changelog_find -type HSM -target-fid $fid -flags 0x7d ||
4180 error "The expected changelog was not emitted"
4182 run_test 221 "Changelog for archive canceled"
4185 # test needs a running copytool
4188 mkdir_on_mdt0 $DIR/$tdir
4189 copy2archive /etc/passwd $tdir/$tfile
4191 local f=$DIR/$tdir/$tfile
4192 copytool import $tdir/$tfile $f
4193 local fid=$(path2fid $f)
4198 wait_request_state $fid RESTORE SUCCEED
4200 changelog_find -type HSM -target-fid $fid -flags 0x80 ||
4201 error "The expected changelog was not emitted"
4203 run_test 222a "Changelog for explicit restore"
4206 # test needs a running copytool
4209 mkdir_on_mdt0 $DIR/$tdir
4210 local f=$DIR/$tdir/$tfile
4211 local fid=$(copy_file /etc/passwd $f)
4214 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4215 wait_request_state $fid ARCHIVE SUCCEED
4220 wait_request_state $fid RESTORE SUCCEED
4222 changelog_find -type HSM -target-fid $fid -flags 0x80 ||
4223 error "The expected changelog was not emitted"
4225 run_test 222b "Changelog for implicit restore"
4228 # test needs a running copytool
4231 mkdir_on_mdt0 $DIR/$tdir
4232 copy2archive /etc/passwd $tdir/$tfile
4234 local f=$DIR/$tdir/$tfile
4235 copytool import $tdir/$tfile $f
4236 local fid=$(path2fid $f)
4240 # block copytool operations to allow for HSM request to be
4241 # submitted and file be unlinked (CDT will find object removed)
4246 # wait request to reach CT
4247 wait_request_state $fid RESTORE STARTED
4253 wait_request_state $fid RESTORE FAILED
4256 changelog_find -type HSM -target-fid $fid -flags 0x82 ||
4257 error "The expected changelog was not emitted"
4259 run_test 222c "Changelog for failed explicit restore"
4262 # test needs a running copytool
4265 mkdir_on_mdt0 $DIR/$tdir
4266 local f=$DIR/$tdir/$tfile
4267 local fid=$(copy_file /etc/passwd $f)
4270 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4271 wait_request_state $fid ARCHIVE SUCCEED
4274 copytool_remove_backend $fid
4277 wait_request_state $fid RESTORE FAILED
4280 changelog_find -type HSM -target-fid $fid -flags 0x82 ||
4281 error "The expected changelog was not emitted"
4283 run_test 222d "Changelog for failed implicit restore"
4286 # test needs a running copytool
4288 mkdir_on_mdt0 $DIR/$tdir
4290 local f=$DIR/$tdir/$tfile
4291 create_archive_file $tdir/$tfile
4295 copytool import $tdir/$tfile $f
4296 local fid=$(path2fid $f)
4299 wait_request_state $fid RESTORE STARTED
4301 wait_request_state $fid RESTORE CANCELED
4302 wait_request_state $fid CANCEL SUCCEED
4304 changelog_find -type HSM -target-fid $fid -flags 0xfd ||
4305 error "The expected changelog was not emitted"
4307 run_test 223a "Changelog for restore canceled (import case)"
4310 mkdir_on_mdt0 $DIR/$tdir
4312 local f=$DIR/$tdir/$tfile
4313 local fid=$(create_empty_file "$f")
4318 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4319 wait_request_state $fid ARCHIVE SUCCEED
4322 # Prevent restore from completing
4325 wait_request_state $fid RESTORE STARTED
4328 wait_request_state $fid RESTORE CANCELED
4331 wait_request_state $fid CANCEL SUCCEED
4333 changelog_find -type HSM -target-fid $fid -flags 0xfd ||
4334 error "The expected changelog was not emitted"
4336 run_test 223b "Changelog for restore canceled (release case)"
4338 test_224A() { # was test_224
4339 # test needs a running copytool
4342 mkdir_on_mdt0 $DIR/$tdir
4344 local f=$DIR/$tdir/$tfile
4345 local fid=$(copy_file /etc/passwd $f)
4348 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4349 wait_request_state $fid ARCHIVE SUCCEED
4352 wait_request_state $fid REMOVE SUCCEED
4354 changelog_find -type HSM -target-fid $fid -flags 0x200 ||
4355 error "The expected changelog was not emitted"
4357 run_test 224A "Changelog for remove"
4360 # test needs a running copytool
4363 mkdir_on_mdt0 $DIR/$tdir
4365 local f=$DIR/$tdir/$tfile
4366 local fid=$(copy_file /etc/passwd $f)
4369 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4370 wait_request_state $fid ARCHIVE SUCCEED
4372 copytool_remove_backend $fid
4374 # block copytool operations to allow for HSM request to be
4375 # submitted and file be unlinked (CDT will find object removed)
4380 # wait for request to reach CT
4381 wait_request_state $fid REMOVE STARTED
4387 wait_request_state $fid REMOVE FAILED
4389 # HE_REMOVE|ENOENT=0x202
4390 changelog_find -type HSM -target-fid $fid -flags 0x202 ||
4391 error "The expected changelog was not emitted"
4393 run_test 224a "Changelog for failed remove"
4396 # test is not usable because remove request is too fast
4397 # so it is always finished before cancel can be done ...
4398 echo "Test disabled"
4401 # test needs a running copytool
4403 mkdir_on_mdt0 $DIR/$tdir
4405 local f=$DIR/$tdir/$tfile
4406 local fid=$(create_empty_file "$f")
4409 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4410 wait_request_state $fid ARCHIVE SUCCEED
4412 # Prevent restore from completing
4417 wait_request_state $fid REMOVE CANCELED
4420 wait_request_state $fid CANCEL SUCCEED
4422 changelog_find -type HSM -target-fid $fid -flags 0x27d
4423 error "The expected changelog was not emitted"
4425 run_test 225 "Changelog for remove canceled"
4428 # test needs a running copytool
4431 mkdir_on_mdt0 $DIR/$tdir
4433 local f1=$DIR/$tdir/$tfile-1
4434 local f2=$DIR/$tdir/$tfile-2
4435 local f3=$DIR/$tdir/$tfile-3
4436 local fid1=$(copy_file /etc/passwd $f1)
4437 local fid2=$(copy_file /etc/passwd $f2)
4438 copy_file /etc/passwd $f3
4441 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
4442 wait_request_state $fid1 ARCHIVE SUCCEED
4444 $LFS hsm_archive $f2
4445 wait_request_state $fid2 ARCHIVE SUCCEED
4447 rm $f1 || error "rm $f1 failed"
4450 changelog_find -type UNLNK -target-fid $fid1 -flags 0x3 ||
4451 error "The expected changelog was not emitted"
4453 mv $f3 $f2 || error "mv $f3 $f2 failed"
4455 changelog_find -type RENME -target-fid $fid2 -flags 0x3 ||
4456 error "The expected changelog was not emitted"
4458 run_test 226 "changelog for last rm/mv with exiting archive"
4460 # This is just a utility function to clarify what test_227 does
4465 "$LFS" "$action" --$flag "$file" ||
4466 error "Cannot ${action#hsm_} $flag on '$file'"
4468 # Only one changelog should be produced
4469 local entries="$(changelog_find -type HSM -target-fid $fid)"
4470 [ $(wc -l <<< "$entries") -eq $((++count)) ] ||
4471 error "lfs $action --$flag '$file' produced more than one" \
4474 # Parse the last changelog record
4475 local entry="$(tail -n 1 <<< "$entries")"
4476 eval local -A changelog=$(changelog2array $entry)
4478 # Also check the flags match what is expected
4479 [[ ${changelog[flags]} == $target ]] ||
4480 error "Changelog flag is '${changelog[flags]}', not $target"
4484 local file="$DIR/$tdir/$tfile"
4485 local fid=$(create_empty_file "$file")
4490 for flag in norelease noarchive exists archived lost; do
4491 if [ "$flag" == lost ]; then
4492 # The flag "lost" only works on an archived file
4493 "$LFS" hsm_set --archived "$file"
4497 action="hsm_set" __test_227
4498 action="hsm_clear" __test_227
4501 run_test 227 "changelog when explicit setting of HSM flags"
4504 # test needs a running copytool
4506 mkdir_on_mdt0 $DIR/$tdir
4508 local fid=$(create_small_sync_file $DIR/$tfile)
4509 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
4510 wait_request_state $fid ARCHIVE SUCCEED
4512 $LFS hsm_release $DIR/$tfile
4513 check_hsm_flags $DIR/$tfile "0x0000000d"
4515 filefrag $DIR/$tfile | grep " 1 extent found" ||
4516 error "filefrag on released file must return only one extent"
4518 # only newer versions of cp detect sparse files by stat/FIEMAP
4520 cp --sparse=auto $DIR/$tfile $DIR/$tfile.2 ||
4521 error "copying $DIR/$tfile"
4522 cmp $DIR/$tfile $DIR/$tfile.2 || error "comparing copied $DIR/$tfile"
4524 $LFS hsm_release $DIR/$tfile
4525 check_hsm_flags $DIR/$tfile "0x0000000d"
4527 mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
4529 tar cf - --sparse $DIR/$tfile | tar xvf - -C $DIR/$tdir ||
4531 cmp $DIR/$tfile $DIR/$tdir/$DIR/$tfile ||
4532 error "comparing untarred $DIR/$tfile"
4534 rm -f $DIR/$tfile $DIR/$tfile.2 ||
4535 error "rm $DIR/$tfile or $DIR/$tfile.2 failed"
4537 run_test 228 "On released file, return extend to FIEMAP. For [cp,tar] --sparse"
4540 local file="$DIR/$tdir/$tfile"
4542 # set max_requests to allow one request of each type to be started (3)
4544 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4545 set_hsm_param max_requests 3
4548 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4549 set_hsm_param loop_period 1
4551 # send 1 requests of each kind twice
4554 mkdir_on_mdt0 $DIR/$tdir
4557 for action in archive restore remove; do
4558 local filepath="$file"-to-$action
4559 local fid=$(create_empty_file "$filepath")
4560 local fid2=$(create_empty_file "$filepath".bis)
4562 if [ "$action" != archive ]; then
4563 "$LFS" hsm_archive "$filepath"
4564 wait_request_state $fid ARCHIVE SUCCEED
4565 "$LFS" hsm_archive "$filepath".bis
4566 wait_request_state $fid2 ARCHIVE SUCCEED
4568 if [ "$action" == restore ]; then
4569 "$LFS" hsm_release "$filepath"
4570 "$LFS" hsm_release "$filepath".bis
4574 # suspend the copytool to prevent requests from completing
4575 stack_trap "copytool_continue" EXIT
4578 # send `max_requests' requests (one of each kind)
4579 for action in archive restore remove; do
4580 filepath="$file"-to-$action
4581 "$LFS" hsm_${action} "$filepath"
4582 wait_request_state $(path2fid "$filepath") "${action^^}" STARTED
4585 # send another batch of requests
4586 for action in archive restore remove; do
4587 "$LFS" hsm_${action} "$file-to-$action".bis
4589 # wait for `loop_period' seconds to make sure the coordinator has time
4590 # to register those, even though it should not
4593 # only the first batch of request should be started
4595 count=$(do_facet $SINGLEMDS "$LCTL" get_param -n $HSM_PARAM.actions |
4599 error "expected 3 STARTED requests, found $count"
4601 run_test 250 "Coordinator max request"
4604 mkdir_on_mdt0 $DIR/$tdir
4606 local f=$DIR/$tdir/$tfile
4607 local fid=$(create_empty_file "$f")
4610 # to have a short test
4611 local old_to=$(get_hsm_param active_request_timeout)
4612 set_hsm_param active_request_timeout 1
4613 # to be sure the cdt will wake up frequently so
4614 # it will be able to cancel the "old" request
4615 local old_loop=$(get_hsm_param loop_period)
4616 set_hsm_param loop_period 1
4621 # Prevent archive from completing
4623 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4624 wait_request_state $fid ARCHIVE STARTED
4626 # Let the request timeout
4627 wait_request_state $fid ARCHIVE CANCELED
4629 set_hsm_param active_request_timeout $old_to
4630 set_hsm_param loop_period $old_loop
4632 run_test 251 "Coordinator request timeout"
4635 mkdir_on_mdt0 $DIR/$tdir
4637 local f=$DIR/$tdir/$tfile
4638 local fid=$(create_empty_file "$f")
4640 # to have a short test
4641 stack_trap "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4642 set_hsm_param loop_period 1
4646 # Prevent archive from completing
4648 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4649 wait_request_state $fid ARCHIVE STARTED
4652 stack_trap "set_hsm_param active_request_timeout \
4653 $(get_hsm_param active_request_timeout)" EXIT
4654 set_hsm_param active_request_timeout 1
4656 wait_request_state $fid ARCHIVE CANCELED
4659 run_test 252 "Timeout'ed running archive of a removed file should be canceled"
4663 # test needs a running copytool
4666 mkdir_on_mdt0 $DIR/$tdir
4667 local f=$DIR/$tdir/$tfile
4669 dd if=/dev/zero of=$f bs=1MB count=10
4670 local fid=$(path2fid $f)
4672 $LFS hsm_archive $f || error "could not archive file"
4673 wait_request_state $fid ARCHIVE SUCCEED
4675 # clear locks to discard inode data
4676 cancel_lru_locks osc
4678 #define OBD_FAIL_MDC_MERGE 0x807
4679 $LCTL set_param fail_loc=0x807
4681 #expect error here, instead of release with wrong size
4684 if ((rc == 0)); then
4685 file_size=$(stat -c '%s' $f)
4686 if ((file_size != 10485760)); then
4687 error "Wrong file size after hsm_release"
4690 echo "could not release file"
4693 run_test 253 "Check for wrong file size after release"
4697 [ $MDS1_VERSION -lt $(version_code 2.10.56) ] &&
4698 skip "need MDS version at least 2.10.56"
4700 # Check that the counters are initialized to 0
4702 for request_type in archive restore remove; do
4703 count="$(get_hsm_param ${request_type}_count)" ||
4704 error "Reading ${request_type}_count failed with $?"
4706 [ "$count" -eq 0 ] ||
4707 error "Expected ${request_type}_count to be " \
4711 run_test 254a "Request counters are initialized to zero"
4715 [ $MDS1_VERSION -lt $(version_code 2.10.56) ] &&
4716 skip "need MDS version at least 2.10.56"
4718 # The number of request to launch (at least 32)
4719 local request_count=$((RANDOM % 32 + 32))
4720 printf "Will launch %i requests of each type\n" "$request_count"
4722 # Launch a copytool to process requests
4725 # Set hsm.max_requests to allow starting all requests at the same time
4727 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4728 set_hsm_param max_requests "$request_count"
4730 mkdir_on_mdt0 $DIR/$tdir
4734 for request_type in archive restore remove; do
4735 printf "Checking %s requests\n" "${request_type}"
4736 # Suspend the copytool to give us time to read the proc files
4739 for ((i = 0; i < $request_count; i++)); do
4740 case $request_type in
4742 create_empty_file "$DIR/$tdir/$tfile-$i" \
4746 lfs hsm_release "$DIR/$tdir/$tfile-$i"
4749 $LFS hsm_${request_type} "$DIR/$tdir/$tfile-$i"
4752 # Give the coordinator 10 seconds to start every request
4754 while get_hsm_param actions | grep -q WAITING; do
4757 [ $timeout -gt 0 ] ||
4758 error "${request_type^} requests took too " \
4762 count="$(get_hsm_param ${request_type}_count)"
4763 [ "$count" -eq "$request_count" ] ||
4764 error "Expected '$request_count' (!= '$count') " \
4765 "active $request_type requests"
4767 # Let the copytool process the requests
4769 # Give it 10 seconds maximum
4771 while get_hsm_param actions | grep -q STARTED; do
4774 [ $timeout -gt 0 ] ||
4775 error "${request_type^} requests took too " \
4779 count="$(get_hsm_param ${request_type}_count)"
4780 [ "$count" -eq 0 ] ||
4781 error "Expected 0 (!= '$count') " \
4782 "active $request_type requests"
4785 run_test 254b "Request counters are correctly incremented and decremented"
4789 [ $MDS1_VERSION -lt $(version_code 2.12.0) ] &&
4790 skip "Need MDS version at least 2.12.0"
4792 mkdir_on_mdt0 $DIR/$tdir
4794 local file="$DIR/$tdir/$tfile"
4795 local fid=$(create_empty_file "$file")
4797 # How do you make sure the coordinator has consumed any outstanding
4798 # event, without triggering an event yourself?
4800 # You wait for a request to disappear from the coordinator's llog.
4802 # Warning: the setup represents 90% of this test
4804 # Create and process an HSM request
4806 "$LFS" hsm_archive "$file"
4807 wait_request_state $fid ARCHIVE SUCCEED
4810 wait_copytools || error "failed to stop copytools"
4812 # Launch a new HSM request
4814 create_empty_file "$file"
4815 "$LFS" hsm_archive "$file"
4819 # Have the completed request be removed as soon as the cdt wakes up
4820 stack_trap "set_hsm_param grace_delay $(get_hsm_param grace_delay)" EXIT
4821 set_hsm_param grace_delay 1
4822 # (Hopefully, time on the MDS will behave nicely)
4823 do_facet $SINGLEMDS sleep 2 &
4825 # Increase `loop_period' as a mean to prevent the coordinator from
4826 # waking itself up to do some housekeeping.
4827 stack_trap "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4828 set_hsm_param loop_period 1000
4830 wait $! || error "waiting failed"
4832 wait_request_state $fid ARCHIVE ""
4833 # The coordinator will not wake up on its own for ~`loop_period' secs...
4835 # ... Unless a copytool registers. Now the real test begins
4837 wait_request_state $(path2fid "$file") ARCHIVE SUCCEED
4839 run_test 255 "Copytool registration wakes the coordinator up"
4841 # tests 260[a-c] rely on the parsing of the copytool's log file, they might
4842 # break in the future because of that.
4845 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
4846 skip "need MDS version 2.11.56 or later"
4848 local -a files=("$DIR/$tdir/$tfile".{0..15})
4851 mkdir_on_mdt0 $DIR/$tdir
4853 for file in "${files[@]}"; do
4854 create_small_file "$file"
4857 # Set a few hsm parameters
4859 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4860 set_hsm_param loop_period 1
4862 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4863 set_hsm_param max_requests 3
4867 "$LFS" hsm_archive "${files[0]}"
4868 wait_request_state "$(path2fid "${files[0]}")" ARCHIVE SUCCEED
4869 "$LFS" hsm_release "${files[0]}"
4873 wait_copytools || error "copytools failed to stop"
4875 # Send several archive requests
4876 for file in "${files[@]:1}"; do
4877 "$LFS" hsm_archive "$file"
4880 # Send one restore request
4881 "$LFS" hsm_restore "${files[0]}"
4886 # Wait for all the requests to complete
4887 wait_request_state "$(path2fid "${files[0]}")" RESTORE SUCCEED
4888 for file in "${files[@]:1}"; do
4889 wait_request_state "$(path2fid "$file")" ARCHIVE SUCCEED
4892 # Collect the actions in the order in which the copytool processed them
4894 $(do_facet "$SINGLEAGT" grep -o '\"RESTORE\\|ARCHIVE\"' \
4895 "$(copytool_logfile "$SINGLEAGT")")
4898 printf '%s\n' "${actions[@]}"
4901 for action in "${actions[@]:0:3}"; do
4902 [ "$action" == RESTORE ] && return
4905 error "Too many ARCHIVE requests were run before the RESTORE request"
4907 run_test 260a "Restore request have priority over other requests"
4909 # This test is very much tied to the implementation of the current priorisation
4910 # mechanism in the coordinator. It might not make sense to keep it in the future
4913 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
4914 skip "need MDS version 2.11.56 or later"
4916 local -a files=("$DIR/$tdir/$tfile".{0..15})
4919 mkdir_on_mdt0 $DIR/$tdir
4921 for file in "${files[@]}"; do
4922 create_small_file "$file"
4925 # Set a few hsm parameters
4927 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4928 set_hsm_param loop_period 1
4930 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4931 set_hsm_param max_requests 3
4934 copytool setup --archive-id 2
4935 "$LFS" hsm_archive --archive 2 "${files[0]}"
4936 wait_request_state "$(path2fid "${files[0]}")" ARCHIVE SUCCEED
4937 "$LFS" hsm_release "${files[0]}"
4941 wait_copytools || error "copytools failed to stop"
4943 # Send several archive requests
4944 for file in "${files[@]:1}"; do
4945 "$LFS" hsm_archive "$file"
4948 # Send one restore request
4949 "$LFS" hsm_restore "${files[0]}"
4953 copytool setup --archive-id 2
4955 # Wait for all the requests to complete
4956 wait_request_state "$(path2fid "${files[0]}")" RESTORE SUCCEED
4957 for file in "${files[@]:1}"; do
4958 wait_request_state "$(path2fid "$file")" ARCHIVE SUCCEED
4961 # Collect the actions in the order in which the copytool processed them
4963 $(do_facet "$SINGLEAGT" grep -o '\"RESTORE\\|ARCHIVE\"' \
4964 "$(copytool_logfile "$SINGLEAGT")")
4967 printf '%s\n' "${actions[@]}"
4970 for action in "${actions[@]:0:3}"; do
4971 [ "$action" == RESTORE ] && return
4974 error "Too many ARCHIVE requests were run before the RESTORE request"
4976 run_test 260b "Restore request have priority over other requests"
4978 # This test is very much tied to the implementation of the current priorisation
4979 # mechanism in the coordinator. It might not make sense to keep it in the future
4982 [ $MDS1_VERSION -lt $(version_code 2.12.0) ] &&
4983 skip "Need MDS version at least 2.12.0"
4985 local -a files=("$DIR/$tdir/$tfile".{0..15})
4988 mkdir_on_mdt0 $DIR/$tdir
4990 for file in "${files[@]}"; do
4991 create_small_file "$file"
4994 # Set a few hsm parameters
4996 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4997 set_hsm_param loop_period 1000
4999 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
5000 set_hsm_param max_requests 3
5003 copytool setup --archive-id 2
5004 "$LFS" hsm_archive --archive 2 "${files[0]}"
5005 wait_request_state "$(path2fid "${files[0]}")" ARCHIVE SUCCEED
5006 "$LFS" hsm_release "${files[0]}"
5010 wait_copytools || error "copytools failed to stop"
5012 # Force the next coordinator run to do housekeeping
5016 "$LFS" hsm_archive "${files[1]}"
5020 copytool setup --archive-id 2
5022 wait_request_state "$(path2fid "${files[1]}")" ARCHIVE SUCCEED
5023 # The coordinator just did a housekeeping run it won't do another one
5024 # for around `loop_period' seconds => requests will not be reordered
5025 # if it costs too much (ie. when the coordinator has to discard a whole
5028 # Send several archive requests
5029 for file in "${files[@]:2}"; do
5030 "$LFS" hsm_archive "$file"
5033 # Send one restore request
5034 "$LFS" hsm_restore "${files[0]}"
5036 # Wait for all the requests to complete
5037 wait_request_state "$(path2fid "${files[0]}")" RESTORE SUCCEED
5038 for file in "${files[@]:2}"; do
5039 wait_request_state "$(path2fid "$file")" ARCHIVE SUCCEED
5042 # Collect the actions in the order in which the copytool processed them
5044 $(do_facet "$SINGLEAGT" grep -o '\"RESTORE\\|ARCHIVE\"' \
5045 "$(copytool_logfile "$SINGLEAGT")")
5048 printf '%s\n' "${actions[@]}"
5051 for action in "${actions[@]:0:3}"; do
5052 [ "$action" == RESTORE ] &&
5053 error "Restore requests should not be prioritised" \
5054 "unless the coordinator is doing housekeeping"
5058 run_test 260c "Requests are not reordered on the 'hot' path of the coordinator"
5061 local file=$DIR/$tdir/$tfile
5066 mkdir_on_mdt0 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
5068 dd if=/dev/zero of=$file bs=4k count=2 || error "Write $file failed"
5069 fid=$(path2fid $file)
5070 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $file
5071 wait_request_state $fid ARCHIVE SUCCEED
5073 $LFS hsm_state $file
5074 $LFS hsm_release $file
5075 $LFS hsm_restore $file
5076 wait_request_state $fid RESTORE SUCCEED
5077 $LFS hsm_release $file
5078 size=$(stat -c %s $file)
5079 [[ $size == 8192 ]] || error "Size after HSM release: $size"
5081 $LFS hsm_release $file
5082 $LFS hsm_restore $file
5083 $LFS hsm_release $file
5084 size=$(stat -c %s $file)
5085 [[ $size == 8192 ]] || error "Size after HSM release: $size"
5086 $LFS hsm_state $file
5088 run_test 261 "Report 0 bytes size after HSM release"
5091 local file=$DIR/$tdir/$tfile
5096 mkdir_on_mdt0 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
5098 dd if=/dev/zero of=$file bs=4k count=2 || error "Write $file failed"
5099 fid=$(path2fid $file)
5100 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $file
5101 wait_request_state $fid ARCHIVE SUCCEED
5103 $LFS hsm_release $file || error "HSM release $file failed"
5104 $LFS hsm_restore $file || error "HSM restore $file failed"
5105 $LFS hsm_release $file || error "HSM release $file failed"
5106 $LFS hsm_release $file || error "HSM release $file failed"
5107 blocks=$(stat -c "%b" $file)
5108 [ $blocks -eq "1" ] || error "wrong block number is $blocks, not 1"
5110 run_test 262 "The client should return 1 block for HSM released files"
5113 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
5115 # the only way to test ondisk conf is to restart MDS ...
5116 echo "Stop coordinator and remove coordinator state at mount"
5119 # clean on disk conf set by default
5120 cdt_clear_mount_state
5121 cdt_check_state stopped
5123 # check cdt still off after umount/remount
5125 cdt_check_state stopped
5127 echo "Set coordinator start at mount, and start coordinator"
5128 cdt_set_mount_state enabled
5131 cdt_check_state enabled
5133 # check cdt still on after umount/remount
5135 cdt_check_state enabled
5137 # we are back to original state (cdt started at mount)
5139 run_test 300 "On disk coordinator state kept between MDT umount/mount"
5142 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
5144 local ai=$(get_hsm_param default_archive_id)
5145 local new=$((ai + 1))
5147 set_hsm_param default_archive_id $new -P
5149 local res=$(get_hsm_param default_archive_id)
5152 set_hsm_param default_archive_id "" "-P -d"
5154 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
5156 run_test 301 "HSM tunnable are persistent"
5159 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
5161 local ai=$(get_hsm_param default_archive_id)
5162 local new=$((ai + 1))
5167 set_hsm_param default_archive_id $new -P
5170 for mdtno in $(seq 1 $MDSCOUNT); do
5175 cdt_check_state enabled
5177 local res=$(get_hsm_param default_archive_id)
5180 set_hsm_param default_archive_id "" "-P -d"
5182 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
5184 run_test 302 "HSM tunnable are persistent when CDT is off"
5187 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5191 mkdir_on_mdt0 $DIR/$tdir
5193 local dir_mdt0=$DIR/$tdir/mdt0
5194 local dir_mdt1=$DIR/$tdir/mdt1
5196 # create 1 dir per MDT
5197 stack_trap "rm -rf $dir_mdt0" EXIT
5198 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
5199 stack_trap "rm -rf $dir_mdt1" EXIT
5200 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
5202 # create 1 file in each MDT
5203 local fid1=$(create_small_file $dir_mdt0/$tfile)
5204 local fid2=$(create_small_file $dir_mdt1/$tfile)
5206 # check that hsm request on mdt0 is sent to the right MDS
5207 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
5208 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
5209 echo "archive successful on mdt0"
5211 # check that hsm request on mdt1 is sent to the right MDS
5212 $LFS hsm_archive $dir_mdt1/$tfile || error "lfs hsm_archive"
5213 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
5214 echo "archive successful on mdt1"
5216 run_test 400 "Single request is sent to the right MDT"
5219 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5223 mkdir_on_mdt0 $DIR/$tdir
5225 local dir_mdt0=$DIR/$tdir/mdt0
5226 local dir_mdt1=$DIR/$tdir/mdt1
5228 # create 1 dir per MDT
5229 stack_trap "rm -rf $dir_mdt0" EXIT
5230 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
5231 stack_trap "rm -rf $dir_mdt1" EXIT
5232 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
5234 # create 1 file in each MDT
5235 local fid1=$(create_small_file $dir_mdt0/$tfile)
5236 local fid2=$(create_small_file $dir_mdt1/$tfile)
5238 # check that compound requests are shunt to the rights MDTs
5239 $LFS hsm_archive $dir_mdt0/$tfile $dir_mdt1/$tfile ||
5240 error "lfs hsm_archive"
5241 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
5242 echo "archive successful on mdt0"
5243 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
5244 echo "archive successful on mdt1"
5246 run_test 401 "Compound requests split and sent to their respective MDTs"
5248 mdc_change_state() # facet, MDT_pattern, activate|deactivate
5253 local node=$(facet_active_host $facet)
5255 for mdc in $(do_facet $facet "$LCTL dl | grep -E ${pattern}-mdc" |
5256 awk '{print $4}'); do
5257 echo "$3 $mdc on $node"
5258 do_facet $facet "$LCTL --device $mdc $state" || return 1
5263 # deactivate all mdc on agent1
5264 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "deactivate"
5266 copytool setup --no-fail
5268 check_agent_unregistered "uuid" # match any agent
5270 # no expected running copytool
5271 search_copytools $agent && error "Copytool start should have failed"
5274 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "activate"
5276 run_test 402a "Copytool start fails if all MDTs are inactive"
5281 mkdir_on_mdt0 $DIR/$tdir
5283 local f=$DIR/$tdir/$tfile
5284 touch $f || error "touch $f failed"
5285 local fid=$(path2fid $f)
5287 #define OBD_FAIL_MDS_HSM_CT_REGISTER_NET 0x14d
5288 do_facet $SINGLEAGT lctl set_param fail_loc=0x14d
5289 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
5291 # give time for CDT to send request and to keep it for retry
5292 wait_for_loop_period
5294 wait_request_state $fid ARCHIVE WAITING
5296 do_facet $SINGLEAGT lctl set_param fail_loc=0
5298 # request should succeed now
5299 wait_request_state $fid ARCHIVE SUCCEED
5301 run_test 402b "CDT must retry request upon slow start of CT"
5304 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5306 local agent=$(facet_active_host $SINGLEAGT)
5308 # deactivate all mdc for MDT0001
5309 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
5312 local uuid=$(get_agent_uuid $agent)
5313 # check the agent is registered on MDT0000, and not on MDT0001
5314 check_agent_registered_by_mdt $uuid 0
5315 check_agent_unregistered_by_mdt $uuid 1
5317 # check running copytool process
5318 search_copytools $agent || error "No running copytools on $agent"
5320 # reactivate all mdc for MDT0001
5321 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
5323 # make sure the copytool is now registered to all MDTs
5324 check_agent_registered $uuid
5326 run_test 403 "Copytool starts with inactive MDT and register on reconnect"
5329 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5333 # create files on both MDT0000 and MDT0001
5334 mkdir_on_mdt0 $DIR/$tdir
5336 local dir_mdt0=$DIR/$tdir/mdt0
5337 stack_trap "rm -rf $dir_mdt0" EXIT
5338 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
5340 # create 1 file on mdt0
5341 local fid1=$(create_small_file $dir_mdt0/$tfile)
5343 # deactivate all mdc for MDT0001
5344 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
5346 # send an HSM request for files in MDT0000
5347 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
5349 # check for completion of files in MDT0000
5350 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
5351 echo "archive successful on mdt0"
5353 # reactivate all mdc for MDT0001
5354 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
5356 run_test 404 "Inactive MDT does not block requests for active MDTs"
5359 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5363 mkdir_on_mdt0 $DIR/$tdir
5365 local striped_dir=$DIR/$tdir/striped_dir
5367 # create striped dir on all of MDTs
5368 $LFS mkdir -i 0 -c $MDSCOUNT $striped_dir || error "lfs mkdir"
5370 local fid1=$(create_small_sync_file $striped_dir/${tfile}_0)
5371 local fid2=$(create_small_sync_file $striped_dir/${tfile}_1)
5372 local fid3=$(create_small_sync_file $striped_dir/${tfile}_2)
5373 local fid4=$(create_small_sync_file $striped_dir/${tfile}_3)
5375 local idx1=$($LFS getstripe -m $striped_dir/${tfile}_0)
5376 local idx2=$($LFS getstripe -m $striped_dir/${tfile}_1)
5377 local idx3=$($LFS getstripe -m $striped_dir/${tfile}_2)
5378 local idx4=$($LFS getstripe -m $striped_dir/${tfile}_3)
5380 # check that compound requests are shunt to the rights MDTs
5381 $LFS hsm_archive $striped_dir/${tfile}_0 $striped_dir/${tfile}_1 \
5382 $striped_dir/${tfile}_2 $striped_dir/${tfile}_3 ||
5383 error "lfs hsm_archive"
5385 wait_request_state $fid1 ARCHIVE SUCCEED $idx1 &&
5386 echo "archive successful on $fid1"
5387 wait_request_state $fid2 ARCHIVE SUCCEED $idx2 &&
5388 echo "archive successful on $fid2"
5389 wait_request_state $fid3 ARCHIVE SUCCEED $idx3 &&
5390 echo "archive successful on $fid3"
5391 wait_request_state $fid4 ARCHIVE SUCCEED $idx4 &&
5392 echo "archive successful on $fid4"
5394 $LFS hsm_release $striped_dir/${tfile}_0 || error "lfs hsm_release 1"
5395 $LFS hsm_release $striped_dir/${tfile}_1 || error "lfs hsm_release 2"
5396 $LFS hsm_release $striped_dir/${tfile}_2 || error "lfs hsm_release 3"
5397 $LFS hsm_release $striped_dir/${tfile}_3 || error "lfs hsm_release 4"
5399 cat $striped_dir/${tfile}_0 > /dev/null || error "cat ${tfile}_0 failed"
5400 cat $striped_dir/${tfile}_1 > /dev/null || error "cat ${tfile}_1 failed"
5401 cat $striped_dir/${tfile}_2 > /dev/null || error "cat ${tfile}_2 failed"
5402 cat $striped_dir/${tfile}_3 > /dev/null || error "cat ${tfile}_3 failed"
5404 run_test 405 "archive and release under striped directory"
5407 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
5409 [ $MDS1_VERSION -lt $(version_code 2.7.64) ] &&
5410 skip "need MDS version at least 2.7.64"
5415 mkdir_on_mdt0 $DIR/$tdir
5417 fid=$(create_small_file $DIR/$tdir/$tfile)
5422 $LFS hsm_archive $DIR/$tdir/$tfile
5423 wait_request_state "$fid" ARCHIVE SUCCEED
5424 $LFS hsm_release $DIR/$tdir/$tfile
5426 # Should migrate $tdir but not $tfile.
5427 $LFS migrate -m1 $DIR/$tdir &&
5428 error "migrating HSM an archived file should fail"
5430 $LFS hsm_restore $DIR/$tdir/$tfile
5431 wait_request_state "$fid" RESTORE SUCCEED
5433 $LFS hsm_remove $DIR/$tdir/$tfile
5434 wait_request_state "$fid" REMOVE SUCCEED
5436 cat $DIR/$tdir/$tfile > /dev/null ||
5437 error "cannot read $DIR/$tdir/$tfile"
5439 $LFS migrate -m1 $DIR/$tdir ||
5440 error "cannot complete migration after HSM remove"
5442 mdt_index=$($LFS getstripe -m $DIR/$tdir)
5443 if ((mdt_index != 1)); then
5444 error "expected MDT index 1, got $mdt_index"
5447 # Refresh fid after migration.
5448 fid=$(path2fid $DIR/$tdir/$tfile)
5451 $LFS hsm_archive $DIR/$tdir/$tfile
5452 wait_request_state "$fid" ARCHIVE SUCCEED 1
5454 lctl set_param debug=+trace
5455 $LFS hsm_release $DIR/$tdir/$tfile ||
5456 error "cannot release $DIR/$tdir/$tfile"
5458 $LFS hsm_restore $DIR/$tdir/$tfile
5459 wait_request_state "$fid" RESTORE SUCCEED 1
5461 cat $DIR/$tdir/$tfile > /dev/null ||
5462 error "cannot read $DIR/$tdir/$tfile"
5464 run_test 406 "attempting to migrate HSM archived files is safe"
5467 mkdir_on_mdt0 $DIR/$tdir
5469 local f=$DIR/$tdir/$tfile
5470 local f2=$DIR2/$tdir/$tfile
5471 local fid=$(create_empty_file "$f")
5475 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
5476 wait_request_state $fid ARCHIVE SUCCEED
5479 #define OBD_FAIL_MDS_HSM_CDT_DELAY 0x164
5480 do_facet $SINGLEMDS $LCTL set_param fail_val=5 fail_loc=0x164
5482 # Prevent restore from completing
5486 # 1st request holds layout lock while appropriate
5487 # RESTORE record is still not added to llog
5491 do_facet $SINGLEMDS "$LCTL get_param $HSM_PARAM.actions"
5492 # after umount hsm_actions->O/x/x log shouldn't have
5493 # double RESTORE records like below
5494 #[0x200000401:0x1:0x0]...0x58d03a0d/0x58d03a0c action=RESTORE...WAITING
5495 #[0x200000401:0x1:0x0]...0x58d03a0c/0x58d03a0d action=RESTORE...WAITING
5497 do_facet $SINGLEMDS "$LCTL get_param $HSM_PARAM.actions"&
5499 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
5501 do_facet $SINGLEMDS "$LCTL get_param $HSM_PARAM.actions"
5504 wait_all_done 100 $fid
5506 run_test 407 "Check for double RESTORE records in llog"
5508 test_408 () { #LU-17110
5509 checkfiemap --test ||
5510 skip "checkfiemap not runnable: $?"
5518 # write data this way: hole - data - hole - data
5519 dd if=/dev/urandom of=$f bs=64K count=1 conv=fsync
5520 [[ "$(facet_fstype ost$(($($LFS getstripe -i $f) + 1)))" != "zfs" ]] ||
5521 skip "ORI-366/LU-1941: FIEMAP unimplemented on ZFS"
5522 dd if=/dev/urandom of=$f bs=64K seek=3 count=1 conv=fsync
5524 echo "disk usage: $(du -B1 $f)"
5525 echo "file size: $(du -b $f)"
5528 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
5529 wait_request_state $fid ARCHIVE SUCCEED
5532 out=$(checkfiemap --corruption_test $f $((4 * 64 * 1024))) ||
5533 error "checkfiemap failed"
5536 grep -q "flags (0x3):" <<< "$out" ||
5537 error "the extent flags of a relase file should be: LAST UNKNOWN"
5540 run_test 408 "Verify fiemap on release file"
5543 (( MDS1_VERSION >= $(version_code 2.15.59) )) ||
5544 skip "need MDS version at least 2.15.59"
5546 mkdir_on_mdt0 $DIR/$tdir
5548 local restore_pid shutdown_pid
5549 local mdt0_hsm_state
5550 local f=$DIR/$tdir/$tfile
5551 local fid=$(create_empty_file "$f")
5555 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
5556 error "could not archive file $f"
5557 wait_request_state $fid ARCHIVE SUCCEED
5558 $LFS hsm_release $f || error "could not release file $f"
5560 #define OBD_FAIL_MDS_HSM_CDT_DELAY 0x164
5561 do_facet $SINGLEMDS $LCTL set_param fail_val=5 fail_loc=0x164
5563 # send a restore request
5564 cat $f > /dev/null & restore_pid=$!
5566 # stop the coordinator while it handle the request
5567 cdt_shutdown & shutdown_pid=$!
5568 stack_trap "cdt_enable" EXIT
5571 mdt0_hsm_state=$(do_facet mds1 "$LCTL get_param -n mdt.*MDT0000.hsm_control")
5572 [[ "$mdt0_hsm_state" == "stopping" ]] ||
5573 error "HSM state of MDT0000 is not 'stopping' (hsm_control=$mdt0_hsm_state)"
5576 cdt_check_state stopped
5577 wait_request_state $fid RESTORE WAITING
5581 # copytool must re-register
5583 wait_copytools || error "copytool failed to stop"
5586 wait $restore_pid || true
5587 wait_request_state $fid RESTORE SUCCEED
5588 cat $f > /dev/null || error "fail to read $f"
5590 run_test 409a "Coordinator should not stop when in use"
5594 mkdir_on_mdt0 $DIR/$tdir
5596 local f=$DIR/$tdir/$tfile
5597 local fid=$(create_empty_file "$f")
5601 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
5602 wait_request_state $fid ARCHIVE SUCCEED
5604 $LFS hsm_release $f || error "cannot release $f"
5605 check_hsm_flags $f "0x0000000d"
5608 wait_copytools || error "copytools failed to stop"
5610 # Remount fs to clear cached file attributes and free restore hash
5611 stack_trap "cdt_check_state enabled" EXIT
5612 stack_trap "cdt_set_mount_state enabled" EXIT
5613 cdt_set_mount_state shutdown
5614 cdt_check_state stopped
5618 # getattr should work even if CDT is stopped
5619 stat $f || error "cannot stat file"
5621 run_test 409b "getattr released file with CDT stopped after remount"
5625 [ "$MDS1_VERSION" -lt $(version_code 2.6.92) ] &&
5626 skip "HSM migrate is not supported"
5628 test_mkdir -p $DIR/$tdir
5630 if [ "$CLIENT_VERSION" -lt $(version_code 2.11.56) ] ||
5631 [ "$MDS1_VERSION" -lt $(version_code 2.11.56) ];
5633 llapi_hsm_test -d $DIR/$tdir -b ||
5634 error "One llapi HSM test failed"
5636 llapi_hsm_test -d $DIR/$tdir ||
5637 error "One llapi HSM test failed"
5640 run_test 500 "various LLAPI HSM tests"
5643 [ "$MDS1_VERSION" -lt $(version_code 2.10.58) ] &&
5644 skip "need MDS version at least 2.10.58"
5648 local f=$DIR/$tdir/$tfile
5651 # set changelog_mask to ALL
5652 changelog_chmask "ALL"
5654 chmod 777 $DIR/$tdir
5655 $RUNAS touch $f || error "touch $f failed as $RUNAS_ID"
5656 local fid=$(path2fid $f)
5659 entry=$(changelog_find -type CREAT -target-fid $fid -uid "$RUNAS_ID" \
5660 -gid "$RUNAS_GID") ||
5661 error "No matching CREAT entry"
5663 # Parse the changelog
5664 eval local -A changelog=$(changelog2array $entry)
5665 local nid="${changelog[nid]}"
5668 echo "Got NID '$nid'"
5669 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5670 error "nid '$nid' does not match any client NID:" \
5673 run_test 600 "Changelog fields 'u=' and 'nid='"
5676 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5677 skip "need MDS version at least 2.10.58"
5681 local f=$DIR/$tdir/$tfile
5684 # set changelog_mask to ALL
5685 changelog_chmask "ALL"
5687 touch $f || error "touch $f failed"
5688 local fid=$(path2fid $f)
5691 cat $f || error "cat $f failed"
5693 changelog_find -type OPEN -target-fid $fid -mode "r--" ||
5694 error "No matching OPEN entry"
5696 run_test 601 "OPEN Changelog entry"
5699 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5700 skip "need MDS version at least 2.10.58"
5702 stack_trap "restore_opencache" EXIT
5707 local f=$DIR/$tdir/$tfile
5710 # set changelog_mask to ALL
5711 changelog_chmask "ALL"
5713 touch $f || error "touch $f failed"
5714 local fid=$(path2fid $f)
5717 cat $f || error "cat $f failed"
5719 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5723 echo f > $f || error "write $f failed"
5726 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5728 # remove OPEN from changelog_mask
5729 changelog_chmask "-OPEN"
5733 cat $f || error "cat $f failed"
5736 changelog_find -type CLOSE -target-fid $fid &&
5737 error "There should be no CLOSE entry"
5741 echo f > $f || error "write $f failed"
5744 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5746 run_test 602 "Changelog record CLOSE only if open+write or OPEN recorded"
5749 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5750 skip "need MDS version at least 2.10.58"
5754 local f=$DIR/$tdir/$tfile
5757 # set changelog_mask to ALL
5758 changelog_chmask "ALL"
5760 touch $f || error "touch $f failed"
5761 local fid=$(path2fid $f)
5763 setfattr -n user.xattr1 -v "value1" $f || error "setfattr $f failed"
5766 getfattr -n user.xattr1 $f || error "getfattr $f failed"
5768 changelog_find -type GXATR -target-fid $fid -xattr "user.xattr1" ||
5769 error "No matching GXATR entry"
5771 run_test 603 "GETXATTR Changelog entry"
5774 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5775 skip "need MDS version at least 2.10.58"
5777 mkdir_on_mdt0 $DIR/$tdir
5779 local f=$DIR/$tdir/$tfile
5780 local f2=$DIR2/$tdir/$tfile
5781 local procname="mdd.$FSNAME-MDT0000.changelog_deniednext"
5783 timeout="$(do_facet mds1 "$LCTL" get_param -n "$procname")"
5784 stack_trap "do_facet mds1 '$LCTL' set_param '$procname=$timeout'" EXIT
5785 do_facet mds1 lctl set_param "$procname=20"
5789 # set changelog_mask to ALL
5790 changelog_chmask "ALL"
5792 touch $f || error "touch $f failed"
5793 local fid=$(path2fid $f)
5799 $RUNAS cat $f2 && error "cat $f2 by user $RUNAS_ID should have failed"
5803 entry=$(changelog_find -type NOPEN -target-fid $fid -uid "$RUNAS_ID" \
5804 -gid "$RUNAS_GID" -mode "r--") ||
5805 error "No matching NOPEN entry"
5807 # Parse the changelog
5808 eval local -A changelog=$(changelog2array $entry)
5809 local nid="${changelog[nid]}"
5812 echo "Got NID '$nid'"
5813 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5814 error "nid '$nid' does not match any client NID:" \
5819 $RUNAS cat $f2 && error "cat $f2 by user $RUNAS_ID should have failed"
5822 changelog_find -type NOPEN -target-fid $fid &&
5823 error "There should be no NOPEN entry"
5825 # Sleep for `changelog_deniednext` seconds
5830 $RUNAS cat $f2 && error "cat $f by user $RUNAS_ID should have failed"
5833 entry=$(changelog_find -type NOPEN -target-fid $fid -uid "$RUNAS_ID" \
5834 -gid "$RUNAS_GID" -mode "r--") ||
5835 error "No matching NOPEN entry"
5837 # Parse the changelog
5838 eval local -A changelog=$(changelog2array $entry)
5839 local nid="${changelog[nid]}"
5842 echo "Got NID '$nid'"
5843 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5844 error "nid '$nid' does not match any client NID:" \
5847 run_test 604 "NOPEN Changelog entry"
5850 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5851 skip "need MDS version at least 2.10.58"
5853 stack_trap "restore_opencache" EXIT
5858 local f=$DIR/$tdir/$tfile
5859 local f2=$DIR2/$tdir/$tfile
5862 # set changelog_mask to ALL
5863 changelog_chmask "ALL"
5865 touch $f || error "touch $f failed"
5866 local fid=$(path2fid $f)
5870 exec 3<> $f || error "open $f failed"
5874 changelog_find -type OPEN -target-fid $fid || error "No OPEN entry"
5878 exec 4<> $f || error "open $f failed"
5881 changelog_find -type OPEN -target-fid $fid &&
5882 error "There should be no OPEN entry"
5884 exec 4>&- || error "close $f failed"
5887 changelog_find -type CLOSE -target-fid $fid &&
5888 error "There should be no CLOSE entry"
5892 # access in rw, so different access mode should generate entries
5893 cat $f || error "cat $f failed"
5896 changelog_find -type OPEN -target-fid $fid || error "No OPEN entry"
5898 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5902 # same access as first one, should not generate new entries
5903 exec 4<> $f || error "open $f failed"
5906 changelog_find -type OPEN -target-fid $fid &&
5907 error "There should be no OPEN entry"
5909 exec 4>&- || error "close $f failed"
5912 changelog_find -type CLOSE -target-fid $fid &&
5913 error "There should be no CLOSE entry"
5917 # access by different user should generate new entries
5918 $RUNAS cat $f || error "cat $f by user $RUNAS_ID failed"
5921 changelog_find -type OPEN -target-fid $fid || error "No OPEN entry"
5923 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5927 exec 3>&- || error "close $f failed"
5930 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5932 run_test 605 "Test OPEN and CLOSE rate limit in Changelogs"
5935 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5936 skip "need MDS version at least 2.10.58"
5938 local llog_reader=$(do_facet mgs "which llog_reader 2> /dev/null")
5939 llog_reader=${llog_reader:-$LUSTRE/utils/llog_reader}
5940 [ -z $(do_facet mgs ls -d $llog_reader 2> /dev/null) ] &&
5941 skip_env "missing llog_reader"
5943 mkdir_on_mdt0 $DIR/$tdir
5945 local f=$DIR/$tdir/$tfile
5948 # set changelog_mask to ALL
5949 changelog_chmask "ALL"
5951 chmod 777 $DIR/$tdir
5952 $RUNAS touch $f || error "touch $f failed as $RUNAS_ID"
5953 local fid=$(path2fid $f)
5954 rm $f || error "rm $f failed"
5956 local mntpt=$(facet_mntpt mds1)
5960 #remount mds1 as ldiskfs or zfs type
5961 stop mds1 || error "stop mds1 failed"
5962 stack_trap "unmount_fstype mds1; start mds1 $(mdsdevname 1)\
5963 $MDS_MOUNT_OPTS" EXIT
5964 mount_fstype mds1 || error "remount mds1 failed"
5966 for ((i = 0; i < 1; i++)); do
5967 do_facet mds1 $llog_reader $mntpt/changelog_catalog
5968 local cat_file=$(do_facet mds1 $llog_reader \
5969 $mntpt/changelog_catalog | awk \
5970 '{match($0,"path=([^ ]+)",a)}END{print a[1]}')
5971 [ -n "$cat_file" ] || error "no catalog file"
5973 entry=$(do_facet mds1 $llog_reader $mntpt/$cat_file |
5974 awk "/CREAT/ && /target:\[$fid\]/ {print}")
5975 [ -n "$entry" ] || error "no CREAT entry"
5978 local uidgid=$(echo $entry |
5979 sed 's+.*\ user:\([0-9][0-9]*:[0-9][0-9]*\)\ .*+\1+')
5980 [ -n "$uidgid" ] || error "uidgid is empty"
5981 echo "Got UID/GID $uidgid"
5982 [ "$uidgid" = "$RUNAS_ID:$RUNAS_GID" ] ||
5983 error "uidgid '$uidgid' != '$RUNAS_ID:$RUNAS_GID'"
5984 local nid=$(echo $entry |
5985 sed 's+.*\ nid:\(\S\S*@\S\S*\)\ .*+\1+')
5986 [ -n "$nid" ] || error "nid is empty"
5988 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5989 error "nid '$nid' does not match any NID ${CLIENT_NIDS[*]}"
5991 run_test 606 "llog_reader groks changelog fields"
5995 getfattr -e text -n trusted.hsm "$1" 2>/dev/null |
5996 sha1sum | awk '{ print $1 }'
5999 test_hsm_migrate_init()
6006 fid=$(create_small_file "$f")
6013 local d="$DIR/$tdir"
6017 (( MDS1_VERSION >= $(version_code 2.15.60) )) ||
6018 skip "need MDS version at least 2.15.60"
6020 (( OSTCOUNT >= 2 )) || skip_env "needs >= 2 OSTs"
6022 fid=$(test_hsm_migrate_init "$d" "$f" | tail -1)
6026 $LFS hsm_archive "$f" || error "could not archive file"
6027 wait_request_state $fid ARCHIVE SUCCEED
6029 $LFS migrate -n -i 1 "$f" ||
6030 error "could not migrate file to OST 1"
6032 $LFS hsm_release "$f" ||
6033 error "could not release file after non blocking migrate"
6034 $LFS hsm_restore "$f" ||
6035 error "could not restore file after non blocking migrate"
6036 wait_request_state $fid RESTORE SUCCEED
6038 run_test 607a "release a file that was migrated after being archived"
6042 local d="$DIR/$tdir"
6043 local f="$DIR/$tdir/$tfile"
6049 (( MDS1_VERSION >= $(version_code 2.15.60) )) ||
6050 skip "need MDS version at least 2.15.60"
6052 (( OSTCOUNT >= 2 )) || skip_env "needs >= 2 OSTs"
6054 fid=$(test_hsm_migrate_init "$d" "$f" | tail -1)
6058 $LFS hsm_archive "$f" || error "could not archive file"
6059 wait_request_state $fid ARCHIVE SUCCEED
6061 saved_params=$($LCTL get_param llite.*.xattr_cache | tr '\n' ' ')
6062 $LCTL set_param llite.*.xattr_cache=0
6063 stack_trap "$LCTL set_param $saved_params" EXIT
6065 # make sure that migrate won't change archive version
6068 old_hsm=$(get_hsm_xattr_sha "$f")
6069 $LFS migrate -n -i 1 "$f" ||
6070 error "could not migrate file to OST 1"
6072 $LFS hsm_state "$f" | grep dirty || error "dirty flag not found"
6074 new_hsm=$(get_hsm_xattr_sha "$f")
6075 [ "$old_hsm" != "$new_hsm" ] &&
6076 error "migrate should not modify data version of dirty files"
6080 run_test 607b "Migrate should not change the HSM attribute of dirty files"
6084 local d="$DIR/$tdir"
6085 local f="$DIR/$tdir/$tfile"
6086 local fid1 fid2 fid3
6089 (( MDS1_VERSION >= $(version_code 2.15.60) )) ||
6090 skip "need MDS version at least 2.15.60"
6093 fid1=$(create_small_file "$f-1")
6094 fid2=$(create_small_file "$f-2")
6095 fid3=$(create_small_file "$f-3")
6099 $LFS hsm_archive "$f-1" || error "could not archive file"
6100 wait_request_state $fid1 ARCHIVE SUCCEED
6102 $LFS hsm_archive "$f-3" || error "could not archive file"
6103 wait_request_state $fid3 ARCHIVE SUCCEED
6105 $LFS swap_layouts "$f-1" "$f-3" |& grep "Operation not permitted" ||
6106 error "swap_layouts should fail with EPERM on 2 archived file"
6108 $LFS swap_layouts "$f-1" "$f-2" ||
6109 error "swap_layout failed on $f-1 and $f-2"
6111 $LFS swap_layouts "$f-2" "$f-3" ||
6112 error "swap_layout failed on $f-2 and $f-3"
6114 nbr_dirty=$($LFS hsm_state "$f-1" "$f-3" | grep -c 'dirty')
6115 ((nbr_dirty == 2)) || error "dirty flag should be set on $f-1 and $f-3"
6118 run_test 607c "'lfs swap_layouts' should set dirty flag on HSM file"
6120 complete_test $SECONDS
6121 check_and_cleanup_lustre