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 json_parser='import json; import fileinput;'
433 json_parser+=' print "\n".join(["local %s=\"%s\"" % tuple for tuple in '
434 json_parser+='json.loads([line for line in '
435 json_parser+='fileinput.input()][0]).items()])'
437 echo $raw_event | python -c "$json_parser"
440 get_agent_by_uuid_mdt() {
443 local mds=mds$(($mdtidx + 1))
444 do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
448 check_agent_registered_by_mdt() {
451 local mds=mds$(($mdtidx + 1))
452 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
453 if [[ ! -z "$agent" ]]; then
454 echo "found agent $agent on $mds"
456 error "uuid $uuid not found in agent list on $mds"
460 check_agent_unregistered_by_mdt() {
463 local mds=mds$(($mdtidx + 1))
464 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
465 if [[ -z "$agent" ]]; then
466 echo "uuid not found in agent list on $mds"
468 error "uuid found in agent list on $mds: $agent"
472 check_agent_registered() {
475 for mdsno in $(seq 1 $MDSCOUNT); do
476 check_agent_registered_by_mdt $uuid $((mdsno - 1))
480 check_agent_unregistered() {
483 for mdsno in $(seq 1 $MDSCOUNT); do
484 check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
489 local agent=${1:-$(facet_active_host $SINGLEAGT)}
491 # Lustre mount-point is mandatory and last parameter on
493 local mntpnt=$(do_rpc_nodes $agent \
494 pgrep --pidfile=$HSMTOOL_PID_FILE --list-full hsmtool |
496 [ -n "$mntpnt" ] || error "Found no Agent or with no mount-point "\
498 do_rpc_nodes $agent get_client_uuid $mntpnt | cut -d' ' -f2
504 # populate MDT device array
507 # cleanup from previous bad setup
510 # for recovery tests, coordinator needs to be started at mount
512 # the lustre conf must be without hsm on (like for sanity.sh)
513 echo "Set HSM on and start"
514 cdt_set_mount_state enabled
515 cdt_check_state enabled
517 echo "Set sanity-hsm HSM policy"
518 cdt_set_sanity_policy
520 # finished requests are quickly removed from list
521 set_hsm_param grace_delay 10
523 CLIENT_NIDS=( $($LCTL list_nids all) )
525 test_1A() { # was test_1
529 local f=$DIR/$tdir/$tfile
533 check_hsm_flags_user $f "0x00000000"
535 $RUNAS $LFS hsm_set --norelease $f ||
536 error "user could not change hsm flags"
537 check_hsm_flags_user $f "0x00000010"
539 $RUNAS $LFS hsm_clear --norelease $f ||
540 error "user could not clear hsm flags"
541 check_hsm_flags_user $f "0x00000000"
543 # User could not change those flags...
544 $RUNAS $LFS hsm_set --exists $f &&
545 error "user should not set this flag"
546 check_hsm_flags_user $f "0x00000000"
549 $LFS hsm_set --exists $f ||
550 error "root could not change hsm flags"
551 check_hsm_flags_user $f "0x00000001"
553 $LFS hsm_clear --exists $f ||
554 error "root could not clear hsm state"
555 check_hsm_flags_user $f "0x00000000"
558 run_test 1A "lfs hsm flags root/non-root access"
561 mkdir_on_mdt0 $DIR/$tdir
563 local f=$DIR/$tdir/$tfile
564 local fid=$(create_small_file $f)
568 $LFS hsm_archive $f || error "could not archive file"
569 wait_request_state $fid ARCHIVE SUCCEED
571 # Release and check states
572 $LFS hsm_release $f || error "could not release file"
573 echo -n "Verifying released state: "
574 check_hsm_flags $f "0x0000000d"
576 $MMAP_CAT $f > /dev/null || error "failed mmap & cat release file"
578 run_test 1a "mmap & cat a HSM released file"
584 dd if=/dev/urandom of=$f bs=1M count=1 conv=sync ||
585 error "failed to create file"
586 local fid=$(path2fid $f)
591 $LFS hsm_archive $f || error "could not archive file"
592 wait_request_state $fid ARCHIVE SUCCEED
595 $LFS hsm_release $f || error "could not release file"
596 echo "verify released state: "
597 check_hsm_flags $f "0x0000000d" && echo "pass"
600 $LFS hsm_restore $f || error "could not restore file"
601 wait_request_state $fid RESTORE SUCCEED
602 echo "verify restored state: "
603 check_hsm_flags $f "0x00000009" && echo "pass"
607 mkdir_on_mdt0 $DIR/$tdir
608 $LFS setstripe -E 1M -S 1M -E 64M -c 2 -E -1 -c 4 $DIR/$tdir ||
609 error "failed to set default stripe"
610 local f=$DIR/$tdir/$tfile
614 run_test 1b "Archive, Release and Restore composite file"
620 local f=$DIR/$tdir/$tfile
623 # Test whether we can set the maximum archive number.
624 local LOCAL_HSM_ARCHIVE_NUMBER=32
625 $LFS hsm_set --exists --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f ||
626 error "root could not change hsm flags"
627 check_hsm_flags_user $f "0x00000001"
628 echo "verifying archive number is $LOCAL_HSM_ARCHIVE_NUMBER"
629 local st=$(get_hsm_archive_id $f)
630 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
631 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
633 # Test whether setting archive number 0 results in no change.
634 $LFS hsm_set --exists --archive-id 0 $f ||
635 error "root could not change hsm flags"
636 check_hsm_flags_user $f "0x00000001"
637 echo "verifying archive number is still $LOCAL_HSM_ARCHIVE_NUMBER"
638 st=$(get_hsm_archive_id $f)
639 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
640 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
642 LOCAL_HSM_ARCHIVE_NUMBER=33
643 if [ "$CLIENT_VERSION" -ge $(version_code 2.11.56) ] &&
644 [ "$MDS1_VERSION" -ge $(version_code 2.11.56) ]; then
645 # lustre in the new version supports unlimited archiveID.
646 # Test whether setting archive number > 32 is supported
647 $LFS hsm_set --exists --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f ||
648 error "archive ID $LOCAL_HSM_ARCHIVE_NUMBER too large?"
649 check_hsm_flags_user $f "0x00000001"
651 echo "verifying archive number is $LOCAL_HSM_ARCHIVE_NUMBER"
652 st=$(get_hsm_archive_id $f)
653 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
654 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
656 # old client or old mds can only support at most 32 archiveID
657 # test whether setting archive number > 32 results in error.
658 $LFS hsm_set --exists --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f &&
659 error "bitmap archive number is larger than 32"
660 check_hsm_flags_user $f "0x00000001"
663 # Test whether setting archive number 16 and archived flag.
664 LOCAL_HSM_ARCHIVE_NUMBER=16
665 $LFS hsm_set --exists --archived \
666 --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f ||
667 error "root could not change hsm flags"
668 check_hsm_flags_user $f "0x00000009"
669 echo "verifying archive number is $LOCAL_HSM_ARCHIVE_NUMBER"
670 st=$(get_hsm_archive_id $f)
671 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
672 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
674 run_test 1c "Check setting archive-id in lfs hsm_set"
677 [ $MDS1_VERSION -lt $(version_code 2.10.59) ] &&
678 skip "need MDS version at least 2.10.59"
680 mkdir_on_mdt0 $DIR/$tdir
681 $LFS setstripe -E 1M -L mdt -E -1 -c 2 $DIR/$tdir ||
682 error "failed to set default stripe"
683 local f=$DIR/$tdir/$tfile
687 run_test 1d "Archive, Release and Restore DoM file"
690 [ "$MDS1_VERSION" -lt $(version_code $SEL_VER) ] &&
691 skip "skipped for lustre < $SEL_VER"
693 mkdir_on_mdt0 $DIR/$tdir
694 $LFS setstripe -E 1G -z 64M -E 10G -z 512M -E -1 -z 1G $DIR/$tdir ||
695 error "failed to set default stripe"
696 local comp_file=$DIR/$tdir/$tfile
698 test_1bde_base $comp_file
700 local flg_opts="--comp-start 0 -E 64M --comp-flags init"
701 local found=$($LFS find $flg_opts $comp_file | wc -l)
702 [ $found -eq 1 ] || error "1st component not found"
704 flg_opts="--comp-start 64M -E 1G --comp-flags extension"
705 found=$($LFS find $flg_opts $comp_file | wc -l)
706 [ $found -eq 1 ] || error "2nd component not found"
708 flg_opts="--comp-start 1G -E 1G --comp-flags ^init"
709 found=$($LFS find $flg_opts $comp_file | wc -l)
710 [ $found -eq 1 ] || error "3rd component not found"
712 flg_opts="--comp-start 1G -E 10G --comp-flags extension"
713 found=$($LFS find $flg_opts $comp_file | wc -l)
714 [ $found -eq 1 ] || error "4th component not found"
716 flg_opts="--comp-start 10G -E 10G --comp-flags ^init"
717 found=$($LFS find $flg_opts $comp_file | wc -l)
718 [ $found -eq 1 ] || error "5th component not found"
720 flg_opts="--comp-start 10G -E EOF --comp-flags extension"
721 found=$($LFS find $flg_opts $comp_file | wc -l)
722 [ $found -eq 1 ] || error "6th component not found"
724 sel_layout_sanity $comp_file 6
726 run_test 1e "Archive, Release and Restore SEL file"
729 local f=$DIR/$tdir/$tfile
731 mkdir_on_mdt0 $DIR/$tdir
732 create_empty_file "$f"
733 # New files are not dirty
734 check_hsm_flags $f "0x00000000"
736 # For test, we simulate an archived file.
737 $LFS hsm_set --exists $f || error "user could not change hsm flags"
738 check_hsm_flags $f "0x00000001"
740 # chmod do not put the file dirty
741 chmod 600 $f || error "could not chmod test file"
742 check_hsm_flags $f "0x00000001"
744 # chown do not put the file dirty
745 chown $RUNAS_ID $f || error "could not chown test file"
746 check_hsm_flags $f "0x00000001"
748 # truncate put the file dirty
749 $TRUNCATE $f 1 || error "could not truncate test file"
750 check_hsm_flags $f "0x00000003"
752 $LFS hsm_clear --dirty $f || error "could not clear hsm flags"
753 check_hsm_flags $f "0x00000001"
755 run_test 2 "Check file dirtyness when doing setattr"
758 mkdir_on_mdt0 $DIR/$tdir
761 # New files are not dirty
763 check_hsm_flags $f "0x00000000"
765 # For test, we simulate an archived file.
766 $LFS hsm_set --exists $f ||
767 error "user could not change hsm flags"
768 check_hsm_flags $f "0x00000001"
770 # Reading a file, does not set dirty
771 cat $f > /dev/null || error "could not read file"
772 check_hsm_flags $f "0x00000001"
774 # Open for write without modifying data, does not set dirty
775 openfile -f O_WRONLY $f || error "could not open test file"
776 check_hsm_flags $f "0x00000001"
778 # Append to a file sets it dirty
779 cp -p /etc/passwd $f.append || error "could not create file"
780 $LFS hsm_set --exists $f.append ||
781 error "user could not change hsm flags"
782 dd if=/etc/passwd of=$f.append bs=1 count=3\
783 conv=notrunc oflag=append status=noxfer ||
784 file_creation_failure dd $f.append $?
785 check_hsm_flags $f.append "0x00000003"
787 # Modify a file sets it dirty
788 cp -p /etc/passwd $f.modify || error "could not create file"
789 $LFS hsm_set --exists $f.modify ||
790 error "user could not change hsm flags"
791 dd if=/dev/zero of=$f.modify bs=1 count=3\
792 conv=notrunc status=noxfer ||
793 file_creation_failure dd $f.modify $?
794 check_hsm_flags $f.modify "0x00000003"
796 # Open O_TRUNC sets dirty
797 cp -p /etc/passwd $f.trunc || error "could not create file"
798 $LFS hsm_set --exists $f.trunc ||
799 error "user could not change hsm flags"
800 cp /etc/group $f.trunc || error "could not override a file"
801 check_hsm_flags $f.trunc "0x00000003"
803 # Mmapped a file sets dirty
804 cp -p /etc/passwd $f.mmap || error "could not create file"
805 $LFS hsm_set --exists $f.mmap ||
806 error "user could not change hsm flags"
807 multiop $f.mmap OSMWUc || error "could not mmap a file"
808 check_hsm_flags $f.mmap "0x00000003"
810 run_test 3 "Check file dirtyness when opening for write"
813 local f=$DIR/$tdir/$tfile
814 local fid=$(create_small_file $f)
817 local st=$(get_request_state $fid CANCEL)
818 [[ -z "$st" ]] || error "hsm_cancel must not be registered (state=$st)"
820 run_test 4 "Useless cancel must not be registered"
823 # test needs a running copytool
826 mkdir_on_mdt0 $DIR/$tdir
827 local f=$DIR/$tdir/$tfile
828 local fid=$(copy_file /etc/passwd $f)
830 wait_request_state $fid ARCHIVE SUCCEED
832 check_hsm_flags $f "0x00000009"
834 run_test 8 "Test default archive number"
836 test_9A() { # was test_9
837 # we do not use the default one to be sure
838 local archive_id=$((HSM_ARCHIVE_NUMBER + 1))
839 copytool setup --archive-id $archive_id
841 # give time for CT to register with MDTs
842 sleep $(($MDSCOUNT*2))
843 local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
844 check_agent_registered $uuid
846 mkdir_on_mdt0 $DIR/$tdir
848 local f=$DIR/$tdir/$tfile
849 local fid=$(copy_file /etc/passwd $f)
850 $LFS hsm_archive --archive $archive_id $f
851 wait_request_state $fid ARCHIVE SUCCEED
853 check_hsm_flags $f "0x00000009"
855 run_test 9A "Use of explicit archive number, with dedicated copytool"
858 needclients 3 || return 0
864 # start all of the copytools
865 for n in $(seq $AGTCOUNT); do
866 copytool setup --facet agt$n
869 mkdir_on_mdt0 $DIR/$tdir
872 for n in $(seq $AGTCOUNT); do
873 file=$DIR/$tdir/$tfile.$n
874 fid=$(create_small_file $file)
876 $LFS hsm_archive $file || error "could not archive file $file"
877 wait_request_state $fid ARCHIVE SUCCEED
878 check_hsm_flags $file "0x00000009"
881 run_test 9a "Multiple remote agents"
884 # test needs a running copytool
887 mkdir_on_mdt0 $DIR/$tdir
888 mkdir -p $DIR/$tdir/d1
889 local f=$DIR/$tdir/$tfile
890 local fid=$(copy_file /etc/hosts $f)
891 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
892 error "hsm_archive failed"
893 wait_request_state $fid ARCHIVE SUCCEED
895 local hsm_root="$(copytool_device $SINGLEAGT)"
896 local archive="$(do_facet $SINGLEAGT \
897 find "$hsm_root" -name "$fid" -print0)"
898 [ -n "$archive" ] || error "fid '$fid' not in archive '$hsm_root'"
900 echo "Verifying content"
901 do_facet $SINGLEAGT diff $f $archive || error "archived file differs"
902 echo "Verifying hsm state "
903 check_hsm_flags $f "0x00000009"
905 echo "Verifying archive number is $HSM_ARCHIVE_NUMBER"
906 local st=$(get_hsm_archive_id $f)
907 [[ $st == $HSM_ARCHIVE_NUMBER ]] ||
908 error "Wrong archive number, $st != $HSM_ARCHIVE_NUMBER"
910 run_test 10a "Archive a file"
913 # test needs a running copytool
916 mkdir_on_mdt0 $DIR/$tdir
917 local f=$DIR/$tdir/$tfile
918 local fid=$(copy_file /etc/hosts $f)
919 $LFS hsm_archive $f || error "archive request failed"
920 wait_request_state $fid ARCHIVE SUCCEED
922 $LFS hsm_archive $f || error "archive of non dirty file failed"
923 local cnt=$(get_request_count $fid ARCHIVE)
924 [[ "$cnt" == "1" ]] ||
925 error "archive of non dirty file must not make a request"
927 run_test 10b "Archive of non dirty file must work without doing request"
930 # test needs a running copytool
934 local f=$DIR/$tdir/$tfile
935 local fid=$(copy_file /etc/hosts $f)
936 $LFS hsm_set --noarchive $f
937 $LFS hsm_archive $f && error "archive a noarchive file must fail"
940 run_test 10c "Check forbidden archive"
943 # test needs a running copytool
946 mkdir_on_mdt0 $DIR/$tdir
947 local f=$DIR/$tdir/$tfile
948 local fid=$(copy_file /etc/hosts $f)
949 $LFS hsm_archive $f || error "cannot archive $f"
950 wait_request_state $fid ARCHIVE SUCCEED
952 local ar=$(get_hsm_archive_id $f)
953 local dflt=$(get_hsm_param default_archive_id)
954 [[ $ar == $dflt ]] ||
955 error "archived file is not on default archive: $ar != $dflt"
957 run_test 10d "Archive a file on the default archive id"
960 mkdir_on_mdt0 $DIR/$tdir
961 copy2archive /etc/hosts $tdir/$tfile
962 local f=$DIR/$tdir/$tfile
964 copytool import $tdir/$tfile $f
965 echo -n "Verifying released state: "
966 check_hsm_flags $f "0x0000000d"
968 local LSZ=$(stat -c "%s" $f)
969 local ASZ=$(do_facet $SINGLEAGT stat -c "%s" "$(hsm_root)/$tdir/$tfile")
971 echo "Verifying imported size $LSZ=$ASZ"
972 [[ $LSZ -eq $ASZ ]] || error "Incorrect size $LSZ != $ASZ"
973 echo -n "Verifying released pattern: "
974 local PTRN=$($LFS getstripe -L $f)
976 [[ $PTRN == released ]] || error "Is not released"
977 local fid=$(path2fid $f)
978 echo "Verifying new fid $fid in archive"
980 do_facet $SINGLEAGT "[ -f \"$(fid2archive "$fid")\" ]" ||
981 error "No archive for fid $fid"
983 run_test 11a "Import a file"
986 # test needs a running copytool
989 mkdir_on_mdt0 $DIR/$tdir
990 local f=$DIR/$tdir/$tfile
991 local fid=$(copy_file /etc/hosts $f)
992 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
993 error "hsm_archive failed"
994 wait_request_state $fid ARCHIVE SUCCEED
996 local FILE_HASH=$(md5sum $f)
999 copytool import $fid $f
1001 echo "$FILE_HASH" | md5sum -c
1003 [[ $? -eq 0 ]] || error "Restored file differs"
1005 run_test 11b "Import a deleted file using its FID"
1008 pool_add $TESTNAME || error "Pool creation failed"
1009 pool_add_targets $TESTNAME 1 1 || error "pool_add_targets failed"
1012 $LFS setstripe -p "$TESTNAME" $DIR/$tdir
1014 copy2archive /etc/hosts $tdir/$tfile
1015 copytool import $tdir/$tfile $DIR/$tdir/$tfile
1017 run_test 11c "Import a file to a directory with a pool"
1020 # test needs a running copytool
1023 mkdir_on_mdt0 $DIR/$tdir
1024 copy2archive /etc/hosts $tdir/$tfile
1026 local f=$DIR/$tdir/$tfile
1027 copytool import $tdir/$tfile $f
1028 local f2=$DIR2/$tdir/$tfile
1029 echo "Verifying released state: "
1030 check_hsm_flags $f2 "0x0000000d"
1032 local fid=$(path2fid $f2)
1033 $LFS hsm_restore $f2
1034 wait_request_state $fid RESTORE SUCCEED
1036 echo "Verifying file state: "
1037 check_hsm_flags $f2 "0x00000009"
1039 do_facet $SINGLEAGT diff -q $(hsm_root)/$tdir/$tfile $f
1041 [[ $? -eq 0 ]] || error "Restored file differs"
1043 run_test 12a "Restore an imported file explicitly"
1046 # test needs a running copytool
1049 mkdir_on_mdt0 $DIR/$tdir
1050 copy2archive /etc/hosts $tdir/$tfile
1052 local f=$DIR/$tdir/$tfile
1053 copytool import $tdir/$tfile $f
1054 echo "Verifying released state: "
1055 check_hsm_flags $f "0x0000000d"
1057 cat $f > /dev/null || error "File read failed"
1059 echo "Verifying file state after restore: "
1060 check_hsm_flags $f "0x00000009"
1062 do_facet $SINGLEAGT diff -q $(hsm_root)/$tdir/$tfile $f
1064 [[ $? -eq 0 ]] || error "Restored file differs"
1066 run_test 12b "Restore an imported file implicitly"
1069 [ "$OSTCOUNT" -lt "2" ] && skip_env "needs >= 2 OSTs" && return
1071 # test needs a running copytool
1074 local f=$DIR/$tdir/$tfile
1075 mkdir_on_mdt0 $DIR/$tdir
1076 $LFS setstripe -c 2 "$f"
1077 local fid=$(create_file "$f" 1M 5)
1079 local FILE_CRC=$(md5sum $f)
1081 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1082 wait_request_state $fid ARCHIVE SUCCEED
1083 $LFS hsm_release $f || error "release $f failed"
1085 echo "$FILE_CRC" | md5sum -c
1087 [[ $? -eq 0 ]] || error "Restored file differs"
1089 run_test 12c "Restore a file with stripe of 2"
1092 # test needs a running copytool
1095 mkdir_on_mdt0 $DIR/$tdir
1097 local f=$DIR/$tdir/$tfile
1098 local fid=$(copy_file /etc/hosts $f)
1099 $LFS hsm_restore $f || error "restore of non archived file failed"
1100 local cnt=$(get_request_count $fid RESTORE)
1101 [[ "$cnt" == "0" ]] ||
1102 error "restore non archived must not make a request"
1103 $LFS hsm_archive $f ||
1104 error "archive request failed"
1105 wait_request_state $fid ARCHIVE SUCCEED
1106 $LFS hsm_restore $f ||
1107 error "restore of non released file failed"
1108 local cnt=$(get_request_count $fid RESTORE)
1109 [[ "$cnt" == "0" ]] ||
1110 error "restore a non dirty file must not make a request"
1112 run_test 12d "Restore of a non archived, non released file must work"\
1113 " without doing request"
1116 # test needs a running copytool
1119 mkdir_on_mdt0 $DIR/$tdir
1120 local f=$DIR/$tdir/$tfile
1121 local fid=$(copy_file /etc/hosts $f)
1122 $LFS hsm_archive $f || error "archive request failed"
1123 wait_request_state $fid ARCHIVE SUCCEED
1126 cat /etc/hosts >> $f
1130 $LFS hsm_restore $f && error "restore a dirty file must fail"
1133 run_test 12e "Check forbidden restore"
1136 # test needs a running copytool
1139 mkdir_on_mdt0 $DIR/$tdir
1140 local f=$DIR/$tdir/$tfile
1141 local fid=$(copy_file /etc/hosts $f)
1143 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1144 wait_request_state $fid ARCHIVE SUCCEED
1145 $LFS hsm_release $f || error "release of $f failed"
1147 wait_request_state $fid RESTORE SUCCEED
1149 echo -n "Verifying file state: "
1150 check_hsm_flags $f "0x00000009"
1152 diff -q /etc/hosts $f
1154 [[ $? -eq 0 ]] || error "Restored file differs"
1156 run_test 12f "Restore a released file explicitly"
1159 # test needs a running copytool
1162 mkdir_on_mdt0 $DIR/$tdir
1163 local f=$DIR/$tdir/$tfile
1164 local fid=$(copy_file /etc/hosts $f)
1166 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1167 wait_request_state $fid ARCHIVE SUCCEED
1168 $LFS hsm_release $f || error "release of $f failed"
1170 diff -q /etc/hosts $f
1173 # we check we had a restore done
1174 wait_request_state $fid RESTORE SUCCEED
1176 [[ $st -eq 0 ]] || error "Restored file differs"
1178 run_test 12g "Restore a released file implicitly"
1181 needclients 2 || return 0
1183 # test needs a running copytool
1186 mkdir_on_mdt0 $DIR/$tdir
1187 local f=$DIR/$tdir/$tfile
1188 local fid=$(copy_file /etc/hosts $f)
1190 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1191 wait_request_state $fid ARCHIVE SUCCEED
1192 $LFS hsm_release $f || error "release of $f failed"
1194 do_node $CLIENT2 diff -q /etc/hosts $f
1197 # we check we had a restore done
1198 wait_request_state $fid RESTORE SUCCEED
1200 [[ $st -eq 0 ]] || error "Restored file differs"
1202 run_test 12h "Restore a released file implicitly from a second node"
1205 # test needs a running copytool
1208 mkdir_on_mdt0 $DIR/$tdir
1209 local f=$DIR/$tdir/$tfile
1210 local fid=$(copy_file /etc/passwd $f)
1211 $LFS hsm_archive $f || error "archive of $f failed"
1212 wait_request_state $fid ARCHIVE SUCCEED
1214 $LFS hsm_release $f || error "release of $f failed"
1218 [[ $? -eq 0 ]] || error "Restored file differs"
1220 run_test 12m "Archive/release/implicit restore"
1223 # test needs a running copytool
1227 copy2archive /etc/hosts $tdir/$tfile
1229 local f=$DIR/$tdir/$tfile
1230 copytool import $tdir/$tfile $f
1232 do_facet $SINGLEAGT cmp /etc/hosts $f ||
1233 error "Restored file differs"
1235 $LFS hsm_release $f || error "release of $f failed"
1237 run_test 12n "Import/implicit restore/release"
1240 # test needs a running copytool
1243 mkdir_on_mdt0 $DIR/$tdir
1244 local f=$DIR/$tdir/$tfile
1245 local fid=$(copy_file /etc/hosts $f)
1247 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1248 wait_request_state $fid ARCHIVE SUCCEED
1249 $LFS hsm_release $f || error "release of $f failed"
1251 #define OBD_FAIL_MDS_HSM_SWAP_LAYOUTS 0x152
1252 do_facet $SINGLEMDS lctl set_param fail_loc=0x152
1254 # set no retry action mode
1257 diff -q /etc/hosts $f
1260 # we check we had a restore failure
1261 wait_request_state $fid RESTORE FAILED
1263 [[ $st -eq 0 ]] && error "Restore must fail"
1265 # remove no retry action mode
1268 # check file is still released
1269 check_hsm_flags $f "0x0000000d"
1271 # retry w/o failure injection
1272 do_facet $SINGLEMDS lctl set_param fail_loc=0
1274 # to be sure previous RESTORE result is gone
1276 wait_for_grace_delay
1278 diff -q /etc/hosts $f
1281 # we check we had a restore done
1282 wait_request_state $fid RESTORE SUCCEED
1284 [[ $st -eq 0 ]] || error "Restored file differs"
1286 run_test 12o "Layout-swap failure during Restore leaves file released"
1289 # test needs a running copytool
1292 mkdir_on_mdt0 $DIR/$tdir
1293 local f=$DIR/$tdir/$tfile
1294 local fid=$(copy_file /etc/hosts $f)
1296 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1297 wait_request_state $fid ARCHIVE SUCCEED
1298 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1299 $LFS hsm_release $f || error "cannot release $f"
1300 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1301 $LFS hsm_release $f || error "cannot release $f"
1302 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1304 run_test 12p "implicit restore of a file on copytool mount point"
1307 [ $MDS1_VERSION -lt $(version_code 2.7.58) ] &&
1308 skip "need MDS version at least 2.7.58"
1310 stack_trap "zconf_umount \"$(facet_host $SINGLEAGT)\" \"$MOUNT3\"" EXIT
1311 zconf_mount $(facet_host $SINGLEAGT) $MOUNT3 ||
1312 error "cannot mount $MOUNT3 on $SINGLEAGT"
1314 # test needs a running copytool
1315 copytool setup -m "$MOUNT3"
1317 mkdir_on_mdt0 $DIR/$tdir
1319 local f=$DIR/$tdir/$tfile
1320 local f2=$DIR2/$tdir/$tfile
1321 local fid=$(create_small_file $f)
1322 local orig_size=$(stat -c "%s" $f)
1324 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1325 wait_request_state $fid ARCHIVE SUCCEED
1327 $LFS hsm_release $f || error "could not release file"
1328 check_hsm_flags $f "0x0000000d"
1331 wait_copytools || error "copytool failed to stop"
1333 cat $f > /dev/null &
1335 # wait a bit to allow implicit restore request to be handled.
1336 # if not, next stat would also block on layout-lock.
1339 local size=$(stat -c "%s" $f2)
1340 [ $size -eq $orig_size ] ||
1341 error "$f2: wrong size after archive: $size != $orig_size"
1343 copytool setup -m "$MOUNT3"
1347 size=$(stat -c "%s" $f)
1348 [ $size -eq $orig_size ] ||
1349 error "$f: wrong size after restore: $size != $orig_size"
1351 size=$(stat -c "%s" $f2)
1352 [ $size -eq $orig_size ] ||
1353 error "$f2: wrong size after restore: $size != $orig_size"
1357 size=$(stat -c "%s" $f)
1359 error "$f: wrong size after overwrite: $size != 0"
1361 size=$(stat -c "%s" $f2)
1363 error "$f2: wrong size after overwrite: $size != 0"
1365 run_test 12q "file attributes are refreshed after restore"
1368 # test needs a running copytool
1371 mkdir_on_mdt0 $DIR/$tdir
1372 local f=$DIR/$tdir/$tfile
1373 local fid=$(copy_file /etc/hosts $f)
1375 $LFS hsm_archive $f || error "archive of $f failed"
1376 wait_request_state $fid ARCHIVE SUCCEED
1377 $LFS hsm_release $f || error "release of $f failed"
1379 offset=$(lseek_test -d 7 $f)
1381 # we check we had a restore done
1382 wait_request_state $fid RESTORE SUCCEED
1383 [[ $offset == 7 ]] || error "offset $offset != 7"
1385 run_test 12r "lseek restores released file"
1389 for i in {1..10}; do
1390 local archive_dir="$(hsm_root)"/subdir/dir.$i
1392 do_facet $SINGLEAGT mkdir -p "$archive_dir"
1393 for j in {1..10}; do
1394 local archive_file="$archive_dir"/file.$j
1396 do_facet $SINGLEAGT "echo $k > \"$archive_dir\"/file.$j"
1402 copytool import "subdir" "$DIR/$tdir"
1404 # To check the import, the test uses diff with the -r flag
1405 # This is nice, but diff only checks files one by one, and triggering
1406 # an implicit restore for one file at a time will consume as many
1407 # seconds as there are files to compare. To speed this up, a restore
1408 # operation is triggered manually first.
1410 find "$DIR/$tdir"/subdir -type f -exec $LFS hsm_restore {} \;
1412 # Compare the imported data
1413 do_facet $SINGLEAGT \
1414 diff -r "$(hsm_root)"/subdir "$DIR/$tdir"/subdir ||
1415 error "imported files differ from archived data"
1417 run_test 13 "Recursively import and restore a directory"
1420 # test needs a running copytool
1423 mkdir_on_mdt0 $DIR/$tdir
1426 local f=$DIR/$tdir/$tfile
1427 local fid=$(create_small_file $f)
1428 local sum=$(md5sum $f | awk '{print $1}')
1429 $LFS hsm_archive $f || error "could not archive file"
1430 wait_request_state $fid ARCHIVE SUCCEED
1432 # create released file (simulate llapi_hsm_import call)
1433 local fid2=$(create_empty_file "$f")
1434 $LFS hsm_set --archived --exists $f || error "could not force hsm flags"
1435 $LFS hsm_release $f || error "could not release file"
1437 # rebind the archive to the newly created file
1438 echo "rebind $fid to $fid2"
1440 copytool rebind $fid $fid2
1442 # restore file and compare md5sum
1443 local sum2=$(md5sum $f | awk '{print $1}')
1445 [[ $sum == $sum2 ]] || error "md5sum mismatch after restore"
1447 run_test 14 "Rebind archived file to a new fid"
1450 # test needs a running copytool
1453 mkdir_on_mdt0 $DIR/$tdir
1456 local f=$DIR/$tdir/$tfile
1458 local tmpfile=$SHARED_DIRECTORY/tmp.$$
1462 for i in $(seq 1 $count); do
1463 fids[$i]=$(create_small_file $f.$i)
1464 sums[$i]=$(md5sum $f.$i | awk '{print $1}')
1465 $LFS hsm_archive $f.$i || error "could not archive file"
1467 wait_all_done $(($count*60))
1469 stack_trap "rm -f $tmpfile" EXIT
1472 for i in $(seq 1 $count); do
1473 local fid2=$(create_empty_file "${f}.${i}")
1474 # add the rebind operation to the list
1475 echo ${fids[$i]} $fid2 >> $tmpfile
1477 # set it released (simulate llapi_hsm_import call)
1478 $LFS hsm_set --archived --exists $f.$i ||
1479 error "could not force hsm flags"
1480 $LFS hsm_release $f.$i || error "could not release file"
1482 nl=$(wc -l < $tmpfile)
1483 [[ $nl == $count ]] || error "$nl files in list, $count expected"
1485 echo "rebind list of files"
1486 copytool rebind "$tmpfile"
1488 # restore files and compare md5sum
1489 for i in $(seq 1 $count); do
1490 local sum2=$(md5sum $f.$i | awk '{print $1}')
1491 [[ $sum2 == ${sums[$i]} ]] ||
1492 error "md5sum mismatch after restore ($sum2 != ${sums[$i]})"
1495 run_test 15 "Rebind a list of files"
1498 # test needs a running copytool
1502 # create a known size file so we can verify transfer speed
1505 dd if=/dev/zero of=$ref bs=1M count=20
1507 mkdir_on_mdt0 $DIR/$tdir
1508 local f=$DIR/$tdir/$tfile
1509 local fid=$(copy_file $ref $f)
1511 local start=$(date +%s)
1513 wait_request_state $fid ARCHIVE SUCCEED
1514 local end=$(date +%s)
1515 # Add 1 to account for rounding errors between start and end (LU-8155)
1516 local duration=$((end - start + 1))
1518 [[ $duration -ge $((goal - 1)) ]] ||
1519 error "Transfer is too fast $duration < $goal"
1521 run_test 16 "Test CT bandwith control option"
1524 local f=$DIR/$tdir/$tfile
1525 create_empty_file "$f"
1527 # Could not release a non-archived file
1528 $LFS hsm_release $f && error "release should not succeed"
1530 # For following tests, we must test them with HS_ARCHIVED set
1531 $LFS hsm_set --exists --archived $f || error "could not add flag"
1533 # Could not release a file if no-release is set
1534 $LFS hsm_set --norelease $f || error "could not add flag"
1535 $LFS hsm_release $f && error "release should not succeed"
1536 $LFS hsm_clear --norelease $f || error "could not remove flag"
1538 # Could not release a file if lost
1539 $LFS hsm_set --lost $f || error "could not add flag"
1540 $LFS hsm_release $f && error "release should not succeed"
1541 $LFS hsm_clear --lost $f || error "could not remove flag"
1543 # Could not release a file if dirty
1544 $LFS hsm_set --dirty $f || error "could not add flag"
1545 $LFS hsm_release $f && error "release should not succeed"
1546 $LFS hsm_clear --dirty $f || error "could not remove flag"
1548 run_test 20 "Release is not permitted"
1551 # test needs a running copytool
1553 mkdir_on_mdt0 $DIR/$tdir
1555 local f=$DIR/$tdir/test_release
1557 # Create a file and check its states
1558 local fid=$(create_small_file $f)
1559 check_hsm_flags $f "0x00000000"
1561 # LU-4388/LU-4389 - ZFS does not report full number of blocks
1562 # used until file is flushed to disk
1563 if [ "$ost1_FSTYPE" == "zfs" ]; then
1564 # this causes an OST_SYNC rpc to be sent
1565 dd if=/dev/zero of=$f bs=512 count=1 oflag=sync conv=notrunc,fsync
1566 # clear locks to reread file data
1567 cancel_lru_locks osc
1570 local orig_size=$(stat -c "%s" $f)
1571 local orig_blocks=$(stat -c "%b" $f)
1573 start_full_debug_logging
1575 $LFS hsm_archive $f || error "could not archive file"
1576 wait_request_state $fid ARCHIVE SUCCEED
1578 local blocks=$(stat -c "%b" $f)
1579 [ $blocks -eq $orig_blocks ] ||
1580 error "$f: wrong block number after archive: " \
1581 "$blocks != $orig_blocks"
1582 local size=$(stat -c "%s" $f)
1583 [ $size -eq $orig_size ] ||
1584 error "$f: wrong size after archive: $size != $orig_size"
1586 # Release and check states
1587 $LFS hsm_release $f || error "could not release file"
1588 check_hsm_flags $f "0x0000000d"
1590 blocks=$(stat -c "%b" $f)
1591 [ $blocks -gt 5 ] &&
1592 error "$f: too many blocks after release: $blocks > 5"
1593 size=$(stat -c "%s" $f)
1594 [ $size -ne $orig_size ] &&
1595 error "$f: wrong size after release: $size != $orig_size"
1597 # Check we can release an file without stripe info
1601 check_hsm_flags $f "0x00000000"
1602 $LFS hsm_archive $f || error "could not archive file"
1603 wait_request_state $fid ARCHIVE SUCCEED
1605 # Release and check states
1606 $LFS hsm_release $f || error "could not release file"
1607 check_hsm_flags $f "0x0000000d"
1609 # Release again a file that is already released is OK
1610 $LFS hsm_release $f || fail "second release should succeed"
1611 check_hsm_flags $f "0x0000000d"
1613 stop_full_debug_logging
1615 run_test 21 "Simple release tests"
1618 # test needs a running copytool
1620 mkdir_on_mdt0 $DIR/$tdir
1622 local f=$DIR/$tdir/test_release
1623 local swap=$DIR/$tdir/test_swap
1625 # Create a file and check its states
1626 local fid=$(create_small_file $f)
1627 check_hsm_flags $f "0x00000000"
1629 $LFS hsm_archive $f || error "could not archive file"
1630 wait_request_state $fid ARCHIVE SUCCEED
1632 # Release and check states
1633 $LFS hsm_release $f || error "could not release file"
1634 check_hsm_flags $f "0x0000000d"
1636 create_small_file $swap
1637 $LFS swap_layouts $swap $f && error "swap_layouts should failed"
1641 run_test 22 "Could not swap a release file"
1644 # test needs a running copytool
1646 mkdir_on_mdt0 $DIR/$tdir
1648 local f=$DIR/$tdir/test_mtime
1650 # Create a file and check its states
1651 local fid=$(create_small_file $f)
1652 check_hsm_flags $f "0x00000000"
1654 $LFS hsm_archive $f || error "could not archive file"
1655 wait_request_state $fid ARCHIVE SUCCEED
1657 # Set modification time in the past
1658 touch -m -a -d @978261179 $f
1660 # Release and check states
1661 $LFS hsm_release $f || error "could not release file"
1662 check_hsm_flags $f "0x0000000d"
1664 local MTIME=$(stat -c "%Y" $f)
1665 local ATIME=$(stat -c "%X" $f)
1666 [ $MTIME -eq "978261179" ] || fail "bad mtime: $MTIME"
1667 [ $ATIME -eq "978261179" ] || fail "bad atime: $ATIME"
1669 run_test 23 "Release does not change a/mtime (utime)"
1672 local file=$DIR/$tdir/$tfile
1681 # test needs a running copytool
1683 mkdir_on_mdt0 $DIR/$tdir
1685 fid=$(create_small_file $file)
1687 # Create a file and check its states
1688 check_hsm_flags $file "0x00000000"
1690 # Ensure atime is less than mtime and ctime.
1694 atime0=$(stat -c "%X" $file)
1695 mtime0=$(stat -c "%Y" $file)
1696 ctime0=$(stat -c "%Z" $file)
1698 [ $atime0 -lt $mtime0 ] ||
1699 error "atime $atime0 is not less than mtime $mtime0"
1701 [ $atime0 -lt $ctime0 ] ||
1702 error "atime $atime0 is not less than ctime $ctime0"
1704 # Archive should not change any timestamps.
1705 $LFS hsm_archive $file || error "cannot archive '$file'"
1706 wait_request_state $fid ARCHIVE SUCCEED
1708 atime1=$(stat -c "%X" $file)
1709 mtime1=$(stat -c "%Y" $file)
1710 ctime1=$(stat -c "%Z" $file)
1712 [ $atime0 -eq $atime1 ] ||
1713 error "archive changed atime from $atime0 to $atime1"
1715 [ $mtime0 -eq $mtime1 ] ||
1716 error "archive changed mtime from $mtime0 to $mtime1"
1718 [ $ctime0 -eq $ctime1 ] ||
1719 error "archive changed ctime from $ctime0 to $ctime1"
1721 # Release should not change any timestamps.
1722 $LFS hsm_release $file || error "cannot release '$file'"
1723 check_hsm_flags $file "0x0000000d"
1725 atime1=$(stat -c "%X" $file)
1726 mtime1=$(stat -c "%Y" $file)
1727 ctime1=$(stat -c "%Z" $file)
1729 [ $atime0 -eq $atime1 ] ||
1730 error "release changed atime from $atime0 to $atime1"
1732 [ $mtime0 -eq $mtime1 ] ||
1733 error "release changed mtime from $mtime0 to $mtime1"
1735 [ $ctime0 -eq $ctime1 ] ||
1736 error "release changed ctime from $ctime0 to $ctime1"
1738 # Restore should not change any timestamps.
1739 $LFS hsm_restore $file
1740 wait_request_state $fid RESTORE SUCCEED
1742 atime1=$(stat -c "%X" $file)
1743 mtime1=$(stat -c "%Y" $file)
1744 ctime1=$(stat -c "%Z" $file)
1746 [ $atime0 -eq $atime1 ] ||
1747 error "restore changed atime from $atime0 to $atime1"
1749 [ $mtime0 -eq $mtime1 ] ||
1750 error "restore changed mtime from $mtime0 to $mtime1"
1752 [ $ctime0 -eq $ctime1 ] ||
1753 error "restore changed ctime from $ctime0 to $ctime1"
1756 wait_copytools || error "Copytools failed to stop"
1758 # Once more, after unmount and mount.
1759 umount_client $MOUNT || error "cannot unmount '$MOUNT'"
1760 mount_client $MOUNT || error "cannot mount '$MOUNT'"
1762 atime1=$(stat -c "%X" $file)
1763 mtime1=$(stat -c "%Y" $file)
1764 ctime1=$(stat -c "%Z" $file)
1766 [ $atime0 -eq $atime1 ] ||
1767 error "remount changed atime from $atime0 to $atime1"
1769 [ $mtime0 -eq $mtime1 ] ||
1770 error "remount changed mtime from $mtime0 to $mtime1"
1772 [ $ctime0 -eq $ctime1 ] ||
1773 error "remount changed ctime from $ctime0 to $ctime1"
1775 run_test 24a "Archive, release, and restore does not change a/mtime (i/o)"
1778 local file=$DIR/$tdir/$tfile
1784 # Test needs a running copytool.
1786 mkdir_on_mdt0 $DIR/$tdir
1788 # Check that root can do HSM actions on a regular user's file.
1789 fid=$(create_small_file $file)
1790 sum0=$(md5sum $file)
1792 chown $RUNAS_ID:$RUNAS_GID $file ||
1793 error "cannot chown '$file' to '$RUNAS_ID'"
1795 chmod ugo-w $DIR/$tdir ||
1796 error "cannot chmod '$DIR/$tdir'"
1798 $LFS hsm_archive $file
1799 wait_request_state $fid ARCHIVE SUCCEED
1801 $LFS hsm_release $file
1802 check_hsm_flags $file "0x0000000d"
1804 $LFS hsm_restore $file
1805 wait_request_state $fid RESTORE SUCCEED
1807 # Check that ordinary user can get HSM state.
1808 $RUNAS $LFS hsm_state $file ||
1809 error "user '$RUNAS_ID' cannot get HSM state of '$file'"
1811 $LFS hsm_release $file
1812 check_hsm_flags $file "0x0000000d"
1814 # Check that ordinary user can accessed released file.
1815 sum1=$($RUNAS md5sum $file) ||
1816 error "user '$RUNAS_ID' cannot read '$file'"
1818 [ "$sum0" == "$sum1" ] ||
1819 error "md5sum mismatch for '$file'"
1821 run_test 24b "root can archive, release, and restore user files"
1824 local file=$DIR/$tdir/$tfile
1825 local action=archive
1830 # test needs a running copytool
1832 mkdir_on_mdt0 $DIR/$tdir
1834 # Save the default masks and check that cleanup_24c will
1835 # restore the request masks correctly.
1836 user_save=$(get_hsm_param user_request_mask)
1837 stack_trap "set_hsm_param user_request_mask $user_save" EXIT
1838 group_save=$(get_hsm_param group_request_mask)
1839 stack_trap "set_hsm_param user_request_mask $group_save" EXIT
1840 other_save=$(get_hsm_param other_request_mask)
1841 stack_trap "set_hsm_param user_request_mask $other_save" EXIT
1843 [ "$user_save" == RESTORE ] ||
1844 error "user_request_mask is '$user_save' expected 'RESTORE'"
1845 [ "$group_save" == RESTORE ] ||
1846 error "group_request_mask is '$group_save' expected 'RESTORE'"
1847 [ "$other_save" == RESTORE ] ||
1848 error "other_request_mask is '$other_save' expected 'RESTORE'"
1851 create_small_file $file
1852 chown $RUNAS_ID:$GROUP $file ||
1853 error "cannot chown '$file' to '$RUNAS_ID:$GROUP'"
1855 $RUNAS $LFS hsm_$action $file &&
1856 error "$action by user should fail"
1858 set_hsm_param user_request_mask $action
1859 $RUNAS $LFS hsm_$action $file ||
1860 error "$action by user should succeed"
1863 create_small_file $file
1864 chown nobody:$RUNAS_GID $file ||
1865 error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
1867 $RUNAS $LFS hsm_$action $file &&
1868 error "$action by group should fail"
1870 set_hsm_param group_request_mask $action
1871 $RUNAS $LFS hsm_$action $file ||
1872 error "$action by group should succeed"
1875 create_small_file $file
1876 chown nobody:$GROUP $file ||
1877 error "cannot chown '$file' to 'nobody:$GROUP'"
1879 $RUNAS $LFS hsm_$action $file &&
1880 error "$action by other should fail"
1882 set_hsm_param other_request_mask $action
1883 $RUNAS $LFS hsm_$action $file ||
1884 error "$action by other should succeed"
1886 run_test 24c "check that user,group,other request masks work"
1889 local file1=$DIR/$tdir/$tfile
1890 local file2=$DIR2/$tdir/$tfile
1894 mkdir_on_mdt0 $DIR/$tdir
1896 fid1=$(create_small_file $file1)
1899 $LFS getstripe $file1
1901 stack_trap "zconf_umount \"$(facet_host $SINGLEAGT)\" \"$MOUNT3\"" EXIT
1902 zconf_mount "$(facet_host $SINGLEAGT)" "$MOUNT3" ||
1903 error "cannot mount '$MOUNT3' on '$SINGLEAGT'"
1905 copytool setup -m "$MOUNT3"
1907 stack_trap "mount -o remount,rw \"$MOUNT2\"" EXIT
1908 mount -o remount,ro $MOUNT2
1910 do_nodes $(comma_list $(nodes_list)) $LCTL clear
1912 fid2=$(path2fid $file2)
1913 [ "$fid1" == "$fid2" ] ||
1914 error "FID mismatch '$fid1' != '$fid2'"
1916 $LFS hsm_archive $file2 &&
1917 error "archive should fail on read-only mount"
1918 check_hsm_flags $file1 "0x00000000"
1920 $LFS hsm_archive $file1 || error "Fail to archive $file1"
1921 wait_request_state $fid1 ARCHIVE SUCCEED
1923 $LFS hsm_release $file1
1924 $LFS hsm_restore $file2
1925 wait_request_state $fid1 RESTORE SUCCEED
1927 $LFS hsm_release $file1 || error "cannot release '$file1'"
1928 dd if=$file2 of=/dev/null bs=1M || error "cannot read '$file2'"
1930 $LFS hsm_release $file2 &&
1931 error "release should fail on read-only mount"
1935 run_test 24d "check that read-only mounts are respected"
1939 mkdir_on_mdt0 $DIR/$tdir
1941 local f=$DIR/$tdir/$tfile
1944 fid=$(create_small_file $f) || error "cannot create $f"
1945 $LFS hsm_archive $f || error "cannot archive $f"
1946 wait_request_state $fid ARCHIVE SUCCEED
1947 $LFS hsm_release $f || error "cannot release $f"
1948 while ! $LFS hsm_state $f | grep released; do
1952 tar -cf $TMP/$tfile.tar $DIR/$tdir || error "cannot tar $DIR/$tdir"
1954 run_test 24e "tar succeeds on HSM released files" # LU-6213
1957 # test needs a running copytool
1959 mkdir_on_mdt0 $DIR/$tdir
1961 mkdir -p $DIR/$tdir/d1
1962 local f=$DIR/$tdir/$tfile
1963 local fid=$(copy_file /etc/hosts $f)
1966 $LFS hsm_archive $f ||
1967 error "hsm_archive failed"
1968 wait_request_state $fid ARCHIVE SUCCEED
1969 $LFS hsm_release $f || error "cannot release $f"
1970 tar --xattrs -cvf $f.tar -C $DIR/$tdir $tfile
1973 tar --xattrs -xvf $f.tar -C $DIR/$tdir ||
1974 error "Can not recover the tar contents"
1976 echo "Sum0 = $sum0, sum1 = $sum1"
1977 [ "$sum0" == "$sum1" ] || error "md5sum mismatch for '$tfile'"
1979 run_test 24f "root can archive, release, and restore tar files"
1982 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
1983 skip "need MDS version 2.11.56 or later"
1985 local file=$DIR/$tdir/$tfile
1988 echo "RUNAS = '$RUNAS'"
1992 mkdir_on_mdt0 $DIR/$tdir
1993 chmod ugo+rwx $DIR/$tdir
1995 echo "Please listen carefully as our options have changed." | tee $file
1996 fid=$(path2fid $file)
1999 $LFS hsm_archive $file
2000 wait_request_state $fid ARCHIVE SUCCEED
2001 check_hsm_flags $file 0x00000009 # exists archived
2003 echo "To be electrocuted by your telephone, press #." | $RUNAS tee $file
2004 check_hsm_flags $file 0x0000000b # exists dirty archived
2006 run_test 24g "write by non-owner still sets dirty" # LU-11369
2009 # test needs a running copytool
2013 copy2archive /etc/hosts $tdir/$tfile
2015 local f=$DIR/$tdir/$tfile
2017 copytool import $tdir/$tfile $f
2019 $LFS hsm_set --lost $f
2024 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2026 run_test 25a "Restore lost file (HS_LOST flag) from import"\
2027 " (Operation not permitted)"
2030 # test needs a running copytool
2033 mkdir_on_mdt0 $DIR/$tdir
2035 local f=$DIR/$tdir/$tfile
2036 local fid=$(copy_file /etc/passwd $f)
2038 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2039 wait_request_state $fid ARCHIVE SUCCEED
2042 $LFS hsm_set --lost $f
2046 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2048 run_test 25b "Restore lost file (HS_LOST flag) after release"\
2049 " (Operation not permitted)"
2051 test_26A() { # was test_26
2052 # test needs a running copytool
2054 mkdir_on_mdt0 $DIR/$tdir
2056 local f=$DIR/$tdir/$tfile
2057 local fid=$(create_empty_file "$f")
2059 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2060 wait_request_state $fid ARCHIVE SUCCEED
2063 wait_request_state $fid REMOVE SUCCEED
2065 check_hsm_flags $f "0x00000000"
2067 run_test 26A "Remove the archive of a valid file"
2070 local raolu=$(get_hsm_param remove_archive_on_last_unlink)
2071 [[ $raolu -eq 0 ]] || error "RAoLU policy should be off"
2073 # test needs a running copytool
2075 mkdir_on_mdt0 $DIR/$tdir
2077 local f=$DIR/$tdir/$tfile
2078 local fid=$(copy_file /etc/passwd $f)
2080 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2081 wait_request_state $fid ARCHIVE SUCCEED
2083 local f2=$DIR/$tdir/${tfile}_2
2084 local fid2=$(copy_file /etc/passwd $f2)
2086 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
2087 wait_request_state $fid2 ARCHIVE SUCCEED
2089 local f3=$DIR/$tdir/${tfile}_3
2090 local fid3=$(copy_file /etc/passwd $f3)
2092 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f3
2093 wait_request_state $fid3 ARCHIVE SUCCEED
2095 # set a long grace_delay vs short loop_period
2096 local orig_loop_period=$(get_hsm_param loop_period)
2097 local orig_grace_delay=$(get_hsm_param grace_delay)
2098 stack_trap "set_hsm_param loop_period $orig_loop_period" EXIT
2099 set_hsm_param loop_period 10
2100 stack_trap "set_hsm_param grace_delay $orig_grace_delay" EXIT
2101 set_hsm_param grace_delay 100
2105 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2106 set_hsm_param remove_archive_on_last_unlink 1
2108 ln "$f3" "$f3"_bis || error "Unable to create hard-link"
2113 wait_request_state $fid2 REMOVE SUCCEED
2115 assert_request_count $fid REMOVE 0 \
2116 "Unexpected archived data remove request for $f"
2117 assert_request_count $fid3 REMOVE 0 \
2118 "Unexpected archived data remove request for $f3"
2120 run_test 26a "Remove Archive On Last Unlink (RAoLU) policy"
2123 # test needs a running copytool
2125 mkdir_on_mdt0 $DIR/$tdir
2127 local f=$DIR/$tdir/$tfile
2128 local fid=$(copy_file /etc/passwd $f)
2130 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2131 wait_request_state $fid ARCHIVE SUCCEED
2133 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2134 set_hsm_param remove_archive_on_last_unlink 1
2137 cdt_check_state stopped
2141 wait_request_state $fid REMOVE WAITING
2145 # copytool must re-register
2147 wait_copytools || error "copytool failed to stop"
2150 wait_request_state $fid REMOVE SUCCEED
2152 run_test 26b "RAoLU policy when CDT off"
2155 # test needs a running copytool
2157 mkdir_on_mdt0 $DIR/$tdir
2159 local f=$DIR/$tdir/$tfile
2160 local fid=$(copy_file /etc/passwd $f)
2162 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2163 wait_request_state $fid ARCHIVE SUCCEED
2165 local f2=$DIR/$tdir/${tfile}_2
2166 local fid2=$(copy_file /etc/passwd $f2)
2168 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
2169 wait_request_state $fid2 ARCHIVE SUCCEED
2171 # set a long grace_delay vs short loop_period
2172 local orig_loop_period=$(get_hsm_param loop_period)
2173 local orig_grace_delay=$(get_hsm_param grace_delay)
2174 stack_trap "set_hsm_param loop_period $orig_loop_period" EXIT
2175 set_hsm_param loop_period 10
2176 stack_trap "set_hsm_param grace_delay $orig_grace_delay" EXIT
2177 set_hsm_param grace_delay 100
2179 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2180 set_hsm_param remove_archive_on_last_unlink 1
2182 multiop_bg_pause $f O_c || error "open $f failed"
2188 wait_request_state $fid2 REMOVE SUCCEED
2189 assert_request_count $fid REMOVE 0 \
2190 "Unexpected archived data remove request for $f"
2192 kill -USR1 $pid || error "multiop early exit"
2193 # should reach autotest timeout if multiop fails to trap
2194 # signal, close file, and exit ...
2195 wait $pid || error "wait PID $PID failed"
2197 wait_request_state $fid REMOVE SUCCEED
2199 run_test 26c "RAoLU effective when file closed"
2202 # test needs a running copytool
2204 mkdir_on_mdt0 $DIR/$tdir
2206 local f=$DIR/$tdir/$tfile
2207 local fid=$(create_small_file $f)
2209 $LFS hsm_archive $f || error "could not archive file"
2210 wait_request_state $fid ARCHIVE SUCCEED
2212 # set a long grace_delay vs short loop_period
2213 local orig_loop_period=$(get_hsm_param loop_period)
2214 local orig_grace_delay=$(get_hsm_param grace_delay)
2215 stack_trap "set_hsm_param loop_period $orig_loop_period" EXIT
2216 set_hsm_param loop_period 10
2217 stack_trap "set_hsm_param grace_delay $orig_grace_delay" EXIT
2218 set_hsm_param grace_delay 100
2220 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2221 set_hsm_param remove_archive_on_last_unlink 1
2223 multiop_bg_pause $f O_c || error "multiop failed"
2230 wait_request_state $fid REMOVE SUCCEED
2232 client_up || client_up || true
2234 kill -USR1 $MULTIPID
2235 wait $MULTIPID || error "multiop close failed"
2237 run_test 26d "RAoLU when Client eviction"
2240 # test needs a running copytool
2243 create_archive_file $tdir/$tfile
2244 local f=$DIR/$tdir/$tfile
2245 copytool import $tdir/$tfile $f
2246 local fid=$(path2fid $f)
2250 [[ $? != 0 ]] || error "Remove of a released file should fail"
2252 run_test 27a "Remove the archive of an imported file (Operation not permitted)"
2255 # test needs a running copytool
2257 mkdir_on_mdt0 $DIR/$tdir
2259 local f=$DIR/$tdir/$tfile
2260 local fid=$(create_empty_file "$f")
2262 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2263 wait_request_state $fid ARCHIVE SUCCEED
2268 [[ $? != 0 ]] || error "Remove of a released file should fail"
2270 run_test 27b "Remove the archive of a relased file (Operation not permitted)"
2273 # test needs a running copytool
2275 mkdir_on_mdt0 $DIR/$tdir
2277 local f=$DIR/$tdir/$tfile
2278 local fid=$(create_empty_file "$f")
2280 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2281 wait_request_state $fid ARCHIVE SUCCEED
2290 wait_request_state $fid REMOVE SUCCEED
2292 run_test 28 "Concurrent archive/file remove"
2295 # Tests --mntpath and --archive options
2298 copytool setup -m "$MOUNT" -a $archive_id
2300 # Bad archive number
2301 $LFS hsm_remove -m "$MOUNT" -a 33 0x857765760:0x8:0x2 2>&1 |
2302 grep "Invalid argument" ||
2303 error "unexpected hsm_remove failure (1)"
2305 # mntpath is present but file is given
2306 $LFS hsm_remove --mntpath "$MOUNT" --archive 30 /qwerty/uyt 2>&1 |
2307 grep "hsm: '/qwerty/uyt' is not a valid FID" ||
2308 error "unexpected hsm_remove failure (2)"
2310 run_test 29a "Tests --mntpath and --archive options"
2313 # test needs a running copytool
2315 mkdir_on_mdt0 $DIR/$tdir
2317 local f=$DIR/$tdir/$tfile
2318 local fid=$(create_small_file $f)
2320 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2321 wait_request_state $fid ARCHIVE SUCCEED
2325 $LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER $fid
2326 wait_request_state $fid REMOVE SUCCEED
2328 run_test 29b "Archive/delete/remove by FID from the archive."
2331 # test needs a running copytool
2333 mkdir_on_mdt0 $DIR/$tdir
2335 local fid1=$(create_small_file $DIR/$tdir/$tfile-1)
2336 local fid2=$(create_small_file $DIR/$tdir/$tfile-2)
2337 local fid3=$(create_small_file $DIR/$tdir/$tfile-3)
2339 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/$tfile-[1-3]
2340 wait_request_state $fid1 ARCHIVE SUCCEED
2341 wait_request_state $fid2 ARCHIVE SUCCEED
2342 wait_request_state $fid3 ARCHIVE SUCCEED
2344 rm -f $DIR/$tdir/$tfile-[1-3]
2346 echo $fid1 > $DIR/$tdir/list
2347 echo $fid2 >> $DIR/$tdir/list
2348 echo $fid3 >> $DIR/$tdir/list
2350 $LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER \
2351 --filelist $DIR/$tdir/list
2352 wait_request_state $fid1 REMOVE SUCCEED
2353 wait_request_state $fid2 REMOVE SUCCEED
2354 wait_request_state $fid3 REMOVE SUCCEED
2356 run_test 29c "Archive/delete/remove by FID, using a file list."
2359 # test needs more than one CT
2360 needclients 3 || return 0
2366 # start all of the copytools
2367 for n in $(seq $AGTCOUNT); do
2368 copytool setup -f agt$n -a $n
2371 mkdir_on_mdt0 $DIR/$tdir
2374 file=$DIR/$tdir/$tfile
2375 fid=$(create_small_file $file)
2377 $LFS hsm_archive $file
2378 wait_request_state $fid ARCHIVE SUCCEED
2379 check_hsm_flags $file "0x00000009"
2383 $LFS hsm_remove --mntpath "$MOUNT" -a 0 $fid ||
2384 error "cannot hsm_remove '$fid'"
2386 # give time for CDT to handle remove request and create broadcasted
2389 # remove request has been broadcasted ?
2390 local cnt=$(get_request_count $fid REMOVE)
2391 # broadcasted requests + original
2392 [[ $cnt -eq $((AGTCOUNT + 1)) ]] ||
2393 error "remove not broadcasted to all CTs"
2395 # give time for CDT and CTs to handle broadcasted
2396 wait_for_loop_period
2398 # each agent serves one different archive_id, so broadcasted
2399 # hsm_remove request should only succeed once and fail at all others
2403 for n in $(seq $AGTCOUNT); do
2404 res=$(do_facet $SINGLEMDS "$LCTL get_param -n \
2405 $HSM_PARAM.actions | awk \
2406 '/'$fid'.*action=REMOVE archive#='$n'/ \
2407 {print \\\$13}' | cut -f2 -d=")
2408 if [[ "$res" == "SUCCEED" ]]; then
2410 elif [[ "$res" == "FAILED" ]]; then
2415 [[ $scnt -eq 1 ]] ||
2416 error "one and only CT should have removed successfully"
2418 [[ $AGTCOUNT -eq $((scnt + fcnt)) ]] ||
2419 error "all but one CT should have failed to remove"
2421 run_test 29d "hsm_remove by FID with archive_id 0 for unlinked file cause "\
2422 "request to be sent once for each registered archive_id"
2425 # restore at exec cannot work on agent node (because of Linux kernel
2426 # protection of executables)
2427 needclients 2 || return 0
2429 # test needs a running copytool
2433 copy2archive /bin/true $tdir/$tfile
2435 local f=$DIR/$tdir/true
2436 copytool import $tdir/$tfile $f
2438 local fid=$(path2fid $f)
2440 stack_trap "cdt_clear_no_retry" EXIT
2441 # set no retry action mode
2448 [[ $st == 0 ]] || error "Failed to exec a released file"
2450 run_test 30a "Restore at exec (import case)"
2453 # restore at exec cannot work on agent node (because of Linux kernel
2454 # protection of executables)
2455 needclients 2 || return 0
2457 # test needs a running copytool
2460 mkdir_on_mdt0 $DIR/$tdir
2461 local f=$DIR/$tdir/true
2462 local fid=$(copy_file /bin/true $f)
2464 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2465 wait_request_state $fid ARCHIVE SUCCEED
2469 stack_trap cdt_clear_no_retry EXIT
2470 # set no retry action mode
2478 [[ $st == 0 ]] || error "Failed to exec a released file"
2480 run_test 30b "Restore at exec (release case)"
2483 needclients 2 || return 0
2485 # test needs a running copytool
2488 mkdir_on_mdt0 $DIR/$tdir
2489 local f=$DIR/$tdir/SLEEP
2490 local slp_sum1=$(md5sum /bin/sleep)
2491 local fid=$(copy_file /bin/sleep $f)
2493 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2494 wait_request_state $fid ARCHIVE SUCCEED
2496 check_hsm_flags $f "0x0000000d"
2498 stack_trap cdt_clear_no_retry EXIT
2499 # set no retry action mode
2502 do_node $CLIENT2 "$f 10" &
2506 [[ $? == 0 ]] && error "Update during exec of released file must fail"
2508 [[ $? == 0 ]] || error "Execution failed during run"
2510 if [[ $? != 0 ]]; then
2511 local slp_sum2=$(md5sum /bin/sleep)
2512 # in case sleep file is modified during the test
2513 [[ $slp_sum1 == $slp_sum2 ]] &&
2514 error "Binary overwritten during exec"
2517 check_hsm_flags $f "0x00000009"
2519 run_test 30c "Update during exec of released file must fail"
2521 restore_and_check_size() {
2524 local s=$(stat -c "%s" $f)
2526 local st=$(get_hsm_flags $f)
2530 while [[ "$st" != "0x00000009" && $cpt -le 10 ]]
2532 n=$(stat -c "%s" $f)
2533 # we echo in both cases to show stat is not hang
2534 if [[ $n != $s ]]; then
2535 echo "size seen is $n != $s"
2538 echo "size seen is right: $n == $s"
2542 st=$(get_hsm_flags $f)
2544 if [[ "$st" = "0x00000009" ]]; then
2547 echo " restore is too long"
2548 wait_request_state $fid RESTORE SUCCEED
2554 # test needs a running copytool
2556 mkdir_on_mdt0 $DIR/$tdir
2558 create_archive_file $tdir/$tfile
2559 local f=$DIR/$tdir/$tfile
2560 copytool import $tdir/$tfile $f
2561 local fid=$($LFS path2fid $f)
2564 restore_and_check_size $f $fid
2567 [[ $err -eq 0 ]] || error "File size changed during restore"
2569 run_test 31a "Import a large file and check size during restore"
2573 # test needs a running copytool
2575 mkdir_on_mdt0 $DIR/$tdir
2577 local f=$DIR/$tdir/$tfile
2578 local fid=$(create_file "$f" 1MB 39)
2580 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2581 wait_request_state $fid ARCHIVE SUCCEED
2584 restore_and_check_size $f $fid
2587 [[ $err -eq 0 ]] || error "File size changed during restore"
2589 run_test 31b "Restore a large unaligned file and check size during restore"
2592 # test needs a running copytool
2594 mkdir_on_mdt0 $DIR/$tdir
2596 local f=$DIR/$tdir/$tfile
2597 local fid=$(create_file "$f" 1M 39)
2599 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2600 wait_request_state $fid ARCHIVE SUCCEED
2603 restore_and_check_size $f $fid
2606 [[ $err -eq 0 ]] || error "File size changed during restore"
2608 run_test 31c "Restore a large aligned file and check size during restore"
2611 local f=$DIR/$tdir/$tfile
2613 mkdir_on_mdt0 $DIR/$tdir
2614 local fid=$(create_empty_file "$f")
2618 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2619 wait_request_state $fid ARCHIVE SUCCEED
2622 # Prevent restore from completing
2626 md5sum $f >/dev/null &
2629 wait_request_state $fid RESTORE STARTED
2634 # Check restore trigger process was killed
2636 [ $? -eq 143 ] || error "md5sum was not 'Terminated'"
2638 run_test 33 "Kill a restore waiting process"
2641 # test needs a running copytool
2643 mkdir_on_mdt0 $DIR/$tdir
2645 local f=$DIR/$tdir/$tfile
2646 local fid=$(create_empty_file "$f")
2648 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2649 wait_request_state $fid ARCHIVE SUCCEED
2652 # Prevent restore from completing
2655 md5sum $f >/dev/null &
2658 wait_request_state $fid RESTORE STARTED
2660 # rm must not block during restore
2661 timeout --signal=KILL 1 rm "$f" || error "rm $f failed"
2664 wait_request_state $fid RESTORE SUCCEED
2666 # Check md5sum pgm finished
2667 kill -0 $pid && error "Restore initiatior still running"
2668 wait $pid || error "Restore initiator failed with $?"
2670 # Check the file was actually deleted
2671 [ ! -f "$f" ] || error "$f was not deleted"
2673 run_test 34 "Remove file during restore"
2676 # test needs a running copytool
2678 mkdir_on_mdt0 $DIR/$tdir
2680 local f=$DIR/$tdir/$tfile
2681 local f1=$DIR/$tdir/$tfile-1
2682 local fid=$(create_empty_file "$f")
2683 local fid1=$(copy_file /etc/passwd $f1)
2685 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2686 wait_request_state $fid ARCHIVE SUCCEED
2689 # Prevent restore from completing
2692 md5sum $f >/dev/null &
2695 wait_request_state $fid RESTORE STARTED
2697 # mv must not block during restore
2698 timeout --signal=KILL 1 mv "$f1" "$f" || error "mv $f1 $f failed"
2701 wait_request_state $fid RESTORE SUCCEED
2703 # Check md5sum pgm finished
2704 kill -0 $pid && error "Restore initiatior still running"
2705 wait $pid || error "Restore initiator failed with $?"
2707 local fid2=$(path2fid $f)
2708 [[ $fid2 == $fid1 ]] || error "Wrong fid after mv $fid2 != $fid1"
2710 run_test 35 "Overwrite file during restore"
2713 # test needs a running copytool
2715 mkdir_on_mdt0 $DIR/$tdir
2717 local f=$DIR/$tdir/$tfile
2718 local fid=$(create_empty_file "$f")
2720 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2721 wait_request_state $fid ARCHIVE SUCCEED
2724 # Prevent restore from completing
2727 md5sum $f >/dev/null &
2730 wait_request_state $fid RESTORE STARTED
2732 # mv must not block during restore
2733 timeout --signal=KILL 10 mv "$f" "$f.new" ||
2734 error "mv '$f' '$f.new' failed with rc=$?"
2737 wait_request_state $fid RESTORE SUCCEED
2739 # Check md5sum pgm finished
2740 kill -0 $pid && error "Restore initiator is still running"
2741 wait $pid || error "Restore initiator failed with $?"
2743 run_test 36 "Move file during restore"
2746 # LU-5683: check that an archived dirty file can be rearchived.
2748 mkdir_on_mdt0 $DIR/$tdir
2750 local f=$DIR/$tdir/$tfile
2753 fid=$(create_small_file $f) || error "cannot create small file"
2755 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2756 wait_request_state $fid ARCHIVE SUCCEED
2757 $LFS hsm_release $f || error "cannot release $f"
2759 # Allow previous archive request to expire from the actions log.
2760 wait_for_grace_delay
2763 dd if=/dev/urandom of=$f bs=1M count=1 || error "cannot dirty file"
2765 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2766 wait_request_state $fid ARCHIVE SUCCEED
2768 run_test 37 "re-archive a dirty file"
2775 for n in $(seq 1 $count); do
2776 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $prefix.$n
2778 echo "$count archive requests submitted"
2782 local stream_count=4
2783 local file_count=100
2785 local f=$DIR/$tdir/$tfile
2789 local max_requests=$(get_hsm_param max_requests)
2791 stack_trap "set_hsm_param max_requests $max_requests" EXIT
2792 # Increase the number of HSM request that can be performed in
2793 # parallel. With the coordinator running once per second, this
2794 # also limits the number of requests per seconds that can be
2795 # performed, so we pick a decent number. But we also need to keep
2796 # that number low because the copytool has no rate limit and will
2797 # fail some requests if if gets too many at once.
2798 set_hsm_param max_requests 300
2800 for i in $(seq 1 $file_count); do
2801 for p in $(seq 1 $stream_count); do
2802 fid=$(copy_file /etc/hosts $f.$p.$i)
2808 # to be sure wait_all_done will not be mislead by previous tests
2810 wait_for_grace_delay
2812 # start archive streams in background (archive files in parallel)
2813 for p in $(seq 1 $stream_count); do
2814 multi_archive $f.$p $file_count &
2817 echo -n "Wait for all requests being enqueued..."
2822 run_test 40 "Parallel archive requests"
2825 # test needs a running copytool
2828 mkdir_on_mdt0 $DIR/$tdir
2829 local f=$DIR/$tdir/$tfile
2830 local fid=$(create_small_file $f)
2832 $LFS hsm_archive $f || error "could not archive file"
2833 wait_request_state $fid ARCHIVE SUCCEED
2834 check_hsm_flags $f "0x00000009"
2836 multiop_bg_pause $f O_c || error "multiop failed"
2840 client_up || client_up || true
2842 kill -USR1 $MULTIPID
2843 wait $MULTIPID || error "multiop close failed"
2845 check_hsm_flags $f "0x0000000b"
2847 run_test 52 "Opened for write file on an evicted client should be set dirty"
2850 # test needs a running copytool
2853 mkdir_on_mdt0 $DIR/$tdir
2854 local f=$DIR/$tdir/$tfile
2855 local fid=$(create_small_file $f)
2857 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2858 error "could not archive file"
2859 wait_request_state $fid ARCHIVE SUCCEED
2860 check_hsm_flags $f "0x00000009"
2862 multiop_bg_pause $f o_c || error "multiop failed"
2866 client_up || client_up || true
2868 kill -USR1 $MULTIPID
2869 wait $MULTIPID || error "multiop close failed"
2871 check_hsm_flags $f "0x00000009"
2873 run_test 53 "Opened for read file on an evicted client should not be set dirty"
2876 mkdir_on_mdt0 $DIR/$tdir
2878 local f=$DIR/$tdir/$tfile
2879 local fid=$(create_file "$f" 1MB 39)
2883 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2884 error "could not archive file"
2885 wait_request_state $fid ARCHIVE STARTED
2887 check_hsm_flags $f "0x00000001"
2889 stack_trap "cdt_clear_no_retry" EXIT
2890 # Avoid coordinator resending this request as soon it has failed.
2895 wait_request_state $fid ARCHIVE FAILED
2897 check_hsm_flags $f "0x00000003"
2899 run_test 54 "Write during an archive cancels it"
2902 mkdir_on_mdt0 $DIR/$tdir
2904 local f=$DIR/$tdir/$tfile
2905 local fid=$(create_file "$f" 1MB 39)
2909 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2910 error "could not archive file"
2911 wait_request_state $fid ARCHIVE STARTED
2913 check_hsm_flags $f "0x00000001"
2915 stack_trap "cdt_clear_no_retry" EXIT
2916 # Avoid coordinator resending this request as soon it has failed.
2919 $TRUNCATE $f 1024 || error "truncate failed"
2921 wait_request_state $fid ARCHIVE FAILED
2923 check_hsm_flags $f "0x00000003"
2925 run_test 55 "Truncate during an archive cancels it"
2928 mkdir_on_mdt0 $DIR/$tdir
2930 local f=$DIR/$tdir/$tfile
2931 local fid=$(create_file "$f" 1MB 39)
2935 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2936 error "could not archive file"
2937 wait_request_state $fid ARCHIVE STARTED
2939 check_hsm_flags $f "0x00000001"
2941 # Change metadata and sync to be sure we are not changing only
2946 wait_request_state $fid ARCHIVE SUCCEED
2948 check_hsm_flags $f "0x00000009"
2950 run_test 56 "Setattr during an archive is ok"
2953 # Need one client for I/O, one for request
2954 needclients 2 || return 0
2956 # test needs a running copytool
2959 mkdir_on_mdt0 $DIR/$tdir
2960 local f=$DIR/$tdir/test_archive_remote
2961 # Create a file on a remote node
2962 do_node $CLIENT2 "dd if=/dev/urandom of=$f bs=1M "\
2963 "count=2 conv=fsync"
2966 do_node $CLIENT2 "$LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f" ||
2967 error "hsm_archive failed"
2968 local fid=$(path2fid $f)
2969 wait_request_state $fid ARCHIVE SUCCEED
2971 # Release and implicit restore it
2972 do_node $CLIENT2 "$LFS hsm_release $f" ||
2973 error "hsm_release failed"
2974 do_node $CLIENT2 "md5sum $f" ||
2975 error "hsm_restore failed"
2977 wait_request_state $fid RESTORE SUCCEED
2979 run_test 57 "Archive a file with dirty cache on another node"
2981 truncate_released_file() {
2985 local sz=$(stat -c %s $src_file)
2986 local f=$DIR/$tdir/$tfile
2987 local fid=$(copy_file $1 $f)
2991 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2992 error "could not archive file"
2993 wait_request_state $fid ARCHIVE SUCCEED
2995 $LFS hsm_release $f || error "could not release file"
2997 $TRUNCATE $f $trunc_to || error "truncate failed"
3000 local sz1=$(stat -c %s $f)
3001 [[ $sz1 == $trunc_to ]] ||
3002 error "size after trunc: $sz1 expect $trunc_to, original $sz"
3005 check_hsm_flags $f "0x0000000b"
3007 local state=$(get_request_state $fid RESTORE)
3008 [[ "$state" == "SUCCEED" ]] ||
3009 error "truncate $sz does not trig restore, state = $state"
3011 $TRUNCATE $ref $trunc_to
3012 cmp $ref $f || error "file data wrong after truncate"
3018 # test needs a running copytool
3021 mkdir_on_mdt0 $DIR/$tdir
3023 local sz=$(stat -c %s /etc/passwd)
3025 echo "truncate up from $sz to $((sz*2))"
3026 truncate_released_file /etc/passwd $((sz*2))
3028 echo "truncate down from $sz to $((sz/2))"
3029 truncate_released_file /etc/passwd $((sz/2))
3031 echo "truncate to 0"
3032 truncate_released_file /etc/passwd 0
3034 run_test 58 "Truncate a released file will trigger restore"
3038 [[ $MDS1_VERSION -lt $(version_code 2.7.63) ]] &&
3039 skip "Need MDS version at least 2.7.63"
3042 $MCREATE $DIR/$tfile || error "mcreate failed"
3043 $TRUNCATE $DIR/$tfile 42 || error "truncate failed"
3044 $LFS hsm_archive $DIR/$tfile || error "archive request failed"
3045 fid=$(path2fid $DIR/$tfile)
3046 wait_request_state $fid ARCHIVE SUCCEED
3047 $LFS hsm_release $DIR/$tfile || error "release failed"
3049 run_test 59 "Release stripeless file with non-zero size"
3052 # This test validates the fix for LU-4512. Ensure that the -u
3053 # option changes the progress reporting interval from the
3054 # default (30 seconds) to the user-specified interval.
3055 mkdir_on_mdt0 $DIR/$tdir
3057 local f=$DIR/$tdir/$tfile
3058 local fid=$(create_file "$f" 1M 10)
3061 local progress_timeout=$((interval * 4))
3062 copytool setup -b 1 --update-interval $interval
3065 local mdt=${MDT_PREFIX}${mdtidx}
3066 local mds=mds$((mdtidx + 1))
3068 # Wait for copytool to register
3069 wait_update_facet $mds \
3070 "$LCTL get_param -n ${mdt}.hsm.agents | grep -o ^uuid" \
3071 uuid 100 || error "coyptool failed to register with $mdt"
3073 local start_at=$(date +%s)
3074 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3075 error "could not archive file"
3077 local agent=$(facet_active_host $SINGLEAGT)
3078 local logfile=$(copytool_logfile $SINGLEAGT)
3080 wait_update $agent \
3081 "grep -o start.copy \"$logfile\"" "start copy" 100 ||
3082 error "copytool failed to start"
3084 local cmd="$LCTL get_param -n ${mdt}.hsm.active_requests"
3085 cmd+=" | awk '/'$fid'.*action=ARCHIVE/ {print \\\$12}' | cut -f2 -d="
3091 echo -n "Expecting a progress update within $progress_timeout seconds... "
3093 RESULT=$(do_node $(facet_active_host $mds) "$cmd")
3094 if [ -n "$RESULT" ] && [ "$RESULT" -gt 0 ]; then
3095 echo "$RESULT bytes copied in $WAIT seconds."
3097 elif [ $WAIT -ge $progress_timeout ]; then
3098 error "Timed out waiting for progress update!"
3101 WAIT=$((WAIT + sleep))
3105 local finish_at=$(date +%s)
3106 local elapsed=$((finish_at - start_at))
3108 # Ensure that the progress update occurred within the expected window.
3109 if [ $elapsed -lt $((interval - 1)) ]; then
3110 error "Expected progress update after at least $interval seconds"
3113 echo "Wait for on going archive hsm action to complete"
3114 wait_update $agent "grep -o copied \"$logfile\"" "copied" 10 ||
3115 echo "File archiving not completed even after 10 secs"
3117 run_test 60 "Changing progress update interval from default"
3120 # test needs a running copytool
3123 mkdir_on_mdt0 $DIR/$tdir
3124 local f=$DIR/$tdir/$tfile
3125 local fid=$(copy_file /etc/passwd $f)
3127 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3130 wait_request_state $fid ARCHIVE FAILED
3132 run_test 61 "Waiting archive of a removed file should fail"
3135 # test needs a new running copytool
3136 stack_trap copytool_monitor_cleanup EXIT
3137 copytool_monitor_setup
3138 copytool setup --event-fifo "$HSMTOOL_MONITOR_DIR/fifo"
3140 # Wait for the copytool to register.
3141 wait_update --verbose $(facet_active_host mds1) \
3142 "$LCTL get_param -n ${MDT_PREFIX}0.hsm.agents | grep -o ^uuid" \
3144 error "copytool failed to register with MDT0000"
3147 wait_copytools || error "Copytools failed to stop"
3149 local REGISTER_EVENT
3150 local UNREGISTER_EVENT
3151 while read event; do
3152 local parsed=$(parse_json_event "$event")
3153 if [ -z "$parsed" ]; then
3154 error "Copytool sent malformed event: $event"
3158 if [ $event_type == "REGISTER" ]; then
3159 REGISTER_EVENT=$event
3160 elif [ $event_type == "UNREGISTER" ]; then
3161 UNREGISTER_EVENT=$event
3163 done < <(echo $"$(get_copytool_event_log)")
3165 if [ -z "$REGISTER_EVENT" ]; then
3166 error "Copytool failed to send register event to FIFO"
3169 if [ -z "$UNREGISTER_EVENT" ]; then
3170 error "Copytool failed to send unregister event to FIFO"
3173 echo "Register/Unregister events look OK."
3175 run_test 70 "Copytool logs JSON register/unregister events to FIFO"
3178 # Bump progress interval for livelier events.
3181 # test needs a new running copytool
3182 stack_trap copytool_monitor_cleanup EXIT
3183 copytool_monitor_setup
3184 copytool setup --update-interval $interval --event-fifo \
3185 "$HSMTOOL_MONITOR_DIR/fifo"
3187 stack_trap "cdt_clear_no_retry" EXIT
3188 # Just start and stop the copytool to generate events.
3191 mkdir_on_mdt0 $DIR/$tdir
3193 local f=$DIR/$tdir/$tfile
3194 local fid=$(create_small_file "$f")
3196 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3197 error "could not archive file"
3198 wait_request_state $fid ARCHIVE SUCCEED
3200 local expected_fields="event_time data_fid source_fid"
3201 expected_fields+=" total_bytes current_bytes"
3204 [ARCHIVE_START]=false
3205 [ARCHIVE_FINISH]=false
3206 [ARCHIVE_RUNNING]=false
3208 while read event; do
3209 # Make sure we're not getting anything from previous events.
3210 for field in $expected_fields; do
3214 local parsed=$(parse_json_event "$event")
3215 if [ -z "$parsed" ]; then
3216 error "Copytool sent malformed event: $event"
3220 events["$event_type"]=true
3222 [ "$event_type" != ARCHIVE_RUNNING ] && continue
3224 # Do some simple checking of the progress update events.
3225 for expected_field in $expected_fields; do
3226 if [ -z ${!expected_field+x} ]; then
3227 error "Missing $expected_field field in event"
3231 [ $total_bytes -gt 0 ] || error "Expected total_bytes to be > 0"
3233 # These should be identical throughout an archive operation
3234 [ $source_fid == $data_fid ] ||
3235 error "Expected source_fid to equal data_fid"
3236 done < <(echo $"$(get_copytool_event_log)")
3238 # Check we received every type of events we were expecting
3239 for event in "${!events[@]}"; do
3240 ${events["$event"]} ||
3241 error "Copytool failed to send '$event' event to FIFO"
3244 echo "Archive events look OK."
3246 run_test 71 "Copytool logs JSON archive events to FIFO"
3249 # Bump progress interval for livelier events.
3252 # test needs a new running copytool
3253 stack_trap copytool_monitor_cleanup EXIT
3254 copytool_monitor_setup
3255 copytool setup --update-interval $interval --event-fifo \
3256 "$HSMTOOL_MONITOR_DIR/fifo"
3257 local test_file=$HSMTOOL_MONITOR_DIR/file
3259 local cmd="dd if=/dev/urandom of=$test_file count=16 bs=1000000 "
3261 do_facet $SINGLEAGT "$cmd" ||
3262 error "cannot create $test_file on $SINGLEAGT"
3263 copy2archive $test_file $tdir/$tfile
3265 mkdir_on_mdt0 $DIR/$tdir
3266 local f=$DIR/$tdir/$tfile
3267 copytool import $tdir/$tfile $f
3268 f=$DIR2/$tdir/$tfile
3269 echo "Verifying released state: "
3270 check_hsm_flags $f "0x0000000d"
3272 local fid=$(path2fid $f)
3274 wait_request_state $fid RESTORE SUCCEED
3276 local expected_fields="event_time data_fid source_fid"
3277 expected_fields+=" total_bytes current_bytes"
3281 while read event; do
3282 # Make sure we're not getting anything from previous events.
3283 for field in $expected_fields; do
3287 local parsed=$(parse_json_event "$event")
3288 if [ -z "$parsed" ]; then
3289 error "Copytool sent malformed event: $event"
3293 if [ $event_type == "RESTORE_START" ]; then
3295 if [ $source_fid != $data_fid ]; then
3296 error "source_fid should == data_fid at start"
3299 elif [ $event_type == "RESTORE_FINISH" ]; then
3301 if [ $source_fid != $data_fid ]; then
3302 error "source_fid should == data_fid at finish"
3305 elif [ $event_type != "RESTORE_RUNNING" ]; then
3309 # Do some simple checking of the progress update events.
3310 for expected_field in $expected_fields; do
3311 if [ -z ${!expected_field+x} ]; then
3312 error "Missing $expected_field field in event"
3316 if [ $total_bytes -eq 0 ]; then
3317 error "Expected total_bytes to be > 0"
3320 # When a restore starts out, the data fid is the same as the
3321 # source fid. After the restore has gotten going, we learn
3322 # the new data fid. Once the restore has finished, the source
3323 # fid is set to the new data fid.
3325 # We test this because some monitoring software may depend on
3326 # this behavior. If it changes, then the consumers of these
3327 # events may need to be modified.
3328 if [ $source_fid == $data_fid ]; then
3329 error "source_fid should != data_fid during restore"
3331 done < <(echo $"$(get_copytool_event_log)")
3333 if [ -z "$START_EVENT" ]; then
3334 error "Copytool failed to send restore start event to FIFO"
3337 if [ -z "$FINISH_EVENT" ]; then
3338 error "Copytool failed to send restore finish event to FIFO"
3341 echo "Restore events look OK."
3343 run_test 72 "Copytool logs JSON restore events to FIFO"
3346 file_count=51 # Max number of files constrained by LNET message size
3347 mkdir_on_mdt0 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3348 local f=$DIR/$tdir/$tfile
3349 local FILELIST=/tmp/filelist.txt
3353 for i in $(seq 1 $file_count); do
3354 fid=$(copy_file /etc/hosts $f.$i)
3355 echo $f.$i >> $FILELIST
3359 # to be sure wait_all_done will not be mislead by previous tests
3361 wait_for_grace_delay
3362 $LFS hsm_archive --filelist $FILELIST ||
3363 error "cannot archive a file list"
3365 $LFS hsm_release --filelist $FILELIST ||
3366 error "cannot release a file list"
3367 $LFS hsm_restore --filelist $FILELIST ||
3368 error "cannot restore a file list"
3371 run_test 90 "Archive/restore a file list"
3373 double_verify_reset_hsm_param() {
3375 echo "Testing $HSM_PARAM.$p"
3376 local val=$(get_hsm_param $p)
3378 local val2=$(($val * 2))
3379 set_hsm_param $p $val2
3380 val=$(get_hsm_param $p)
3381 [[ $val == $val2 ]] ||
3382 error "$HSM_PARAM.$p: $val != $val2 should be (2 * $save)"
3383 echo "Set $p to 0 must failed"
3387 set_hsm_param $p $save
3389 if [[ $rc == 0 ]]; then
3390 error "we must not be able to set $HSM_PARAM.$p to 0"
3395 double_verify_reset_hsm_param loop_period
3396 double_verify_reset_hsm_param grace_delay
3397 double_verify_reset_hsm_param active_request_timeout
3398 double_verify_reset_hsm_param max_requests
3399 double_verify_reset_hsm_param default_archive_id
3401 run_test 100 "Set coordinator /proc tunables"
3408 run_test 102 "Verify coordinator control"
3411 # test needs a running copytool
3418 for i in $(seq 1 20); do
3419 fid=$(copy_file /etc/passwd $DIR/$tdir/$i)
3421 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/*
3425 echo "Current requests"
3426 local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3427 $HSM_PARAM.actions |\
3428 grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
3430 [[ -z "$res" ]] || error "Some request have not been canceled"
3432 run_test 103 "Purge all requests"
3437 mkdir_on_mdt0 $DIR/$tdir
3439 local f=$DIR/$tdir/$tfile
3440 local fid=$(create_empty_file "$f")
3442 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
3443 local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3444 $HSM_PARAM.actions |\
3445 grep $fid | cut -f16 -d=")
3447 [[ "$data1" == "$DATAHEX" ]] ||
3448 error "Data field in records is ($data1) and not ($DATAHEX)"
3452 run_test 104 "Copy tool data field"
3455 local max_requests=$(get_hsm_param max_requests)
3456 mkdir_on_mdt0 $DIR/$tdir
3459 stack_trap "set_hsm_param max_requests $max_requests" EXIT
3460 set_hsm_param max_requests 300
3463 for i in $(seq -w 1 10); do
3464 cp /etc/passwd $DIR/$tdir/$i
3465 $LFS hsm_archive $DIR/$tdir/$i
3467 local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3468 $HSM_PARAM.actions |\
3469 grep WAITING | wc -l")
3473 local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3474 $HSM_PARAM.actions |\
3475 grep WAITING | wc -l")
3478 [[ "$reqcnt1" == "$reqcnt2" ]] ||
3479 error "Requests count after shutdown $reqcnt2 != "\
3480 "before shutdown $reqcnt1"
3482 run_test 105 "Restart of coordinator"
3485 # test needs a running copytool
3488 local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
3490 check_agent_registered $uuid
3492 search_copytools || error "No copytool found"
3495 wait_copytools || error "Copytool failed to stop"
3497 check_agent_unregistered $uuid
3500 uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
3501 check_agent_registered $uuid
3503 run_test 106 "Copytool register/unregister"
3506 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
3508 # test needs a running copytool
3510 # create and archive file
3511 mkdir_on_mdt0 $DIR/$tdir
3512 local f1=$DIR/$tdir/$tfile
3513 local fid=$(copy_file /etc/passwd $f1)
3514 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
3515 wait_request_state $fid ARCHIVE SUCCEED
3516 # shutdown and restart MDS
3518 # check the copytool still gets messages from MDT
3519 local f2=$DIR/$tdir/2
3520 local fid=$(copy_file /etc/passwd $f2)
3521 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
3522 # main check of this sanity: this request MUST succeed
3523 wait_request_state $fid ARCHIVE SUCCEED
3525 run_test 107 "Copytool re-register after MDS restart"
3527 policy_set_and_test()
3531 do_facet $SINGLEMDS $LCTL set_param "$HSM_PARAM.policy=\\\"$change\\\""
3532 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3533 [[ "$policy" == "$target" ]] ||
3534 error "Wrong policy after '$change': '$policy' != '$target'"
3538 # to force default policy setting if error
3539 CDT_POLICY_HAD_CHANGED=true
3541 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3542 local default="NonBlockingRestore [NoRetryAction]"
3543 [[ "$policy" == "$default" ]] ||
3544 error "default policy has changed,"\
3545 " '$policy' != '$default' update the test"
3546 policy_set_and_test "+NBR" "[NonBlockingRestore] [NoRetryAction]"
3547 policy_set_and_test "+NRA" "[NonBlockingRestore] [NoRetryAction]"
3548 policy_set_and_test "-NBR" "NonBlockingRestore [NoRetryAction]"
3549 policy_set_and_test "-NRA" "NonBlockingRestore NoRetryAction"
3550 policy_set_and_test "NRA NBR" "[NonBlockingRestore] [NoRetryAction]"
3551 # useless bacause we know but safer for futur changes to use real value
3552 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3553 echo "Next set_param must failed"
3554 policy_set_and_test "wrong" "$policy"
3557 echo "Back to default policy"
3558 cdt_set_sanity_policy
3560 run_test 109 "Policy display/change"
3563 # test needs a running copytool
3566 mkdir_on_mdt0 $DIR/$tdir
3568 copy2archive /etc/passwd $tdir/$tfile
3570 local f=$DIR/$tdir/$tfile
3571 copytool import $tdir/$tfile $f
3572 local fid=$(path2fid $f)
3574 cdt_set_non_blocking_restore
3579 wait_request_state $fid RESTORE SUCCEED
3580 cdt_clear_non_blocking_restore
3584 error "md5sum returns $st != 1, "\
3585 "should also perror ENODATA (No data available)"
3587 run_test 110a "Non blocking restore policy (import case)"
3590 # test needs a running copytool
3593 mkdir_on_mdt0 $DIR/$tdir
3594 local f=$DIR/$tdir/$tfile
3595 local fid=$(copy_file /etc/passwd $f)
3596 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3597 wait_request_state $fid ARCHIVE SUCCEED
3600 cdt_set_non_blocking_restore
3605 wait_request_state $fid RESTORE SUCCEED
3606 cdt_clear_non_blocking_restore
3610 error "md5sum returns $st != 1, "\
3611 "should also perror ENODATA (No data available)"
3613 run_test 110b "Non blocking restore policy (release case)"
3616 # test needs a running copytool
3619 mkdir_on_mdt0 $DIR/$tdir
3620 copy2archive /etc/passwd $tdir/$tfile
3622 local f=$DIR/$tdir/$tfile
3624 copytool import $tdir/$tfile $f
3625 local fid=$(path2fid $f)
3629 copytool_remove_backend $fid
3632 wait_request_state $fid RESTORE FAILED
3639 [[ $st == 0 ]] || error "Restore does not failed"
3641 run_test 111a "No retry policy (import case), restore will error"\
3642 " (No such file or directory)"
3645 # test needs a running copytool
3648 mkdir_on_mdt0 $DIR/$tdir
3649 local f=$DIR/$tdir/$tfile
3650 local fid=$(copy_file /etc/passwd $f)
3651 stack_trap cdt_clear_no_retry EXIT
3653 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3654 wait_request_state $fid ARCHIVE SUCCEED
3657 copytool_remove_backend $fid
3660 wait_request_state $fid RESTORE FAILED
3664 [[ $st == 0 ]] || error "Restore does not failed"
3666 run_test 111b "No retry policy (release case), restore will error"\
3667 " (No such file or directory)"
3670 # test needs a running copytool
3673 mkdir_on_mdt0 $DIR/$tdir
3674 local f=$DIR/$tdir/$tfile
3675 local fid=$(copy_file /etc/passwd $f)
3677 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3678 local l=$($LFS hsm_action $f)
3680 local res=$(echo $l | cut -f 2- -d" " | grep ARCHIVE)
3683 wait_request_state $fid ARCHIVE SUCCEED
3686 [[ ! -z "$res" ]] || error "action is $l which is not an ARCHIVE"
3688 run_test 112 "State of recorded request"
3691 mkdir_on_mdt0 $DIR/$tdir
3693 local file1=$DIR/$tdir/$tfile
3694 local file2=$DIR2/$tdir/$tfile
3696 local fid=$(create_small_sync_file $file1)
3698 stack_trap "zconf_umount \"$(facet_host $SINGLEAGT)\" \"$MOUNT3\"" EXIT
3699 zconf_mount "$(facet_host $SINGLEAGT)" "$MOUNT3" ||
3700 error "cannot mount '$MOUNT3' on '$SINGLEAGT'"
3702 copytool setup -m "$MOUNT3"
3704 do_nodes $(comma_list $(nodes_list)) $LCTL clear
3706 $LFS hsm_archive $file1 || error "Fail to archive $file1"
3707 wait_request_state $fid ARCHIVE SUCCEED
3709 $LFS hsm_release $file1
3710 echo "Verifying released state: "
3711 check_hsm_flags $file1 "0x0000000d"
3713 multiop_bg_pause $file1 oO_WRONLY:O_APPEND:_w4c || error "multiop failed"
3716 kill -USR1 $MULTIPID
3721 local size1=$(stat -c "%s" $file1)
3722 local size2=$(stat -c "%s" $file2)
3724 [ $size1 -eq $size2 ] || error "sizes are different $size1 $size2"
3726 run_test 113 "wrong stat after restore"
3729 mkdir_on_mdt0 $DIR/$tdir
3731 local f=$DIR/$tdir/$tfile
3732 local fid=$(create_empty_file "$f")
3736 # Prevent archive from completing
3739 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3740 # wait archive to register at CDT
3741 wait_request_state $fid ARCHIVE STARTED
3743 # Cancel the archive
3744 $LFS hsm_cancel "$f"
3746 wait_request_state $fid ARCHIVE CANCELED
3749 wait_request_state $fid CANCEL SUCCEED
3751 run_test 200 "Register/Cancel archive"
3754 # test needs a running copytool
3756 mkdir_on_mdt0 $DIR/$tdir
3758 local f=$DIR/$tdir/$tfile
3759 create_archive_file $tdir/$tfile
3760 copytool import $tdir/$tfile $f
3761 local fid=$(path2fid $f)
3763 # test with cdt on is made in test_222
3766 # wait restore to register at CDT
3767 wait_request_state $fid RESTORE WAITING
3770 wait_request_state $fid RESTORE CANCELED
3771 wait_request_state $fid CANCEL SUCCEED
3773 run_test 201 "Register/Cancel restore"
3776 mkdir_on_mdt0 $DIR/$tdir
3778 local f=$DIR/$tdir/$tfile
3779 local fid=$(create_empty_file "$f")
3781 # test needs a running copytool
3784 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3785 wait_request_state $fid ARCHIVE SUCCEED
3789 # wait remove to register at CDT
3790 wait_request_state $fid REMOVE STARTED
3793 wait_request_state $fid REMOVE CANCELED
3795 run_test 202 "Register/Cancel remove"
3797 test_220A() { # was test_220
3798 # test needs a running copytool
3801 mkdir_on_mdt0 $DIR/$tdir
3803 local f=$DIR/$tdir/$tfile
3804 local fid=$(copy_file /etc/passwd $f)
3808 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3809 wait_request_state $fid ARCHIVE SUCCEED
3811 changelog_find -type HSM -target-fid $fid -flags 0x0 ||
3812 error "The expected changelog was not emitted"
3814 run_test 220A "Changelog for archive"
3817 # test needs a running copytool
3820 mkdir_on_mdt0 $DIR/$tdir
3822 local f=$DIR/$tdir/$tfile
3823 local fid=$(copy_file /etc/passwd $f)
3827 # block copytool operations to allow for HSM request to be
3828 # submitted and file be unlinked (CDT will find object removed)
3831 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3833 # wait request to reach CT
3834 wait_request_state $fid ARCHIVE STARTED
3840 wait_request_state $fid ARCHIVE FAILED
3843 changelog_find -type HSM -target-fid $fid -flags 0x2 ||
3844 error "The expected changelog was not emitted"
3846 run_test 220a "Changelog for failed archive"
3849 mkdir_on_mdt0 $DIR/$tdir
3851 local f=$DIR/$tdir/$tfile
3852 local fid=$(create_empty_file "$f")
3857 # Prevent archive from completing
3859 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3860 wait_request_state $fid ARCHIVE STARTED
3863 wait_request_state $fid ARCHIVE CANCELED
3866 wait_request_state $fid CANCEL SUCCEED
3868 changelog_find -type HSM -target-fid $fid -flags 0x7d ||
3869 error "The expected changelog was not emitted"
3871 run_test 221 "Changelog for archive canceled"
3874 # test needs a running copytool
3877 mkdir_on_mdt0 $DIR/$tdir
3878 copy2archive /etc/passwd $tdir/$tfile
3880 local f=$DIR/$tdir/$tfile
3881 copytool import $tdir/$tfile $f
3882 local fid=$(path2fid $f)
3887 wait_request_state $fid RESTORE SUCCEED
3889 changelog_find -type HSM -target-fid $fid -flags 0x80 ||
3890 error "The expected changelog was not emitted"
3892 run_test 222a "Changelog for explicit restore"
3895 # test needs a running copytool
3898 mkdir_on_mdt0 $DIR/$tdir
3899 local f=$DIR/$tdir/$tfile
3900 local fid=$(copy_file /etc/passwd $f)
3903 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3904 wait_request_state $fid ARCHIVE SUCCEED
3909 wait_request_state $fid RESTORE SUCCEED
3911 changelog_find -type HSM -target-fid $fid -flags 0x80 ||
3912 error "The expected changelog was not emitted"
3914 run_test 222b "Changelog for implicit restore"
3917 # test needs a running copytool
3920 mkdir_on_mdt0 $DIR/$tdir
3921 copy2archive /etc/passwd $tdir/$tfile
3923 local f=$DIR/$tdir/$tfile
3924 copytool import $tdir/$tfile $f
3925 local fid=$(path2fid $f)
3929 # block copytool operations to allow for HSM request to be
3930 # submitted and file be unlinked (CDT will find object removed)
3935 # wait request to reach CT
3936 wait_request_state $fid RESTORE STARTED
3942 wait_request_state $fid RESTORE FAILED
3945 changelog_find -type HSM -target-fid $fid -flags 0x82 ||
3946 error "The expected changelog was not emitted"
3948 run_test 222c "Changelog for failed explicit restore"
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)
3959 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3960 wait_request_state $fid ARCHIVE SUCCEED
3963 copytool_remove_backend $fid
3966 wait_request_state $fid RESTORE FAILED
3969 changelog_find -type HSM -target-fid $fid -flags 0x82 ||
3970 error "The expected changelog was not emitted"
3972 run_test 222d "Changelog for failed implicit restore"
3975 # test needs a running copytool
3977 mkdir_on_mdt0 $DIR/$tdir
3979 local f=$DIR/$tdir/$tfile
3980 create_archive_file $tdir/$tfile
3984 copytool import $tdir/$tfile $f
3985 local fid=$(path2fid $f)
3988 wait_request_state $fid RESTORE STARTED
3990 wait_request_state $fid RESTORE CANCELED
3991 wait_request_state $fid CANCEL SUCCEED
3993 changelog_find -type HSM -target-fid $fid -flags 0xfd ||
3994 error "The expected changelog was not emitted"
3996 run_test 223a "Changelog for restore canceled (import case)"
3999 mkdir_on_mdt0 $DIR/$tdir
4001 local f=$DIR/$tdir/$tfile
4002 local fid=$(create_empty_file "$f")
4007 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4008 wait_request_state $fid ARCHIVE SUCCEED
4011 # Prevent restore from completing
4014 wait_request_state $fid RESTORE STARTED
4017 wait_request_state $fid RESTORE CANCELED
4020 wait_request_state $fid CANCEL SUCCEED
4022 changelog_find -type HSM -target-fid $fid -flags 0xfd ||
4023 error "The expected changelog was not emitted"
4025 run_test 223b "Changelog for restore canceled (release case)"
4027 test_224A() { # was test_224
4028 # test needs a running copytool
4031 mkdir_on_mdt0 $DIR/$tdir
4033 local f=$DIR/$tdir/$tfile
4034 local fid=$(copy_file /etc/passwd $f)
4037 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4038 wait_request_state $fid ARCHIVE SUCCEED
4041 wait_request_state $fid REMOVE SUCCEED
4043 changelog_find -type HSM -target-fid $fid -flags 0x200 ||
4044 error "The expected changelog was not emitted"
4046 run_test 224A "Changelog for remove"
4049 # test needs a running copytool
4052 mkdir_on_mdt0 $DIR/$tdir
4054 local f=$DIR/$tdir/$tfile
4055 local fid=$(copy_file /etc/passwd $f)
4058 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4059 wait_request_state $fid ARCHIVE SUCCEED
4061 copytool_remove_backend $fid
4063 # block copytool operations to allow for HSM request to be
4064 # submitted and file be unlinked (CDT will find object removed)
4069 # wait for request to reach CT
4070 wait_request_state $fid REMOVE STARTED
4076 wait_request_state $fid REMOVE FAILED
4078 # HE_REMOVE|ENOENT=0x202
4079 changelog_find -type HSM -target-fid $fid -flags 0x202 ||
4080 error "The expected changelog was not emitted"
4082 run_test 224a "Changelog for failed remove"
4085 # test is not usable because remove request is too fast
4086 # so it is always finished before cancel can be done ...
4087 echo "Test disabled"
4090 # test needs a running copytool
4092 mkdir_on_mdt0 $DIR/$tdir
4094 local f=$DIR/$tdir/$tfile
4095 local fid=$(create_empty_file "$f")
4098 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4099 wait_request_state $fid ARCHIVE SUCCEED
4101 # Prevent restore from completing
4106 wait_request_state $fid REMOVE CANCELED
4109 wait_request_state $fid CANCEL SUCCEED
4111 changelog_find -type HSM -target-fid $fid -flags 0x27d
4112 error "The expected changelog was not emitted"
4114 run_test 225 "Changelog for remove canceled"
4117 # test needs a running copytool
4120 mkdir_on_mdt0 $DIR/$tdir
4122 local f1=$DIR/$tdir/$tfile-1
4123 local f2=$DIR/$tdir/$tfile-2
4124 local f3=$DIR/$tdir/$tfile-3
4125 local fid1=$(copy_file /etc/passwd $f1)
4126 local fid2=$(copy_file /etc/passwd $f2)
4127 copy_file /etc/passwd $f3
4130 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
4131 wait_request_state $fid1 ARCHIVE SUCCEED
4133 $LFS hsm_archive $f2
4134 wait_request_state $fid2 ARCHIVE SUCCEED
4136 rm $f1 || error "rm $f1 failed"
4139 changelog_find -type UNLNK -target-fid $fid1 -flags 0x3 ||
4140 error "The expected changelog was not emitted"
4142 mv $f3 $f2 || error "mv $f3 $f2 failed"
4144 changelog_find -type RENME -target-fid $fid2 -flags 0x3 ||
4145 error "The expected changelog was not emitted"
4147 run_test 226 "changelog for last rm/mv with exiting archive"
4149 # This is just a utility function to clarify what test_227 does
4154 "$LFS" "$action" --$flag "$file" ||
4155 error "Cannot ${action#hsm_} $flag on '$file'"
4157 # Only one changelog should be produced
4158 local entries="$(changelog_find -type HSM -target-fid $fid)"
4159 [ $(wc -l <<< "$entries") -eq $((++count)) ] ||
4160 error "lfs $action --$flag '$file' produced more than one" \
4163 # Parse the last changelog record
4164 local entry="$(tail -n 1 <<< "$entries")"
4165 eval local -A changelog=$(changelog2array $entry)
4167 # Also check the flags match what is expected
4168 [[ ${changelog[flags]} == $target ]] ||
4169 error "Changelog flag is '${changelog[flags]}', not $target"
4173 local file="$DIR/$tdir/$tfile"
4174 local fid=$(create_empty_file "$file")
4179 for flag in norelease noarchive exists archived lost; do
4180 if [ "$flag" == lost ]; then
4181 # The flag "lost" only works on an archived file
4182 "$LFS" hsm_set --archived "$file"
4186 action="hsm_set" __test_227
4187 action="hsm_clear" __test_227
4190 run_test 227 "changelog when explicit setting of HSM flags"
4193 # test needs a running copytool
4195 mkdir_on_mdt0 $DIR/$tdir
4197 local fid=$(create_small_sync_file $DIR/$tfile)
4198 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
4199 wait_request_state $fid ARCHIVE SUCCEED
4201 $LFS hsm_release $DIR/$tfile
4202 check_hsm_flags $DIR/$tfile "0x0000000d"
4204 filefrag $DIR/$tfile | grep " 1 extent found" ||
4205 error "filefrag on released file must return only one extent"
4207 # only newer versions of cp detect sparse files by stat/FIEMAP
4209 cp --sparse=auto $DIR/$tfile $DIR/$tfile.2 ||
4210 error "copying $DIR/$tfile"
4211 cmp $DIR/$tfile $DIR/$tfile.2 || error "comparing copied $DIR/$tfile"
4213 $LFS hsm_release $DIR/$tfile
4214 check_hsm_flags $DIR/$tfile "0x0000000d"
4216 mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
4218 tar cf - --sparse $DIR/$tfile | tar xvf - -C $DIR/$tdir ||
4220 cmp $DIR/$tfile $DIR/$tdir/$DIR/$tfile ||
4221 error "comparing untarred $DIR/$tfile"
4223 rm -f $DIR/$tfile $DIR/$tfile.2 ||
4224 error "rm $DIR/$tfile or $DIR/$tfile.2 failed"
4226 run_test 228 "On released file, return extend to FIEMAP. For [cp,tar] --sparse"
4229 local file="$DIR/$tdir/$tfile"
4231 # set max_requests to allow one request of each type to be started (3)
4233 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4234 set_hsm_param max_requests 3
4237 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4238 set_hsm_param loop_period 1
4240 # send 1 requests of each kind twice
4243 mkdir_on_mdt0 $DIR/$tdir
4246 for action in archive restore remove; do
4247 local filepath="$file"-to-$action
4248 local fid=$(create_empty_file "$filepath")
4249 local fid2=$(create_empty_file "$filepath".bis)
4251 if [ "$action" != archive ]; then
4252 "$LFS" hsm_archive "$filepath"
4253 wait_request_state $fid ARCHIVE SUCCEED
4254 "$LFS" hsm_archive "$filepath".bis
4255 wait_request_state $fid2 ARCHIVE SUCCEED
4257 if [ "$action" == restore ]; then
4258 "$LFS" hsm_release "$filepath"
4259 "$LFS" hsm_release "$filepath".bis
4263 # suspend the copytool to prevent requests from completing
4264 stack_trap "copytool_continue" EXIT
4267 # send `max_requests' requests (one of each kind)
4268 for action in archive restore remove; do
4269 filepath="$file"-to-$action
4270 "$LFS" hsm_${action} "$filepath"
4271 wait_request_state $(path2fid "$filepath") "${action^^}" STARTED
4274 # send another batch of requests
4275 for action in archive restore remove; do
4276 "$LFS" hsm_${action} "$file-to-$action".bis
4278 # wait for `loop_period' seconds to make sure the coordinator has time
4279 # to register those, even though it should not
4282 # only the first batch of request should be started
4284 count=$(do_facet $SINGLEMDS "$LCTL" get_param -n $HSM_PARAM.actions |
4288 error "expected 3 STARTED requests, found $count"
4290 run_test 250 "Coordinator max request"
4293 mkdir_on_mdt0 $DIR/$tdir
4295 local f=$DIR/$tdir/$tfile
4296 local fid=$(create_empty_file "$f")
4299 # to have a short test
4300 local old_to=$(get_hsm_param active_request_timeout)
4301 set_hsm_param active_request_timeout 1
4302 # to be sure the cdt will wake up frequently so
4303 # it will be able to cancel the "old" request
4304 local old_loop=$(get_hsm_param loop_period)
4305 set_hsm_param loop_period 1
4310 # Prevent archive from completing
4312 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4313 wait_request_state $fid ARCHIVE STARTED
4315 # Let the request timeout
4316 wait_request_state $fid ARCHIVE CANCELED
4318 set_hsm_param active_request_timeout $old_to
4319 set_hsm_param loop_period $old_loop
4321 run_test 251 "Coordinator request timeout"
4324 mkdir_on_mdt0 $DIR/$tdir
4326 local f=$DIR/$tdir/$tfile
4327 local fid=$(create_empty_file "$f")
4329 # to have a short test
4330 stack_trap "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4331 set_hsm_param loop_period 1
4335 # Prevent archive from completing
4337 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4338 wait_request_state $fid ARCHIVE STARTED
4341 stack_trap "set_hsm_param active_request_timeout \
4342 $(get_hsm_param active_request_timeout)" EXIT
4343 set_hsm_param active_request_timeout 1
4345 wait_request_state $fid ARCHIVE CANCELED
4348 run_test 252 "Timeout'ed running archive of a removed file should be canceled"
4352 # test needs a running copytool
4355 mkdir_on_mdt0 $DIR/$tdir
4356 local f=$DIR/$tdir/$tfile
4358 dd if=/dev/zero of=$f bs=1MB count=10
4359 local fid=$(path2fid $f)
4361 $LFS hsm_archive $f || error "could not archive file"
4362 wait_request_state $fid ARCHIVE SUCCEED
4364 # clear locks to discard inode data
4365 cancel_lru_locks osc
4367 #define OBD_FAIL_MDC_MERGE 0x807
4368 $LCTL set_param fail_loc=0x807
4370 #expect error here, instead of release with wrong size
4373 if ((rc == 0)); then
4374 file_size=$(stat -c '%s' $f)
4375 if ((file_size != 10485760)); then
4376 error "Wrong file size after hsm_release"
4379 echo "could not release file"
4382 run_test 253 "Check for wrong file size after release"
4386 [ $MDS1_VERSION -lt $(version_code 2.10.56) ] &&
4387 skip "need MDS version at least 2.10.56"
4389 # Check that the counters are initialized to 0
4391 for request_type in archive restore remove; do
4392 count="$(get_hsm_param ${request_type}_count)" ||
4393 error "Reading ${request_type}_count failed with $?"
4395 [ "$count" -eq 0 ] ||
4396 error "Expected ${request_type}_count to be " \
4400 run_test 254a "Request counters are initialized to zero"
4404 [ $MDS1_VERSION -lt $(version_code 2.10.56) ] &&
4405 skip "need MDS version at least 2.10.56"
4407 # The number of request to launch (at least 32)
4408 local request_count=$((RANDOM % 32 + 32))
4409 printf "Will launch %i requests of each type\n" "$request_count"
4411 # Launch a copytool to process requests
4414 # Set hsm.max_requests to allow starting all requests at the same time
4416 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4417 set_hsm_param max_requests "$request_count"
4419 mkdir_on_mdt0 $DIR/$tdir
4423 for request_type in archive restore remove; do
4424 printf "Checking %s requests\n" "${request_type}"
4425 # Suspend the copytool to give us time to read the proc files
4428 for ((i = 0; i < $request_count; i++)); do
4429 case $request_type in
4431 create_empty_file "$DIR/$tdir/$tfile-$i" \
4435 lfs hsm_release "$DIR/$tdir/$tfile-$i"
4438 $LFS hsm_${request_type} "$DIR/$tdir/$tfile-$i"
4441 # Give the coordinator 10 seconds to start every request
4443 while get_hsm_param actions | grep -q WAITING; do
4446 [ $timeout -gt 0 ] ||
4447 error "${request_type^} requests took too " \
4451 count="$(get_hsm_param ${request_type}_count)"
4452 [ "$count" -eq "$request_count" ] ||
4453 error "Expected '$request_count' (!= '$count') " \
4454 "active $request_type requests"
4456 # Let the copytool process the requests
4458 # Give it 10 seconds maximum
4460 while get_hsm_param actions | grep -q STARTED; do
4463 [ $timeout -gt 0 ] ||
4464 error "${request_type^} requests took too " \
4468 count="$(get_hsm_param ${request_type}_count)"
4469 [ "$count" -eq 0 ] ||
4470 error "Expected 0 (!= '$count') " \
4471 "active $request_type requests"
4474 run_test 254b "Request counters are correctly incremented and decremented"
4478 [ $MDS1_VERSION -lt $(version_code 2.12.0) ] &&
4479 skip "Need MDS version at least 2.12.0"
4481 mkdir_on_mdt0 $DIR/$tdir
4483 local file="$DIR/$tdir/$tfile"
4484 local fid=$(create_empty_file "$file")
4486 # How do you make sure the coordinator has consumed any outstanding
4487 # event, without triggering an event yourself?
4489 # You wait for a request to disappear from the coordinator's llog.
4491 # Warning: the setup represents 90% of this test
4493 # Create and process an HSM request
4495 "$LFS" hsm_archive "$file"
4496 wait_request_state $fid ARCHIVE SUCCEED
4499 wait_copytools || error "failed to stop copytools"
4501 # Launch a new HSM request
4503 create_empty_file "$file"
4504 "$LFS" hsm_archive "$file"
4508 # Have the completed request be removed as soon as the cdt wakes up
4509 stack_trap "set_hsm_param grace_delay $(get_hsm_param grace_delay)" EXIT
4510 set_hsm_param grace_delay 1
4511 # (Hopefully, time on the MDS will behave nicely)
4512 do_facet $SINGLEMDS sleep 2 &
4514 # Increase `loop_period' as a mean to prevent the coordinator from
4515 # waking itself up to do some housekeeping.
4516 stack_trap "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4517 set_hsm_param loop_period 1000
4519 wait $! || error "waiting failed"
4521 wait_request_state $fid ARCHIVE ""
4522 # The coordinator will not wake up on its own for ~`loop_period' secs...
4524 # ... Unless a copytool registers. Now the real test begins
4526 wait_request_state $(path2fid "$file") ARCHIVE SUCCEED
4528 run_test 255 "Copytool registration wakes the coordinator up"
4530 # tests 260[a-c] rely on the parsing of the copytool's log file, they might
4531 # break in the future because of that.
4534 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
4535 skip "need MDS version 2.11.56 or later"
4537 local -a files=("$DIR/$tdir/$tfile".{0..15})
4540 mkdir_on_mdt0 $DIR/$tdir
4542 for file in "${files[@]}"; do
4543 create_small_file "$file"
4546 # Set a few hsm parameters
4548 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4549 set_hsm_param loop_period 1
4551 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4552 set_hsm_param max_requests 3
4556 "$LFS" hsm_archive "${files[0]}"
4557 wait_request_state "$(path2fid "${files[0]}")" ARCHIVE SUCCEED
4558 "$LFS" hsm_release "${files[0]}"
4562 wait_copytools || error "copytools failed to stop"
4564 # Send several archive requests
4565 for file in "${files[@]:1}"; do
4566 "$LFS" hsm_archive "$file"
4569 # Send one restore request
4570 "$LFS" hsm_restore "${files[0]}"
4575 # Wait for all the requests to complete
4576 wait_request_state "$(path2fid "${files[0]}")" RESTORE SUCCEED
4577 for file in "${files[@]:1}"; do
4578 wait_request_state "$(path2fid "$file")" ARCHIVE SUCCEED
4581 # Collect the actions in the order in which the copytool processed them
4583 $(do_facet "$SINGLEAGT" grep -o '\"RESTORE\\|ARCHIVE\"' \
4584 "$(copytool_logfile "$SINGLEAGT")")
4587 printf '%s\n' "${actions[@]}"
4590 for action in "${actions[@]:0:3}"; do
4591 [ "$action" == RESTORE ] && return
4594 error "Too many ARCHIVE requests were run before the RESTORE request"
4596 run_test 260a "Restore request have priority over other requests"
4598 # This test is very much tied to the implementation of the current priorisation
4599 # mechanism in the coordinator. It might not make sense to keep it in the future
4602 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
4603 skip "need MDS version 2.11.56 or later"
4605 local -a files=("$DIR/$tdir/$tfile".{0..15})
4608 mkdir_on_mdt0 $DIR/$tdir
4610 for file in "${files[@]}"; do
4611 create_small_file "$file"
4614 # Set a few hsm parameters
4616 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4617 set_hsm_param loop_period 1
4619 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4620 set_hsm_param max_requests 3
4623 copytool setup --archive-id 2
4624 "$LFS" hsm_archive --archive 2 "${files[0]}"
4625 wait_request_state "$(path2fid "${files[0]}")" ARCHIVE SUCCEED
4626 "$LFS" hsm_release "${files[0]}"
4630 wait_copytools || error "copytools failed to stop"
4632 # Send several archive requests
4633 for file in "${files[@]:1}"; do
4634 "$LFS" hsm_archive "$file"
4637 # Send one restore request
4638 "$LFS" hsm_restore "${files[0]}"
4642 copytool setup --archive-id 2
4644 # Wait for all the requests to complete
4645 wait_request_state "$(path2fid "${files[0]}")" RESTORE SUCCEED
4646 for file in "${files[@]:1}"; do
4647 wait_request_state "$(path2fid "$file")" ARCHIVE SUCCEED
4650 # Collect the actions in the order in which the copytool processed them
4652 $(do_facet "$SINGLEAGT" grep -o '\"RESTORE\\|ARCHIVE\"' \
4653 "$(copytool_logfile "$SINGLEAGT")")
4656 printf '%s\n' "${actions[@]}"
4659 for action in "${actions[@]:0:3}"; do
4660 [ "$action" == RESTORE ] && return
4663 error "Too many ARCHIVE requests were run before the RESTORE request"
4665 run_test 260b "Restore request have priority over other requests"
4667 # This test is very much tied to the implementation of the current priorisation
4668 # mechanism in the coordinator. It might not make sense to keep it in the future
4671 [ $MDS1_VERSION -lt $(version_code 2.12.0) ] &&
4672 skip "Need MDS version at least 2.12.0"
4674 local -a files=("$DIR/$tdir/$tfile".{0..15})
4677 mkdir_on_mdt0 $DIR/$tdir
4679 for file in "${files[@]}"; do
4680 create_small_file "$file"
4683 # Set a few hsm parameters
4685 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4686 set_hsm_param loop_period 1000
4688 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4689 set_hsm_param max_requests 3
4692 copytool setup --archive-id 2
4693 "$LFS" hsm_archive --archive 2 "${files[0]}"
4694 wait_request_state "$(path2fid "${files[0]}")" ARCHIVE SUCCEED
4695 "$LFS" hsm_release "${files[0]}"
4699 wait_copytools || error "copytools failed to stop"
4701 # Force the next coordinator run to do housekeeping
4705 "$LFS" hsm_archive "${files[1]}"
4709 copytool setup --archive-id 2
4711 wait_request_state "$(path2fid "${files[1]}")" ARCHIVE SUCCEED
4712 # The coordinator just did a housekeeping run it won't do another one
4713 # for around `loop_period' seconds => requests will not be reordered
4714 # if it costs too much (ie. when the coordinator has to discard a whole
4717 # Send several archive requests
4718 for file in "${files[@]:2}"; do
4719 "$LFS" hsm_archive "$file"
4722 # Send one restore request
4723 "$LFS" hsm_restore "${files[0]}"
4725 # Wait for all the requests to complete
4726 wait_request_state "$(path2fid "${files[0]}")" RESTORE SUCCEED
4727 for file in "${files[@]:2}"; do
4728 wait_request_state "$(path2fid "$file")" ARCHIVE SUCCEED
4731 # Collect the actions in the order in which the copytool processed them
4733 $(do_facet "$SINGLEAGT" grep -o '\"RESTORE\\|ARCHIVE\"' \
4734 "$(copytool_logfile "$SINGLEAGT")")
4737 printf '%s\n' "${actions[@]}"
4740 for action in "${actions[@]:0:3}"; do
4741 [ "$action" == RESTORE ] &&
4742 error "Restore requests should not be prioritised" \
4743 "unless the coordinator is doing housekeeping"
4747 run_test 260c "Requests are not reordered on the 'hot' path of the coordinator"
4750 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
4752 # the only way to test ondisk conf is to restart MDS ...
4753 echo "Stop coordinator and remove coordinator state at mount"
4756 # clean on disk conf set by default
4757 cdt_clear_mount_state
4758 cdt_check_state stopped
4760 # check cdt still off after umount/remount
4762 cdt_check_state stopped
4764 echo "Set coordinator start at mount, and start coordinator"
4765 cdt_set_mount_state enabled
4768 cdt_check_state enabled
4770 # check cdt still on after umount/remount
4772 cdt_check_state enabled
4774 # we are back to original state (cdt started at mount)
4776 run_test 300 "On disk coordinator state kept between MDT umount/mount"
4779 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
4781 local ai=$(get_hsm_param default_archive_id)
4782 local new=$((ai + 1))
4784 set_hsm_param default_archive_id $new -P
4786 local res=$(get_hsm_param default_archive_id)
4789 set_hsm_param default_archive_id "" "-P -d"
4791 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
4793 run_test 301 "HSM tunnable are persistent"
4796 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
4798 local ai=$(get_hsm_param default_archive_id)
4799 local new=$((ai + 1))
4804 set_hsm_param default_archive_id $new -P
4807 for mdtno in $(seq 1 $MDSCOUNT); do
4812 cdt_check_state enabled
4814 local res=$(get_hsm_param default_archive_id)
4817 set_hsm_param default_archive_id "" "-P -d"
4819 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
4821 run_test 302 "HSM tunnable are persistent when CDT is off"
4824 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4828 mkdir_on_mdt0 $DIR/$tdir
4830 local dir_mdt0=$DIR/$tdir/mdt0
4831 local dir_mdt1=$DIR/$tdir/mdt1
4833 # create 1 dir per MDT
4834 stack_trap "rm -rf $dir_mdt0" EXIT
4835 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4836 stack_trap "rm -rf $dir_mdt1" EXIT
4837 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
4839 # create 1 file in each MDT
4840 local fid1=$(create_small_file $dir_mdt0/$tfile)
4841 local fid2=$(create_small_file $dir_mdt1/$tfile)
4843 # check that hsm request on mdt0 is sent to the right MDS
4844 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
4845 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4846 echo "archive successful on mdt0"
4848 # check that hsm request on mdt1 is sent to the right MDS
4849 $LFS hsm_archive $dir_mdt1/$tfile || error "lfs hsm_archive"
4850 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
4851 echo "archive successful on mdt1"
4853 run_test 400 "Single request is sent to the right MDT"
4856 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4860 mkdir_on_mdt0 $DIR/$tdir
4862 local dir_mdt0=$DIR/$tdir/mdt0
4863 local dir_mdt1=$DIR/$tdir/mdt1
4865 # create 1 dir per MDT
4866 stack_trap "rm -rf $dir_mdt0" EXIT
4867 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4868 stack_trap "rm -rf $dir_mdt1" EXIT
4869 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
4871 # create 1 file in each MDT
4872 local fid1=$(create_small_file $dir_mdt0/$tfile)
4873 local fid2=$(create_small_file $dir_mdt1/$tfile)
4875 # check that compound requests are shunt to the rights MDTs
4876 $LFS hsm_archive $dir_mdt0/$tfile $dir_mdt1/$tfile ||
4877 error "lfs hsm_archive"
4878 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4879 echo "archive successful on mdt0"
4880 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
4881 echo "archive successful on mdt1"
4883 run_test 401 "Compound requests split and sent to their respective MDTs"
4885 mdc_change_state() # facet, MDT_pattern, activate|deactivate
4890 local node=$(facet_active_host $facet)
4892 for mdc in $(do_facet $facet "$LCTL dl | grep -E ${pattern}-mdc" |
4893 awk '{print $4}'); do
4894 echo "$3 $mdc on $node"
4895 do_facet $facet "$LCTL --device $mdc $state" || return 1
4900 # deactivate all mdc on agent1
4901 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "deactivate"
4903 copytool setup --no-fail
4905 check_agent_unregistered "uuid" # match any agent
4907 # no expected running copytool
4908 search_copytools $agent && error "Copytool start should have failed"
4911 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "activate"
4913 run_test 402a "Copytool start fails if all MDTs are inactive"
4918 mkdir_on_mdt0 $DIR/$tdir
4920 local f=$DIR/$tdir/$tfile
4921 touch $f || error "touch $f failed"
4922 local fid=$(path2fid $f)
4924 #define OBD_FAIL_MDS_HSM_CT_REGISTER_NET 0x14d
4925 do_facet $SINGLEAGT lctl set_param fail_loc=0x14d
4926 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4928 # give time for CDT to send request and to keep it for retry
4929 wait_for_loop_period
4931 wait_request_state $fid ARCHIVE WAITING
4933 do_facet $SINGLEAGT lctl set_param fail_loc=0
4935 # request should succeed now
4936 wait_request_state $fid ARCHIVE SUCCEED
4938 run_test 402b "CDT must retry request upon slow start of CT"
4941 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4943 local agent=$(facet_active_host $SINGLEAGT)
4945 # deactivate all mdc for MDT0001
4946 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
4949 local uuid=$(get_agent_uuid $agent)
4950 # check the agent is registered on MDT0000, and not on MDT0001
4951 check_agent_registered_by_mdt $uuid 0
4952 check_agent_unregistered_by_mdt $uuid 1
4954 # check running copytool process
4955 search_copytools $agent || error "No running copytools on $agent"
4957 # reactivate all mdc for MDT0001
4958 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
4960 # make sure the copytool is now registered to all MDTs
4961 check_agent_registered $uuid
4963 run_test 403 "Copytool starts with inactive MDT and register on reconnect"
4966 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4970 # create files on both MDT0000 and MDT0001
4971 mkdir_on_mdt0 $DIR/$tdir
4973 local dir_mdt0=$DIR/$tdir/mdt0
4974 stack_trap "rm -rf $dir_mdt0" EXIT
4975 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4977 # create 1 file on mdt0
4978 local fid1=$(create_small_file $dir_mdt0/$tfile)
4980 # deactivate all mdc for MDT0001
4981 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
4983 # send an HSM request for files in MDT0000
4984 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
4986 # check for completion of files in MDT0000
4987 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4988 echo "archive successful on mdt0"
4990 # reactivate all mdc for MDT0001
4991 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
4993 run_test 404 "Inactive MDT does not block requests for active MDTs"
4996 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5000 mkdir_on_mdt0 $DIR/$tdir
5002 local striped_dir=$DIR/$tdir/striped_dir
5004 # create striped dir on all of MDTs
5005 $LFS mkdir -i 0 -c $MDSCOUNT $striped_dir || error "lfs mkdir"
5007 local fid1=$(create_small_sync_file $striped_dir/${tfile}_0)
5008 local fid2=$(create_small_sync_file $striped_dir/${tfile}_1)
5009 local fid3=$(create_small_sync_file $striped_dir/${tfile}_2)
5010 local fid4=$(create_small_sync_file $striped_dir/${tfile}_3)
5012 local idx1=$($LFS getstripe -m $striped_dir/${tfile}_0)
5013 local idx2=$($LFS getstripe -m $striped_dir/${tfile}_1)
5014 local idx3=$($LFS getstripe -m $striped_dir/${tfile}_2)
5015 local idx4=$($LFS getstripe -m $striped_dir/${tfile}_3)
5017 # check that compound requests are shunt to the rights MDTs
5018 $LFS hsm_archive $striped_dir/${tfile}_0 $striped_dir/${tfile}_1 \
5019 $striped_dir/${tfile}_2 $striped_dir/${tfile}_3 ||
5020 error "lfs hsm_archive"
5022 wait_request_state $fid1 ARCHIVE SUCCEED $idx1 &&
5023 echo "archive successful on $fid1"
5024 wait_request_state $fid2 ARCHIVE SUCCEED $idx2 &&
5025 echo "archive successful on $fid2"
5026 wait_request_state $fid3 ARCHIVE SUCCEED $idx3 &&
5027 echo "archive successful on $fid3"
5028 wait_request_state $fid4 ARCHIVE SUCCEED $idx4 &&
5029 echo "archive successful on $fid4"
5031 $LFS hsm_release $striped_dir/${tfile}_0 || error "lfs hsm_release 1"
5032 $LFS hsm_release $striped_dir/${tfile}_1 || error "lfs hsm_release 2"
5033 $LFS hsm_release $striped_dir/${tfile}_2 || error "lfs hsm_release 3"
5034 $LFS hsm_release $striped_dir/${tfile}_3 || error "lfs hsm_release 4"
5036 cat $striped_dir/${tfile}_0 > /dev/null || error "cat ${tfile}_0 failed"
5037 cat $striped_dir/${tfile}_1 > /dev/null || error "cat ${tfile}_1 failed"
5038 cat $striped_dir/${tfile}_2 > /dev/null || error "cat ${tfile}_2 failed"
5039 cat $striped_dir/${tfile}_3 > /dev/null || error "cat ${tfile}_3 failed"
5041 run_test 405 "archive and release under striped directory"
5044 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
5046 [ $MDS1_VERSION -lt $(version_code 2.7.64) ] &&
5047 skip "need MDS version at least 2.7.64"
5052 mkdir_on_mdt0 $DIR/$tdir
5054 fid=$(create_small_file $DIR/$tdir/$tfile)
5059 $LFS hsm_archive $DIR/$tdir/$tfile
5060 wait_request_state "$fid" ARCHIVE SUCCEED
5061 $LFS hsm_release $DIR/$tdir/$tfile
5063 # Should migrate $tdir but not $tfile.
5064 $LFS migrate -m1 $DIR/$tdir &&
5065 error "migrating HSM an archived file should fail"
5067 $LFS hsm_restore $DIR/$tdir/$tfile
5068 wait_request_state "$fid" RESTORE SUCCEED
5070 $LFS hsm_remove $DIR/$tdir/$tfile
5071 wait_request_state "$fid" REMOVE SUCCEED
5073 cat $DIR/$tdir/$tfile > /dev/null ||
5074 error "cannot read $DIR/$tdir/$tfile"
5076 $LFS migrate -m1 $DIR/$tdir ||
5077 error "cannot complete migration after HSM remove"
5079 mdt_index=$($LFS getstripe -m $DIR/$tdir)
5080 if ((mdt_index != 1)); then
5081 error "expected MDT index 1, got $mdt_index"
5084 # Refresh fid after migration.
5085 fid=$(path2fid $DIR/$tdir/$tfile)
5088 $LFS hsm_archive $DIR/$tdir/$tfile
5089 wait_request_state "$fid" ARCHIVE SUCCEED 1
5091 lctl set_param debug=+trace
5092 $LFS hsm_release $DIR/$tdir/$tfile ||
5093 error "cannot release $DIR/$tdir/$tfile"
5095 $LFS hsm_restore $DIR/$tdir/$tfile
5096 wait_request_state "$fid" RESTORE SUCCEED 1
5098 cat $DIR/$tdir/$tfile > /dev/null ||
5099 error "cannot read $DIR/$tdir/$tfile"
5101 run_test 406 "attempting to migrate HSM archived files is safe"
5104 mkdir_on_mdt0 $DIR/$tdir
5106 local f=$DIR/$tdir/$tfile
5107 local f2=$DIR2/$tdir/$tfile
5108 local fid=$(create_empty_file "$f")
5112 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
5113 wait_request_state $fid ARCHIVE SUCCEED
5116 #define OBD_FAIL_MDS_HSM_CDT_DELAY 0x164
5117 do_facet $SINGLEMDS $LCTL set_param fail_val=5 fail_loc=0x164
5119 # Prevent restore from completing
5123 # 1st request holds layout lock while appropriate
5124 # RESTORE record is still not added to llog
5128 do_facet $SINGLEMDS "$LCTL get_param $HSM_PARAM.actions"
5129 # after umount hsm_actions->O/x/x log shouldn't have
5130 # double RESTORE records like below
5131 #[0x200000401:0x1:0x0]...0x58d03a0d/0x58d03a0c action=RESTORE...WAITING
5132 #[0x200000401:0x1:0x0]...0x58d03a0c/0x58d03a0d action=RESTORE...WAITING
5134 do_facet $SINGLEMDS "$LCTL get_param $HSM_PARAM.actions"&
5136 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
5138 do_facet $SINGLEMDS "$LCTL get_param $HSM_PARAM.actions"
5141 wait_all_done 100 $fid
5143 run_test 407 "Check for double RESTORE records in llog"
5147 [ "$MDS1_VERSION" -lt $(version_code 2.6.92) ] &&
5148 skip "HSM migrate is not supported"
5150 test_mkdir -p $DIR/$tdir
5152 if [ "$CLIENT_VERSION" -lt $(version_code 2.11.56) ] ||
5153 [ "$MDS1_VERSION" -lt $(version_code 2.11.56) ];
5155 llapi_hsm_test -d $DIR/$tdir -b ||
5156 error "One llapi HSM test failed"
5158 llapi_hsm_test -d $DIR/$tdir ||
5159 error "One llapi HSM test failed"
5162 run_test 500 "various LLAPI HSM tests"
5165 [ "$MDS1_VERSION" -lt $(version_code 2.10.58) ] &&
5166 skip "need MDS version at least 2.10.58"
5170 local f=$DIR/$tdir/$tfile
5173 # set changelog_mask to ALL
5174 changelog_chmask "ALL"
5176 chmod 777 $DIR/$tdir
5177 $RUNAS touch $f || error "touch $f failed as $RUNAS_ID"
5178 local fid=$(path2fid $f)
5181 entry=$(changelog_find -type CREAT -target-fid $fid -uid "$RUNAS_ID" \
5182 -gid "$RUNAS_GID") ||
5183 error "No matching CREAT entry"
5185 # Parse the changelog
5186 eval local -A changelog=$(changelog2array $entry)
5187 local nid="${changelog[nid]}"
5190 echo "Got NID '$nid'"
5191 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5192 error "nid '$nid' does not match any client NID:" \
5195 run_test 600 "Changelog fields 'u=' and 'nid='"
5198 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5199 skip "need MDS version at least 2.10.58"
5203 local f=$DIR/$tdir/$tfile
5206 # set changelog_mask to ALL
5207 changelog_chmask "ALL"
5209 touch $f || error "touch $f failed"
5210 local fid=$(path2fid $f)
5213 cat $f || error "cat $f failed"
5215 changelog_find -type OPEN -target-fid $fid -mode "r--" ||
5216 error "No matching OPEN entry"
5218 run_test 601 "OPEN Changelog entry"
5221 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5222 skip "need MDS version at least 2.10.58"
5224 stack_trap "restore_opencache" EXIT
5229 local f=$DIR/$tdir/$tfile
5232 # set changelog_mask to ALL
5233 changelog_chmask "ALL"
5235 touch $f || error "touch $f failed"
5236 local fid=$(path2fid $f)
5239 cat $f || error "cat $f failed"
5241 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5245 echo f > $f || error "write $f failed"
5248 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5250 # remove OPEN from changelog_mask
5251 changelog_chmask "-OPEN"
5255 cat $f || error "cat $f failed"
5258 changelog_find -type CLOSE -target-fid $fid &&
5259 error "There should be no CLOSE entry"
5263 echo f > $f || error "write $f failed"
5266 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5268 run_test 602 "Changelog record CLOSE only if open+write or OPEN recorded"
5271 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5272 skip "need MDS version at least 2.10.58"
5276 local f=$DIR/$tdir/$tfile
5279 # set changelog_mask to ALL
5280 changelog_chmask "ALL"
5282 touch $f || error "touch $f failed"
5283 local fid=$(path2fid $f)
5285 setfattr -n user.xattr1 -v "value1" $f || error "setfattr $f failed"
5288 getfattr -n user.xattr1 $f || error "getfattr $f failed"
5290 changelog_find -type GXATR -target-fid $fid -xattr "user.xattr1" ||
5291 error "No matching GXATR entry"
5293 run_test 603 "GETXATTR Changelog entry"
5296 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5297 skip "need MDS version at least 2.10.58"
5299 mkdir_on_mdt0 $DIR/$tdir
5301 local f=$DIR/$tdir/$tfile
5302 local f2=$DIR2/$tdir/$tfile
5303 local procname="mdd.$FSNAME-MDT0000.changelog_deniednext"
5305 timeout="$(do_facet mds1 "$LCTL" get_param -n "$procname")"
5306 stack_trap "do_facet mds1 '$LCTL' set_param '$procname=$timeout'" EXIT
5307 do_facet mds1 lctl set_param "$procname=20"
5311 # set changelog_mask to ALL
5312 changelog_chmask "ALL"
5314 touch $f || error "touch $f failed"
5315 local fid=$(path2fid $f)
5321 $RUNAS cat $f2 && error "cat $f2 by user $RUNAS_ID should have failed"
5325 entry=$(changelog_find -type NOPEN -target-fid $fid -uid "$RUNAS_ID" \
5326 -gid "$RUNAS_GID" -mode "r--") ||
5327 error "No matching NOPEN entry"
5329 # Parse the changelog
5330 eval local -A changelog=$(changelog2array $entry)
5331 local nid="${changelog[nid]}"
5334 echo "Got NID '$nid'"
5335 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5336 error "nid '$nid' does not match any client NID:" \
5341 $RUNAS cat $f2 && error "cat $f2 by user $RUNAS_ID should have failed"
5344 changelog_find -type NOPEN -target-fid $fid &&
5345 error "There should be no NOPEN entry"
5347 # Sleep for `changelog_deniednext` seconds
5352 $RUNAS cat $f2 && error "cat $f by user $RUNAS_ID should have failed"
5355 entry=$(changelog_find -type NOPEN -target-fid $fid -uid "$RUNAS_ID" \
5356 -gid "$RUNAS_GID" -mode "r--") ||
5357 error "No matching NOPEN entry"
5359 # Parse the changelog
5360 eval local -A changelog=$(changelog2array $entry)
5361 local nid="${changelog[nid]}"
5364 echo "Got NID '$nid'"
5365 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5366 error "nid '$nid' does not match any client NID:" \
5369 run_test 604 "NOPEN Changelog entry"
5372 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5373 skip "need MDS version at least 2.10.58"
5375 stack_trap "restore_opencache" EXIT
5380 local f=$DIR/$tdir/$tfile
5381 local f2=$DIR2/$tdir/$tfile
5384 # set changelog_mask to ALL
5385 changelog_chmask "ALL"
5387 touch $f || error "touch $f failed"
5388 local fid=$(path2fid $f)
5392 exec 3<> $f || error "open $f failed"
5396 changelog_find -type OPEN -target-fid $fid || error "No OPEN entry"
5400 exec 4<> $f || error "open $f failed"
5403 changelog_find -type OPEN -target-fid $fid &&
5404 error "There should be no OPEN entry"
5406 exec 4>&- || error "close $f failed"
5409 changelog_find -type CLOSE -target-fid $fid &&
5410 error "There should be no CLOSE entry"
5414 # access in rw, so different access mode should generate entries
5415 cat $f || error "cat $f failed"
5418 changelog_find -type OPEN -target-fid $fid || error "No OPEN entry"
5420 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5424 # same access as first one, should not generate new entries
5425 exec 4<> $f || error "open $f failed"
5428 changelog_find -type OPEN -target-fid $fid &&
5429 error "There should be no OPEN entry"
5431 exec 4>&- || error "close $f failed"
5434 changelog_find -type CLOSE -target-fid $fid &&
5435 error "There should be no CLOSE entry"
5439 # access by different user should generate new entries
5440 $RUNAS cat $f || error "cat $f by user $RUNAS_ID failed"
5443 changelog_find -type OPEN -target-fid $fid || error "No OPEN entry"
5445 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5449 exec 3>&- || error "close $f failed"
5452 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5454 run_test 605 "Test OPEN and CLOSE rate limit in Changelogs"
5457 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5458 skip "need MDS version at least 2.10.58"
5460 local llog_reader=$(do_facet mgs "which llog_reader 2> /dev/null")
5461 llog_reader=${llog_reader:-$LUSTRE/utils/llog_reader}
5462 [ -z $(do_facet mgs ls -d $llog_reader 2> /dev/null) ] &&
5463 skip_env "missing llog_reader"
5465 mkdir_on_mdt0 $DIR/$tdir
5467 local f=$DIR/$tdir/$tfile
5470 # set changelog_mask to ALL
5471 changelog_chmask "ALL"
5473 chmod 777 $DIR/$tdir
5474 $RUNAS touch $f || error "touch $f failed as $RUNAS_ID"
5475 local fid=$(path2fid $f)
5476 rm $f || error "rm $f failed"
5478 local mntpt=$(facet_mntpt mds1)
5482 #remount mds1 as ldiskfs or zfs type
5483 stop mds1 || error "stop mds1 failed"
5484 stack_trap "unmount_fstype mds1; start mds1 $(mdsdevname 1)\
5485 $MDS_MOUNT_OPTS" EXIT
5486 mount_fstype mds1 || error "remount mds1 failed"
5488 for ((i = 0; i < 1; i++)); do
5489 do_facet mds1 $llog_reader $mntpt/changelog_catalog
5490 local cat_file=$(do_facet mds1 $llog_reader \
5491 $mntpt/changelog_catalog | awk \
5492 '{match($0,"path=([^ ]+)",a)}END{print a[1]}')
5493 [ -n "$cat_file" ] || error "no catalog file"
5495 entry=$(do_facet mds1 $llog_reader $mntpt/$cat_file |
5496 awk "/CREAT/ && /target:\[$fid\]/ {print}")
5497 [ -n "$entry" ] || error "no CREAT entry"
5500 local uidgid=$(echo $entry |
5501 sed 's+.*\ user:\([0-9][0-9]*:[0-9][0-9]*\)\ .*+\1+')
5502 [ -n "$uidgid" ] || error "uidgid is empty"
5503 echo "Got UID/GID $uidgid"
5504 [ "$uidgid" = "$RUNAS_ID:$RUNAS_GID" ] ||
5505 error "uidgid '$uidgid' != '$RUNAS_ID:$RUNAS_GID'"
5506 local nid=$(echo $entry |
5507 sed 's+.*\ nid:\(\S\S*@\S\S*\)\ .*+\1+')
5508 [ -n "$nid" ] || error "nid is empty"
5510 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5511 error "nid '$nid' does not match any NID ${CLIENT_NIDS[@]}"
5513 run_test 606 "llog_reader groks changelog fields"
5516 check_and_cleanup_lustre