3 # Run select tests by setting ONLY, or as arguments to the script.
4 # Skip specific tests by setting EXCEPT.
12 LUSTRE=${LUSTRE:-$(dirname $0)/..}
13 . $LUSTRE/tests/test-framework.sh
17 ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT "
19 # bug number for skipped tests: LU-9795
20 ALWAYS_EXCEPT+=" 402b "
21 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
24 # Skip tests for PPC that fail frequently
25 if [[ $(uname -m) = ppc64 ]]; then
26 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
27 ALWAYS_EXCEPT+=" 1a 1b 1d 1e 12c 12f "
28 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
29 ALWAYS_EXCEPT+=" 12g 12h 12m 12n 12o 12p "
30 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
31 ALWAYS_EXCEPT+=" 12q 21 22 23 24a 24b "
32 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
33 ALWAYS_EXCEPT+=" 24d 24e 24f 25b 30c 37 "
34 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
35 ALWAYS_EXCEPT+=" 57 58 90 110b 111b 113 "
36 # bug number: LU-12251 LU-12251 LU-12251 LU-12251 LU-12251 LU-12251
37 ALWAYS_EXCEPT+=" 222b 222d 228 260a 260b 260c "
38 # bug number: LU-12252 LU-12252 LU-12252 LU-12252 LU-12252 LU-12252
39 ALWAYS_EXCEPT+=" 220A 220a 221 222a 222c 223a "
40 # bug number: LU-12252 LU-12252 LU-12252 LU-12252 LU-12252 LU-12252
41 ALWAYS_EXCEPT+=" 223b 224A 224a 226 227 600"
42 # bug number: LU-12252 LU-12252 LU-12252 LU-12252 LU-12252 LU-12252
43 ALWAYS_EXCEPT+=" 601 602 603 604 605 "
48 [ -n "$FILESET" ] && skip "Not functional for FILESET set"
50 OPENFILE=${OPENFILE:-openfile}
51 MOUNT_2=${MOUNT_2:-"yes"}
54 # script only handles up to 10 MDTs (because of MDT_PREFIX)
55 [ $MDSCOUNT -gt 9 ] &&
56 error "script cannot handle more than 9 MDTs, please fix"
58 check_and_setup_lustre
60 if [[ $MDS1_VERSION -lt $(version_code 2.4.53) ]]; then
61 skip_env "Need MDS version at least 2.4.53"
64 # $RUNAS_ID may get set incorrectly somewhere else
65 if [[ $UID -eq 0 && $RUNAS_ID -eq 0 ]]; then
66 skip_env "\$RUNAS_ID set to 0, but \$UID is also 0!"
68 check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS
69 if getent group nobody; then
71 elif getent group nogroup; then
74 error "No generic nobody group"
77 # if there is no CLIENT1 defined, some tests can be ran on localhost
78 CLIENT1=${CLIENT1:-$HOSTNAME}
79 # if CLIENT2 doesn't exist then use CLIENT1 instead
80 # All tests should use CLIENT2 with MOUNT2 only therefore it will work if
82 # Exception is the test which need two separate nodes
83 CLIENT2=${CLIENT2:-$CLIENT1}
86 local hosts=${1:-$(facet_active_host $SINGLEAGT)}
87 do_nodesv $hosts "pgrep --pidfile=$HSMTOOL_PID_FILE hsmtool"
91 local hosts=${1:-$(facet_active_host $SINGLEAGT)}
92 local wait_timeout=200
93 local wait_start=$SECONDS
94 local wait_end=$((wait_start + wait_timeout))
97 while ((SECONDS < wait_end)); do
98 if ! search_copytools $hosts; then
99 echo "copytools stopped in $((SECONDS - wait_start))s"
103 echo "copytools still running on $hosts"
105 [ $sleep_time -lt 5 ] && sleep_time=$((sleep_time + 1))
108 # try to dump Copytool's stack
109 do_nodesv $hosts "echo 1 >/proc/sys/kernel/sysrq ; " \
110 "echo t >/proc/sysrq-trigger"
112 echo "copytools failed to stop in ${wait_timeout}s"
117 copytool_monitor_setup() {
118 local facet=${1:-$SINGLEAGT}
119 local agent=$(facet_active_host $facet)
121 local cmd="mktemp --tmpdir=/tmp -d ${TESTSUITE}.${TESTNAME}.XXXX"
122 local test_dir=$(do_node $agent "$cmd") ||
123 error "Failed to create tempdir on $agent"
124 export HSMTOOL_MONITOR_DIR=$test_dir
126 # Create the fifo and a monitor (cat dies when copytool dies)
127 do_node $agent "mkfifo -m 0644 $test_dir/fifo" ||
128 error "failed to create copytool fifo on $agent"
129 cmd="cat $test_dir/fifo > $test_dir/events &"
130 cmd+=" echo \\\$! > $test_dir/monitor_pid"
132 # This background subshell nonsense is required when pdsh/ssh decides
133 # to wait for the cat process to exit on the remote client
134 (do_node $agent "$cmd") &
135 export HSMTOOL_MONITOR_PDSH=$!
137 # Slightly racy, but just making a best-effort to catch obvious
140 do_node $agent "stat $HSMTOOL_MONITOR_DIR/monitor_pid 2>&1 > /dev/null"
142 error "Failed to start copytool monitor on $agent"
150 case "$HSMTOOL_ARCHIVE_FORMAT" in
152 printf "%s" "$(hsm_root)/*/*/*/*/*/*/$fid"
155 printf "%s" "$(hsm_root)/*/$fid"
160 get_copytool_event_log() {
161 local facet=${1:-$SINGLEAGT}
162 local agent=$(facet_active_host $facet)
164 [ -z "$HSMTOOL_MONITOR_DIR" ] &&
165 error "Can't get event log: No monitor directory!"
167 do_node $agent "cat $HSMTOOL_MONITOR_DIR/events" ||
168 error "Could not collect event log from $agent"
172 local agents=${1:-$(facet_active_host $SINGLEAGT)}
174 stack_trap "pkill_copytools $agents CONT || true" EXIT
175 pkill_copytools $agents STOP || return 0
176 echo "Copytool is suspended on $agents"
179 copytool_remove_backend() {
181 local be=$(do_facet $SINGLEAGT find "$(hsm_root)" -name $fid)
182 echo "Remove from backend: $fid = $be"
183 do_facet $SINGLEAGT rm -f $be
186 file_creation_failure() {
193 df $MOUNT $MOUNT2 >&2
194 error "Not enough space to create $file with $cmd"
197 error "cannot create $file with $cmd, status=$err"
202 # Creates a file using dd
208 local source=${5:-/dev/zero}
212 if [ -n "$conv" ]; then
216 # Create the directory in case it does not exist
217 mkdir -p "$(dirname "$file")"
218 # Delete the file in case it already exist
221 if dd if="$source" of="$file" count="$count" bs="$bs" $args; then
222 path2fid "$file" || error "cannot get FID of '$file'"
225 echo "cannot create file '$file'" >&2;
226 # Let the caller decide what to do on error
231 create_empty_file() {
232 create_file "${1/$DIR/$DIR2}" 1M 0 ||
233 file_creation_failure dd "${1/$DIR/$DIR2}" $?
236 create_small_file() {
237 local source_file=/dev/urandom
240 local conv=${2:-fsync}
242 create_file "${1/$DIR/$DIR2}" $bs $count $conv $source_file ||
243 file_creation_failure dd "${1/$DIR/$DIR2}" $?
246 create_small_sync_file() {
247 create_small_file "$1" sync
250 create_archive_file() {
251 local file="$(hsm_root)/$1"
253 local source=/dev/urandom
255 # Create the counterpart directory of the archive
256 do_facet "$SINGLEAGT" mkdir -p "$(dirname "$file")" ||
257 error "cannot create archive directory '$(dirname "$file")'"
259 do_facet "$SINGLEAGT" dd if=$source of="$file" bs=1M count=$count ||
260 error "cannot create archive file '$file'"
264 local hsm_root="$(hsm_root)"
265 local file="$hsm_root/$2"
267 stack_trap "do_facet $SINGLEAGT rm -rf '$hsm_root'" EXIT
268 do_facet $SINGLEAGT mkdir -p "$(dirname "$file")" ||
269 error "mkdir '$(dirname "$file")' failed"
270 do_facet $SINGLEAGT cp -p "$1" "$file" ||
271 error "cannot copy '$1' to '$file'"
276 local val=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.$param)
283 mdts_set_param "" hsm_control "$cmd"
284 mdts_check_param hsm_control "$target" 10
289 mdts_set_param "" hsm.policy "+NRA"
290 CDT_POLICY_HAD_CHANGED=true
293 cdt_clear_no_retry() {
294 mdts_set_param "" hsm.policy "-NRA"
295 CDT_POLICY_HAD_CHANGED=true
298 cdt_set_non_blocking_restore() {
299 mdts_set_param "" hsm.policy "+NBR"
300 CDT_POLICY_HAD_CHANGED=true
303 cdt_clear_non_blocking_restore() {
304 mdts_set_param "" hsm.policy "-NBR"
305 CDT_POLICY_HAD_CHANGED=true
308 cdt_clear_mount_state() {
309 mdts_set_param "-P -d" hsm_control ""
313 set_test_state disabled disabled
317 set_test_state enabled enabled
321 set_test_state shutdown stopped
325 set_test_state purge enabled
331 cdt_set_sanity_policy
335 get_hsm_archive_id() {
338 st=$($LFS hsm_state $f)
339 [[ $? == 0 ]] || error "$LFS hsm_state $f failed"
341 local ar=$(echo $st | grep -oP '(?<=archive_id:).*')
345 check_hsm_flags_user() {
349 local st=$(get_hsm_flags $f user)
350 [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
362 if [[ "$3" != 1 ]]; then
366 cp $1 $f || file_creation_failure cp $f $?
368 path2fid $f || error "cannot get fid on $f"
371 # Delete any file bigger than 10M under $MOUNT and wait for deletes to complete
373 # Note that this might lead to surprising behaviours such as deleting an
374 # important file for the currently running test
375 delete_large_files() {
376 printf "Deleting large files...\n" >&2
377 find $MOUNT -size +10M -delete
378 wait_delete_completed
381 get_request_state() {
385 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
386 "awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
389 get_request_count() {
393 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
394 "awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
397 # Ensure the number of HSM request for a given FID is correct
398 # assert_request_count FID REQUEST_TYPE COUNT [ERROR_MSG]
399 assert_request_count() {
400 local request_count=$(get_request_count $1 $2)
401 local default_error_msg=("expected $3 '$2' request(s) for '$1', found "
403 [ $request_count -eq $3 ] || error "${4:-"${default_error_msg[@]}"}"
410 local cmd="$LCTL get_param -n $HSM_PARAM.actions"
411 [[ -n $fid ]] && cmd+=" | grep '$fid'"
412 cmd+=" | egrep 'WAITING|STARTED'"
414 wait_update_facet --verbose mds1 "$cmd" "" $timeout ||
415 error "requests did not complete"
418 wait_for_grace_delay() {
419 local val=$(get_hsm_param grace_delay)
423 wait_for_loop_period() {
424 local val=$(get_hsm_param loop_period)
431 # python2.6 in EL6 includes an internal json module
432 local PYTHON='python'
433 local json_parser='import json; import fileinput;'
434 json_parser+=' print("\n".join(["local %s=\"%s\"" % tuple for tuple in '
435 json_parser+='json.loads([line for line in '
436 json_parser+='fileinput.input()][0]).items()]))'
438 # check if python/python2/python3 is available
439 if ! which $PYTHON > /dev/null 2>&1 ; then
441 if ! which $PYTHON > /dev/null 2>&1 ; then
445 echo $raw_event | $PYTHON -c "$json_parser"
448 get_agent_by_uuid_mdt() {
451 local mds=mds$(($mdtidx + 1))
452 do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
456 check_agent_registered_by_mdt() {
459 local mds=mds$(($mdtidx + 1))
460 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
461 if [[ ! -z "$agent" ]]; then
462 echo "found agent $agent on $mds"
464 error "uuid $uuid not found in agent list on $mds"
468 check_agent_unregistered_by_mdt() {
471 local mds=mds$(($mdtidx + 1))
472 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
473 if [[ -z "$agent" ]]; then
474 echo "uuid not found in agent list on $mds"
476 error "uuid found in agent list on $mds: $agent"
480 check_agent_registered() {
483 for mdsno in $(seq 1 $MDSCOUNT); do
484 check_agent_registered_by_mdt $uuid $((mdsno - 1))
488 check_agent_unregistered() {
491 for mdsno in $(seq 1 $MDSCOUNT); do
492 check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
497 local agent=${1:-$(facet_active_host $SINGLEAGT)}
499 # Lustre mount-point is mandatory and last parameter on
501 local mntpnt=$(do_rpc_nodes $agent \
502 pgrep --pidfile=$HSMTOOL_PID_FILE --list-full hsmtool |
504 [ -n "$mntpnt" ] || error "Found no Agent or with no mount-point "\
506 do_rpc_nodes $agent get_client_uuid $mntpnt | cut -d' ' -f2
512 # populate MDT device array
515 # cleanup from previous bad setup
518 # for recovery tests, coordinator needs to be started at mount
520 # the lustre conf must be without hsm on (like for sanity.sh)
521 echo "Set HSM on and start"
522 cdt_set_mount_state enabled
523 cdt_check_state enabled
525 echo "Set sanity-hsm HSM policy"
526 cdt_set_sanity_policy
528 # finished requests are quickly removed from list
529 set_hsm_param grace_delay 10
531 CLIENT_NIDS=( $($LCTL list_nids all) )
533 test_1A() { # was test_1
537 local f=$DIR/$tdir/$tfile
541 check_hsm_flags_user $f "0x00000000"
543 $RUNAS $LFS hsm_set --norelease $f ||
544 error "user could not change hsm flags"
545 check_hsm_flags_user $f "0x00000010"
547 $RUNAS $LFS hsm_clear --norelease $f ||
548 error "user could not clear hsm flags"
549 check_hsm_flags_user $f "0x00000000"
551 # User could not change those flags...
552 $RUNAS $LFS hsm_set --exists $f &&
553 error "user should not set this flag"
554 check_hsm_flags_user $f "0x00000000"
557 $LFS hsm_set --exists $f ||
558 error "root could not change hsm flags"
559 check_hsm_flags_user $f "0x00000001"
561 $LFS hsm_clear --exists $f ||
562 error "root could not clear hsm state"
563 check_hsm_flags_user $f "0x00000000"
566 run_test 1A "lfs hsm flags root/non-root access"
569 mkdir_on_mdt0 $DIR/$tdir
571 local f=$DIR/$tdir/$tfile
572 local fid=$(create_small_file $f)
576 $LFS hsm_archive $f || error "could not archive file"
577 wait_request_state $fid ARCHIVE SUCCEED
579 # Release and check states
580 $LFS hsm_release $f || error "could not release file"
581 echo -n "Verifying released state: "
582 check_hsm_flags $f "0x0000000d"
584 $MMAP_CAT $f > /dev/null || error "failed mmap & cat release file"
586 run_test 1a "mmap & cat a HSM released file"
592 dd if=/dev/urandom of=$f bs=1M count=1 conv=sync ||
593 error "failed to create file"
594 local fid=$(path2fid $f)
599 $LFS hsm_archive $f || error "could not archive file"
600 wait_request_state $fid ARCHIVE SUCCEED
603 $LFS hsm_release $f || error "could not release file"
604 echo "verify released state: "
605 check_hsm_flags $f "0x0000000d" && echo "pass"
608 $LFS hsm_restore $f || error "could not restore file"
609 wait_request_state $fid RESTORE SUCCEED
610 echo "verify restored state: "
611 check_hsm_flags $f "0x00000009" && echo "pass"
615 mkdir_on_mdt0 $DIR/$tdir
616 $LFS setstripe -E 1M -S 1M -E 64M -c 2 -E -1 -c 4 $DIR/$tdir ||
617 error "failed to set default stripe"
618 local f=$DIR/$tdir/$tfile
622 run_test 1b "Archive, Release and Restore composite file"
628 local f=$DIR/$tdir/$tfile
631 # Test whether we can set the maximum archive number.
632 local LOCAL_HSM_ARCHIVE_NUMBER=32
633 $LFS hsm_set --exists --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f ||
634 error "root could not change hsm flags"
635 check_hsm_flags_user $f "0x00000001"
636 echo "verifying archive number is $LOCAL_HSM_ARCHIVE_NUMBER"
637 local st=$(get_hsm_archive_id $f)
638 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
639 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
641 # Test whether setting archive number 0 results in no change.
642 $LFS hsm_set --exists --archive-id 0 $f ||
643 error "root could not change hsm flags"
644 check_hsm_flags_user $f "0x00000001"
645 echo "verifying archive number is still $LOCAL_HSM_ARCHIVE_NUMBER"
646 st=$(get_hsm_archive_id $f)
647 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
648 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
650 LOCAL_HSM_ARCHIVE_NUMBER=33
651 if [ "$CLIENT_VERSION" -ge $(version_code 2.11.56) ] &&
652 [ "$MDS1_VERSION" -ge $(version_code 2.11.56) ]; then
653 # lustre in the new version supports unlimited archiveID.
654 # Test whether setting archive number > 32 is supported
655 $LFS hsm_set --exists --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f ||
656 error "archive ID $LOCAL_HSM_ARCHIVE_NUMBER too large?"
657 check_hsm_flags_user $f "0x00000001"
659 echo "verifying archive number is $LOCAL_HSM_ARCHIVE_NUMBER"
660 st=$(get_hsm_archive_id $f)
661 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
662 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
664 # old client or old mds can only support at most 32 archiveID
665 # test whether setting archive number > 32 results in error.
666 $LFS hsm_set --exists --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f &&
667 error "bitmap archive number is larger than 32"
668 check_hsm_flags_user $f "0x00000001"
671 # Test whether setting archive number 16 and archived flag.
672 LOCAL_HSM_ARCHIVE_NUMBER=16
673 $LFS hsm_set --exists --archived \
674 --archive-id $LOCAL_HSM_ARCHIVE_NUMBER $f ||
675 error "root could not change hsm flags"
676 check_hsm_flags_user $f "0x00000009"
677 echo "verifying archive number is $LOCAL_HSM_ARCHIVE_NUMBER"
678 st=$(get_hsm_archive_id $f)
679 [[ $st == $LOCAL_HSM_ARCHIVE_NUMBER ]] ||
680 error "wrong archive number, $st != $LOCAL_HSM_ARCHIVE_NUMBER"
682 run_test 1c "Check setting archive-id in lfs hsm_set"
685 [ $MDS1_VERSION -lt $(version_code 2.10.59) ] &&
686 skip "need MDS version at least 2.10.59"
688 mkdir_on_mdt0 $DIR/$tdir
689 $LFS setstripe -E 1M -L mdt -E -1 -c 2 $DIR/$tdir ||
690 error "failed to set default stripe"
691 local f=$DIR/$tdir/$tfile
695 run_test 1d "Archive, Release and Restore DoM file"
698 [ "$MDS1_VERSION" -lt $(version_code $SEL_VER) ] &&
699 skip "skipped for lustre < $SEL_VER"
701 mkdir_on_mdt0 $DIR/$tdir
702 $LFS setstripe -E 1G -z 64M -E 10G -z 512M -E -1 -z 1G $DIR/$tdir ||
703 error "failed to set default stripe"
704 local comp_file=$DIR/$tdir/$tfile
706 test_1bde_base $comp_file
708 local flg_opts="--comp-start 0 -E 64M --comp-flags init"
709 local found=$($LFS find $flg_opts $comp_file | wc -l)
710 [ $found -eq 1 ] || error "1st component not found"
712 flg_opts="--comp-start 64M -E 1G --comp-flags extension"
713 found=$($LFS find $flg_opts $comp_file | wc -l)
714 [ $found -eq 1 ] || error "2nd component not found"
716 flg_opts="--comp-start 1G -E 1G --comp-flags ^init"
717 found=$($LFS find $flg_opts $comp_file | wc -l)
718 [ $found -eq 1 ] || error "3rd component not found"
720 flg_opts="--comp-start 1G -E 10G --comp-flags extension"
721 found=$($LFS find $flg_opts $comp_file | wc -l)
722 [ $found -eq 1 ] || error "4th component not found"
724 flg_opts="--comp-start 10G -E 10G --comp-flags ^init"
725 found=$($LFS find $flg_opts $comp_file | wc -l)
726 [ $found -eq 1 ] || error "5th component not found"
728 flg_opts="--comp-start 10G -E EOF --comp-flags extension"
729 found=$($LFS find $flg_opts $comp_file | wc -l)
730 [ $found -eq 1 ] || error "6th component not found"
732 sel_layout_sanity $comp_file 6
734 run_test 1e "Archive, Release and Restore SEL file"
737 (( $MDS1_VERSION >= $(version_code 2.15.55.203) )) ||
738 skip "need MDS version at least 2.15.55.203"
739 local dom=$DIR/$tdir/$tfile
741 mkdir_on_mdt0 $DIR/$tdir
742 $LFS setstripe -E 512K -L mdt -E -1 -c 2 $DIR/$tdir ||
743 error "failed to set default stripe"
747 [[ $($LFS getstripe --component-start=0 -L $dom) == 'mdt' ]] ||
748 error "MDT stripe isn't set"
750 # check that metadata change doesn't prevent second release
751 chmod 600 $dom || error "chmod failed"
753 echo "release again $dom"
754 $LFS hsm_release $dom || error "second release failed"
756 echo "verify released state: "
757 check_hsm_flags $dom "0x0000000d" && echo "pass"
759 run_test 1f "DoM file release after restore"
762 local f=$DIR/$tdir/$tfile
764 mkdir_on_mdt0 $DIR/$tdir
765 create_empty_file "$f"
766 # New files are not dirty
767 check_hsm_flags $f "0x00000000"
769 # For test, we simulate an archived file.
770 $LFS hsm_set --exists $f || error "user could not change hsm flags"
771 check_hsm_flags $f "0x00000001"
773 # chmod do not put the file dirty
774 chmod 600 $f || error "could not chmod test file"
775 check_hsm_flags $f "0x00000001"
777 # chown do not put the file dirty
778 chown $RUNAS_ID $f || error "could not chown test file"
779 check_hsm_flags $f "0x00000001"
781 # truncate put the file dirty
782 $TRUNCATE $f 1 || error "could not truncate test file"
783 check_hsm_flags $f "0x00000003"
785 $LFS hsm_clear --dirty $f || error "could not clear hsm flags"
786 check_hsm_flags $f "0x00000001"
788 run_test 2 "Check file dirtyness when doing setattr"
791 mkdir_on_mdt0 $DIR/$tdir
794 # New files are not dirty
796 check_hsm_flags $f "0x00000000"
798 # For test, we simulate an archived file.
799 $LFS hsm_set --exists $f ||
800 error "user could not change hsm flags"
801 check_hsm_flags $f "0x00000001"
803 # Reading a file, does not set dirty
804 cat $f > /dev/null || error "could not read file"
805 check_hsm_flags $f "0x00000001"
807 # Open for write without modifying data, does not set dirty
808 openfile -f O_WRONLY $f || error "could not open test file"
809 check_hsm_flags $f "0x00000001"
811 # Append to a file sets it dirty
812 cp -p /etc/passwd $f.append || error "could not create file"
813 $LFS hsm_set --exists $f.append ||
814 error "user could not change hsm flags"
815 dd if=/etc/passwd of=$f.append bs=1 count=3\
816 conv=notrunc oflag=append status=noxfer ||
817 file_creation_failure dd $f.append $?
818 check_hsm_flags $f.append "0x00000003"
820 # Modify a file sets it dirty
821 cp -p /etc/passwd $f.modify || error "could not create file"
822 $LFS hsm_set --exists $f.modify ||
823 error "user could not change hsm flags"
824 dd if=/dev/zero of=$f.modify bs=1 count=3\
825 conv=notrunc status=noxfer ||
826 file_creation_failure dd $f.modify $?
827 check_hsm_flags $f.modify "0x00000003"
829 # Open O_TRUNC sets dirty
830 cp -p /etc/passwd $f.trunc || error "could not create file"
831 $LFS hsm_set --exists $f.trunc ||
832 error "user could not change hsm flags"
833 cp /etc/group $f.trunc || error "could not override a file"
834 check_hsm_flags $f.trunc "0x00000003"
836 # Mmapped a file sets dirty
837 cp -p /etc/passwd $f.mmap || error "could not create file"
838 $LFS hsm_set --exists $f.mmap ||
839 error "user could not change hsm flags"
840 multiop $f.mmap OSMWUc || error "could not mmap a file"
841 check_hsm_flags $f.mmap "0x00000003"
843 run_test 3 "Check file dirtyness when opening for write"
846 local f=$DIR/$tdir/$tfile
847 local fid=$(create_small_file $f)
850 local st=$(get_request_state $fid CANCEL)
851 [[ -z "$st" ]] || error "hsm_cancel must not be registered (state=$st)"
853 run_test 4 "Useless cancel must not be registered"
856 # test needs a running copytool
859 mkdir_on_mdt0 $DIR/$tdir
860 local f=$DIR/$tdir/$tfile
861 local fid=$(copy_file /etc/passwd $f)
863 wait_request_state $fid ARCHIVE SUCCEED
865 check_hsm_flags $f "0x00000009"
867 run_test 8 "Test default archive number"
869 test_9A() { # was test_9
870 # we do not use the default one to be sure
871 local archive_id=$((HSM_ARCHIVE_NUMBER + 1))
872 copytool setup --archive-id $archive_id
874 # give time for CT to register with MDTs
875 sleep $(($MDSCOUNT*2))
876 local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
877 check_agent_registered $uuid
879 mkdir_on_mdt0 $DIR/$tdir
881 local f=$DIR/$tdir/$tfile
882 local fid=$(copy_file /etc/passwd $f)
883 $LFS hsm_archive --archive $archive_id $f
884 wait_request_state $fid ARCHIVE SUCCEED
886 check_hsm_flags $f "0x00000009"
888 run_test 9A "Use of explicit archive number, with dedicated copytool"
891 needclients 3 || return 0
897 # start all of the copytools
898 for n in $(seq $AGTCOUNT); do
899 copytool setup --facet agt$n
902 mkdir_on_mdt0 $DIR/$tdir
905 for n in $(seq $AGTCOUNT); do
906 file=$DIR/$tdir/$tfile.$n
907 fid=$(create_small_file $file)
909 $LFS hsm_archive $file || error "could not archive file $file"
910 wait_request_state $fid ARCHIVE SUCCEED
911 check_hsm_flags $file "0x00000009"
914 run_test 9a "Multiple remote agents"
917 # test needs a running copytool
920 mkdir_on_mdt0 $DIR/$tdir
921 mkdir -p $DIR/$tdir/d1
922 local f=$DIR/$tdir/$tfile
923 local fid=$(copy_file /etc/hosts $f)
924 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
925 error "hsm_archive failed"
926 wait_request_state $fid ARCHIVE SUCCEED
928 local hsm_root="$(copytool_device $SINGLEAGT)"
929 local archive="$(do_facet $SINGLEAGT \
930 find "$hsm_root" -name "$fid" -print0)"
931 [ -n "$archive" ] || error "fid '$fid' not in archive '$hsm_root'"
933 echo "Verifying content"
934 do_facet $SINGLEAGT diff $f $archive || error "archived file differs"
935 echo "Verifying hsm state "
936 check_hsm_flags $f "0x00000009"
938 echo "Verifying archive number is $HSM_ARCHIVE_NUMBER"
939 local st=$(get_hsm_archive_id $f)
940 [[ $st == $HSM_ARCHIVE_NUMBER ]] ||
941 error "Wrong archive number, $st != $HSM_ARCHIVE_NUMBER"
943 run_test 10a "Archive a file"
946 # test needs a running copytool
949 mkdir_on_mdt0 $DIR/$tdir
950 local f=$DIR/$tdir/$tfile
951 local fid=$(copy_file /etc/hosts $f)
952 $LFS hsm_archive $f || error "archive request failed"
953 wait_request_state $fid ARCHIVE SUCCEED
955 $LFS hsm_archive $f || error "archive of non dirty file failed"
956 local cnt=$(get_request_count $fid ARCHIVE)
957 [[ "$cnt" == "1" ]] ||
958 error "archive of non dirty file must not make a request"
960 run_test 10b "Archive of non dirty file must work without doing request"
963 # test needs a running copytool
967 local f=$DIR/$tdir/$tfile
968 local fid=$(copy_file /etc/hosts $f)
969 $LFS hsm_set --noarchive $f
970 $LFS hsm_archive $f && error "archive a noarchive file must fail"
973 run_test 10c "Check forbidden archive"
976 # test needs a running copytool
979 mkdir_on_mdt0 $DIR/$tdir
980 local f=$DIR/$tdir/$tfile
981 local fid=$(copy_file /etc/hosts $f)
982 $LFS hsm_archive $f || error "cannot archive $f"
983 wait_request_state $fid ARCHIVE SUCCEED
985 local ar=$(get_hsm_archive_id $f)
986 local dflt=$(get_hsm_param default_archive_id)
987 [[ $ar == $dflt ]] ||
988 error "archived file is not on default archive: $ar != $dflt"
990 run_test 10d "Archive a file on the default archive id"
993 mkdir_on_mdt0 $DIR/$tdir
994 copy2archive /etc/hosts $tdir/$tfile
995 local f=$DIR/$tdir/$tfile
997 copytool import $tdir/$tfile $f
998 echo -n "Verifying released state: "
999 check_hsm_flags $f "0x0000000d"
1001 local LSZ=$(stat -c "%s" $f)
1002 local ASZ=$(do_facet $SINGLEAGT stat -c "%s" "$(hsm_root)/$tdir/$tfile")
1004 echo "Verifying imported size $LSZ=$ASZ"
1005 [[ $LSZ -eq $ASZ ]] || error "Incorrect size $LSZ != $ASZ"
1006 echo -n "Verifying released pattern: "
1007 local PTRN=$($LFS getstripe -L $f)
1009 [[ $PTRN == released ]] || error "Is not released"
1010 local fid=$(path2fid $f)
1011 echo "Verifying new fid $fid in archive"
1013 do_facet $SINGLEAGT "[ -f \"$(fid2archive "$fid")\" ]" ||
1014 error "No archive for fid $fid"
1016 run_test 11a "Import a file"
1019 # test needs a running copytool
1022 mkdir_on_mdt0 $DIR/$tdir
1023 local f=$DIR/$tdir/$tfile
1024 local fid=$(copy_file /etc/hosts $f)
1025 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
1026 error "hsm_archive failed"
1027 wait_request_state $fid ARCHIVE SUCCEED
1029 local FILE_HASH=$(md5sum $f)
1032 copytool import $fid $f
1034 echo "$FILE_HASH" | md5sum -c
1036 [[ $? -eq 0 ]] || error "Restored file differs"
1038 run_test 11b "Import a deleted file using its FID"
1041 pool_add $TESTNAME || error "Pool creation failed"
1042 pool_add_targets $TESTNAME 1 1 || error "pool_add_targets failed"
1045 $LFS setstripe -p "$TESTNAME" $DIR/$tdir
1047 copy2archive /etc/hosts $tdir/$tfile
1048 copytool import $tdir/$tfile $DIR/$tdir/$tfile
1050 run_test 11c "Import a file to a directory with a pool"
1053 # test needs a running copytool
1056 mkdir_on_mdt0 $DIR/$tdir
1057 copy2archive /etc/hosts $tdir/$tfile
1059 local f=$DIR/$tdir/$tfile
1060 copytool import $tdir/$tfile $f
1061 local f2=$DIR2/$tdir/$tfile
1062 echo "Verifying released state: "
1063 check_hsm_flags $f2 "0x0000000d"
1065 local fid=$(path2fid $f2)
1066 $LFS hsm_restore $f2
1067 wait_request_state $fid RESTORE SUCCEED
1069 echo "Verifying file state: "
1070 check_hsm_flags $f2 "0x00000009"
1072 do_facet $SINGLEAGT diff -q $(hsm_root)/$tdir/$tfile $f
1074 [[ $? -eq 0 ]] || error "Restored file differs"
1076 run_test 12a "Restore an imported file explicitly"
1079 # test needs a running copytool
1082 mkdir_on_mdt0 $DIR/$tdir
1083 copy2archive /etc/hosts $tdir/$tfile
1085 local f=$DIR/$tdir/$tfile
1086 copytool import $tdir/$tfile $f
1087 echo "Verifying released state: "
1088 check_hsm_flags $f "0x0000000d"
1090 cat $f > /dev/null || error "File read failed"
1092 echo "Verifying file state after restore: "
1093 check_hsm_flags $f "0x00000009"
1095 do_facet $SINGLEAGT diff -q $(hsm_root)/$tdir/$tfile $f
1097 [[ $? -eq 0 ]] || error "Restored file differs"
1099 run_test 12b "Restore an imported file implicitly"
1102 [ "$OSTCOUNT" -lt "2" ] && skip_env "needs >= 2 OSTs" && return
1104 # test needs a running copytool
1107 local f=$DIR/$tdir/$tfile
1108 mkdir_on_mdt0 $DIR/$tdir
1109 $LFS setstripe -c 2 "$f"
1110 local fid=$(create_file "$f" 1M 5)
1112 local FILE_CRC=$(md5sum $f)
1114 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1115 wait_request_state $fid ARCHIVE SUCCEED
1116 $LFS hsm_release $f || error "release $f failed"
1118 echo "$FILE_CRC" | md5sum -c
1120 [[ $? -eq 0 ]] || error "Restored file differs"
1122 run_test 12c "Restore a file with stripe of 2"
1125 # test needs a running copytool
1128 mkdir_on_mdt0 $DIR/$tdir
1130 local f=$DIR/$tdir/$tfile
1131 local fid=$(copy_file /etc/hosts $f)
1132 $LFS hsm_restore $f || error "restore of non archived file failed"
1133 local cnt=$(get_request_count $fid RESTORE)
1134 [[ "$cnt" == "0" ]] ||
1135 error "restore non archived must not make a request"
1136 $LFS hsm_archive $f ||
1137 error "archive request failed"
1138 wait_request_state $fid ARCHIVE SUCCEED
1139 $LFS hsm_restore $f ||
1140 error "restore of non released file failed"
1141 local cnt=$(get_request_count $fid RESTORE)
1142 [[ "$cnt" == "0" ]] ||
1143 error "restore a non dirty file must not make a request"
1145 run_test 12d "Restore of a non archived, non released file must work"\
1146 " without doing request"
1149 # test needs a running copytool
1152 mkdir_on_mdt0 $DIR/$tdir
1153 local f=$DIR/$tdir/$tfile
1154 local fid=$(copy_file /etc/hosts $f)
1155 $LFS hsm_archive $f || error "archive request failed"
1156 wait_request_state $fid ARCHIVE SUCCEED
1159 cat /etc/hosts >> $f
1163 $LFS hsm_restore $f && error "restore a dirty file must fail"
1166 run_test 12e "Check forbidden restore"
1169 # test needs a running copytool
1172 mkdir_on_mdt0 $DIR/$tdir
1173 local f=$DIR/$tdir/$tfile
1174 local fid=$(copy_file /etc/hosts $f)
1176 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1177 wait_request_state $fid ARCHIVE SUCCEED
1178 $LFS hsm_release $f || error "release of $f failed"
1180 wait_request_state $fid RESTORE SUCCEED
1182 echo -n "Verifying file state: "
1183 check_hsm_flags $f "0x00000009"
1185 diff -q /etc/hosts $f
1187 [[ $? -eq 0 ]] || error "Restored file differs"
1189 run_test 12f "Restore a released file explicitly"
1192 # test needs a running copytool
1195 mkdir_on_mdt0 $DIR/$tdir
1196 local f=$DIR/$tdir/$tfile
1197 local fid=$(copy_file /etc/hosts $f)
1199 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1200 wait_request_state $fid ARCHIVE SUCCEED
1201 $LFS hsm_release $f || error "release of $f failed"
1203 diff -q /etc/hosts $f
1206 # we check we had a restore done
1207 wait_request_state $fid RESTORE SUCCEED
1209 [[ $st -eq 0 ]] || error "Restored file differs"
1211 run_test 12g "Restore a released file implicitly"
1214 needclients 2 || return 0
1216 # test needs a running copytool
1219 mkdir_on_mdt0 $DIR/$tdir
1220 local f=$DIR/$tdir/$tfile
1221 local fid=$(copy_file /etc/hosts $f)
1223 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1224 wait_request_state $fid ARCHIVE SUCCEED
1225 $LFS hsm_release $f || error "release of $f failed"
1227 do_node $CLIENT2 diff -q /etc/hosts $f
1230 # we check we had a restore done
1231 wait_request_state $fid RESTORE SUCCEED
1233 [[ $st -eq 0 ]] || error "Restored file differs"
1235 run_test 12h "Restore a released file implicitly from a second node"
1238 # test needs a running copytool
1241 mkdir_on_mdt0 $DIR/$tdir
1242 local f=$DIR/$tdir/$tfile
1243 local fid=$(copy_file /etc/passwd $f)
1244 $LFS hsm_archive $f || error "archive of $f failed"
1245 wait_request_state $fid ARCHIVE SUCCEED
1247 $LFS hsm_release $f || error "release of $f failed"
1251 [[ $? -eq 0 ]] || error "Restored file differs"
1253 run_test 12m "Archive/release/implicit restore"
1256 # test needs a running copytool
1260 copy2archive /etc/hosts $tdir/$tfile
1262 local f=$DIR/$tdir/$tfile
1263 copytool import $tdir/$tfile $f
1265 do_facet $SINGLEAGT cmp /etc/hosts $f ||
1266 error "Restored file differs"
1268 $LFS hsm_release $f || error "release of $f failed"
1270 run_test 12n "Import/implicit restore/release"
1273 # test needs a running copytool
1276 mkdir_on_mdt0 $DIR/$tdir
1277 local f=$DIR/$tdir/$tfile
1278 local fid=$(copy_file /etc/hosts $f)
1280 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1281 wait_request_state $fid ARCHIVE SUCCEED
1282 $LFS hsm_release $f || error "release of $f failed"
1284 #define OBD_FAIL_MDS_HSM_SWAP_LAYOUTS 0x152
1285 do_facet $SINGLEMDS lctl set_param fail_loc=0x152
1287 # set no retry action mode
1290 diff -q /etc/hosts $f
1293 # we check we had a restore failure
1294 wait_request_state $fid RESTORE FAILED
1296 [[ $st -eq 0 ]] && error "Restore must fail"
1298 # remove no retry action mode
1301 # check file is still released
1302 check_hsm_flags $f "0x0000000d"
1304 # retry w/o failure injection
1305 do_facet $SINGLEMDS lctl set_param fail_loc=0
1307 # to be sure previous RESTORE result is gone
1309 wait_for_grace_delay
1311 diff -q /etc/hosts $f
1314 # we check we had a restore done
1315 wait_request_state $fid RESTORE SUCCEED
1317 [[ $st -eq 0 ]] || error "Restored file differs"
1319 run_test 12o "Layout-swap failure during Restore leaves file released"
1322 # test needs a running copytool
1325 mkdir_on_mdt0 $DIR/$tdir
1326 local f=$DIR/$tdir/$tfile
1327 local fid=$(copy_file /etc/hosts $f)
1329 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1330 wait_request_state $fid ARCHIVE SUCCEED
1331 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1332 $LFS hsm_release $f || error "cannot release $f"
1333 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1334 $LFS hsm_release $f || error "cannot release $f"
1335 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1337 run_test 12p "implicit restore of a file on copytool mount point"
1340 [ $MDS1_VERSION -lt $(version_code 2.7.58) ] &&
1341 skip "need MDS version at least 2.7.58"
1343 stack_trap "zconf_umount \"$(facet_host $SINGLEAGT)\" \"$MOUNT3\"" EXIT
1344 zconf_mount $(facet_host $SINGLEAGT) $MOUNT3 ||
1345 error "cannot mount $MOUNT3 on $SINGLEAGT"
1347 # test needs a running copytool
1348 copytool setup -m "$MOUNT3"
1350 mkdir_on_mdt0 $DIR/$tdir
1352 local f=$DIR/$tdir/$tfile
1353 local f2=$DIR2/$tdir/$tfile
1354 local fid=$(create_small_file $f)
1355 local orig_size=$(stat -c "%s" $f)
1357 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1358 wait_request_state $fid ARCHIVE SUCCEED
1360 $LFS hsm_release $f || error "could not release file"
1361 check_hsm_flags $f "0x0000000d"
1364 wait_copytools || error "copytool failed to stop"
1366 cat $f > /dev/null &
1368 # wait a bit to allow implicit restore request to be handled.
1369 # if not, next stat would also block on layout-lock.
1372 local size=$(stat -c "%s" $f2)
1373 [ $size -eq $orig_size ] ||
1374 error "$f2: wrong size after archive: $size != $orig_size"
1376 copytool setup -m "$MOUNT3"
1380 size=$(stat -c "%s" $f)
1381 [ $size -eq $orig_size ] ||
1382 error "$f: wrong size after restore: $size != $orig_size"
1384 size=$(stat -c "%s" $f2)
1385 [ $size -eq $orig_size ] ||
1386 error "$f2: wrong size after restore: $size != $orig_size"
1390 size=$(stat -c "%s" $f)
1392 error "$f: wrong size after overwrite: $size != 0"
1394 size=$(stat -c "%s" $f2)
1396 error "$f2: wrong size after overwrite: $size != 0"
1398 run_test 12q "file attributes are refreshed after restore"
1401 # test needs a running copytool
1404 mkdir_on_mdt0 $DIR/$tdir
1405 local f=$DIR/$tdir/$tfile
1406 local fid=$(copy_file /etc/hosts $f)
1408 $LFS hsm_archive $f || error "archive of $f failed"
1409 wait_request_state $fid ARCHIVE SUCCEED
1410 $LFS hsm_release $f || error "release of $f failed"
1412 offset=$(lseek_test -d 7 $f)
1414 # we check we had a restore done
1415 wait_request_state $fid RESTORE SUCCEED
1416 [[ $offset == 7 ]] || error "offset $offset != 7"
1418 run_test 12r "lseek restores released file"
1421 local f=$DIR/$tdir/$tfile
1425 (( MDS1_VERSION >= $(version_code 2.15.50) )) ||
1426 skip "Need MDS version newer than 2.15.50"
1428 # test needs a running copytool
1431 mkdir_on_mdt0 $DIR/$tdir
1432 fid=$(copy_file /etc/hosts $f)
1434 $LFS hsm_archive $f || error "archive of $f failed"
1435 wait_request_state $fid ARCHIVE SUCCEED
1436 $LFS hsm_release $f || error "release of $f failed"
1438 #define CFS_FAIL_ONCE|OBD_FAIL_MDS_HSM_RESTORE_RACE 0x8000018b
1439 do_facet mds1 $LCTL set_param fail_loc=0x8000018b
1440 cat $f > /dev/null & pid1=$!
1441 cat $f > /dev/null & pid2=$!
1443 wait $pid1 || error "cat process 1 fail (pid: $pid1)"
1444 wait $pid2 || error "cat process 2 fail (pid: $pid2)"
1446 # Race exists if more than 1 restore requests is registered
1447 assert_request_count $fid RESTORE 1
1449 run_test 12s "race between restore requests"
1453 for i in {1..10}; do
1454 local archive_dir="$(hsm_root)"/subdir/dir.$i
1456 do_facet $SINGLEAGT mkdir -p "$archive_dir"
1457 for j in {1..10}; do
1458 local archive_file="$archive_dir"/file.$j
1460 do_facet $SINGLEAGT "echo $k > \"$archive_dir\"/file.$j"
1466 copytool import "subdir" "$DIR/$tdir"
1468 # To check the import, the test uses diff with the -r flag
1469 # This is nice, but diff only checks files one by one, and triggering
1470 # an implicit restore for one file at a time will consume as many
1471 # seconds as there are files to compare. To speed this up, a restore
1472 # operation is triggered manually first.
1474 find "$DIR/$tdir"/subdir -type f -exec $LFS hsm_restore {} \;
1476 # Compare the imported data
1477 do_facet $SINGLEAGT \
1478 diff -r "$(hsm_root)"/subdir "$DIR/$tdir"/subdir ||
1479 error "imported files differ from archived data"
1481 run_test 13 "Recursively import and restore a directory"
1484 # test needs a running copytool
1487 mkdir_on_mdt0 $DIR/$tdir
1490 local f=$DIR/$tdir/$tfile
1491 local fid=$(create_small_file $f)
1492 local sum=$(md5sum $f | awk '{print $1}')
1493 $LFS hsm_archive $f || error "could not archive file"
1494 wait_request_state $fid ARCHIVE SUCCEED
1496 # create released file (simulate llapi_hsm_import call)
1497 local fid2=$(create_empty_file "$f")
1498 $LFS hsm_set --archived --exists $f || error "could not force hsm flags"
1499 $LFS hsm_release $f || error "could not release file"
1501 # rebind the archive to the newly created file
1502 echo "rebind $fid to $fid2"
1504 copytool rebind $fid $fid2
1506 # restore file and compare md5sum
1507 local sum2=$(md5sum $f | awk '{print $1}')
1509 [[ $sum == $sum2 ]] || error "md5sum mismatch after restore"
1511 run_test 14 "Rebind archived file to a new fid"
1514 # test needs a running copytool
1517 mkdir_on_mdt0 $DIR/$tdir
1520 local f=$DIR/$tdir/$tfile
1522 local tmpfile=$SHARED_DIRECTORY/tmp.$$
1526 for i in $(seq 1 $count); do
1527 fids[$i]=$(create_small_file $f.$i)
1528 sums[$i]=$(md5sum $f.$i | awk '{print $1}')
1529 $LFS hsm_archive $f.$i || error "could not archive file"
1531 wait_all_done $(($count*60))
1533 stack_trap "rm -f $tmpfile" EXIT
1536 for i in $(seq 1 $count); do
1537 local fid2=$(create_empty_file "${f}.${i}")
1538 # add the rebind operation to the list
1539 echo ${fids[$i]} $fid2 >> $tmpfile
1541 # set it released (simulate llapi_hsm_import call)
1542 $LFS hsm_set --archived --exists $f.$i ||
1543 error "could not force hsm flags"
1544 $LFS hsm_release $f.$i || error "could not release file"
1546 nl=$(wc -l < $tmpfile)
1547 [[ $nl == $count ]] || error "$nl files in list, $count expected"
1549 echo "rebind list of files"
1550 copytool rebind "$tmpfile"
1552 # restore files and compare md5sum
1553 for i in $(seq 1 $count); do
1554 local sum2=$(md5sum $f.$i | awk '{print $1}')
1555 [[ $sum2 == ${sums[$i]} ]] ||
1556 error "md5sum mismatch after restore ($sum2 != ${sums[$i]})"
1559 run_test 15 "Rebind a list of files"
1562 # test needs a running copytool
1566 # create a known size file so we can verify transfer speed
1569 dd if=/dev/zero of=$ref bs=1M count=20
1571 mkdir_on_mdt0 $DIR/$tdir
1572 local f=$DIR/$tdir/$tfile
1573 local fid=$(copy_file $ref $f)
1575 local start=$(date +%s)
1577 wait_request_state $fid ARCHIVE SUCCEED
1578 local end=$(date +%s)
1579 # Add 1 to account for rounding errors between start and end (LU-8155)
1580 local duration=$((end - start + 1))
1582 [[ $duration -ge $((goal - 1)) ]] ||
1583 error "Transfer is too fast $duration < $goal"
1585 run_test 16 "Test CT bandwith control option"
1588 local f=$DIR/$tdir/$tfile
1589 create_empty_file "$f"
1591 # Could not release a non-archived file
1592 $LFS hsm_release $f && error "release should not succeed"
1594 # For following tests, we must test them with HS_ARCHIVED set
1595 $LFS hsm_set --exists --archived $f || error "could not add flag"
1597 # Could not release a file if no-release is set
1598 $LFS hsm_set --norelease $f || error "could not add flag"
1599 $LFS hsm_release $f && error "release should not succeed"
1600 $LFS hsm_clear --norelease $f || error "could not remove flag"
1602 # Could not release a file if lost
1603 $LFS hsm_set --lost $f || error "could not add flag"
1604 $LFS hsm_release $f && error "release should not succeed"
1605 $LFS hsm_clear --lost $f || error "could not remove flag"
1607 # Could not release a file if dirty
1608 $LFS hsm_set --dirty $f || error "could not add flag"
1609 $LFS hsm_release $f && error "release should not succeed"
1610 $LFS hsm_clear --dirty $f || error "could not remove flag"
1612 run_test 20 "Release is not permitted"
1615 # test needs a running copytool
1617 mkdir_on_mdt0 $DIR/$tdir
1619 local f=$DIR/$tdir/test_release
1621 # Create a file and check its states
1622 local fid=$(create_small_file $f)
1623 check_hsm_flags $f "0x00000000"
1625 # LU-4388/LU-4389 - ZFS does not report full number of blocks
1626 # used until file is flushed to disk
1627 if [ "$ost1_FSTYPE" == "zfs" ]; then
1628 # this causes an OST_SYNC rpc to be sent
1629 dd if=/dev/zero of=$f bs=512 count=1 oflag=sync conv=notrunc,fsync
1630 # clear locks to reread file data
1631 cancel_lru_locks osc
1634 local orig_size=$(stat -c "%s" $f)
1635 local orig_blocks=$(stat -c "%b" $f)
1637 start_full_debug_logging
1639 $LFS hsm_archive $f || error "could not archive file"
1640 wait_request_state $fid ARCHIVE SUCCEED
1642 local blocks=$(stat -c "%b" $f)
1643 [ $blocks -eq $orig_blocks ] ||
1644 error "$f: wrong block number after archive: " \
1645 "$blocks != $orig_blocks"
1646 local size=$(stat -c "%s" $f)
1647 [ $size -eq $orig_size ] ||
1648 error "$f: wrong size after archive: $size != $orig_size"
1650 # Release and check states
1651 $LFS hsm_release $f || error "could not release file"
1652 check_hsm_flags $f "0x0000000d"
1654 blocks=$(stat -c "%b" $f)
1655 [ $blocks -gt 5 ] &&
1656 error "$f: too many blocks after release: $blocks > 5"
1657 size=$(stat -c "%s" $f)
1658 [ $size -ne $orig_size ] &&
1659 error "$f: wrong size after release: $size != $orig_size"
1661 # Check we can release an file without stripe info
1665 check_hsm_flags $f "0x00000000"
1666 $LFS hsm_archive $f || error "could not archive file"
1667 wait_request_state $fid ARCHIVE SUCCEED
1669 # Release and check states
1670 $LFS hsm_release $f || error "could not release file"
1671 check_hsm_flags $f "0x0000000d"
1673 # Release again a file that is already released is OK
1674 $LFS hsm_release $f || fail "second release should succeed"
1675 check_hsm_flags $f "0x0000000d"
1677 stop_full_debug_logging
1679 run_test 21 "Simple release tests"
1682 # test needs a running copytool
1684 mkdir_on_mdt0 $DIR/$tdir
1686 local f=$DIR/$tdir/test_release
1687 local swap=$DIR/$tdir/test_swap
1689 # Create a file and check its states
1690 local fid=$(create_small_file $f)
1691 check_hsm_flags $f "0x00000000"
1693 $LFS hsm_archive $f || error "could not archive file"
1694 wait_request_state $fid ARCHIVE SUCCEED
1696 # Release and check states
1697 $LFS hsm_release $f || error "could not release file"
1698 check_hsm_flags $f "0x0000000d"
1700 create_small_file $swap
1701 $LFS swap_layouts $swap $f && error "swap_layouts should failed"
1705 run_test 22 "Could not swap a release file"
1708 # test needs a running copytool
1710 mkdir_on_mdt0 $DIR/$tdir
1712 local f=$DIR/$tdir/test_mtime
1714 # Create a file and check its states
1715 local fid=$(create_small_file $f)
1716 check_hsm_flags $f "0x00000000"
1718 $LFS hsm_archive $f || error "could not archive file"
1719 wait_request_state $fid ARCHIVE SUCCEED
1721 # Set modification time in the past
1722 touch -m -a -d @978261179 $f
1724 # Release and check states
1725 $LFS hsm_release $f || error "could not release file"
1726 check_hsm_flags $f "0x0000000d"
1728 local MTIME=$(stat -c "%Y" $f)
1729 local ATIME=$(stat -c "%X" $f)
1730 [ $MTIME -eq "978261179" ] || fail "bad mtime: $MTIME"
1731 [ $ATIME -eq "978261179" ] || fail "bad atime: $ATIME"
1733 run_test 23 "Release does not change a/mtime (utime)"
1736 local file=$DIR/$tdir/$tfile
1745 # test needs a running copytool
1747 mkdir_on_mdt0 $DIR/$tdir
1749 fid=$(create_small_file $file)
1751 # Create a file and check its states
1752 check_hsm_flags $file "0x00000000"
1754 # Ensure atime is less than mtime and ctime.
1758 atime0=$(stat -c "%X" $file)
1759 mtime0=$(stat -c "%Y" $file)
1760 ctime0=$(stat -c "%Z" $file)
1762 [ $atime0 -lt $mtime0 ] ||
1763 error "atime $atime0 is not less than mtime $mtime0"
1765 [ $atime0 -lt $ctime0 ] ||
1766 error "atime $atime0 is not less than ctime $ctime0"
1768 # Archive should not change any timestamps.
1769 $LFS hsm_archive $file || error "cannot archive '$file'"
1770 wait_request_state $fid ARCHIVE SUCCEED
1772 atime1=$(stat -c "%X" $file)
1773 mtime1=$(stat -c "%Y" $file)
1774 ctime1=$(stat -c "%Z" $file)
1776 [ $atime0 -eq $atime1 ] ||
1777 error "archive changed atime from $atime0 to $atime1"
1779 [ $mtime0 -eq $mtime1 ] ||
1780 error "archive changed mtime from $mtime0 to $mtime1"
1782 [ $ctime0 -eq $ctime1 ] ||
1783 error "archive changed ctime from $ctime0 to $ctime1"
1785 # Release should not change any timestamps.
1786 $LFS hsm_release $file || error "cannot release '$file'"
1787 check_hsm_flags $file "0x0000000d"
1789 atime1=$(stat -c "%X" $file)
1790 mtime1=$(stat -c "%Y" $file)
1791 ctime1=$(stat -c "%Z" $file)
1793 [ $atime0 -eq $atime1 ] ||
1794 error "release changed atime from $atime0 to $atime1"
1796 [ $mtime0 -eq $mtime1 ] ||
1797 error "release changed mtime from $mtime0 to $mtime1"
1799 [ $ctime0 -eq $ctime1 ] ||
1800 error "release changed ctime from $ctime0 to $ctime1"
1802 # Restore should not change any timestamps.
1803 $LFS hsm_restore $file
1804 wait_request_state $fid RESTORE SUCCEED
1806 atime1=$(stat -c "%X" $file)
1807 mtime1=$(stat -c "%Y" $file)
1808 ctime1=$(stat -c "%Z" $file)
1810 [ $atime0 -eq $atime1 ] ||
1811 error "restore changed atime from $atime0 to $atime1"
1813 [ $mtime0 -eq $mtime1 ] ||
1814 error "restore changed mtime from $mtime0 to $mtime1"
1816 [ $ctime0 -eq $ctime1 ] ||
1817 error "restore changed ctime from $ctime0 to $ctime1"
1820 wait_copytools || error "Copytools failed to stop"
1822 # Once more, after unmount and mount.
1823 umount_client $MOUNT || error "cannot unmount '$MOUNT'"
1824 mount_client $MOUNT || error "cannot mount '$MOUNT'"
1826 atime1=$(stat -c "%X" $file)
1827 mtime1=$(stat -c "%Y" $file)
1828 ctime1=$(stat -c "%Z" $file)
1830 [ $atime0 -eq $atime1 ] ||
1831 error "remount changed atime from $atime0 to $atime1"
1833 [ $mtime0 -eq $mtime1 ] ||
1834 error "remount changed mtime from $mtime0 to $mtime1"
1836 [ $ctime0 -eq $ctime1 ] ||
1837 error "remount changed ctime from $ctime0 to $ctime1"
1839 run_test 24a "Archive, release, and restore does not change a/mtime (i/o)"
1842 local file=$DIR/$tdir/$tfile
1848 # Test needs a running copytool.
1850 mkdir_on_mdt0 $DIR/$tdir
1852 # Check that root can do HSM actions on a regular user's file.
1853 fid=$(create_small_file $file)
1854 sum0=$(md5sum $file)
1856 chown $RUNAS_ID:$RUNAS_GID $file ||
1857 error "cannot chown '$file' to '$RUNAS_ID'"
1859 chmod ugo-w $DIR/$tdir ||
1860 error "cannot chmod '$DIR/$tdir'"
1862 $LFS hsm_archive $file
1863 wait_request_state $fid ARCHIVE SUCCEED
1865 $LFS hsm_release $file
1866 check_hsm_flags $file "0x0000000d"
1868 $LFS hsm_restore $file
1869 wait_request_state $fid RESTORE SUCCEED
1871 # Check that ordinary user can get HSM state.
1872 $RUNAS $LFS hsm_state $file ||
1873 error "user '$RUNAS_ID' cannot get HSM state of '$file'"
1875 $LFS hsm_release $file
1876 check_hsm_flags $file "0x0000000d"
1878 # Check that ordinary user can accessed released file.
1879 sum1=$($RUNAS md5sum $file) ||
1880 error "user '$RUNAS_ID' cannot read '$file'"
1882 [ "$sum0" == "$sum1" ] ||
1883 error "md5sum mismatch for '$file'"
1885 run_test 24b "root can archive, release, and restore user files"
1888 local file=$DIR/$tdir/$tfile
1889 local action=archive
1894 # test needs a running copytool
1896 mkdir_on_mdt0 $DIR/$tdir
1898 # Save the default masks and check that cleanup_24c will
1899 # restore the request masks correctly.
1900 user_save=$(get_hsm_param user_request_mask)
1901 stack_trap "set_hsm_param user_request_mask '$user_save'" EXIT
1902 group_save=$(get_hsm_param group_request_mask)
1903 stack_trap "set_hsm_param group_request_mask '$group_save'" EXIT
1904 other_save=$(get_hsm_param other_request_mask)
1905 stack_trap "set_hsm_param other_request_mask '$other_save'" EXIT
1907 [ "$user_save" == RESTORE ] ||
1908 error "user_request_mask is '$user_save' expected 'RESTORE'"
1909 [ "$group_save" == RESTORE ] ||
1910 error "group_request_mask is '$group_save' expected 'RESTORE'"
1911 [ "$other_save" == RESTORE ] ||
1912 error "other_request_mask is '$other_save' expected 'RESTORE'"
1915 create_small_file $file
1916 chown $RUNAS_ID:$GROUP $file ||
1917 error "cannot chown '$file' to '$RUNAS_ID:$GROUP'"
1919 $RUNAS $LFS hsm_$action $file &&
1920 error "$action by user should fail"
1922 set_hsm_param user_request_mask $action
1923 $RUNAS $LFS hsm_$action $file ||
1924 error "$action by user should succeed"
1927 create_small_file $file
1928 chown nobody:$RUNAS_GID $file ||
1929 error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
1931 $RUNAS $LFS hsm_$action $file &&
1932 error "$action by group should fail"
1934 set_hsm_param group_request_mask $action
1935 $RUNAS $LFS hsm_$action $file ||
1936 error "$action by group should succeed"
1939 create_small_file $file
1940 chown nobody:$GROUP $file ||
1941 error "cannot chown '$file' to 'nobody:$GROUP'"
1943 $RUNAS $LFS hsm_$action $file &&
1944 error "$action by other should fail"
1946 set_hsm_param other_request_mask $action
1947 $RUNAS $LFS hsm_$action $file ||
1948 error "$action by other should succeed"
1950 run_test 24c "check that user,group,other request masks work"
1953 local file1=$DIR/$tdir/$tfile
1954 local file2=$DIR2/$tdir/$tfile
1958 mkdir_on_mdt0 $DIR/$tdir
1960 fid1=$(create_small_file $file1)
1963 $LFS getstripe $file1
1965 stack_trap "zconf_umount \"$(facet_host $SINGLEAGT)\" \"$MOUNT3\"" EXIT
1966 zconf_mount "$(facet_host $SINGLEAGT)" "$MOUNT3" ||
1967 error "cannot mount '$MOUNT3' on '$SINGLEAGT'"
1969 copytool setup -m "$MOUNT3"
1971 stack_trap "mount -o remount,rw \"$MOUNT2\"" EXIT
1972 mount -o remount,ro $MOUNT2
1974 do_nodes $(comma_list $(nodes_list)) $LCTL clear
1976 fid2=$(path2fid $file2)
1977 [ "$fid1" == "$fid2" ] ||
1978 error "FID mismatch '$fid1' != '$fid2'"
1980 $LFS hsm_archive $file2 &&
1981 error "archive should fail on read-only mount"
1982 check_hsm_flags $file1 "0x00000000"
1984 $LFS hsm_archive $file1 || error "Fail to archive $file1"
1985 wait_request_state $fid1 ARCHIVE SUCCEED
1987 $LFS hsm_release $file1
1988 $LFS hsm_restore $file2
1989 wait_request_state $fid1 RESTORE SUCCEED
1991 $LFS hsm_release $file1 || error "cannot release '$file1'"
1992 dd if=$file2 of=/dev/null bs=1M || error "cannot read '$file2'"
1994 $LFS hsm_release $file2 &&
1995 error "release should fail on read-only mount"
1999 run_test 24d "check that read-only mounts are respected"
2003 mkdir_on_mdt0 $DIR/$tdir
2005 local f=$DIR/$tdir/$tfile
2008 fid=$(create_small_file $f) || error "cannot create $f"
2009 $LFS hsm_archive $f || error "cannot archive $f"
2010 wait_request_state $fid ARCHIVE SUCCEED
2011 $LFS hsm_release $f || error "cannot release $f"
2012 while ! $LFS hsm_state $f | grep released; do
2016 tar -cf $TMP/$tfile.tar $DIR/$tdir || error "cannot tar $DIR/$tdir"
2018 run_test 24e "tar succeeds on HSM released files" # LU-6213
2021 # test needs a running copytool
2023 mkdir_on_mdt0 $DIR/$tdir
2025 mkdir -p $DIR/$tdir/d1
2026 local f=$DIR/$tdir/$tfile
2027 local fid=$(copy_file /etc/hosts $f)
2030 $LFS hsm_archive $f ||
2031 error "hsm_archive failed"
2032 wait_request_state $fid ARCHIVE SUCCEED
2033 $LFS hsm_release $f || error "cannot release $f"
2034 tar --xattrs -cvf $f.tar -C $DIR/$tdir $tfile
2037 tar --xattrs -xvf $f.tar -C $DIR/$tdir ||
2038 error "Can not recover the tar contents"
2040 echo "Sum0 = $sum0, sum1 = $sum1"
2041 [ "$sum0" == "$sum1" ] || error "md5sum mismatch for '$tfile'"
2043 run_test 24f "root can archive, release, and restore tar files"
2046 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
2047 skip "need MDS version 2.11.56 or later"
2049 local file=$DIR/$tdir/$tfile
2052 echo "RUNAS = '$RUNAS'"
2056 mkdir_on_mdt0 $DIR/$tdir
2057 chmod ugo+rwx $DIR/$tdir
2059 echo "Please listen carefully as our options have changed." | tee $file
2060 fid=$(path2fid $file)
2063 $LFS hsm_archive $file
2064 wait_request_state $fid ARCHIVE SUCCEED
2065 check_hsm_flags $file 0x00000009 # exists archived
2067 echo "To be electrocuted by your telephone, press #." | $RUNAS tee $file
2068 check_hsm_flags $file 0x0000000b # exists dirty archived
2070 run_test 24g "write by non-owner still sets dirty" # LU-11369
2073 # test needs a running copytool
2077 copy2archive /etc/hosts $tdir/$tfile
2079 local f=$DIR/$tdir/$tfile
2081 copytool import $tdir/$tfile $f
2083 $LFS hsm_set --lost $f
2088 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2090 run_test 25a "Restore lost file (HS_LOST flag) from import"\
2091 " (Operation not permitted)"
2094 # test needs a running copytool
2097 mkdir_on_mdt0 $DIR/$tdir
2099 local f=$DIR/$tdir/$tfile
2100 local fid=$(copy_file /etc/passwd $f)
2102 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2103 wait_request_state $fid ARCHIVE SUCCEED
2106 $LFS hsm_set --lost $f
2110 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2112 run_test 25b "Restore lost file (HS_LOST flag) after release"\
2113 " (Operation not permitted)"
2115 test_26A() { # was test_26
2116 # test needs a running copytool
2118 mkdir_on_mdt0 $DIR/$tdir
2120 local f=$DIR/$tdir/$tfile
2121 local fid=$(create_empty_file "$f")
2123 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2124 wait_request_state $fid ARCHIVE SUCCEED
2127 wait_request_state $fid REMOVE SUCCEED
2129 check_hsm_flags $f "0x00000000"
2131 run_test 26A "Remove the archive of a valid file"
2134 local raolu=$(get_hsm_param remove_archive_on_last_unlink)
2135 [[ $raolu -eq 0 ]] || error "RAoLU policy should be off"
2137 # test needs a running copytool
2139 mkdir_on_mdt0 $DIR/$tdir
2141 local f=$DIR/$tdir/$tfile
2142 local fid=$(copy_file /etc/passwd $f)
2144 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2145 wait_request_state $fid ARCHIVE SUCCEED
2147 local f2=$DIR/$tdir/${tfile}_2
2148 local fid2=$(copy_file /etc/passwd $f2)
2150 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
2151 wait_request_state $fid2 ARCHIVE SUCCEED
2153 local f3=$DIR/$tdir/${tfile}_3
2154 local fid3=$(copy_file /etc/passwd $f3)
2156 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f3
2157 wait_request_state $fid3 ARCHIVE SUCCEED
2159 # set a long grace_delay vs short loop_period
2160 local orig_loop_period=$(get_hsm_param loop_period)
2161 local orig_grace_delay=$(get_hsm_param grace_delay)
2162 stack_trap "set_hsm_param loop_period $orig_loop_period" EXIT
2163 set_hsm_param loop_period 10
2164 stack_trap "set_hsm_param grace_delay $orig_grace_delay" EXIT
2165 set_hsm_param grace_delay 100
2169 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2170 set_hsm_param remove_archive_on_last_unlink 1
2172 ln "$f3" "$f3"_bis || error "Unable to create hard-link"
2177 wait_request_state $fid2 REMOVE SUCCEED
2179 assert_request_count $fid REMOVE 0 \
2180 "Unexpected archived data remove request for $f"
2181 assert_request_count $fid3 REMOVE 0 \
2182 "Unexpected archived data remove request for $f3"
2184 run_test 26a "Remove Archive On Last Unlink (RAoLU) policy"
2187 # test needs a running copytool
2189 mkdir_on_mdt0 $DIR/$tdir
2191 local f=$DIR/$tdir/$tfile
2192 local fid=$(copy_file /etc/passwd $f)
2194 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2195 wait_request_state $fid ARCHIVE SUCCEED
2197 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2198 set_hsm_param remove_archive_on_last_unlink 1
2201 cdt_check_state stopped
2205 wait_request_state $fid REMOVE WAITING
2209 # copytool must re-register
2211 wait_copytools || error "copytool failed to stop"
2214 wait_request_state $fid REMOVE SUCCEED
2216 run_test 26b "RAoLU policy when CDT off"
2219 # test needs a running copytool
2221 mkdir_on_mdt0 $DIR/$tdir
2223 local f=$DIR/$tdir/$tfile
2224 local fid=$(copy_file /etc/passwd $f)
2226 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2227 wait_request_state $fid ARCHIVE SUCCEED
2229 local f2=$DIR/$tdir/${tfile}_2
2230 local fid2=$(copy_file /etc/passwd $f2)
2232 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
2233 wait_request_state $fid2 ARCHIVE SUCCEED
2235 # set a long grace_delay vs short loop_period
2236 local orig_loop_period=$(get_hsm_param loop_period)
2237 local orig_grace_delay=$(get_hsm_param grace_delay)
2238 stack_trap "set_hsm_param loop_period $orig_loop_period" EXIT
2239 set_hsm_param loop_period 10
2240 stack_trap "set_hsm_param grace_delay $orig_grace_delay" EXIT
2241 set_hsm_param grace_delay 100
2243 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2244 set_hsm_param remove_archive_on_last_unlink 1
2246 multiop_bg_pause $f O_c || error "open $f failed"
2252 wait_request_state $fid2 REMOVE SUCCEED
2253 assert_request_count $fid REMOVE 0 \
2254 "Unexpected archived data remove request for $f"
2256 kill -USR1 $pid || error "multiop early exit"
2257 # should reach autotest timeout if multiop fails to trap
2258 # signal, close file, and exit ...
2259 wait $pid || error "wait PID $PID failed"
2261 wait_request_state $fid REMOVE SUCCEED
2263 run_test 26c "RAoLU effective when file closed"
2266 # test needs a running copytool
2268 mkdir_on_mdt0 $DIR/$tdir
2270 local f=$DIR/$tdir/$tfile
2271 local fid=$(create_small_file $f)
2273 $LFS hsm_archive $f || error "could not archive file"
2274 wait_request_state $fid ARCHIVE SUCCEED
2276 # set a long grace_delay vs short loop_period
2277 local orig_loop_period=$(get_hsm_param loop_period)
2278 local orig_grace_delay=$(get_hsm_param grace_delay)
2279 stack_trap "set_hsm_param loop_period $orig_loop_period" EXIT
2280 set_hsm_param loop_period 10
2281 stack_trap "set_hsm_param grace_delay $orig_grace_delay" EXIT
2282 set_hsm_param grace_delay 100
2284 stack_trap "set_hsm_param remove_archive_on_last_unlink 0" EXIT
2285 set_hsm_param remove_archive_on_last_unlink 1
2287 multiop_bg_pause $f O_c || error "multiop failed"
2294 wait_request_state $fid REMOVE SUCCEED
2296 client_up || client_up || true
2298 kill -USR1 $MULTIPID
2299 wait $MULTIPID || error "multiop close failed"
2301 run_test 26d "RAoLU when Client eviction"
2304 # test needs a running copytool
2307 create_archive_file $tdir/$tfile
2308 local f=$DIR/$tdir/$tfile
2309 copytool import $tdir/$tfile $f
2310 local fid=$(path2fid $f)
2314 [[ $? != 0 ]] || error "Remove of a released file should fail"
2316 run_test 27a "Remove the archive of an imported file (Operation not permitted)"
2319 # test needs a running copytool
2321 mkdir_on_mdt0 $DIR/$tdir
2323 local f=$DIR/$tdir/$tfile
2324 local fid=$(create_empty_file "$f")
2326 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2327 wait_request_state $fid ARCHIVE SUCCEED
2332 [[ $? != 0 ]] || error "Remove of a released file should fail"
2334 run_test 27b "Remove the archive of a relased file (Operation not permitted)"
2337 # test needs a running copytool
2339 mkdir_on_mdt0 $DIR/$tdir
2341 local f=$DIR/$tdir/$tfile
2342 local fid=$(create_empty_file "$f")
2344 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2345 wait_request_state $fid ARCHIVE SUCCEED
2354 wait_request_state $fid REMOVE SUCCEED
2356 run_test 28 "Concurrent archive/file remove"
2359 # Tests --mntpath and --archive options
2362 copytool setup -m "$MOUNT" -a $archive_id
2364 # Bad archive number
2365 $LFS hsm_remove -m "$MOUNT" -a 33 0x857765760:0x8:0x2 2>&1 |
2366 grep "Invalid argument" ||
2367 error "unexpected hsm_remove failure (1)"
2369 # mntpath is present but file is given
2370 $LFS hsm_remove --mntpath "$MOUNT" --archive 30 /qwerty/uyt 2>&1 |
2371 grep "hsm: '/qwerty/uyt' is not a valid FID" ||
2372 error "unexpected hsm_remove failure (2)"
2374 run_test 29a "Tests --mntpath and --archive options"
2377 # test needs a running copytool
2379 mkdir_on_mdt0 $DIR/$tdir
2381 local f=$DIR/$tdir/$tfile
2382 local fid=$(create_small_file $f)
2384 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2385 wait_request_state $fid ARCHIVE SUCCEED
2389 $LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER $fid
2390 wait_request_state $fid REMOVE SUCCEED
2392 run_test 29b "Archive/delete/remove by FID from the archive."
2395 # test needs a running copytool
2397 mkdir_on_mdt0 $DIR/$tdir
2399 local fid1=$(create_small_file $DIR/$tdir/$tfile-1)
2400 local fid2=$(create_small_file $DIR/$tdir/$tfile-2)
2401 local fid3=$(create_small_file $DIR/$tdir/$tfile-3)
2403 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/$tfile-[1-3]
2404 wait_request_state $fid1 ARCHIVE SUCCEED
2405 wait_request_state $fid2 ARCHIVE SUCCEED
2406 wait_request_state $fid3 ARCHIVE SUCCEED
2408 rm -f $DIR/$tdir/$tfile-[1-3]
2410 echo $fid1 > $DIR/$tdir/list
2411 echo $fid2 >> $DIR/$tdir/list
2412 echo $fid3 >> $DIR/$tdir/list
2414 $LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER \
2415 --filelist $DIR/$tdir/list
2416 wait_request_state $fid1 REMOVE SUCCEED
2417 wait_request_state $fid2 REMOVE SUCCEED
2418 wait_request_state $fid3 REMOVE SUCCEED
2420 run_test 29c "Archive/delete/remove by FID, using a file list."
2423 # test needs more than one CT
2424 needclients 3 || return 0
2430 # start all of the copytools
2431 for n in $(seq $AGTCOUNT); do
2432 copytool setup -f agt$n -a $n
2435 mkdir_on_mdt0 $DIR/$tdir
2438 file=$DIR/$tdir/$tfile
2439 fid=$(create_small_file $file)
2441 $LFS hsm_archive $file
2442 wait_request_state $fid ARCHIVE SUCCEED
2443 check_hsm_flags $file "0x00000009"
2447 $LFS hsm_remove --mntpath "$MOUNT" -a 0 $fid ||
2448 error "cannot hsm_remove '$fid'"
2450 # give time for CDT to handle remove request and create broadcasted
2453 # remove request has been broadcasted ?
2454 local cnt=$(get_request_count $fid REMOVE)
2455 # broadcasted requests + original
2456 [[ $cnt -eq $((AGTCOUNT + 1)) ]] ||
2457 error "remove not broadcasted to all CTs"
2459 # give time for CDT and CTs to handle broadcasted
2460 wait_for_loop_period
2462 # each agent serves one different archive_id, so broadcasted
2463 # hsm_remove request should only succeed once and fail at all others
2467 for n in $(seq $AGTCOUNT); do
2468 res=$(do_facet $SINGLEMDS "$LCTL get_param -n \
2469 $HSM_PARAM.actions | awk \
2470 '/'$fid'.*action=REMOVE archive#='$n'/ \
2471 {print \\\$13}' | cut -f2 -d=")
2472 if [[ "$res" == "SUCCEED" ]]; then
2474 elif [[ "$res" == "FAILED" ]]; then
2479 [[ $scnt -eq 1 ]] ||
2480 error "one and only CT should have removed successfully"
2482 [[ $AGTCOUNT -eq $((scnt + fcnt)) ]] ||
2483 error "all but one CT should have failed to remove"
2485 run_test 29d "hsm_remove by FID with archive_id 0 for unlinked file cause "\
2486 "request to be sent once for each registered archive_id"
2489 # restore at exec cannot work on agent node (because of Linux kernel
2490 # protection of executables)
2491 needclients 2 || return 0
2493 # test needs a running copytool
2497 copy2archive /bin/true $tdir/$tfile
2499 local f=$DIR/$tdir/true
2500 copytool import $tdir/$tfile $f
2502 local fid=$(path2fid $f)
2504 stack_trap "cdt_clear_no_retry" EXIT
2505 # set no retry action mode
2512 [[ $st == 0 ]] || error "Failed to exec a released file"
2514 run_test 30a "Restore at exec (import case)"
2517 # restore at exec cannot work on agent node (because of Linux kernel
2518 # protection of executables)
2519 needclients 2 || return 0
2521 # test needs a running copytool
2524 mkdir_on_mdt0 $DIR/$tdir
2525 local f=$DIR/$tdir/true
2526 local fid=$(copy_file /bin/true $f)
2528 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2529 wait_request_state $fid ARCHIVE SUCCEED
2533 stack_trap cdt_clear_no_retry EXIT
2534 # set no retry action mode
2542 [[ $st == 0 ]] || error "Failed to exec a released file"
2544 run_test 30b "Restore at exec (release case)"
2547 needclients 2 || return 0
2549 # test needs a running copytool
2552 mkdir_on_mdt0 $DIR/$tdir
2553 local f=$DIR/$tdir/SLEEP
2554 local slp_sum1=$(md5sum /bin/sleep)
2555 local fid=$(copy_file /bin/sleep $f)
2557 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2558 wait_request_state $fid ARCHIVE SUCCEED
2560 check_hsm_flags $f "0x0000000d"
2562 stack_trap cdt_clear_no_retry EXIT
2563 # set no retry action mode
2566 do_node $CLIENT2 "$f 10" &
2570 [[ $? == 0 ]] && error "Update during exec of released file must fail"
2572 [[ $? == 0 ]] || error "Execution failed during run"
2574 if [[ $? != 0 ]]; then
2575 local slp_sum2=$(md5sum /bin/sleep)
2576 # in case sleep file is modified during the test
2577 [[ $slp_sum1 == $slp_sum2 ]] &&
2578 error "Binary overwritten during exec"
2581 check_hsm_flags $f "0x00000009"
2583 run_test 30c "Update during exec of released file must fail"
2585 restore_and_check_size() {
2588 local s=$(stat -c "%s" $f)
2590 local st=$(get_hsm_flags $f)
2594 while [[ "$st" != "0x00000009" && $cpt -le 10 ]]
2596 n=$(stat -c "%s" $f)
2597 # we echo in both cases to show stat is not hang
2598 if [[ $n != $s ]]; then
2599 echo "size seen is $n != $s"
2602 echo "size seen is right: $n == $s"
2606 st=$(get_hsm_flags $f)
2608 if [[ "$st" = "0x00000009" ]]; then
2611 echo " restore is too long"
2612 wait_request_state $fid RESTORE SUCCEED
2618 # test needs a running copytool
2620 mkdir_on_mdt0 $DIR/$tdir
2622 create_archive_file $tdir/$tfile
2623 local f=$DIR/$tdir/$tfile
2624 copytool import $tdir/$tfile $f
2625 local fid=$($LFS path2fid $f)
2628 restore_and_check_size $f $fid
2631 [[ $err -eq 0 ]] || error "File size changed during restore"
2633 run_test 31a "Import a large file and check size during restore"
2637 # test needs a running copytool
2639 mkdir_on_mdt0 $DIR/$tdir
2641 local f=$DIR/$tdir/$tfile
2642 local fid=$(create_file "$f" 1MB 39)
2644 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2645 wait_request_state $fid ARCHIVE SUCCEED
2648 restore_and_check_size $f $fid
2651 [[ $err -eq 0 ]] || error "File size changed during restore"
2653 run_test 31b "Restore a large unaligned file and check size during restore"
2656 # test needs a running copytool
2658 mkdir_on_mdt0 $DIR/$tdir
2660 local f=$DIR/$tdir/$tfile
2661 local fid=$(create_file "$f" 1M 39)
2663 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2664 wait_request_state $fid ARCHIVE SUCCEED
2667 restore_and_check_size $f $fid
2670 [[ $err -eq 0 ]] || error "File size changed during restore"
2672 run_test 31c "Restore a large aligned file and check size during restore"
2675 local f=$DIR/$tdir/$tfile
2677 mkdir_on_mdt0 $DIR/$tdir
2678 local fid=$(create_empty_file "$f")
2682 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2683 wait_request_state $fid ARCHIVE SUCCEED
2686 # Prevent restore from completing
2690 md5sum $f >/dev/null &
2693 wait_request_state $fid RESTORE STARTED
2698 # Check restore trigger process was killed
2700 [ $? -eq 143 ] || error "md5sum was not 'Terminated'"
2702 run_test 33 "Kill a restore waiting process"
2705 # test needs a running copytool
2707 mkdir_on_mdt0 $DIR/$tdir
2709 local f=$DIR/$tdir/$tfile
2710 local fid=$(create_empty_file "$f")
2712 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2713 wait_request_state $fid ARCHIVE SUCCEED
2716 # Prevent restore from completing
2719 md5sum $f >/dev/null &
2722 wait_request_state $fid RESTORE STARTED
2724 # rm must not block during restore
2725 timeout --signal=KILL 1 rm "$f" || error "rm $f failed"
2728 wait_request_state $fid RESTORE SUCCEED
2730 # Check md5sum pgm finished
2731 kill -0 $pid && error "Restore initiatior still running"
2732 wait $pid || error "Restore initiator failed with $?"
2734 # Check the file was actually deleted
2735 [ ! -f "$f" ] || error "$f was not deleted"
2737 run_test 34 "Remove file during restore"
2740 # test needs a running copytool
2742 mkdir_on_mdt0 $DIR/$tdir
2744 local f=$DIR/$tdir/$tfile
2745 local f1=$DIR/$tdir/$tfile-1
2746 local fid=$(create_empty_file "$f")
2747 local fid1=$(copy_file /etc/passwd $f1)
2749 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2750 wait_request_state $fid ARCHIVE SUCCEED
2753 # Prevent restore from completing
2756 md5sum $f >/dev/null &
2759 wait_request_state $fid RESTORE STARTED
2761 # mv must not block during restore
2762 timeout --signal=KILL 1 mv "$f1" "$f" || error "mv $f1 $f failed"
2765 wait_request_state $fid RESTORE SUCCEED
2767 # Check md5sum pgm finished
2768 kill -0 $pid && error "Restore initiatior still running"
2769 wait $pid || error "Restore initiator failed with $?"
2771 local fid2=$(path2fid $f)
2772 [[ $fid2 == $fid1 ]] || error "Wrong fid after mv $fid2 != $fid1"
2774 run_test 35 "Overwrite file during restore"
2777 # test needs a running copytool
2779 mkdir_on_mdt0 $DIR/$tdir
2781 local f=$DIR/$tdir/$tfile
2782 local fid=$(create_empty_file "$f")
2784 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2785 wait_request_state $fid ARCHIVE SUCCEED
2788 # Prevent restore from completing
2791 md5sum $f >/dev/null &
2794 wait_request_state $fid RESTORE STARTED
2796 # mv must not block during restore
2797 timeout --signal=KILL 10 mv "$f" "$f.new" ||
2798 error "mv '$f' '$f.new' failed with rc=$?"
2801 wait_request_state $fid RESTORE SUCCEED
2803 # Check md5sum pgm finished
2804 kill -0 $pid && error "Restore initiator is still running"
2805 wait $pid || error "Restore initiator failed with $?"
2807 run_test 36 "Move file during restore"
2810 # LU-5683: check that an archived dirty file can be rearchived.
2812 mkdir_on_mdt0 $DIR/$tdir
2814 local f=$DIR/$tdir/$tfile
2817 fid=$(create_small_file $f) || error "cannot create small file"
2819 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2820 wait_request_state $fid ARCHIVE SUCCEED
2821 $LFS hsm_release $f || error "cannot release $f"
2823 # Allow previous archive request to expire from the actions log.
2824 wait_for_grace_delay
2827 dd if=/dev/urandom of=$f bs=1M count=1 || error "cannot dirty file"
2829 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2830 wait_request_state $fid ARCHIVE SUCCEED
2832 run_test 37 "re-archive a dirty file"
2839 for n in $(seq 1 $count); do
2840 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $prefix.$n
2842 echo "$count archive requests submitted"
2846 local stream_count=4
2847 local file_count=100
2849 local f=$DIR/$tdir/$tfile
2853 local max_requests=$(get_hsm_param max_requests)
2855 stack_trap "set_hsm_param max_requests $max_requests" EXIT
2856 # Increase the number of HSM request that can be performed in
2857 # parallel. With the coordinator running once per second, this
2858 # also limits the number of requests per seconds that can be
2859 # performed, so we pick a decent number. But we also need to keep
2860 # that number low because the copytool has no rate limit and will
2861 # fail some requests if if gets too many at once.
2862 set_hsm_param max_requests 300
2864 for i in $(seq 1 $file_count); do
2865 for p in $(seq 1 $stream_count); do
2866 fid=$(copy_file /etc/hosts $f.$p.$i)
2872 # to be sure wait_all_done will not be mislead by previous tests
2874 wait_for_grace_delay
2876 # start archive streams in background (archive files in parallel)
2877 for p in $(seq 1 $stream_count); do
2878 multi_archive $f.$p $file_count &
2881 echo -n "Wait for all requests being enqueued..."
2886 run_test 40 "Parallel archive requests"
2888 hsm_archive_batch() {
2896 while [ $i -lt $files_num ]; do
2897 if [ $batch_num -eq $batch_max ]; then
2898 $LFS hsm_archive $fileset || error "HSM archive failed"
2899 # Reset the batch container.
2904 fileset+="${filebase}$i "
2905 batch_num=$(( batch_num + 1 ))
2909 if [ $batch_num -ne 0 ]; then
2910 $LFS hsm_archive $fileset || error "HSM archive failed"
2917 local dir=$DIR/$tdir
2920 set_hsm_param max_requests 1000000
2921 mkdir $dir || error "mkdir $dir failed"
2930 filebase="$dir/$tfile.start."
2931 createmany -m $filebase $files_num ||
2932 error "createmany -m $filebase failed: $?"
2935 hsm_archive_batch $files_num $batch_max "$filebase"
2936 elapsed=$((SECONDS - start))
2937 do_facet $SINGLEMDS "$LCTL get_param -n \
2938 $HSM_PARAM.actions | grep WAITING | wc -l"
2939 unlinkmany $filebase $files_num || error "unlinkmany $filabase failed"
2940 echo "Start Phase files_num: $files_num time: $elapsed"
2943 filebase="$dir/$tfile.in."
2944 createmany -m $filebase $files_num ||
2945 error "createmany -m $filebase failed: $?"
2947 hsm_archive_batch $files_num $batch_max "$filebase"
2948 elapsed=$((SECONDS - start))
2949 unlinkmany $filebase $files_num || error "unlinkmany $filabase failed"
2950 echo "Middle Phase files_num: $files_num time: $elapsed"
2953 filebase="$dir/$tfile.end."
2954 createmany -m $filebase $files_num ||
2955 error "createmany -m $filebase failed: $?"
2958 hsm_archive_batch $files_num $batch_max "$filebase"
2959 elapsed=$((SECONDS - start))
2960 do_facet $SINGLEMDS "$LCTL get_param -n \
2961 $HSM_PARAM.actions | grep WAITING | wc -l"
2963 unlinkmany $filebase $files_num || error "unlinkmany $filebase failed"
2964 echo "End Phase files_num: $files_num time: $elapsed"
2966 do_facet $SINGLEMDS "$LCTL get_param -n \
2967 $HSM_PARAM.actions | grep WAITING | wc -l"
2971 run_test 50 "Archive with large number of pending HSM actions"
2974 # test needs a running copytool
2977 mkdir_on_mdt0 $DIR/$tdir
2978 local f=$DIR/$tdir/$tfile
2979 local fid=$(create_small_file $f)
2981 $LFS hsm_archive $f || error "could not archive file"
2982 wait_request_state $fid ARCHIVE SUCCEED
2983 check_hsm_flags $f "0x00000009"
2985 multiop_bg_pause $f O_c || error "multiop failed"
2989 client_up || client_up || true
2991 kill -USR1 $MULTIPID
2992 wait $MULTIPID || error "multiop close failed"
2994 check_hsm_flags $f "0x0000000b"
2996 run_test 52 "Opened for write file on an evicted client should be set dirty"
2999 # test needs a running copytool
3002 mkdir_on_mdt0 $DIR/$tdir
3003 local f=$DIR/$tdir/$tfile
3004 local fid=$(create_small_file $f)
3006 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3007 error "could not archive file"
3008 wait_request_state $fid ARCHIVE SUCCEED
3009 check_hsm_flags $f "0x00000009"
3011 multiop_bg_pause $f o_c || error "multiop failed"
3015 client_up || client_up || true
3017 kill -USR1 $MULTIPID
3018 wait $MULTIPID || error "multiop close failed"
3020 check_hsm_flags $f "0x00000009"
3022 run_test 53 "Opened for read file on an evicted client should not be set dirty"
3025 mkdir_on_mdt0 $DIR/$tdir
3027 local f=$DIR/$tdir/$tfile
3028 local fid=$(create_file "$f" 1MB 39)
3032 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3033 error "could not archive file"
3034 wait_request_state $fid ARCHIVE STARTED
3036 check_hsm_flags $f "0x00000001"
3038 stack_trap "cdt_clear_no_retry" EXIT
3039 # Avoid coordinator resending this request as soon it has failed.
3044 wait_request_state $fid ARCHIVE FAILED
3046 check_hsm_flags $f "0x00000003"
3048 run_test 54 "Write during an archive cancels it"
3051 mkdir_on_mdt0 $DIR/$tdir
3053 local f=$DIR/$tdir/$tfile
3054 local fid=$(create_file "$f" 1MB 39)
3058 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3059 error "could not archive file"
3060 wait_request_state $fid ARCHIVE STARTED
3062 check_hsm_flags $f "0x00000001"
3064 stack_trap "cdt_clear_no_retry" EXIT
3065 # Avoid coordinator resending this request as soon it has failed.
3068 $TRUNCATE $f 1024 || error "truncate failed"
3070 wait_request_state $fid ARCHIVE FAILED
3072 check_hsm_flags $f "0x00000003"
3074 run_test 55 "Truncate during an archive cancels it"
3077 mkdir_on_mdt0 $DIR/$tdir
3079 local f=$DIR/$tdir/$tfile
3080 local fid=$(create_file "$f" 1MB 39)
3084 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3085 error "could not archive file"
3086 wait_request_state $fid ARCHIVE STARTED
3088 check_hsm_flags $f "0x00000001"
3090 # Change metadata and sync to be sure we are not changing only
3095 wait_request_state $fid ARCHIVE SUCCEED
3097 check_hsm_flags $f "0x00000009"
3099 run_test 56 "Setattr during an archive is ok"
3102 # Need one client for I/O, one for request
3103 needclients 2 || return 0
3105 # test needs a running copytool
3108 mkdir_on_mdt0 $DIR/$tdir
3109 local f=$DIR/$tdir/test_archive_remote
3110 # Create a file on a remote node
3111 do_node $CLIENT2 "dd if=/dev/urandom of=$f bs=1M "\
3112 "count=2 conv=fsync"
3115 do_node $CLIENT2 "$LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f" ||
3116 error "hsm_archive failed"
3117 local fid=$(path2fid $f)
3118 wait_request_state $fid ARCHIVE SUCCEED
3120 # Release and implicit restore it
3121 do_node $CLIENT2 "$LFS hsm_release $f" ||
3122 error "hsm_release failed"
3123 do_node $CLIENT2 "md5sum $f" ||
3124 error "hsm_restore failed"
3126 wait_request_state $fid RESTORE SUCCEED
3128 run_test 57 "Archive a file with dirty cache on another node"
3130 truncate_released_file() {
3134 local sz=$(stat -c %s $src_file)
3135 local f=$DIR/$tdir/$tfile
3136 local fid=$(copy_file $1 $f)
3140 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3141 error "could not archive file"
3142 wait_request_state $fid ARCHIVE SUCCEED
3144 $LFS hsm_release $f || error "could not release file"
3146 $TRUNCATE $f $trunc_to || error "truncate failed"
3149 local sz1=$(stat -c %s $f)
3150 [[ $sz1 == $trunc_to ]] ||
3151 error "size after trunc: $sz1 expect $trunc_to, original $sz"
3154 check_hsm_flags $f "0x0000000b"
3156 local state=$(get_request_state $fid RESTORE)
3157 [[ "$state" == "SUCCEED" ]] ||
3158 error "truncate $sz does not trig restore, state = $state"
3160 $TRUNCATE $ref $trunc_to
3161 cmp $ref $f || error "file data wrong after truncate"
3167 # test needs a running copytool
3170 mkdir_on_mdt0 $DIR/$tdir
3172 local sz=$(stat -c %s /etc/passwd)
3174 echo "truncate up from $sz to $((sz*2))"
3175 truncate_released_file /etc/passwd $((sz*2))
3177 echo "truncate down from $sz to $((sz/2))"
3178 truncate_released_file /etc/passwd $((sz/2))
3180 echo "truncate to 0"
3181 truncate_released_file /etc/passwd 0
3183 run_test 58 "Truncate a released file will trigger restore"
3187 [[ $MDS1_VERSION -lt $(version_code 2.7.63) ]] &&
3188 skip "Need MDS version at least 2.7.63"
3191 $MCREATE $DIR/$tfile || error "mcreate failed"
3192 $TRUNCATE $DIR/$tfile 42 || error "truncate failed"
3193 $LFS hsm_archive $DIR/$tfile || error "archive request failed"
3194 fid=$(path2fid $DIR/$tfile)
3195 wait_request_state $fid ARCHIVE SUCCEED
3196 $LFS hsm_release $DIR/$tfile || error "release failed"
3198 run_test 59 "Release stripeless file with non-zero size"
3201 # This test validates the fix for LU-4512. Ensure that the -u
3202 # option changes the progress reporting interval from the
3203 # default (30 seconds) to the user-specified interval.
3204 mkdir_on_mdt0 $DIR/$tdir
3206 local f=$DIR/$tdir/$tfile
3207 local fid=$(create_file "$f" 1M 10)
3210 local progress_timeout=$((interval * 4))
3211 copytool setup -b 1 --update-interval $interval
3214 local mdt=${MDT_PREFIX}${mdtidx}
3215 local mds=mds$((mdtidx + 1))
3217 # Wait for copytool to register
3218 wait_update_facet $mds \
3219 "$LCTL get_param -n ${mdt}.hsm.agents | grep -o ^uuid" \
3220 uuid 100 || error "coyptool failed to register with $mdt"
3222 local start_at=$(date +%s)
3223 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3224 error "could not archive file"
3226 local agent=$(facet_active_host $SINGLEAGT)
3227 local logfile=$(copytool_logfile $SINGLEAGT)
3229 wait_update $agent \
3230 "grep -o start.copy \"$logfile\"" "start copy" 100 ||
3231 error "copytool failed to start"
3233 local cmd="$LCTL get_param -n ${mdt}.hsm.active_requests"
3234 cmd+=" | awk '/'$fid'.*action=ARCHIVE/ {print \\\$12}' | cut -f2 -d="
3240 echo -n "Expecting a progress update within $progress_timeout seconds... "
3242 RESULT=$(do_node $(facet_active_host $mds) "$cmd")
3243 if [ -n "$RESULT" ] && [ "$RESULT" -gt 0 ]; then
3244 echo "$RESULT bytes copied in $WAIT seconds."
3246 elif [ $WAIT -ge $progress_timeout ]; then
3247 error "Timed out waiting for progress update!"
3250 WAIT=$((WAIT + sleep))
3254 local finish_at=$(date +%s)
3255 local elapsed=$((finish_at - start_at))
3257 # Ensure that the progress update occurred within the expected window.
3258 if [ $elapsed -lt $((interval - 1)) ]; then
3259 error "Expected progress update after at least $interval seconds"
3262 echo "Wait for on going archive hsm action to complete"
3263 wait_update $agent "grep -o copied \"$logfile\"" "copied" 10 ||
3264 echo "File archiving not completed even after 10 secs"
3266 run_test 60 "Changing progress update interval from default"
3269 # test needs a running copytool
3272 mkdir_on_mdt0 $DIR/$tdir
3273 local f=$DIR/$tdir/$tfile
3274 local fid=$(copy_file /etc/passwd $f)
3276 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3279 wait_request_state $fid ARCHIVE FAILED
3281 run_test 61 "Waiting archive of a removed file should fail"
3284 # test needs a new running copytool
3285 stack_trap copytool_monitor_cleanup EXIT
3286 copytool_monitor_setup
3287 copytool setup --event-fifo "$HSMTOOL_MONITOR_DIR/fifo"
3289 # Wait for the copytool to register.
3290 wait_update --verbose $(facet_active_host mds1) \
3291 "$LCTL get_param -n ${MDT_PREFIX}0.hsm.agents | grep -o ^uuid" \
3293 error "copytool failed to register with MDT0000"
3296 wait_copytools || error "Copytools failed to stop"
3298 local REGISTER_EVENT
3299 local UNREGISTER_EVENT
3300 while read event; do
3301 local parsed=$(parse_json_event "$event")
3302 if [ -z "$parsed" ]; then
3303 error "Copytool sent malformed event: $event"
3307 if [ $event_type == "REGISTER" ]; then
3308 REGISTER_EVENT=$event
3309 elif [ $event_type == "UNREGISTER" ]; then
3310 UNREGISTER_EVENT=$event
3312 done < <(echo $"$(get_copytool_event_log)")
3314 if [ -z "$REGISTER_EVENT" ]; then
3315 error "Copytool failed to send register event to FIFO"
3318 if [ -z "$UNREGISTER_EVENT" ]; then
3319 error "Copytool failed to send unregister event to FIFO"
3322 echo "Register/Unregister events look OK."
3324 run_test 70 "Copytool logs JSON register/unregister events to FIFO"
3327 # Bump progress interval for livelier events.
3330 # test needs a new running copytool
3331 stack_trap copytool_monitor_cleanup EXIT
3332 copytool_monitor_setup
3333 copytool setup --update-interval $interval --event-fifo \
3334 "$HSMTOOL_MONITOR_DIR/fifo"
3336 stack_trap "cdt_clear_no_retry" EXIT
3337 # Just start and stop the copytool to generate events.
3340 mkdir_on_mdt0 $DIR/$tdir
3342 local f=$DIR/$tdir/$tfile
3343 local fid=$(create_small_file "$f")
3345 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3346 error "could not archive file"
3347 wait_request_state $fid ARCHIVE SUCCEED
3349 local expected_fields="event_time data_fid source_fid"
3350 expected_fields+=" total_bytes current_bytes"
3353 [ARCHIVE_START]=false
3354 [ARCHIVE_FINISH]=false
3355 [ARCHIVE_RUNNING]=false
3357 while read event; do
3358 # Make sure we're not getting anything from previous events.
3359 for field in $expected_fields; do
3363 local parsed=$(parse_json_event "$event")
3364 if [ -z "$parsed" ]; then
3365 error "Copytool sent malformed event: $event"
3369 events["$event_type"]=true
3371 [ "$event_type" != ARCHIVE_RUNNING ] && continue
3373 # Do some simple checking of the progress update events.
3374 for expected_field in $expected_fields; do
3375 if [ -z ${!expected_field+x} ]; then
3376 error "Missing $expected_field field in event"
3380 [ $total_bytes -gt 0 ] || error "Expected total_bytes to be > 0"
3382 # These should be identical throughout an archive operation
3383 [ $source_fid == $data_fid ] ||
3384 error "Expected source_fid to equal data_fid"
3385 done < <(echo $"$(get_copytool_event_log)")
3387 # Check we received every type of events we were expecting
3388 for event in "${!events[@]}"; do
3389 ${events["$event"]} ||
3390 error "Copytool failed to send '$event' event to FIFO"
3393 echo "Archive events look OK."
3395 run_test 71 "Copytool logs JSON archive events to FIFO"
3398 # Bump progress interval for livelier events.
3401 # test needs a new running copytool
3402 stack_trap copytool_monitor_cleanup EXIT
3403 copytool_monitor_setup
3404 copytool setup --update-interval $interval --event-fifo \
3405 "$HSMTOOL_MONITOR_DIR/fifo"
3406 local test_file=$HSMTOOL_MONITOR_DIR/file
3408 local cmd="dd if=/dev/urandom of=$test_file count=16 bs=1000000 "
3410 do_facet $SINGLEAGT "$cmd" ||
3411 error "cannot create $test_file on $SINGLEAGT"
3412 copy2archive $test_file $tdir/$tfile
3414 mkdir_on_mdt0 $DIR/$tdir
3415 local f=$DIR/$tdir/$tfile
3416 copytool import $tdir/$tfile $f
3417 f=$DIR2/$tdir/$tfile
3418 echo "Verifying released state: "
3419 check_hsm_flags $f "0x0000000d"
3421 local fid=$(path2fid $f)
3423 wait_request_state $fid RESTORE SUCCEED
3425 local expected_fields="event_time data_fid source_fid"
3426 expected_fields+=" total_bytes current_bytes"
3430 while read event; do
3431 # Make sure we're not getting anything from previous events.
3432 for field in $expected_fields; do
3436 local parsed=$(parse_json_event "$event")
3437 if [ -z "$parsed" ]; then
3438 error "Copytool sent malformed event: $event"
3442 if [ $event_type == "RESTORE_START" ]; then
3444 if [ $source_fid != $data_fid ]; then
3445 error "source_fid should == data_fid at start"
3448 elif [ $event_type == "RESTORE_FINISH" ]; then
3450 if [ $source_fid != $data_fid ]; then
3451 error "source_fid should == data_fid at finish"
3454 elif [ $event_type != "RESTORE_RUNNING" ]; then
3458 # Do some simple checking of the progress update events.
3459 for expected_field in $expected_fields; do
3460 if [ -z ${!expected_field+x} ]; then
3461 error "Missing $expected_field field in event"
3465 if [ $total_bytes -eq 0 ]; then
3466 error "Expected total_bytes to be > 0"
3469 # When a restore starts out, the data fid is the same as the
3470 # source fid. After the restore has gotten going, we learn
3471 # the new data fid. Once the restore has finished, the source
3472 # fid is set to the new data fid.
3474 # We test this because some monitoring software may depend on
3475 # this behavior. If it changes, then the consumers of these
3476 # events may need to be modified.
3477 if [ $source_fid == $data_fid ]; then
3478 error "source_fid should != data_fid during restore"
3480 done < <(echo $"$(get_copytool_event_log)")
3482 if [ -z "$START_EVENT" ]; then
3483 error "Copytool failed to send restore start event to FIFO"
3486 if [ -z "$FINISH_EVENT" ]; then
3487 error "Copytool failed to send restore finish event to FIFO"
3490 echo "Restore events look OK."
3492 run_test 72 "Copytool logs JSON restore events to FIFO"
3495 file_count=51 # Max number of files constrained by LNET message size
3496 mkdir_on_mdt0 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3497 local f=$DIR/$tdir/$tfile
3498 local FILELIST=/tmp/filelist.txt
3502 for i in $(seq 1 $file_count); do
3503 fid=$(copy_file /etc/hosts $f.$i)
3504 echo $f.$i >> $FILELIST
3508 # to be sure wait_all_done will not be mislead by previous tests
3510 wait_for_grace_delay
3511 $LFS hsm_archive --filelist $FILELIST ||
3512 error "cannot archive a file list"
3514 $LFS hsm_release --filelist $FILELIST ||
3515 error "cannot release a file list"
3516 $LFS hsm_restore --filelist $FILELIST ||
3517 error "cannot restore a file list"
3520 run_test 90 "Archive/restore a file list"
3522 double_verify_reset_hsm_param() {
3524 echo "Testing $HSM_PARAM.$p"
3525 local val=$(get_hsm_param $p)
3527 local val2=$(($val * 2))
3528 set_hsm_param $p $val2
3529 val=$(get_hsm_param $p)
3530 [[ $val == $val2 ]] ||
3531 error "$HSM_PARAM.$p: $val != $val2 should be (2 * $save)"
3532 echo "Set $p to 0 must failed"
3536 set_hsm_param $p $save
3538 if [[ $rc == 0 ]]; then
3539 error "we must not be able to set $HSM_PARAM.$p to 0"
3544 double_verify_reset_hsm_param loop_period
3545 double_verify_reset_hsm_param grace_delay
3546 double_verify_reset_hsm_param active_request_timeout
3547 double_verify_reset_hsm_param max_requests
3548 double_verify_reset_hsm_param default_archive_id
3550 run_test 100 "Set coordinator /proc tunables"
3557 run_test 102 "Verify coordinator control"
3560 # test needs a running copytool
3567 for i in $(seq 1 20); do
3568 fid=$(copy_file /etc/passwd $DIR/$tdir/$i)
3570 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/*
3574 echo "Current requests"
3575 local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3576 $HSM_PARAM.actions |\
3577 grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
3579 [[ -z "$res" ]] || error "Some request have not been canceled"
3581 run_test 103 "Purge all requests"
3584 (( MDS1_VERSION >= $(version_code 2.14.56) )) ||
3585 skip "Need MDS version at least 2.14.56"
3587 cdt_clear_non_blocking_restore
3589 # test needs a running copytool
3594 local rpcs_inflight=$($LCTL get_param -n \
3595 "mdc.$(facet_svc mds1)*.max_rpcs_in_flight" |
3598 mkdir_on_mdt0 $DIR/$tdir
3599 for ((i=0; i < rpcs_inflight; i++)); do
3600 fids+=( $(copy_file /etc/passwd $DIR/$tdir/${tfile}_$i) )
3602 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/*
3606 local grep_regex="($(tr ' ' '|' <<< "${fids[*]}")).*action=ARCHIVE.*status=SUCCEED"
3608 while [[ $time -lt 5 ]] && [[ $cnt -ne ${#fids[@]} ]]; do
3609 cnt=$(do_facet mds1 "$LCTL get_param -n $HSM_PARAM.actions |
3610 grep -c -E '$grep_regex'")
3614 [[ $cnt -eq ${#fids[@]} ]] || error "Fail to archive files $cnt/${#fids[@]}"
3616 $LFS hsm_release $DIR/$tdir/*
3619 wait_copytools || error "Copytool failed to stop"
3622 for i in "${fids[@]}"; do
3623 cat $DIR/.lustre/fid/$i > /dev/null & pids+=($!)
3627 grep_regex="($(tr ' ' '|' <<< "${fids[*]}")).*action=RESTORE.*status=CANCELED"
3628 cnt=$(do_facet mds1 "$LCTL get_param -n $HSM_PARAM.actions |
3629 grep -cE '$grep_regex'")
3631 [[ "$cnt" -eq ${#fids[@]} ]] ||
3632 error "Some request have not been canceled ($cnt/${#fids[@]} canceled)"
3634 # cat cmds should not hang and should fail
3635 for i in "${!pids[@]}"; do
3637 error "Restore for ${tfile}_$i (${pids[$i]}) should fail" ||
3641 run_test 103a "Purge pending restore requests"
3646 mkdir_on_mdt0 $DIR/$tdir
3648 local f=$DIR/$tdir/$tfile
3649 local fid=$(create_empty_file "$f")
3651 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
3652 local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3653 $HSM_PARAM.actions |\
3654 grep $fid | cut -f16 -d=")
3656 [[ "$data1" == "$DATAHEX" ]] ||
3657 error "Data field in records is ($data1) and not ($DATAHEX)"
3661 run_test 104 "Copy tool data field"
3664 local max_requests=$(get_hsm_param max_requests)
3665 mkdir_on_mdt0 $DIR/$tdir
3668 stack_trap "set_hsm_param max_requests $max_requests" EXIT
3669 set_hsm_param max_requests 300
3672 for i in $(seq -w 1 10); do
3673 cp /etc/passwd $DIR/$tdir/$i
3674 $LFS hsm_archive $DIR/$tdir/$i
3676 local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3677 $HSM_PARAM.actions |\
3678 grep WAITING | wc -l")
3682 local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3683 $HSM_PARAM.actions |\
3684 grep WAITING | wc -l")
3687 [[ "$reqcnt1" == "$reqcnt2" ]] ||
3688 error "Requests count after shutdown $reqcnt2 != "\
3689 "before shutdown $reqcnt1"
3691 run_test 105 "Restart of coordinator"
3694 # test needs a running copytool
3697 local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
3699 check_agent_registered $uuid
3701 search_copytools || error "No copytool found"
3704 wait_copytools || error "Copytool failed to stop"
3706 check_agent_unregistered $uuid
3709 uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
3710 check_agent_registered $uuid
3712 run_test 106 "Copytool register/unregister"
3715 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
3717 # test needs a running copytool
3719 # create and archive file
3720 mkdir_on_mdt0 $DIR/$tdir
3721 local f1=$DIR/$tdir/$tfile
3722 local fid=$(copy_file /etc/passwd $f1)
3723 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
3724 wait_request_state $fid ARCHIVE SUCCEED
3725 # shutdown and restart MDS
3727 # check the copytool still gets messages from MDT
3728 local f2=$DIR/$tdir/2
3729 local fid=$(copy_file /etc/passwd $f2)
3730 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
3731 # main check of this sanity: this request MUST succeed
3732 wait_request_state $fid ARCHIVE SUCCEED
3734 run_test 107 "Copytool re-register after MDS restart"
3736 policy_set_and_test()
3740 do_facet $SINGLEMDS $LCTL set_param "$HSM_PARAM.policy=\\\"$change\\\""
3741 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3742 [[ "$policy" == "$target" ]] ||
3743 error "Wrong policy after '$change': '$policy' != '$target'"
3747 # to force default policy setting if error
3748 CDT_POLICY_HAD_CHANGED=true
3750 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3751 local default="NonBlockingRestore [NoRetryAction]"
3752 [[ "$policy" == "$default" ]] ||
3753 error "default policy has changed,"\
3754 " '$policy' != '$default' update the test"
3755 policy_set_and_test "+NBR" "[NonBlockingRestore] [NoRetryAction]"
3756 policy_set_and_test "+NRA" "[NonBlockingRestore] [NoRetryAction]"
3757 policy_set_and_test "-NBR" "NonBlockingRestore [NoRetryAction]"
3758 policy_set_and_test "-NRA" "NonBlockingRestore NoRetryAction"
3759 policy_set_and_test "NRA NBR" "[NonBlockingRestore] [NoRetryAction]"
3760 # useless bacause we know but safer for futur changes to use real value
3761 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3762 echo "Next set_param must failed"
3763 policy_set_and_test "wrong" "$policy"
3766 echo "Back to default policy"
3767 cdt_set_sanity_policy
3769 run_test 109 "Policy display/change"
3772 # test needs a running copytool
3775 mkdir_on_mdt0 $DIR/$tdir
3777 copy2archive /etc/passwd $tdir/$tfile
3779 local f=$DIR/$tdir/$tfile
3780 copytool import $tdir/$tfile $f
3781 local fid=$(path2fid $f)
3783 cdt_set_non_blocking_restore
3788 wait_request_state $fid RESTORE SUCCEED
3789 cdt_clear_non_blocking_restore
3793 error "md5sum returns $st != 1, "\
3794 "should also perror ENODATA (No data available)"
3796 run_test 110a "Non blocking restore policy (import case)"
3799 # test needs a running copytool
3802 mkdir_on_mdt0 $DIR/$tdir
3803 local f=$DIR/$tdir/$tfile
3804 local fid=$(copy_file /etc/passwd $f)
3805 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3806 wait_request_state $fid ARCHIVE SUCCEED
3809 cdt_set_non_blocking_restore
3814 wait_request_state $fid RESTORE SUCCEED
3815 cdt_clear_non_blocking_restore
3819 error "md5sum returns $st != 1, "\
3820 "should also perror ENODATA (No data available)"
3822 run_test 110b "Non blocking restore policy (release case)"
3825 # test needs a running copytool
3828 mkdir_on_mdt0 $DIR/$tdir
3829 copy2archive /etc/passwd $tdir/$tfile
3831 local f=$DIR/$tdir/$tfile
3833 copytool import $tdir/$tfile $f
3834 local fid=$(path2fid $f)
3838 copytool_remove_backend $fid
3841 wait_request_state $fid RESTORE FAILED
3848 [[ $st == 0 ]] || error "Restore does not failed"
3850 run_test 111a "No retry policy (import case), restore will error"\
3851 " (No such file or directory)"
3854 # test needs a running copytool
3857 mkdir_on_mdt0 $DIR/$tdir
3858 local f=$DIR/$tdir/$tfile
3859 local fid=$(copy_file /etc/passwd $f)
3860 stack_trap cdt_clear_no_retry EXIT
3862 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3863 wait_request_state $fid ARCHIVE SUCCEED
3866 copytool_remove_backend $fid
3869 wait_request_state $fid RESTORE FAILED
3873 [[ $st == 0 ]] || error "Restore does not failed"
3875 run_test 111b "No retry policy (release case), restore will error"\
3876 " (No such file or directory)"
3879 # test needs a running copytool
3882 mkdir_on_mdt0 $DIR/$tdir
3883 local f=$DIR/$tdir/$tfile
3884 local fid=$(copy_file /etc/passwd $f)
3886 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3887 local l=$($LFS hsm_action $f)
3889 local res=$(echo $l | cut -f 2- -d" " | grep ARCHIVE)
3892 wait_request_state $fid ARCHIVE SUCCEED
3895 [[ ! -z "$res" ]] || error "action is $l which is not an ARCHIVE"
3897 run_test 112 "State of recorded request"
3900 mkdir_on_mdt0 $DIR/$tdir
3902 local file1=$DIR/$tdir/$tfile
3903 local file2=$DIR2/$tdir/$tfile
3905 local fid=$(create_small_sync_file $file1)
3907 stack_trap "zconf_umount \"$(facet_host $SINGLEAGT)\" \"$MOUNT3\"" EXIT
3908 zconf_mount "$(facet_host $SINGLEAGT)" "$MOUNT3" ||
3909 error "cannot mount '$MOUNT3' on '$SINGLEAGT'"
3911 copytool setup -m "$MOUNT3"
3913 do_nodes $(comma_list $(nodes_list)) $LCTL clear
3915 $LFS hsm_archive $file1 || error "Fail to archive $file1"
3916 wait_request_state $fid ARCHIVE SUCCEED
3918 $LFS hsm_release $file1
3919 echo "Verifying released state: "
3920 check_hsm_flags $file1 "0x0000000d"
3922 multiop_bg_pause $file1 oO_WRONLY:O_APPEND:_w4c || error "multiop failed"
3925 kill -USR1 $MULTIPID
3930 local size1=$(stat -c "%s" $file1)
3931 local size2=$(stat -c "%s" $file2)
3933 [ $size1 -eq $size2 ] || error "sizes are different $size1 $size2"
3935 run_test 113 "wrong stat after restore"
3938 mkdir_on_mdt0 $DIR/$tdir
3940 local f1=$DIR/$tdir/${tfile}1
3941 local f2=$DIR/$tdir/${tfile}2
3942 local fid1=$(create_empty_file "$f1")
3943 local fid2=$(create_empty_file "$f2")
3947 # Prevent archive from completing
3950 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1 $f2
3951 # wait archive to register at CDT
3952 wait_request_state "$fid1" ARCHIVE WAITING
3954 # Set f2 in an incompatible state
3955 $LFS hsm_set --noarchive $f2
3959 wait_request_state "$fid1" ARCHIVE SUCCEED
3960 wait_request_state "$fid2" ARCHIVE FAILED
3962 run_test 114 "Incompatible request does not set other requests as STARTED"
3965 mkdir_on_mdt0 $DIR/$tdir
3967 local f=$DIR/$tdir/$tfile
3968 local fid=$(create_empty_file "$f")
3972 # Prevent archive from completing
3975 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3976 # wait archive to register at CDT
3977 wait_request_state $fid ARCHIVE STARTED
3979 # Cancel the archive
3980 $LFS hsm_cancel "$f"
3982 wait_request_state $fid ARCHIVE CANCELED
3985 wait_request_state $fid CANCEL SUCCEED
3987 run_test 200 "Register/Cancel archive"
3990 # test needs a running copytool
3992 mkdir_on_mdt0 $DIR/$tdir
3994 local f=$DIR/$tdir/$tfile
3995 create_archive_file $tdir/$tfile
3996 copytool import $tdir/$tfile $f
3997 local fid=$(path2fid $f)
3999 # test with cdt on is made in test_222
4002 # wait restore to register at CDT
4003 wait_request_state $fid RESTORE WAITING
4006 wait_request_state $fid RESTORE CANCELED
4007 wait_request_state $fid CANCEL SUCCEED
4009 run_test 201 "Register/Cancel restore"
4012 mkdir_on_mdt0 $DIR/$tdir
4014 local f=$DIR/$tdir/$tfile
4015 local fid=$(create_empty_file "$f")
4017 # test needs a running copytool
4020 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4021 wait_request_state $fid ARCHIVE SUCCEED
4025 # wait remove to register at CDT
4026 wait_request_state $fid REMOVE STARTED
4029 wait_request_state $fid REMOVE CANCELED
4031 run_test 202 "Register/Cancel remove"
4033 test_220A() { # was test_220
4034 # test needs a running copytool
4037 mkdir_on_mdt0 $DIR/$tdir
4039 local f=$DIR/$tdir/$tfile
4040 local fid=$(copy_file /etc/passwd $f)
4044 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4045 wait_request_state $fid ARCHIVE SUCCEED
4047 changelog_find -type HSM -target-fid $fid -flags 0x0 ||
4048 error "The expected changelog was not emitted"
4050 run_test 220A "Changelog for archive"
4053 # test needs a running copytool
4056 mkdir_on_mdt0 $DIR/$tdir
4058 local f=$DIR/$tdir/$tfile
4059 local fid=$(copy_file /etc/passwd $f)
4063 # block copytool operations to allow for HSM request to be
4064 # submitted and file be unlinked (CDT will find object removed)
4067 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4069 # wait request to reach CT
4070 wait_request_state $fid ARCHIVE STARTED
4076 wait_request_state $fid ARCHIVE FAILED
4079 changelog_find -type HSM -target-fid $fid -flags 0x2 ||
4080 error "The expected changelog was not emitted"
4082 run_test 220a "Changelog for failed archive"
4085 mkdir_on_mdt0 $DIR/$tdir
4087 local f=$DIR/$tdir/$tfile
4088 local fid=$(create_empty_file "$f")
4093 # Prevent archive from completing
4095 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4096 wait_request_state $fid ARCHIVE STARTED
4099 wait_request_state $fid ARCHIVE CANCELED
4102 wait_request_state $fid CANCEL SUCCEED
4104 changelog_find -type HSM -target-fid $fid -flags 0x7d ||
4105 error "The expected changelog was not emitted"
4107 run_test 221 "Changelog for archive canceled"
4110 # test needs a running copytool
4113 mkdir_on_mdt0 $DIR/$tdir
4114 copy2archive /etc/passwd $tdir/$tfile
4116 local f=$DIR/$tdir/$tfile
4117 copytool import $tdir/$tfile $f
4118 local fid=$(path2fid $f)
4123 wait_request_state $fid RESTORE SUCCEED
4125 changelog_find -type HSM -target-fid $fid -flags 0x80 ||
4126 error "The expected changelog was not emitted"
4128 run_test 222a "Changelog for explicit restore"
4131 # test needs a running copytool
4134 mkdir_on_mdt0 $DIR/$tdir
4135 local f=$DIR/$tdir/$tfile
4136 local fid=$(copy_file /etc/passwd $f)
4139 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4140 wait_request_state $fid ARCHIVE SUCCEED
4145 wait_request_state $fid RESTORE SUCCEED
4147 changelog_find -type HSM -target-fid $fid -flags 0x80 ||
4148 error "The expected changelog was not emitted"
4150 run_test 222b "Changelog for implicit restore"
4153 # test needs a running copytool
4156 mkdir_on_mdt0 $DIR/$tdir
4157 copy2archive /etc/passwd $tdir/$tfile
4159 local f=$DIR/$tdir/$tfile
4160 copytool import $tdir/$tfile $f
4161 local fid=$(path2fid $f)
4165 # block copytool operations to allow for HSM request to be
4166 # submitted and file be unlinked (CDT will find object removed)
4171 # wait request to reach CT
4172 wait_request_state $fid RESTORE STARTED
4178 wait_request_state $fid RESTORE FAILED
4181 changelog_find -type HSM -target-fid $fid -flags 0x82 ||
4182 error "The expected changelog was not emitted"
4184 run_test 222c "Changelog for failed explicit restore"
4187 # test needs a running copytool
4190 mkdir_on_mdt0 $DIR/$tdir
4191 local f=$DIR/$tdir/$tfile
4192 local fid=$(copy_file /etc/passwd $f)
4195 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4196 wait_request_state $fid ARCHIVE SUCCEED
4199 copytool_remove_backend $fid
4202 wait_request_state $fid RESTORE FAILED
4205 changelog_find -type HSM -target-fid $fid -flags 0x82 ||
4206 error "The expected changelog was not emitted"
4208 run_test 222d "Changelog for failed implicit restore"
4211 # test needs a running copytool
4213 mkdir_on_mdt0 $DIR/$tdir
4215 local f=$DIR/$tdir/$tfile
4216 create_archive_file $tdir/$tfile
4220 copytool import $tdir/$tfile $f
4221 local fid=$(path2fid $f)
4224 wait_request_state $fid RESTORE STARTED
4226 wait_request_state $fid RESTORE CANCELED
4227 wait_request_state $fid CANCEL SUCCEED
4229 changelog_find -type HSM -target-fid $fid -flags 0xfd ||
4230 error "The expected changelog was not emitted"
4232 run_test 223a "Changelog for restore canceled (import case)"
4235 mkdir_on_mdt0 $DIR/$tdir
4237 local f=$DIR/$tdir/$tfile
4238 local fid=$(create_empty_file "$f")
4243 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4244 wait_request_state $fid ARCHIVE SUCCEED
4247 # Prevent restore from completing
4250 wait_request_state $fid RESTORE STARTED
4253 wait_request_state $fid RESTORE CANCELED
4256 wait_request_state $fid CANCEL SUCCEED
4258 changelog_find -type HSM -target-fid $fid -flags 0xfd ||
4259 error "The expected changelog was not emitted"
4261 run_test 223b "Changelog for restore canceled (release case)"
4263 test_224A() { # was test_224
4264 # test needs a running copytool
4267 mkdir_on_mdt0 $DIR/$tdir
4269 local f=$DIR/$tdir/$tfile
4270 local fid=$(copy_file /etc/passwd $f)
4273 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4274 wait_request_state $fid ARCHIVE SUCCEED
4277 wait_request_state $fid REMOVE SUCCEED
4279 changelog_find -type HSM -target-fid $fid -flags 0x200 ||
4280 error "The expected changelog was not emitted"
4282 run_test 224A "Changelog for remove"
4285 # test needs a running copytool
4288 mkdir_on_mdt0 $DIR/$tdir
4290 local f=$DIR/$tdir/$tfile
4291 local fid=$(copy_file /etc/passwd $f)
4294 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4295 wait_request_state $fid ARCHIVE SUCCEED
4297 copytool_remove_backend $fid
4299 # block copytool operations to allow for HSM request to be
4300 # submitted and file be unlinked (CDT will find object removed)
4305 # wait for request to reach CT
4306 wait_request_state $fid REMOVE STARTED
4312 wait_request_state $fid REMOVE FAILED
4314 # HE_REMOVE|ENOENT=0x202
4315 changelog_find -type HSM -target-fid $fid -flags 0x202 ||
4316 error "The expected changelog was not emitted"
4318 run_test 224a "Changelog for failed remove"
4321 # test is not usable because remove request is too fast
4322 # so it is always finished before cancel can be done ...
4323 echo "Test disabled"
4326 # test needs a running copytool
4328 mkdir_on_mdt0 $DIR/$tdir
4330 local f=$DIR/$tdir/$tfile
4331 local fid=$(create_empty_file "$f")
4334 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4335 wait_request_state $fid ARCHIVE SUCCEED
4337 # Prevent restore from completing
4342 wait_request_state $fid REMOVE CANCELED
4345 wait_request_state $fid CANCEL SUCCEED
4347 changelog_find -type HSM -target-fid $fid -flags 0x27d
4348 error "The expected changelog was not emitted"
4350 run_test 225 "Changelog for remove canceled"
4353 # test needs a running copytool
4356 mkdir_on_mdt0 $DIR/$tdir
4358 local f1=$DIR/$tdir/$tfile-1
4359 local f2=$DIR/$tdir/$tfile-2
4360 local f3=$DIR/$tdir/$tfile-3
4361 local fid1=$(copy_file /etc/passwd $f1)
4362 local fid2=$(copy_file /etc/passwd $f2)
4363 copy_file /etc/passwd $f3
4366 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
4367 wait_request_state $fid1 ARCHIVE SUCCEED
4369 $LFS hsm_archive $f2
4370 wait_request_state $fid2 ARCHIVE SUCCEED
4372 rm $f1 || error "rm $f1 failed"
4375 changelog_find -type UNLNK -target-fid $fid1 -flags 0x3 ||
4376 error "The expected changelog was not emitted"
4378 mv $f3 $f2 || error "mv $f3 $f2 failed"
4380 changelog_find -type RENME -target-fid $fid2 -flags 0x3 ||
4381 error "The expected changelog was not emitted"
4383 run_test 226 "changelog for last rm/mv with exiting archive"
4385 # This is just a utility function to clarify what test_227 does
4390 "$LFS" "$action" --$flag "$file" ||
4391 error "Cannot ${action#hsm_} $flag on '$file'"
4393 # Only one changelog should be produced
4394 local entries="$(changelog_find -type HSM -target-fid $fid)"
4395 [ $(wc -l <<< "$entries") -eq $((++count)) ] ||
4396 error "lfs $action --$flag '$file' produced more than one" \
4399 # Parse the last changelog record
4400 local entry="$(tail -n 1 <<< "$entries")"
4401 eval local -A changelog=$(changelog2array $entry)
4403 # Also check the flags match what is expected
4404 [[ ${changelog[flags]} == $target ]] ||
4405 error "Changelog flag is '${changelog[flags]}', not $target"
4409 local file="$DIR/$tdir/$tfile"
4410 local fid=$(create_empty_file "$file")
4415 for flag in norelease noarchive exists archived lost; do
4416 if [ "$flag" == lost ]; then
4417 # The flag "lost" only works on an archived file
4418 "$LFS" hsm_set --archived "$file"
4422 action="hsm_set" __test_227
4423 action="hsm_clear" __test_227
4426 run_test 227 "changelog when explicit setting of HSM flags"
4429 # test needs a running copytool
4431 mkdir_on_mdt0 $DIR/$tdir
4433 local fid=$(create_small_sync_file $DIR/$tfile)
4434 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
4435 wait_request_state $fid ARCHIVE SUCCEED
4437 $LFS hsm_release $DIR/$tfile
4438 check_hsm_flags $DIR/$tfile "0x0000000d"
4440 filefrag $DIR/$tfile | grep " 1 extent found" ||
4441 error "filefrag on released file must return only one extent"
4443 # only newer versions of cp detect sparse files by stat/FIEMAP
4445 cp --sparse=auto $DIR/$tfile $DIR/$tfile.2 ||
4446 error "copying $DIR/$tfile"
4447 cmp $DIR/$tfile $DIR/$tfile.2 || error "comparing copied $DIR/$tfile"
4449 $LFS hsm_release $DIR/$tfile
4450 check_hsm_flags $DIR/$tfile "0x0000000d"
4452 mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
4454 tar cf - --sparse $DIR/$tfile | tar xvf - -C $DIR/$tdir ||
4456 cmp $DIR/$tfile $DIR/$tdir/$DIR/$tfile ||
4457 error "comparing untarred $DIR/$tfile"
4459 rm -f $DIR/$tfile $DIR/$tfile.2 ||
4460 error "rm $DIR/$tfile or $DIR/$tfile.2 failed"
4462 run_test 228 "On released file, return extend to FIEMAP. For [cp,tar] --sparse"
4465 local file="$DIR/$tdir/$tfile"
4467 # set max_requests to allow one request of each type to be started (3)
4469 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4470 set_hsm_param max_requests 3
4473 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4474 set_hsm_param loop_period 1
4476 # send 1 requests of each kind twice
4479 mkdir_on_mdt0 $DIR/$tdir
4482 for action in archive restore remove; do
4483 local filepath="$file"-to-$action
4484 local fid=$(create_empty_file "$filepath")
4485 local fid2=$(create_empty_file "$filepath".bis)
4487 if [ "$action" != archive ]; then
4488 "$LFS" hsm_archive "$filepath"
4489 wait_request_state $fid ARCHIVE SUCCEED
4490 "$LFS" hsm_archive "$filepath".bis
4491 wait_request_state $fid2 ARCHIVE SUCCEED
4493 if [ "$action" == restore ]; then
4494 "$LFS" hsm_release "$filepath"
4495 "$LFS" hsm_release "$filepath".bis
4499 # suspend the copytool to prevent requests from completing
4500 stack_trap "copytool_continue" EXIT
4503 # send `max_requests' requests (one of each kind)
4504 for action in archive restore remove; do
4505 filepath="$file"-to-$action
4506 "$LFS" hsm_${action} "$filepath"
4507 wait_request_state $(path2fid "$filepath") "${action^^}" STARTED
4510 # send another batch of requests
4511 for action in archive restore remove; do
4512 "$LFS" hsm_${action} "$file-to-$action".bis
4514 # wait for `loop_period' seconds to make sure the coordinator has time
4515 # to register those, even though it should not
4518 # only the first batch of request should be started
4520 count=$(do_facet $SINGLEMDS "$LCTL" get_param -n $HSM_PARAM.actions |
4524 error "expected 3 STARTED requests, found $count"
4526 run_test 250 "Coordinator max request"
4529 mkdir_on_mdt0 $DIR/$tdir
4531 local f=$DIR/$tdir/$tfile
4532 local fid=$(create_empty_file "$f")
4535 # to have a short test
4536 local old_to=$(get_hsm_param active_request_timeout)
4537 set_hsm_param active_request_timeout 1
4538 # to be sure the cdt will wake up frequently so
4539 # it will be able to cancel the "old" request
4540 local old_loop=$(get_hsm_param loop_period)
4541 set_hsm_param loop_period 1
4546 # Prevent archive from completing
4548 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4549 wait_request_state $fid ARCHIVE STARTED
4551 # Let the request timeout
4552 wait_request_state $fid ARCHIVE CANCELED
4554 set_hsm_param active_request_timeout $old_to
4555 set_hsm_param loop_period $old_loop
4557 run_test 251 "Coordinator request timeout"
4560 mkdir_on_mdt0 $DIR/$tdir
4562 local f=$DIR/$tdir/$tfile
4563 local fid=$(create_empty_file "$f")
4565 # to have a short test
4566 stack_trap "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4567 set_hsm_param loop_period 1
4571 # Prevent archive from completing
4573 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4574 wait_request_state $fid ARCHIVE STARTED
4577 stack_trap "set_hsm_param active_request_timeout \
4578 $(get_hsm_param active_request_timeout)" EXIT
4579 set_hsm_param active_request_timeout 1
4581 wait_request_state $fid ARCHIVE CANCELED
4584 run_test 252 "Timeout'ed running archive of a removed file should be canceled"
4588 # test needs a running copytool
4591 mkdir_on_mdt0 $DIR/$tdir
4592 local f=$DIR/$tdir/$tfile
4594 dd if=/dev/zero of=$f bs=1MB count=10
4595 local fid=$(path2fid $f)
4597 $LFS hsm_archive $f || error "could not archive file"
4598 wait_request_state $fid ARCHIVE SUCCEED
4600 # clear locks to discard inode data
4601 cancel_lru_locks osc
4603 #define OBD_FAIL_MDC_MERGE 0x807
4604 $LCTL set_param fail_loc=0x807
4606 #expect error here, instead of release with wrong size
4609 if ((rc == 0)); then
4610 file_size=$(stat -c '%s' $f)
4611 if ((file_size != 10485760)); then
4612 error "Wrong file size after hsm_release"
4615 echo "could not release file"
4618 run_test 253 "Check for wrong file size after release"
4622 [ $MDS1_VERSION -lt $(version_code 2.10.56) ] &&
4623 skip "need MDS version at least 2.10.56"
4625 # Check that the counters are initialized to 0
4627 for request_type in archive restore remove; do
4628 count="$(get_hsm_param ${request_type}_count)" ||
4629 error "Reading ${request_type}_count failed with $?"
4631 [ "$count" -eq 0 ] ||
4632 error "Expected ${request_type}_count to be " \
4636 run_test 254a "Request counters are initialized to zero"
4640 [ $MDS1_VERSION -lt $(version_code 2.10.56) ] &&
4641 skip "need MDS version at least 2.10.56"
4643 # The number of request to launch (at least 32)
4644 local request_count=$((RANDOM % 32 + 32))
4645 printf "Will launch %i requests of each type\n" "$request_count"
4647 # Launch a copytool to process requests
4650 # Set hsm.max_requests to allow starting all requests at the same time
4652 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4653 set_hsm_param max_requests "$request_count"
4655 mkdir_on_mdt0 $DIR/$tdir
4659 for request_type in archive restore remove; do
4660 printf "Checking %s requests\n" "${request_type}"
4661 # Suspend the copytool to give us time to read the proc files
4664 for ((i = 0; i < $request_count; i++)); do
4665 case $request_type in
4667 create_empty_file "$DIR/$tdir/$tfile-$i" \
4671 lfs hsm_release "$DIR/$tdir/$tfile-$i"
4674 $LFS hsm_${request_type} "$DIR/$tdir/$tfile-$i"
4677 # Give the coordinator 10 seconds to start every request
4679 while get_hsm_param actions | grep -q WAITING; do
4682 [ $timeout -gt 0 ] ||
4683 error "${request_type^} requests took too " \
4687 count="$(get_hsm_param ${request_type}_count)"
4688 [ "$count" -eq "$request_count" ] ||
4689 error "Expected '$request_count' (!= '$count') " \
4690 "active $request_type requests"
4692 # Let the copytool process the requests
4694 # Give it 10 seconds maximum
4696 while get_hsm_param actions | grep -q STARTED; do
4699 [ $timeout -gt 0 ] ||
4700 error "${request_type^} requests took too " \
4704 count="$(get_hsm_param ${request_type}_count)"
4705 [ "$count" -eq 0 ] ||
4706 error "Expected 0 (!= '$count') " \
4707 "active $request_type requests"
4710 run_test 254b "Request counters are correctly incremented and decremented"
4714 [ $MDS1_VERSION -lt $(version_code 2.12.0) ] &&
4715 skip "Need MDS version at least 2.12.0"
4717 mkdir_on_mdt0 $DIR/$tdir
4719 local file="$DIR/$tdir/$tfile"
4720 local fid=$(create_empty_file "$file")
4722 # How do you make sure the coordinator has consumed any outstanding
4723 # event, without triggering an event yourself?
4725 # You wait for a request to disappear from the coordinator's llog.
4727 # Warning: the setup represents 90% of this test
4729 # Create and process an HSM request
4731 "$LFS" hsm_archive "$file"
4732 wait_request_state $fid ARCHIVE SUCCEED
4735 wait_copytools || error "failed to stop copytools"
4737 # Launch a new HSM request
4739 create_empty_file "$file"
4740 "$LFS" hsm_archive "$file"
4744 # Have the completed request be removed as soon as the cdt wakes up
4745 stack_trap "set_hsm_param grace_delay $(get_hsm_param grace_delay)" EXIT
4746 set_hsm_param grace_delay 1
4747 # (Hopefully, time on the MDS will behave nicely)
4748 do_facet $SINGLEMDS sleep 2 &
4750 # Increase `loop_period' as a mean to prevent the coordinator from
4751 # waking itself up to do some housekeeping.
4752 stack_trap "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4753 set_hsm_param loop_period 1000
4755 wait $! || error "waiting failed"
4757 wait_request_state $fid ARCHIVE ""
4758 # The coordinator will not wake up on its own for ~`loop_period' secs...
4760 # ... Unless a copytool registers. Now the real test begins
4762 wait_request_state $(path2fid "$file") ARCHIVE SUCCEED
4764 run_test 255 "Copytool registration wakes the coordinator up"
4766 # tests 260[a-c] rely on the parsing of the copytool's log file, they might
4767 # break in the future because of that.
4770 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
4771 skip "need MDS version 2.11.56 or later"
4773 local -a files=("$DIR/$tdir/$tfile".{0..15})
4776 mkdir_on_mdt0 $DIR/$tdir
4778 for file in "${files[@]}"; do
4779 create_small_file "$file"
4782 # Set a few hsm parameters
4784 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4785 set_hsm_param loop_period 1
4787 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4788 set_hsm_param max_requests 3
4792 "$LFS" hsm_archive "${files[0]}"
4793 wait_request_state "$(path2fid "${files[0]}")" ARCHIVE SUCCEED
4794 "$LFS" hsm_release "${files[0]}"
4798 wait_copytools || error "copytools failed to stop"
4800 # Send several archive requests
4801 for file in "${files[@]:1}"; do
4802 "$LFS" hsm_archive "$file"
4805 # Send one restore request
4806 "$LFS" hsm_restore "${files[0]}"
4811 # Wait for all the requests to complete
4812 wait_request_state "$(path2fid "${files[0]}")" RESTORE SUCCEED
4813 for file in "${files[@]:1}"; do
4814 wait_request_state "$(path2fid "$file")" ARCHIVE SUCCEED
4817 # Collect the actions in the order in which the copytool processed them
4819 $(do_facet "$SINGLEAGT" grep -o '\"RESTORE\\|ARCHIVE\"' \
4820 "$(copytool_logfile "$SINGLEAGT")")
4823 printf '%s\n' "${actions[@]}"
4826 for action in "${actions[@]:0:3}"; do
4827 [ "$action" == RESTORE ] && return
4830 error "Too many ARCHIVE requests were run before the RESTORE request"
4832 run_test 260a "Restore request have priority over other requests"
4834 # This test is very much tied to the implementation of the current priorisation
4835 # mechanism in the coordinator. It might not make sense to keep it in the future
4838 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
4839 skip "need MDS version 2.11.56 or later"
4841 local -a files=("$DIR/$tdir/$tfile".{0..15})
4844 mkdir_on_mdt0 $DIR/$tdir
4846 for file in "${files[@]}"; do
4847 create_small_file "$file"
4850 # Set a few hsm parameters
4852 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4853 set_hsm_param loop_period 1
4855 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4856 set_hsm_param max_requests 3
4859 copytool setup --archive-id 2
4860 "$LFS" hsm_archive --archive 2 "${files[0]}"
4861 wait_request_state "$(path2fid "${files[0]}")" ARCHIVE SUCCEED
4862 "$LFS" hsm_release "${files[0]}"
4866 wait_copytools || error "copytools failed to stop"
4868 # Send several archive requests
4869 for file in "${files[@]:1}"; do
4870 "$LFS" hsm_archive "$file"
4873 # Send one restore request
4874 "$LFS" hsm_restore "${files[0]}"
4878 copytool setup --archive-id 2
4880 # Wait for all the requests to complete
4881 wait_request_state "$(path2fid "${files[0]}")" RESTORE SUCCEED
4882 for file in "${files[@]:1}"; do
4883 wait_request_state "$(path2fid "$file")" ARCHIVE SUCCEED
4886 # Collect the actions in the order in which the copytool processed them
4888 $(do_facet "$SINGLEAGT" grep -o '\"RESTORE\\|ARCHIVE\"' \
4889 "$(copytool_logfile "$SINGLEAGT")")
4892 printf '%s\n' "${actions[@]}"
4895 for action in "${actions[@]:0:3}"; do
4896 [ "$action" == RESTORE ] && return
4899 error "Too many ARCHIVE requests were run before the RESTORE request"
4901 run_test 260b "Restore request have priority over other requests"
4903 # This test is very much tied to the implementation of the current priorisation
4904 # mechanism in the coordinator. It might not make sense to keep it in the future
4907 [ $MDS1_VERSION -lt $(version_code 2.12.0) ] &&
4908 skip "Need MDS version at least 2.12.0"
4910 local -a files=("$DIR/$tdir/$tfile".{0..15})
4913 mkdir_on_mdt0 $DIR/$tdir
4915 for file in "${files[@]}"; do
4916 create_small_file "$file"
4919 # Set a few hsm parameters
4921 "set_hsm_param loop_period $(get_hsm_param loop_period)" EXIT
4922 set_hsm_param loop_period 1000
4924 "set_hsm_param max_requests $(get_hsm_param max_requests)" EXIT
4925 set_hsm_param max_requests 3
4928 copytool setup --archive-id 2
4929 "$LFS" hsm_archive --archive 2 "${files[0]}"
4930 wait_request_state "$(path2fid "${files[0]}")" ARCHIVE SUCCEED
4931 "$LFS" hsm_release "${files[0]}"
4935 wait_copytools || error "copytools failed to stop"
4937 # Force the next coordinator run to do housekeeping
4941 "$LFS" hsm_archive "${files[1]}"
4945 copytool setup --archive-id 2
4947 wait_request_state "$(path2fid "${files[1]}")" ARCHIVE SUCCEED
4948 # The coordinator just did a housekeeping run it won't do another one
4949 # for around `loop_period' seconds => requests will not be reordered
4950 # if it costs too much (ie. when the coordinator has to discard a whole
4953 # Send several archive requests
4954 for file in "${files[@]:2}"; do
4955 "$LFS" hsm_archive "$file"
4958 # Send one restore request
4959 "$LFS" hsm_restore "${files[0]}"
4961 # Wait for all the requests to complete
4962 wait_request_state "$(path2fid "${files[0]}")" RESTORE SUCCEED
4963 for file in "${files[@]:2}"; do
4964 wait_request_state "$(path2fid "$file")" ARCHIVE SUCCEED
4967 # Collect the actions in the order in which the copytool processed them
4969 $(do_facet "$SINGLEAGT" grep -o '\"RESTORE\\|ARCHIVE\"' \
4970 "$(copytool_logfile "$SINGLEAGT")")
4973 printf '%s\n' "${actions[@]}"
4976 for action in "${actions[@]:0:3}"; do
4977 [ "$action" == RESTORE ] &&
4978 error "Restore requests should not be prioritised" \
4979 "unless the coordinator is doing housekeeping"
4983 run_test 260c "Requests are not reordered on the 'hot' path of the coordinator"
4986 local file=$DIR/$tdir/$tfile
4991 mkdir_on_mdt0 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
4993 dd if=/dev/zero of=$file bs=4k count=2 || error "Write $file failed"
4994 fid=$(path2fid $file)
4995 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $file
4996 wait_request_state $fid ARCHIVE SUCCEED
4998 $LFS hsm_state $file
4999 $LFS hsm_release $file
5000 $LFS hsm_restore $file
5001 wait_request_state $fid RESTORE SUCCEED
5002 $LFS hsm_release $file
5003 size=$(stat -c %s $file)
5004 [[ $size == 8192 ]] || error "Size after HSM release: $size"
5006 $LFS hsm_release $file
5007 $LFS hsm_restore $file
5008 $LFS hsm_release $file
5009 size=$(stat -c %s $file)
5010 [[ $size == 8192 ]] || error "Size after HSM release: $size"
5011 $LFS hsm_state $file
5013 run_test 261 "Report 0 bytes size after HSM release"
5016 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
5018 # the only way to test ondisk conf is to restart MDS ...
5019 echo "Stop coordinator and remove coordinator state at mount"
5022 # clean on disk conf set by default
5023 cdt_clear_mount_state
5024 cdt_check_state stopped
5026 # check cdt still off after umount/remount
5028 cdt_check_state stopped
5030 echo "Set coordinator start at mount, and start coordinator"
5031 cdt_set_mount_state enabled
5034 cdt_check_state enabled
5036 # check cdt still on after umount/remount
5038 cdt_check_state enabled
5040 # we are back to original state (cdt started at mount)
5042 run_test 300 "On disk coordinator state kept between MDT umount/mount"
5045 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
5047 local ai=$(get_hsm_param default_archive_id)
5048 local new=$((ai + 1))
5050 set_hsm_param default_archive_id $new -P
5052 local res=$(get_hsm_param default_archive_id)
5055 set_hsm_param default_archive_id "" "-P -d"
5057 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
5059 run_test 301 "HSM tunnable are persistent"
5062 [ "$CLIENTONLY" ] && skip "CLIENTONLY mode" && return
5064 local ai=$(get_hsm_param default_archive_id)
5065 local new=$((ai + 1))
5070 set_hsm_param default_archive_id $new -P
5073 for mdtno in $(seq 1 $MDSCOUNT); do
5078 cdt_check_state enabled
5080 local res=$(get_hsm_param default_archive_id)
5083 set_hsm_param default_archive_id "" "-P -d"
5085 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
5087 run_test 302 "HSM tunnable are persistent when CDT is off"
5090 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5094 mkdir_on_mdt0 $DIR/$tdir
5096 local dir_mdt0=$DIR/$tdir/mdt0
5097 local dir_mdt1=$DIR/$tdir/mdt1
5099 # create 1 dir per MDT
5100 stack_trap "rm -rf $dir_mdt0" EXIT
5101 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
5102 stack_trap "rm -rf $dir_mdt1" EXIT
5103 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
5105 # create 1 file in each MDT
5106 local fid1=$(create_small_file $dir_mdt0/$tfile)
5107 local fid2=$(create_small_file $dir_mdt1/$tfile)
5109 # check that hsm request on mdt0 is sent to the right MDS
5110 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
5111 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
5112 echo "archive successful on mdt0"
5114 # check that hsm request on mdt1 is sent to the right MDS
5115 $LFS hsm_archive $dir_mdt1/$tfile || error "lfs hsm_archive"
5116 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
5117 echo "archive successful on mdt1"
5119 run_test 400 "Single request is sent to the right MDT"
5122 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5126 mkdir_on_mdt0 $DIR/$tdir
5128 local dir_mdt0=$DIR/$tdir/mdt0
5129 local dir_mdt1=$DIR/$tdir/mdt1
5131 # create 1 dir per MDT
5132 stack_trap "rm -rf $dir_mdt0" EXIT
5133 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
5134 stack_trap "rm -rf $dir_mdt1" EXIT
5135 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
5137 # create 1 file in each MDT
5138 local fid1=$(create_small_file $dir_mdt0/$tfile)
5139 local fid2=$(create_small_file $dir_mdt1/$tfile)
5141 # check that compound requests are shunt to the rights MDTs
5142 $LFS hsm_archive $dir_mdt0/$tfile $dir_mdt1/$tfile ||
5143 error "lfs hsm_archive"
5144 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
5145 echo "archive successful on mdt0"
5146 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
5147 echo "archive successful on mdt1"
5149 run_test 401 "Compound requests split and sent to their respective MDTs"
5151 mdc_change_state() # facet, MDT_pattern, activate|deactivate
5156 local node=$(facet_active_host $facet)
5158 for mdc in $(do_facet $facet "$LCTL dl | grep -E ${pattern}-mdc" |
5159 awk '{print $4}'); do
5160 echo "$3 $mdc on $node"
5161 do_facet $facet "$LCTL --device $mdc $state" || return 1
5166 # deactivate all mdc on agent1
5167 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "deactivate"
5169 copytool setup --no-fail
5171 check_agent_unregistered "uuid" # match any agent
5173 # no expected running copytool
5174 search_copytools $agent && error "Copytool start should have failed"
5177 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "activate"
5179 run_test 402a "Copytool start fails if all MDTs are inactive"
5184 mkdir_on_mdt0 $DIR/$tdir
5186 local f=$DIR/$tdir/$tfile
5187 touch $f || error "touch $f failed"
5188 local fid=$(path2fid $f)
5190 #define OBD_FAIL_MDS_HSM_CT_REGISTER_NET 0x14d
5191 do_facet $SINGLEAGT lctl set_param fail_loc=0x14d
5192 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
5194 # give time for CDT to send request and to keep it for retry
5195 wait_for_loop_period
5197 wait_request_state $fid ARCHIVE WAITING
5199 do_facet $SINGLEAGT lctl set_param fail_loc=0
5201 # request should succeed now
5202 wait_request_state $fid ARCHIVE SUCCEED
5204 run_test 402b "CDT must retry request upon slow start of CT"
5207 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5209 local agent=$(facet_active_host $SINGLEAGT)
5211 # deactivate all mdc for MDT0001
5212 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
5215 local uuid=$(get_agent_uuid $agent)
5216 # check the agent is registered on MDT0000, and not on MDT0001
5217 check_agent_registered_by_mdt $uuid 0
5218 check_agent_unregistered_by_mdt $uuid 1
5220 # check running copytool process
5221 search_copytools $agent || error "No running copytools on $agent"
5223 # reactivate all mdc for MDT0001
5224 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
5226 # make sure the copytool is now registered to all MDTs
5227 check_agent_registered $uuid
5229 run_test 403 "Copytool starts with inactive MDT and register on reconnect"
5232 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5236 # create files on both MDT0000 and MDT0001
5237 mkdir_on_mdt0 $DIR/$tdir
5239 local dir_mdt0=$DIR/$tdir/mdt0
5240 stack_trap "rm -rf $dir_mdt0" EXIT
5241 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
5243 # create 1 file on mdt0
5244 local fid1=$(create_small_file $dir_mdt0/$tfile)
5246 # deactivate all mdc for MDT0001
5247 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
5249 # send an HSM request for files in MDT0000
5250 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
5252 # check for completion of files in MDT0000
5253 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
5254 echo "archive successful on mdt0"
5256 # reactivate all mdc for MDT0001
5257 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
5259 run_test 404 "Inactive MDT does not block requests for active MDTs"
5262 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
5266 mkdir_on_mdt0 $DIR/$tdir
5268 local striped_dir=$DIR/$tdir/striped_dir
5270 # create striped dir on all of MDTs
5271 $LFS mkdir -i 0 -c $MDSCOUNT $striped_dir || error "lfs mkdir"
5273 local fid1=$(create_small_sync_file $striped_dir/${tfile}_0)
5274 local fid2=$(create_small_sync_file $striped_dir/${tfile}_1)
5275 local fid3=$(create_small_sync_file $striped_dir/${tfile}_2)
5276 local fid4=$(create_small_sync_file $striped_dir/${tfile}_3)
5278 local idx1=$($LFS getstripe -m $striped_dir/${tfile}_0)
5279 local idx2=$($LFS getstripe -m $striped_dir/${tfile}_1)
5280 local idx3=$($LFS getstripe -m $striped_dir/${tfile}_2)
5281 local idx4=$($LFS getstripe -m $striped_dir/${tfile}_3)
5283 # check that compound requests are shunt to the rights MDTs
5284 $LFS hsm_archive $striped_dir/${tfile}_0 $striped_dir/${tfile}_1 \
5285 $striped_dir/${tfile}_2 $striped_dir/${tfile}_3 ||
5286 error "lfs hsm_archive"
5288 wait_request_state $fid1 ARCHIVE SUCCEED $idx1 &&
5289 echo "archive successful on $fid1"
5290 wait_request_state $fid2 ARCHIVE SUCCEED $idx2 &&
5291 echo "archive successful on $fid2"
5292 wait_request_state $fid3 ARCHIVE SUCCEED $idx3 &&
5293 echo "archive successful on $fid3"
5294 wait_request_state $fid4 ARCHIVE SUCCEED $idx4 &&
5295 echo "archive successful on $fid4"
5297 $LFS hsm_release $striped_dir/${tfile}_0 || error "lfs hsm_release 1"
5298 $LFS hsm_release $striped_dir/${tfile}_1 || error "lfs hsm_release 2"
5299 $LFS hsm_release $striped_dir/${tfile}_2 || error "lfs hsm_release 3"
5300 $LFS hsm_release $striped_dir/${tfile}_3 || error "lfs hsm_release 4"
5302 cat $striped_dir/${tfile}_0 > /dev/null || error "cat ${tfile}_0 failed"
5303 cat $striped_dir/${tfile}_1 > /dev/null || error "cat ${tfile}_1 failed"
5304 cat $striped_dir/${tfile}_2 > /dev/null || error "cat ${tfile}_2 failed"
5305 cat $striped_dir/${tfile}_3 > /dev/null || error "cat ${tfile}_3 failed"
5307 run_test 405 "archive and release under striped directory"
5310 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
5312 [ $MDS1_VERSION -lt $(version_code 2.7.64) ] &&
5313 skip "need MDS version at least 2.7.64"
5318 mkdir_on_mdt0 $DIR/$tdir
5320 fid=$(create_small_file $DIR/$tdir/$tfile)
5325 $LFS hsm_archive $DIR/$tdir/$tfile
5326 wait_request_state "$fid" ARCHIVE SUCCEED
5327 $LFS hsm_release $DIR/$tdir/$tfile
5329 # Should migrate $tdir but not $tfile.
5330 $LFS migrate -m1 $DIR/$tdir &&
5331 error "migrating HSM an archived file should fail"
5333 $LFS hsm_restore $DIR/$tdir/$tfile
5334 wait_request_state "$fid" RESTORE SUCCEED
5336 $LFS hsm_remove $DIR/$tdir/$tfile
5337 wait_request_state "$fid" REMOVE SUCCEED
5339 cat $DIR/$tdir/$tfile > /dev/null ||
5340 error "cannot read $DIR/$tdir/$tfile"
5342 $LFS migrate -m1 $DIR/$tdir ||
5343 error "cannot complete migration after HSM remove"
5345 mdt_index=$($LFS getstripe -m $DIR/$tdir)
5346 if ((mdt_index != 1)); then
5347 error "expected MDT index 1, got $mdt_index"
5350 # Refresh fid after migration.
5351 fid=$(path2fid $DIR/$tdir/$tfile)
5354 $LFS hsm_archive $DIR/$tdir/$tfile
5355 wait_request_state "$fid" ARCHIVE SUCCEED 1
5357 lctl set_param debug=+trace
5358 $LFS hsm_release $DIR/$tdir/$tfile ||
5359 error "cannot release $DIR/$tdir/$tfile"
5361 $LFS hsm_restore $DIR/$tdir/$tfile
5362 wait_request_state "$fid" RESTORE SUCCEED 1
5364 cat $DIR/$tdir/$tfile > /dev/null ||
5365 error "cannot read $DIR/$tdir/$tfile"
5367 run_test 406 "attempting to migrate HSM archived files is safe"
5370 mkdir_on_mdt0 $DIR/$tdir
5372 local f=$DIR/$tdir/$tfile
5373 local f2=$DIR2/$tdir/$tfile
5374 local fid=$(create_empty_file "$f")
5378 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
5379 wait_request_state $fid ARCHIVE SUCCEED
5382 #define OBD_FAIL_MDS_HSM_CDT_DELAY 0x164
5383 do_facet $SINGLEMDS $LCTL set_param fail_val=5 fail_loc=0x164
5385 # Prevent restore from completing
5389 # 1st request holds layout lock while appropriate
5390 # RESTORE record is still not added to llog
5394 do_facet $SINGLEMDS "$LCTL get_param $HSM_PARAM.actions"
5395 # after umount hsm_actions->O/x/x log shouldn't have
5396 # double RESTORE records like below
5397 #[0x200000401:0x1:0x0]...0x58d03a0d/0x58d03a0c action=RESTORE...WAITING
5398 #[0x200000401:0x1:0x0]...0x58d03a0c/0x58d03a0d action=RESTORE...WAITING
5400 do_facet $SINGLEMDS "$LCTL get_param $HSM_PARAM.actions"&
5402 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
5404 do_facet $SINGLEMDS "$LCTL get_param $HSM_PARAM.actions"
5407 wait_all_done 100 $fid
5409 run_test 407 "Check for double RESTORE records in llog"
5413 [ "$MDS1_VERSION" -lt $(version_code 2.6.92) ] &&
5414 skip "HSM migrate is not supported"
5416 test_mkdir -p $DIR/$tdir
5418 if [ "$CLIENT_VERSION" -lt $(version_code 2.11.56) ] ||
5419 [ "$MDS1_VERSION" -lt $(version_code 2.11.56) ];
5421 llapi_hsm_test -d $DIR/$tdir -b ||
5422 error "One llapi HSM test failed"
5424 llapi_hsm_test -d $DIR/$tdir ||
5425 error "One llapi HSM test failed"
5428 run_test 500 "various LLAPI HSM tests"
5431 [ "$MDS1_VERSION" -lt $(version_code 2.10.58) ] &&
5432 skip "need MDS version at least 2.10.58"
5436 local f=$DIR/$tdir/$tfile
5439 # set changelog_mask to ALL
5440 changelog_chmask "ALL"
5442 chmod 777 $DIR/$tdir
5443 $RUNAS touch $f || error "touch $f failed as $RUNAS_ID"
5444 local fid=$(path2fid $f)
5447 entry=$(changelog_find -type CREAT -target-fid $fid -uid "$RUNAS_ID" \
5448 -gid "$RUNAS_GID") ||
5449 error "No matching CREAT entry"
5451 # Parse the changelog
5452 eval local -A changelog=$(changelog2array $entry)
5453 local nid="${changelog[nid]}"
5456 echo "Got NID '$nid'"
5457 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5458 error "nid '$nid' does not match any client NID:" \
5461 run_test 600 "Changelog fields 'u=' and 'nid='"
5464 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5465 skip "need MDS version at least 2.10.58"
5469 local f=$DIR/$tdir/$tfile
5472 # set changelog_mask to ALL
5473 changelog_chmask "ALL"
5475 touch $f || error "touch $f failed"
5476 local fid=$(path2fid $f)
5479 cat $f || error "cat $f failed"
5481 changelog_find -type OPEN -target-fid $fid -mode "r--" ||
5482 error "No matching OPEN entry"
5484 run_test 601 "OPEN Changelog entry"
5487 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5488 skip "need MDS version at least 2.10.58"
5490 stack_trap "restore_opencache" EXIT
5495 local f=$DIR/$tdir/$tfile
5498 # set changelog_mask to ALL
5499 changelog_chmask "ALL"
5501 touch $f || error "touch $f failed"
5502 local fid=$(path2fid $f)
5505 cat $f || error "cat $f failed"
5507 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5511 echo f > $f || error "write $f failed"
5514 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5516 # remove OPEN from changelog_mask
5517 changelog_chmask "-OPEN"
5521 cat $f || error "cat $f failed"
5524 changelog_find -type CLOSE -target-fid $fid &&
5525 error "There should be no CLOSE entry"
5529 echo f > $f || error "write $f failed"
5532 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5534 run_test 602 "Changelog record CLOSE only if open+write or OPEN recorded"
5537 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5538 skip "need MDS version at least 2.10.58"
5542 local f=$DIR/$tdir/$tfile
5545 # set changelog_mask to ALL
5546 changelog_chmask "ALL"
5548 touch $f || error "touch $f failed"
5549 local fid=$(path2fid $f)
5551 setfattr -n user.xattr1 -v "value1" $f || error "setfattr $f failed"
5554 getfattr -n user.xattr1 $f || error "getfattr $f failed"
5556 changelog_find -type GXATR -target-fid $fid -xattr "user.xattr1" ||
5557 error "No matching GXATR entry"
5559 run_test 603 "GETXATTR Changelog entry"
5562 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5563 skip "need MDS version at least 2.10.58"
5565 mkdir_on_mdt0 $DIR/$tdir
5567 local f=$DIR/$tdir/$tfile
5568 local f2=$DIR2/$tdir/$tfile
5569 local procname="mdd.$FSNAME-MDT0000.changelog_deniednext"
5571 timeout="$(do_facet mds1 "$LCTL" get_param -n "$procname")"
5572 stack_trap "do_facet mds1 '$LCTL' set_param '$procname=$timeout'" EXIT
5573 do_facet mds1 lctl set_param "$procname=20"
5577 # set changelog_mask to ALL
5578 changelog_chmask "ALL"
5580 touch $f || error "touch $f failed"
5581 local fid=$(path2fid $f)
5587 $RUNAS cat $f2 && error "cat $f2 by user $RUNAS_ID should have failed"
5591 entry=$(changelog_find -type NOPEN -target-fid $fid -uid "$RUNAS_ID" \
5592 -gid "$RUNAS_GID" -mode "r--") ||
5593 error "No matching NOPEN entry"
5595 # Parse the changelog
5596 eval local -A changelog=$(changelog2array $entry)
5597 local nid="${changelog[nid]}"
5600 echo "Got NID '$nid'"
5601 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5602 error "nid '$nid' does not match any client NID:" \
5607 $RUNAS cat $f2 && error "cat $f2 by user $RUNAS_ID should have failed"
5610 changelog_find -type NOPEN -target-fid $fid &&
5611 error "There should be no NOPEN entry"
5613 # Sleep for `changelog_deniednext` seconds
5618 $RUNAS cat $f2 && error "cat $f by user $RUNAS_ID should have failed"
5621 entry=$(changelog_find -type NOPEN -target-fid $fid -uid "$RUNAS_ID" \
5622 -gid "$RUNAS_GID" -mode "r--") ||
5623 error "No matching NOPEN entry"
5625 # Parse the changelog
5626 eval local -A changelog=$(changelog2array $entry)
5627 local nid="${changelog[nid]}"
5630 echo "Got NID '$nid'"
5631 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5632 error "nid '$nid' does not match any client NID:" \
5635 run_test 604 "NOPEN Changelog entry"
5638 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5639 skip "need MDS version at least 2.10.58"
5641 stack_trap "restore_opencache" EXIT
5646 local f=$DIR/$tdir/$tfile
5647 local f2=$DIR2/$tdir/$tfile
5650 # set changelog_mask to ALL
5651 changelog_chmask "ALL"
5653 touch $f || error "touch $f failed"
5654 local fid=$(path2fid $f)
5658 exec 3<> $f || error "open $f failed"
5662 changelog_find -type OPEN -target-fid $fid || error "No OPEN entry"
5666 exec 4<> $f || error "open $f failed"
5669 changelog_find -type OPEN -target-fid $fid &&
5670 error "There should be no OPEN entry"
5672 exec 4>&- || error "close $f failed"
5675 changelog_find -type CLOSE -target-fid $fid &&
5676 error "There should be no CLOSE entry"
5680 # access in rw, so different access mode should generate entries
5681 cat $f || error "cat $f failed"
5684 changelog_find -type OPEN -target-fid $fid || error "No OPEN entry"
5686 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5690 # same access as first one, should not generate new entries
5691 exec 4<> $f || error "open $f failed"
5694 changelog_find -type OPEN -target-fid $fid &&
5695 error "There should be no OPEN entry"
5697 exec 4>&- || error "close $f failed"
5700 changelog_find -type CLOSE -target-fid $fid &&
5701 error "There should be no CLOSE entry"
5705 # access by different user should generate new entries
5706 $RUNAS cat $f || error "cat $f by user $RUNAS_ID failed"
5709 changelog_find -type OPEN -target-fid $fid || error "No OPEN entry"
5711 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5715 exec 3>&- || error "close $f failed"
5718 changelog_find -type CLOSE -target-fid $fid || error "No CLOSE entry"
5720 run_test 605 "Test OPEN and CLOSE rate limit in Changelogs"
5723 [ $MDS1_VERSION -lt $(version_code 2.10.58) ] &&
5724 skip "need MDS version at least 2.10.58"
5726 local llog_reader=$(do_facet mgs "which llog_reader 2> /dev/null")
5727 llog_reader=${llog_reader:-$LUSTRE/utils/llog_reader}
5728 [ -z $(do_facet mgs ls -d $llog_reader 2> /dev/null) ] &&
5729 skip_env "missing llog_reader"
5731 mkdir_on_mdt0 $DIR/$tdir
5733 local f=$DIR/$tdir/$tfile
5736 # set changelog_mask to ALL
5737 changelog_chmask "ALL"
5739 chmod 777 $DIR/$tdir
5740 $RUNAS touch $f || error "touch $f failed as $RUNAS_ID"
5741 local fid=$(path2fid $f)
5742 rm $f || error "rm $f failed"
5744 local mntpt=$(facet_mntpt mds1)
5748 #remount mds1 as ldiskfs or zfs type
5749 stop mds1 || error "stop mds1 failed"
5750 stack_trap "unmount_fstype mds1; start mds1 $(mdsdevname 1)\
5751 $MDS_MOUNT_OPTS" EXIT
5752 mount_fstype mds1 || error "remount mds1 failed"
5754 for ((i = 0; i < 1; i++)); do
5755 do_facet mds1 $llog_reader $mntpt/changelog_catalog
5756 local cat_file=$(do_facet mds1 $llog_reader \
5757 $mntpt/changelog_catalog | awk \
5758 '{match($0,"path=([^ ]+)",a)}END{print a[1]}')
5759 [ -n "$cat_file" ] || error "no catalog file"
5761 entry=$(do_facet mds1 $llog_reader $mntpt/$cat_file |
5762 awk "/CREAT/ && /target:\[$fid\]/ {print}")
5763 [ -n "$entry" ] || error "no CREAT entry"
5766 local uidgid=$(echo $entry |
5767 sed 's+.*\ user:\([0-9][0-9]*:[0-9][0-9]*\)\ .*+\1+')
5768 [ -n "$uidgid" ] || error "uidgid is empty"
5769 echo "Got UID/GID $uidgid"
5770 [ "$uidgid" = "$RUNAS_ID:$RUNAS_GID" ] ||
5771 error "uidgid '$uidgid' != '$RUNAS_ID:$RUNAS_GID'"
5772 local nid=$(echo $entry |
5773 sed 's+.*\ nid:\(\S\S*@\S\S*\)\ .*+\1+')
5774 [ -n "$nid" ] || error "nid is empty"
5776 [ -n "$nid" ] && [[ "${CLIENT_NIDS[*]}" =~ $nid ]] ||
5777 error "nid '$nid' does not match any NID ${CLIENT_NIDS[*]}"
5779 run_test 606 "llog_reader groks changelog fields"
5781 complete_test $SECONDS
5782 check_and_cleanup_lustre