3 # Run select tests by setting ONLY, or as arguments to the script.
4 # Skip specific tests by setting EXCEPT.
11 export PATH=$PWD/$SRCDIR:$SRCDIR:$PWD/$SRCDIR/utils:$PATH:/sbin:/usr/sbin
14 # bug number for skipped test: LU-3815
15 ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36"
16 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
18 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
20 . $LUSTRE/tests/test-framework.sh
22 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
25 MULTIOP=${MULTIOP:-multiop}
26 OPENFILE=${OPENFILE:-openfile}
27 MMAP_CAT=${MMAP_CAT:-mmap_cat}
28 MOUNT_2=${MOUNT_2:-"yes"}
31 # script only handles up to 10 MDTs (because of MDT_PREFIX)
32 [ $MDSCOUNT -gt 9 ] &&
33 error "script cannot handle more than 9 MDTs, please fix" && exit
35 check_and_setup_lustre
37 if [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.53) ]]; then
38 skip_env "Need MDS version at least 2.4.53" && exit
41 # $RUNAS_ID may get set incorrectly somewhere else
42 if [[ $UID -eq 0 && $RUNAS_ID -eq 0 ]]; then
43 skip_env "\$RUNAS_ID set to 0, but \$UID is also 0!" && exit
45 check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS
49 # if there is no CLIENT1 defined, some tests can be ran on localhost
50 CLIENT1=${CLIENT1:-$HOSTNAME}
51 # if CLIENT2 doesn't exist then use CLIENT1 instead
52 # All tests should use CLIENT2 with MOUNT2 only therefore it will work if
54 # Exception is the test which need two separate nodes
55 CLIENT2=${CLIENT2:-$CLIENT1}
58 # In order to test multiple remote HSM agents, a new facet type named "AGT" and
59 # the following associated variables are added:
61 # AGTCOUNT: number of agents
62 # AGTDEV{N}: target HSM mount point (root path of the backend)
63 # agt{N}_HOST: hostname of the agent agt{N}
64 # SINGLEAGT: facet of the single agent
66 # The number of agents is initialized as the number of remote client nodes.
67 # By default, only single copytool is started on a remote client/agent. If there
68 # was no remote client, then the copytool will be started on the local client.
74 export AGTCOUNT=${AGTCOUNT:-$((CLIENTCOUNT - 1))}
75 [[ $AGTCOUNT -gt 0 ]] || AGTCOUNT=1
77 export SHARED_DIRECTORY=${SHARED_DIRECTORY:-$TMP}
78 if [[ $CLIENTCOUNT -gt 1 ]] &&
79 ! check_shared_dir $SHARED_DIRECTORY $CLIENTS; then
80 skip_env "SHARED_DIRECTORY should be accessible"\
85 # We used to put the HSM archive in $SHARED_DIRECTORY but that
86 # meant NFS issues could hose sanity-hsm sessions. So now we
88 for n in $(seq $AGTCOUNT); do
89 eval export AGTDEV$n=\$\{AGTDEV$n:-"$TMP/arc$n"\}
90 agent=CLIENT$((n + 1))
91 if [[ -z "${!agent}" ]]; then
92 [[ $CLIENTCOUNT -eq 1 ]] && agent=CLIENT1 ||
95 eval export agt${n}_HOST=\$\{agt${n}_HOST:-${!agent}\}
98 export SINGLEAGT=${SINGLEAGT:-agt1}
100 export HSMTOOL=${HSMTOOL:-"lhsmtool_posix"}
101 export HSMTOOL_VERBOSE=${HSMTOOL_VERBOSE:-""}
102 export HSMTOOL_UPDATE_INTERVAL=${HSMTOOL_UPDATE_INTERVAL:=""}
103 export HSMTOOL_EVENT_FIFO=${HSMTOOL_EVENT_FIFO:=""}
104 export HSMTOOL_TESTDIR
105 export HSMTOOL_BASE=$(basename "$HSMTOOL" | cut -f1 -d" ")
106 # $hsm_root/$HSMTMP Makes $hsm_root dir path less generic to ensure
107 # rm -rf $hsm_root/* is safe even if $hsm_root becomes unset to avoid
108 # deleting everything in filesystem, independent of any copytool.
109 export HSMTMP=${HSMTMP:-"shsm"}
111 HSM_ARCHIVE=$(copytool_device $SINGLEAGT)
113 [ -z "${HSM_ARCHIVE// /}" ] && error "HSM_ARCHIVE is empty!"
114 HSM_ARCHIVE=$HSM_ARCHIVE/$HSMTMP
118 # The test only support up to 10 MDTs
119 MDT_PREFIX="mdt.$FSNAME-MDT000"
120 HSM_PARAM="${MDT_PREFIX}0.hsm"
122 # archive is purged at copytool setup
123 HSM_ARCHIVE_PURGE=true
125 # Don't allow copytool error upon start/setup
126 HSMTOOL_NOERROR=false
129 # Get the backend root path for the given agent facet.
132 local dev=AGTDEV$(facet_number $facet)
137 # Stop copytool and unregister an existing changelog user.
139 copytool_monitor_cleanup
142 cdt_set_sanity_policy
147 # get MDT device for each mdc
148 for mdtno in $(seq 1 $MDSCOUNT); do
149 local idx=$(($mdtno - 1))
150 MDT[$idx]=$($LCTL get_param -n \
151 mdc.$FSNAME-MDT000${idx}-mdc-*.mds_server_uuid |
152 awk '{gsub(/_UUID/,""); print $1}' | head -n1)
157 local hosts=${1:-$(facet_active_host $SINGLEAGT)}
158 do_nodesv $hosts "pgrep -x $HSMTOOL_BASE"
162 local hosts=${1:-$(facet_active_host $SINGLEAGT)}
164 echo "Killing existing copytools on $hosts"
165 do_nodesv $hosts "killall -q $HSMTOOL_BASE" || true
169 local hosts=${1:-$(facet_active_host $SINGLEAGT)}
170 local wait_timeout=200
171 local wait_start=$SECONDS
172 local wait_end=$((wait_start + wait_timeout))
174 while ((SECONDS < wait_end)); do
176 if ! search_copytools $hosts; then
177 echo "copytools stopped in $((SECONDS - wait_start))s"
181 echo "copytools still running on $hosts"
184 # try to dump Copytool's stack
185 do_nodesv $hosts "echo 1 >/proc/sys/kernel/sysrq ; " \
186 "echo t >/proc/sysrq-trigger"
188 echo "copytools failed to stop in ${wait_timeout}s"
193 copytool_monitor_setup() {
194 local facet=${1:-$SINGLEAGT}
195 local agent=$(facet_active_host $facet)
197 local cmd="mktemp --tmpdir=/tmp -d ${TESTSUITE}.${TESTNAME}.XXXX"
198 local test_dir=$(do_node $agent "$cmd") ||
199 error "Failed to create tempdir on $agent"
200 export HSMTOOL_MONITOR_DIR=$test_dir
202 # Create the fifo and a monitor (cat dies when copytool dies)
203 do_node $agent "mkfifo -m 0644 $test_dir/fifo" ||
204 error "failed to create copytool fifo on $agent"
205 cmd="cat $test_dir/fifo > $test_dir/events &"
206 cmd+=" echo \\\$! > $test_dir/monitor_pid"
208 if [[ $PDSH == *Rmrsh* ]]; then
209 # This is required for pdsh -Rmrsh and its handling of remote
211 # Regular ssh and pdsh -Rssh work fine without this
212 # backgrounded subshell nonsense.
213 (do_node $agent "$cmd") &
214 export HSMTOOL_MONITOR_PDSH=$!
216 # Slightly racy, but just making a best-effort to catch obvious
219 ps -p $HSMTOOL_MONITOR_PDSH > /dev/null ||
220 error "Failed to start copytool monitor on $agent"
222 do_node $agent "$cmd"
224 error "Failed to start copytool monitor on $agent"
229 copytool_monitor_cleanup() {
230 local facet=${1:-$SINGLEAGT}
231 local agent=$(facet_active_host $facet)
233 if [ -n "$HSMTOOL_MONITOR_DIR" ]; then
234 # Should die when the copytool dies, but just in case.
235 local cmd="kill \\\$(cat $HSMTOOL_MONITOR_DIR/monitor_pid)"
236 cmd+=" 2>/dev/null || true"
237 do_node $agent "$cmd"
238 do_node $agent "rm -fr $HSMTOOL_MONITOR_DIR"
239 export HSMTOOL_MONITOR_DIR=
242 # The pdsh should die on its own when the monitor dies. Just
243 # in case, though, try to clean up to avoid any cruft.
244 if [ -n "$HSMTOOL_MONITOR_PDSH" ]; then
245 kill $HSMTOOL_MONITOR_PDSH 2>/dev/null
246 export HSMTOOL_MONITOR_PDSH=
251 local facet=${1:-$SINGLEAGT}
252 # Use MOUNT2 by default if defined
253 local lustre_mntpnt=${2:-${MOUNT2:-$MOUNT}}
255 local hsm_root=${4:-$(copytool_device $facet)}
257 [ -z "${hsm_root// /}" ] && error "copytool_setup: hsm_root empty!"
259 local agent=$(facet_active_host $facet)
261 if [[ -z "$arc_id" ]] &&
262 do_facet $facet "pkill -CONT -x $HSMTOOL_BASE"; then
263 echo "Only wakeup running copytool $facet on $agent"
267 if $HSM_ARCHIVE_PURGE; then
268 echo "Purging archive on $agent"
269 do_facet $facet "rm -rf $hsm_root/$HSMTMP/*"
272 echo "Starting copytool $facet on $agent"
273 do_facet $facet "mkdir -p $hsm_root/$HSMTMP/" ||
274 error "mkdir '$hsm_root/$HSMTMP' failed"
275 # bandwidth is limited to 1MB/s so the copy time is known and
276 # independent of hardware
277 local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon"
278 cmd+=" --hsm-root $hsm_root/$HSMTMP"
279 [[ -z "$arc_id" ]] || cmd+=" --archive $arc_id"
280 [[ -z "$HSMTOOL_UPDATE_INTERVAL" ]] ||
281 cmd+=" --update-interval $HSMTOOL_UPDATE_INTERVAL"
282 [[ -z "$HSMTOOL_EVENT_FIFO" ]] ||
283 cmd+=" --event-fifo $HSMTOOL_EVENT_FIFO"
284 cmd+=" --bandwidth 1 $lustre_mntpnt"
286 # Redirect the standard output and error to a log file which
287 # can be uploaded to Maloo.
288 local prefix=$TESTLOG_PREFIX
289 [[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME
290 local copytool_log=$prefix.copytool${arc_id}_log.$agent.log
292 do_facet $facet "$cmd < /dev/null > $copytool_log 2>&1"
293 if [[ $? != 0 ]]; then
294 [[ $HSMTOOL_NOERROR == true ]] ||
295 error "start copytool $facet on $agent failed"
296 echo "start copytool $facet on $agent failed"
302 get_copytool_event_log() {
303 local facet=${1:-$SINGLEAGT}
304 local agent=$(facet_active_host $facet)
306 [ -z "$HSMTOOL_MONITOR_DIR" ] &&
307 error "Can't get event log: No monitor directory!"
309 do_node $agent "cat $HSMTOOL_MONITOR_DIR/events" ||
310 error "Could not collect event log from $agent"
315 local agt_facet=$SINGLEAGT
316 local agt_hosts=${1:-$(facet_active_host $agt_facet)}
317 local hsm_root=$(copytool_device $agt_facet)
319 [ -z "${hsm_root// /}" ] && error "copytool_cleanup: hsm_root empty!"
326 kill_copytools $agt_hosts
327 wait_copytools $agt_hosts || error "copytools failed to stop"
329 # Clean all CDTs orphans requests from previous tests that
330 # would otherwise need to timeout to clear.
331 for ((i = 0; i < MDSCOUNT; i++)); do
333 param=$(printf 'mdt.%s-MDT%04x.hsm_control' $FSNAME $i)
334 state[$i]=$(do_facet $facet "$LCTL get_param -n $param")
336 # Skip already stopping or stopped CDTs.
337 [[ "${state[$i]}" =~ ^stop ]] && continue
339 do_facet $facet "$LCTL set_param $param=shutdown"
342 for ((i = 0; i < MDSCOUNT; i++)); do
343 # Only check and restore CDTs that we stopped in the first loop.
344 [[ "${state[$i]}" =~ ^stop ]] && continue
347 param=$(printf 'mdt.%s-MDT%04x.hsm_control' $FSNAME $i)
349 wait_result $facet "$LCTL get_param -n $param" stopped 20 ||
350 error "$facet CDT state is not stopped"
352 # Restore old CDT state.
353 do_facet $facet "$LCTL set_param $param=${state[$i]}"
356 for ((i = 0; i < MDSCOUNT; i++)); do
357 # Only check CDTs that we stopped in the first loop.
358 [[ "${state[$i]}" =~ ^stop ]] && continue
361 param=$(printf 'mdt.%s-MDT%04x.hsm_control' $FSNAME $i)
363 # Check that the old CDT state was restored.
364 wait_result $facet "$LCTL get_param -n $param" "${state[$i]}" \
365 20 || error "$facet CDT state is not '${state[$i]}'"
368 if do_facet $agt_facet "df $hsm_root" >/dev/null 2>&1 ; then
369 do_facet $agt_facet "rm -rf $hsm_root/$HSMTMP/*"
374 local agents=${1:-$(facet_active_host $SINGLEAGT)}
376 do_nodesv $agents "pkill -STOP -x $HSMTOOL_BASE" || return 0
377 echo "Copytool is suspended on $agents"
380 copytool_remove_backend() {
382 local be=$(do_facet $SINGLEAGT find $HSM_ARCHIVE -name $fid)
383 echo "Remove from backend: $fid = $be"
384 do_facet $SINGLEAGT rm -f $be
388 do_facet $SINGLEAGT \
389 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
390 --import $1 $2 $MOUNT" ||
391 error "import of $1 to $2 failed"
395 local file=$HSM_ARCHIVE/$1
396 do_facet $SINGLEAGT mkdir -p $(dirname $file)
397 do_facet $SINGLEAGT dd if=/dev/urandom of=$file count=32 bs=1000000 ||
398 file_creation_failure dd $file $?
402 local file=$HSM_ARCHIVE/$2
403 do_facet $SINGLEAGT mkdir -p $(dirname $file)
404 do_facet $SINGLEAGT cp -p $1 $file || error "cannot copy $1 to $file"
413 if [[ "$value" != "" ]]; then
416 for mdtno in $(seq 1 $MDSCOUNT); do
417 local idx=$(($mdtno - 1))
418 local facet=mds${mdtno}
419 # if $arg include -P option, run 1 set_param per MDT on the MGS
420 # else, run set_param on each MDT
421 [[ $arg = *"-P"* ]] && facet=mgs
422 do_facet $facet $LCTL set_param $arg mdt.${MDT[$idx]}.$key$value
423 [[ $? != 0 ]] && rc=1
433 for mdtno in $(seq 1 $MDSCOUNT); do
434 local idx=$(($mdtno - 1))
435 wait_result mds${mdtno} \
436 "$LCTL get_param -n $MDT_PREFIX${idx}.$key" "$target" \
438 error "$key state is not '$target' on mds${mdtno}"
445 for mdtno in $(seq 1 $MDSCOUNT); do
446 local idx=$(($mdtno - 1))
447 local cl_user=$(do_facet mds${mdtno} $LCTL \
448 --device ${MDT[$idx]} \
449 changelog_register -n)
451 do_facet mds${mdtno} lctl set_param \
452 mdd.${MDT[$idx]}.changelog_mask="+hsm"
453 $LFS changelog_clear ${MDT[$idx]} $cl_user 0
457 changelog_cleanup() {
459 for mdtno in $(seq 1 $MDSCOUNT); do
460 local idx=$(($mdtno - 1))
461 [[ -z ${CL_USERS[$idx]} ]] && continue
462 $LFS changelog_clear ${MDT[$idx]} ${CL_USERS[$idx]} 0
463 do_facet mds${mdtno} lctl --device ${MDT[$idx]} \
464 changelog_deregister ${CL_USERS[$idx]}
469 changelog_get_flags() {
474 $LFS changelog $mdt | awk "/$cltype/ && /t=\[$fid\]/ {print \$5}"
479 local val=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.$param)
487 mdts_set_param "$opt -n" "hsm.$param" "$value"
494 mdts_set_param "" hsm_control "$cmd"
495 mdts_check_param hsm_control "$target" 10
498 cdt_set_sanity_policy() {
499 if [[ "$CDT_POLICY_HAD_CHANGED" ]]
502 mdts_set_param "" hsm.policy "+NRA"
503 mdts_set_param "" hsm.policy "-NBR"
504 CDT_POLICY_HAD_CHANGED=
509 mdts_set_param "" hsm.policy "+NRA"
510 CDT_POLICY_HAD_CHANGED=true
513 cdt_clear_no_retry() {
514 mdts_set_param "" hsm.policy "-NRA"
515 CDT_POLICY_HAD_CHANGED=true
518 cdt_set_non_blocking_restore() {
519 mdts_set_param "" hsm.policy "+NBR"
520 CDT_POLICY_HAD_CHANGED=true
523 cdt_clear_non_blocking_restore() {
524 mdts_set_param "" hsm.policy "-NBR"
525 CDT_POLICY_HAD_CHANGED=true
528 cdt_clear_mount_state() {
529 mdts_set_param "-P -d" hsm_control ""
532 cdt_set_mount_state() {
533 mdts_set_param "-P" hsm_control "$1"
534 # set_param -P is asynchronous operation and could race with set_param.
535 # In such case configs could be retrieved and applied at mgc after
536 # set_param -P completion. Sleep here to avoid race with set_param.
537 # We need at least 20 seconds. 10 for mgc_requeue_thread to wake up
538 # MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC(5 + 5)
539 # and 10 seconds to retrieve config from server.
544 mdts_check_param hsm_control "$1" 20
548 set_test_state disabled disabled
552 set_test_state enabled enabled
556 set_test_state shutdown stopped
560 set_test_state purge enabled
566 cdt_set_sanity_policy
570 local client_count=$1
571 if [[ $CLIENTCOUNT -lt $client_count ]]; then
572 skip "Need $client_count or more clients, have $CLIENTCOUNT"
579 $LFS path2fid $1 | tr -d '[]'
580 return ${PIPESTATUS[0]}
588 if [[ $u == "user" ]]; then
589 st=$($RUNAS $LFS hsm_state $f)
592 st=$($LFS hsm_state $f)
595 [[ $? == 0 ]] || error "$LFS hsm_state $f failed (run as $u)"
597 st=$(echo $st | cut -f 2 -d" " | tr -d "()," )
601 get_hsm_archive_id() {
604 st=$($LFS hsm_state $f)
605 [[ $? == 0 ]] || error "$LFS hsm_state $f failed"
607 local ar=$(echo $st | grep "archive_id" | cut -f5 -d" " |
616 local st=$(get_hsm_flags $f)
617 [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
620 check_hsm_flags_user() {
624 local st=$(get_hsm_flags $f user)
625 [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
628 file_creation_failure() {
633 df $MOUNT $MOUNT2 >&2
634 error "cannot create $f with $cmd, status=$err"
646 if [[ "$3" != 1 ]]; then
650 cp $1 $f || file_creation_failure cp $f $?
652 path2fid $f || error "cannot get fid on $f"
656 local file2=${1/$DIR/$DIR2}
657 dd if=/dev/urandom of=$file2 count=2 bs=1M conv=fsync ||
658 file_creation_failure dd $file2 $?
660 path2fid $1 || error "cannot get fid on $1"
664 dd if=/dev/urandom of=$1 count=1 bs=1M conv=sync ||
665 file_creation_failure dd $1 $?
666 path2fid $1 || error "cannot get fid on $1"
669 cleanup_large_files() {
670 local ratio=$(df -P $MOUNT | tail -1 | awk '{print $5}' |
672 [ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \;
675 check_enough_free_space() {
678 local need=$((nb * unit /1024))
679 local free=$(df -kP $MOUNT | tail -1 | awk '{print $4}')
680 (( $need >= $free )) && return 1
684 make_custom_file_for_progress() {
685 local file2=${1/$DIR/$DIR2}
686 local fsize=${2:-"39"}
687 local blksz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1)
690 [[ $fsize -gt 0 ]] || error "Invalid file size"
691 [[ $blksz -gt 0 ]] || error "Invalid stripe size"
694 check_enough_free_space $fsize $blksz
695 [ $? != 0 ] && return $?
696 dd if=/dev/zero of=$file2 count=$fsize bs=$blksz conv=fsync ||
697 file_creation_failure dd $file2 $?
698 path2fid $1 || error "cannot get fid on $1"
704 wait_update --verbose $(facet_active_host $facet) "$@"
707 wait_request_state() {
711 # 4th arg (mdt index) is optional
713 local mds=mds$(($mdtidx + 1))
715 local cmd="$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.actions"
716 cmd+=" | awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
718 wait_result $mds "$cmd" $state 200 ||
719 error "request on $fid is not $state on $mds"
722 get_request_state() {
726 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
727 "awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
730 get_request_count() {
734 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
735 "awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
742 local cmd="$LCTL get_param -n $HSM_PARAM.actions"
743 [[ -n $fid ]] && cmd+=" | grep '$fid'"
744 cmd+=" | egrep 'WAITING|STARTED'"
746 wait_result $SINGLEMDS "$cmd" "" $timeout ||
747 error "requests did not complete"
750 wait_for_grace_delay() {
751 local val=$(get_hsm_param grace_delay)
758 # python2.6 in EL6 includes an internal json module
759 local json_parser='import json; import fileinput;'
760 json_parser+=' print "\n".join(["local %s=\"%s\"" % tuple for tuple in '
761 json_parser+='json.loads([line for line in '
762 json_parser+='fileinput.input()][0]).items()])'
764 echo $raw_event | python -c "$json_parser"
767 get_agent_by_uuid_mdt() {
770 local mds=mds$(($mdtidx + 1))
771 do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
775 check_agent_registered_by_mdt() {
778 local mds=mds$(($mdtidx + 1))
779 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
780 if [[ ! -z "$agent" ]]; then
781 echo "found agent $agent on $mds"
783 error "uuid $uuid not found in agent list on $mds"
787 check_agent_unregistered_by_mdt() {
790 local mds=mds$(($mdtidx + 1))
791 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
792 if [[ -z "$agent" ]]; then
793 echo "uuid not found in agent list on $mds"
795 error "uuid found in agent list on $mds: $agent"
799 check_agent_registered() {
802 for mdsno in $(seq 1 $MDSCOUNT); do
803 check_agent_registered_by_mdt $uuid $((mdsno - 1))
807 check_agent_unregistered() {
810 for mdsno in $(seq 1 $MDSCOUNT); do
811 check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
816 local agent=${1:-$(facet_active_host $SINGLEAGT)}
818 # Lustre mount-point is mandatory and last parameter on
820 local mntpnt=$(do_rpc_nodes $agent ps -C $HSMTOOL_BASE -o args= |
822 [ -n "$mntpnt" ] || error "Found no Agent or with no mount-point "\
824 do_rpc_nodes $agent get_client_uuid $mntpnt | cut -d' ' -f2
830 # populate MDT device array
833 # cleanup from previous bad setup
836 # for recovery tests, coordinator needs to be started at mount
838 # the lustre conf must be without hsm on (like for sanity.sh)
839 echo "Set HSM on and start"
840 cdt_set_mount_state enabled
841 cdt_check_state enabled
843 echo "Start copytool"
846 echo "Set sanity-hsm HSM policy"
847 cdt_set_sanity_policy
849 # finished requests are quickly removed from list
850 set_hsm_param grace_delay 10
856 local f=$DIR/$tdir/$tfile
860 check_hsm_flags_user $f "0x00000000"
862 $RUNAS $LFS hsm_set --norelease $f ||
863 error "user could not change hsm flags"
864 check_hsm_flags_user $f "0x00000010"
866 $RUNAS $LFS hsm_clear --norelease $f ||
867 error "user could not clear hsm flags"
868 check_hsm_flags_user $f "0x00000000"
870 # User could not change those flags...
871 $RUNAS $LFS hsm_set --exists $f &&
872 error "user should not set this flag"
873 check_hsm_flags_user $f "0x00000000"
876 $LFS hsm_set --exists $f ||
877 error "root could not change hsm flags"
878 check_hsm_flags_user $f "0x00000001"
880 $LFS hsm_clear --exists $f ||
881 error "root could not clear hsm state"
882 check_hsm_flags_user $f "0x00000000"
885 run_test 1 "lfs hsm flags root/non-root access"
889 local f=$DIR/$tdir/$tfile
890 local fid=$(make_small $f)
892 $LFS hsm_archive $f || error "could not archive file"
893 wait_request_state $fid ARCHIVE SUCCEED
895 # Release and check states
896 $LFS hsm_release $f || error "could not release file"
897 echo -n "Verifying released state: "
898 check_hsm_flags $f "0x0000000d"
900 $MMAP_CAT $f > /dev/null || error "failed mmap & cat release file"
902 run_test 1a "mmap & cat a HSM released file"
906 local f=$DIR/$tdir/$tfile
908 # New files are not dirty
909 check_hsm_flags $f "0x00000000"
911 # For test, we simulate an archived file.
912 $LFS hsm_set --exists $f || error "user could not change hsm flags"
913 check_hsm_flags $f "0x00000001"
915 # chmod do not put the file dirty
916 chmod 600 $f || error "could not chmod test file"
917 check_hsm_flags $f "0x00000001"
919 # chown do not put the file dirty
920 chown $RUNAS_ID $f || error "could not chown test file"
921 check_hsm_flags $f "0x00000001"
923 # truncate put the file dirty
924 $TRUNCATE $f 1 || error "could not truncate test file"
925 check_hsm_flags $f "0x00000003"
927 $LFS hsm_clear --dirty $f || error "could not clear hsm flags"
928 check_hsm_flags $f "0x00000001"
930 run_test 2 "Check file dirtyness when doing setattr"
936 # New files are not dirty
938 check_hsm_flags $f "0x00000000"
940 # For test, we simulate an archived file.
941 $LFS hsm_set --exists $f ||
942 error "user could not change hsm flags"
943 check_hsm_flags $f "0x00000001"
945 # Reading a file, does not set dirty
946 cat $f > /dev/null || error "could not read file"
947 check_hsm_flags $f "0x00000001"
949 # Open for write without modifying data, does not set dirty
950 openfile -f O_WRONLY $f || error "could not open test file"
951 check_hsm_flags $f "0x00000001"
953 # Append to a file sets it dirty
954 cp -p /etc/passwd $f.append || error "could not create file"
955 $LFS hsm_set --exists $f.append ||
956 error "user could not change hsm flags"
957 dd if=/etc/passwd of=$f.append bs=1 count=3\
958 conv=notrunc oflag=append status=noxfer ||
959 file_creation_failure dd $f.append $?
960 check_hsm_flags $f.append "0x00000003"
962 # Modify a file sets it dirty
963 cp -p /etc/passwd $f.modify || error "could not create file"
964 $LFS hsm_set --exists $f.modify ||
965 error "user could not change hsm flags"
966 dd if=/dev/zero of=$f.modify bs=1 count=3\
967 conv=notrunc status=noxfer ||
968 file_creation_failure dd $f.modify $?
969 check_hsm_flags $f.modify "0x00000003"
971 # Open O_TRUNC sets dirty
972 cp -p /etc/passwd $f.trunc || error "could not create file"
973 $LFS hsm_set --exists $f.trunc ||
974 error "user could not change hsm flags"
975 cp /etc/group $f.trunc || error "could not override a file"
976 check_hsm_flags $f.trunc "0x00000003"
978 # Mmapped a file sets dirty
979 cp -p /etc/passwd $f.mmap || error "could not create file"
980 $LFS hsm_set --exists $f.mmap ||
981 error "user could not change hsm flags"
982 multiop $f.mmap OSMWUc || error "could not mmap a file"
983 check_hsm_flags $f.mmap "0x00000003"
985 run_test 3 "Check file dirtyness when opening for write"
989 local f=$DIR/$tdir/$tfile
990 local fid=$(make_small $f)
993 local st=$(get_request_state $fid CANCEL)
994 [[ -z "$st" ]] || error "hsm_cancel must not be registered (state=$st)"
996 run_test 4 "Useless cancel must not be registered"
999 # test needs a running copytool
1003 local f=$DIR/$tdir/$tfile
1004 local fid=$(copy_file /etc/passwd $f)
1006 wait_request_state $fid ARCHIVE SUCCEED
1008 check_hsm_flags $f "0x00000009"
1012 run_test 8 "Test default archive number"
1015 # we do not use the default one to be sure
1016 local new_an=$((HSM_ARCHIVE_NUMBER + 1))
1018 copytool_setup $SINGLEAGT $MOUNT $new_an
1020 # give time for CT to register with MDTs
1021 sleep $(($MDSCOUNT*2))
1022 local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
1023 check_agent_registered $uuid
1026 local f=$DIR/$tdir/$tfile
1027 local fid=$(copy_file /etc/passwd $f)
1028 $LFS hsm_archive --archive $new_an $f
1029 wait_request_state $fid ARCHIVE SUCCEED
1031 check_hsm_flags $f "0x00000009"
1035 run_test 9 "Use of explicit archive number, with dedicated copytool"
1038 needclients 3 || return 0
1044 copytool_cleanup $(comma_list $(agts_nodes))
1046 # start all of the copytools
1047 for n in $(seq $AGTCOUNT); do
1048 copytool_setup agt$n
1051 trap "copytool_cleanup $(comma_list $(agts_nodes))" EXIT
1054 for n in $(seq $AGTCOUNT); do
1055 file=$DIR/$tdir/$tfile.$n
1056 fid=$(make_small $file)
1058 $LFS hsm_archive $file || error "could not archive file $file"
1059 wait_request_state $fid ARCHIVE SUCCEED
1060 check_hsm_flags $file "0x00000009"
1064 copytool_cleanup $(comma_list $(agts_nodes))
1066 run_test 9a "Multiple remote agents"
1069 # test needs a running copytool
1072 mkdir -p $DIR/$tdir/d1
1073 local f=$DIR/$tdir/$tfile
1074 local fid=$(copy_file /etc/hosts $f)
1075 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
1076 error "hsm_archive failed"
1077 wait_request_state $fid ARCHIVE SUCCEED
1079 local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
1080 error "fid $fid not in archive $HSM_ARCHIVE"
1081 echo "Verifying content"
1082 do_facet $SINGLEAGT diff $f $AFILE || error "archived file differs"
1083 echo "Verifying hsm state "
1084 check_hsm_flags $f "0x00000009"
1086 echo "Verifying archive number is $HSM_ARCHIVE_NUMBER"
1087 local st=$(get_hsm_archive_id $f)
1088 [[ $st == $HSM_ARCHIVE_NUMBER ]] ||
1089 error "Wrong archive number, $st != $HSM_ARCHIVE_NUMBER"
1094 run_test 10a "Archive a file"
1097 # test needs a running copytool
1101 local f=$DIR/$tdir/$tfile
1102 local fid=$(copy_file /etc/hosts $f)
1103 $LFS hsm_archive $f || error "archive request failed"
1104 wait_request_state $fid ARCHIVE SUCCEED
1106 $LFS hsm_archive $f || error "archive of non dirty file failed"
1107 local cnt=$(get_request_count $fid ARCHIVE)
1108 [[ "$cnt" == "1" ]] ||
1109 error "archive of non dirty file must not make a request"
1113 run_test 10b "Archive of non dirty file must work without doing request"
1116 # test needs a running copytool
1120 local f=$DIR/$tdir/$tfile
1121 local fid=$(copy_file /etc/hosts $f)
1122 $LFS hsm_set --noarchive $f
1123 $LFS hsm_archive $f && error "archive a noarchive file must fail"
1127 run_test 10c "Check forbidden archive"
1130 # test needs a running copytool
1134 local f=$DIR/$tdir/$tfile
1135 local fid=$(copy_file /etc/hosts $f)
1136 $LFS hsm_archive $f || error "cannot archive $f"
1137 wait_request_state $fid ARCHIVE SUCCEED
1139 local ar=$(get_hsm_archive_id $f)
1140 local dflt=$(get_hsm_param default_archive_id)
1141 [[ $ar == $dflt ]] ||
1142 error "archived file is not on default archive: $ar != $dflt"
1146 run_test 10d "Archive a file on the default archive id"
1150 copy2archive /etc/hosts $tdir/$tfile
1151 local f=$DIR/$tdir/$tfile
1153 import_file $tdir/$tfile $f
1154 echo -n "Verifying released state: "
1155 check_hsm_flags $f "0x0000000d"
1157 local LSZ=$(stat -c "%s" $f)
1158 local ASZ=$(do_facet $SINGLEAGT stat -c "%s" $HSM_ARCHIVE/$tdir/$tfile)
1160 echo "Verifying imported size $LSZ=$ASZ"
1161 [[ $LSZ -eq $ASZ ]] || error "Incorrect size $LSZ != $ASZ"
1162 echo -n "Verifying released pattern: "
1163 local PTRN=$($GETSTRIPE -L $f)
1165 [[ $PTRN == 80000001 ]] || error "Is not released"
1166 local fid=$(path2fid $f)
1167 echo "Verifying new fid $fid in archive"
1169 local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
1170 error "fid $fid not in archive $HSM_ARCHIVE"
1172 run_test 11a "Import a file"
1175 # test needs a running copytool
1179 local f=$DIR/$tdir/$tfile
1180 local fid=$(copy_file /etc/hosts $f)
1181 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
1182 error "hsm_archive failed"
1183 wait_request_state $fid ARCHIVE SUCCEED
1185 local FILE_HASH=$(md5sum $f)
1190 echo "$FILE_HASH" | md5sum -c
1192 [[ $? -eq 0 ]] || error "Restored file differs"
1196 run_test 11b "Import a deleted file using its FID"
1199 # test needs a running copytool
1203 copy2archive /etc/hosts $tdir/$tfile
1205 local f=$DIR/$tdir/$tfile
1206 import_file $tdir/$tfile $f
1207 local f2=$DIR2/$tdir/$tfile
1208 echo "Verifying released state: "
1209 check_hsm_flags $f2 "0x0000000d"
1211 local fid=$(path2fid $f2)
1212 $LFS hsm_restore $f2
1213 wait_request_state $fid RESTORE SUCCEED
1215 echo "Verifying file state: "
1216 check_hsm_flags $f2 "0x00000009"
1218 do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
1220 [[ $? -eq 0 ]] || error "Restored file differs"
1224 run_test 12a "Restore an imported file explicitly"
1227 # test needs a running copytool
1231 copy2archive /etc/hosts $tdir/$tfile
1233 local f=$DIR/$tdir/$tfile
1234 import_file $tdir/$tfile $f
1235 echo "Verifying released state: "
1236 check_hsm_flags $f "0x0000000d"
1238 cat $f > /dev/null || error "File read failed"
1240 echo "Verifying file state after restore: "
1241 check_hsm_flags $f "0x00000009"
1243 do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
1245 [[ $? -eq 0 ]] || error "Restored file differs"
1249 run_test 12b "Restore an imported file implicitly"
1252 [ "$OSTCOUNT" -lt "2" ] && skip_env "skipping 2-stripe test" && return
1254 # test needs a running copytool
1258 local f=$DIR/$tdir/$tfile
1259 $LFS setstripe -c 2 $f
1261 fid=$(make_custom_file_for_progress $f 5)
1262 [ $? != 0 ] && skip "not enough free space" && return
1264 local FILE_CRC=$(md5sum $f)
1266 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1267 wait_request_state $fid ARCHIVE SUCCEED
1268 $LFS hsm_release $f || error "release $f failed"
1270 echo "$FILE_CRC" | md5sum -c
1272 [[ $? -eq 0 ]] || error "Restored file differs"
1276 run_test 12c "Restore a file with stripe of 2"
1279 # test needs a running copytool
1284 local f=$DIR/$tdir/$tfile
1285 local fid=$(copy_file /etc/hosts $f)
1286 $LFS hsm_restore $f || error "restore of non archived file failed"
1287 local cnt=$(get_request_count $fid RESTORE)
1288 [[ "$cnt" == "0" ]] ||
1289 error "restore non archived must not make a request"
1290 $LFS hsm_archive $f ||
1291 error "archive request failed"
1292 wait_request_state $fid ARCHIVE SUCCEED
1293 $LFS hsm_restore $f ||
1294 error "restore of non released file failed"
1295 local cnt=$(get_request_count $fid RESTORE)
1296 [[ "$cnt" == "0" ]] ||
1297 error "restore a non dirty file must not make a request"
1301 run_test 12d "Restore of a non archived, non released file must work"\
1302 " without doing request"
1305 # test needs a running copytool
1308 mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
1309 local f=$DIR/$tdir/$tfile
1310 local fid=$(copy_file /etc/hosts $f)
1311 $LFS hsm_archive $f || error "archive request failed"
1312 wait_request_state $fid ARCHIVE SUCCEED
1315 cat /etc/hosts >> $f
1319 $LFS hsm_restore $f && error "restore a dirty file must fail"
1323 run_test 12e "Check forbidden restore"
1326 # test needs a running copytool
1330 local f=$DIR/$tdir/$tfile
1331 local fid=$(copy_file /etc/hosts $f)
1333 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1334 wait_request_state $fid ARCHIVE SUCCEED
1335 $LFS hsm_release $f || error "release of $f failed"
1337 wait_request_state $fid RESTORE SUCCEED
1339 echo -n "Verifying file state: "
1340 check_hsm_flags $f "0x00000009"
1342 diff -q /etc/hosts $f
1344 [[ $? -eq 0 ]] || error "Restored file differs"
1348 run_test 12f "Restore a released file explicitly"
1351 # test needs a running copytool
1355 local f=$DIR/$tdir/$tfile
1356 local fid=$(copy_file /etc/hosts $f)
1358 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1359 wait_request_state $fid ARCHIVE SUCCEED
1360 $LFS hsm_release $f || error "release of $f failed"
1362 diff -q /etc/hosts $f
1365 # we check we had a restore done
1366 wait_request_state $fid RESTORE SUCCEED
1368 [[ $st -eq 0 ]] || error "Restored file differs"
1372 run_test 12g "Restore a released file implicitly"
1375 needclients 2 || return 0
1377 # test needs a running copytool
1381 local f=$DIR/$tdir/$tfile
1382 local fid=$(copy_file /etc/hosts $f)
1384 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1385 wait_request_state $fid ARCHIVE SUCCEED
1386 $LFS hsm_release $f || error "release of $f failed"
1388 do_node $CLIENT2 diff -q /etc/hosts $f
1391 # we check we had a restore done
1392 wait_request_state $fid RESTORE SUCCEED
1394 [[ $st -eq 0 ]] || error "Restored file differs"
1398 run_test 12h "Restore a released file implicitly from a second node"
1401 # test needs a running copytool
1405 local f=$DIR/$tdir/$tfile
1406 local fid=$(copy_file /etc/passwd $f)
1407 $LFS hsm_archive $f || error "archive of $f failed"
1408 wait_request_state $fid ARCHIVE SUCCEED
1410 $LFS hsm_release $f || error "release of $f failed"
1414 [[ $? -eq 0 ]] || error "Restored file differs"
1418 run_test 12m "Archive/release/implicit restore"
1421 # test needs a running copytool
1425 copy2archive /etc/hosts $tdir/$tfile
1427 local f=$DIR/$tdir/$tfile
1428 import_file $tdir/$tfile $f
1430 do_facet $SINGLEAGT cmp /etc/hosts $f ||
1431 error "Restored file differs"
1433 $LFS hsm_release $f || error "release of $f failed"
1437 run_test 12n "Import/implicit restore/release"
1440 # test needs a running copytool
1444 local f=$DIR/$tdir/$tfile
1445 local fid=$(copy_file /etc/hosts $f)
1447 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1448 wait_request_state $fid ARCHIVE SUCCEED
1449 $LFS hsm_release $f || error "release of $f failed"
1451 #define OBD_FAIL_MDS_HSM_SWAP_LAYOUTS 0x152
1452 do_facet $SINGLEMDS lctl set_param fail_loc=0x152
1454 # set no retry action mode
1457 diff -q /etc/hosts $f
1460 # we check we had a restore failure
1461 wait_request_state $fid RESTORE FAILED
1463 [[ $st -eq 0 ]] && error "Restore must fail"
1465 # remove no retry action mode
1468 # check file is still released
1469 check_hsm_flags $f "0x0000000d"
1471 # retry w/o failure injection
1472 do_facet $SINGLEMDS lctl set_param fail_loc=0
1474 # to be sure previous RESTORE result is gone
1476 wait_for_grace_delay
1478 diff -q /etc/hosts $f
1481 # we check we had a restore done
1482 wait_request_state $fid RESTORE SUCCEED
1484 [[ $st -eq 0 ]] || error "Restored file differs"
1488 run_test 12o "Layout-swap failure during Restore leaves file released"
1491 # test needs a running copytool
1495 local f=$DIR/$tdir/$tfile
1496 local fid=$(copy_file /etc/hosts $f)
1498 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1499 wait_request_state $fid ARCHIVE SUCCEED
1500 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1501 $LFS hsm_release $f || error "cannot release $f"
1502 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1503 $LFS hsm_release $f || error "cannot release $f"
1504 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1508 run_test 12p "implicit restore of a file on copytool mount point"
1510 cleanup_test_12q() {
1512 zconf_umount $(facet_host $SINGLEAGT) $MOUNT3 ||
1513 error "cannot umount $MOUNT3 on $SINGLEAGT"
1517 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.58) ] &&
1518 skip "need MDS version at least 2.7.58" && return 0
1520 zconf_mount $(facet_host $SINGLEAGT) $MOUNT3 ||
1521 error "cannot mount $MOUNT3 on $SINGLEAGT"
1523 trap cleanup_test_12q EXIT
1525 # test needs a running copytool
1526 copytool_setup $SINGLEAGT $MOUNT3
1529 local f=$DIR/$tdir/$tfile
1530 local f2=$DIR2/$tdir/$tfile
1531 local fid=$(make_small $f)
1532 local orig_size=$(stat -c "%s" $f)
1534 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1535 wait_request_state $fid ARCHIVE SUCCEED
1537 $LFS hsm_release $f || error "could not release file"
1538 check_hsm_flags $f "0x0000000d"
1541 wait_copytools || error "copytool failed to stop"
1543 cat $f > /dev/null &
1545 # wait a bit to allow implicit restore request to be handled.
1546 # if not, next stat would also block on layout-lock.
1549 local size=$(stat -c "%s" $f2)
1550 [ $size -eq $orig_size ] ||
1551 error "$f2: wrong size after archive: $size != $orig_size"
1553 HSM_ARCHIVE_PURGE=false copytool_setup $SINGLEAGT /mnt/lustre3
1557 size=$(stat -c "%s" $f)
1558 [ $size -eq $orig_size ] ||
1559 error "$f: wrong size after restore: $size != $orig_size"
1561 size=$(stat -c "%s" $f2)
1562 [ $size -eq $orig_size ] ||
1563 error "$f2: wrong size after restore: $size != $orig_size"
1567 size=$(stat -c "%s" $f)
1569 error "$f: wrong size after overwrite: $size != 0"
1571 size=$(stat -c "%s" $f2)
1573 error "$f2: wrong size after overwrite: $size != 0"
1576 zconf_umount $(facet_host $SINGLEAGT) $MOUNT3 ||
1577 error "cannot umount $MOUNT3 on $SINGLEAGT"
1579 run_test 12q "file attributes are refreshed after restore"
1582 # test needs a running copytool
1585 local ARC_SUBDIR="import.orig"
1589 # populate directory to be imported
1590 for d in $(seq 1 10); do
1591 local CURR_DIR="$HSM_ARCHIVE/$ARC_SUBDIR/dir.$d"
1592 do_facet $SINGLEAGT mkdir -p "$CURR_DIR"
1593 for f in $(seq 1 10); do
1594 CURR_FILE="$CURR_DIR/$tfile.$f"
1595 # write file-specific data
1596 do_facet $SINGLEAGT \
1597 "echo d=$d, f=$f, dir=$CURR_DIR, "\
1598 "file=$CURR_FILE > $CURR_FILE"
1602 import_file "$ARC_SUBDIR" $DIR/$tdir
1603 # diff lustre content and origin (triggers file restoration)
1604 # there must be 10x10 identical files, and no difference
1605 local cnt_ok=$(do_facet $SINGLEAGT diff -rs $HSM_ARCHIVE/$ARC_SUBDIR \
1606 $DIR/$tdir/$ARC_SUBDIR | grep identical | wc -l)
1607 local cnt_diff=$(do_facet $SINGLEAGT diff -r $HSM_ARCHIVE/$ARC_SUBDIR \
1608 $DIR/$tdir/$ARC_SUBDIR | wc -l)
1610 [ $cnt_diff -eq 0 ] ||
1611 error "$cnt_diff imported files differ from read data"
1612 [ $cnt_ok -eq 100 ] ||
1613 error "not enough identical files ($cnt_ok != 100)"
1617 run_test 13 "Recursively import and restore a directory"
1620 # test needs a running copytool
1625 local f=$DIR/$tdir/$tfile
1626 local fid=$(make_small $f)
1627 local sum=$(md5sum $f | awk '{print $1}')
1628 $LFS hsm_archive $f || error "could not archive file"
1629 wait_request_state $fid ARCHIVE SUCCEED
1633 # create released file (simulate llapi_hsm_import call)
1635 local fid2=$(path2fid $f)
1636 $LFS hsm_set --archived --exists $f || error "could not force hsm flags"
1637 $LFS hsm_release $f || error "could not release file"
1639 # rebind the archive to the newly created file
1640 echo "rebind $fid to $fid2"
1642 do_facet $SINGLEAGT \
1643 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
1644 --rebind $fid $fid2 $DIR" || error "could not rebind file"
1646 # restore file and compare md5sum
1647 local sum2=$(md5sum $f | awk '{print $1}')
1649 [[ $sum == $sum2 ]] || error "md5sum mismatch after restore"
1653 run_test 14 "Rebind archived file to a new fid"
1656 # test needs a running copytool
1661 local f=$DIR/$tdir/$tfile
1663 local tmpfile=$SHARED_DIRECTORY/tmp.$$
1667 for i in $(seq 1 $count); do
1668 fids[$i]=$(make_small $f.$i)
1669 sums[$i]=$(md5sum $f.$i | awk '{print $1}')
1670 $LFS hsm_archive $f.$i || error "could not archive file"
1672 wait_all_done $(($count*60))
1676 for i in $(seq 1 $count); do
1679 local fid2=$(path2fid $f.$i)
1680 # add the rebind operation to the list
1681 echo ${fids[$i]} $fid2 >> $tmpfile
1683 # set it released (simulate llapi_hsm_import call)
1684 $LFS hsm_set --archived --exists $f.$i ||
1685 error "could not force hsm flags"
1686 $LFS hsm_release $f.$i || error "could not release file"
1688 nl=$(wc -l < $tmpfile)
1689 [[ $nl == $count ]] || error "$nl files in list, $count expected"
1691 echo "rebind list of files"
1692 do_facet $SINGLEAGT \
1693 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
1694 --rebind $tmpfile $DIR" || error "could not rebind file list"
1696 # restore files and compare md5sum
1697 for i in $(seq 1 $count); do
1698 local sum2=$(md5sum $f.$i | awk '{print $1}')
1699 [[ $sum2 == ${sums[$i]} ]] ||
1700 error "md5sum mismatch after restore ($sum2 != ${sums[$i]})"
1706 run_test 15 "Rebind a list of files"
1709 # test needs a running copytool
1713 # create a known size file so we can verify transfer speed
1716 dd if=/dev/zero of=$ref bs=1M count=20
1719 local f=$DIR/$tdir/$tfile
1720 local fid=$(copy_file $ref $f)
1722 local start=$(date +%s)
1724 wait_request_state $fid ARCHIVE SUCCEED
1725 local end=$(date +%s)
1726 # Add 1 to account for rounding errors between start and end (LU-8155)
1727 local duration=$((end - start + 1))
1729 [[ $duration -ge $goal ]] ||
1730 error "Transfer is too fast $duration < $goal"
1734 run_test 16 "Test CT bandwith control option"
1739 local f=$DIR/$tdir/$tfile
1740 touch $f || error "touch $f failed"
1742 # Could not release a non-archived file
1743 $LFS hsm_release $f && error "release should not succeed"
1745 # For following tests, we must test them with HS_ARCHIVED set
1746 $LFS hsm_set --exists --archived $f || error "could not add flag"
1748 # Could not release a file if no-release is set
1749 $LFS hsm_set --norelease $f || error "could not add flag"
1750 $LFS hsm_release $f && error "release should not succeed"
1751 $LFS hsm_clear --norelease $f || error "could not remove flag"
1753 # Could not release a file if lost
1754 $LFS hsm_set --lost $f || error "could not add flag"
1755 $LFS hsm_release $f && error "release should not succeed"
1756 $LFS hsm_clear --lost $f || error "could not remove flag"
1758 # Could not release a file if dirty
1759 $LFS hsm_set --dirty $f || error "could not add flag"
1760 $LFS hsm_release $f && error "release should not succeed"
1761 $LFS hsm_clear --dirty $f || error "could not remove flag"
1763 run_test 20 "Release is not permitted"
1766 # test needs a running copytool
1770 local f=$DIR/$tdir/test_release
1772 # Create a file and check its states
1773 local fid=$(make_small $f)
1774 check_hsm_flags $f "0x00000000"
1776 # LU-4388/LU-4389 - ZFS does not report full number of blocks
1777 # used until file is flushed to disk
1778 if [ $(facet_fstype ost1) == "zfs" ]; then
1779 # this causes an OST_SYNC rpc to be sent
1780 dd if=/dev/zero of=$f bs=512 count=1 oflag=sync conv=notrunc,fsync
1781 # clear locks to reread file data
1782 cancel_lru_locks osc
1785 local orig_size=$(stat -c "%s" $f)
1786 local orig_blocks=$(stat -c "%b" $f)
1788 start_full_debug_logging
1790 $LFS hsm_archive $f || error "could not archive file"
1791 wait_request_state $fid ARCHIVE SUCCEED
1793 local blocks=$(stat -c "%b" $f)
1794 [ $blocks -eq $orig_blocks ] ||
1795 error "$f: wrong block number after archive: " \
1796 "$blocks != $orig_blocks"
1797 local size=$(stat -c "%s" $f)
1798 [ $size -eq $orig_size ] ||
1799 error "$f: wrong size after archive: $size != $orig_size"
1801 # Release and check states
1802 $LFS hsm_release $f || error "could not release file"
1803 check_hsm_flags $f "0x0000000d"
1805 blocks=$(stat -c "%b" $f)
1806 [ $blocks -gt 5 ] &&
1807 error "$f: too many blocks after release: $blocks > 5"
1808 size=$(stat -c "%s" $f)
1809 [ $size -ne $orig_size ] &&
1810 error "$f: wrong size after release: $size != $orig_size"
1812 # Check we can release an file without stripe info
1816 check_hsm_flags $f "0x00000000"
1817 $LFS hsm_archive $f || error "could not archive file"
1818 wait_request_state $fid ARCHIVE SUCCEED
1820 # Release and check states
1821 $LFS hsm_release $f || error "could not release file"
1822 check_hsm_flags $f "0x0000000d"
1824 # Release again a file that is already released is OK
1825 $LFS hsm_release $f || fail "second release should succeed"
1826 check_hsm_flags $f "0x0000000d"
1828 stop_full_debug_logging
1832 run_test 21 "Simple release tests"
1835 # test needs a running copytool
1840 local f=$DIR/$tdir/test_release
1841 local swap=$DIR/$tdir/test_swap
1843 # Create a file and check its states
1844 local fid=$(make_small $f)
1845 check_hsm_flags $f "0x00000000"
1847 $LFS hsm_archive $f || error "could not archive file"
1848 wait_request_state $fid ARCHIVE SUCCEED
1850 # Release and check states
1851 $LFS hsm_release $f || error "could not release file"
1852 check_hsm_flags $f "0x0000000d"
1855 $LFS swap_layouts $swap $f && error "swap_layouts should failed"
1860 run_test 22 "Could not swap a release file"
1863 # test needs a running copytool
1868 local f=$DIR/$tdir/test_mtime
1870 # Create a file and check its states
1871 local fid=$(make_small $f)
1872 check_hsm_flags $f "0x00000000"
1874 $LFS hsm_archive $f || error "could not archive file"
1875 wait_request_state $fid ARCHIVE SUCCEED
1877 # Set modification time in the past
1878 touch -m -a -d @978261179 $f
1880 # Release and check states
1881 $LFS hsm_release $f || error "could not release file"
1882 check_hsm_flags $f "0x0000000d"
1884 local MTIME=$(stat -c "%Y" $f)
1885 local ATIME=$(stat -c "%X" $f)
1886 [ $MTIME -eq "978261179" ] || fail "bad mtime: $MTIME"
1887 [ $ATIME -eq "978261179" ] || fail "bad atime: $ATIME"
1891 run_test 23 "Release does not change a/mtime (utime)"
1894 local file=$DIR/$tdir/$tfile
1903 # test needs a running copytool
1908 fid=$(make_small $file)
1910 # Create a file and check its states
1911 check_hsm_flags $file "0x00000000"
1913 # Ensure atime is less than mtime and ctime.
1917 atime0=$(stat -c "%X" $file)
1918 mtime0=$(stat -c "%Y" $file)
1919 ctime0=$(stat -c "%Z" $file)
1921 [ $atime0 -lt $mtime0 ] ||
1922 error "atime $atime0 is not less than mtime $mtime0"
1924 [ $atime0 -lt $ctime0 ] ||
1925 error "atime $atime0 is not less than ctime $ctime0"
1927 # Archive should not change any timestamps.
1928 $LFS hsm_archive $file || error "cannot archive '$file'"
1929 wait_request_state $fid ARCHIVE SUCCEED
1931 atime1=$(stat -c "%X" $file)
1932 mtime1=$(stat -c "%Y" $file)
1933 ctime1=$(stat -c "%Z" $file)
1935 [ $atime0 -eq $atime1 ] ||
1936 error "archive changed atime from $atime0 to $atime1"
1938 [ $mtime0 -eq $mtime1 ] ||
1939 error "archive changed mtime from $mtime0 to $mtime1"
1941 [ $ctime0 -eq $ctime1 ] ||
1942 error "archive changed ctime from $ctime0 to $ctime1"
1944 # Release should not change any timestamps.
1945 $LFS hsm_release $file || error "cannot release '$file'"
1946 check_hsm_flags $file "0x0000000d"
1948 atime1=$(stat -c "%X" $file)
1949 mtime1=$(stat -c "%Y" $file)
1950 ctime1=$(stat -c "%Z" $file)
1952 [ $atime0 -eq $atime1 ] ||
1953 error "release changed atime from $atime0 to $atime1"
1955 [ $mtime0 -eq $mtime1 ] ||
1956 error "release changed mtime from $mtime0 to $mtime1"
1958 [ $ctime0 -eq $ctime1 ] ||
1959 error "release changed ctime from $ctime0 to $ctime1"
1961 # Restore should not change any timestamps.
1962 $LFS hsm_restore $file
1963 wait_request_state $fid RESTORE SUCCEED
1965 atime1=$(stat -c "%X" $file)
1966 mtime1=$(stat -c "%Y" $file)
1967 ctime1=$(stat -c "%Z" $file)
1969 [ $atime0 -eq $atime1 ] ||
1970 error "restore changed atime from $atime0 to $atime1"
1972 [ $mtime0 -eq $mtime1 ] ||
1973 error "restore changed mtime from $mtime0 to $mtime1"
1975 [ $ctime0 -eq $ctime1 ] ||
1976 error "restore changed ctime from $ctime0 to $ctime1"
1980 # Once more, after unmount and mount.
1981 umount_client $MOUNT || error "cannot unmount '$MOUNT'"
1982 mount_client $MOUNT || error "cannot mount '$MOUNT'"
1984 atime1=$(stat -c "%X" $file)
1985 mtime1=$(stat -c "%Y" $file)
1986 ctime1=$(stat -c "%Z" $file)
1988 [ $atime0 -eq $atime1 ] ||
1989 error "remount changed atime from $atime0 to $atime1"
1991 [ $mtime0 -eq $mtime1 ] ||
1992 error "remount changed mtime from $mtime0 to $mtime1"
1994 [ $ctime0 -eq $ctime1 ] ||
1995 error "remount changed ctime from $ctime0 to $ctime1"
1997 run_test 24a "Archive, release, and restore does not change a/mtime (i/o)"
2000 local file=$DIR/$tdir/$tfile
2006 # Test needs a running copytool.
2010 # Check that root can do HSM actions on a regular user's file.
2012 fid=$(make_small $file)
2013 sum0=$(md5sum $file)
2015 chown $RUNAS_ID:$RUNAS_GID $file ||
2016 error "cannot chown '$file' to '$RUNAS_ID'"
2018 chmod ugo-w $DIR/$tdir ||
2019 error "cannot chmod '$DIR/$tdir'"
2021 $LFS hsm_archive $file
2022 wait_request_state $fid ARCHIVE SUCCEED
2024 $LFS hsm_release $file
2025 check_hsm_flags $file "0x0000000d"
2027 $LFS hsm_restore $file
2028 wait_request_state $fid RESTORE SUCCEED
2030 # Check that ordinary user can get HSM state.
2031 $RUNAS $LFS hsm_state $file ||
2032 error "user '$RUNAS_ID' cannot get HSM state of '$file'"
2034 $LFS hsm_release $file
2035 check_hsm_flags $file "0x0000000d"
2037 # Check that ordinary user can accessed released file.
2038 sum1=$($RUNAS md5sum $file) ||
2039 error "user '$RUNAS_ID' cannot read '$file'"
2041 [ "$sum0" == "$sum1" ] ||
2042 error "md5sum mismatch for '$file'"
2046 run_test 24b "root can archive, release, and restore user files"
2048 cleanup_test_24c() {
2050 set_hsm_param user_request_mask RESTORE
2051 set_hsm_param group_request_mask RESTORE
2052 set_hsm_param other_request_mask RESTORE
2056 local file=$DIR/$tdir/$tfile
2057 local action=archive
2062 # test needs a running copytool
2067 # Save the default masks and check that cleanup_24c will
2068 # restore the request masks correctly.
2069 user_save=$(get_hsm_param user_request_mask)
2070 group_save=$(get_hsm_param group_request_mask)
2071 other_save=$(get_hsm_param other_request_mask)
2073 [ "$user_save" == RESTORE ] ||
2074 error "user_request_mask is '$user_save' expected 'RESTORE'"
2075 [ "$group_save" == RESTORE ] ||
2076 error "group_request_mask is '$group_save' expected 'RESTORE'"
2077 [ "$other_save" == RESTORE ] ||
2078 error "other_request_mask is '$other_save' expected 'RESTORE'"
2080 trap cleanup_test_24c EXIT
2085 chown $RUNAS_ID:nobody $file ||
2086 error "cannot chown '$file' to '$RUNAS_ID:nobody'"
2088 $RUNAS $LFS hsm_$action $file &&
2089 error "$action by user should fail"
2091 set_hsm_param user_request_mask $action
2092 $RUNAS $LFS hsm_$action $file ||
2093 error "$action by user should succeed"
2098 chown nobody:$RUNAS_GID $file ||
2099 error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
2101 $RUNAS $LFS hsm_$action $file &&
2102 error "$action by group should fail"
2104 set_hsm_param group_request_mask $action
2105 $RUNAS $LFS hsm_$action $file ||
2106 error "$action by group should succeed"
2111 chown nobody:nobody $file ||
2112 error "cannot chown '$file' to 'nobody:nobody'"
2114 $RUNAS $LFS hsm_$action $file &&
2115 error "$action by other should fail"
2117 set_hsm_param other_request_mask $action
2118 $RUNAS $LFS hsm_$action $file ||
2119 error "$action by other should succeed"
2124 run_test 24c "check that user,group,other request masks work"
2126 cleanup_test_24d() {
2128 mount -o remount,rw $MOUNT2
2132 local file1=$DIR/$tdir/$tfile
2133 local file2=$DIR2/$tdir/$tfile
2141 fid1=$(make_small $file1)
2143 trap cleanup_test_24d EXIT
2145 mount -o remount,ro $MOUNT2
2147 fid2=$(path2fid $file2)
2148 [ "$fid1" == "$fid2" ] ||
2149 error "FID mismatch '$fid1' != '$fid2'"
2151 $LFS hsm_archive $file2 &&
2152 error "archive should fail on read-only mount"
2153 check_hsm_flags $file1 "0x00000000"
2155 $LFS hsm_archive $file1
2156 wait_request_state $fid1 ARCHIVE SUCCEED
2158 $LFS hsm_release $file1
2159 $LFS hsm_restore $file2
2160 wait_request_state $fid1 RESTORE SUCCEED
2162 $LFS hsm_release $file1 || error "cannot release '$file1'"
2163 dd if=$file2 of=/dev/null bs=1M || error "cannot read '$file2'"
2165 $LFS hsm_release $file2 &&
2166 error "release should fail on read-only mount"
2171 run_test 24d "check that read-only mounts are respected"
2178 local f=$DIR/$tdir/$tfile
2181 fid=$(make_small $f) || error "cannot create $f"
2182 $LFS hsm_archive $f || error "cannot archive $f"
2183 wait_request_state $fid ARCHIVE SUCCEED
2184 $LFS hsm_release $f || error "cannot release $f"
2185 while ! $LFS hsm_state $f | grep released; do
2189 tar -cf $TMP/$tfile.tar $DIR/$tdir || error "cannot tar $DIR/$tdir"
2193 run_test 24e "tar succeeds on HSM released files" # LU-6213
2197 # test needs a running copytool
2200 mkdir -p $DIR/$tdir/d1
2201 local f=$DIR/$tdir/$tfile
2202 local fid=$(copy_file /etc/hosts $f)
2205 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
2206 error "hsm_archive failed"
2207 wait_request_state $fid ARCHIVE SUCCEED
2208 $LFS hsm_release $f || error "cannot release $f"
2209 tar --xattrs -cvf $f.tar -C $DIR/$tdir $tfile
2212 tar --xattrs -xvf $f.tar -C $DIR/$tdir ||
2213 error "Can not recover the tar contents"
2215 echo "Sum0 = $sum0, sum1 = $sum1"
2216 [ "$sum0" == "$sum1" ] || error "md5sum mismatch for '$tfile'"
2220 run_test 24f "root can archive, release, and restore tar files"
2223 # test needs a running copytool
2227 copy2archive /etc/hosts $tdir/$tfile
2229 local f=$DIR/$tdir/$tfile
2231 import_file $tdir/$tfile $f
2233 $LFS hsm_set --lost $f
2238 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2242 run_test 25a "Restore lost file (HS_LOST flag) from import"\
2243 " (Operation not permitted)"
2246 # test needs a running copytool
2251 local f=$DIR/$tdir/$tfile
2252 local fid=$(copy_file /etc/passwd $f)
2254 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2255 wait_request_state $fid ARCHIVE SUCCEED
2258 $LFS hsm_set --lost $f
2262 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2266 run_test 25b "Restore lost file (HS_LOST flag) after release"\
2267 " (Operation not permitted)"
2270 # test needs a running copytool
2274 local f=$DIR/$tdir/$tfile
2276 fid=$(make_custom_file_for_progress $f 39 1000000)
2277 [ $? != 0 ] && skip "not enough free space" && return
2279 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2280 wait_request_state $fid ARCHIVE SUCCEED
2283 wait_request_state $fid REMOVE SUCCEED
2285 check_hsm_flags $f "0x00000000"
2289 run_test 26 "Remove the archive of a valid file"
2292 # test needs a running copytool
2296 make_archive $tdir/$tfile
2297 local f=$DIR/$tdir/$tfile
2298 import_file $tdir/$tfile $f
2299 local fid=$(path2fid $f)
2303 [[ $? != 0 ]] || error "Remove of a released file should fail"
2307 run_test 27a "Remove the archive of an imported file (Operation not permitted)"
2310 # test needs a running copytool
2314 local f=$DIR/$tdir/$tfile
2316 fid=$(make_custom_file_for_progress $f 39 1000000)
2317 [ $? != 0 ] && skip "not enough free space" && return
2319 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2320 wait_request_state $fid ARCHIVE SUCCEED
2325 [[ $? != 0 ]] || error "Remove of a released file should fail"
2329 run_test 27b "Remove the archive of a relased file (Operation not permitted)"
2332 # test needs a running copytool
2336 local f=$DIR/$tdir/$tfile
2338 fid=$(make_custom_file_for_progress $f 39 1000000)
2339 [ $? != 0 ] && skip "not enough free space" && return
2341 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2342 wait_request_state $fid ARCHIVE SUCCEED
2351 wait_request_state $fid REMOVE SUCCEED
2355 run_test 28 "Concurrent archive/file remove"
2358 # Tests --mntpath and --archive options
2361 copytool_setup $SINGLEAGT $MOUNT $archive_id
2363 # Bad archive number
2364 $LFS hsm_remove -m $MOUNT -a 33 0x857765760:0x8:0x2 2>&1 |
2365 grep "Invalid argument" ||
2366 error "unexpected hsm_remove failure (1)"
2368 # mntpath is present but file is given
2369 $LFS hsm_remove --mntpath $MOUNT --archive 30 /qwerty/uyt 2>&1 |
2370 grep "hsm: '/qwerty/uyt' is not a valid FID" ||
2371 error "unexpected hsm_remove failure (2)"
2375 run_test 29a "Tests --mntpath and --archive options"
2378 # test needs a running copytool
2382 local f=$DIR/$tdir/$tfile
2383 local fid=$(make_small $f)
2385 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2386 wait_request_state $fid ARCHIVE SUCCEED
2390 $LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER $fid
2391 wait_request_state $fid REMOVE SUCCEED
2395 run_test 29b "Archive/delete/remove by FID from the archive."
2398 # test needs a running copytool
2402 local fid1=$(make_small $DIR/$tdir/$tfile-1)
2403 local fid2=$(make_small $DIR/$tdir/$tfile-2)
2404 local fid3=$(make_small $DIR/$tdir/$tfile-3)
2406 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/$tfile-[1-3]
2407 wait_request_state $fid1 ARCHIVE SUCCEED
2408 wait_request_state $fid2 ARCHIVE SUCCEED
2409 wait_request_state $fid3 ARCHIVE SUCCEED
2411 rm -f $DIR/$tdir/$tfile-[1-3]
2413 echo $fid1 > $DIR/$tdir/list
2414 echo $fid2 >> $DIR/$tdir/list
2415 echo $fid3 >> $DIR/$tdir/list
2417 $LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER \
2418 --filelist $DIR/$tdir/list
2419 wait_request_state $fid1 REMOVE SUCCEED
2420 wait_request_state $fid2 REMOVE SUCCEED
2421 wait_request_state $fid3 REMOVE SUCCEED
2425 run_test 29c "Archive/delete/remove by FID, using a file list."
2428 # restore at exec cannot work on agent node (because of Linux kernel
2429 # protection of executables)
2430 needclients 2 || return 0
2432 # test needs a running copytool
2436 copy2archive /bin/true $tdir/$tfile
2438 local f=$DIR/$tdir/true
2439 import_file $tdir/$tfile $f
2441 local fid=$(path2fid $f)
2443 # set no retry action mode
2449 # remove no try action mode
2453 [[ $st == 0 ]] || error "Failed to exec a released file"
2457 run_test 30a "Restore at exec (import case)"
2460 # restore at exec cannot work on agent node (because of Linux kernel
2461 # protection of executables)
2462 needclients 2 || return 0
2464 # test needs a running copytool
2468 local f=$DIR/$tdir/true
2469 local fid=$(copy_file /bin/true $f)
2471 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2472 wait_request_state $fid ARCHIVE SUCCEED
2475 # set no retry action mode
2481 # remove no try action mode
2485 [[ $st == 0 ]] || error "Failed to exec a released file"
2489 run_test 30b "Restore at exec (release case)"
2492 needclients 2 || return 0
2494 # test needs a running copytool
2498 local f=$DIR/$tdir/SLEEP
2499 local slp_sum1=$(md5sum /bin/sleep)
2500 local fid=$(copy_file /bin/sleep $f)
2502 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2503 wait_request_state $fid ARCHIVE SUCCEED
2505 check_hsm_flags $f "0x0000000d"
2506 # set no retry action mode
2508 do_node $CLIENT2 "$f 10" &
2512 [[ $? == 0 ]] && error "Update during exec of released file must fail"
2514 [[ $? == 0 ]] || error "Execution failed during run"
2516 if [[ $? != 0 ]]; then
2517 local slp_sum2=$(md5sum /bin/sleep)
2518 # in case sleep file is modified during the test
2519 [[ $slp_sum1 == $slp_sum2 ]] &&
2520 error "Binary overwritten during exec"
2524 # remove no try action mode
2526 check_hsm_flags $f "0x00000009"
2530 run_test 30c "Update during exec of released file must fail"
2532 restore_and_check_size() {
2535 local s=$(stat -c "%s" $f)
2537 local st=$(get_hsm_flags $f)
2541 while [[ "$st" != "0x00000009" && $cpt -le 10 ]]
2543 n=$(stat -c "%s" $f)
2544 # we echo in both cases to show stat is not
2546 if [[ $n != $s ]]; then
2547 echo "size seen is $n != $s"
2550 echo "size seen is right: $n == $s"
2552 st=$(get_hsm_flags $f)
2556 if [[ $cpt -lt 10 ]]; then
2559 echo " restore is too long"
2560 wait_request_state $fid RESTORE SUCCEED
2566 # test needs a running copytool
2571 make_archive $tdir/$tfile
2572 local f=$DIR/$tdir/$tfile
2573 import_file $tdir/$tfile $f
2574 local fid=$($LFS path2fid $f)
2575 HSM_ARCHIVE_PURGE=false copytool_setup
2577 restore_and_check_size $f $fid
2580 [[ $err -eq 0 ]] || error "File size changed during restore"
2584 run_test 31a "Import a large file and check size during restore"
2588 # test needs a running copytool
2593 local f=$DIR/$tdir/$tfile
2595 fid=$(make_custom_file_for_progress $f 39 1000000)
2596 [ $? != 0 ] && skip "not enough free space" && return
2598 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2599 wait_request_state $fid ARCHIVE SUCCEED
2602 restore_and_check_size $f $fid
2605 [[ $err -eq 0 ]] || error "File size changed during restore"
2609 run_test 31b "Restore a large unaligned file and check size during restore"
2612 # test needs a running copytool
2617 local f=$DIR/$tdir/$tfile
2619 fid=$(make_custom_file_for_progress $f 33 1048576)
2620 [ $? != 0 ] && skip "not enough free space" && return
2622 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2623 wait_request_state $fid ARCHIVE SUCCEED
2626 restore_and_check_size $f $fid
2629 [[ $err -eq 0 ]] || error "File size changed during restore"
2633 run_test 31c "Restore a large aligned file and check size during restore"
2636 # test needs a running copytool
2641 local f=$DIR/$tdir/$tfile
2643 fid=$(make_custom_file_for_progress $f 39 1000000)
2644 [ $? != 0 ] && skip "not enough free space" && return
2646 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2647 wait_request_state $fid ARCHIVE SUCCEED
2650 # to be sure wait_all_done will not be mislead by previous tests
2653 wait_for_grace_delay
2654 # Also raise grace_delay significantly so the Canceled
2655 # Restore action will stay enough long avail.
2656 local old_grace=$(get_hsm_param grace_delay)
2657 set_hsm_param grace_delay 100
2659 md5sum $f >/dev/null &
2661 wait_request_state $fid RESTORE STARTED
2666 # Check restore trigger process was killed
2667 local killed=$(ps -o pid,comm hp $pid >/dev/null)
2671 # instead of waiting+checking both Restore and Cancel ops
2672 # sequentially, wait for both to be finished and then check
2674 wait_all_done 100 $fid
2675 local rstate=$(get_request_state $fid RESTORE)
2676 local cstate=$(get_request_state $fid CANCEL)
2678 # restore orig grace_delay.
2679 set_hsm_param grace_delay $old_grace
2681 if [[ "$rstate" == "CANCELED" ]] ; then
2682 [[ "$cstate" == "SUCCEED" ]] ||
2683 error "Restore state is CANCELED and Cancel state " \
2684 "is not SUCCEED but $cstate"
2685 echo "Restore state is CANCELED, Cancel state is SUCCEED"
2686 elif [[ "$rstate" == "SUCCEED" ]] ; then
2687 [[ "$cstate" == "FAILED" ]] ||
2688 error "Restore state is SUCCEED and Cancel state " \
2689 "is not FAILED but $cstate"
2690 echo "Restore state is SUCCEED, Cancel state is FAILED"
2692 error "Restore state is $rstate and Cancel state is $cstate"
2696 error "Cannot kill process waiting for restore ($killed)"
2700 run_test 33 "Kill a restore waiting process"
2703 # test needs a running copytool
2708 local f=$DIR/$tdir/$tfile
2710 fid=$(make_custom_file_for_progress $f 39 1000000)
2711 [ $? != 0 ] && skip "not enough free space" && return
2713 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2714 wait_request_state $fid ARCHIVE SUCCEED
2717 md5sum $f >/dev/null &
2719 wait_request_state $fid RESTORE STARTED
2721 rm $f || error "rm $f failed"
2722 # rm must not block during restore
2723 wait_request_state $fid RESTORE STARTED
2725 wait_request_state $fid RESTORE SUCCEED
2726 # check md5sum pgm finished
2727 local there=$(ps -o pid,comm hp $pid >/dev/null)
2728 [[ -z $there ]] || error "Restore initiator does not exit"
2730 local rc=$(wait $pid)
2731 [[ $rc -eq 0 ]] || error "Restore initiator failed with $rc"
2735 run_test 34 "Remove file during restore"
2738 # test needs a running copytool
2743 local f=$DIR/$tdir/$tfile
2744 local f1=$DIR/$tdir/$tfile-1
2746 fid=$(make_custom_file_for_progress $f 39 1000000)
2747 [ $? != 0 ] && skip "not enough free space" && return
2749 local fid1=$(copy_file /etc/passwd $f1)
2750 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2751 wait_request_state $fid ARCHIVE SUCCEED
2754 md5sum $f >/dev/null &
2756 wait_request_state $fid RESTORE STARTED
2758 mv $f1 $f || error "mv $f1 $f failed"
2759 # mv must not block during restore
2760 wait_request_state $fid RESTORE STARTED
2762 wait_request_state $fid RESTORE SUCCEED
2763 # check md5sum pgm finished
2764 local there=$(ps -o pid,comm hp $pid >/dev/null)
2765 [[ -z $there ]] || error "Restore initiator does not exit"
2767 local rc=$(wait $pid)
2768 [[ $rc -eq 0 ]] || error "Restore initiator failed with $rc"
2771 [[ $fid2 == $fid1 ]] || error "Wrong fid after mv $fid2 != $fid1"
2775 run_test 35 "Overwrite file during restore"
2778 # test needs a running copytool
2783 local f=$DIR/$tdir/$tfile
2785 fid=$(make_custom_file_for_progress $f 39 1000000)
2786 [ $? != 0 ] && skip "not enough free space" && return
2788 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2789 wait_request_state $fid ARCHIVE SUCCEED
2792 md5sum $f >/dev/null &
2794 wait_request_state $fid RESTORE STARTED
2797 # rm must not block during restore
2798 wait_request_state $fid RESTORE STARTED
2800 wait_request_state $fid RESTORE SUCCEED
2801 # check md5sum pgm finished
2802 local there=$(ps -o pid,comm hp $pid >/dev/null)
2804 error "Restore initiator does not exit"
2806 local rc=$(wait $pid)
2808 error "Restore initiator failed with $rc"
2812 run_test 36 "Move file during restore"
2815 # LU-5683: check that an archived dirty file can be rearchived.
2817 copytool_setup $SINGLEAGT $MOUNT2
2820 local f=$DIR/$tdir/$tfile
2823 fid=$(make_small $f) || error "cannot create small file"
2825 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2826 wait_request_state $fid ARCHIVE SUCCEED
2827 $LFS hsm_release $f || error "cannot release $f"
2830 dd if=/dev/urandom of=$f bs=1M count=1 || error "cannot dirty file"
2832 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2833 wait_request_state $fid ARCHIVE SUCCEED
2837 run_test 37 "re-archive a dirty file"
2844 for n in $(seq 1 $count); do
2845 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $prefix.$n
2847 echo "$count archive requests submitted"
2851 local stream_count=4
2852 local file_count=100
2854 local f=$DIR/$tdir/$tfile
2859 for i in $(seq 1 $file_count); do
2860 for p in $(seq 1 $stream_count); do
2861 fid=$(copy_file /etc/hosts $f.$p.$i)
2864 # force copytool to use a local/temp archive dir to ensure best
2865 # performance vs remote/NFS mounts used in auto-tests
2866 if do_facet $SINGLEAGT "df --local $HSM_ARCHIVE" >/dev/null 2>&1 ; then
2869 copytool_setup $SINGLEAGT $MOUNT $HSM_ARCHIVE_NUMBER $TMP/$tdir
2871 # to be sure wait_all_done will not be mislead by previous tests
2873 wait_for_grace_delay
2875 # start archive streams in background (archive files in parallel)
2876 for p in $(seq 1 $stream_count); do
2877 multi_archive $f.$p $file_count &
2880 echo -n "Wait for all requests being enqueued..."
2886 run_test 40 "Parallel archive requests"
2889 # test needs a running copytool
2893 local f=$DIR/$tdir/$tfile
2894 local fid=$(copy_file /etc/motd $f 1)
2896 $LFS hsm_archive $f || error "could not archive file"
2897 wait_request_state $fid ARCHIVE SUCCEED
2898 check_hsm_flags $f "0x00000009"
2900 multiop_bg_pause $f O_c || error "multiop failed"
2904 client_up || client_up || true
2906 kill -USR1 $MULTIPID
2907 wait $MULTIPID || error "multiop close failed"
2909 check_hsm_flags $f "0x0000000b"
2913 run_test 52 "Opened for write file on an evicted client should be set dirty"
2916 # test needs a running copytool
2920 local f=$DIR/$tdir/$tfile
2921 local fid=$(copy_file /etc/motd $f 1)
2923 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2924 error "could not archive file"
2925 wait_request_state $fid ARCHIVE SUCCEED
2926 check_hsm_flags $f "0x00000009"
2928 multiop_bg_pause $f o_c || error "multiop failed"
2932 client_up || client_up || true
2934 kill -USR1 $MULTIPID
2935 wait $MULTIPID || error "multiop close failed"
2937 check_hsm_flags $f "0x00000009"
2941 run_test 53 "Opened for read file on an evicted client should not be set dirty"
2944 # test needs a running copytool
2948 local f=$DIR/$tdir/$tfile
2949 local fid=$(make_custom_file_for_progress $f 39 1000000)
2951 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2952 error "could not archive file"
2953 wait_request_state $fid ARCHIVE STARTED
2955 check_hsm_flags $f "0x00000001"
2957 # Avoid coordinator resending this request as soon it has failed.
2962 wait_request_state $fid ARCHIVE FAILED
2964 check_hsm_flags $f "0x00000003"
2969 run_test 54 "Write during an archive cancels it"
2972 # test needs a running copytool
2976 local f=$DIR/$tdir/$tfile
2977 local fid=$(make_custom_file_for_progress $f 39 1000000)
2979 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2980 error "could not archive file"
2981 wait_request_state $fid ARCHIVE STARTED
2983 check_hsm_flags $f "0x00000001"
2985 # Avoid coordinator resending this request as soon it has failed.
2988 $TRUNCATE $f 1024 || error "truncate failed"
2990 wait_request_state $fid ARCHIVE FAILED
2992 check_hsm_flags $f "0x00000003"
2997 run_test 55 "Truncate during an archive cancels it"
3000 # test needs a running copytool
3004 local f=$DIR/$tdir/$tfile
3006 fid=$(make_custom_file_for_progress $f 39 1000000)
3007 [ $? != 0 ] && skip "not enough free space" && return
3009 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3010 error "could not archive file"
3011 wait_request_state $fid ARCHIVE STARTED
3013 check_hsm_flags $f "0x00000001"
3015 # Change metadata and sync to be sure we are not changing only
3020 wait_request_state $fid ARCHIVE SUCCEED
3022 check_hsm_flags $f "0x00000009"
3026 run_test 56 "Setattr during an archive is ok"
3029 # Need one client for I/O, one for request
3030 needclients 2 || return 0
3032 # test needs a running copytool
3036 local f=$DIR/$tdir/test_archive_remote
3037 # Create a file on a remote node
3038 do_node $CLIENT2 "dd if=/dev/urandom of=$f bs=1M "\
3039 "count=2 conv=fsync"
3042 do_node $CLIENT2 "$LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f" ||
3043 error "hsm_archive failed"
3044 local fid=$(path2fid $f)
3045 wait_request_state $fid ARCHIVE SUCCEED
3047 # Release and implicit restore it
3048 do_node $CLIENT2 "$LFS hsm_release $f" ||
3049 error "hsm_release failed"
3050 do_node $CLIENT2 "md5sum $f" ||
3051 error "hsm_restore failed"
3053 wait_request_state $fid RESTORE SUCCEED
3057 run_test 57 "Archive a file with dirty cache on another node"
3059 truncate_released_file() {
3063 local sz=$(stat -c %s $src_file)
3064 local f=$DIR/$tdir/$tfile
3065 local fid=$(copy_file $1 $f)
3069 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3070 error "could not archive file"
3071 wait_request_state $fid ARCHIVE SUCCEED
3073 $LFS hsm_release $f || error "could not release file"
3075 $TRUNCATE $f $trunc_to || error "truncate failed"
3078 local sz1=$(stat -c %s $f)
3079 [[ $sz1 == $trunc_to ]] ||
3080 error "size after trunc: $sz1 expect $trunc_to, original $sz"
3083 check_hsm_flags $f "0x0000000b"
3085 local state=$(get_request_state $fid RESTORE)
3086 [[ "$state" == "SUCCEED" ]] ||
3087 error "truncate $sz does not trig restore, state = $state"
3089 $TRUNCATE $ref $trunc_to
3090 cmp $ref $f || error "file data wrong after truncate"
3096 # test needs a running copytool
3101 local sz=$(stat -c %s /etc/passwd)
3103 echo "truncate up from $sz to $((sz*2))"
3104 truncate_released_file /etc/passwd $((sz*2))
3106 echo "truncate down from $sz to $((sz/2))"
3107 truncate_released_file /etc/passwd $((sz/2))
3109 echo "truncate to 0"
3110 truncate_released_file /etc/passwd 0
3114 run_test 58 "Truncate a released file will trigger restore"
3118 local server_version=$(lustre_version_code $SINGLEMDS)
3119 [[ $server_version -lt $(version_code 2.7.63) ]] &&
3120 skip "Need MDS version at least 2.7.63" && return
3123 $MCREATE $DIR/$tfile || error "mcreate failed"
3124 $TRUNCATE $DIR/$tfile 42 || error "truncate failed"
3125 $LFS hsm_archive $DIR/$tfile || error "archive request failed"
3126 fid=$(path2fid $DIR/$tfile)
3127 wait_request_state $fid ARCHIVE SUCCEED
3128 $LFS hsm_release $DIR/$tfile || error "release failed"
3131 run_test 59 "Release stripeless file with non-zero size"
3134 # This test validates the fix for LU-4512. Ensure that the -u
3135 # option changes the progress reporting interval from the
3136 # default (30 seconds) to the user-specified interval.
3138 local progress_timeout=$((interval * 4))
3140 # test needs a new running copytool
3142 HSMTOOL_UPDATE_INTERVAL=$interval copytool_setup
3145 local f=$DIR/$tdir/$tfile
3147 fid=$(make_custom_file_for_progress $f 10)
3148 [ $? != 0 ] && skip "not enough free space" && return
3151 local mdt=${MDT_PREFIX}${mdtidx}
3152 local mds=mds$((mdtidx + 1))
3154 # Wait for copytool to register
3155 wait_update_facet $mds \
3156 "$LCTL get_param -n ${mdt}.hsm.agents | grep -o ^uuid" \
3157 uuid 100 || error "coyptool failed to register with $mdt"
3159 local start_at=$(date +%s)
3160 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3161 error "could not archive file"
3163 local agent=$(facet_active_host $SINGLEAGT)
3164 local prefix=$TESTLOG_PREFIX
3165 [[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME
3166 local copytool_log=$prefix.copytool_log.$agent.log
3169 wait_update $agent \
3170 "grep -o start.copy $copytool_log" "start copy" 100 ||
3171 error "copytool failed to start"
3173 local cmd="$LCTL get_param -n ${mdt}.hsm.active_requests"
3174 cmd+=" | awk '/'$fid'.*action=ARCHIVE/ {print \\\$12}' | cut -f2 -d="
3180 echo -n "Expecting a progress update within $progress_timeout seconds... "
3182 RESULT=$(do_node $(facet_active_host $mds) "$cmd")
3183 if [ $RESULT -gt 0 ]; then
3184 echo "$RESULT bytes copied in $WAIT seconds."
3186 elif [ $WAIT -ge $progress_timeout ]; then
3187 error "Timed out waiting for progress update!"
3190 WAIT=$((WAIT + sleep))
3194 local finish_at=$(date +%s)
3195 local elapsed=$((finish_at - start_at))
3197 # Ensure that the progress update occurred within the expected window.
3198 if [ $elapsed -lt $((interval - 1)) ]; then
3199 error "Expected progress update after at least $interval seconds"
3202 echo "Wait for on going archive hsm action to complete"
3203 wait_update $agent "grep -o copied $copytool_log" "copied" 10 ||
3204 echo "File archiving not completed even after 10 secs"
3209 run_test 60 "Changing progress update interval from default"
3212 # test needs a new running copytool
3214 copytool_monitor_setup
3215 HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
3217 # Just start and stop the copytool to generate events.
3220 # Wait for the copytool to register.
3221 wait_update --verbose $(facet_active_host mds1) \
3222 "$LCTL get_param -n ${MDT_PREFIX}0.hsm.agents | grep -o ^uuid" \
3224 error "copytool failed to register with MDT0000"
3228 local REGISTER_EVENT
3229 local UNREGISTER_EVENT
3230 while read event; do
3231 local parsed=$(parse_json_event "$event")
3232 if [ -z "$parsed" ]; then
3233 error "Copytool sent malformed event: $event"
3237 if [ $event_type == "REGISTER" ]; then
3238 REGISTER_EVENT=$event
3239 elif [ $event_type == "UNREGISTER" ]; then
3240 UNREGISTER_EVENT=$event
3242 done < <(echo $"$(get_copytool_event_log)")
3244 if [ -z "$REGISTER_EVENT" ]; then
3245 error "Copytool failed to send register event to FIFO"
3248 if [ -z "$UNREGISTER_EVENT" ]; then
3249 error "Copytool failed to send unregister event to FIFO"
3252 copytool_monitor_cleanup
3253 echo "Register/Unregister events look OK."
3255 run_test 70 "Copytool logs JSON register/unregister events to FIFO"
3258 # Bump progress interval for livelier events.
3261 # test needs a new running copytool
3263 copytool_monitor_setup
3264 HSMTOOL_UPDATE_INTERVAL=$interval \
3265 HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
3268 local f=$DIR/$tdir/$tfile
3270 fid=$(make_custom_file_for_progress $f 39 1000000)
3271 [ $? != 0 ] && skip "not enough free space" && return
3273 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3274 error "could not archive file"
3275 wait_request_state $fid ARCHIVE SUCCEED
3277 local expected_fields="event_time data_fid source_fid"
3278 expected_fields+=" total_bytes current_bytes"
3282 while read event; do
3283 # Make sure we're not getting anything from previous events.
3284 for field in $expected_fields; do
3288 local parsed=$(parse_json_event "$event")
3289 if [ -z "$parsed" ]; then
3290 error "Copytool sent malformed event: $event"
3294 if [ $event_type == "ARCHIVE_START" ]; then
3297 elif [ $event_type == "ARCHIVE_FINISH" ]; then
3300 elif [ $event_type != "ARCHIVE_RUNNING" ]; then
3304 # Do some simple checking of the progress update events.
3305 for expected_field in $expected_fields; do
3306 if [ -z ${!expected_field+x} ]; then
3307 error "Missing $expected_field field in event"
3311 if [ $total_bytes -eq 0 ]; then
3312 error "Expected total_bytes to be > 0"
3315 # These should be identical throughout an archive
3317 if [ $source_fid != $data_fid ]; then
3318 error "Expected source_fid to equal data_fid"
3320 done < <(echo $"$(get_copytool_event_log)")
3322 if [ -z "$START_EVENT" ]; then
3323 error "Copytool failed to send archive start event to FIFO"
3326 if [ -z "$FINISH_EVENT" ]; then
3327 error "Copytool failed to send archive finish event to FIFO"
3330 echo "Archive events look OK."
3334 copytool_monitor_cleanup
3336 run_test 71 "Copytool logs JSON archive events to FIFO"
3339 # Bump progress interval for livelier events.
3342 # test needs a new running copytool
3344 copytool_monitor_setup
3345 HSMTOOL_UPDATE_INTERVAL=$interval \
3346 HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
3347 local test_file=$HSMTOOL_MONITOR_DIR/file
3349 local cmd="dd if=/dev/urandom of=$test_file count=16 bs=1000000 "
3351 do_facet $SINGLEAGT "$cmd" ||
3352 error "cannot create $test_file on $SINGLEAGT"
3353 copy2archive $test_file $tdir/$tfile
3356 local f=$DIR/$tdir/$tfile
3357 import_file $tdir/$tfile $f
3358 f=$DIR2/$tdir/$tfile
3359 echo "Verifying released state: "
3360 check_hsm_flags $f "0x0000000d"
3362 local fid=$(path2fid $f)
3364 wait_request_state $fid RESTORE SUCCEED
3366 local expected_fields="event_time data_fid source_fid"
3367 expected_fields+=" total_bytes current_bytes"
3371 while read event; do
3372 # Make sure we're not getting anything from previous events.
3373 for field in $expected_fields; do
3377 local parsed=$(parse_json_event "$event")
3378 if [ -z "$parsed" ]; then
3379 error "Copytool sent malformed event: $event"
3383 if [ $event_type == "RESTORE_START" ]; then
3385 if [ $source_fid != $data_fid ]; then
3386 error "source_fid should == data_fid at start"
3389 elif [ $event_type == "RESTORE_FINISH" ]; then
3391 if [ $source_fid != $data_fid ]; then
3392 error "source_fid should == data_fid at finish"
3395 elif [ $event_type != "RESTORE_RUNNING" ]; then
3399 # Do some simple checking of the progress update events.
3400 for expected_field in $expected_fields; do
3401 if [ -z ${!expected_field+x} ]; then
3402 error "Missing $expected_field field in event"
3406 if [ $total_bytes -eq 0 ]; then
3407 error "Expected total_bytes to be > 0"
3410 # When a restore starts out, the data fid is the same as the
3411 # source fid. After the restore has gotten going, we learn
3412 # the new data fid. Once the restore has finished, the source
3413 # fid is set to the new data fid.
3415 # We test this because some monitoring software may depend on
3416 # this behavior. If it changes, then the consumers of these
3417 # events may need to be modified.
3418 if [ $source_fid == $data_fid ]; then
3419 error "source_fid should != data_fid during restore"
3421 done < <(echo $"$(get_copytool_event_log)")
3423 if [ -z "$START_EVENT" ]; then
3424 error "Copytool failed to send restore start event to FIFO"
3427 if [ -z "$FINISH_EVENT" ]; then
3428 error "Copytool failed to send restore finish event to FIFO"
3431 echo "Restore events look OK."
3435 copytool_monitor_cleanup
3439 run_test 72 "Copytool logs JSON restore events to FIFO"
3442 file_count=51 # Max number of files constrained by LNET message size
3443 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3444 local f=$DIR/$tdir/$tfile
3445 local FILELIST=/tmp/filelist.txt
3449 for i in $(seq 1 $file_count); do
3450 fid=$(copy_file /etc/hosts $f.$i)
3451 echo $f.$i >> $FILELIST
3453 # force copytool to use a local/temp archive dir to ensure best
3454 # performance vs remote/NFS mounts used in auto-tests
3455 if do_facet $SINGLEAGT "df --local $HSM_ARCHIVE" >/dev/null 2>&1 ; then
3458 local dai=$(get_hsm_param default_archive_id)
3459 copytool_setup $SINGLEAGT $MOUNT $dai $TMP/$tdir
3461 # to be sure wait_all_done will not be mislead by previous tests
3463 wait_for_grace_delay
3464 $LFS hsm_archive --filelist $FILELIST ||
3465 error "cannot archive a file list"
3467 $LFS hsm_release --filelist $FILELIST ||
3468 error "cannot release a file list"
3469 $LFS hsm_restore --filelist $FILELIST ||
3470 error "cannot restore a file list"
3474 run_test 90 "Archive/restore a file list"
3476 double_verify_reset_hsm_param() {
3478 echo "Testing $HSM_PARAM.$p"
3479 local val=$(get_hsm_param $p)
3481 local val2=$(($val * 2))
3482 set_hsm_param $p $val2
3483 val=$(get_hsm_param $p)
3484 [[ $val == $val2 ]] ||
3485 error "$HSM_PARAM.$p: $val != $val2 should be (2 * $save)"
3486 echo "Set $p to 0 must failed"
3490 set_hsm_param $p $save
3492 if [[ $rc == 0 ]]; then
3493 error "we must not be able to set $HSM_PARAM.$p to 0"
3498 double_verify_reset_hsm_param loop_period
3499 double_verify_reset_hsm_param grace_delay
3500 double_verify_reset_hsm_param active_request_timeout
3501 double_verify_reset_hsm_param max_requests
3502 double_verify_reset_hsm_param default_archive_id
3504 run_test 100 "Set coordinator /proc tunables"
3511 run_test 102 "Verify coordinator control"
3514 # test needs a running copytool
3521 for i in $(seq 1 20); do
3522 fid=$(copy_file /etc/passwd $DIR/$tdir/$i)
3524 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/*
3528 echo "Current requests"
3529 local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3530 $HSM_PARAM.actions |\
3531 grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
3533 [[ -z "$res" ]] || error "Some request have not been canceled"
3537 run_test 103 "Purge all requests"
3542 # test needs a running copytool
3546 local f=$DIR/$tdir/$tfile
3548 fid=$(make_custom_file_for_progress $f 39 1000000)
3549 [ $? != 0 ] && skip "not enough free space" && return
3551 # if cdt is on, it can serve too quickly the request
3553 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
3554 local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3555 $HSM_PARAM.actions |\
3556 grep $fid | cut -f16 -d=")
3559 [[ "$data1" == "$DATAHEX" ]] ||
3560 error "Data field in records is ($data1) and not ($DATAHEX)"
3564 run_test 104 "Copy tool data field"
3571 for i in $(seq -w 1 10); do
3572 cp /etc/passwd $DIR/$tdir/$i
3573 $LFS hsm_archive $DIR/$tdir/$i
3575 local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3576 $HSM_PARAM.actions |\
3577 grep WAITING | wc -l")
3580 local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3581 $HSM_PARAM.actions |\
3582 grep WAITING | wc -l")
3585 [[ "$reqcnt1" == "$reqcnt2" ]] ||
3586 error "Requests count after shutdown $reqcnt2 != "\
3587 "before shutdown $reqcnt1"
3589 run_test 105 "Restart of coordinator"
3592 # test needs a running copytool
3595 local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
3597 check_agent_registered $uuid
3599 search_copytools || error "No copytool found"
3602 check_agent_unregistered $uuid
3605 uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
3606 check_agent_registered $uuid
3610 run_test 106 "Copytool register/unregister"
3613 # test needs a running copytool
3615 # create and archive file
3617 local f1=$DIR/$tdir/$tfile
3618 local fid=$(copy_file /etc/passwd $f1)
3619 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
3620 wait_request_state $fid ARCHIVE SUCCEED
3621 # shutdown and restart MDS
3623 # check the copytool still gets messages from MDT
3624 local f2=$DIR/$tdir/2
3625 local fid=$(copy_file /etc/passwd $f2)
3626 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
3627 # main check of this sanity: this request MUST succeed
3628 wait_request_state $fid ARCHIVE SUCCEED
3631 run_test 107 "Copytool re-register after MDS restart"
3633 policy_set_and_test()
3637 do_facet $SINGLEMDS $LCTL set_param "$HSM_PARAM.policy=\\\"$change\\\""
3638 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3639 [[ "$policy" == "$target" ]] ||
3640 error "Wrong policy after '$change': '$policy' != '$target'"
3644 # to force default policy setting if error
3645 CDT_POLICY_HAD_CHANGED=true
3647 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3648 local default="NonBlockingRestore [NoRetryAction]"
3649 [[ "$policy" == "$default" ]] ||
3650 error "default policy has changed,"\
3651 " '$policy' != '$default' update the test"
3652 policy_set_and_test "+NBR" "[NonBlockingRestore] [NoRetryAction]"
3653 policy_set_and_test "+NRA" "[NonBlockingRestore] [NoRetryAction]"
3654 policy_set_and_test "-NBR" "NonBlockingRestore [NoRetryAction]"
3655 policy_set_and_test "-NRA" "NonBlockingRestore NoRetryAction"
3656 policy_set_and_test "NRA NBR" "[NonBlockingRestore] [NoRetryAction]"
3657 # useless bacause we know but safer for futur changes to use real value
3658 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3659 echo "Next set_param must failed"
3660 policy_set_and_test "wrong" "$policy"
3663 echo "Back to default policy"
3664 cdt_set_sanity_policy
3666 run_test 109 "Policy display/change"
3669 # test needs a running copytool
3674 copy2archive /etc/passwd $tdir/$tfile
3676 local f=$DIR/$tdir/$tfile
3677 import_file $tdir/$tfile $f
3678 local fid=$(path2fid $f)
3680 cdt_set_non_blocking_restore
3685 wait_request_state $fid RESTORE SUCCEED
3686 cdt_clear_non_blocking_restore
3690 error "md5sum returns $st != 1, "\
3691 "should also perror ENODATA (No data available)"
3695 run_test 110a "Non blocking restore policy (import case)"
3698 # test needs a running copytool
3702 local f=$DIR/$tdir/$tfile
3703 local fid=$(copy_file /etc/passwd $f)
3704 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3705 wait_request_state $fid ARCHIVE SUCCEED
3708 cdt_set_non_blocking_restore
3713 wait_request_state $fid RESTORE SUCCEED
3714 cdt_clear_non_blocking_restore
3718 error "md5sum returns $st != 1, "\
3719 "should also perror ENODATA (No data available)"
3723 run_test 110b "Non blocking restore policy (release case)"
3726 # test needs a running copytool
3730 copy2archive /etc/passwd $tdir/$tfile
3732 local f=$DIR/$tdir/$tfile
3734 import_file $tdir/$tfile $f
3735 local fid=$(path2fid $f)
3739 copytool_remove_backend $fid
3742 wait_request_state $fid RESTORE FAILED
3749 [[ $st == 0 ]] || error "Restore does not failed"
3753 run_test 111a "No retry policy (import case), restore will error"\
3754 " (No such file or directory)"
3757 # test needs a running copytool
3761 local f=$DIR/$tdir/$tfile
3762 local fid=$(copy_file /etc/passwd $f)
3764 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3765 wait_request_state $fid ARCHIVE SUCCEED
3768 copytool_remove_backend $fid
3771 wait_request_state $fid RESTORE FAILED
3778 [[ $st == 0 ]] || error "Restore does not failed"
3782 run_test 111b "No retry policy (release case), restore will error"\
3783 " (No such file or directory)"
3786 # test needs a running copytool
3790 local f=$DIR/$tdir/$tfile
3791 local fid=$(copy_file /etc/passwd $f)
3793 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3794 local l=$($LFS hsm_action $f)
3796 local res=$(echo $l | cut -f 2- -d" " | grep ARCHIVE)
3800 wait_request_state $fid ARCHIVE SUCCEED
3803 [[ ! -z "$res" ]] || error "action is $l which is not an ARCHIVE"
3807 run_test 112 "State of recorded request"
3810 # test needs a running copytool
3814 local f=$DIR/$tdir/$tfile
3816 fid=$(make_custom_file_for_progress $f 103 1048576)
3817 [ $? != 0 ] && skip "not enough free space" && return
3819 # test with cdt on is made in test_221
3821 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3822 # wait archive to register at CDT
3823 wait_request_state $fid ARCHIVE WAITING
3826 wait_request_state $fid ARCHIVE CANCELED
3827 wait_request_state $fid CANCEL SUCCEED
3831 run_test 200 "Register/Cancel archive"
3834 # test needs a running copytool
3838 local f=$DIR/$tdir/$tfile
3839 make_archive $tdir/$tfile
3840 import_file $tdir/$tfile $f
3841 local fid=$(path2fid $f)
3843 # test with cdt on is made in test_222
3846 # wait restore to register at CDT
3847 wait_request_state $fid RESTORE WAITING
3850 wait_request_state $fid RESTORE CANCELED
3851 wait_request_state $fid CANCEL SUCCEED
3855 run_test 201 "Register/Cancel restore"
3858 # test needs a running copytool
3862 local f=$DIR/$tdir/$tfile
3864 fid=$(make_custom_file_for_progress $f 39 1000000)
3865 [ $? != 0 ] && skip "not enough free space" && return
3867 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3868 wait_request_state $fid ARCHIVE SUCCEED
3872 # wait remove to register at CDT
3873 wait_request_state $fid REMOVE WAITING
3876 wait_request_state $fid REMOVE CANCELED
3880 run_test 202 "Register/Cancel remove"
3883 # test needs a running copytool
3888 local f=$DIR/$tdir/$tfile
3889 local fid=$(copy_file /etc/passwd $f)
3893 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3894 wait_request_state $fid ARCHIVE SUCCEED
3896 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3900 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3904 run_test 220 "Changelog for archive"
3907 # test needs a running copytool
3912 local f=$DIR/$tdir/$tfile
3914 fid=$(make_custom_file_for_progress $f 103 1048576)
3915 [ $? != 0 ] && skip "not enough free space" && return
3919 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3920 wait_request_state $fid ARCHIVE STARTED
3922 wait_request_state $fid ARCHIVE CANCELED
3923 wait_request_state $fid CANCEL SUCCEED
3925 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3928 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3932 run_test 221 "Changelog for archive canceled"
3935 # test needs a running copytool
3939 copy2archive /etc/passwd $tdir/$tfile
3941 local f=$DIR/$tdir/$tfile
3942 import_file $tdir/$tfile $f
3943 local fid=$(path2fid $f)
3948 wait_request_state $fid RESTORE SUCCEED
3950 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3953 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3957 run_test 222a "Changelog for explicit restore"
3960 # test needs a running copytool
3964 local f=$DIR/$tdir/$tfile
3965 local fid=$(copy_file /etc/passwd $f)
3968 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3969 wait_request_state $fid ARCHIVE SUCCEED
3974 wait_request_state $fid RESTORE SUCCEED
3976 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3979 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3983 run_test 222b "Changelog for implicit restore"
3986 # test needs a running copytool
3991 local f=$DIR/$tdir/$tfile
3992 make_archive $tdir/$tfile
3996 import_file $tdir/$tfile $f
3997 local fid=$(path2fid $f)
4000 wait_request_state $fid RESTORE STARTED
4002 wait_request_state $fid RESTORE CANCELED
4003 wait_request_state $fid CANCEL SUCCEED
4005 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
4008 [[ $flags == $target ]] ||
4009 error "Changelog flag is $flags not $target"
4013 run_test 223a "Changelog for restore canceled (import case)"
4016 # test needs a running copytool
4021 local f=$DIR/$tdir/$tfile
4023 fid=$(make_custom_file_for_progress $f 39 1000000)
4024 [ $? != 0 ] && skip "not enough free space" && return
4027 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4028 wait_request_state $fid ARCHIVE SUCCEED
4031 wait_request_state $fid RESTORE STARTED
4033 wait_request_state $fid RESTORE CANCELED
4034 wait_request_state $fid CANCEL SUCCEED
4036 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
4039 [[ $flags == $target ]] ||
4040 error "Changelog flag is $flags not $target"
4044 run_test 223b "Changelog for restore canceled (release case)"
4047 # test needs a running copytool
4052 local f=$DIR/$tdir/$tfile
4053 local fid=$(copy_file /etc/passwd $f)
4056 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4057 wait_request_state $fid ARCHIVE SUCCEED
4060 wait_request_state $fid REMOVE SUCCEED
4062 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -n 1)
4065 [[ $flags == $target ]] ||
4066 error "Changelog flag is $flags not $target"
4070 run_test 224 "Changelog for remove"
4073 # test needs a running copytool
4076 # test is not usable because remove request is too fast
4077 # so it is always finished before cancel can be done ...
4078 echo "Test disabled"
4083 local f=$DIR/$tdir/$tfile
4085 fid=$(make_custom_file_for_progress $f 39 1000000)
4086 [ $? != 0 ] && skip "not enough free space" && return
4089 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4090 wait_request_state $fid ARCHIVE SUCCEED
4092 # if cdt is on, it can serve too quickly the request
4097 wait_request_state $fid REMOVE CANCELED
4098 wait_request_state $fid CANCEL SUCCEED
4100 flags=$(changelog_get_flags ${MDT[0]} RENME $fid2)
4101 local flags=$($LFS changelog ${MDT[0]} | grep HSM | grep $fid |
4102 tail -n 1 | awk '{print $5}')
4105 [[ $flags == $target ]] ||
4106 error "Changelog flag is $flags not $target"
4110 run_test 225 "Changelog for remove canceled"
4113 # test needs a running copytool
4118 local f1=$DIR/$tdir/$tfile-1
4119 local f2=$DIR/$tdir/$tfile-2
4120 local f3=$DIR/$tdir/$tfile-3
4121 local fid1=$(copy_file /etc/passwd $f1)
4122 local fid2=$(copy_file /etc/passwd $f2)
4123 copy_file /etc/passwd $f3
4126 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
4127 wait_request_state $fid1 ARCHIVE SUCCEED
4129 $LFS hsm_archive $f2
4130 wait_request_state $fid2 ARCHIVE SUCCEED
4132 rm $f1 || error "rm $f1 failed"
4134 local flags=$(changelog_get_flags ${MDT[0]} UNLNK $fid1)
4137 [[ $flags == $target ]] ||
4138 error "Changelog flag is $flags not $target"
4140 mv $f3 $f2 || error "mv $f3 $f2 failed"
4142 flags=$(changelog_get_flags ${MDT[0]} RENME $fid2)
4145 [[ $flags == $target ]] ||
4146 error "Changelog flag is $flags not $target"
4150 run_test 226 "changelog for last rm/mv with exiting archive"
4152 check_flags_changes() {
4160 $LFS hsm_set --$hsm_flag $f ||
4161 error "Cannot set $hsm_flag on $f"
4162 local flags=($(changelog_get_flags ${MDT[0]} HSM $fid))
4163 local seen=${#flags[*]}
4165 [[ $seen == $cnt ]] ||
4166 error "set $hsm_flag: Changelog events $seen != $cnt"
4167 [[ ${flags[$((cnt - 1))]} == $target ]] ||
4168 error "set $hsm_flag: Changelog flags are "\
4169 "${flags[$((cnt - 1))]} not $target"
4171 $LFS hsm_clear --$hsm_flag $f ||
4172 error "Cannot clear $hsm_flag on $f"
4173 flags=($(changelog_get_flags ${MDT[0]} HSM $fid))
4176 [[ $cnt == $seen ]] ||
4177 error "clear $hsm_flag: Changelog events $seen != $cnt"
4179 [[ ${flags[$((cnt - 1))]} == $target ]] ||
4180 error "clear $hsm_flag: Changelog flag is "\
4181 "${flags[$((cnt - 1))]} not $target"
4185 # test needs a running copytool
4192 for i in norelease noarchive exists archived
4194 local f=$DIR/$tdir/$tfile-$i
4195 local fid=$(copy_file /etc/passwd $f)
4196 check_flags_changes $f $fid $i 0 1
4199 f=$DIR/$tdir/$tfile---lost
4200 fid=$(copy_file /etc/passwd $f)
4201 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4202 wait_request_state $fid ARCHIVE SUCCEED
4203 check_flags_changes $f $fid lost 3 1
4207 run_test 227 "changelog when explicit setting of HSM flags"
4210 # test needs a running copytool
4213 local fid=$(make_small_sync $DIR/$tfile)
4214 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
4215 wait_request_state $fid ARCHIVE SUCCEED
4217 $LFS hsm_release $DIR/$tfile
4218 check_hsm_flags $DIR/$tfile "0x0000000d"
4220 filefrag $DIR/$tfile | grep " 1 extent found" ||
4221 error "filefrag on released file must return only one extent"
4223 # only newer versions of cp detect sparse files by stat/FIEMAP
4225 cp --sparse=auto $DIR/$tfile $DIR/$tfile.2 ||
4226 error "copying $DIR/$tfile"
4227 cmp $DIR/$tfile $DIR/$tfile.2 || error "comparing copied $DIR/$tfile"
4229 $LFS hsm_release $DIR/$tfile
4230 check_hsm_flags $DIR/$tfile "0x0000000d"
4232 mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
4234 tar cf - --sparse $DIR/$tfile | tar xvf - -C $DIR/$tdir ||
4236 cmp $DIR/$tfile $DIR/$tdir/$DIR/$tfile ||
4237 error "comparing untarred $DIR/$tfile"
4239 rm -f $DIR/$tfile $DIR/$tfile.2 ||
4240 error "rm $DIR/$tfile or $DIR/$tfile.2 failed"
4243 run_test 228 "On released file, return extend to FIEMAP. For [cp,tar] --sparse"
4246 # test needs a running copytool
4250 local maxrequest=$(get_hsm_param max_requests)
4251 local rqcnt=$(($maxrequest * 3))
4255 for i in $(seq -w 1 $rqcnt); do
4257 dd if=/dev/urandom of=$DIR/$tdir/$i bs=1M count=10 conv=fsync
4259 # we do it in 2 steps, so all requests arrive at the same time
4260 for i in $(seq -w 1 $rqcnt); do
4261 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/$i
4266 while [[ $cnt != 0 || $wt != 0 ]]; do
4268 cnt=$(do_facet $SINGLEMDS "$LCTL get_param -n\
4269 $HSM_PARAM.actions |\
4270 grep STARTED | grep -v CANCEL | wc -l")
4271 [[ $cnt -le $maxrequest ]] ||
4272 error "$cnt > $maxrequest too many started requests"
4273 wt=$(do_facet $SINGLEMDS "$LCTL get_param\
4274 $HSM_PARAM.actions |\
4275 grep WAITING | wc -l")
4276 echo "max=$maxrequest started=$cnt waiting=$wt"
4281 run_test 250 "Coordinator max request"
4284 # test needs a running copytool
4288 local f=$DIR/$tdir/$tfile
4290 fid=$(make_custom_file_for_progress $f 103 1048576)
4291 [ $? != 0 ] && skip "not enough free space" && return
4294 # to have a short test
4295 local old_to=$(get_hsm_param active_request_timeout)
4296 set_hsm_param active_request_timeout 4
4297 # to be sure the cdt will wake up frequently so
4298 # it will be able to cancel the "old" request
4299 local old_loop=$(get_hsm_param loop_period)
4300 set_hsm_param loop_period 2
4303 # clear locks to avoid extra delay caused by flush/cancel
4304 # and thus prevent early copytool death to timeout.
4305 cancel_lru_locks osc
4307 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4308 wait_request_state $fid ARCHIVE STARTED
4310 wait_request_state $fid ARCHIVE CANCELED
4312 set_hsm_param active_request_timeout $old_to
4313 set_hsm_param loop_period $old_loop
4317 run_test 251 "Coordinator request timeout"
4320 # the only way to test ondisk conf is to restart MDS ...
4321 echo "Stop coordinator and remove coordinator state at mount"
4324 # clean on disk conf set by default
4325 cdt_clear_mount_state
4326 cdt_check_state stopped
4328 # check cdt still off after umount/remount
4330 cdt_check_state stopped
4332 echo "Set coordinator start at mount, and start coordinator"
4333 cdt_set_mount_state enabled
4336 cdt_check_state enabled
4338 # check cdt still on after umount/remount
4340 cdt_check_state enabled
4342 # we are back to original state (cdt started at mount)
4344 run_test 300 "On disk coordinator state kept between MDT umount/mount"
4347 local ai=$(get_hsm_param default_archive_id)
4348 local new=$((ai + 1))
4350 set_hsm_param default_archive_id $new -P
4352 local res=$(get_hsm_param default_archive_id)
4355 set_hsm_param default_archive_id "" "-P -d"
4357 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
4359 run_test 301 "HSM tunnable are persistent"
4362 local ai=$(get_hsm_param default_archive_id)
4363 local new=$((ai + 1))
4368 set_hsm_param default_archive_id $new -P
4371 for mdtno in $(seq 1 $MDSCOUNT); do
4376 cdt_check_state enabled
4378 local res=$(get_hsm_param default_archive_id)
4381 set_hsm_param default_archive_id "" "-P -d"
4383 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
4385 run_test 302 "HSM tunnable are persistent when CDT is off"
4388 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4394 local dir_mdt0=$DIR/$tdir/mdt0
4395 local dir_mdt1=$DIR/$tdir/mdt1
4397 # create 1 dir per MDT
4398 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4399 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
4401 # create 1 file in each MDT
4402 local fid1=$(make_small $dir_mdt0/$tfile)
4403 local fid2=$(make_small $dir_mdt1/$tfile)
4405 # check that hsm request on mdt0 is sent to the right MDS
4406 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
4407 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4408 echo "archive successful on mdt0"
4410 # check that hsm request on mdt1 is sent to the right MDS
4411 $LFS hsm_archive $dir_mdt1/$tfile || error "lfs hsm_archive"
4412 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
4413 echo "archive successful on mdt1"
4416 # clean test files and directories
4417 rm -rf $dir_mdt0 $dir_mdt1
4419 run_test 400 "Single request is sent to the right MDT"
4422 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4428 local dir_mdt0=$DIR/$tdir/mdt0
4429 local dir_mdt1=$DIR/$tdir/mdt1
4431 # create 1 dir per MDT
4432 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4433 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
4435 # create 1 file in each MDT
4436 local fid1=$(make_small $dir_mdt0/$tfile)
4437 local fid2=$(make_small $dir_mdt1/$tfile)
4439 # check that compound requests are shunt to the rights MDTs
4440 $LFS hsm_archive $dir_mdt0/$tfile $dir_mdt1/$tfile ||
4441 error "lfs hsm_archive"
4442 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4443 echo "archive successful on mdt0"
4444 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
4445 echo "archive successful on mdt1"
4448 # clean test files and directories
4449 rm -rf $dir_mdt0 $dir_mdt1
4451 run_test 401 "Compound requests split and sent to their respective MDTs"
4453 mdc_change_state() # facet, MDT_pattern, activate|deactivate
4458 local node=$(facet_active_host $facet)
4460 for mdc in $(do_facet $facet "$LCTL dl | grep -E ${pattern}-mdc" |
4461 awk '{print $4}'); do
4462 echo "$3 $mdc on $node"
4463 do_facet $facet "$LCTL --device $mdc $state" || return 1
4468 # make sure there is no running copytool
4471 # deactivate all mdc on agent1
4472 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "deactivate"
4474 HSMTOOL_NOERROR=true copytool_setup $SINGLEAGT
4476 check_agent_unregistered "uuid" # match any agent
4478 # no expected running copytool
4479 search_copytools $agent && error "Copytool start should have failed"
4482 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "activate"
4484 run_test 402 "Copytool start fails if all MDTs are inactive"
4487 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4489 # make sure there is no running copytool
4492 local agent=$(facet_active_host $SINGLEAGT)
4494 # deactivate all mdc for MDT0001
4495 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
4498 local uuid=$(get_agent_uuid $agent)
4499 # check the agent is registered on MDT0000, and not on MDT0001
4500 check_agent_registered_by_mdt $uuid 0
4501 check_agent_unregistered_by_mdt $uuid 1
4503 # check running copytool process
4504 search_copytools $agent || error "No running copytools on $agent"
4506 # reactivate all mdc for MDT0001
4507 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
4509 # make sure the copytool is now registered to all MDTs
4510 check_agent_registered $uuid
4514 run_test 403 "Copytool starts with inactive MDT and register on reconnect"
4517 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4521 # create files on both MDT0000 and MDT0001
4524 local dir_mdt0=$DIR/$tdir/mdt0
4525 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4527 # create 1 file on mdt0
4528 local fid1=$(make_small $dir_mdt0/$tfile)
4530 # deactivate all mdc for MDT0001
4531 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
4533 # send an HSM request for files in MDT0000
4534 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
4536 # check for completion of files in MDT0000
4537 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4538 echo "archive successful on mdt0"
4540 # reactivate all mdc for MDT0001
4541 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
4544 # clean test files and directories
4547 run_test 404 "Inactive MDT does not block requests for active MDTs"
4550 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4556 local striped_dir=$DIR/$tdir/striped_dir
4558 # create striped dir on all of MDTs
4559 $LFS mkdir -i 0 -c $MDSCOUNT $striped_dir || error "lfs mkdir"
4561 local fid1=$(make_small_sync $striped_dir/${tfile}_0)
4562 local fid2=$(make_small_sync $striped_dir/${tfile}_1)
4563 local fid3=$(make_small_sync $striped_dir/${tfile}_2)
4564 local fid4=$(make_small_sync $striped_dir/${tfile}_3)
4566 local idx1=$($LFS getstripe -M $striped_dir/${tfile}_0)
4567 local idx2=$($LFS getstripe -M $striped_dir/${tfile}_1)
4568 local idx3=$($LFS getstripe -M $striped_dir/${tfile}_2)
4569 local idx4=$($LFS getstripe -M $striped_dir/${tfile}_3)
4571 # check that compound requests are shunt to the rights MDTs
4572 $LFS hsm_archive $striped_dir/${tfile}_0 $striped_dir/${tfile}_1 \
4573 $striped_dir/${tfile}_2 $striped_dir/${tfile}_3 ||
4574 error "lfs hsm_archive"
4576 wait_request_state $fid1 ARCHIVE SUCCEED $idx1 &&
4577 echo "archive successful on $fid1"
4578 wait_request_state $fid2 ARCHIVE SUCCEED $idx2 &&
4579 echo "archive successful on $fid2"
4580 wait_request_state $fid3 ARCHIVE SUCCEED $idx3 &&
4581 echo "archive successful on $fid3"
4582 wait_request_state $fid4 ARCHIVE SUCCEED $idx4 &&
4583 echo "archive successful on $fid4"
4585 $LFS hsm_release $striped_dir/${tfile}_0 || error "lfs hsm_release 1"
4586 $LFS hsm_release $striped_dir/${tfile}_1 || error "lfs hsm_release 2"
4587 $LFS hsm_release $striped_dir/${tfile}_2 || error "lfs hsm_release 3"
4588 $LFS hsm_release $striped_dir/${tfile}_3 || error "lfs hsm_release 4"
4590 cat $striped_dir/${tfile}_0 > /dev/null || error "cat ${tfile}_0 failed"
4591 cat $striped_dir/${tfile}_1 > /dev/null || error "cat ${tfile}_1 failed"
4592 cat $striped_dir/${tfile}_2 > /dev/null || error "cat ${tfile}_2 failed"
4593 cat $striped_dir/${tfile}_3 > /dev/null || error "cat ${tfile}_3 failed"
4597 run_test 405 "archive and release under striped directory"
4600 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4602 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
4603 skip "need MDS version at least 2.7.64" && return 0
4610 fid=$(make_small $DIR/$tdir/$tfile)
4613 $LFS hsm_archive $DIR/$tdir/$tfile
4614 wait_request_state "$fid" ARCHIVE SUCCEED
4615 $LFS hsm_release $DIR/$tdir/$tfile
4617 # Should migrate $tdir but not $tfile.
4618 $LFS mv -M1 $DIR/$tdir &&
4619 error "migrating HSM an archived file should fail"
4621 $LFS hsm_restore $DIR/$tdir/$tfile
4622 wait_request_state "$fid" RESTORE SUCCEED
4624 $LFS hsm_remove $DIR/$tdir/$tfile
4625 wait_request_state "$fid" REMOVE SUCCEED
4627 cat $DIR/$tdir/$tfile > /dev/null ||
4628 error "cannot read $DIR/$tdir/$tfile"
4630 $LFS mv -M1 $DIR/$tdir ||
4631 error "cannot complete migration after HSM remove"
4633 mdt_index=$($LFS getstripe -M $DIR/$tdir)
4634 if ((mdt_index != 1)); then
4635 error "expected MDT index 1, got $mdt_index"
4638 # Refresh fid after migration.
4639 fid=$(path2fid $DIR/$tdir/$tfile)
4642 $LFS hsm_archive $DIR/$tdir/$tfile
4643 wait_request_state "$fid" ARCHIVE SUCCEED 1
4645 lctl set_param debug=+trace
4646 $LFS hsm_release $DIR/$tdir/$tfile ||
4647 error "cannot release $DIR/$tdir/$tfile"
4649 $LFS hsm_restore $DIR/$tdir/$tfile
4650 wait_request_state "$fid" RESTORE SUCCEED 1
4652 cat $DIR/$tdir/$tfile > /dev/null ||
4653 error "cannot read $DIR/$tdir/$tfile"
4657 run_test 406 "attempting to migrate HSM archived files is safe"
4661 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.92) ] &&
4662 skip "HSM migrate is not supported" && return
4664 # Stop the existing copytool
4667 test_mkdir -p $DIR/$tdir
4668 llapi_hsm_test -d $DIR/$tdir || error "One llapi HSM test failed"
4670 run_test 500 "various LLAPI HSM tests"
4675 check_and_cleanup_lustre