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=${OPENFILE:-openfile}
27 MOUNT_2=${MOUNT_2:-"yes"}
30 # script only handles up to 10 MDTs (because of MDT_PREFIX)
31 [ $MDSCOUNT -gt 9 ] &&
32 error "script cannot handle more than 9 MDTs, please fix" && exit
34 check_and_setup_lustre
36 if [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.53) ]]; then
37 skip_env "Need MDS version at least 2.4.53" && exit
40 # $RUNAS_ID may get set incorrectly somewhere else
41 if [[ $UID -eq 0 && $RUNAS_ID -eq 0 ]]; then
42 skip_env "\$RUNAS_ID set to 0, but \$UID is also 0!" && exit
44 check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS
48 # if there is no CLIENT1 defined, some tests can be ran on localhost
49 CLIENT1=${CLIENT1:-$HOSTNAME}
50 # if CLIENT2 doesn't exist then use CLIENT1 instead
51 # All tests should use CLIENT2 with MOUNT2 only therefore it will work if
53 # Exception is the test which need two separate nodes
54 CLIENT2=${CLIENT2:-$CLIENT1}
57 # In order to test multiple remote HSM agents, a new facet type named "AGT" and
58 # the following associated variables are added:
60 # AGTCOUNT: number of agents
61 # AGTDEV{N}: target HSM mount point (root path of the backend)
62 # agt{N}_HOST: hostname of the agent agt{N}
63 # SINGLEAGT: facet of the single agent
65 # The number of agents is initialized as the number of remote client nodes.
66 # By default, only single copytool is started on a remote client/agent. If there
67 # was no remote client, then the copytool will be started on the local client.
73 export AGTCOUNT=${AGTCOUNT:-$((CLIENTCOUNT - 1))}
74 [[ $AGTCOUNT -gt 0 ]] || AGTCOUNT=1
76 export SHARED_DIRECTORY=${SHARED_DIRECTORY:-$TMP}
77 if [[ $CLIENTCOUNT -gt 1 ]] &&
78 ! check_shared_dir $SHARED_DIRECTORY $CLIENTS; then
79 skip_env "SHARED_DIRECTORY should be accessible"\
84 # We used to put the HSM archive in $SHARED_DIRECTORY but that
85 # meant NFS issues could hose sanity-hsm sessions. So now we
87 for n in $(seq $AGTCOUNT); do
88 eval export AGTDEV$n=\$\{AGTDEV$n:-"$TMP/arc$n"\}
89 agent=CLIENT$((n + 1))
90 if [[ -z "${!agent}" ]]; then
91 [[ $CLIENTCOUNT -eq 1 ]] && agent=CLIENT1 ||
94 eval export agt${n}_HOST=\$\{agt${n}_HOST:-${!agent}\}
97 export SINGLEAGT=${SINGLEAGT:-agt1}
99 export HSMTOOL=${HSMTOOL:-"lhsmtool_posix"}
100 export HSMTOOL_VERBOSE=${HSMTOOL_VERBOSE:-""}
101 export HSMTOOL_UPDATE_INTERVAL=${HSMTOOL_UPDATE_INTERVAL:=""}
102 export HSMTOOL_EVENT_FIFO=${HSMTOOL_EVENT_FIFO:=""}
103 export HSMTOOL_TESTDIR
104 export HSMTOOL_BASE=$(basename "$HSMTOOL" | cut -f1 -d" ")
105 HSM_ARCHIVE=$(copytool_device $SINGLEAGT)
108 # The test only support up to 10 MDTs
109 MDT_PREFIX="mdt.$FSNAME-MDT000"
110 HSM_PARAM="${MDT_PREFIX}0.hsm"
112 # archive is purged at copytool setup
113 HSM_ARCHIVE_PURGE=true
115 # Don't allow copytool error upon start/setup
116 HSMTOOL_NOERROR=false
119 # Get the backend root path for the given agent facet.
122 local dev=AGTDEV$(facet_number $facet)
127 # Stop copytool and unregister an existing changelog user.
129 copytool_monitor_cleanup
132 cdt_set_sanity_policy
137 # get MDT device for each mdc
138 for mdtno in $(seq 1 $MDSCOUNT); do
139 local idx=$(($mdtno - 1))
140 MDT[$idx]=$($LCTL get_param -n \
141 mdc.$FSNAME-MDT000${idx}-mdc-*.mds_server_uuid |
142 awk '{gsub(/_UUID/,""); print $1}' | head -n1)
147 local agents=${1:-$(facet_active_host $SINGLEAGT)}
148 do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
151 search_and_kill_copytool() {
152 local agents=${1:-$(facet_active_host $SINGLEAGT)}
154 echo "Killing existing copytools on $agents"
155 do_nodesv $agents "killall -q $HSMTOOL_BASE" || true
158 copytool_monitor_setup() {
159 local facet=${1:-$SINGLEAGT}
160 local agent=$(facet_active_host $facet)
162 local cmd="mktemp --tmpdir=/tmp -d ${TESTSUITE}.${TESTNAME}.XXXX"
163 local test_dir=$(do_node $agent "$cmd") ||
164 error "Failed to create tempdir on $agent"
165 export HSMTOOL_MONITOR_DIR=$test_dir
167 # Create the fifo and a monitor (cat dies when copytool dies)
168 do_node $agent "mkfifo -m 0644 $test_dir/fifo" ||
169 error "failed to create copytool fifo on $agent"
170 cmd="cat $test_dir/fifo > $test_dir/events &"
171 cmd+=" echo \\\$! > $test_dir/monitor_pid"
173 if [[ $PDSH == *Rmrsh* ]]; then
174 # This is required for pdsh -Rmrsh and its handling of remote
176 # Regular ssh and pdsh -Rssh work fine without this
177 # backgrounded subshell nonsense.
178 (do_node $agent "$cmd") &
179 export HSMTOOL_MONITOR_PDSH=$!
181 # Slightly racy, but just making a best-effort to catch obvious
184 ps -p $HSMTOOL_MONITOR_PDSH > /dev/null ||
185 error "Failed to start copytool monitor on $agent"
187 do_node $agent "$cmd"
189 error "Failed to start copytool monitor on $agent"
194 copytool_monitor_cleanup() {
195 local facet=${1:-$SINGLEAGT}
196 local agent=$(facet_active_host $facet)
198 if [ -n "$HSMTOOL_MONITOR_DIR" ]; then
199 # Should die when the copytool dies, but just in case.
200 local cmd="kill \\\$(cat $HSMTOOL_MONITOR_DIR/monitor_pid)"
201 cmd+=" 2>/dev/null || true"
202 do_node $agent "$cmd"
203 do_node $agent "rm -fr $HSMTOOL_MONITOR_DIR"
204 export HSMTOOL_MONITOR_DIR=
207 # The pdsh should die on its own when the monitor dies. Just
208 # in case, though, try to clean up to avoid any cruft.
209 if [ -n "$HSMTOOL_MONITOR_PDSH" ]; then
210 kill $HSMTOOL_MONITOR_PDSH 2>/dev/null
211 export HSMTOOL_MONITOR_PDSH=
216 local facet=${1:-$SINGLEAGT}
217 # Use MOUNT2 by default if defined
218 local lustre_mntpnt=${2:-${MOUNT2:-$MOUNT}}
220 local hsm_root=${4:-$(copytool_device $facet)}
221 local agent=$(facet_active_host $facet)
223 if [[ -z "$arc_id" ]] &&
224 do_facet $facet "pkill -CONT -x $HSMTOOL_BASE"; then
225 echo "Only wakeup running copytool $facet on $agent"
229 if $HSM_ARCHIVE_PURGE; then
230 echo "Purging archive on $agent"
231 do_facet $facet "rm -rf $hsm_root/*"
234 echo "Starting copytool $facet on $agent"
235 do_facet $facet "mkdir -p $hsm_root" || error "mkdir '$hsm_root' failed"
236 # bandwidth is limited to 1MB/s so the copy time is known and
237 # independent of hardware
238 local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon --hsm-root $hsm_root"
239 [[ -z "$arc_id" ]] || cmd+=" --archive $arc_id"
240 [[ -z "$HSMTOOL_UPDATE_INTERVAL" ]] ||
241 cmd+=" --update-interval $HSMTOOL_UPDATE_INTERVAL"
242 [[ -z "$HSMTOOL_EVENT_FIFO" ]] ||
243 cmd+=" --event-fifo $HSMTOOL_EVENT_FIFO"
244 cmd+=" --bandwidth 1 $lustre_mntpnt"
246 # Redirect the standard output and error to a log file which
247 # can be uploaded to Maloo.
248 local prefix=$TESTLOG_PREFIX
249 [[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME
250 local copytool_log=$prefix.copytool${arc_id}_log.$agent.log
252 do_facet $facet "$cmd < /dev/null > $copytool_log 2>&1"
253 if [[ $? != 0 ]]; then
254 [[ $HSMTOOL_NOERROR == true ]] ||
255 error "start copytool $facet on $agent failed"
256 echo "start copytool $facet on $agent failed"
262 get_copytool_event_log() {
263 local facet=${1:-$SINGLEAGT}
264 local agent=$(facet_active_host $facet)
266 [ -z "$HSMTOOL_MONITOR_DIR" ] &&
267 error "Can't get event log: No monitor directory!"
269 do_node $agent "cat $HSMTOOL_MONITOR_DIR/events" ||
270 error "Could not collect event log from $agent"
275 local facet=$SINGLEAGT
276 local agents=${1:-$(facet_active_host $facet)}
281 local hsm_root=$(copytool_device $facet)
282 local end_wait=$(( SECONDS + TIMEOUT ))
284 do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
286 while (( SECONDS < end_wait )); do
288 do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
289 if [ $? -ne 0 ]; then
290 echo "Copytool is stopped on $agents"
293 echo "Copytool still running on $agents"
295 if do_nodesv $agents "pgrep -x $HSMTOOL_BASE"; then
296 error "Copytool failed to stop in ${TIMEOUT}s ..."
298 echo "Copytool has stopped in " \
299 "$((TIMEOUT - (end_wait - SECONDS)))s."
302 # clean all CDTs orphans requests from previous tests
303 # that would otherwise need to timeout to clear.
304 for mdtno in $(seq 1 $MDSCOUNT); do
306 mdt_hsmctrl="mdt.$FSNAME-MDT000${idx}.hsm_control"
307 oldstate=$(do_facet mds${mdtno} "$LCTL get_param -n " \
309 # skip already stop[ed,ing] CDTs
310 echo $oldstate | grep stop && continue
312 do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=shutdown"
313 wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
315 error "mds${mdtno} cdt state is not stopped"
316 do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=$oldstate"
317 wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
319 error "mds${mdtno} cdt state is not $oldstate"
321 if do_facet $facet "df $hsm_root" >/dev/null 2>&1 ; then
322 do_facet $facet "rm -rf $hsm_root/*"
327 local agents=${1:-$(facet_active_host $SINGLEAGT)}
329 do_nodesv $agents "pkill -STOP -x $HSMTOOL_BASE" || return 0
330 echo "Copytool is suspended on $agents"
333 copytool_remove_backend() {
335 local be=$(do_facet $SINGLEAGT find $HSM_ARCHIVE -name $fid)
336 echo "Remove from backend: $fid = $be"
337 do_facet $SINGLEAGT rm -f $be
341 do_facet $SINGLEAGT \
342 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
343 --import $1 $2 $MOUNT" ||
344 error "import of $1 to $2 failed"
348 local file=$HSM_ARCHIVE/$1
349 do_facet $SINGLEAGT mkdir -p $(dirname $file)
350 do_facet $SINGLEAGT dd if=/dev/urandom of=$file count=32 bs=1000000 ||
351 file_creation_failure dd $file $?
355 local file=$HSM_ARCHIVE/$2
356 do_facet $SINGLEAGT mkdir -p $(dirname $file)
357 do_facet $SINGLEAGT cp -p $1 $file || error "cannot copy $1 to $file"
366 if [[ "$value" != "" ]]; then
369 for mdtno in $(seq 1 $MDSCOUNT); do
370 local idx=$(($mdtno - 1))
371 local facet=mds${mdtno}
372 # if $arg include -P option, run 1 set_param per MDT on the MGS
373 # else, run set_param on each MDT
374 [[ $arg = *"-P"* ]] && facet=mgs
375 do_facet $facet $LCTL set_param $arg mdt.${MDT[$idx]}.$key$value
376 [[ $? != 0 ]] && rc=1
386 for mdtno in $(seq 1 $MDSCOUNT); do
387 local idx=$(($mdtno - 1))
388 wait_result mds${mdtno} \
389 "$LCTL get_param -n $MDT_PREFIX${idx}.$key" "$target" \
391 error "$key state is not '$target' on mds${mdtno}"
398 for mdtno in $(seq 1 $MDSCOUNT); do
399 local idx=$(($mdtno - 1))
400 local cl_user=$(do_facet mds${mdtno} $LCTL \
401 --device ${MDT[$idx]} \
402 changelog_register -n)
404 do_facet mds${mdtno} lctl set_param \
405 mdd.${MDT[$idx]}.changelog_mask="+hsm"
406 $LFS changelog_clear ${MDT[$idx]} $cl_user 0
410 changelog_cleanup() {
412 for mdtno in $(seq 1 $MDSCOUNT); do
413 local idx=$(($mdtno - 1))
414 [[ -z ${CL_USERS[$idx]} ]] && continue
415 $LFS changelog_clear ${MDT[$idx]} ${CL_USERS[$idx]} 0
416 do_facet mds${mdtno} lctl --device ${MDT[$idx]} \
417 changelog_deregister ${CL_USERS[$idx]}
422 changelog_get_flags() {
427 $LFS changelog $mdt | awk "/$cltype/ && /t=\[$fid\]/ {print \$5}"
432 local val=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.$param)
440 mdts_set_param "$opt -n" "hsm.$param" "$value"
447 mdts_set_param "" hsm_control "$cmd"
448 mdts_check_param hsm_control "$target" 10
451 cdt_set_sanity_policy() {
452 if [[ "$CDT_POLICY_HAD_CHANGED" ]]
455 mdts_set_param "" hsm.policy "+NRA"
456 mdts_set_param "" hsm.policy "-NBR"
457 CDT_POLICY_HAD_CHANGED=
462 mdts_set_param "" hsm.policy "+NRA"
463 CDT_POLICY_HAD_CHANGED=true
466 cdt_clear_no_retry() {
467 mdts_set_param "" hsm.policy "-NRA"
468 CDT_POLICY_HAD_CHANGED=true
471 cdt_set_non_blocking_restore() {
472 mdts_set_param "" hsm.policy "+NBR"
473 CDT_POLICY_HAD_CHANGED=true
476 cdt_clear_non_blocking_restore() {
477 mdts_set_param "" hsm.policy "-NBR"
478 CDT_POLICY_HAD_CHANGED=true
481 cdt_clear_mount_state() {
482 mdts_set_param "-P -d" hsm_control ""
485 cdt_set_mount_state() {
486 mdts_set_param "-P" hsm_control "$1"
490 mdts_check_param hsm_control "$1" 20
494 set_test_state disabled disabled
498 set_test_state enabled enabled
502 set_test_state shutdown stopped
506 set_test_state purge enabled
512 cdt_set_sanity_policy
516 local client_count=$1
517 if [[ $CLIENTCOUNT -lt $client_count ]]; then
518 skip "Need $client_count or more clients, have $CLIENTCOUNT"
525 $LFS path2fid $1 | tr -d '[]'
526 return ${PIPESTATUS[0]}
534 if [[ $u == "user" ]]; then
535 st=$($RUNAS $LFS hsm_state $f)
538 st=$($LFS hsm_state $f)
541 [[ $? == 0 ]] || error "$LFS hsm_state $f failed (run as $u)"
543 st=$(echo $st | cut -f 2 -d" " | tr -d "()," )
547 get_hsm_archive_id() {
550 st=$($LFS hsm_state $f)
551 [[ $? == 0 ]] || error "$LFS hsm_state $f failed"
553 local ar=$(echo $st | grep "archive_id" | cut -f5 -d" " |
562 local st=$(get_hsm_flags $f)
563 [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
566 check_hsm_flags_user() {
570 local st=$(get_hsm_flags $f user)
571 [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
574 file_creation_failure() {
579 df $MOUNT $MOUNT2 >&2
580 error "cannot create $f with $cmd, status=$err"
592 if [[ "$3" != 1 ]]; then
596 cp $1 $f || file_creation_failure cp $f $?
598 path2fid $f || error "cannot get fid on $f"
602 local file2=${1/$DIR/$DIR2}
603 dd if=/dev/urandom of=$file2 count=2 bs=1M conv=fsync ||
604 file_creation_failure dd $file2 $?
606 path2fid $1 || error "cannot get fid on $1"
610 dd if=/dev/urandom of=$1 count=1 bs=1M conv=sync ||
611 file_creation_failure dd $1 $?
612 path2fid $1 || error "cannot get fid on $1"
615 cleanup_large_files() {
616 local ratio=$(df -P $MOUNT | tail -1 | awk '{print $5}' |
618 [ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \;
621 check_enough_free_space() {
624 local need=$((nb * unit /1024))
625 local free=$(df -kP $MOUNT | tail -1 | awk '{print $4}')
626 (( $need >= $free )) && return 1
630 make_large_for_striping() {
631 local file2=${1/$DIR/$DIR2}
632 local sz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1)
636 check_enough_free_space 5 $sz
637 [ $? != 0 ] && return $?
639 dd if=/dev/urandom of=$file2 count=5 bs=$sz conv=fsync ||
640 file_creation_failure dd $file2 $?
642 path2fid $1 || error "cannot get fid on $1"
645 make_large_for_progress() {
646 local file2=${1/$DIR/$DIR2}
650 check_enough_free_space 39 1000000
651 [ $? != 0 ] && return $?
653 # big file is large enough, so copy time is > 30s
654 # so copytool make 1 progress
655 # size is not a multiple of 1M to avoid stripe
657 dd if=/dev/urandom of=$file2 count=39 bs=1000000 conv=fsync ||
658 file_creation_failure dd $file2 $?
660 path2fid $1 || error "cannot get fid on $1"
663 make_large_for_progress_aligned() {
664 local file2=${1/$DIR/$DIR2}
668 check_enough_free_space 33 1048576
669 [ $? != 0 ] && return $?
671 # big file is large enough, so copy time is > 30s
672 # so copytool make 1 progress
673 # size is a multiple of 1M to have stripe
675 dd if=/dev/urandom of=$file2 count=33 bs=1M conv=fsync ||
676 file_creation_failure dd $file2 $?
677 path2fid $1 || error "cannot get fid on $1"
680 make_large_for_cancel() {
681 local file2=${1/$DIR/$DIR2}
685 check_enough_free_space 103 1048576
686 [ $? != 0 ] && return $?
688 # Copy timeout is 100s. 105MB => 105s
689 dd if=/dev/urandom of=$file2 count=103 bs=1M conv=fsync ||
690 file_creation_failure dd $file2 $?
691 path2fid $1 || error "cannot get fid on $1"
697 wait_update --verbose $(facet_active_host $facet) "$@"
700 wait_request_state() {
704 # 4th arg (mdt index) is optional
706 local mds=mds$(($mdtidx + 1))
708 local cmd="$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.actions"
709 cmd+=" | awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
711 wait_result $mds "$cmd" $state 200 ||
712 error "request on $fid is not $state on $mds"
715 get_request_state() {
719 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
720 "awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
723 get_request_count() {
727 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
728 "awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
735 local cmd="$LCTL get_param -n $HSM_PARAM.actions"
736 [[ -n $fid ]] && cmd+=" | grep '$fid'"
737 cmd+=" | egrep 'WAITING|STARTED'"
739 wait_result $SINGLEMDS "$cmd" "" $timeout ||
740 error "requests did not complete"
743 wait_for_grace_delay() {
744 local val=$(get_hsm_param grace_delay)
751 # python2.6 in EL6 includes an internal json module
752 local json_parser='import json; import fileinput;'
753 json_parser+=' print "\n".join(["local %s=\"%s\"" % tuple for tuple in '
754 json_parser+='json.loads([line for line in '
755 json_parser+='fileinput.input()][0]).items()])'
757 echo $raw_event | python -c "$json_parser"
760 # populate MDT device array
766 # cleanup from previous bad setup
767 search_and_kill_copytool
769 # for recovery tests, coordinator needs to be started at mount
771 # the lustre conf must be without hsm on (like for sanity.sh)
772 echo "Set HSM on and start"
773 cdt_set_mount_state enabled
774 cdt_check_state enabled
776 echo "Start copytool"
779 echo "Set sanity-hsm HSM policy"
780 cdt_set_sanity_policy
782 # finished requests are quickly removed from list
783 set_hsm_param grace_delay 10
789 local f=$DIR/$tdir/$tfile
793 check_hsm_flags_user $f "0x00000000"
795 $RUNAS $LFS hsm_set --norelease $f ||
796 error "user could not change hsm flags"
797 check_hsm_flags_user $f "0x00000010"
799 $RUNAS $LFS hsm_clear --norelease $f ||
800 error "user could not clear hsm flags"
801 check_hsm_flags_user $f "0x00000000"
803 # User could not change those flags...
804 $RUNAS $LFS hsm_set --exists $f &&
805 error "user should not set this flag"
806 check_hsm_flags_user $f "0x00000000"
809 $LFS hsm_set --exists $f ||
810 error "root could not change hsm flags"
811 check_hsm_flags_user $f "0x00000001"
813 $LFS hsm_clear --exists $f ||
814 error "root could not clear hsm state"
815 check_hsm_flags_user $f "0x00000000"
818 run_test 1 "lfs hsm flags root/non-root access"
822 local f=$DIR/$tdir/$tfile
824 # New files are not dirty
825 check_hsm_flags $f "0x00000000"
827 # For test, we simulate an archived file.
828 $LFS hsm_set --exists $f || error "user could not change hsm flags"
829 check_hsm_flags $f "0x00000001"
831 # chmod do not put the file dirty
832 chmod 600 $f || error "could not chmod test file"
833 check_hsm_flags $f "0x00000001"
835 # chown do not put the file dirty
836 chown $RUNAS_ID $f || error "could not chown test file"
837 check_hsm_flags $f "0x00000001"
839 # truncate put the file dirty
840 $TRUNCATE $f 1 || error "could not truncate test file"
841 check_hsm_flags $f "0x00000003"
843 $LFS hsm_clear --dirty $f || error "could not clear hsm flags"
844 check_hsm_flags $f "0x00000001"
846 run_test 2 "Check file dirtyness when doing setattr"
852 # New files are not dirty
854 check_hsm_flags $f "0x00000000"
856 # For test, we simulate an archived file.
857 $LFS hsm_set --exists $f ||
858 error "user could not change hsm flags"
859 check_hsm_flags $f "0x00000001"
861 # Reading a file, does not set dirty
862 cat $f > /dev/null || error "could not read file"
863 check_hsm_flags $f "0x00000001"
865 # Open for write without modifying data, does not set dirty
866 openfile -f O_WRONLY $f || error "could not open test file"
867 check_hsm_flags $f "0x00000001"
869 # Append to a file sets it dirty
870 cp -p /etc/passwd $f.append || error "could not create file"
871 $LFS hsm_set --exists $f.append ||
872 error "user could not change hsm flags"
873 dd if=/etc/passwd of=$f.append bs=1 count=3\
874 conv=notrunc oflag=append status=noxfer ||
875 file_creation_failure dd $f.append $?
876 check_hsm_flags $f.append "0x00000003"
878 # Modify a file sets it dirty
879 cp -p /etc/passwd $f.modify || error "could not create file"
880 $LFS hsm_set --exists $f.modify ||
881 error "user could not change hsm flags"
882 dd if=/dev/zero of=$f.modify bs=1 count=3\
883 conv=notrunc status=noxfer ||
884 file_creation_failure dd $f.modify $?
885 check_hsm_flags $f.modify "0x00000003"
887 # Open O_TRUNC sets dirty
888 cp -p /etc/passwd $f.trunc || error "could not create file"
889 $LFS hsm_set --exists $f.trunc ||
890 error "user could not change hsm flags"
891 cp /etc/group $f.trunc || error "could not override a file"
892 check_hsm_flags $f.trunc "0x00000003"
894 # Mmapped a file sets dirty
895 cp -p /etc/passwd $f.mmap || error "could not create file"
896 $LFS hsm_set --exists $f.mmap ||
897 error "user could not change hsm flags"
898 multiop $f.mmap OSMWUc || error "could not mmap a file"
899 check_hsm_flags $f.mmap "0x00000003"
901 run_test 3 "Check file dirtyness when opening for write"
905 local f=$DIR/$tdir/$tfile
906 local fid=$(make_small $f)
909 local st=$(get_request_state $fid CANCEL)
910 [[ -z "$st" ]] || error "hsm_cancel must not be registered (state=$st)"
912 run_test 4 "Useless cancel must not be registered"
915 # test needs a running copytool
919 local f=$DIR/$tdir/$tfile
920 local fid=$(copy_file /etc/passwd $f)
922 wait_request_state $fid ARCHIVE SUCCEED
924 check_hsm_flags $f "0x00000009"
928 run_test 8 "Test default archive number"
932 local f=$DIR/$tdir/$tfile
933 local fid=$(copy_file /etc/passwd $f)
934 # we do not use the default one to be sure
935 local new_an=$((HSM_ARCHIVE_NUMBER + 1))
937 copytool_setup $SINGLEAGT $MOUNT $new_an
938 $LFS hsm_archive --archive $new_an $f
939 wait_request_state $fid ARCHIVE SUCCEED
941 check_hsm_flags $f "0x00000009"
945 run_test 9 "Use of explicit archive number, with dedicated copytool"
948 needclients 3 || return 0
954 copytool_cleanup $(comma_list $(agts_nodes))
956 # start all of the copytools
957 for n in $(seq $AGTCOUNT); do
961 trap "copytool_cleanup $(comma_list $(agts_nodes))" EXIT
964 for n in $(seq $AGTCOUNT); do
965 file=$DIR/$tdir/$tfile.$n
966 fid=$(make_small $file)
968 $LFS hsm_archive $file || error "could not archive file $file"
969 wait_request_state $fid ARCHIVE SUCCEED
970 check_hsm_flags $file "0x00000009"
974 copytool_cleanup $(comma_list $(agts_nodes))
976 run_test 9a "Multiple remote agents"
979 # test needs a running copytool
982 mkdir -p $DIR/$tdir/d1
983 local f=$DIR/$tdir/$tfile
984 local fid=$(copy_file /etc/hosts $f)
985 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
986 error "hsm_archive failed"
987 wait_request_state $fid ARCHIVE SUCCEED
989 local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
990 error "fid $fid not in archive $HSM_ARCHIVE"
991 echo "Verifying content"
992 do_facet $SINGLEAGT diff $f $AFILE || error "archived file differs"
993 echo "Verifying hsm state "
994 check_hsm_flags $f "0x00000009"
996 echo "Verifying archive number is $HSM_ARCHIVE_NUMBER"
997 local st=$(get_hsm_archive_id $f)
998 [[ $st == $HSM_ARCHIVE_NUMBER ]] ||
999 error "Wrong archive number, $st != $HSM_ARCHIVE_NUMBER"
1004 run_test 10a "Archive a file"
1007 # test needs a running copytool
1011 local f=$DIR/$tdir/$tfile
1012 local fid=$(copy_file /etc/hosts $f)
1013 $LFS hsm_archive $f || error "archive request failed"
1014 wait_request_state $fid ARCHIVE SUCCEED
1016 $LFS hsm_archive $f || error "archive of non dirty file failed"
1017 local cnt=$(get_request_count $fid ARCHIVE)
1018 [[ "$cnt" == "1" ]] ||
1019 error "archive of non dirty file must not make a request"
1023 run_test 10b "Archive of non dirty file must work without doing request"
1026 # test needs a running copytool
1030 local f=$DIR/$tdir/$tfile
1031 local fid=$(copy_file /etc/hosts $f)
1032 $LFS hsm_set --noarchive $f
1033 $LFS hsm_archive $f && error "archive a noarchive file must fail"
1037 run_test 10c "Check forbidden archive"
1040 # test needs a running copytool
1044 local f=$DIR/$tdir/$tfile
1045 local fid=$(copy_file /etc/hosts $f)
1046 $LFS hsm_archive $f || error "cannot archive $f"
1047 wait_request_state $fid ARCHIVE SUCCEED
1049 local ar=$(get_hsm_archive_id $f)
1050 local dflt=$(get_hsm_param default_archive_id)
1051 [[ $ar == $dflt ]] ||
1052 error "archived file is not on default archive: $ar != $dflt"
1056 run_test 10d "Archive a file on the default archive id"
1060 copy2archive /etc/hosts $tdir/$tfile
1061 local f=$DIR/$tdir/$tfile
1063 import_file $tdir/$tfile $f
1064 echo -n "Verifying released state: "
1065 check_hsm_flags $f "0x0000000d"
1067 local LSZ=$(stat -c "%s" $f)
1068 local ASZ=$(do_facet $SINGLEAGT stat -c "%s" $HSM_ARCHIVE/$tdir/$tfile)
1070 echo "Verifying imported size $LSZ=$ASZ"
1071 [[ $LSZ -eq $ASZ ]] || error "Incorrect size $LSZ != $ASZ"
1072 echo -n "Verifying released pattern: "
1073 local PTRN=$($GETSTRIPE -L $f)
1075 [[ $PTRN == 80000001 ]] || error "Is not released"
1076 local fid=$(path2fid $f)
1077 echo "Verifying new fid $fid in archive"
1079 local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
1080 error "fid $fid not in archive $HSM_ARCHIVE"
1082 run_test 11a "Import a file"
1085 # test needs a running copytool
1089 local f=$DIR/$tdir/$tfile
1090 local fid=$(copy_file /etc/hosts $f)
1091 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
1092 error "hsm_archive failed"
1093 wait_request_state $fid ARCHIVE SUCCEED
1095 local FILE_HASH=$(md5sum $f)
1100 echo "$FILE_HASH" | md5sum -c
1102 [[ $? -eq 0 ]] || error "Restored file differs"
1106 run_test 11b "Import a deleted file using its FID"
1109 # test needs a running copytool
1113 copy2archive /etc/hosts $tdir/$tfile
1115 local f=$DIR/$tdir/$tfile
1116 import_file $tdir/$tfile $f
1117 local f2=$DIR2/$tdir/$tfile
1118 echo "Verifying released state: "
1119 check_hsm_flags $f2 "0x0000000d"
1121 local fid=$(path2fid $f2)
1122 $LFS hsm_restore $f2
1123 wait_request_state $fid RESTORE SUCCEED
1125 echo "Verifying file state: "
1126 check_hsm_flags $f2 "0x00000009"
1128 do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
1130 [[ $? -eq 0 ]] || error "Restored file differs"
1134 run_test 12a "Restore an imported file explicitly"
1137 # test needs a running copytool
1141 copy2archive /etc/hosts $tdir/$tfile
1143 local f=$DIR/$tdir/$tfile
1144 import_file $tdir/$tfile $f
1145 echo "Verifying released state: "
1146 check_hsm_flags $f "0x0000000d"
1148 cat $f > /dev/null || error "File read failed"
1150 echo "Verifying file state after restore: "
1151 check_hsm_flags $f "0x00000009"
1153 do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
1155 [[ $? -eq 0 ]] || error "Restored file differs"
1159 run_test 12b "Restore an imported file implicitly"
1162 [ "$OSTCOUNT" -lt "2" ] && skip_env "skipping 2-stripe test" && return
1164 # test needs a running copytool
1168 local f=$DIR/$tdir/$tfile
1169 $LFS setstripe -c 2 $f
1171 fid=$(make_large_for_striping $f)
1172 [ $? != 0 ] && skip "not enough free space" && return
1174 local FILE_CRC=$(md5sum $f)
1176 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1177 wait_request_state $fid ARCHIVE SUCCEED
1178 $LFS hsm_release $f || error "release $f failed"
1180 echo "$FILE_CRC" | md5sum -c
1182 [[ $? -eq 0 ]] || error "Restored file differs"
1186 run_test 12c "Restore a file with stripe of 2"
1189 # test needs a running copytool
1194 local f=$DIR/$tdir/$tfile
1195 local fid=$(copy_file /etc/hosts $f)
1196 $LFS hsm_restore $f || error "restore of non archived file failed"
1197 local cnt=$(get_request_count $fid RESTORE)
1198 [[ "$cnt" == "0" ]] ||
1199 error "restore non archived must not make a request"
1200 $LFS hsm_archive $f ||
1201 error "archive request failed"
1202 wait_request_state $fid ARCHIVE SUCCEED
1203 $LFS hsm_restore $f ||
1204 error "restore of non released file failed"
1205 local cnt=$(get_request_count $fid RESTORE)
1206 [[ "$cnt" == "0" ]] ||
1207 error "restore a non dirty file must not make a request"
1211 run_test 12d "Restore of a non archived, non released file must work"\
1212 " without doing request"
1215 # test needs a running copytool
1218 mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
1219 local f=$DIR/$tdir/$tfile
1220 local fid=$(copy_file /etc/hosts $f)
1221 $LFS hsm_archive $f || error "archive request failed"
1222 wait_request_state $fid ARCHIVE SUCCEED
1225 cat /etc/hosts >> $f
1229 $LFS hsm_restore $f && error "restore a dirty file must fail"
1233 run_test 12e "Check forbidden restore"
1236 # test needs a running copytool
1240 local f=$DIR/$tdir/$tfile
1241 local fid=$(copy_file /etc/hosts $f)
1243 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1244 wait_request_state $fid ARCHIVE SUCCEED
1245 $LFS hsm_release $f || error "release of $f failed"
1247 wait_request_state $fid RESTORE SUCCEED
1249 echo -n "Verifying file state: "
1250 check_hsm_flags $f "0x00000009"
1252 diff -q /etc/hosts $f
1254 [[ $? -eq 0 ]] || error "Restored file differs"
1258 run_test 12f "Restore a released file explicitly"
1261 # test needs a running copytool
1265 local f=$DIR/$tdir/$tfile
1266 local fid=$(copy_file /etc/hosts $f)
1268 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1269 wait_request_state $fid ARCHIVE SUCCEED
1270 $LFS hsm_release $f || error "release of $f failed"
1272 diff -q /etc/hosts $f
1275 # we check we had a restore done
1276 wait_request_state $fid RESTORE SUCCEED
1278 [[ $st -eq 0 ]] || error "Restored file differs"
1282 run_test 12g "Restore a released file implicitly"
1285 needclients 2 || return 0
1287 # test needs a running copytool
1291 local f=$DIR/$tdir/$tfile
1292 local fid=$(copy_file /etc/hosts $f)
1294 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1295 wait_request_state $fid ARCHIVE SUCCEED
1296 $LFS hsm_release $f || error "release of $f failed"
1298 do_node $CLIENT2 diff -q /etc/hosts $f
1301 # we check we had a restore done
1302 wait_request_state $fid RESTORE SUCCEED
1304 [[ $st -eq 0 ]] || error "Restored file differs"
1308 run_test 12h "Restore a released file implicitly from a second node"
1311 # test needs a running copytool
1315 local f=$DIR/$tdir/$tfile
1316 local fid=$(copy_file /etc/passwd $f)
1317 $LFS hsm_archive $f || error "archive of $f failed"
1318 wait_request_state $fid ARCHIVE SUCCEED
1320 $LFS hsm_release $f || error "release of $f failed"
1324 [[ $? -eq 0 ]] || error "Restored file differs"
1328 run_test 12m "Archive/release/implicit restore"
1331 # test needs a running copytool
1335 copy2archive /etc/hosts $tdir/$tfile
1337 local f=$DIR/$tdir/$tfile
1338 import_file $tdir/$tfile $f
1340 do_facet $SINGLEAGT cmp /etc/hosts $f ||
1341 error "Restored file differs"
1343 $LFS hsm_release $f || error "release of $f failed"
1347 run_test 12n "Import/implicit restore/release"
1350 # test needs a running copytool
1354 local f=$DIR/$tdir/$tfile
1355 local fid=$(copy_file /etc/hosts $f)
1357 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1358 wait_request_state $fid ARCHIVE SUCCEED
1359 $LFS hsm_release $f || error "release of $f failed"
1361 #define OBD_FAIL_MDS_HSM_SWAP_LAYOUTS 0x152
1362 do_facet $SINGLEMDS lctl set_param fail_loc=0x152
1364 # set no retry action mode
1367 diff -q /etc/hosts $f
1370 # we check we had a restore failure
1371 wait_request_state $fid RESTORE FAILED
1373 [[ $st -eq 0 ]] && error "Restore must fail"
1375 # remove no retry action mode
1378 # check file is still released
1379 check_hsm_flags $f "0x0000000d"
1381 # retry w/o failure injection
1382 do_facet $SINGLEMDS lctl set_param fail_loc=0
1384 # to be sure previous RESTORE result is gone
1386 wait_for_grace_delay
1388 diff -q /etc/hosts $f
1391 # we check we had a restore done
1392 wait_request_state $fid RESTORE SUCCEED
1394 [[ $st -eq 0 ]] || error "Restored file differs"
1398 run_test 12o "Layout-swap failure during Restore leaves file released"
1401 # test needs a running copytool
1405 local f=$DIR/$tdir/$tfile
1406 local fid=$(copy_file /etc/hosts $f)
1408 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1409 wait_request_state $fid ARCHIVE SUCCEED
1410 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1411 $LFS hsm_release $f || error "cannot release $f"
1412 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1413 $LFS hsm_release $f || error "cannot release $f"
1414 do_facet $SINGLEAGT cat $f > /dev/null || error "cannot cat $f"
1418 run_test 12p "implicit restore of a file on copytool mount point"
1420 cleanup_test_12q() {
1422 zconf_umount $(facet_host $SINGLEAGT) $MOUNT3 ||
1423 error "cannot umount $MOUNT3 on $SINGLEAGT"
1428 zconf_mount $(facet_host $SINGLEAGT) $MOUNT3 ||
1429 error "cannot mount $MOUNT3 on $SINGLEAGT"
1431 trap cleanup_test_12q EXIT
1433 # test needs a running copytool
1434 copytool_setup $SINGLEAGT $MOUNT3
1437 local f=$DIR/$tdir/$tfile
1438 local f2=$DIR2/$tdir/$tfile
1439 local fid=$(make_small $f)
1440 local orig_size=$(stat -c "%s" $f)
1442 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1443 wait_request_state $fid ARCHIVE SUCCEED
1445 $LFS hsm_release $f || error "could not release file"
1446 check_hsm_flags $f "0x0000000d"
1448 search_and_kill_copytool
1450 search_copytools && error "Copytool should have stopped"
1452 cat $f > /dev/null &
1454 # wait a bit to allow implicit restore request to be handled.
1455 # if not, next stat would also block on layout-lock.
1458 local size=$(stat -c "%s" $f2)
1459 [ $size -eq $orig_size ] ||
1460 error "$f2: wrong size after archive: $size != $orig_size"
1462 HSM_ARCHIVE_PURGE=false copytool_setup $SINGLEAGT /mnt/lustre3
1466 size=$(stat -c "%s" $f)
1467 [ $size -eq $orig_size ] ||
1468 error "$f: wrong size after restore: $size != $orig_size"
1470 size=$(stat -c "%s" $f2)
1471 [ $size -eq $orig_size ] ||
1472 error "$f2: wrong size after restore: $size != $orig_size"
1476 size=$(stat -c "%s" $f)
1478 error "$f: wrong size after overwrite: $size != 0"
1480 size=$(stat -c "%s" $f2)
1482 error "$f2: wrong size after overwrite: $size != 0"
1485 zconf_umount $(facet_host $SINGLEAGT) $MOUNT3 ||
1486 error "cannot umount $MOUNT3 on $SINGLEAGT"
1488 run_test 12q "file attributes are refreshed after restore"
1491 # test needs a running copytool
1494 local ARC_SUBDIR="import.orig"
1498 # populate directory to be imported
1499 for d in $(seq 1 10); do
1500 local CURR_DIR="$HSM_ARCHIVE/$ARC_SUBDIR/dir.$d"
1501 do_facet $SINGLEAGT mkdir -p "$CURR_DIR"
1502 for f in $(seq 1 10); do
1503 CURR_FILE="$CURR_DIR/$tfile.$f"
1504 # write file-specific data
1505 do_facet $SINGLEAGT \
1506 "echo d=$d, f=$f, dir=$CURR_DIR, "\
1507 "file=$CURR_FILE > $CURR_FILE"
1511 import_file "$ARC_SUBDIR" $DIR/$tdir
1512 # diff lustre content and origin (triggers file restoration)
1513 # there must be 10x10 identical files, and no difference
1514 local cnt_ok=$(do_facet $SINGLEAGT diff -rs $HSM_ARCHIVE/$ARC_SUBDIR \
1515 $DIR/$tdir/$ARC_SUBDIR | grep identical | wc -l)
1516 local cnt_diff=$(do_facet $SINGLEAGT diff -r $HSM_ARCHIVE/$ARC_SUBDIR \
1517 $DIR/$tdir/$ARC_SUBDIR | wc -l)
1519 [ $cnt_diff -eq 0 ] ||
1520 error "$cnt_diff imported files differ from read data"
1521 [ $cnt_ok -eq 100 ] ||
1522 error "not enough identical files ($cnt_ok != 100)"
1526 run_test 13 "Recursively import and restore a directory"
1529 # test needs a running copytool
1534 local f=$DIR/$tdir/$tfile
1535 local fid=$(make_small $f)
1536 local sum=$(md5sum $f | awk '{print $1}')
1537 $LFS hsm_archive $f || error "could not archive file"
1538 wait_request_state $fid ARCHIVE SUCCEED
1542 # create released file (simulate llapi_hsm_import call)
1544 local fid2=$(path2fid $f)
1545 $LFS hsm_set --archived --exists $f || error "could not force hsm flags"
1546 $LFS hsm_release $f || error "could not release file"
1548 # rebind the archive to the newly created file
1549 echo "rebind $fid to $fid2"
1551 do_facet $SINGLEAGT \
1552 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
1553 --rebind $fid $fid2 $DIR" || error "could not rebind file"
1555 # restore file and compare md5sum
1556 local sum2=$(md5sum $f | awk '{print $1}')
1558 [[ $sum == $sum2 ]] || error "md5sum mismatch after restore"
1562 run_test 14 "Rebind archived file to a new fid"
1565 # test needs a running copytool
1570 local f=$DIR/$tdir/$tfile
1572 local tmpfile=$SHARED_DIRECTORY/tmp.$$
1576 for i in $(seq 1 $count); do
1577 fids[$i]=$(make_small $f.$i)
1578 sums[$i]=$(md5sum $f.$i | awk '{print $1}')
1579 $LFS hsm_archive $f.$i || error "could not archive file"
1581 wait_all_done $(($count*60))
1585 for i in $(seq 1 $count); do
1588 local fid2=$(path2fid $f.$i)
1589 # add the rebind operation to the list
1590 echo ${fids[$i]} $fid2 >> $tmpfile
1592 # set it released (simulate llapi_hsm_import call)
1593 $LFS hsm_set --archived --exists $f.$i ||
1594 error "could not force hsm flags"
1595 $LFS hsm_release $f.$i || error "could not release file"
1597 nl=$(wc -l < $tmpfile)
1598 [[ $nl == $count ]] || error "$nl files in list, $count expected"
1600 echo "rebind list of files"
1601 do_facet $SINGLEAGT \
1602 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
1603 --rebind $tmpfile $DIR" || error "could not rebind file list"
1605 # restore files and compare md5sum
1606 for i in $(seq 1 $count); do
1607 local sum2=$(md5sum $f.$i | awk '{print $1}')
1608 [[ $sum2 == ${sums[$i]} ]] ||
1609 error "md5sum mismatch after restore ($sum2 != ${sums[$i]})"
1615 run_test 15 "Rebind a list of files"
1618 # test needs a running copytool
1622 # create a known size file so we can verify transfer speed
1625 dd if=/dev/zero of=$ref bs=1M count=20
1628 local f=$DIR/$tdir/$tfile
1629 local fid=$(copy_file $ref $f)
1631 local start=$(date +%s)
1633 wait_request_state $fid ARCHIVE SUCCEED
1634 local end=$(date +%s)
1635 local duration=$((end - start))
1637 [[ $duration -ge $goal ]] ||
1638 error "Transfer is too fast $duration < $goal"
1642 run_test 16 "Test CT bandwith control option"
1647 local f=$DIR/$tdir/$tfile
1648 touch $f || error "touch $f failed"
1650 # Could not release a non-archived file
1651 $LFS hsm_release $f && error "release should not succeed"
1653 # For following tests, we must test them with HS_ARCHIVED set
1654 $LFS hsm_set --exists --archived $f || error "could not add flag"
1656 # Could not release a file if no-release is set
1657 $LFS hsm_set --norelease $f || error "could not add flag"
1658 $LFS hsm_release $f && error "release should not succeed"
1659 $LFS hsm_clear --norelease $f || error "could not remove flag"
1661 # Could not release a file if lost
1662 $LFS hsm_set --lost $f || error "could not add flag"
1663 $LFS hsm_release $f && error "release should not succeed"
1664 $LFS hsm_clear --lost $f || error "could not remove flag"
1666 # Could not release a file if dirty
1667 $LFS hsm_set --dirty $f || error "could not add flag"
1668 $LFS hsm_release $f && error "release should not succeed"
1669 $LFS hsm_clear --dirty $f || error "could not remove flag"
1671 run_test 20 "Release is not permitted"
1674 # test needs a running copytool
1678 local f=$DIR/$tdir/test_release
1680 # Create a file and check its states
1681 local fid=$(make_small $f)
1682 check_hsm_flags $f "0x00000000"
1684 # LU-4388/LU-4389 - ZFS does not report full number of blocks
1685 # used until file is flushed to disk
1686 if [ $(facet_fstype ost1) == "zfs" ]; then
1687 # this causes an OST_SYNC rpc to be sent
1688 dd if=/dev/zero of=$f bs=512 count=1 oflag=sync conv=notrunc,fsync
1689 # clear locks to reread file data
1690 cancel_lru_locks osc
1693 local orig_size=$(stat -c "%s" $f)
1694 local orig_blocks=$(stat -c "%b" $f)
1696 start_full_debug_logging
1698 $LFS hsm_archive $f || error "could not archive file"
1699 wait_request_state $fid ARCHIVE SUCCEED
1701 local blocks=$(stat -c "%b" $f)
1702 [ $blocks -eq $orig_blocks ] ||
1703 error "$f: wrong block number after archive: " \
1704 "$blocks != $orig_blocks"
1705 local size=$(stat -c "%s" $f)
1706 [ $size -eq $orig_size ] ||
1707 error "$f: wrong size after archive: $size != $orig_size"
1709 # Release and check states
1710 $LFS hsm_release $f || error "could not release file"
1711 check_hsm_flags $f "0x0000000d"
1713 blocks=$(stat -c "%b" $f)
1714 [ $blocks -gt 5 ] &&
1715 error "$f: too many blocks after release: $blocks > 5"
1716 size=$(stat -c "%s" $f)
1717 [ $size -ne $orig_size ] &&
1718 error "$f: wrong size after release: $size != $orig_size"
1720 # Check we can release an file without stripe info
1724 check_hsm_flags $f "0x00000000"
1725 $LFS hsm_archive $f || error "could not archive file"
1726 wait_request_state $fid ARCHIVE SUCCEED
1728 # Release and check states
1729 $LFS hsm_release $f || error "could not release file"
1730 check_hsm_flags $f "0x0000000d"
1732 # Release again a file that is already released is OK
1733 $LFS hsm_release $f || fail "second release should succeed"
1734 check_hsm_flags $f "0x0000000d"
1736 stop_full_debug_logging
1740 run_test 21 "Simple release tests"
1743 # test needs a running copytool
1748 local f=$DIR/$tdir/test_release
1749 local swap=$DIR/$tdir/test_swap
1751 # Create a file and check its states
1752 local fid=$(make_small $f)
1753 check_hsm_flags $f "0x00000000"
1755 $LFS hsm_archive $f || error "could not archive file"
1756 wait_request_state $fid ARCHIVE SUCCEED
1758 # Release and check states
1759 $LFS hsm_release $f || error "could not release file"
1760 check_hsm_flags $f "0x0000000d"
1763 $LFS swap_layouts $swap $f && error "swap_layouts should failed"
1768 run_test 22 "Could not swap a release file"
1771 # test needs a running copytool
1776 local f=$DIR/$tdir/test_mtime
1778 # Create a file and check its states
1779 local fid=$(make_small $f)
1780 check_hsm_flags $f "0x00000000"
1782 $LFS hsm_archive $f || error "could not archive file"
1783 wait_request_state $fid ARCHIVE SUCCEED
1785 # Set modification time in the past
1786 touch -m -a -d @978261179 $f
1788 # Release and check states
1789 $LFS hsm_release $f || error "could not release file"
1790 check_hsm_flags $f "0x0000000d"
1792 local MTIME=$(stat -c "%Y" $f)
1793 local ATIME=$(stat -c "%X" $f)
1794 [ $MTIME -eq "978261179" ] || fail "bad mtime: $MTIME"
1795 [ $ATIME -eq "978261179" ] || fail "bad atime: $ATIME"
1799 run_test 23 "Release does not change a/mtime (utime)"
1802 local file=$DIR/$tdir/$tfile
1811 # test needs a running copytool
1816 fid=$(make_small $file)
1818 # Create a file and check its states
1819 check_hsm_flags $file "0x00000000"
1821 # Ensure atime is less than mtime and ctime.
1825 atime0=$(stat -c "%X" $file)
1826 mtime0=$(stat -c "%Y" $file)
1827 ctime0=$(stat -c "%Z" $file)
1829 [ $atime0 -lt $mtime0 ] ||
1830 error "atime $atime0 is not less than mtime $mtime0"
1832 [ $atime0 -lt $ctime0 ] ||
1833 error "atime $atime0 is not less than ctime $ctime0"
1835 # Archive should not change any timestamps.
1836 $LFS hsm_archive $file || error "cannot archive '$file'"
1837 wait_request_state $fid ARCHIVE SUCCEED
1839 atime1=$(stat -c "%X" $file)
1840 mtime1=$(stat -c "%Y" $file)
1841 ctime1=$(stat -c "%Z" $file)
1843 [ $atime0 -eq $atime1 ] ||
1844 error "archive changed atime from $atime0 to $atime1"
1846 [ $mtime0 -eq $mtime1 ] ||
1847 error "archive changed mtime from $mtime0 to $mtime1"
1849 [ $ctime0 -eq $ctime1 ] ||
1850 error "archive changed ctime from $ctime0 to $ctime1"
1852 # Release should not change any timestamps.
1853 $LFS hsm_release $file || error "cannot release '$file'"
1854 check_hsm_flags $file "0x0000000d"
1856 atime1=$(stat -c "%X" $file)
1857 mtime1=$(stat -c "%Y" $file)
1858 ctime1=$(stat -c "%Z" $file)
1860 [ $atime0 -eq $atime1 ] ||
1861 error "release changed atime from $atime0 to $atime1"
1863 [ $mtime0 -eq $mtime1 ] ||
1864 error "release changed mtime from $mtime0 to $mtime1"
1866 [ $ctime0 -eq $ctime1 ] ||
1867 error "release changed ctime from $ctime0 to $ctime1"
1869 # Restore should not change atime or mtime and should not
1871 $LFS hsm_restore $file
1872 wait_request_state $fid RESTORE SUCCEED
1874 atime1=$(stat -c "%X" $file)
1875 mtime1=$(stat -c "%Y" $file)
1876 ctime1=$(stat -c "%Z" $file)
1878 [ $atime0 -eq $atime1 ] ||
1879 error "restore changed atime from $atime0 to $atime1"
1881 [ $mtime0 -eq $mtime1 ] ||
1882 error "restore changed mtime from $mtime0 to $mtime1"
1884 [ $ctime0 -eq $ctime1 ] ||
1885 error "restore changed ctime from $ctime0 to $ctime1"
1889 # Once more, after unmount and mount.
1890 umount_client $MOUNT || error "cannot unmount '$MOUNT'"
1891 mount_client $MOUNT || error "cannot mount '$MOUNT'"
1893 atime1=$(stat -c "%X" $file)
1894 mtime1=$(stat -c "%Y" $file)
1895 ctime1=$(stat -c "%Z" $file)
1897 [ $atime0 -eq $atime1 ] ||
1898 error "remount changed atime from $atime0 to $atime1"
1900 [ $mtime0 -eq $mtime1 ] ||
1901 error "remount changed mtime from $mtime0 to $mtime1"
1903 [ $ctime0 -eq $ctime1 ] ||
1904 error "remount changed ctime from $ctime0 to $ctime1"
1906 run_test 24a "Archive, release, and restore does not change a/mtime (i/o)"
1909 local file=$DIR/$tdir/$tfile
1915 # Test needs a running copytool.
1919 # Check that root can do HSM actions on a regular user's file.
1921 fid=$(make_small $file)
1922 sum0=$(md5sum $file)
1924 chown $RUNAS_ID:$RUNAS_GID $file ||
1925 error "cannot chown '$file' to '$RUNAS_ID'"
1927 chmod ugo-w $DIR/$tdir ||
1928 error "cannot chmod '$DIR/$tdir'"
1930 $LFS hsm_archive $file
1931 wait_request_state $fid ARCHIVE SUCCEED
1933 $LFS hsm_release $file
1934 check_hsm_flags $file "0x0000000d"
1936 $LFS hsm_restore $file
1937 wait_request_state $fid RESTORE SUCCEED
1939 # Check that ordinary user can get HSM state.
1940 $RUNAS $LFS hsm_state $file ||
1941 error "user '$RUNAS_ID' cannot get HSM state of '$file'"
1943 $LFS hsm_release $file
1944 check_hsm_flags $file "0x0000000d"
1946 # Check that ordinary user can accessed released file.
1947 sum1=$($RUNAS md5sum $file) ||
1948 error "user '$RUNAS_ID' cannot read '$file'"
1950 [ "$sum0" == "$sum1" ] ||
1951 error "md5sum mismatch for '$file'"
1955 run_test 24b "root can archive, release, and restore user files"
1957 cleanup_test_24c() {
1959 set_hsm_param user_request_mask RESTORE
1960 set_hsm_param group_request_mask RESTORE
1961 set_hsm_param other_request_mask RESTORE
1965 local file=$DIR/$tdir/$tfile
1966 local action=archive
1971 # test needs a running copytool
1976 # Save the default masks and check that cleanup_24c will
1977 # restore the request masks correctly.
1978 user_save=$(get_hsm_param user_request_mask)
1979 group_save=$(get_hsm_param group_request_mask)
1980 other_save=$(get_hsm_param other_request_mask)
1982 [ "$user_save" == RESTORE ] ||
1983 error "user_request_mask is '$user_save' expected 'RESTORE'"
1984 [ "$group_save" == RESTORE ] ||
1985 error "group_request_mask is '$group_save' expected 'RESTORE'"
1986 [ "$other_save" == RESTORE ] ||
1987 error "other_request_mask is '$other_save' expected 'RESTORE'"
1989 trap cleanup_test_24c EXIT
1994 chown $RUNAS_ID:nobody $file ||
1995 error "cannot chown '$file' to '$RUNAS_ID:nobody'"
1997 set_hsm_param user_request_mask ""
1998 $RUNAS $LFS hsm_$action $file &&
1999 error "$action by user should fail"
2001 set_hsm_param user_request_mask $action
2002 $RUNAS $LFS hsm_$action $file ||
2003 error "$action by user should succeed"
2008 chown nobody:$RUNAS_GID $file ||
2009 error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
2011 set_hsm_param group_request_mask ""
2012 $RUNAS $LFS hsm_$action $file &&
2013 error "$action by group should fail"
2015 set_hsm_param group_request_mask $action
2016 $RUNAS $LFS hsm_$action $file ||
2017 error "$action by group should succeed"
2022 chown nobody:nobody $file ||
2023 error "cannot chown '$file' to 'nobody:nobody'"
2025 set_hsm_param other_request_mask ""
2026 $RUNAS $LFS hsm_$action $file &&
2027 error "$action by other should fail"
2029 set_hsm_param other_request_mask $action
2030 $RUNAS $LFS hsm_$action $file ||
2031 error "$action by other should succeed"
2036 run_test 24c "check that user,group,other request masks work"
2038 cleanup_test_24d() {
2040 mount -o remount,rw $MOUNT2
2044 local file1=$DIR/$tdir/$tfile
2045 local file2=$DIR2/$tdir/$tfile
2053 fid1=$(make_small $file1)
2055 trap cleanup_test_24d EXIT
2057 mount -o remount,ro $MOUNT2
2059 fid2=$(path2fid $file2)
2060 [ "$fid1" == "$fid2" ] ||
2061 error "FID mismatch '$fid1' != '$fid2'"
2063 $LFS hsm_archive $file2 &&
2064 error "archive should fail on read-only mount"
2065 check_hsm_flags $file1 "0x00000000"
2067 $LFS hsm_archive $file1
2068 wait_request_state $fid1 ARCHIVE SUCCEED
2070 $LFS hsm_release $file1
2071 $LFS hsm_restore $file2
2072 wait_request_state $fid1 RESTORE SUCCEED
2074 $LFS hsm_release $file1 || error "cannot release '$file1'"
2075 dd if=$file2 of=/dev/null bs=1M || "cannot read '$file2'"
2077 $LFS hsm_release $file2 &&
2078 error "release should fail on read-only mount"
2083 run_test 24d "check that read-only mounts are respected"
2090 local f=$DIR/$tdir/$tfile
2093 fid=$(make_small $f) || error "cannot create $f"
2094 $LFS hsm_archive $f || error "cannot archive $f"
2095 wait_request_state $fid ARCHIVE SUCCEED
2096 $LFS hsm_release $f || error "cannot release $f"
2097 while ! $LFS hsm_state $f | grep released; do
2101 tar -cf $TMP/$tfile.tar $DIR/$tdir || error "cannot tar $DIR/$tdir"
2105 run_test 24e "tar succeeds on HSM released files" # LU-6213
2109 # test needs a running copytool
2112 mkdir -p $DIR/$tdir/d1
2113 local f=$DIR/$tdir/$tfile
2114 local fid=$(copy_file /etc/hosts $f)
2117 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
2118 error "hsm_archive failed"
2119 wait_request_state $fid ARCHIVE SUCCEED
2120 $LFS hsm_release $f || error "cannot release $f"
2121 tar --xattrs -cvf $f.tar -C $DIR/$tdir $tfile
2124 tar --xattrs -xvf $f.tar -C $DIR/$tdir ||
2125 error "Can not recover the tar contents"
2127 echo "Sum0 = $sum0, sum1 = $sum1"
2128 [ "$sum0" == "$sum1" ] || error "md5sum mismatch for '$tfile'"
2132 run_test 24f "root can archive, release, and restore tar files"
2135 # test needs a running copytool
2139 copy2archive /etc/hosts $tdir/$tfile
2141 local f=$DIR/$tdir/$tfile
2143 import_file $tdir/$tfile $f
2145 $LFS hsm_set --lost $f
2150 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2154 run_test 25a "Restore lost file (HS_LOST flag) from import"\
2155 " (Operation not permitted)"
2158 # test needs a running copytool
2163 local f=$DIR/$tdir/$tfile
2164 local fid=$(copy_file /etc/passwd $f)
2166 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2167 wait_request_state $fid ARCHIVE SUCCEED
2170 $LFS hsm_set --lost $f
2174 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2178 run_test 25b "Restore lost file (HS_LOST flag) after release"\
2179 " (Operation not permitted)"
2182 # test needs a running copytool
2186 local f=$DIR/$tdir/$tfile
2188 fid=$(make_large_for_progress $f)
2189 [ $? != 0 ] && skip "not enough free space" && return
2191 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2192 wait_request_state $fid ARCHIVE SUCCEED
2195 wait_request_state $fid REMOVE SUCCEED
2197 check_hsm_flags $f "0x00000000"
2201 run_test 26 "Remove the archive of a valid file"
2204 # test needs a running copytool
2208 make_archive $tdir/$tfile
2209 local f=$DIR/$tdir/$tfile
2210 import_file $tdir/$tfile $f
2211 local fid=$(path2fid $f)
2215 [[ $? != 0 ]] || error "Remove of a released file should fail"
2219 run_test 27a "Remove the archive of an imported file (Operation not permitted)"
2222 # test needs a running copytool
2226 local f=$DIR/$tdir/$tfile
2228 fid=$(make_large_for_progress $f)
2229 [ $? != 0 ] && skip "not enough free space" && return
2231 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2232 wait_request_state $fid ARCHIVE SUCCEED
2237 [[ $? != 0 ]] || error "Remove of a released file should fail"
2241 run_test 27b "Remove the archive of a relased file (Operation not permitted)"
2244 # test needs a running copytool
2248 local f=$DIR/$tdir/$tfile
2250 fid=$(make_large_for_progress $f)
2251 [ $? != 0 ] && skip "not enough free space" && return
2253 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2254 wait_request_state $fid ARCHIVE SUCCEED
2263 wait_request_state $fid REMOVE SUCCEED
2267 run_test 28 "Concurrent archive/file remove"
2270 # Tests --mntpath and --archive options
2273 copytool_setup $SINGLEAGT $MOUNT $archive_id
2275 # Bad archive number
2276 $LFS hsm_remove -m $MOUNT -a 33 0x857765760:0x8:0x2 2>&1 |
2277 grep "Invalid argument" ||
2278 error "unexpected hsm_remove failure (1)"
2280 # mntpath is present but file is given
2281 $LFS hsm_remove --mntpath $MOUNT --archive 30 /qwerty/uyt 2>&1 |
2282 grep "hsm: '/qwerty/uyt' is not a valid FID" ||
2283 error "unexpected hsm_remove failure (2)"
2287 run_test 29a "Tests --mntpath and --archive options"
2290 # test needs a running copytool
2294 local f=$DIR/$tdir/$tfile
2295 local fid=$(make_small $f)
2297 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2298 wait_request_state $fid ARCHIVE SUCCEED
2302 $LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER $fid
2303 wait_request_state $fid REMOVE SUCCEED
2307 run_test 29b "Archive/delete/remove by FID from the archive."
2310 # test needs a running copytool
2314 local fid1=$(make_small $DIR/$tdir/$tfile-1)
2315 local fid2=$(make_small $DIR/$tdir/$tfile-2)
2316 local fid3=$(make_small $DIR/$tdir/$tfile-3)
2318 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/$tfile-[1-3]
2319 wait_request_state $fid1 ARCHIVE SUCCEED
2320 wait_request_state $fid2 ARCHIVE SUCCEED
2321 wait_request_state $fid3 ARCHIVE SUCCEED
2323 rm -f $DIR/$tdir/$tfile-[1-3]
2325 echo $fid1 > $DIR/$tdir/list
2326 echo $fid2 >> $DIR/$tdir/list
2327 echo $fid3 >> $DIR/$tdir/list
2329 $LFS hsm_remove -m $MOUNT -a $HSM_ARCHIVE_NUMBER \
2330 --filelist $DIR/$tdir/list
2331 wait_request_state $fid1 REMOVE SUCCEED
2332 wait_request_state $fid2 REMOVE SUCCEED
2333 wait_request_state $fid3 REMOVE SUCCEED
2337 run_test 29c "Archive/delete/remove by FID, using a file list."
2340 # restore at exec cannot work on agent node (because of Linux kernel
2341 # protection of executables)
2342 needclients 2 || return 0
2344 # test needs a running copytool
2348 copy2archive /bin/true $tdir/$tfile
2350 local f=$DIR/$tdir/true
2351 import_file $tdir/$tfile $f
2353 local fid=$(path2fid $f)
2355 # set no retry action mode
2361 # remove no try action mode
2365 [[ $st == 0 ]] || error "Failed to exec a released file"
2369 run_test 30a "Restore at exec (import case)"
2372 # restore at exec cannot work on agent node (because of Linux kernel
2373 # protection of executables)
2374 needclients 2 || return 0
2376 # test needs a running copytool
2380 local f=$DIR/$tdir/true
2381 local fid=$(copy_file /bin/true $f)
2383 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2384 wait_request_state $fid ARCHIVE SUCCEED
2387 # set no retry action mode
2393 # remove no try action mode
2397 [[ $st == 0 ]] || error "Failed to exec a released file"
2401 run_test 30b "Restore at exec (release case)"
2404 needclients 2 || return 0
2406 # test needs a running copytool
2410 local f=$DIR/$tdir/SLEEP
2411 local slp_sum1=$(md5sum /bin/sleep)
2412 local fid=$(copy_file /bin/sleep $f)
2414 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2415 wait_request_state $fid ARCHIVE SUCCEED
2417 check_hsm_flags $f "0x0000000d"
2418 # set no retry action mode
2420 do_node $CLIENT2 "$f 10" &
2424 [[ $? == 0 ]] && error "Update during exec of released file must fail"
2426 [[ $? == 0 ]] || error "Execution failed during run"
2428 if [[ $? != 0 ]]; then
2429 local slp_sum2=$(md5sum /bin/sleep)
2430 # in case sleep file is modified during the test
2431 [[ $slp_sum1 == $slp_sum2 ]] &&
2432 error "Binary overwritten during exec"
2436 # remove no try action mode
2438 check_hsm_flags $f "0x00000009"
2442 run_test 30c "Update during exec of released file must fail"
2444 restore_and_check_size() {
2447 local s=$(stat -c "%s" $f)
2449 local st=$(get_hsm_flags $f)
2453 while [[ "$st" != "0x00000009" && $cpt -le 10 ]]
2455 n=$(stat -c "%s" $f)
2456 # we echo in both cases to show stat is not
2458 if [[ $n != $s ]]; then
2459 echo "size seen is $n != $s"
2462 echo "size seen is right: $n == $s"
2464 st=$(get_hsm_flags $f)
2468 if [[ $cpt -lt 10 ]]; then
2471 echo " restore is too long"
2472 wait_request_state $fid RESTORE SUCCEED
2478 # test needs a running copytool
2483 make_archive $tdir/$tfile
2484 local f=$DIR/$tdir/$tfile
2485 import_file $tdir/$tfile $f
2486 local fid=$($LFS path2fid $f)
2487 HSM_ARCHIVE_PURGE=false copytool_setup
2489 restore_and_check_size $f $fid
2492 [[ $err -eq 0 ]] || error "File size changed during restore"
2496 run_test 31a "Import a large file and check size during restore"
2500 # test needs a running copytool
2505 local f=$DIR/$tdir/$tfile
2507 fid=$(make_large_for_progress $f)
2508 [ $? != 0 ] && skip "not enough free space" && return
2510 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2511 wait_request_state $fid ARCHIVE SUCCEED
2514 restore_and_check_size $f $fid
2517 [[ $err -eq 0 ]] || error "File size changed during restore"
2521 run_test 31b "Restore a large unaligned file and check size during restore"
2524 # test needs a running copytool
2529 local f=$DIR/$tdir/$tfile
2531 fid=$(make_large_for_progress_aligned $f)
2532 [ $? != 0 ] && skip "not enough free space" && return
2534 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2535 wait_request_state $fid ARCHIVE SUCCEED
2538 restore_and_check_size $f $fid
2541 [[ $err -eq 0 ]] || error "File size changed during restore"
2545 run_test 31c "Restore a large aligned file and check size during restore"
2548 # test needs a running copytool
2553 local f=$DIR/$tdir/$tfile
2555 fid=$(make_large_for_progress $f)
2556 [ $? != 0 ] && skip "not enough free space" && return
2558 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2559 wait_request_state $fid ARCHIVE SUCCEED
2562 # to be sure wait_all_done will not be mislead by previous tests
2565 wait_for_grace_delay
2566 # Also raise grace_delay significantly so the Canceled
2567 # Restore action will stay enough long avail.
2568 local old_grace=$(get_hsm_param grace_delay)
2569 set_hsm_param grace_delay 100
2571 md5sum $f >/dev/null &
2573 wait_request_state $fid RESTORE STARTED
2578 # Check restore trigger process was killed
2579 local killed=$(ps -o pid,comm hp $pid >/dev/null)
2583 # instead of waiting+checking both Restore and Cancel ops
2584 # sequentially, wait for both to be finished and then check
2586 wait_all_done 100 $fid
2587 local rstate=$(get_request_state $fid RESTORE)
2588 local cstate=$(get_request_state $fid CANCEL)
2590 # restore orig grace_delay.
2591 set_hsm_param grace_delay $old_grace
2593 if [[ "$rstate" == "CANCELED" ]] ; then
2594 [[ "$cstate" == "SUCCEED" ]] ||
2595 error "Restore state is CANCELED and Cancel state " \
2596 "is not SUCCEED but $cstate"
2597 echo "Restore state is CANCELED, Cancel state is SUCCEED"
2598 elif [[ "$rstate" == "SUCCEED" ]] ; then
2599 [[ "$cstate" == "FAILED" ]] ||
2600 error "Restore state is SUCCEED and Cancel state " \
2601 "is not FAILED but $cstate"
2602 echo "Restore state is SUCCEED, Cancel state is FAILED"
2604 error "Restore state is $rstate and Cancel state is $cstate"
2608 error "Cannot kill process waiting for restore ($killed)"
2612 run_test 33 "Kill a restore waiting process"
2615 # test needs a running copytool
2620 local f=$DIR/$tdir/$tfile
2622 fid=$(make_large_for_progress $f)
2623 [ $? != 0 ] && skip "not enough free space" && return
2625 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2626 wait_request_state $fid ARCHIVE SUCCEED
2629 md5sum $f >/dev/null &
2631 wait_request_state $fid RESTORE STARTED
2633 rm $f || error "rm $f failed"
2634 # rm must not block during restore
2635 wait_request_state $fid RESTORE STARTED
2637 wait_request_state $fid RESTORE SUCCEED
2638 # check md5sum pgm finished
2639 local there=$(ps -o pid,comm hp $pid >/dev/null)
2640 [[ -z $there ]] || error "Restore initiator does not exit"
2642 local rc=$(wait $pid)
2643 [[ $rc -eq 0 ]] || error "Restore initiator failed with $rc"
2647 run_test 34 "Remove file during restore"
2650 # test needs a running copytool
2655 local f=$DIR/$tdir/$tfile
2656 local f1=$DIR/$tdir/$tfile-1
2658 fid=$(make_large_for_progress $f)
2659 [ $? != 0 ] && skip "not enough free space" && return
2661 local fid1=$(copy_file /etc/passwd $f1)
2662 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2663 wait_request_state $fid ARCHIVE SUCCEED
2666 md5sum $f >/dev/null &
2668 wait_request_state $fid RESTORE STARTED
2670 mv $f1 $f || error "mv $f1 $f failed"
2671 # mv must not block during restore
2672 wait_request_state $fid RESTORE STARTED
2674 wait_request_state $fid RESTORE SUCCEED
2675 # check md5sum pgm finished
2676 local there=$(ps -o pid,comm hp $pid >/dev/null)
2677 [[ -z $there ]] || error "Restore initiator does not exit"
2679 local rc=$(wait $pid)
2680 [[ $rc -eq 0 ]] || error "Restore initiator failed with $rc"
2683 [[ $fid2 == $fid1 ]] || error "Wrong fid after mv $fid2 != $fid1"
2687 run_test 35 "Overwrite file during restore"
2690 # test needs a running copytool
2695 local f=$DIR/$tdir/$tfile
2697 fid=$(make_large_for_progress $f)
2698 [ $? != 0 ] && skip "not enough free space" && return
2700 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2701 wait_request_state $fid ARCHIVE SUCCEED
2704 md5sum $f >/dev/null &
2706 wait_request_state $fid RESTORE STARTED
2709 # rm must not block during restore
2710 wait_request_state $fid RESTORE STARTED
2712 wait_request_state $fid RESTORE SUCCEED
2713 # check md5sum pgm finished
2714 local there=$(ps -o pid,comm hp $pid >/dev/null)
2716 error "Restore initiator does not exit"
2718 local rc=$(wait $pid)
2720 error "Restore initiator failed with $rc"
2724 run_test 36 "Move file during restore"
2727 # LU-5683: check that an archived dirty file can be rearchived.
2729 copytool_setup $SINGLEAGT $MOUNT2
2732 local f=$DIR/$tdir/$tfile
2735 fid=$(make_small $f) || error "cannot create small file"
2737 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2738 wait_request_state $fid ARCHIVE SUCCEED
2739 $LFS hsm_release $f || error "cannot release $f"
2742 dd if=/dev/urandom of=$f bs=1M count=1 || error "cannot dirty file"
2744 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2745 wait_request_state $fid ARCHIVE SUCCEED
2749 run_test 37 "re-archive a dirty file"
2756 for n in $(seq 1 $count); do
2757 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $prefix.$n
2759 echo "$count archive requests submitted"
2763 local stream_count=4
2764 local file_count=100
2766 local f=$DIR/$tdir/$tfile
2771 for i in $(seq 1 $file_count); do
2772 for p in $(seq 1 $stream_count); do
2773 fid=$(copy_file /etc/hosts $f.$p.$i)
2776 # force copytool to use a local/temp archive dir to ensure best
2777 # performance vs remote/NFS mounts used in auto-tests
2778 if do_facet $SINGLEAGT "df --local $HSM_ARCHIVE" >/dev/null 2>&1 ; then
2781 copytool_setup $SINGLEAGT $MOUNT $HSM_ARCHIVE_NUMBER $TMP/$tdir
2783 # to be sure wait_all_done will not be mislead by previous tests
2785 wait_for_grace_delay
2787 # start archive streams in background (archive files in parallel)
2788 for p in $(seq 1 $stream_count); do
2789 multi_archive $f.$p $file_count &
2792 echo -n "Wait for all requests being enqueued..."
2798 run_test 40 "Parallel archive requests"
2801 # test needs a running copytool
2805 local f=$DIR/$tdir/$tfile
2806 local fid=$(copy_file /etc/motd $f 1)
2808 $LFS hsm_archive $f || error "could not archive file"
2809 wait_request_state $fid ARCHIVE SUCCEED
2810 check_hsm_flags $f "0x00000009"
2812 multiop_bg_pause $f O_c || error "multiop failed"
2816 client_up || client_up || true
2818 kill -USR1 $MULTIPID
2819 wait $MULTIPID || error "multiop close failed"
2821 check_hsm_flags $f "0x0000000b"
2825 run_test 52 "Opened for write file on an evicted client should be set dirty"
2828 # test needs a running copytool
2832 local f=$DIR/$tdir/$tfile
2833 local fid=$(copy_file /etc/motd $f 1)
2835 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2836 error "could not archive file"
2837 wait_request_state $fid ARCHIVE SUCCEED
2838 check_hsm_flags $f "0x00000009"
2840 multiop_bg_pause $f o_c || error "multiop failed"
2844 client_up || client_up || true
2846 kill -USR1 $MULTIPID
2847 wait $MULTIPID || error "multiop close failed"
2849 check_hsm_flags $f "0x00000009"
2853 run_test 53 "Opened for read file on an evicted client should not be set dirty"
2856 # test needs a running copytool
2860 local f=$DIR/$tdir/$tfile
2861 local fid=$(make_large_for_progress $f)
2863 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2864 error "could not archive file"
2865 wait_request_state $fid ARCHIVE STARTED
2867 check_hsm_flags $f "0x00000001"
2869 # Avoid coordinator resending this request as soon it has failed.
2874 wait_request_state $fid ARCHIVE FAILED
2876 check_hsm_flags $f "0x00000003"
2881 run_test 54 "Write during an archive cancels it"
2884 # test needs a running copytool
2888 local f=$DIR/$tdir/$tfile
2889 local fid=$(make_large_for_progress $f)
2891 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2892 error "could not archive file"
2893 wait_request_state $fid ARCHIVE STARTED
2895 check_hsm_flags $f "0x00000001"
2897 # Avoid coordinator resending this request as soon it has failed.
2900 $TRUNCATE $f 1024 || error "truncate failed"
2902 wait_request_state $fid ARCHIVE FAILED
2904 check_hsm_flags $f "0x00000003"
2909 run_test 55 "Truncate during an archive cancels it"
2912 # test needs a running copytool
2916 local f=$DIR/$tdir/$tfile
2918 fid=$(make_large_for_progress $f)
2919 [ $? != 0 ] && skip "not enough free space" && return
2921 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2922 error "could not archive file"
2923 wait_request_state $fid ARCHIVE STARTED
2925 check_hsm_flags $f "0x00000001"
2927 # Change metadata and sync to be sure we are not changing only
2932 wait_request_state $fid ARCHIVE SUCCEED
2934 check_hsm_flags $f "0x00000009"
2938 run_test 56 "Setattr during an archive is ok"
2941 # Need one client for I/O, one for request
2942 needclients 2 || return 0
2944 # test needs a running copytool
2948 local f=$DIR/$tdir/test_archive_remote
2949 # Create a file on a remote node
2950 do_node $CLIENT2 "dd if=/dev/urandom of=$f bs=1M "\
2951 "count=2 conv=fsync"
2954 do_node $CLIENT2 "$LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f" ||
2955 error "hsm_archive failed"
2956 local fid=$(path2fid $f)
2957 wait_request_state $fid ARCHIVE SUCCEED
2959 # Release and implicit restore it
2960 do_node $CLIENT2 "$LFS hsm_release $f" ||
2961 error "hsm_release failed"
2962 do_node $CLIENT2 "md5sum $f" ||
2963 error "hsm_restore failed"
2965 wait_request_state $fid RESTORE SUCCEED
2969 run_test 57 "Archive a file with dirty cache on another node"
2971 truncate_released_file() {
2975 local sz=$(stat -c %s $src_file)
2976 local f=$DIR/$tdir/$tfile
2977 local fid=$(copy_file $1 $f)
2981 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2982 error "could not archive file"
2983 wait_request_state $fid ARCHIVE SUCCEED
2985 $LFS hsm_release $f || error "could not release file"
2987 $TRUNCATE $f $trunc_to || error "truncate failed"
2990 local sz1=$(stat -c %s $f)
2991 [[ $sz1 == $trunc_to ]] ||
2992 error "size after trunc: $sz1 expect $trunc_to, original $sz"
2995 check_hsm_flags $f "0x0000000b"
2997 local state=$(get_request_state $fid RESTORE)
2998 [[ "$state" == "SUCCEED" ]] ||
2999 error "truncate $sz does not trig restore, state = $state"
3001 $TRUNCATE $ref $trunc_to
3002 cmp $ref $f || error "file data wrong after truncate"
3008 # test needs a running copytool
3013 local sz=$(stat -c %s /etc/passwd)
3015 echo "truncate up from $sz to $((sz*2))"
3016 truncate_released_file /etc/passwd $((sz*2))
3018 echo "truncate down from $sz to $((sz/2))"
3019 truncate_released_file /etc/passwd $((sz/2))
3021 echo "truncate to 0"
3022 truncate_released_file /etc/passwd 0
3026 run_test 58 "Truncate a released file will trigger restore"
3029 # This test validates the fix for LU-4512. Ensure that the -u
3030 # option changes the progress reporting interval from the
3031 # default (30 seconds) to the user-specified interval.
3033 local progress_timeout=$((interval * 4))
3035 # test needs a new running copytool
3037 HSMTOOL_UPDATE_INTERVAL=$interval copytool_setup
3040 local f=$DIR/$tdir/$tfile
3042 fid=$(make_large_for_progress $f)
3043 [ $? != 0 ] && skip "not enough free space" && return
3046 local mdt=${MDT_PREFIX}${mdtidx}
3047 local mds=mds$((mdtidx + 1))
3049 # Wait for copytool to register
3050 wait_update_facet $mds \
3051 "$LCTL get_param -n ${mdt}.hsm.agents | grep -o ^uuid" \
3052 uuid 100 || error "coyptool failed to register with $mdt"
3054 local start_at=$(date +%s)
3055 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3056 error "could not archive file"
3058 local agent=$(facet_active_host $SINGLEAGT)
3059 local prefix=$TESTLOG_PREFIX
3060 [[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME
3061 local copytool_log=$prefix.copytool_log.$agent.log
3064 wait_update $agent \
3065 "grep -o start.copy $copytool_log" "start copy" 100 ||
3066 error "copytool failed to start"
3068 local cmd="$LCTL get_param -n ${mdt}.hsm.active_requests"
3069 cmd+=" | awk '/'$fid'.*action=ARCHIVE/ {print \\\$12}' | cut -f2 -d="
3075 echo -n "Expecting a progress update within $progress_timeout seconds... "
3077 RESULT=$(do_node $(facet_active_host $mds) "$cmd")
3078 if [ $RESULT -gt 0 ]; then
3079 echo "$RESULT bytes copied in $WAIT seconds."
3081 elif [ $WAIT -ge $progress_timeout ]; then
3082 error "Timed out waiting for progress update!"
3085 WAIT=$((WAIT + sleep))
3089 local finish_at=$(date +%s)
3090 local elapsed=$((finish_at - start_at))
3092 # Ensure that the progress update occurred within the expected window.
3093 if [ $elapsed -lt $interval ]; then
3094 error "Expected progress update after at least $interval seconds"
3100 run_test 60 "Changing progress update interval from default"
3103 # test needs a new running copytool
3105 copytool_monitor_setup
3106 HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
3108 # Just start and stop the copytool to generate events.
3111 # Wait for the copytool to register.
3112 wait_update --verbose $(facet_active_host mds1) \
3113 "$LCTL get_param -n ${MDT_PREFIX}0.hsm.agents | grep -o ^uuid" \
3115 error "copytool failed to register with MDT0000"
3119 local REGISTER_EVENT
3120 local UNREGISTER_EVENT
3121 while read event; do
3122 local parsed=$(parse_json_event "$event")
3123 if [ -z "$parsed" ]; then
3124 error "Copytool sent malformed event: $event"
3128 if [ $event_type == "REGISTER" ]; then
3129 REGISTER_EVENT=$event
3130 elif [ $event_type == "UNREGISTER" ]; then
3131 UNREGISTER_EVENT=$event
3133 done < <(echo $"$(get_copytool_event_log)")
3135 if [ -z "$REGISTER_EVENT" ]; then
3136 error "Copytool failed to send register event to FIFO"
3139 if [ -z "$UNREGISTER_EVENT" ]; then
3140 error "Copytool failed to send unregister event to FIFO"
3143 copytool_monitor_cleanup
3144 echo "Register/Unregister events look OK."
3146 run_test 70 "Copytool logs JSON register/unregister events to FIFO"
3149 # Bump progress interval for livelier events.
3152 # test needs a new running copytool
3154 copytool_monitor_setup
3155 HSMTOOL_UPDATE_INTERVAL=$interval \
3156 HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
3159 local f=$DIR/$tdir/$tfile
3161 fid=$(make_large_for_progress $f)
3162 [ $? != 0 ] && skip "not enough free space" && return
3164 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
3165 error "could not archive file"
3166 wait_request_state $fid ARCHIVE SUCCEED
3168 local expected_fields="event_time data_fid source_fid"
3169 expected_fields+=" total_bytes current_bytes"
3173 while read event; do
3174 # Make sure we're not getting anything from previous events.
3175 for field in $expected_fields; do
3179 local parsed=$(parse_json_event "$event")
3180 if [ -z "$parsed" ]; then
3181 error "Copytool sent malformed event: $event"
3185 if [ $event_type == "ARCHIVE_START" ]; then
3188 elif [ $event_type == "ARCHIVE_FINISH" ]; then
3191 elif [ $event_type != "ARCHIVE_RUNNING" ]; then
3195 # Do some simple checking of the progress update events.
3196 for expected_field in $expected_fields; do
3197 if [ -z ${!expected_field+x} ]; then
3198 error "Missing $expected_field field in event"
3202 if [ $total_bytes -eq 0 ]; then
3203 error "Expected total_bytes to be > 0"
3206 # These should be identical throughout an archive
3208 if [ $source_fid != $data_fid ]; then
3209 error "Expected source_fid to equal data_fid"
3211 done < <(echo $"$(get_copytool_event_log)")
3213 if [ -z "$START_EVENT" ]; then
3214 error "Copytool failed to send archive start event to FIFO"
3217 if [ -z "$FINISH_EVENT" ]; then
3218 error "Copytool failed to send archive finish event to FIFO"
3221 echo "Archive events look OK."
3225 copytool_monitor_cleanup
3227 run_test 71 "Copytool logs JSON archive events to FIFO"
3230 # Bump progress interval for livelier events.
3233 # test needs a new running copytool
3235 copytool_monitor_setup
3236 HSMTOOL_UPDATE_INTERVAL=$interval \
3237 HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
3238 local test_file=$HSMTOOL_MONITOR_DIR/file
3240 local cmd="dd if=/dev/urandom of=$test_file count=16 bs=1000000 "
3242 do_facet $SINGLEAGT "$cmd" ||
3243 error "cannot create $test_file on $SINGLEAGT"
3244 copy2archive $test_file $tdir/$tfile
3247 local f=$DIR/$tdir/$tfile
3248 import_file $tdir/$tfile $f
3249 f=$DIR2/$tdir/$tfile
3250 echo "Verifying released state: "
3251 check_hsm_flags $f "0x0000000d"
3253 local fid=$(path2fid $f)
3255 wait_request_state $fid RESTORE SUCCEED
3257 local expected_fields="event_time data_fid source_fid"
3258 expected_fields+=" total_bytes current_bytes"
3262 while read event; do
3263 # Make sure we're not getting anything from previous events.
3264 for field in $expected_fields; do
3268 local parsed=$(parse_json_event "$event")
3269 if [ -z "$parsed" ]; then
3270 error "Copytool sent malformed event: $event"
3274 if [ $event_type == "RESTORE_START" ]; then
3276 if [ $source_fid != $data_fid ]; then
3277 error "source_fid should == data_fid at start"
3280 elif [ $event_type == "RESTORE_FINISH" ]; then
3282 if [ $source_fid != $data_fid ]; then
3283 error "source_fid should == data_fid at finish"
3286 elif [ $event_type != "RESTORE_RUNNING" ]; then
3290 # Do some simple checking of the progress update events.
3291 for expected_field in $expected_fields; do
3292 if [ -z ${!expected_field+x} ]; then
3293 error "Missing $expected_field field in event"
3297 if [ $total_bytes -eq 0 ]; then
3298 error "Expected total_bytes to be > 0"
3301 # When a restore starts out, the data fid is the same as the
3302 # source fid. After the restore has gotten going, we learn
3303 # the new data fid. Once the restore has finished, the source
3304 # fid is set to the new data fid.
3306 # We test this because some monitoring software may depend on
3307 # this behavior. If it changes, then the consumers of these
3308 # events may need to be modified.
3309 if [ $source_fid == $data_fid ]; then
3310 error "source_fid should != data_fid during restore"
3312 done < <(echo $"$(get_copytool_event_log)")
3314 if [ -z "$START_EVENT" ]; then
3315 error "Copytool failed to send restore start event to FIFO"
3318 if [ -z "$FINISH_EVENT" ]; then
3319 error "Copytool failed to send restore finish event to FIFO"
3322 echo "Restore events look OK."
3326 copytool_monitor_cleanup
3330 run_test 72 "Copytool logs JSON restore events to FIFO"
3333 file_count=51 # Max number of files constrained by LNET message size
3334 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3335 local f=$DIR/$tdir/$tfile
3336 local FILELIST=/tmp/filelist.txt
3340 for i in $(seq 1 $file_count); do
3341 fid=$(copy_file /etc/hosts $f.$i)
3342 echo $f.$i >> $FILELIST
3344 # force copytool to use a local/temp archive dir to ensure best
3345 # performance vs remote/NFS mounts used in auto-tests
3346 if do_facet $SINGLEAGT "df --local $HSM_ARCHIVE" >/dev/null 2>&1 ; then
3349 local dai=$(get_hsm_param default_archive_id)
3350 copytool_setup $SINGLEAGT $MOUNT $dai $TMP/$tdir
3352 # to be sure wait_all_done will not be mislead by previous tests
3354 wait_for_grace_delay
3355 $LFS hsm_archive --filelist $FILELIST ||
3356 error "cannot archive a file list"
3358 $LFS hsm_release --filelist $FILELIST ||
3359 error "cannot release a file list"
3360 $LFS hsm_restore --filelist $FILELIST ||
3361 error "cannot restore a file list"
3365 run_test 90 "Archive/restore a file list"
3367 double_verify_reset_hsm_param() {
3369 echo "Testing $HSM_PARAM.$p"
3370 local val=$(get_hsm_param $p)
3372 local val2=$(($val * 2))
3373 set_hsm_param $p $val2
3374 val=$(get_hsm_param $p)
3375 [[ $val == $val2 ]] ||
3376 error "$HSM_PARAM.$p: $val != $val2 should be (2 * $save)"
3377 echo "Set $p to 0 must failed"
3381 set_hsm_param $p $save
3383 if [[ $rc == 0 ]]; then
3384 error "we must not be able to set $HSM_PARAM.$p to 0"
3389 double_verify_reset_hsm_param loop_period
3390 double_verify_reset_hsm_param grace_delay
3391 double_verify_reset_hsm_param active_request_timeout
3392 double_verify_reset_hsm_param max_requests
3393 double_verify_reset_hsm_param default_archive_id
3395 run_test 100 "Set coordinator /proc tunables"
3402 run_test 102 "Verify coordinator control"
3405 # test needs a running copytool
3412 for i in $(seq 1 20); do
3413 fid=$(copy_file /etc/passwd $DIR/$tdir/$i)
3415 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/*
3419 echo "Current requests"
3420 local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3421 $HSM_PARAM.actions |\
3422 grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
3424 [[ -z "$res" ]] || error "Some request have not been canceled"
3428 run_test 103 "Purge all requests"
3433 # test needs a running copytool
3437 local f=$DIR/$tdir/$tfile
3439 fid=$(make_large_for_progress $f)
3440 [ $? != 0 ] && skip "not enough free space" && return
3442 # if cdt is on, it can serve too quickly the request
3444 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
3445 local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3446 $HSM_PARAM.actions |\
3447 grep $fid | cut -f16 -d=")
3450 [[ "$data1" == "$DATAHEX" ]] ||
3451 error "Data field in records is ($data1) and not ($DATAHEX)"
3455 run_test 104 "Copy tool data field"
3462 for i in $(seq -w 1 10); do
3463 cp /etc/passwd $DIR/$tdir/$i
3464 $LFS hsm_archive $DIR/$tdir/$i
3466 local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3467 $HSM_PARAM.actions |\
3468 grep WAITING | wc -l")
3471 local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3472 $HSM_PARAM.actions |\
3473 grep WAITING | wc -l")
3476 [[ "$reqcnt1" == "$reqcnt2" ]] ||
3477 error "Requests count after shutdown $reqcnt2 != "\
3478 "before shutdown $reqcnt1"
3480 run_test 105 "Restart of coordinator"
3482 get_agent_by_uuid_mdt() {
3485 local mds=mds$(($mdtidx + 1))
3486 do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
3490 check_agent_registered_by_mdt() {
3493 local mds=mds$(($mdtidx + 1))
3494 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
3495 if [[ ! -z "$agent" ]]; then
3496 echo "found agent $agent on $mds"
3498 error "uuid $uuid not found in agent list on $mds"
3502 check_agent_unregistered_by_mdt() {
3505 local mds=mds$(($mdtidx + 1))
3506 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
3507 if [[ -z "$agent" ]]; then
3508 echo "uuid not found in agent list on $mds"
3510 error "uuid found in agent list on $mds: $agent"
3514 check_agent_registered() {
3517 for mdsno in $(seq 1 $MDSCOUNT); do
3518 check_agent_registered_by_mdt $uuid $((mdsno - 1))
3522 check_agent_unregistered() {
3525 for mdsno in $(seq 1 $MDSCOUNT); do
3526 check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
3531 local agent=${1:-$(facet_active_host $SINGLEAGT)}
3533 # Lustre mount-point is mandatory and last parameter on
3534 # copytool cmd-line.
3535 local mntpnt=$(do_rpc_nodes $agent pgrep -fl $HSMTOOL_BASE |
3536 grep -v pgrep | awk '{print $NF}')
3537 [ -n "$mntpnt" ] || error "Found no Agent or with no mount-point "\
3539 do_rpc_nodes $agent get_client_uuid $mntpnt | cut -d' ' -f2
3543 # test needs a running copytool
3546 local uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
3548 check_agent_registered $uuid
3550 search_copytools || error "No copytool found"
3553 check_agent_unregistered $uuid
3556 uuid=$(get_agent_uuid $(facet_active_host $SINGLEAGT))
3557 check_agent_registered $uuid
3561 run_test 106 "Copytool register/unregister"
3564 # test needs a running copytool
3566 # create and archive file
3568 local f1=$DIR/$tdir/$tfile
3569 local fid=$(copy_file /etc/passwd $f1)
3570 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
3571 wait_request_state $fid ARCHIVE SUCCEED
3572 # shutdown and restart MDS
3574 # check the copytool still gets messages from MDT
3575 local f2=$DIR/$tdir/2
3576 local fid=$(copy_file /etc/passwd $f2)
3577 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
3578 # main check of this sanity: this request MUST succeed
3579 wait_request_state $fid ARCHIVE SUCCEED
3582 run_test 107 "Copytool re-register after MDS restart"
3584 policy_set_and_test()
3588 do_facet $SINGLEMDS $LCTL set_param "$HSM_PARAM.policy=\\\"$change\\\""
3589 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3590 [[ "$policy" == "$target" ]] ||
3591 error "Wrong policy after '$change': '$policy' != '$target'"
3595 # to force default policy setting if error
3596 CDT_POLICY_HAD_CHANGED=true
3598 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3599 local default="NonBlockingRestore [NoRetryAction]"
3600 [[ "$policy" == "$default" ]] ||
3601 error "default policy has changed,"\
3602 " '$policy' != '$default' update the test"
3603 policy_set_and_test "+NBR" "[NonBlockingRestore] [NoRetryAction]"
3604 policy_set_and_test "+NRA" "[NonBlockingRestore] [NoRetryAction]"
3605 policy_set_and_test "-NBR" "NonBlockingRestore [NoRetryAction]"
3606 policy_set_and_test "-NRA" "NonBlockingRestore NoRetryAction"
3607 policy_set_and_test "NRA NBR" "[NonBlockingRestore] [NoRetryAction]"
3608 # useless bacause we know but safer for futur changes to use real value
3609 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3610 echo "Next set_param must failed"
3611 policy_set_and_test "wrong" "$policy"
3614 echo "Back to default policy"
3615 cdt_set_sanity_policy
3617 run_test 109 "Policy display/change"
3620 # test needs a running copytool
3625 copy2archive /etc/passwd $tdir/$tfile
3627 local f=$DIR/$tdir/$tfile
3628 import_file $tdir/$tfile $f
3629 local fid=$(path2fid $f)
3631 cdt_set_non_blocking_restore
3636 wait_request_state $fid RESTORE SUCCEED
3637 cdt_clear_non_blocking_restore
3641 error "md5sum returns $st != 1, "\
3642 "should also perror ENODATA (No data available)"
3646 run_test 110a "Non blocking restore policy (import case)"
3649 # test needs a running copytool
3653 local f=$DIR/$tdir/$tfile
3654 local fid=$(copy_file /etc/passwd $f)
3655 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3656 wait_request_state $fid ARCHIVE SUCCEED
3659 cdt_set_non_blocking_restore
3664 wait_request_state $fid RESTORE SUCCEED
3665 cdt_clear_non_blocking_restore
3669 error "md5sum returns $st != 1, "\
3670 "should also perror ENODATA (No data available)"
3674 run_test 110b "Non blocking restore policy (release case)"
3677 # test needs a running copytool
3681 copy2archive /etc/passwd $tdir/$tfile
3683 local f=$DIR/$tdir/$tfile
3685 import_file $tdir/$tfile $f
3686 local fid=$(path2fid $f)
3690 copytool_remove_backend $fid
3693 wait_request_state $fid RESTORE FAILED
3700 [[ $st == 0 ]] || error "Restore does not failed"
3704 run_test 111a "No retry policy (import case), restore will error"\
3705 " (No such file or directory)"
3708 # test needs a running copytool
3712 local f=$DIR/$tdir/$tfile
3713 local fid=$(copy_file /etc/passwd $f)
3715 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3716 wait_request_state $fid ARCHIVE SUCCEED
3719 copytool_remove_backend $fid
3722 wait_request_state $fid RESTORE FAILED
3729 [[ $st == 0 ]] || error "Restore does not failed"
3733 run_test 111b "No retry policy (release case), restore will error"\
3734 " (No such file or directory)"
3737 # test needs a running copytool
3741 local f=$DIR/$tdir/$tfile
3742 local fid=$(copy_file /etc/passwd $f)
3744 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3745 local l=$($LFS hsm_action $f)
3747 local res=$(echo $l | cut -f 2- -d" " | grep ARCHIVE)
3751 wait_request_state $fid ARCHIVE SUCCEED
3754 [[ ! -z "$res" ]] || error "action is $l which is not an ARCHIVE"
3758 run_test 112 "State of recorded request"
3761 # test needs a running copytool
3765 local f=$DIR/$tdir/$tfile
3767 fid=$(make_large_for_cancel $f)
3768 [ $? != 0 ] && skip "not enough free space" && return
3770 # test with cdt on is made in test_221
3772 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3775 wait_request_state $fid ARCHIVE CANCELED
3776 wait_request_state $fid CANCEL SUCCEED
3780 run_test 200 "Register/Cancel archive"
3783 # test needs a running copytool
3787 local f=$DIR/$tdir/$tfile
3788 make_archive $tdir/$tfile
3789 import_file $tdir/$tfile $f
3790 local fid=$(path2fid $f)
3792 # test with cdt on is made in test_222
3797 wait_request_state $fid RESTORE CANCELED
3798 wait_request_state $fid CANCEL SUCCEED
3802 run_test 201 "Register/Cancel restore"
3805 # test needs a running copytool
3809 local f=$DIR/$tdir/$tfile
3811 fid=$(make_large_for_progress $f)
3812 [ $? != 0 ] && skip "not enough free space" && return
3814 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3815 wait_request_state $fid ARCHIVE SUCCEED
3821 wait_request_state $fid REMOVE CANCELED
3825 run_test 202 "Register/Cancel remove"
3828 # test needs a running copytool
3833 local f=$DIR/$tdir/$tfile
3834 local fid=$(copy_file /etc/passwd $f)
3838 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3839 wait_request_state $fid ARCHIVE SUCCEED
3841 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3845 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3849 run_test 220 "Changelog for archive"
3852 # test needs a running copytool
3857 local f=$DIR/$tdir/$tfile
3859 fid=$(make_large_for_cancel $f)
3860 [ $? != 0 ] && skip "not enough free space" && return
3864 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3865 wait_request_state $fid ARCHIVE STARTED
3867 wait_request_state $fid ARCHIVE CANCELED
3868 wait_request_state $fid CANCEL SUCCEED
3870 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3873 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3877 run_test 221 "Changelog for archive canceled"
3880 # test needs a running copytool
3884 copy2archive /etc/passwd $tdir/$tfile
3886 local f=$DIR/$tdir/$tfile
3887 import_file $tdir/$tfile $f
3888 local fid=$(path2fid $f)
3893 wait_request_state $fid RESTORE SUCCEED
3895 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3898 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3902 run_test 222a "Changelog for explicit restore"
3905 # test needs a running copytool
3909 local f=$DIR/$tdir/$tfile
3910 local fid=$(copy_file /etc/passwd $f)
3913 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3914 wait_request_state $fid ARCHIVE SUCCEED
3919 wait_request_state $fid RESTORE SUCCEED
3921 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3924 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3928 run_test 222b "Changelog for implicit restore"
3931 # test needs a running copytool
3936 local f=$DIR/$tdir/$tfile
3937 make_archive $tdir/$tfile
3941 import_file $tdir/$tfile $f
3942 local fid=$(path2fid $f)
3945 wait_request_state $fid RESTORE STARTED
3947 wait_request_state $fid RESTORE CANCELED
3948 wait_request_state $fid CANCEL SUCCEED
3950 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3953 [[ $flags == $target ]] ||
3954 error "Changelog flag is $flags not $target"
3958 run_test 223a "Changelog for restore canceled (import case)"
3961 # test needs a running copytool
3966 local f=$DIR/$tdir/$tfile
3968 fid=$(make_large_for_progress $f)
3969 [ $? != 0 ] && skip "not enough free space" && return
3972 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3973 wait_request_state $fid ARCHIVE SUCCEED
3976 wait_request_state $fid RESTORE STARTED
3978 wait_request_state $fid RESTORE CANCELED
3979 wait_request_state $fid CANCEL SUCCEED
3981 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3984 [[ $flags == $target ]] ||
3985 error "Changelog flag is $flags not $target"
3989 run_test 223b "Changelog for restore canceled (release case)"
3992 # test needs a running copytool
3997 local f=$DIR/$tdir/$tfile
3998 local fid=$(copy_file /etc/passwd $f)
4001 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4002 wait_request_state $fid ARCHIVE SUCCEED
4005 wait_request_state $fid REMOVE SUCCEED
4007 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -n 1)
4010 [[ $flags == $target ]] ||
4011 error "Changelog flag is $flags not $target"
4015 run_test 224 "Changelog for remove"
4018 # test needs a running copytool
4021 # test is not usable because remove request is too fast
4022 # so it is always finished before cancel can be done ...
4023 echo "Test disabled"
4028 local f=$DIR/$tdir/$tfile
4030 fid=$(make_large_for_progress $f)
4031 [ $? != 0 ] && skip "not enough free space" && return
4034 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4035 wait_request_state $fid ARCHIVE SUCCEED
4037 # if cdt is on, it can serve too quickly the request
4042 wait_request_state $fid REMOVE CANCELED
4043 wait_request_state $fid CANCEL SUCCEED
4045 flags=$(changelog_get_flags ${MDT[0]} RENME $fid2)
4046 local flags=$($LFS changelog ${MDT[0]} | grep HSM | grep $fid |
4047 tail -n 1 | awk '{print $5}')
4050 [[ $flags == $target ]] ||
4051 error "Changelog flag is $flags not $target"
4055 run_test 225 "Changelog for remove canceled"
4058 # test needs a running copytool
4063 local f1=$DIR/$tdir/$tfile-1
4064 local f2=$DIR/$tdir/$tfile-2
4065 local f3=$DIR/$tdir/$tfile-3
4066 local fid1=$(copy_file /etc/passwd $f1)
4067 local fid2=$(copy_file /etc/passwd $f2)
4068 copy_file /etc/passwd $f3
4071 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
4072 wait_request_state $fid1 ARCHIVE SUCCEED
4074 $LFS hsm_archive $f2
4075 wait_request_state $fid2 ARCHIVE SUCCEED
4077 rm $f1 || error "rm $f1 failed"
4079 local flags=$(changelog_get_flags ${MDT[0]} UNLNK $fid1)
4082 [[ $flags == $target ]] ||
4083 error "Changelog flag is $flags not $target"
4085 mv $f3 $f2 || error "mv $f3 $f2 failed"
4087 flags=$(changelog_get_flags ${MDT[0]} RENME $fid2)
4090 [[ $flags == $target ]] ||
4091 error "Changelog flag is $flags not $target"
4095 run_test 226 "changelog for last rm/mv with exiting archive"
4097 check_flags_changes() {
4105 $LFS hsm_set --$hsm_flag $f ||
4106 error "Cannot set $hsm_flag on $f"
4107 local flags=($(changelog_get_flags ${MDT[0]} HSM $fid))
4108 local seen=${#flags[*]}
4110 [[ $seen == $cnt ]] ||
4111 error "set $hsm_flag: Changelog events $seen != $cnt"
4112 [[ ${flags[$((cnt - 1))]} == $target ]] ||
4113 error "set $hsm_flag: Changelog flags are "\
4114 "${flags[$((cnt - 1))]} not $target"
4116 $LFS hsm_clear --$hsm_flag $f ||
4117 error "Cannot clear $hsm_flag on $f"
4118 flags=($(changelog_get_flags ${MDT[0]} HSM $fid))
4121 [[ $cnt == $seen ]] ||
4122 error "clear $hsm_flag: Changelog events $seen != $cnt"
4124 [[ ${flags[$((cnt - 1))]} == $target ]] ||
4125 error "clear $hsm_flag: Changelog flag is "\
4126 "${flags[$((cnt - 1))]} not $target"
4130 # test needs a running copytool
4137 for i in norelease noarchive exists archived
4139 local f=$DIR/$tdir/$tfile-$i
4140 local fid=$(copy_file /etc/passwd $f)
4141 check_flags_changes $f $fid $i 0 1
4144 f=$DIR/$tdir/$tfile---lost
4145 fid=$(copy_file /etc/passwd $f)
4146 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4147 wait_request_state $fid ARCHIVE SUCCEED
4148 check_flags_changes $f $fid lost 3 1
4152 run_test 227 "changelog when explicit setting of HSM flags"
4155 # test needs a running copytool
4158 local fid=$(make_small_sync $DIR/$tfile)
4159 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
4160 wait_request_state $fid ARCHIVE SUCCEED
4162 $LFS hsm_release $DIR/$tfile
4163 check_hsm_flags $DIR/$tfile "0x0000000d"
4165 filefrag $DIR/$tfile | grep " 1 extent found" ||
4166 error "filefrag on released file must return only one extent"
4168 # only newer versions of cp detect sparse files by stat/FIEMAP
4170 cp --sparse=auto $DIR/$tfile $DIR/$tfile.2 ||
4171 error "copying $DIR/$tfile"
4172 cmp $DIR/$tfile $DIR/$tfile.2 || error "comparing copied $DIR/$tfile"
4174 $LFS hsm_release $DIR/$tfile
4175 check_hsm_flags $DIR/$tfile "0x0000000d"
4177 mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
4179 tar cf - --sparse $DIR/$tfile | tar xvf - -C $DIR/$tdir ||
4181 cmp $DIR/$tfile $DIR/$tdir/$DIR/$tfile ||
4182 error "comparing untarred $DIR/$tfile"
4184 rm -f $DIR/$tfile $DIR/$tfile.2 ||
4185 error "rm $DIR/$tfile or $DIR/$tfile.2 failed"
4188 run_test 228 "On released file, return extend to FIEMAP. For [cp,tar] --sparse"
4191 # test needs a running copytool
4195 local maxrequest=$(get_hsm_param max_requests)
4196 local rqcnt=$(($maxrequest * 3))
4200 for i in $(seq -w 1 $rqcnt); do
4202 dd if=/dev/urandom of=$DIR/$tdir/$i bs=1M count=10 conv=fsync
4204 # we do it in 2 steps, so all requests arrive at the same time
4205 for i in $(seq -w 1 $rqcnt); do
4206 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/$i
4211 while [[ $cnt != 0 || $wt != 0 ]]; do
4213 cnt=$(do_facet $SINGLEMDS "$LCTL get_param -n\
4214 $HSM_PARAM.actions |\
4215 grep STARTED | grep -v CANCEL | wc -l")
4216 [[ $cnt -le $maxrequest ]] ||
4217 error "$cnt > $maxrequest too many started requests"
4218 wt=$(do_facet $SINGLEMDS "$LCTL get_param\
4219 $HSM_PARAM.actions |\
4220 grep WAITING | wc -l")
4221 echo "max=$maxrequest started=$cnt waiting=$wt"
4226 run_test 250 "Coordinator max request"
4229 # test needs a running copytool
4233 local f=$DIR/$tdir/$tfile
4235 fid=$(make_large_for_cancel $f)
4236 [ $? != 0 ] && skip "not enough free space" && return
4239 # to have a short test
4240 local old_to=$(get_hsm_param active_request_timeout)
4241 set_hsm_param active_request_timeout 4
4242 # to be sure the cdt will wake up frequently so
4243 # it will be able to cancel the "old" request
4244 local old_loop=$(get_hsm_param loop_period)
4245 set_hsm_param loop_period 2
4248 # clear locks to avoid extra delay caused by flush/cancel
4249 # and thus prevent early copytool death to timeout.
4250 cancel_lru_locks osc
4252 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
4253 wait_request_state $fid ARCHIVE STARTED
4255 wait_request_state $fid ARCHIVE CANCELED
4257 set_hsm_param active_request_timeout $old_to
4258 set_hsm_param loop_period $old_loop
4262 run_test 251 "Coordinator request timeout"
4265 # the only way to test ondisk conf is to restart MDS ...
4266 echo "Stop coordinator and remove coordinator state at mount"
4269 # clean on disk conf set by default
4270 cdt_clear_mount_state
4271 cdt_check_state stopped
4273 # check cdt still off after umount/remount
4275 cdt_check_state stopped
4277 echo "Set coordinator start at mount, and start coordinator"
4278 cdt_set_mount_state enabled
4281 cdt_check_state enabled
4283 # check cdt still on after umount/remount
4285 cdt_check_state enabled
4287 # we are back to original state (cdt started at mount)
4289 run_test 300 "On disk coordinator state kept between MDT umount/mount"
4292 local ai=$(get_hsm_param default_archive_id)
4293 local new=$((ai + 1))
4295 set_hsm_param default_archive_id $new -P
4297 local res=$(get_hsm_param default_archive_id)
4300 set_hsm_param default_archive_id "" "-P -d"
4302 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
4304 run_test 301 "HSM tunnable are persistent"
4307 local ai=$(get_hsm_param default_archive_id)
4308 local new=$((ai + 1))
4313 set_hsm_param default_archive_id $new -P
4316 for mdtno in $(seq 1 $MDSCOUNT); do
4321 cdt_check_state enabled
4323 local res=$(get_hsm_param default_archive_id)
4326 set_hsm_param default_archive_id "" "-P -d"
4328 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
4330 run_test 302 "HSM tunnable are persistent when CDT is off"
4333 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4339 local dir_mdt0=$DIR/$tdir/mdt0
4340 local dir_mdt1=$DIR/$tdir/mdt1
4342 # create 1 dir per MDT
4343 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4344 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
4346 # create 1 file in each MDT
4347 local fid1=$(make_small $dir_mdt0/$tfile)
4348 local fid2=$(make_small $dir_mdt1/$tfile)
4350 # check that hsm request on mdt0 is sent to the right MDS
4351 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
4352 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4353 echo "archive successful on mdt0"
4355 # check that hsm request on mdt1 is sent to the right MDS
4356 $LFS hsm_archive $dir_mdt1/$tfile || error "lfs hsm_archive"
4357 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
4358 echo "archive successful on mdt1"
4361 # clean test files and directories
4362 rm -rf $dir_mdt0 $dir_mdt1
4364 run_test 400 "Single request is sent to the right MDT"
4367 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4373 local dir_mdt0=$DIR/$tdir/mdt0
4374 local dir_mdt1=$DIR/$tdir/mdt1
4376 # create 1 dir per MDT
4377 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4378 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
4380 # create 1 file in each MDT
4381 local fid1=$(make_small $dir_mdt0/$tfile)
4382 local fid2=$(make_small $dir_mdt1/$tfile)
4384 # check that compound requests are shunt to the rights MDTs
4385 $LFS hsm_archive $dir_mdt0/$tfile $dir_mdt1/$tfile ||
4386 error "lfs hsm_archive"
4387 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4388 echo "archive successful on mdt0"
4389 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
4390 echo "archive successful on mdt1"
4393 # clean test files and directories
4394 rm -rf $dir_mdt0 $dir_mdt1
4396 run_test 401 "Compound requests split and sent to their respective MDTs"
4398 mdc_change_state() # facet, MDT_pattern, activate|deactivate
4403 local node=$(facet_active_host $facet)
4405 for mdc in $(do_facet $facet "$LCTL dl | grep -E ${pattern}-mdc" |
4406 awk '{print $4}'); do
4407 echo "$3 $mdc on $node"
4408 do_facet $facet "$LCTL --device $mdc $state" || return 1
4413 # make sure there is no running copytool
4416 # deactivate all mdc on agent1
4417 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "deactivate"
4419 HSMTOOL_NOERROR=true copytool_setup $SINGLEAGT
4421 check_agent_unregistered "uuid" # match any agent
4423 # no expected running copytool
4424 search_copytools $agent && error "Copytool start should have failed"
4427 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "activate"
4429 run_test 402 "Copytool start fails if all MDTs are inactive"
4432 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4434 # make sure there is no running copytool
4437 local agent=$(facet_active_host $SINGLEAGT)
4439 # deactivate all mdc for MDT0001
4440 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
4443 local uuid=$(get_agent_uuid $agent)
4444 # check the agent is registered on MDT0000, and not on MDT0001
4445 check_agent_registered_by_mdt $uuid 0
4446 check_agent_unregistered_by_mdt $uuid 1
4448 # check running copytool process
4449 search_copytools $agent || error "No running copytools on $agent"
4451 # reactivate all mdc for MDT0001
4452 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
4454 # make sure the copytool is now registered to all MDTs
4455 check_agent_registered $uuid
4459 run_test 403 "Copytool starts with inactive MDT and register on reconnect"
4462 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4466 # create files on both MDT0000 and MDT0001
4469 local dir_mdt0=$DIR/$tdir/mdt0
4470 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4472 # create 1 file on mdt0
4473 local fid1=$(make_small $dir_mdt0/$tfile)
4475 # deactivate all mdc for MDT0001
4476 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
4478 # send an HSM request for files in MDT0000
4479 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
4481 # check for completion of files in MDT0000
4482 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4483 echo "archive successful on mdt0"
4485 # reactivate all mdc for MDT0001
4486 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
4489 # clean test files and directories
4492 run_test 404 "Inactive MDT does not block requests for active MDTs"
4495 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4501 local striped_dir=$DIR/$tdir/striped_dir
4503 # create striped dir on all of MDTs
4504 $LFS mkdir -i 0 -c $MDSCOUNT $striped_dir || error "lfs mkdir"
4506 local fid1=$(make_small_sync $striped_dir/${tfile}_0)
4507 local fid2=$(make_small_sync $striped_dir/${tfile}_1)
4508 local fid3=$(make_small_sync $striped_dir/${tfile}_2)
4509 local fid4=$(make_small_sync $striped_dir/${tfile}_3)
4511 local idx1=$($LFS getstripe -M $striped_dir/${tfile}_0)
4512 local idx2=$($LFS getstripe -M $striped_dir/${tfile}_1)
4513 local idx3=$($LFS getstripe -M $striped_dir/${tfile}_2)
4514 local idx4=$($LFS getstripe -M $striped_dir/${tfile}_3)
4516 # check that compound requests are shunt to the rights MDTs
4517 $LFS hsm_archive $striped_dir/${tfile}_0 $striped_dir/${tfile}_1 \
4518 $striped_dir/${tfile}_2 $striped_dir/${tfile}_3 ||
4519 error "lfs hsm_archive"
4521 wait_request_state $fid1 ARCHIVE SUCCEED $idx1 &&
4522 echo "archive successful on $fid1"
4523 wait_request_state $fid2 ARCHIVE SUCCEED $idx2 &&
4524 echo "archive successful on $fid2"
4525 wait_request_state $fid3 ARCHIVE SUCCEED $idx3 &&
4526 echo "archive successful on $fid3"
4527 wait_request_state $fid4 ARCHIVE SUCCEED $idx4 &&
4528 echo "archive successful on $fid4"
4530 $LFS hsm_release $striped_dir/${tfile}_0 || error "lfs hsm_release 1"
4531 $LFS hsm_release $striped_dir/${tfile}_1 || error "lfs hsm_release 2"
4532 $LFS hsm_release $striped_dir/${tfile}_2 || error "lfs hsm_release 3"
4533 $LFS hsm_release $striped_dir/${tfile}_3 || error "lfs hsm_release 4"
4535 cat $striped_dir/${tfile}_0 > /dev/null || error "cat ${tfile}_0 failed"
4536 cat $striped_dir/${tfile}_1 > /dev/null || error "cat ${tfile}_1 failed"
4537 cat $striped_dir/${tfile}_2 > /dev/null || error "cat ${tfile}_2 failed"
4538 cat $striped_dir/${tfile}_3 > /dev/null || error "cat ${tfile}_3 failed"
4542 run_test 405 "archive and release under striped directory"
4546 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.92) ] &&
4547 skip "HSM migrate is not supported" && return
4549 # Stop the existing copytool
4552 test_mkdir -p $DIR/$tdir
4553 llapi_hsm_test -d $DIR/$tdir || error "One llapi HSM test failed"
4555 run_test 500 "various LLAPI HSM tests"
4560 check_and_cleanup_lustre