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 $RUNAS
49 # In order to test multiple remote HSM agents, a new facet type named "AGT" and
50 # the following associated variables are added:
52 # AGTCOUNT: number of agents
53 # AGTDEV{N}: target HSM mount point (root path of the backend)
54 # agt{N}_HOST: hostname of the agent agt{N}
55 # SINGLEAGT: facet of the single agent
57 # The number of agents is initialized as the number of remote client nodes.
58 # By default, only single copytool is started on a remote client/agent. If there
59 # was no remote client, then the copytool will be started on the local client.
65 export AGTCOUNT=${AGTCOUNT:-$((CLIENTCOUNT - 1))}
66 [[ $AGTCOUNT -gt 0 ]] || AGTCOUNT=1
68 export SHARED_DIRECTORY=${SHARED_DIRECTORY:-$TMP}
69 if [[ $CLIENTCOUNT -gt 1 ]] &&
70 ! check_shared_dir $SHARED_DIRECTORY $CLIENTS; then
71 skip_env "SHARED_DIRECTORY should be accessible"\
76 for n in $(seq $AGTCOUNT); do
77 eval export AGTDEV$n=\$\{AGTDEV$n:-"$SHARED_DIRECTORY/arc$n"\}
78 agent=CLIENT$((n + 1))
79 if [[ -z "${!agent}" ]]; then
80 [[ $CLIENTCOUNT -eq 1 ]] && agent=CLIENT1 ||
83 eval export agt${n}_HOST=\$\{agt${n}_HOST:-${!agent}\}
86 export SINGLEAGT=${SINGLEAGT:-agt1}
88 export HSMTOOL=${HSMTOOL:-"lhsmtool_posix"}
89 export HSMTOOL_VERBOSE=${HSMTOOL_VERBOSE:-""}
90 export HSMTOOL_UPDATE_INTERVAL=${HSMTOOL_UPDATE_INTERVAL:=""}
91 export HSMTOOL_EVENT_FIFO=${HSMTOOL_EVENT_FIFO:=""}
92 export HSMTOOL_TESTDIR
93 export HSMTOOL_BASE=$(basename "$HSMTOOL" | cut -f1 -d" ")
94 HSM_ARCHIVE=$(copytool_device $SINGLEAGT)
97 # The test only support up to 10 MDTs
98 MDT_PREFIX="mdt.$FSNAME-MDT000"
99 HSM_PARAM="${MDT_PREFIX}0.hsm"
101 # archive is purged at copytool setup
102 HSM_ARCHIVE_PURGE=true
104 # Don't allow copytool error upon start/setup
105 HSMTOOL_NOERROR=false
108 # Get the backend root path for the given agent facet.
111 local dev=AGTDEV$(facet_number $facet)
116 # Stop copytool and unregister an existing changelog user.
118 copytool_monitor_cleanup
121 cdt_set_sanity_policy
126 # get MDT device for each mdc
127 for mdtno in $(seq 1 $MDSCOUNT); do
128 local idx=$(($mdtno - 1))
129 MDT[$idx]=$($LCTL get_param -n \
130 mdc.$FSNAME-MDT000${idx}-mdc-*.mds_server_uuid |
131 awk '{gsub(/_UUID/,""); print $1}' | head -n1)
136 local agents=${1:-$(facet_active_host $SINGLEAGT)}
137 do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
140 search_and_kill_copytool() {
141 local agents=${1:-$(facet_active_host $SINGLEAGT)}
143 echo "Killing existing copytools on $agents"
144 do_nodesv $agents "killall -q $HSMTOOL_BASE" || true
147 copytool_monitor_setup() {
148 local facet=${1:-$SINGLEAGT}
149 local agent=$(facet_active_host $facet)
151 local cmd="mktemp --tmpdir=/tmp -d ${TESTSUITE}.${TESTNAME}.XXXX"
152 local test_dir=$(do_node $agent "$cmd") ||
153 error "Failed to create tempdir on $agent"
154 export HSMTOOL_MONITOR_DIR=$test_dir
156 # Create the fifo and a monitor (cat dies when copytool dies)
157 do_node $agent "mkfifo -m 0644 $test_dir/fifo" ||
158 error "failed to create copytool fifo on $agent"
159 cmd="cat $test_dir/fifo > $test_dir/events &"
160 cmd+=" echo \\\$! > $test_dir/monitor_pid"
162 if [[ $PDSH == *Rmrsh* ]]; then
163 # This is required for pdsh -Rmrsh and its handling of remote
165 # Regular ssh and pdsh -Rssh work fine without this
166 # backgrounded subshell nonsense.
167 (do_node $agent "$cmd") &
168 export HSMTOOL_MONITOR_PDSH=$!
170 # Slightly racy, but just making a best-effort to catch obvious
173 ps -p $HSMTOOL_MONITOR_PDSH >&- ||
174 error "Failed to start copytool monitor on $agent"
176 do_node $agent "$cmd"
178 error "Failed to start copytool monitor on $agent"
183 copytool_monitor_cleanup() {
184 local facet=${1:-$SINGLEAGT}
185 local agent=$(facet_active_host $facet)
187 if [ -n "$HSMTOOL_MONITOR_DIR" ]; then
188 # Should die when the copytool dies, but just in case.
189 local cmd="kill \\\$(cat $HSMTOOL_MONITOR_DIR/monitor_pid)"
190 cmd+=" 2>/dev/null || true"
191 do_node $agent "$cmd"
192 do_node $agent "rm -fr $HSMTOOL_MONITOR_DIR"
193 export HSMTOOL_MONITOR_DIR=
196 # The pdsh should die on its own when the monitor dies. Just
197 # in case, though, try to clean up to avoid any cruft.
198 if [ -n "$HSMTOOL_MONITOR_PDSH" ]; then
199 kill $HSMTOOL_MONITOR_PDSH 2>/dev/null
200 export HSMTOOL_MONITOR_PDSH=
205 local facet=${1:-$SINGLEAGT}
206 local lustre_mntpnt=${2:-$MOUNT}
208 local hsm_root=${4:-$(copytool_device $facet)}
209 local agent=$(facet_active_host $facet)
211 if [[ -z "$arc_id" ]] &&
212 do_facet $facet "pkill -CONT -x $HSMTOOL_BASE"; then
213 echo "Only wakeup running copytool $facet on $agent"
217 if $HSM_ARCHIVE_PURGE; then
218 echo "Purging archive on $agent"
219 do_facet $facet "rm -rf $hsm_root/*"
222 echo "Starting copytool $facet on $agent"
223 do_facet $facet "mkdir -p $hsm_root" || error "mkdir '$hsm_root' failed"
224 # bandwidth is limited to 1MB/s so the copy time is known and
225 # independent of hardware
226 local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon --hsm-root $hsm_root"
227 [[ -z "$arc_id" ]] || cmd+=" --archive $arc_id"
228 [[ -z "$HSMTOOL_UPDATE_INTERVAL" ]] ||
229 cmd+=" --update-interval $HSMTOOL_UPDATE_INTERVAL"
230 [[ -z "$HSMTOOL_EVENT_FIFO" ]] ||
231 cmd+=" --event-fifo $HSMTOOL_EVENT_FIFO"
232 cmd+=" --bandwidth 1 $lustre_mntpnt"
234 # Redirect the standard output and error to a log file which
235 # can be uploaded to Maloo.
236 local prefix=$TESTLOG_PREFIX
237 [[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME
238 local copytool_log=$prefix.copytool${arc_id}_log.$agent.log
240 do_facet $facet "$cmd < /dev/null > $copytool_log 2>&1"
241 if [[ $? != 0 ]]; then
242 [[ $HSMTOOL_NOERROR == true ]] ||
243 error "start copytool $facet on $agent failed"
244 echo "start copytool $facet on $agent failed"
250 get_copytool_event_log() {
251 local facet=${1:-$SINGLEAGT}
252 local agent=$(facet_active_host $facet)
254 [ -z "$HSMTOOL_MONITOR_DIR" ] &&
255 error "Can't get event log: No monitor directory!"
257 do_node $agent "cat $HSMTOOL_MONITOR_DIR/events" ||
258 error "Could not collect event log from $agent"
263 local facet=$SINGLEAGT
264 local agents=${1:-$(facet_active_host $facet)}
269 local hsm_root=$(copytool_device $facet)
270 local end_wait=$(( SECONDS + TIMEOUT ))
272 do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
274 while (( SECONDS < end_wait )); do
276 do_nodesv $agents "pgrep -x $HSMTOOL_BASE"
277 if [ $? -ne 0 ]; then
278 echo "Copytool is stopped on $agents"
281 echo "Copytool still running on $agents"
283 if do_nodesv $agents "pgrep -x $HSMTOOL_BASE"; then
284 error "Copytool failed to stop in ${TIMEOUT}s ..."
286 echo "Copytool has stopped in " \
287 "$((TIMEOUT - (end_wait - SECONDS)))s."
290 # clean all CDTs orphans requests from previous tests
291 # that would otherwise need to timeout to clear.
292 for mdtno in $(seq 1 $MDSCOUNT); do
294 mdt_hsmctrl="mdt.$FSNAME-MDT000${idx}.hsm_control"
295 oldstate=$(do_facet mds${mdtno} "$LCTL get_param -n " \
297 # skip already stop[ed,ing] CDTs
298 echo $oldstate | grep stop && continue
300 do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=shutdown"
301 wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
303 error "mds${mdtno} cdt state is not stopped"
304 do_facet mds${mdtno} "$LCTL set_param $mdt_hsmctrl=$oldstate"
305 wait_result mds${mdtno} "$LCTL get_param -n $mdt_hsmctrl" \
307 error "mds${mdtno} cdt state is not $oldstate"
309 if do_facet $facet "df $hsm_root" >/dev/null 2>&1 ; then
310 do_facet $facet "rm -rf $hsm_root/*"
315 local agents=${1:-$(facet_active_host $SINGLEAGT)}
317 do_nodesv $agents "pkill -STOP -x $HSMTOOL_BASE" || return 0
318 echo "Copytool is suspended on $agents"
321 copytool_remove_backend() {
323 local be=$(do_facet $SINGLEAGT find $HSM_ARCHIVE -name $fid)
324 echo "Remove from backend: $fid = $be"
325 do_facet $SINGLEAGT rm -f $be
329 do_facet $SINGLEAGT \
330 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
331 --import $1 $2 $MOUNT" ||
332 error "import of $1 to $2 failed"
336 local file=$HSM_ARCHIVE/$1
337 do_facet $SINGLEAGT mkdir -p $(dirname $file)
338 do_facet $SINGLEAGT dd if=/dev/urandom of=$file count=32 bs=1000000 ||
339 file_creation_failure dd $file $?
343 local file=$HSM_ARCHIVE/$2
344 do_facet $SINGLEAGT mkdir -p $(dirname $file)
345 do_facet $SINGLEAGT cp -p $1 $file || error "cannot copy $1 to $file"
354 if [[ "$value" != "" ]]; then
357 for mdtno in $(seq 1 $MDSCOUNT); do
358 local idx=$(($mdtno - 1))
359 local facet=mds${mdtno}
360 # if $arg include -P option, run 1 set_param per MDT on the MGS
361 # else, run set_param on each MDT
362 [[ $arg = *"-P"* ]] && facet=mgs
363 do_facet $facet $LCTL set_param $arg mdt.${MDT[$idx]}.$key$value
364 [[ $? != 0 ]] && rc=1
374 for mdtno in $(seq 1 $MDSCOUNT); do
375 local idx=$(($mdtno - 1))
376 wait_result mds${mdtno} \
377 "$LCTL get_param -n $MDT_PREFIX${idx}.$key" "$target" \
379 error "$key state is not '$target' on mds${mdtno}"
386 for mdtno in $(seq 1 $MDSCOUNT); do
387 local idx=$(($mdtno - 1))
388 local cl_user=$(do_facet mds${mdtno} $LCTL \
389 --device ${MDT[$idx]} \
390 changelog_register -n)
392 do_facet mds${mdtno} lctl set_param \
393 mdd.${MDT[$idx]}.changelog_mask="+hsm"
394 $LFS changelog_clear ${MDT[$idx]} $cl_user 0
398 changelog_cleanup() {
400 for mdtno in $(seq 1 $MDSCOUNT); do
401 local idx=$(($mdtno - 1))
402 [[ -z ${CL_USERS[$idx]} ]] && continue
403 $LFS changelog_clear ${MDT[$idx]} ${CL_USERS[$idx]} 0
404 do_facet mds${mdtno} lctl --device ${MDT[$idx]} \
405 changelog_deregister ${CL_USERS[$idx]}
410 changelog_get_flags() {
415 $LFS changelog $mdt | awk "/$cltype/ && /t=\[$fid\]/ {print \$5}"
420 local val=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.$param)
428 mdts_set_param "$opt -n" "hsm.$param" "$value"
435 mdts_set_param "" hsm_control "$cmd"
436 mdts_check_param hsm_control "$target" 10
439 cdt_set_sanity_policy() {
440 if [[ "$CDT_POLICY_HAD_CHANGED" ]]
443 mdts_set_param "" hsm.policy "+NRA"
444 mdts_set_param "" hsm.policy "-NBR"
445 CDT_POLICY_HAD_CHANGED=
450 mdts_set_param "" hsm.policy "+NRA"
451 CDT_POLICY_HAD_CHANGED=true
454 cdt_clear_no_retry() {
455 mdts_set_param "" hsm.policy "-NRA"
456 CDT_POLICY_HAD_CHANGED=true
459 cdt_set_non_blocking_restore() {
460 mdts_set_param "" hsm.policy "+NBR"
461 CDT_POLICY_HAD_CHANGED=true
464 cdt_clear_non_blocking_restore() {
465 mdts_set_param "" hsm.policy "-NBR"
466 CDT_POLICY_HAD_CHANGED=true
469 cdt_clear_mount_state() {
470 mdts_set_param "-P -d" hsm_control ""
473 cdt_set_mount_state() {
474 mdts_set_param "-P" hsm_control "$1"
478 mdts_check_param hsm_control "$1" 20
482 set_test_state disabled disabled
486 set_test_state enabled enabled
490 set_test_state shutdown stopped
494 set_test_state purge enabled
500 cdt_set_sanity_policy
504 local client_count=$1
505 if [[ $CLIENTCOUNT -lt $client_count ]]; then
506 skip "Need $client_count or more clients, have $CLIENTCOUNT"
513 $LFS path2fid $1 | tr -d '[]'
514 return ${PIPESTATUS[0]}
522 if [[ $u == "user" ]]; then
523 st=$($RUNAS $LFS hsm_state $f)
526 st=$($LFS hsm_state $f)
529 [[ $? == 0 ]] || error "$LFS hsm_state $f failed (run as $u)"
531 st=$(echo $st | cut -f 2 -d" " | tr -d "()," )
535 get_hsm_archive_id() {
538 st=$($LFS hsm_state $f)
539 [[ $? == 0 ]] || error "$LFS hsm_state $f failed"
541 local ar=$(echo $st | grep "archive_id" | cut -f5 -d" " |
550 local st=$(get_hsm_flags $f)
551 [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
554 check_hsm_flags_user() {
558 local st=$(get_hsm_flags $f user)
559 [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
562 file_creation_failure() {
567 df $MOUNT $MOUNT2 >&2
568 error "cannot create $f with $cmd, status=$err"
580 if [[ "$3" != 1 ]]; then
584 cp $1 $f || file_creation_failure cp $f $?
586 path2fid $f || error "cannot get fid on $f"
590 local file2=${1/$DIR/$DIR2}
591 dd if=/dev/urandom of=$file2 count=2 bs=1M conv=fsync ||
592 file_creation_failure dd $file2 $?
594 path2fid $1 || error "cannot get fid on $1"
598 dd if=/dev/urandom of=$1 count=1 bs=1M conv=sync ||
599 file_creation_failure dd $1 $?
600 path2fid $1 || error "cannot get fid on $1"
603 cleanup_large_files() {
604 local ratio=$(df -P $MOUNT | tail -1 | awk '{print $5}' |
606 [ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \;
609 check_enough_free_space() {
612 local need=$((nb * unit /1024))
613 local free=$(df -kP $MOUNT | tail -1 | awk '{print $4}')
614 (( $need >= $free )) && return 1
618 make_large_for_striping() {
619 local file2=${1/$DIR/$DIR2}
620 local sz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -n1)
624 check_enough_free_space 5 $sz
625 [ $? != 0 ] && return $?
627 dd if=/dev/urandom of=$file2 count=5 bs=$sz conv=fsync ||
628 file_creation_failure dd $file2 $?
630 path2fid $1 || error "cannot get fid on $1"
633 make_large_for_progress() {
634 local file2=${1/$DIR/$DIR2}
638 check_enough_free_space 39 1000000
639 [ $? != 0 ] && return $?
641 # big file is large enough, so copy time is > 30s
642 # so copytool make 1 progress
643 # size is not a multiple of 1M to avoid stripe
645 dd if=/dev/urandom of=$file2 count=39 bs=1000000 conv=fsync ||
646 file_creation_failure dd $file2 $?
648 path2fid $1 || error "cannot get fid on $1"
651 make_large_for_progress_aligned() {
652 local file2=${1/$DIR/$DIR2}
656 check_enough_free_space 33 1048576
657 [ $? != 0 ] && return $?
659 # big file is large enough, so copy time is > 30s
660 # so copytool make 1 progress
661 # size is a multiple of 1M to have stripe
663 dd if=/dev/urandom of=$file2 count=33 bs=1M conv=fsync ||
664 file_creation_failure dd $file2 $?
665 path2fid $1 || error "cannot get fid on $1"
668 make_large_for_cancel() {
669 local file2=${1/$DIR/$DIR2}
673 check_enough_free_space 103 1048576
674 [ $? != 0 ] && return $?
676 # Copy timeout is 100s. 105MB => 105s
677 dd if=/dev/urandom of=$file2 count=103 bs=1M conv=fsync ||
678 file_creation_failure dd $file2 $?
679 path2fid $1 || error "cannot get fid on $1"
685 wait_update --verbose $(facet_active_host $facet) "$@"
688 wait_request_state() {
692 # 4th arg (mdt index) is optional
694 local mds=mds$(($mdtidx + 1))
696 local cmd="$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.actions"
697 cmd+=" | awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
699 wait_result $mds "$cmd" $state 200 ||
700 error "request on $fid is not $state on $mds"
703 get_request_state() {
707 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
708 "awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
711 get_request_count() {
715 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
716 "awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
723 local cmd="$LCTL get_param -n $HSM_PARAM.actions"
724 [[ -n $fid ]] && cmd+=" | grep '$fid'"
725 cmd+=" | egrep 'WAITING|STARTED'"
727 wait_result $SINGLEMDS "$cmd" "" $timeout ||
728 error "requests did not complete"
731 wait_for_grace_delay() {
732 local val=$(get_hsm_param grace_delay)
739 # python2.6 in EL6 includes an internal json module
740 local json_parser='import json; import fileinput;'
741 json_parser+=' print "\n".join(["local %s=\"%s\"" % tuple for tuple in '
742 json_parser+='json.loads([line for line in '
743 json_parser+='fileinput.input()][0]).items()])'
745 echo $raw_event | python -c "$json_parser"
748 # populate MDT device array
754 # cleanup from previous bad setup
755 search_and_kill_copytool
757 # for recovery tests, coordinator needs to be started at mount
759 # the lustre conf must be without hsm on (like for sanity.sh)
760 echo "Set HSM on and start"
761 cdt_set_mount_state enabled
762 cdt_check_state enabled
764 echo "Start copytool"
767 echo "Set sanity-hsm HSM policy"
768 cdt_set_sanity_policy
770 # finished requests are quickly removed from list
771 set_hsm_param grace_delay 10
777 local f=$DIR/$tdir/$tfile
781 check_hsm_flags_user $f "0x00000000"
783 $RUNAS $LFS hsm_set --norelease $f ||
784 error "user could not change hsm flags"
785 check_hsm_flags_user $f "0x00000010"
787 $RUNAS $LFS hsm_clear --norelease $f ||
788 error "user could not clear hsm flags"
789 check_hsm_flags_user $f "0x00000000"
791 # User could not change those flags...
792 $RUNAS $LFS hsm_set --exists $f &&
793 error "user should not set this flag"
794 check_hsm_flags_user $f "0x00000000"
797 $LFS hsm_set --exists $f ||
798 error "root could not change hsm flags"
799 check_hsm_flags_user $f "0x00000001"
801 $LFS hsm_clear --exists $f ||
802 error "root could not clear hsm state"
803 check_hsm_flags_user $f "0x00000000"
806 run_test 1 "lfs hsm flags root/non-root access"
810 local f=$DIR/$tdir/$tfile
812 # New files are not dirty
813 check_hsm_flags $f "0x00000000"
815 # For test, we simulate an archived file.
816 $LFS hsm_set --exists $f || error "user could not change hsm flags"
817 check_hsm_flags $f "0x00000001"
819 # chmod do not put the file dirty
820 chmod 600 $f || error "could not chmod test file"
821 check_hsm_flags $f "0x00000001"
823 # chown do not put the file dirty
824 chown $RUNAS_ID $f || error "could not chown test file"
825 check_hsm_flags $f "0x00000001"
827 # truncate put the file dirty
828 $TRUNCATE $f 1 || error "could not truncate test file"
829 check_hsm_flags $f "0x00000003"
831 $LFS hsm_clear --dirty $f || error "could not clear hsm flags"
832 check_hsm_flags $f "0x00000001"
834 run_test 2 "Check file dirtyness when doing setattr"
840 # New files are not dirty
842 check_hsm_flags $f "0x00000000"
844 # For test, we simulate an archived file.
845 $LFS hsm_set --exists $f ||
846 error "user could not change hsm flags"
847 check_hsm_flags $f "0x00000001"
849 # Reading a file, does not set dirty
850 cat $f > /dev/null || error "could not read file"
851 check_hsm_flags $f "0x00000001"
853 # Open for write without modifying data, does not set dirty
854 openfile -f O_WRONLY $f || error "could not open test file"
855 check_hsm_flags $f "0x00000001"
857 # Append to a file sets it dirty
858 cp -p /etc/passwd $f.append || error "could not create file"
859 $LFS hsm_set --exists $f.append ||
860 error "user could not change hsm flags"
861 dd if=/etc/passwd of=$f.append bs=1 count=3\
862 conv=notrunc oflag=append status=noxfer ||
863 file_creation_failure dd $f.append $?
864 check_hsm_flags $f.append "0x00000003"
866 # Modify a file sets it dirty
867 cp -p /etc/passwd $f.modify || error "could not create file"
868 $LFS hsm_set --exists $f.modify ||
869 error "user could not change hsm flags"
870 dd if=/dev/zero of=$f.modify bs=1 count=3\
871 conv=notrunc status=noxfer ||
872 file_creation_failure dd $f.modify $?
873 check_hsm_flags $f.modify "0x00000003"
875 # Open O_TRUNC sets dirty
876 cp -p /etc/passwd $f.trunc || error "could not create file"
877 $LFS hsm_set --exists $f.trunc ||
878 error "user could not change hsm flags"
879 cp /etc/group $f.trunc || error "could not override a file"
880 check_hsm_flags $f.trunc "0x00000003"
882 # Mmapped a file sets dirty
883 cp -p /etc/passwd $f.mmap || error "could not create file"
884 $LFS hsm_set --exists $f.mmap ||
885 error "user could not change hsm flags"
886 multiop $f.mmap OSMWUc || error "could not mmap a file"
887 check_hsm_flags $f.mmap "0x00000003"
889 run_test 3 "Check file dirtyness when opening for write"
893 local f=$DIR/$tdir/$tfile
894 local fid=$(make_small $f)
897 local st=$(get_request_state $fid CANCEL)
898 [[ -z "$st" ]] || error "hsm_cancel must not be registered (state=$st)"
900 run_test 4 "Useless cancel must not be registered"
903 # test needs a running copytool
907 local f=$DIR/$tdir/$tfile
908 local fid=$(copy_file /etc/passwd $f)
910 wait_request_state $fid ARCHIVE SUCCEED
912 check_hsm_flags $f "0x00000009"
916 run_test 8 "Test default archive number"
920 local f=$DIR/$tdir/$tfile
921 local fid=$(copy_file /etc/passwd $f)
922 # we do not use the default one to be sure
923 local new_an=$((HSM_ARCHIVE_NUMBER + 1))
925 copytool_setup $SINGLEAGT $MOUNT $new_an
926 $LFS hsm_archive --archive $new_an $f
927 wait_request_state $fid ARCHIVE SUCCEED
929 check_hsm_flags $f "0x00000009"
933 run_test 9 "Use of explict archive number, with dedicated copytool"
936 needclients 3 || return 0
942 copytool_cleanup $(comma_list $(agts_nodes))
944 # start all of the copytools
945 for n in $(seq $AGTCOUNT); do
949 trap "copytool_cleanup $(comma_list $(agts_nodes))" EXIT
952 for n in $(seq $AGTCOUNT); do
953 file=$DIR/$tdir/$tfile.$n
954 fid=$(make_small $file)
956 $LFS hsm_archive $file || error "could not archive file $file"
957 wait_request_state $fid ARCHIVE SUCCEED
958 check_hsm_flags $file "0x00000009"
962 copytool_cleanup $(comma_list $(agts_nodes))
964 run_test 9a "Multiple remote agents"
967 # test needs a running copytool
970 mkdir -p $DIR/$tdir/d1
971 local f=$DIR/$tdir/$tfile
972 local fid=$(copy_file /etc/hosts $f)
973 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
974 error "hsm_archive failed"
975 wait_request_state $fid ARCHIVE SUCCEED
977 local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
978 error "fid $fid not in archive $HSM_ARCHIVE"
979 echo "Verifying content"
980 do_facet $SINGLEAGT diff $f $AFILE || error "archived file differs"
981 echo "Verifying hsm state "
982 check_hsm_flags $f "0x00000009"
984 echo "Verifying archive number is $HSM_ARCHIVE_NUMBER"
985 local st=$(get_hsm_archive_id $f)
986 [[ $st == $HSM_ARCHIVE_NUMBER ]] ||
987 error "Wrong archive number, $st != $HSM_ARCHIVE_NUMBER"
992 run_test 10a "Archive a file"
995 # test needs a running copytool
999 local f=$DIR/$tdir/$tfile
1000 local fid=$(copy_file /etc/hosts $f)
1001 $LFS hsm_archive $f || error "archive request failed"
1002 wait_request_state $fid ARCHIVE SUCCEED
1004 $LFS hsm_archive $f || error "archive of non dirty file failed"
1005 local cnt=$(get_request_count $fid ARCHIVE)
1006 [[ "$cnt" == "1" ]] ||
1007 error "archive of non dirty file must not make a request"
1011 run_test 10b "Archive of non dirty file must work without doing request"
1014 # test needs a running copytool
1018 local f=$DIR/$tdir/$tfile
1019 local fid=$(copy_file /etc/hosts $f)
1020 $LFS hsm_set --noarchive $f
1021 $LFS hsm_archive $f && error "archive a noarchive file must fail"
1025 run_test 10c "Check forbidden archive"
1028 # test needs a running copytool
1032 local f=$DIR/$tdir/$tfile
1033 local fid=$(copy_file /etc/hosts $f)
1034 $LFS hsm_archive $f || error "cannot archive $f"
1035 wait_request_state $fid ARCHIVE SUCCEED
1037 local ar=$(get_hsm_archive_id $f)
1038 local dflt=$(get_hsm_param default_archive_id)
1039 [[ $ar == $dflt ]] ||
1040 error "archived file is not on default archive: $ar != $dflt"
1044 run_test 10d "Archive a file on the default archive id"
1048 copy2archive /etc/hosts $tdir/$tfile
1049 local f=$DIR/$tdir/$tfile
1051 import_file $tdir/$tfile $f
1052 echo -n "Verifying released state: "
1053 check_hsm_flags $f "0x0000000d"
1055 local LSZ=$(stat -c "%s" $f)
1056 local ASZ=$(do_facet $SINGLEAGT stat -c "%s" $HSM_ARCHIVE/$tdir/$tfile)
1058 echo "Verifying imported size $LSZ=$ASZ"
1059 [[ $LSZ -eq $ASZ ]] || error "Incorrect size $LSZ != $ASZ"
1060 echo -n "Verifying released pattern: "
1061 local PTRN=$($GETSTRIPE -L $f)
1063 [[ $PTRN == 80000001 ]] || error "Is not released"
1064 local fid=$(path2fid $f)
1065 echo "Verifying new fid $fid in archive"
1067 local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
1068 error "fid $fid not in archive $HSM_ARCHIVE"
1070 run_test 11a "Import a file"
1073 # test needs a running copytool
1077 local f=$DIR/$tdir/$tfile
1078 local fid=$(copy_file /etc/hosts $f)
1079 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
1080 error "hsm_archive failed"
1081 wait_request_state $fid ARCHIVE SUCCEED
1083 local FILE_HASH=$(md5sum $f)
1088 echo "$FILE_HASH" | md5sum -c
1090 [[ $? -eq 0 ]] || error "Restored file differs"
1094 run_test 11b "Import a deleted file using its FID"
1097 # test needs a running copytool
1101 copy2archive /etc/hosts $tdir/$tfile
1103 local f=$DIR/$tdir/$tfile
1104 import_file $tdir/$tfile $f
1105 local f2=$DIR2/$tdir/$tfile
1106 echo "Verifying released state: "
1107 check_hsm_flags $f2 "0x0000000d"
1109 local fid=$(path2fid $f2)
1110 $LFS hsm_restore $f2
1111 wait_request_state $fid RESTORE SUCCEED
1113 echo "Verifying file state: "
1114 check_hsm_flags $f2 "0x00000009"
1116 do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
1118 [[ $? -eq 0 ]] || error "Restored file differs"
1122 run_test 12a "Restore an imported file explicitly"
1125 # test needs a running copytool
1129 copy2archive /etc/hosts $tdir/$tfile
1131 local f=$DIR/$tdir/$tfile
1132 import_file $tdir/$tfile $f
1133 echo "Verifying released state: "
1134 check_hsm_flags $f "0x0000000d"
1136 cat $f > /dev/null || error "File read failed"
1138 echo "Verifying file state after restore: "
1139 check_hsm_flags $f "0x00000009"
1141 do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
1143 [[ $? -eq 0 ]] || error "Restored file differs"
1147 run_test 12b "Restore an imported file implicitly"
1150 [ "$OSTCOUNT" -lt "2" ] && skip_env "skipping 2-stripe test" && return
1152 # test needs a running copytool
1156 local f=$DIR/$tdir/$tfile
1157 $LFS setstripe -c 2 $f
1159 fid=$(make_large_for_striping $f)
1160 [ $? != 0 ] && skip "not enough free space" && return
1162 local FILE_CRC=$(md5sum $f)
1164 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1165 wait_request_state $fid ARCHIVE SUCCEED
1166 $LFS hsm_release $f || error "release $f failed"
1168 echo "$FILE_CRC" | md5sum -c
1170 [[ $? -eq 0 ]] || error "Restored file differs"
1174 run_test 12c "Restore a file with stripe of 2"
1177 # test needs a running copytool
1182 local f=$DIR/$tdir/$tfile
1183 local fid=$(copy_file /etc/hosts $f)
1184 $LFS hsm_restore $f || error "restore of non archived file failed"
1185 local cnt=$(get_request_count $fid RESTORE)
1186 [[ "$cnt" == "0" ]] ||
1187 error "restore non archived must not make a request"
1188 $LFS hsm_archive $f ||
1189 error "archive request failed"
1190 wait_request_state $fid ARCHIVE SUCCEED
1191 $LFS hsm_restore $f ||
1192 error "restore of non released file failed"
1193 local cnt=$(get_request_count $fid RESTORE)
1194 [[ "$cnt" == "0" ]] ||
1195 error "restore a non dirty file must not make a request"
1199 run_test 12d "Restore of a non archived, non released file must work"\
1200 " without doing request"
1203 # test needs a running copytool
1206 mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
1207 local f=$DIR/$tdir/$tfile
1208 local fid=$(copy_file /etc/hosts $f)
1209 $LFS hsm_archive $f || error "archive request failed"
1210 wait_request_state $fid ARCHIVE SUCCEED
1213 cat /etc/hosts >> $f
1217 $LFS hsm_restore $f && error "restore a dirty file must fail"
1221 run_test 12e "Check forbidden restore"
1224 # test needs a running copytool
1228 local f=$DIR/$tdir/$tfile
1229 local fid=$(copy_file /etc/hosts $f)
1231 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1232 wait_request_state $fid ARCHIVE SUCCEED
1233 $LFS hsm_release $f || error "release of $f failed"
1235 wait_request_state $fid RESTORE SUCCEED
1237 echo -n "Verifying file state: "
1238 check_hsm_flags $f "0x00000009"
1240 diff -q /etc/hosts $f
1242 [[ $? -eq 0 ]] || error "Restored file differs"
1246 run_test 12f "Restore a released file explicitly"
1249 # test needs a running copytool
1253 local f=$DIR/$tdir/$tfile
1254 local fid=$(copy_file /etc/hosts $f)
1256 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1257 wait_request_state $fid ARCHIVE SUCCEED
1258 $LFS hsm_release $f || error "release of $f failed"
1260 diff -q /etc/hosts $f
1263 # we check we had a restore done
1264 wait_request_state $fid RESTORE SUCCEED
1266 [[ $st -eq 0 ]] || error "Restored file differs"
1270 run_test 12g "Restore a released file implicitly"
1273 needclients 2 || return 0
1275 # test needs a running copytool
1279 local f=$DIR/$tdir/$tfile
1280 local fid=$(copy_file /etc/hosts $f)
1282 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1283 wait_request_state $fid ARCHIVE SUCCEED
1284 $LFS hsm_release $f || error "release of $f failed"
1286 do_node $CLIENT2 diff -q /etc/hosts $f
1289 # we check we had a restore done
1290 wait_request_state $fid RESTORE SUCCEED
1292 [[ $st -eq 0 ]] || error "Restored file differs"
1296 run_test 12h "Restore a released file implicitly from a second node"
1299 # test needs a running copytool
1303 local f=$DIR/$tdir/$tfile
1304 local fid=$(copy_file /etc/passwd $f)
1305 $LFS hsm_archive $f || error "archive of $f failed"
1306 wait_request_state $fid ARCHIVE SUCCEED
1308 $LFS hsm_release $f || error "release of $f failed"
1312 [[ $? -eq 0 ]] || error "Restored file differs"
1316 run_test 12m "Archive/release/implicit restore"
1319 # test needs a running copytool
1323 copy2archive /etc/hosts $tdir/$tfile
1325 local f=$DIR/$tdir/$tfile
1326 import_file $tdir/$tfile $f
1328 do_facet $SINGLEAGT cmp /etc/hosts $f ||
1329 error "Restored file differs"
1331 $LFS hsm_release $f || error "release of $f failed"
1335 run_test 12n "Import/implicit restore/release"
1338 # test needs a running copytool
1342 local f=$DIR/$tdir/$tfile
1343 local fid=$(copy_file /etc/hosts $f)
1345 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1346 wait_request_state $fid ARCHIVE SUCCEED
1347 $LFS hsm_release $f || error "release of $f failed"
1349 #define OBD_FAIL_MDS_HSM_SWAP_LAYOUTS 0x152
1350 do_facet $SINGLEMDS lctl set_param fail_loc=0x152
1352 # set no retry action mode
1355 diff -q /etc/hosts $f
1358 # we check we had a restore failure
1359 wait_request_state $fid RESTORE FAILED
1361 [[ $st -eq 0 ]] && error "Restore must fail"
1363 # remove no retry action mode
1366 # check file is still released
1367 check_hsm_flags $f "0x0000000d"
1369 # retry w/o failure injection
1370 do_facet $SINGLEMDS lctl set_param fail_loc=0
1372 # to be sure previous RESTORE result is gone
1374 wait_for_grace_delay
1376 diff -q /etc/hosts $f
1379 # we check we had a restore done
1380 wait_request_state $fid RESTORE SUCCEED
1382 [[ $st -eq 0 ]] || error "Restored file differs"
1386 run_test 12o "Layout-swap failure during Restore leaves file released"
1389 # test needs a running copytool
1392 local ARC_SUBDIR="import.orig"
1396 # populate directory to be imported
1397 for d in $(seq 1 10); do
1398 local CURR_DIR="$HSM_ARCHIVE/$ARC_SUBDIR/dir.$d"
1399 do_facet $SINGLEAGT mkdir -p "$CURR_DIR"
1400 for f in $(seq 1 10); do
1401 CURR_FILE="$CURR_DIR/$tfile.$f"
1402 # write file-specific data
1403 do_facet $SINGLEAGT \
1404 "echo d=$d, f=$f, dir=$CURR_DIR, "\
1405 "file=$CURR_FILE > $CURR_FILE"
1409 import_file "$ARC_SUBDIR" $DIR/$tdir
1410 # diff lustre content and origin (triggers file restoration)
1411 # there must be 10x10 identical files, and no difference
1412 local cnt_ok=$(do_facet $SINGLEAGT diff -rs $HSM_ARCHIVE/$ARC_SUBDIR \
1413 $DIR/$tdir/$ARC_SUBDIR | grep identical | wc -l)
1414 local cnt_diff=$(do_facet $SINGLEAGT diff -r $HSM_ARCHIVE/$ARC_SUBDIR \
1415 $DIR/$tdir/$ARC_SUBDIR | wc -l)
1417 [ $cnt_diff -eq 0 ] ||
1418 error "$cnt_diff imported files differ from read data"
1419 [ $cnt_ok -eq 100 ] ||
1420 error "not enough identical files ($cnt_ok != 100)"
1424 run_test 13 "Recursively import and restore a directory"
1427 # test needs a running copytool
1432 local f=$DIR/$tdir/$tfile
1433 local fid=$(make_small $f)
1434 local sum=$(md5sum $f | awk '{print $1}')
1435 $LFS hsm_archive $f || error "could not archive file"
1436 wait_request_state $fid ARCHIVE SUCCEED
1440 # create released file (simulate llapi_hsm_import call)
1442 local fid2=$(path2fid $f)
1443 $LFS hsm_set --archived --exists $f || error "could not force hsm flags"
1444 $LFS hsm_release $f || error "could not release file"
1446 # rebind the archive to the newly created file
1447 echo "rebind $fid to $fid2"
1449 do_facet $SINGLEAGT \
1450 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
1451 --rebind $fid $fid2 $DIR" || error "could not rebind file"
1453 # restore file and compare md5sum
1454 local sum2=$(md5sum $f | awk '{print $1}')
1456 [[ $sum == $sum2 ]] || error "md5sum mismatch after restore"
1460 run_test 14 "Rebind archived file to a new fid"
1463 # test needs a running copytool
1468 local f=$DIR/$tdir/$tfile
1470 local tmpfile=$SHARED_DIRECTORY/tmp.$$
1474 for i in $(seq 1 $count); do
1475 fids[$i]=$(make_small $f.$i)
1476 sums[$i]=$(md5sum $f.$i | awk '{print $1}')
1477 $LFS hsm_archive $f.$i || error "could not archive file"
1479 wait_all_done $(($count*60))
1483 for i in $(seq 1 $count); do
1486 local fid2=$(path2fid $f.$i)
1487 # add the rebind operation to the list
1488 echo ${fids[$i]} $fid2 >> $tmpfile
1490 # set it released (simulate llapi_hsm_import call)
1491 $LFS hsm_set --archived --exists $f.$i ||
1492 error "could not force hsm flags"
1493 $LFS hsm_release $f.$i || error "could not release file"
1495 nl=$(wc -l < $tmpfile)
1496 [[ $nl == $count ]] || error "$nl files in list, $count expected"
1498 echo "rebind list of files"
1499 do_facet $SINGLEAGT \
1500 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
1501 --rebind $tmpfile $DIR" || error "could not rebind file list"
1503 # restore files and compare md5sum
1504 for i in $(seq 1 $count); do
1505 local sum2=$(md5sum $f.$i | awk '{print $1}')
1506 [[ $sum2 == ${sums[$i]} ]] ||
1507 error "md5sum mismatch after restore ($sum2 != ${sums[$i]})"
1513 run_test 15 "Rebind a list of files"
1516 # test needs a running copytool
1520 # create a known size file so we can verify transfer speed
1523 dd if=/dev/zero of=$ref bs=1M count=20
1526 local f=$DIR/$tdir/$tfile
1527 local fid=$(copy_file $ref $f)
1529 local start=$(date +%s)
1531 wait_request_state $fid ARCHIVE SUCCEED
1532 local end=$(date +%s)
1533 local duration=$((end - start))
1535 [[ $duration -ge $goal ]] ||
1536 error "Transfer is too fast $duration < $goal"
1540 run_test 16 "Test CT bandwith control option"
1545 local f=$DIR/$tdir/$tfile
1546 touch $f || error "touch $f failed"
1548 # Could not release a non-archived file
1549 $LFS hsm_release $f && error "release should not succeed"
1551 # For following tests, we must test them with HS_ARCHIVED set
1552 $LFS hsm_set --exists --archived $f || error "could not add flag"
1554 # Could not release a file if no-release is set
1555 $LFS hsm_set --norelease $f || error "could not add flag"
1556 $LFS hsm_release $f && error "release should not succeed"
1557 $LFS hsm_clear --norelease $f || error "could not remove flag"
1559 # Could not release a file if lost
1560 $LFS hsm_set --lost $f || error "could not add flag"
1561 $LFS hsm_release $f && error "release should not succeed"
1562 $LFS hsm_clear --lost $f || error "could not remove flag"
1564 # Could not release a file if dirty
1565 $LFS hsm_set --dirty $f || error "could not add flag"
1566 $LFS hsm_release $f && error "release should not succeed"
1567 $LFS hsm_clear --dirty $f || error "could not remove flag"
1569 run_test 20 "Release is not permitted"
1572 # test needs a running copytool
1576 local f=$DIR/$tdir/test_release
1578 # Create a file and check its states
1579 local fid=$(make_small $f)
1580 check_hsm_flags $f "0x00000000"
1582 # LU-4388/LU-4389 - ZFS does not report full number of blocks
1583 # used until file is flushed to disk
1584 if [ $(facet_fstype ost1) == "zfs" ]; then
1585 # this causes an OST_SYNC rpc to be sent
1586 dd if=/dev/zero of=$f bs=512 count=1 oflag=sync conv=notrunc,fsync
1587 # clear locks to reread file data
1588 cancel_lru_locks osc
1591 local orig_size=$(stat -c "%s" $f)
1592 local orig_blocks=$(stat -c "%b" $f)
1594 start_full_debug_logging
1596 $LFS hsm_archive $f || error "could not archive file"
1597 wait_request_state $fid ARCHIVE SUCCEED
1599 local blocks=$(stat -c "%b" $f)
1600 [ $blocks -eq $orig_blocks ] ||
1601 error "$f: wrong block number after archive: " \
1602 "$blocks != $orig_blocks"
1603 local size=$(stat -c "%s" $f)
1604 [ $size -eq $orig_size ] ||
1605 error "$f: wrong size after archive: $size != $orig_size"
1607 # Release and check states
1608 $LFS hsm_release $f || error "could not release file"
1609 check_hsm_flags $f "0x0000000d"
1611 blocks=$(stat -c "%b" $f)
1612 [ $blocks -gt 5 ] &&
1613 error "$f: too many blocks after release: $blocks > 5"
1614 size=$(stat -c "%s" $f)
1615 [ $size -ne $orig_size ] &&
1616 error "$f: wrong size after release: $size != $orig_size"
1618 # Check we can release an file without stripe info
1622 check_hsm_flags $f "0x00000000"
1623 $LFS hsm_archive $f || error "could not archive file"
1624 wait_request_state $fid ARCHIVE SUCCEED
1626 # Release and check states
1627 $LFS hsm_release $f || error "could not release file"
1628 check_hsm_flags $f "0x0000000d"
1630 # Release again a file that is already released is OK
1631 $LFS hsm_release $f || fail "second release should succeed"
1632 check_hsm_flags $f "0x0000000d"
1634 stop_full_debug_logging
1638 run_test 21 "Simple release tests"
1641 # test needs a running copytool
1646 local f=$DIR/$tdir/test_release
1647 local swap=$DIR/$tdir/test_swap
1649 # Create a file and check its states
1650 local fid=$(make_small $f)
1651 check_hsm_flags $f "0x00000000"
1653 $LFS hsm_archive $f || error "could not archive file"
1654 wait_request_state $fid ARCHIVE SUCCEED
1656 # Release and check states
1657 $LFS hsm_release $f || error "could not release file"
1658 check_hsm_flags $f "0x0000000d"
1661 $LFS swap_layouts $swap $f && error "swap_layouts should failed"
1666 run_test 22 "Could not swap a release file"
1669 # test needs a running copytool
1674 local f=$DIR/$tdir/test_mtime
1676 # Create a file and check its states
1677 local fid=$(make_small $f)
1678 check_hsm_flags $f "0x00000000"
1680 $LFS hsm_archive $f || error "could not archive file"
1681 wait_request_state $fid ARCHIVE SUCCEED
1683 # Set modification time in the past
1684 touch -m -a -d @978261179 $f
1686 # Release and check states
1687 $LFS hsm_release $f || error "could not release file"
1688 check_hsm_flags $f "0x0000000d"
1690 local MTIME=$(stat -c "%Y" $f)
1691 local ATIME=$(stat -c "%X" $f)
1692 [ $MTIME -eq "978261179" ] || fail "bad mtime: $MTIME"
1693 [ $ATIME -eq "978261179" ] || fail "bad atime: $ATIME"
1697 run_test 23 "Release does not change a/mtime (utime)"
1700 local file=$DIR/$tdir/$tfile
1709 # test needs a running copytool
1714 fid=$(make_small $file)
1716 # Create a file and check its states
1717 check_hsm_flags $file "0x00000000"
1719 # Ensure atime is less than mtime and ctime.
1723 atime0=$(stat -c "%X" $file)
1724 mtime0=$(stat -c "%Y" $file)
1725 ctime0=$(stat -c "%Z" $file)
1727 [ $atime0 -lt $mtime0 ] ||
1728 error "atime $atime0 is not less than mtime $mtime0"
1730 [ $atime0 -lt $ctime0 ] ||
1731 error "atime $atime0 is not less than ctime $ctime0"
1733 # Archive should not change any timestamps.
1734 $LFS hsm_archive $file || error "cannot archive '$file'"
1735 wait_request_state $fid ARCHIVE SUCCEED
1737 atime1=$(stat -c "%X" $file)
1738 mtime1=$(stat -c "%Y" $file)
1739 ctime1=$(stat -c "%Z" $file)
1741 [ $atime0 -eq $atime1 ] ||
1742 error "archive changed atime from $atime0 to $atime1"
1744 [ $mtime0 -eq $mtime1 ] ||
1745 error "archive changed mtime from $mtime0 to $mtime1"
1747 [ $ctime0 -eq $ctime1 ] ||
1748 error "archive changed ctime from $ctime0 to $ctime1"
1750 # Release should not change any timestamps.
1751 $LFS hsm_release $file || error "cannot release '$file'"
1752 check_hsm_flags $file "0x0000000d"
1754 atime1=$(stat -c "%X" $file)
1755 mtime1=$(stat -c "%Y" $file)
1756 ctime1=$(stat -c "%Z" $file)
1758 [ $atime0 -eq $atime1 ] ||
1759 error "release changed atime from $atime0 to $atime1"
1761 [ $mtime0 -eq $mtime1 ] ||
1762 error "release changed mtime from $mtime0 to $mtime1"
1764 [ $ctime0 -eq $ctime1 ] ||
1765 error "release changed ctime from $ctime0 to $ctime1"
1767 # Restore should not change atime or mtime and should not
1769 $LFS hsm_restore $file
1770 wait_request_state $fid RESTORE SUCCEED
1772 atime1=$(stat -c "%X" $file)
1773 mtime1=$(stat -c "%Y" $file)
1774 ctime1=$(stat -c "%Z" $file)
1776 [ $atime0 -eq $atime1 ] ||
1777 error "restore changed atime from $atime0 to $atime1"
1779 [ $mtime0 -eq $mtime1 ] ||
1780 error "restore changed mtime from $mtime0 to $mtime1"
1782 [ $ctime0 -le $ctime1 ] ||
1783 error "restore changed ctime from $ctime0 to $ctime1"
1787 # Once more, after unmount and mount.
1788 umount_client $MOUNT || error "cannot unmount '$MOUNT'"
1789 mount_client $MOUNT || error "cannot mount '$MOUNT'"
1791 atime1=$(stat -c "%X" $file)
1792 mtime1=$(stat -c "%Y" $file)
1793 ctime1=$(stat -c "%Z" $file)
1795 [ $atime0 -eq $atime1 ] ||
1796 error "remount changed atime from $atime0 to $atime1"
1798 [ $mtime0 -eq $mtime1 ] ||
1799 error "remount changed mtime from $mtime0 to $mtime1"
1801 [ $ctime0 -le $ctime1 ] ||
1802 error "remount changed ctime from $ctime0 to $ctime1"
1804 run_test 24a "Archive, release, and restore does not change a/mtime (i/o)"
1807 local file=$DIR/$tdir/$tfile
1813 # Test needs a running copytool.
1817 # Check that root can do HSM actions on a ordinary user's file.
1819 fid=$(make_small $file)
1820 sum0=$(md5sum $file)
1822 chown $RUNAS_ID:$RUNAS_GID $file ||
1823 error "cannot chown '$file' to '$RUNAS_ID'"
1825 chmod ugo-w $DIR/$tdir ||
1826 error "cannot chmod '$DIR/$tdir'"
1828 $LFS hsm_archive $file
1829 wait_request_state $fid ARCHIVE SUCCEED
1831 $LFS hsm_release $file
1832 check_hsm_flags $file "0x0000000d"
1834 $LFS hsm_restore $file
1835 wait_request_state $fid RESTORE SUCCEED
1837 # Check that ordinary user can get HSM state.
1838 $RUNAS $LFS hsm_state $file ||
1839 error "user '$RUNAS_ID' cannot get HSM state of '$file'"
1841 $LFS hsm_release $file
1842 check_hsm_flags $file "0x0000000d"
1844 # Check that ordinary user can accessed released file.
1845 sum1=$($RUNAS md5sum $file) ||
1846 error "user '$RUNAS_ID' cannot read '$file'"
1848 [ "$sum0" == "$sum1" ] ||
1849 error "md5sum mismatch for '$file'"
1853 run_test 24b "root can archive, release, and restore user files"
1855 cleanup_test_24c() {
1857 set_hsm_param user_request_mask RESTORE
1858 set_hsm_param group_request_mask RESTORE
1859 set_hsm_param other_request_mask RESTORE
1863 local file=$DIR/$tdir/$tfile
1864 local action=archive
1869 # test needs a running copytool
1874 # Save the default masks and check that cleanup_24c will
1875 # restore the request masks correctly.
1876 user_save=$(get_hsm_param user_request_mask)
1877 group_save=$(get_hsm_param group_request_mask)
1878 other_save=$(get_hsm_param other_request_mask)
1880 [ "$user_save" == RESTORE ] ||
1881 error "user_request_mask is '$user_save' expected 'RESTORE'"
1882 [ "$group_save" == RESTORE ] ||
1883 error "group_request_mask is '$group_save' expected 'RESTORE'"
1884 [ "$other_save" == RESTORE ] ||
1885 error "other_request_mask is '$other_save' expected 'RESTORE'"
1887 trap cleanup_test_24c EXIT
1892 chown $RUNAS_ID:nobody $file ||
1893 error "cannot chown '$file' to '$RUNAS_ID:nobody'"
1895 set_hsm_param user_request_mask ""
1896 $RUNAS $LFS hsm_$action $file &&
1897 error "$action by user should fail"
1899 set_hsm_param user_request_mask $action
1900 $RUNAS $LFS hsm_$action $file ||
1901 error "$action by user should succeed"
1906 chown nobody:$RUNAS_GID $file ||
1907 error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
1909 set_hsm_param group_request_mask ""
1910 $RUNAS $LFS hsm_$action $file &&
1911 error "$action by group should fail"
1913 set_hsm_param group_request_mask $action
1914 $RUNAS $LFS hsm_$action $file ||
1915 error "$action by group should succeed"
1920 chown nobody:nobody $file ||
1921 error "cannot chown '$file' to 'nobody:nobody'"
1923 set_hsm_param other_request_mask ""
1924 $RUNAS $LFS hsm_$action $file &&
1925 error "$action by other should fail"
1927 set_hsm_param other_request_mask $action
1928 $RUNAS $LFS hsm_$action $file ||
1929 error "$action by other should succeed"
1934 run_test 24c "check that user,group,other request masks work"
1936 cleanup_test_24d() {
1938 mount -o remount,rw $MOUNT2
1942 local file1=$DIR/$tdir/$tfile
1943 local file2=$DIR2/$tdir/$tfile
1951 fid1=$(make_small $file1)
1953 trap cleanup_test_24d EXIT
1955 mount -o remount,ro $MOUNT2
1957 fid2=$(path2fid $file2)
1958 [ "$fid1" == "$fid2" ] ||
1959 error "FID mismatch '$fid1' != '$fid2'"
1961 $LFS hsm_archive $file2 &&
1962 error "archive should fail on read-only mount"
1963 check_hsm_flags $file1 "0x00000000"
1965 $LFS hsm_archive $file1
1966 wait_request_state $fid1 ARCHIVE SUCCEED
1968 $LFS hsm_release $file1
1969 $LFS hsm_restore $file2
1970 wait_request_state $fid1 RESTORE SUCCEED
1972 $LFS hsm_release $file1 || error "cannot release '$file1'"
1973 dd if=$file2 of=/dev/null bs=1M || "cannot read '$file2'"
1975 $LFS hsm_release $file2 &&
1976 error "release should fail on read-only mount"
1981 run_test 24d "check that read-only mounts are respected"
1984 # test needs a running copytool
1988 copy2archive /etc/hosts $tdir/$tfile
1990 local f=$DIR/$tdir/$tfile
1992 import_file $tdir/$tfile $f
1994 $LFS hsm_set --lost $f
1999 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2003 run_test 25a "Restore lost file (HS_LOST flag) from import"\
2004 " (Operation not permitted)"
2007 # test needs a running copytool
2012 local f=$DIR/$tdir/$tfile
2013 local fid=$(copy_file /etc/passwd $f)
2015 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2016 wait_request_state $fid ARCHIVE SUCCEED
2019 $LFS hsm_set --lost $f
2023 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
2027 run_test 25b "Restore lost file (HS_LOST flag) after release"\
2028 " (Operation not permitted)"
2031 # test needs a running copytool
2035 local f=$DIR/$tdir/$tfile
2037 fid=$(make_large_for_progress $f)
2038 [ $? != 0 ] && skip "not enough free space" && return
2040 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2041 wait_request_state $fid ARCHIVE SUCCEED
2044 wait_request_state $fid REMOVE SUCCEED
2046 check_hsm_flags $f "0x00000000"
2050 run_test 26 "Remove the archive of a valid file"
2053 # test needs a running copytool
2057 make_archive $tdir/$tfile
2058 local f=$DIR/$tdir/$tfile
2059 import_file $tdir/$tfile $f
2060 local fid=$(path2fid $f)
2064 [[ $? != 0 ]] || error "Remove of a released file should fail"
2068 run_test 27a "Remove the archive of an imported file (Operation not permitted)"
2071 # test needs a running copytool
2075 local f=$DIR/$tdir/$tfile
2077 fid=$(make_large_for_progress $f)
2078 [ $? != 0 ] && skip "not enough free space" && return
2080 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2081 wait_request_state $fid ARCHIVE SUCCEED
2086 [[ $? != 0 ]] || error "Remove of a released file should fail"
2090 run_test 27b "Remove the archive of a relased file (Operation not permitted)"
2093 # test needs a running copytool
2097 local f=$DIR/$tdir/$tfile
2099 fid=$(make_large_for_progress $f)
2100 [ $? != 0 ] && skip "not enough free space" && return
2102 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2103 wait_request_state $fid ARCHIVE SUCCEED
2112 wait_request_state $fid REMOVE SUCCEED
2116 run_test 28 "Concurrent archive/file remove"
2119 # restore at exec cannot work on agent node (because of Linux kernel
2120 # protection of executables)
2121 needclients 2 || return 0
2123 # test needs a running copytool
2127 copy2archive /bin/true $tdir/$tfile
2129 local f=$DIR/$tdir/true
2130 import_file $tdir/$tfile $f
2132 local fid=$(path2fid $f)
2134 # set no retry action mode
2140 # remove no try action mode
2144 [[ $st == 0 ]] || error "Failed to exec a released file"
2148 run_test 30a "Restore at exec (import case)"
2151 # restore at exec cannot work on agent node (because of Linux kernel
2152 # protection of executables)
2153 needclients 2 || return 0
2155 # test needs a running copytool
2159 local f=$DIR/$tdir/true
2160 local fid=$(copy_file /bin/true $f)
2162 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2163 wait_request_state $fid ARCHIVE SUCCEED
2166 # set no retry action mode
2172 # remove no try action mode
2176 [[ $st == 0 ]] || error "Failed to exec a released file"
2180 run_test 30b "Restore at exec (release case)"
2183 needclients 2 || return 0
2185 # test needs a running copytool
2189 local f=$DIR/$tdir/SLEEP
2190 local fid=$(copy_file /bin/sleep $f)
2192 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2193 wait_request_state $fid ARCHIVE SUCCEED
2195 check_hsm_flags $f "0x0000000d"
2196 # set no retry action mode
2198 do_node $CLIENT2 "$f 10" &
2202 [[ $? == 0 ]] && error "Update during exec of released file must fail"
2204 [[ $? == 0 ]] || error "Execution failed during run"
2206 [[ $? == 0 ]] || error "Binary overwritten during exec"
2209 # remove no try action mode
2211 check_hsm_flags $f "0x00000009"
2215 run_test 30c "Update during exec of released file must fail"
2217 restore_and_check_size() {
2220 local s=$(stat -c "%s" $f)
2222 local st=$(get_hsm_flags $f)
2226 while [[ "$st" != "0x00000009" && $cpt -le 10 ]]
2228 n=$(stat -c "%s" $f)
2229 # we echo in both cases to show stat is not
2231 if [[ $n != $s ]]; then
2232 echo "size seen is $n != $s"
2235 echo "size seen is right: $n == $s"
2237 st=$(get_hsm_flags $f)
2241 if [[ $cpt -lt 10 ]]; then
2244 echo " restore is too long"
2245 wait_request_state $fid RESTORE SUCCEED
2251 # test needs a running copytool
2256 make_archive $tdir/$tfile
2257 local f=$DIR/$tdir/$tfile
2258 import_file $tdir/$tfile $f
2259 local fid=$($LFS path2fid $f)
2260 HSM_ARCHIVE_PURGE=false copytool_setup
2262 restore_and_check_size $f $fid
2265 [[ $err -eq 0 ]] || error "File size changed during restore"
2269 run_test 31a "Import a large file and check size during restore"
2273 # test needs a running copytool
2278 local f=$DIR/$tdir/$tfile
2280 fid=$(make_large_for_progress $f)
2281 [ $? != 0 ] && skip "not enough free space" && return
2283 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2284 wait_request_state $fid ARCHIVE SUCCEED
2287 restore_and_check_size $f $fid
2290 [[ $err -eq 0 ]] || error "File size changed during restore"
2294 run_test 31b "Restore a large unaligned file and check size during restore"
2297 # test needs a running copytool
2302 local f=$DIR/$tdir/$tfile
2304 fid=$(make_large_for_progress_aligned $f)
2305 [ $? != 0 ] && skip "not enough free space" && return
2307 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2308 wait_request_state $fid ARCHIVE SUCCEED
2311 restore_and_check_size $f $fid
2314 [[ $err -eq 0 ]] || error "File size changed during restore"
2318 run_test 31c "Restore a large aligned file and check size during restore"
2321 # test needs a running copytool
2326 local f=$DIR/$tdir/$tfile
2328 fid=$(make_large_for_progress $f)
2329 [ $? != 0 ] && skip "not enough free space" && return
2331 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2332 wait_request_state $fid ARCHIVE SUCCEED
2335 # to be sure wait_all_done will not be mislead by previous tests
2338 wait_for_grace_delay
2339 # Also raise grace_delay significantly so the Canceled
2340 # Restore action will stay enough long avail.
2341 local old_grace=$(get_hsm_param grace_delay)
2342 set_hsm_param grace_delay 100
2344 md5sum $f >/dev/null &
2346 wait_request_state $fid RESTORE STARTED
2351 # Check restore trigger process was killed
2352 local killed=$(ps -o pid,comm hp $pid >/dev/null)
2356 # instead of waiting+checking both Restore and Cancel ops
2357 # sequentially, wait for both to be finished and then check
2359 wait_all_done 100 $fid
2360 local rstate=$(get_request_state $fid RESTORE)
2361 local cstate=$(get_request_state $fid CANCEL)
2363 # restore orig grace_delay.
2364 set_hsm_param grace_delay $old_grace
2366 if [[ "$rstate" == "CANCELED" ]] ; then
2367 [[ "$cstate" == "SUCCEED" ]] ||
2368 error "Restore state is CANCELED and Cancel state " \
2369 "is not SUCCEED but $cstate"
2370 echo "Restore state is CANCELED, Cancel state is SUCCEED"
2371 elif [[ "$rstate" == "SUCCEED" ]] ; then
2372 [[ "$cstate" == "FAILED" ]] ||
2373 error "Restore state is SUCCEED and Cancel state " \
2374 "is not FAILED but $cstate"
2375 echo "Restore state is SUCCEED, Cancel state is FAILED"
2377 error "Restore state is $rstate and Cancel state is $cstate"
2381 error "Cannot kill process waiting for restore ($killed)"
2385 run_test 33 "Kill a restore waiting process"
2388 # test needs a running copytool
2393 local f=$DIR/$tdir/$tfile
2395 fid=$(make_large_for_progress $f)
2396 [ $? != 0 ] && skip "not enough free space" && return
2398 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2399 wait_request_state $fid ARCHIVE SUCCEED
2402 md5sum $f >/dev/null &
2404 wait_request_state $fid RESTORE STARTED
2406 rm $f || error "rm $f failed"
2407 # rm must not block during restore
2408 wait_request_state $fid RESTORE STARTED
2410 wait_request_state $fid RESTORE SUCCEED
2411 # check md5sum pgm finished
2412 local there=$(ps -o pid,comm hp $pid >/dev/null)
2413 [[ -z $there ]] || error "Restore initiator does not exit"
2415 local rc=$(wait $pid)
2416 [[ $rc -eq 0 ]] || error "Restore initiator failed with $rc"
2420 run_test 34 "Remove file during restore"
2423 # test needs a running copytool
2428 local f=$DIR/$tdir/$tfile
2429 local f1=$DIR/$tdir/$tfile-1
2431 fid=$(make_large_for_progress $f)
2432 [ $? != 0 ] && skip "not enough free space" && return
2434 local fid1=$(copy_file /etc/passwd $f1)
2435 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2436 wait_request_state $fid ARCHIVE SUCCEED
2439 md5sum $f >/dev/null &
2441 wait_request_state $fid RESTORE STARTED
2443 mv $f1 $f || error "mv $f1 $f failed"
2444 # mv must not block during restore
2445 wait_request_state $fid RESTORE STARTED
2447 wait_request_state $fid RESTORE SUCCEED
2448 # check md5sum pgm finished
2449 local there=$(ps -o pid,comm hp $pid >/dev/null)
2450 [[ -z $there ]] || error "Restore initiator does not exit"
2452 local rc=$(wait $pid)
2453 [[ $rc -eq 0 ]] || error "Restore initiator failed with $rc"
2456 [[ $fid2 == $fid1 ]] || error "Wrong fid after mv $fid2 != $fid1"
2460 run_test 35 "Overwrite file during restore"
2463 # test needs a running copytool
2468 local f=$DIR/$tdir/$tfile
2470 fid=$(make_large_for_progress $f)
2471 [ $? != 0 ] && skip "not enough free space" && return
2473 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2474 wait_request_state $fid ARCHIVE SUCCEED
2477 md5sum $f >/dev/null &
2479 wait_request_state $fid RESTORE STARTED
2482 # rm must not block during restore
2483 wait_request_state $fid RESTORE STARTED
2485 wait_request_state $fid RESTORE SUCCEED
2486 # check md5sum pgm finished
2487 local there=$(ps -o pid,comm hp $pid >/dev/null)
2489 error "Restore initiator does not exit"
2491 local rc=$(wait $pid)
2493 error "Restore initiator failed with $rc"
2497 run_test 36 "Move file during restore"
2504 for n in $(seq 1 $count); do
2505 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $prefix.$n
2507 echo "$count archive requests submitted"
2511 local stream_count=4
2512 local file_count=100
2514 local f=$DIR/$tdir/$tfile
2519 for i in $(seq 1 $file_count); do
2520 for p in $(seq 1 $stream_count); do
2521 fid=$(copy_file /etc/hosts $f.$p.$i)
2524 # force copytool to use a local/temp archive dir to ensure best
2525 # performance vs remote/NFS mounts used in auto-tests
2526 if do_facet $SINGLEAGT "df --local $HSM_ARCHIVE" >/dev/null 2>&1 ; then
2529 copytool_setup $SINGLEAGT $MOUNT $HSM_ARCHIVE_NUMBER $TMP/$tdir
2531 # to be sure wait_all_done will not be mislead by previous tests
2533 wait_for_grace_delay
2535 # start archive streams in background (archive files in parallel)
2536 for p in $(seq 1 $stream_count); do
2537 multi_archive $f.$p $file_count &
2540 echo -n "Wait for all requests being enqueued..."
2546 run_test 40 "Parallel archive requests"
2549 # test needs a running copytool
2553 local f=$DIR/$tdir/$tfile
2554 local fid=$(copy_file /etc/motd $f 1)
2556 $LFS hsm_archive $f || error "could not archive file"
2557 wait_request_state $fid ARCHIVE SUCCEED
2558 check_hsm_flags $f "0x00000009"
2560 multiop_bg_pause $f O_c || error "multiop failed"
2564 client_up || client_up || true
2566 kill -USR1 $MULTIPID
2567 wait $MULTIPID || error "multiop close failed"
2569 check_hsm_flags $f "0x0000000b"
2573 run_test 52 "Opened for write file on an evicted client should be set dirty"
2576 # test needs a running copytool
2580 local f=$DIR/$tdir/$tfile
2581 local fid=$(copy_file /etc/motd $f 1)
2583 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2584 error "could not archive file"
2585 wait_request_state $fid ARCHIVE SUCCEED
2586 check_hsm_flags $f "0x00000009"
2588 multiop_bg_pause $f o_c || error "multiop failed"
2592 client_up || client_up || true
2594 kill -USR1 $MULTIPID
2595 wait $MULTIPID || error "multiop close failed"
2597 check_hsm_flags $f "0x00000009"
2601 run_test 53 "Opened for read file on an evicted client should not be set dirty"
2604 # test needs a running copytool
2608 local f=$DIR/$tdir/$tfile
2609 local fid=$(make_small $f)
2611 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2612 error "could not archive file"
2613 wait_request_state $fid ARCHIVE STARTED
2615 check_hsm_flags $f "0x00000001"
2617 # Avoid coordinator resending this request as soon it has failed.
2622 wait_request_state $fid ARCHIVE FAILED
2624 check_hsm_flags $f "0x00000003"
2629 run_test 54 "Write during an archive cancels it"
2632 # test needs a running copytool
2636 local f=$DIR/$tdir/$tfile
2637 local fid=$(make_small $f)
2639 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2640 error "could not archive file"
2641 wait_request_state $fid ARCHIVE STARTED
2643 check_hsm_flags $f "0x00000001"
2645 # Avoid coordinator resending this request as soon it has failed.
2648 $TRUNCATE $f 1024 || error "truncate failed"
2650 wait_request_state $fid ARCHIVE FAILED
2652 check_hsm_flags $f "0x00000003"
2657 run_test 55 "Truncate during an archive cancels it"
2660 # test needs a running copytool
2664 local f=$DIR/$tdir/$tfile
2666 fid=$(make_large_for_progress $f)
2667 [ $? != 0 ] && skip "not enough free space" && return
2669 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2670 error "could not archive file"
2671 wait_request_state $fid ARCHIVE STARTED
2673 check_hsm_flags $f "0x00000001"
2675 # Change metadata and sync to be sure we are not changing only
2680 wait_request_state $fid ARCHIVE SUCCEED
2682 check_hsm_flags $f "0x00000009"
2686 run_test 56 "Setattr during an archive is ok"
2689 # Need one client for I/O, one for request
2690 needclients 2 || return 0
2692 # test needs a running copytool
2696 local f=$DIR/$tdir/test_archive_remote
2697 # Create a file on a remote node
2698 do_node $CLIENT2 "dd if=/dev/urandom of=$f bs=1M "\
2699 "count=2 conv=fsync"
2702 do_node $CLIENT2 "$LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f" ||
2703 error "hsm_archive failed"
2704 local fid=$(path2fid $f)
2705 wait_request_state $fid ARCHIVE SUCCEED
2707 # Release and implicit restore it
2708 do_node $CLIENT2 "$LFS hsm_release $f" ||
2709 error "hsm_release failed"
2710 do_node $CLIENT2 "md5sum $f" ||
2711 error "hsm_restore failed"
2713 wait_request_state $fid RESTORE SUCCEED
2717 run_test 57 "Archive a file with dirty cache on another node"
2719 truncate_released_file() {
2723 local sz=$(stat -c %s $src_file)
2724 local f=$DIR/$tdir/$tfile
2725 local fid=$(copy_file $1 $f)
2729 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2730 error "could not archive file"
2731 wait_request_state $fid ARCHIVE SUCCEED
2733 $LFS hsm_release $f || error "could not release file"
2735 $TRUNCATE $f $trunc_to || error "truncate failed"
2738 local sz1=$(stat -c %s $f)
2739 [[ $sz1 == $trunc_to ]] ||
2740 error "size after trunc: $sz1 expect $trunc_to, original $sz"
2743 check_hsm_flags $f "0x0000000b"
2745 local state=$(get_request_state $fid RESTORE)
2746 [[ "$state" == "SUCCEED" ]] ||
2747 error "truncate $sz does not trig restore, state = $state"
2749 $TRUNCATE $ref $trunc_to
2750 cmp $ref $f || error "file data wrong after truncate"
2756 # test needs a running copytool
2761 local sz=$(stat -c %s /etc/passwd)
2763 echo "truncate up from $sz to $((sz*2))"
2764 truncate_released_file /etc/passwd $((sz*2))
2766 echo "truncate down from $sz to $((sz/2))"
2767 truncate_released_file /etc/passwd $((sz/2))
2769 echo "truncate to 0"
2770 truncate_released_file /etc/passwd 0
2774 run_test 58 "Truncate a released file will trigger restore"
2777 # This test validates the fix for LU-4512. Ensure that the -u
2778 # option changes the progress reporting interval from the
2779 # default (30 seconds) to the user-specified interval.
2781 local progress_timeout=$((interval * 4))
2783 # test needs a new running copytool
2785 HSMTOOL_UPDATE_INTERVAL=$interval copytool_setup
2788 local f=$DIR/$tdir/$tfile
2790 fid=$(make_large_for_progress $f)
2791 [ $? != 0 ] && skip "not enough free space" && return
2794 local mdt=${MDT_PREFIX}${mdtidx}
2795 local mds=mds$((mdtidx + 1))
2797 # Wait for copytool to register
2798 wait_update_facet $mds \
2799 "$LCTL get_param -n ${mdt}.hsm.agents | grep -o ^uuid" \
2800 uuid 100 || error "coyptool failed to register with $mdt"
2802 local start_at=$(date +%s)
2803 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2804 error "could not archive file"
2806 local cmd="$LCTL get_param -n ${mdt}.hsm.active_requests"
2807 cmd+=" | awk '/'$fid'.*action=ARCHIVE/ {print \\\$12}' | cut -f2 -d="
2813 echo -n "Expecting a progress update within $progress_timeout seconds... "
2815 RESULT=$(do_node $(facet_active_host $mds) "$cmd")
2816 if [ $RESULT -gt 0 ]; then
2817 echo "$RESULT bytes copied in $WAIT seconds."
2819 elif [ $WAIT -ge $progress_timeout ]; then
2820 error "Timed out waiting for progress update!"
2823 WAIT=$((WAIT + sleep))
2827 local finish_at=$(date +%s)
2828 local elapsed=$((finish_at - start_at))
2830 # Ensure that the progress update occurred within the expected window.
2831 if [ $elapsed -lt $interval ]; then
2832 error "Expected progress update after at least $interval seconds"
2838 run_test 60 "Changing progress update interval from default"
2841 # test needs a new running copytool
2843 copytool_monitor_setup
2844 HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
2846 # Just start and stop the copytool to generate events.
2849 # Wait for the copytool to register.
2850 wait_update --verbose $(facet_active_host mds1) \
2851 "$LCTL get_param -n ${MDT_PREFIX}0.hsm.agents | grep -o ^uuid" \
2853 error "copytool failed to register with MDT0000"
2857 local REGISTER_EVENT
2858 local UNREGISTER_EVENT
2859 while read event; do
2860 local parsed=$(parse_json_event "$event")
2861 if [ -z "$parsed" ]; then
2862 error "Copytool sent malformed event: $event"
2866 if [ $event_type == "REGISTER" ]; then
2867 REGISTER_EVENT=$event
2868 elif [ $event_type == "UNREGISTER" ]; then
2869 UNREGISTER_EVENT=$event
2871 done < <(echo $"$(get_copytool_event_log)")
2873 if [ -z "$REGISTER_EVENT" ]; then
2874 error "Copytool failed to send register event to FIFO"
2877 if [ -z "$UNREGISTER_EVENT" ]; then
2878 error "Copytool failed to send unregister event to FIFO"
2881 copytool_monitor_cleanup
2882 echo "Register/Unregister events look OK."
2884 run_test 70 "Copytool logs JSON register/unregister events to FIFO"
2887 # Bump progress interval for livelier events.
2890 # test needs a new running copytool
2892 copytool_monitor_setup
2893 HSMTOOL_UPDATE_INTERVAL=$interval \
2894 HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
2897 local f=$DIR/$tdir/$tfile
2899 fid=$(make_large_for_progress $f)
2900 [ $? != 0 ] && skip "not enough free space" && return
2902 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2903 error "could not archive file"
2904 wait_request_state $fid ARCHIVE SUCCEED
2906 local expected_fields="event_time data_fid source_fid"
2907 expected_fields+=" total_bytes current_bytes"
2911 while read event; do
2912 # Make sure we're not getting anything from previous events.
2913 for field in $expected_fields; do
2917 local parsed=$(parse_json_event "$event")
2918 if [ -z "$parsed" ]; then
2919 error "Copytool sent malformed event: $event"
2923 if [ $event_type == "ARCHIVE_START" ]; then
2926 elif [ $event_type == "ARCHIVE_FINISH" ]; then
2929 elif [ $event_type != "ARCHIVE_RUNNING" ]; then
2933 # Do some simple checking of the progress update events.
2934 for expected_field in $expected_fields; do
2935 if [ -z ${!expected_field+x} ]; then
2936 error "Missing $expected_field field in event"
2940 if [ $total_bytes -eq 0 ]; then
2941 error "Expected total_bytes to be > 0"
2944 # These should be identical throughout an archive
2946 if [ $source_fid != $data_fid ]; then
2947 error "Expected source_fid to equal data_fid"
2949 done < <(echo $"$(get_copytool_event_log)")
2951 if [ -z "$START_EVENT" ]; then
2952 error "Copytool failed to send archive start event to FIFO"
2955 if [ -z "$FINISH_EVENT" ]; then
2956 error "Copytool failed to send archive finish event to FIFO"
2959 echo "Archive events look OK."
2963 copytool_monitor_cleanup
2965 run_test 71 "Copytool logs JSON archive events to FIFO"
2968 # Bump progress interval for livelier events.
2971 # test needs a new running copytool
2973 copytool_monitor_setup
2974 HSMTOOL_UPDATE_INTERVAL=$interval \
2975 HSMTOOL_EVENT_FIFO=$HSMTOOL_MONITOR_DIR/fifo copytool_setup
2976 local test_file=$HSMTOOL_MONITOR_DIR/file
2978 local cmd="dd if=/dev/urandom of=$test_file count=16 bs=1000000 "
2980 do_facet $SINGLEAGT "$cmd" ||
2981 error "cannot create $test_file on $SINGLEAGT"
2982 copy2archive $test_file $tdir/$tfile
2985 local f=$DIR/$tdir/$tfile
2986 import_file $tdir/$tfile $f
2987 f=$DIR2/$tdir/$tfile
2988 echo "Verifying released state: "
2989 check_hsm_flags $f "0x0000000d"
2991 local fid=$(path2fid $f)
2993 wait_request_state $fid RESTORE SUCCEED
2995 local expected_fields="event_time data_fid source_fid"
2996 expected_fields+=" total_bytes current_bytes"
3000 while read event; do
3001 # Make sure we're not getting anything from previous events.
3002 for field in $expected_fields; do
3006 local parsed=$(parse_json_event "$event")
3007 if [ -z "$parsed" ]; then
3008 error "Copytool sent malformed event: $event"
3012 if [ $event_type == "RESTORE_START" ]; then
3014 if [ $source_fid != $data_fid ]; then
3015 error "source_fid should == data_fid at start"
3018 elif [ $event_type == "RESTORE_FINISH" ]; then
3020 if [ $source_fid != $data_fid ]; then
3021 error "source_fid should == data_fid at finish"
3024 elif [ $event_type != "RESTORE_RUNNING" ]; then
3028 # Do some simple checking of the progress update events.
3029 for expected_field in $expected_fields; do
3030 if [ -z ${!expected_field+x} ]; then
3031 error "Missing $expected_field field in event"
3035 if [ $total_bytes -eq 0 ]; then
3036 error "Expected total_bytes to be > 0"
3039 # When a restore starts out, the data fid is the same as the
3040 # source fid. After the restore has gotten going, we learn
3041 # the new data fid. Once the restore has finished, the source
3042 # fid is set to the new data fid.
3044 # We test this because some monitoring software may depend on
3045 # this behavior. If it changes, then the consumers of these
3046 # events may need to be modified.
3047 if [ $source_fid == $data_fid ]; then
3048 error "source_fid should != data_fid during restore"
3050 done < <(echo $"$(get_copytool_event_log)")
3052 if [ -z "$START_EVENT" ]; then
3053 error "Copytool failed to send restore start event to FIFO"
3056 if [ -z "$FINISH_EVENT" ]; then
3057 error "Copytool failed to send restore finish event to FIFO"
3060 echo "Restore events look OK."
3064 copytool_monitor_cleanup
3068 run_test 72 "Copytool logs JSON restore events to FIFO"
3071 file_count=51 # Max number of files constrained by LNET message size
3072 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3073 local f=$DIR/$tdir/$tfile
3074 local FILELIST=/tmp/filelist.txt
3078 for i in $(seq 1 $file_count); do
3079 fid=$(copy_file /etc/hosts $f.$i)
3080 echo $f.$i >> $FILELIST
3082 # force copytool to use a local/temp archive dir to ensure best
3083 # performance vs remote/NFS mounts used in auto-tests
3084 if do_facet $SINGLEAGT "df --local $HSM_ARCHIVE" >/dev/null 2>&1 ; then
3087 local dai=$(get_hsm_param default_archive_id)
3088 copytool_setup $SINGLEAGT $MOUNT $dai $TMP/$tdir
3090 # to be sure wait_all_done will not be mislead by previous tests
3092 wait_for_grace_delay
3093 $LFS hsm_archive --filelist $FILELIST ||
3094 error "cannot archive a file list"
3096 $LFS hsm_release --filelist $FILELIST ||
3097 error "cannot release a file list"
3098 $LFS hsm_restore --filelist $FILELIST ||
3099 error "cannot restore a file list"
3103 run_test 90 "Archive/restore a file list"
3105 double_verify_reset_hsm_param() {
3107 echo "Testing $HSM_PARAM.$p"
3108 local val=$(get_hsm_param $p)
3110 local val2=$(($val * 2))
3111 set_hsm_param $p $val2
3112 val=$(get_hsm_param $p)
3113 [[ $val == $val2 ]] ||
3114 error "$HSM_PARAM.$p: $val != $val2 should be (2 * $save)"
3115 echo "Set $p to 0 must failed"
3119 set_hsm_param $p $save
3121 if [[ $rc == 0 ]]; then
3122 error "we must not be able to set $HSM_PARAM.$p to 0"
3127 double_verify_reset_hsm_param loop_period
3128 double_verify_reset_hsm_param grace_delay
3129 double_verify_reset_hsm_param active_request_timeout
3130 double_verify_reset_hsm_param max_requests
3131 double_verify_reset_hsm_param default_archive_id
3133 run_test 100 "Set coordinator /proc tunables"
3140 run_test 102 "Verify coordinator control"
3143 # test needs a running copytool
3150 for i in $(seq 1 20); do
3151 fid=$(copy_file /etc/passwd $DIR/$tdir/$i)
3153 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/*
3157 echo "Current requests"
3158 local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3159 $HSM_PARAM.actions |\
3160 grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
3162 [[ -z "$res" ]] || error "Some request have not been canceled"
3166 run_test 103 "Purge all requests"
3171 # test needs a running copytool
3175 local f=$DIR/$tdir/$tfile
3177 fid=$(make_large_for_progress $f)
3178 [ $? != 0 ] && skip "not enough free space" && return
3180 # if cdt is on, it can serve too quickly the request
3182 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
3183 local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3184 $HSM_PARAM.actions |\
3185 grep $fid | cut -f16 -d=")
3188 [[ "$data1" == "$DATAHEX" ]] ||
3189 error "Data field in records is ($data1) and not ($DATAHEX)"
3193 run_test 104 "Copy tool data field"
3200 for i in $(seq -w 1 10); do
3201 cp /etc/passwd $DIR/$tdir/$i
3202 $LFS hsm_archive $DIR/$tdir/$i
3204 local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3205 $HSM_PARAM.actions |\
3206 grep WAITING | wc -l")
3209 local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3210 $HSM_PARAM.actions |\
3211 grep WAITING | wc -l")
3214 [[ "$reqcnt1" == "$reqcnt2" ]] ||
3215 error "Requests count after shutdown $reqcnt2 != "\
3216 "before shutdown $reqcnt1"
3218 run_test 105 "Restart of coordinator"
3220 get_agent_by_uuid_mdt() {
3223 local mds=mds$(($mdtidx + 1))
3224 do_facet $mds "$LCTL get_param -n ${MDT_PREFIX}${mdtidx}.hsm.agents |\
3228 check_agent_registered_by_mdt() {
3231 local mds=mds$(($mdtidx + 1))
3232 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
3233 if [[ ! -z "$agent" ]]; then
3234 echo "found agent $agent on $mds"
3236 error "uuid $uuid not found in agent list on $mds"
3240 check_agent_unregistered_by_mdt() {
3243 local mds=mds$(($mdtidx + 1))
3244 local agent=$(get_agent_by_uuid_mdt $uuid $mdtidx)
3245 if [[ -z "$agent" ]]; then
3246 echo "uuid not found in agent list on $mds"
3248 error "uuid found in agent list on $mds: $agent"
3252 check_agent_registered() {
3255 for mdsno in $(seq 1 $MDSCOUNT); do
3256 check_agent_registered_by_mdt $uuid $((mdsno - 1))
3260 check_agent_unregistered() {
3263 for mdsno in $(seq 1 $MDSCOUNT); do
3264 check_agent_unregistered_by_mdt $uuid $((mdsno - 1))
3269 local uuid=$(do_rpc_nodes $(facet_active_host $SINGLEAGT) \
3270 get_client_uuid $MOUNT | cut -d' ' -f2)
3273 check_agent_registered $uuid
3275 search_copytools || error "No copytool found"
3278 check_agent_unregistered $uuid
3281 check_agent_registered $uuid
3285 run_test 106 "Copytool register/unregister"
3288 # test needs a running copytool
3290 # create and archive file
3292 local f1=$DIR/$tdir/$tfile
3293 local fid=$(copy_file /etc/passwd $f1)
3294 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
3295 wait_request_state $fid ARCHIVE SUCCEED
3296 # shutdown and restart MDS
3298 # check the copytool still gets messages from MDT
3299 local f2=$DIR/$tdir/2
3300 local fid=$(copy_file /etc/passwd $f2)
3301 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
3302 # main check of this sanity: this request MUST succeed
3303 wait_request_state $fid ARCHIVE SUCCEED
3306 run_test 107 "Copytool re-register after MDS restart"
3308 policy_set_and_test()
3312 do_facet $SINGLEMDS $LCTL set_param "$HSM_PARAM.policy=\\\"$change\\\""
3313 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3314 [[ "$policy" == "$target" ]] ||
3315 error "Wrong policy after '$change': '$policy' != '$target'"
3319 # to force default policy setting if error
3320 CDT_POLICY_HAD_CHANGED=true
3322 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3323 local default="NonBlockingRestore [NoRetryAction]"
3324 [[ "$policy" == "$default" ]] ||
3325 error "default policy has changed,"\
3326 " '$policy' != '$default' update the test"
3327 policy_set_and_test "+NBR" "[NonBlockingRestore] [NoRetryAction]"
3328 policy_set_and_test "+NRA" "[NonBlockingRestore] [NoRetryAction]"
3329 policy_set_and_test "-NBR" "NonBlockingRestore [NoRetryAction]"
3330 policy_set_and_test "-NRA" "NonBlockingRestore NoRetryAction"
3331 policy_set_and_test "NRA NBR" "[NonBlockingRestore] [NoRetryAction]"
3332 # useless bacause we know but safer for futur changes to use real value
3333 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
3334 echo "Next set_param must failed"
3335 policy_set_and_test "wrong" "$policy"
3338 echo "Back to default policy"
3339 cdt_set_sanity_policy
3341 run_test 109 "Policy display/change"
3344 # test needs a running copytool
3349 copy2archive /etc/passwd $tdir/$tfile
3351 local f=$DIR/$tdir/$tfile
3352 import_file $tdir/$tfile $f
3353 local fid=$(path2fid $f)
3355 cdt_set_non_blocking_restore
3360 wait_request_state $fid RESTORE SUCCEED
3361 cdt_clear_non_blocking_restore
3365 error "md5sum returns $st != 1, "\
3366 "should also perror ENODATA (No data available)"
3370 run_test 110a "Non blocking restore policy (import case)"
3373 # test needs a running copytool
3377 local f=$DIR/$tdir/$tfile
3378 local fid=$(copy_file /etc/passwd $f)
3379 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3380 wait_request_state $fid ARCHIVE SUCCEED
3383 cdt_set_non_blocking_restore
3388 wait_request_state $fid RESTORE SUCCEED
3389 cdt_clear_non_blocking_restore
3393 error "md5sum returns $st != 1, "\
3394 "should also perror ENODATA (No data available)"
3398 run_test 110b "Non blocking restore policy (release case)"
3401 # test needs a running copytool
3405 copy2archive /etc/passwd $tdir/$tfile
3407 local f=$DIR/$tdir/$tfile
3409 import_file $tdir/$tfile $f
3410 local fid=$(path2fid $f)
3414 copytool_remove_backend $fid
3417 wait_request_state $fid RESTORE FAILED
3424 [[ $st == 0 ]] || error "Restore does not failed"
3428 run_test 111a "No retry policy (import case), restore will error"\
3429 " (No such file or directory)"
3432 # test needs a running copytool
3436 local f=$DIR/$tdir/$tfile
3437 local fid=$(copy_file /etc/passwd $f)
3439 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3440 wait_request_state $fid ARCHIVE SUCCEED
3443 copytool_remove_backend $fid
3446 wait_request_state $fid RESTORE FAILED
3453 [[ $st == 0 ]] || error "Restore does not failed"
3457 run_test 111b "No retry policy (release case), restore will error"\
3458 " (No such file or directory)"
3461 # test needs a running copytool
3465 local f=$DIR/$tdir/$tfile
3466 local fid=$(copy_file /etc/passwd $f)
3468 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3469 local l=$($LFS hsm_action $f)
3471 local res=$(echo $l | cut -f 2- -d" " | grep ARCHIVE)
3475 wait_request_state $fid ARCHIVE SUCCEED
3478 [[ ! -z "$res" ]] || error "action is $l which is not an ARCHIVE"
3482 run_test 112 "State of recorded request"
3485 # test needs a running copytool
3489 local f=$DIR/$tdir/$tfile
3491 fid=$(make_large_for_cancel $f)
3492 [ $? != 0 ] && skip "not enough free space" && return
3494 # test with cdt on is made in test_221
3496 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3499 wait_request_state $fid ARCHIVE CANCELED
3500 wait_request_state $fid CANCEL SUCCEED
3504 run_test 200 "Register/Cancel archive"
3507 # test needs a running copytool
3511 local f=$DIR/$tdir/$tfile
3512 make_archive $tdir/$tfile
3513 import_file $tdir/$tfile $f
3514 local fid=$(path2fid $f)
3516 # test with cdt on is made in test_222
3521 wait_request_state $fid RESTORE CANCELED
3522 wait_request_state $fid CANCEL SUCCEED
3526 run_test 201 "Register/Cancel restore"
3529 # test needs a running copytool
3533 local f=$DIR/$tdir/$tfile
3535 fid=$(make_large_for_progress $f)
3536 [ $? != 0 ] && skip "not enough free space" && return
3538 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3539 wait_request_state $fid ARCHIVE SUCCEED
3545 wait_request_state $fid REMOVE CANCELED
3549 run_test 202 "Register/Cancel remove"
3552 # test needs a running copytool
3557 local f=$DIR/$tdir/$tfile
3558 local fid=$(copy_file /etc/passwd $f)
3562 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3563 wait_request_state $fid ARCHIVE SUCCEED
3565 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3569 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3573 run_test 220 "Changelog for archive"
3576 # test needs a running copytool
3581 local f=$DIR/$tdir/$tfile
3583 fid=$(make_large_for_cancel $f)
3584 [ $? != 0 ] && skip "not enough free space" && return
3588 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3589 wait_request_state $fid ARCHIVE STARTED
3591 wait_request_state $fid ARCHIVE CANCELED
3592 wait_request_state $fid CANCEL SUCCEED
3594 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3597 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3601 run_test 221 "Changelog for archive canceled"
3604 # test needs a running copytool
3608 copy2archive /etc/passwd $tdir/$tfile
3610 local f=$DIR/$tdir/$tfile
3611 import_file $tdir/$tfile $f
3612 local fid=$(path2fid $f)
3617 wait_request_state $fid RESTORE SUCCEED
3619 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3622 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3626 run_test 222a "Changelog for explicit restore"
3629 # test needs a running copytool
3633 local f=$DIR/$tdir/$tfile
3634 local fid=$(copy_file /etc/passwd $f)
3637 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3638 wait_request_state $fid ARCHIVE SUCCEED
3643 wait_request_state $fid RESTORE SUCCEED
3645 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3648 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
3652 run_test 222b "Changelog for implicit restore"
3655 # test needs a running copytool
3660 local f=$DIR/$tdir/$tfile
3661 make_archive $tdir/$tfile
3665 import_file $tdir/$tfile $f
3666 local fid=$(path2fid $f)
3669 wait_request_state $fid RESTORE STARTED
3671 wait_request_state $fid RESTORE CANCELED
3672 wait_request_state $fid CANCEL SUCCEED
3674 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3677 [[ $flags == $target ]] ||
3678 error "Changelog flag is $flags not $target"
3682 run_test 223a "Changelog for restore canceled (import case)"
3685 # test needs a running copytool
3690 local f=$DIR/$tdir/$tfile
3692 fid=$(make_large_for_progress $f)
3693 [ $? != 0 ] && skip "not enough free space" && return
3696 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3697 wait_request_state $fid ARCHIVE SUCCEED
3700 wait_request_state $fid RESTORE STARTED
3702 wait_request_state $fid RESTORE CANCELED
3703 wait_request_state $fid CANCEL SUCCEED
3705 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -1)
3708 [[ $flags == $target ]] ||
3709 error "Changelog flag is $flags not $target"
3713 run_test 223b "Changelog for restore canceled (release case)"
3716 # test needs a running copytool
3721 local f=$DIR/$tdir/$tfile
3722 local fid=$(copy_file /etc/passwd $f)
3725 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3726 wait_request_state $fid ARCHIVE SUCCEED
3729 wait_request_state $fid REMOVE SUCCEED
3731 local flags=$(changelog_get_flags ${MDT[0]} HSM $fid | tail -n 1)
3734 [[ $flags == $target ]] ||
3735 error "Changelog flag is $flags not $target"
3739 run_test 224 "Changelog for remove"
3742 # test needs a running copytool
3745 # test is not usable because remove request is too fast
3746 # so it is always finished before cancel can be done ...
3747 echo "Test disabled"
3752 local f=$DIR/$tdir/$tfile
3754 fid=$(make_large_for_progress $f)
3755 [ $? != 0 ] && skip "not enough free space" && return
3758 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3759 wait_request_state $fid ARCHIVE SUCCEED
3761 # if cdt is on, it can serve too quickly the request
3766 wait_request_state $fid REMOVE CANCELED
3767 wait_request_state $fid CANCEL SUCCEED
3769 flags=$(changelog_get_flags ${MDT[0]} RENME $fid2)
3770 local flags=$($LFS changelog ${MDT[0]} | grep HSM | grep $fid |
3771 tail -n 1 | awk '{print $5}')
3774 [[ $flags == $target ]] ||
3775 error "Changelog flag is $flags not $target"
3779 run_test 225 "Changelog for remove canceled"
3782 # test needs a running copytool
3787 local f1=$DIR/$tdir/$tfile-1
3788 local f2=$DIR/$tdir/$tfile-2
3789 local f3=$DIR/$tdir/$tfile-3
3790 local fid1=$(copy_file /etc/passwd $f1)
3791 local fid2=$(copy_file /etc/passwd $f2)
3792 copy_file /etc/passwd $f3
3795 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
3796 wait_request_state $fid1 ARCHIVE SUCCEED
3798 $LFS hsm_archive $f2
3799 wait_request_state $fid2 ARCHIVE SUCCEED
3801 rm $f1 || error "rm $f1 failed"
3803 local flags=$(changelog_get_flags ${MDT[0]} UNLNK $fid1)
3806 [[ $flags == $target ]] ||
3807 error "Changelog flag is $flags not $target"
3809 mv $f3 $f2 || error "mv $f3 $f2 failed"
3811 flags=$(changelog_get_flags ${MDT[0]} RENME $fid2)
3814 [[ $flags == $target ]] ||
3815 error "Changelog flag is $flags not $target"
3819 run_test 226 "changelog for last rm/mv with exiting archive"
3821 check_flags_changes() {
3829 $LFS hsm_set --$hsm_flag $f ||
3830 error "Cannot set $hsm_flag on $f"
3831 local flags=($(changelog_get_flags ${MDT[0]} HSM $fid))
3832 local seen=${#flags[*]}
3834 [[ $seen == $cnt ]] ||
3835 error "set $hsm_flag: Changelog events $seen != $cnt"
3836 [[ ${flags[$((cnt - 1))]} == $target ]] ||
3837 error "set $hsm_flag: Changelog flags are "\
3838 "${flags[$((cnt - 1))]} not $target"
3840 $LFS hsm_clear --$hsm_flag $f ||
3841 error "Cannot clear $hsm_flag on $f"
3842 flags=($(changelog_get_flags ${MDT[0]} HSM $fid))
3845 [[ $cnt == $seen ]] ||
3846 error "clear $hsm_flag: Changelog events $seen != $cnt"
3848 [[ ${flags[$((cnt - 1))]} == $target ]] ||
3849 error "clear $hsm_flag: Changelog flag is "\
3850 "${flags[$((cnt - 1))]} not $target"
3854 # test needs a running copytool
3861 for i in norelease noarchive exists archived
3863 local f=$DIR/$tdir/$tfile-$i
3864 local fid=$(copy_file /etc/passwd $f)
3865 check_flags_changes $f $fid $i 0 1
3868 f=$DIR/$tdir/$tfile---lost
3869 fid=$(copy_file /etc/passwd $f)
3870 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3871 wait_request_state $fid ARCHIVE SUCCEED
3872 check_flags_changes $f $fid lost 3 1
3876 run_test 227 "changelog when explicit setting of HSM flags"
3879 # test needs a running copytool
3882 local fid=$(make_small_sync $DIR/$tfile)
3883 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
3884 wait_request_state $fid ARCHIVE SUCCEED
3886 $LFS hsm_release $DIR/$tfile
3887 check_hsm_flags $DIR/$tfile "0x0000000d"
3889 filefrag $DIR/$tfile | grep " 1 extent found" ||
3890 error "filefrag on released file must return only one extent"
3892 # only newer versions of cp detect sparse files by stat/FIEMAP
3894 cp --sparse=auto $DIR/$tfile $DIR/$tfile.2 ||
3895 error "copying $DIR/$tfile"
3896 cmp $DIR/$tfile $DIR/$tfile.2 || error "comparing copied $DIR/$tfile"
3898 $LFS hsm_release $DIR/$tfile
3899 check_hsm_flags $DIR/$tfile "0x0000000d"
3901 mkdir -p $DIR/$tdir || error "mkdir $tdir failed"
3903 tar cf - --sparse $DIR/$tfile | tar xvf - -C $DIR/$tdir ||
3905 cmp $DIR/$tfile $DIR/$tdir/$DIR/$tfile ||
3906 error "comparing untarred $DIR/$tfile"
3908 rm -f $DIR/$tfile $DIR/$tfile.2 ||
3909 error "rm $DIR/$tfile or $DIR/$tfile.2 failed"
3912 run_test 228 "On released file, return extend to FIEMAP. For [cp,tar] --sparse"
3915 # test needs a running copytool
3919 local maxrequest=$(get_hsm_param max_requests)
3920 local rqcnt=$(($maxrequest * 3))
3924 for i in $(seq -w 1 $rqcnt); do
3926 dd if=/dev/urandom of=$DIR/$tdir/$i bs=1M count=10 conv=fsync
3928 # we do it in 2 steps, so all requests arrive at the same time
3929 for i in $(seq -w 1 $rqcnt); do
3930 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/$i
3935 while [[ $cnt != 0 || $wt != 0 ]]; do
3937 cnt=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3938 $HSM_PARAM.actions |\
3939 grep STARTED | grep -v CANCEL | wc -l")
3940 [[ $cnt -le $maxrequest ]] ||
3941 error "$cnt > $maxrequest too many started requests"
3942 wt=$(do_facet $SINGLEMDS "$LCTL get_param\
3943 $HSM_PARAM.actions |\
3944 grep WAITING | wc -l")
3945 echo "max=$maxrequest started=$cnt waiting=$wt"
3950 run_test 250 "Coordinator max request"
3953 # test needs a running copytool
3957 local f=$DIR/$tdir/$tfile
3959 fid=$(make_large_for_cancel $f)
3960 [ $? != 0 ] && skip "not enough free space" && return
3963 # to have a short test
3964 local old_to=$(get_hsm_param active_request_timeout)
3965 set_hsm_param active_request_timeout 4
3966 # to be sure the cdt will wake up frequently so
3967 # it will be able to cancel the "old" request
3968 local old_loop=$(get_hsm_param loop_period)
3969 set_hsm_param loop_period 2
3972 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3973 wait_request_state $fid ARCHIVE STARTED
3975 wait_request_state $fid ARCHIVE CANCELED
3977 set_hsm_param active_request_timeout $old_to
3978 set_hsm_param loop_period $old_loop
3982 run_test 251 "Coordinator request timeout"
3985 # the only way to test ondisk conf is to restart MDS ...
3986 echo "Stop coordinator and remove coordinator state at mount"
3989 # clean on disk conf set by default
3990 cdt_clear_mount_state
3991 cdt_check_state stopped
3993 # check cdt still off after umount/remount
3995 cdt_check_state stopped
3997 echo "Set coordinator start at mount, and start coordinator"
3998 cdt_set_mount_state enabled
4001 cdt_check_state enabled
4003 # check cdt still on after umount/remount
4005 cdt_check_state enabled
4007 # we are back to original state (cdt started at mount)
4009 run_test 300 "On disk coordinator state kept between MDT umount/mount"
4012 local ai=$(get_hsm_param default_archive_id)
4013 local new=$((ai + 1))
4015 set_hsm_param default_archive_id $new -P
4017 local res=$(get_hsm_param default_archive_id)
4020 set_hsm_param default_archive_id "" "-P -d"
4022 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
4024 run_test 301 "HSM tunnable are persistent"
4027 local ai=$(get_hsm_param default_archive_id)
4028 local new=$((ai + 1))
4033 set_hsm_param default_archive_id $new -P
4036 for mdtno in $(seq 1 $MDSCOUNT); do
4041 cdt_check_state enabled
4043 local res=$(get_hsm_param default_archive_id)
4046 set_hsm_param default_archive_id "" "-P -d"
4048 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
4050 run_test 302 "HSM tunnable are persistent when CDT is off"
4053 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4059 local dir_mdt0=$DIR/$tdir/mdt0
4060 local dir_mdt1=$DIR/$tdir/mdt1
4062 # create 1 dir per MDT
4063 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4064 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
4066 # create 1 file in each MDT
4067 local fid1=$(make_small $dir_mdt0/$tfile)
4068 local fid2=$(make_small $dir_mdt1/$tfile)
4070 # check that hsm request on mdt0 is sent to the right MDS
4071 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
4072 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4073 echo "archive successful on mdt0"
4075 # check that hsm request on mdt1 is sent to the right MDS
4076 $LFS hsm_archive $dir_mdt1/$tfile || error "lfs hsm_archive"
4077 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
4078 echo "archive successful on mdt1"
4081 # clean test files and directories
4082 rm -rf $dir_mdt0 $dir_mdt1
4084 run_test 400 "Single request is sent to the right MDT"
4087 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4093 local dir_mdt0=$DIR/$tdir/mdt0
4094 local dir_mdt1=$DIR/$tdir/mdt1
4096 # create 1 dir per MDT
4097 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4098 $LFS mkdir -i 1 $dir_mdt1 || error "lfs mkdir"
4100 # create 1 file in each MDT
4101 local fid1=$(make_small $dir_mdt0/$tfile)
4102 local fid2=$(make_small $dir_mdt1/$tfile)
4104 # check that compound requests are shunt to the rights MDTs
4105 $LFS hsm_archive $dir_mdt0/$tfile $dir_mdt1/$tfile ||
4106 error "lfs hsm_archive"
4107 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4108 echo "archive successful on mdt0"
4109 wait_request_state $fid2 ARCHIVE SUCCEED 1 &&
4110 echo "archive successful on mdt1"
4113 # clean test files and directories
4114 rm -rf $dir_mdt0 $dir_mdt1
4116 run_test 401 "Compound requests split and sent to their respective MDTs"
4118 mdc_change_state() # facet, MDT_pattern, activate|deactivate
4123 local node=$(facet_active_host $facet)
4125 for mdc in $(do_facet $facet "$LCTL dl | grep -E ${pattern}-mdc" |
4126 awk '{print $4}'); do
4127 echo "$3 $mdc on $node"
4128 do_facet $facet "$LCTL --device $mdc $state" || return 1
4133 # make sure there is no running copytool
4136 # deactivate all mdc on agent1
4137 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "deactivate"
4139 HSMTOOL_NOERROR=true copytool_setup $SINGLEAGT
4141 check_agent_unregistered "uuid" # match any agent
4143 # no expected running copytool
4144 search_copytools $agent && error "Copytool start should have failed"
4147 mdc_change_state $SINGLEAGT "$FSNAME-MDT000." "activate"
4149 run_test 402 "Copytool start fails if all MDTs are inactive"
4152 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4154 # make sure there is no running copytool
4157 local agent=$(facet_active_host $SINGLEAGT)
4158 local uuid=$(do_rpc_nodes $agent get_client_uuid | cut -d' ' -f2)
4160 # deactivate all mdc for MDT0001
4161 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
4164 # check the agent is registered on MDT0000, and not on MDT0001
4165 check_agent_registered_by_mdt $uuid 0
4166 check_agent_unregistered_by_mdt $uuid 1
4168 # check running copytool process
4169 search_copytools $agent || error "No running copytools on $agent"
4171 # reactivate all mdc for MDT0001
4172 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
4174 # make sure the copytool is now registered to all MDTs
4175 check_agent_registered $uuid
4179 run_test 403 "Copytool starts with inactive MDT and register on reconnect"
4182 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4186 # create files on both MDT0000 and MDT0001
4189 local dir_mdt0=$DIR/$tdir/mdt0
4190 $LFS mkdir -i 0 $dir_mdt0 || error "lfs mkdir"
4192 # create 1 file on mdt0
4193 local fid1=$(make_small $dir_mdt0/$tfile)
4195 # deactivate all mdc for MDT0001
4196 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "deactivate"
4198 # send an HSM request for files in MDT0000
4199 $LFS hsm_archive $dir_mdt0/$tfile || error "lfs hsm_archive"
4201 # check for completion of files in MDT0000
4202 wait_request_state $fid1 ARCHIVE SUCCEED 0 &&
4203 echo "archive successful on mdt0"
4205 # reactivate all mdc for MDT0001
4206 mdc_change_state $SINGLEAGT "$FSNAME-MDT0001" "activate"
4209 # clean test files and directories
4212 run_test 404 "Inactive MDT does not block requests for active MDTs"
4215 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return
4221 local striped_dir=$DIR/$tdir/striped_dir
4223 # create striped dir on all of MDTs
4224 $LFS mkdir -i 0 -c $MDSCOUNT $striped_dir || error "lfs mkdir"
4226 local fid1=$(make_small_sync $striped_dir/${tfile}_0)
4227 local fid2=$(make_small_sync $striped_dir/${tfile}_1)
4228 local fid3=$(make_small_sync $striped_dir/${tfile}_2)
4229 local fid4=$(make_small_sync $striped_dir/${tfile}_3)
4231 local idx1=$($LFS getstripe -M $striped_dir/${tfile}_0)
4232 local idx2=$($LFS getstripe -M $striped_dir/${tfile}_1)
4233 local idx3=$($LFS getstripe -M $striped_dir/${tfile}_2)
4234 local idx4=$($LFS getstripe -M $striped_dir/${tfile}_3)
4236 # check that compound requests are shunt to the rights MDTs
4237 $LFS hsm_archive $striped_dir/${tfile}_0 $striped_dir/${tfile}_1 \
4238 $striped_dir/${tfile}_2 $striped_dir/${tfile}_3 ||
4239 error "lfs hsm_archive"
4241 wait_request_state $fid1 ARCHIVE SUCCEED $idx1 &&
4242 echo "archive successful on $fid1"
4243 wait_request_state $fid2 ARCHIVE SUCCEED $idx2 &&
4244 echo "archive successful on $fid2"
4245 wait_request_state $fid3 ARCHIVE SUCCEED $idx3 &&
4246 echo "archive successful on $fid3"
4247 wait_request_state $fid4 ARCHIVE SUCCEED $idx4 &&
4248 echo "archive successful on $fid4"
4250 $LFS hsm_release $striped_dir/${tfile}_0 || error "lfs hsm_release 1"
4251 $LFS hsm_release $striped_dir/${tfile}_1 || error "lfs hsm_release 2"
4252 $LFS hsm_release $striped_dir/${tfile}_2 || error "lfs hsm_release 3"
4253 $LFS hsm_release $striped_dir/${tfile}_3 || error "lfs hsm_release 4"
4255 cat $striped_dir/${tfile}_0 > /dev/null || error "cat ${tfile}_0 failed"
4256 cat $striped_dir/${tfile}_1 > /dev/null || error "cat ${tfile}_1 failed"
4257 cat $striped_dir/${tfile}_2 > /dev/null || error "cat ${tfile}_2 failed"
4258 cat $striped_dir/${tfile}_3 > /dev/null || error "cat ${tfile}_3 failed"
4262 run_test 405 "archive and release under striped directory"
4266 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.92) ] &&
4267 skip "HSM migrate is not supported" && return
4269 # Stop the existing copytool
4272 test_mkdir -p $DIR/$tdir
4273 llapi_hsm_test -d $DIR/$tdir || error "One llapi HSM test failed"
4275 run_test 500 "various LLAPI HSM tests"
4280 check_and_cleanup_lustre