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: 3815 3939
15 ALWAYS_EXCEPT="$SANITY_HSM_EXCEPT 34 35 36 40"
16 # bug number for skipped test:4178 4176
17 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 200 221 223b 31a"
18 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
20 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
22 . $LUSTRE/tests/test-framework.sh
24 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
27 MULTIOP=${MULTIOP:-multiop}
28 OPENFILE=${OPENFILE:-openfile}
29 MCREATE=${MCREATE:-mcreate}
30 MOUNT_2=${MOUNT_2:-"yes"}
33 if [[ $MDSCOUNT -ge 2 ]]; then
34 skip_env "Only run with single MDT for now" && exit
37 check_and_setup_lustre
39 if [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.4.53) ]]; then
40 skip_env "Need MDS version at least 2.4.53" && exit
43 # $RUNAS_ID may get set incorrectly somewhere else
44 if [[ $UID -eq 0 && $RUNAS_ID -eq 0 ]]; then
45 skip_env "\$RUNAS_ID set to 0, but \$UID is also 0!" && exit
47 check_runas_id $RUNAS_ID $RUNAS_GID $RUNAS
52 # In order to test multiple remote HSM agents, a new facet type named "AGT" and
53 # the following associated variables are added:
55 # AGTCOUNT: number of agents
56 # AGTDEV{N}: target HSM mount point (root path of the backend)
57 # agt{N}_HOST: hostname of the agent agt{N}
58 # SINGLEAGT: facet of the single agent
60 # The number of agents is initialized as the number of remote client nodes.
61 # By default, only single copytool is started on a remote client/agent. If there
62 # was no remote client, then the copytool will be started on the local client.
68 export AGTCOUNT=${AGTCOUNT:-$((CLIENTCOUNT - 1))}
69 [[ $AGTCOUNT -gt 0 ]] || AGTCOUNT=1
71 export SHARED_DIRECTORY=${SHARED_DIRECTORY:-$TMP}
72 if [[ $CLIENTCOUNT -gt 1 ]] &&
73 ! check_shared_dir $SHARED_DIRECTORY $CLIENTS; then
74 skip_env "SHARED_DIRECTORY should be accessible"\
79 for n in $(seq $AGTCOUNT); do
80 eval export AGTDEV$n=\$\{AGTDEV$n:-"$SHARED_DIRECTORY/arc$n"\}
81 agent=CLIENT$((n + 1))
82 if [[ -z "${!agent}" ]]; then
83 [[ $CLIENTCOUNT -eq 1 ]] && agent=CLIENT1 ||
86 eval export agt${n}_HOST=\$\{agt${n}_HOST:-${!agent}\}
89 export SINGLEAGT=${SINGLEAGT:-agt1}
91 export HSMTOOL=${HSMTOOL:-"lhsmtool_posix"}
92 export HSMTOOL_VERBOSE=${HSMTOOL_VERBOSE:-""}
93 export HSMTOOL_BASE=$(basename "$HSMTOOL" | cut -f1 -d" ")
94 HSM_ARCHIVE=$(copytool_device $SINGLEAGT)
97 MDT_PARAM="mdt.$FSNAME-MDT0000"
98 HSM_PARAM="$MDT_PARAM.hsm"
100 # archive is purged at copytool setup
101 HSM_ARCHIVE_PURGE=true
104 # Get the backend root path for the given agent facet.
107 local dev=AGTDEV$(facet_number $facet)
112 # Stop copytool and unregister an existing changelog user.
116 cdt_set_sanity_policy
119 search_and_kill_copytool() {
120 local agents=${1:-$(facet_active_host $SINGLEAGT)}
122 echo "Killing existing copytools on $agents"
123 do_nodesv $agents "killall -q $HSMTOOL_BASE" || true
127 local facet=${1:-$SINGLEAGT}
128 local lustre_mntpnt=${2:-$MOUNT}
130 local hsm_root=$(copytool_device $facet)
131 local agent=$(facet_active_host $facet)
133 if [[ -z "$arc_id" ]] &&
134 do_facet $facet "pkill -CONT -x $HSMTOOL_BASE"; then
135 echo "Wakeup copytool $facet on $agent"
139 if $HSM_ARCHIVE_PURGE; then
140 echo "Purging archive on $agent"
141 do_facet $facet "rm -rf $hsm_root/*"
144 echo "Starting copytool $facet on $agent"
145 do_facet $facet "mkdir -p $hsm_root" || error "mkdir '$hsm_root' failed"
146 # bandwidth is limited to 1MB/s so the copy time is known and
147 # independent of hardware
148 local cmd="$HSMTOOL $HSMTOOL_VERBOSE --daemon --hsm-root $hsm_root"
149 [[ -z "$arc_id" ]] || cmd+=" --archive $arc_id"
150 cmd+=" --bandwidth 1 $lustre_mntpnt"
152 # Redirect the standard output and error to a log file which
153 # can be uploaded to Maloo.
154 local prefix=$TESTLOG_PREFIX
155 [[ -z "$TESTNAME" ]] || prefix=$prefix.$TESTNAME
156 local copytool_log=$prefix.copytool${arc_id}_log.$agent.log
158 do_facet $facet "$cmd < /dev/null > $copytool_log 2>&1" ||
159 error "start copytool $facet on $agent failed"
165 local agents=${1:-$(facet_active_host $SINGLEAGT)}
167 do_nodesv $agents "pkill -INT -x $HSMTOOL_BASE" || return 0
169 echo "Copytool is stopped on $agents"
173 local agents=${1:-$(facet_active_host $SINGLEAGT)}
175 do_nodesv $agents "pkill -STOP -x $HSMTOOL_BASE" || return 0
176 echo "Copytool is suspended on $agents"
179 copytool_remove_backend() {
181 local be=$(do_facet $SINGLEAGT find $HSM_ARCHIVE -name $fid)
182 echo "Remove from backend: $fid = $be"
183 do_facet $SINGLEAGT rm -f $be
187 do_facet $SINGLEAGT \
188 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
189 --import $1 $2 $MOUNT" ||
190 error "import of $1 to $2 failed"
194 local file=$HSM_ARCHIVE/$1
195 do_facet $SINGLEAGT mkdir -p $(dirname $file)
196 do_facet $SINGLEAGT dd if=/dev/urandom of=$file count=32 bs=1000000 ||
197 error "cannot create $file"
201 local file=$HSM_ARCHIVE/$2
202 do_facet $SINGLEAGT mkdir -p $(dirname $file)
203 do_facet $SINGLEAGT cp -p $1 $file || error "cannot copy $1 to $file"
207 CL_USER=$(do_facet $SINGLEMDS $LCTL --device $MDT0\
208 changelog_register -n)
209 do_facet $SINGLEMDS lctl set_param mdd.$MDT0.changelog_mask="+hsm"
210 $LFS changelog_clear $MDT0 $CL_USER 0
213 changelog_cleanup() {
214 # $LFS changelog $MDT0
215 [[ -n "$CL_USER" ]] || return 0
217 $LFS changelog_clear $MDT0 $CL_USER 0
218 do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $CL_USER
222 changelog_get_flags() {
227 $LFS changelog $mdt | awk "/$cltype/ && /t=\[$fid\]/ {print \$5}"
232 local val=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.$param)
240 if [[ "$value" != "" ]]; then
243 do_facet $SINGLEMDS $LCTL set_param $opt -n $HSM_PARAM.$param$value
250 do_facet $SINGLEMDS $LCTL set_param $MDT_PARAM.hsm_control=$cmd
251 wait_result $SINGLEMDS "$LCTL get_param -n $MDT_PARAM.hsm_control"\
252 $target 10 || error "cdt state is not $target"
255 cdt_set_sanity_policy() {
256 if [[ "$CDT_POLICY_HAD_CHANGED" ]]
259 do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=+NRA
260 do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-NBR
261 CDT_POLICY_HAD_CHANGED=
266 do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=+NRA
267 CDT_POLICY_HAD_CHANGED=true
270 cdt_clear_no_retry() {
271 do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-NRA
272 CDT_POLICY_HAD_CHANGED=true
275 cdt_set_non_blocking_restore() {
276 do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=+NBR
277 CDT_POLICY_HAD_CHANGED=true
280 cdt_clear_non_blocking_restore() {
281 do_facet $SINGLEMDS $LCTL set_param $HSM_PARAM.policy=-NBR
282 CDT_POLICY_HAD_CHANGED=true
285 cdt_clear_mount_state() {
286 do_facet $SINGLEMDS $LCTL set_param -d -P $MDT_PARAM.hsm_control
289 cdt_set_mount_state() {
290 do_facet $SINGLEMDS $LCTL set_param -P $MDT_PARAM.hsm_control=$1
295 wait_result $SINGLEMDS\
296 "$LCTL get_param -n $MDT_PARAM.hsm_control" "$target" 20 ||
297 error "cdt state is not $target"
301 set_test_state disabled disabled
305 set_test_state enabled enabled
309 set_test_state shutdown stopped
313 set_test_state purge enabled
319 cdt_set_sanity_policy
323 local client_count=$1
324 if [[ $CLIENTCOUNT -lt $client_count ]]; then
325 skip "Need $client_count or more clients, have $CLIENTCOUNT"
332 $LFS path2fid $1 | tr -d '[]'
339 if [[ $u == "user" ]]; then
340 local st=$($RUNAS $LFS hsm_state $f)
342 local st=$($LFS hsm_state $f)
346 [[ $? == 0 ]] || error "$LFS hsm_state $f failed (run as $u)"
348 st=$(echo $st | cut -f 2 -d" " | tr -d "()," )
352 get_hsm_archive_id() {
354 local st=$($LFS hsm_state $f)
355 [[ $? == 0 ]] || error "$LFS hsm_state $f failed"
357 local ar=$(echo $st | grep "archive_id" | cut -f5 -d" " |
366 local st=$(get_hsm_flags $f)
367 [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
370 check_hsm_flags_user() {
374 local st=$(get_hsm_flags $f user)
375 [[ $st == $fl ]] || error "hsm flags on $f are $st != $fl"
387 if [[ "$3" != 1 ]]; then
391 cp $1 $f || error "cannot copy $1 to $f"
392 path2fid $f || error "cannot get fid on $f"
396 local file2=${1/$DIR/$DIR2}
397 dd if=/dev/urandom of=$file2 count=2 bs=1M conv=fsync ||
398 error "cannot create $file2"
399 path2fid $1 || error "cannot get fid on $1"
402 cleanup_large_files() {
403 local ratio=$(df -P $MOUNT | tail -1 | awk '{print $5}' |
405 [ $ratio -gt 50 ] && find $MOUNT -size +10M -exec rm -f {} \;
408 make_large_for_striping() {
409 local file2=${1/$DIR/$DIR2}
410 local sz=$($LCTL get_param -n lov.*-clilov-*.stripesize | head -1)
414 dd if=/dev/urandom of=$file2 count=5 bs=$sz conv=fsync ||
415 error "cannot create $file2"
416 path2fid $1 || error "cannot get fid on $1"
419 make_large_for_progress() {
420 local file2=${1/$DIR/$DIR2}
424 # big file is large enough, so copy time is > 30s
425 # so copytool make 1 progress
426 # size is not a multiple of 1M to avoid stripe
428 dd if=/dev/urandom of=$file2 count=39 bs=1000000 conv=fsync ||
429 error "cannot create $file2"
430 path2fid $1 || error "cannot get fid on $1"
433 make_large_for_progress_aligned() {
434 local file2=${1/$DIR/$DIR2}
438 # big file is large enough, so copy time is > 30s
439 # so copytool make 1 progress
440 # size is a multiple of 1M to have stripe
442 dd if=/dev/urandom of=$file2 count=33 bs=1M conv=fsync ||
443 error "cannot create $file2"
444 path2fid $1 || error "cannot get fid on $1"
447 make_large_for_cancel() {
448 local file2=${1/$DIR/$DIR2}
452 # Copy timeout is 100s. 105MB => 105s
453 dd if=/dev/urandom of=$file2 count=103 bs=1M conv=fsync ||
454 error "cannot create $file2"
455 path2fid $1 || error "cannot get fid on $1"
461 wait_update --verbose $(facet_active_host $facet) "$@"
464 wait_request_state() {
469 local cmd="$LCTL get_param -n $HSM_PARAM.actions"
470 cmd+=" | awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
472 wait_result $SINGLEMDS "$cmd" $state 100 ||
473 error "request on $fid is not $state"
476 get_request_state() {
480 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
481 "awk '/'$fid'.*action='$request'/ {print \\\$13}' | cut -f2 -d="
484 get_request_count() {
488 do_facet $SINGLEMDS "$LCTL get_param -n $HSM_PARAM.actions |"\
489 "awk -vn=0 '/'$fid'.*action='$request'/ {n++}; END {print n}'"
495 local cmd="$LCTL get_param -n $HSM_PARAM.actions"
496 cmd+=" | egrep 'WAITING|STARTED'"
498 wait_result $SINGLEMDS "$cmd" "" $timeout ||
499 error "requests did not complete"
502 wait_for_grace_delay() {
503 local val=$(get_hsm_param grace_delay)
507 MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
508 awk '{gsub(/_UUID/,""); print $1}' | head -1)
513 # cleanup from previous bad setup
514 search_and_kill_copytool
516 # for recovery tests, coordinator needs to be started at mount
518 # the lustre conf must be without hsm on (like for sanity.sh)
519 echo "Set HSM on and start"
520 cdt_set_mount_state enabled
521 cdt_check_state enabled
523 echo "Start copytool"
526 echo "Set sanity-hsm HSM policy"
527 cdt_set_sanity_policy
529 # finished requests are quickly removed from list
530 set_hsm_param grace_delay 10
536 local f=$DIR/$tdir/$tfile
540 check_hsm_flags_user $f "0x00000000"
542 $RUNAS $LFS hsm_set --norelease $f ||
543 error "user could not change hsm flags"
544 check_hsm_flags_user $f "0x00000010"
546 $RUNAS $LFS hsm_clear --norelease $f ||
547 error "user could not clear hsm flags"
548 check_hsm_flags_user $f "0x00000000"
550 # User could not change those flags...
551 $RUNAS $LFS hsm_set --exists $f &&
552 error "user should not set this flag"
553 check_hsm_flags_user $f "0x00000000"
556 $LFS hsm_set --exists $f ||
557 error "root could not change hsm flags"
558 check_hsm_flags_user $f "0x00000001"
560 $LFS hsm_clear --exists $f ||
561 error "root could not clear hsm state"
562 check_hsm_flags_user $f "0x00000000"
565 run_test 1 "lfs hsm flags root/non-root access"
569 local f=$DIR/$tdir/$tfile
571 # New files are not dirty
572 check_hsm_flags $f "0x00000000"
574 # For test, we simulate an archived file.
575 $LFS hsm_set --exists $f || error "user could not change hsm flags"
576 check_hsm_flags $f "0x00000001"
578 # chmod do not put the file dirty
579 chmod 600 $f || error "could not chmod test file"
580 check_hsm_flags $f "0x00000001"
582 # chown do not put the file dirty
583 chown $RUNAS_ID $f || error "could not chown test file"
584 check_hsm_flags $f "0x00000001"
586 # truncate put the file dirty
587 $TRUNCATE $f 1 || error "could not truncate test file"
588 check_hsm_flags $f "0x00000003"
590 $LFS hsm_clear --dirty $f || error "could not clear hsm flags"
591 check_hsm_flags $f "0x00000001"
593 run_test 2 "Check file dirtyness when doing setattr"
599 # New files are not dirty
601 check_hsm_flags $f "0x00000000"
603 # For test, we simulate an archived file.
604 $LFS hsm_set --exists $f ||
605 error "user could not change hsm flags"
606 check_hsm_flags $f "0x00000001"
608 # Reading a file, does not set dirty
609 cat $f > /dev/null || error "could not read file"
610 check_hsm_flags $f "0x00000001"
612 # Open for write without modifying data, does not set dirty
613 openfile -f O_WRONLY $f || error "could not open test file"
614 check_hsm_flags $f "0x00000001"
616 # Append to a file sets it dirty
617 cp -p /etc/passwd $f.append || error "could not create file"
618 $LFS hsm_set --exists $f.append ||
619 error "user could not change hsm flags"
620 dd if=/etc/passwd of=$f.append bs=1 count=3\
621 conv=notrunc oflag=append status=noxfer ||
622 error "could not append to test file"
623 check_hsm_flags $f.append "0x00000003"
625 # Modify a file sets it dirty
626 cp -p /etc/passwd $f.modify || error "could not create file"
627 $LFS hsm_set --exists $f.modify ||
628 error "user could not change hsm flags"
629 dd if=/dev/zero of=$f.modify bs=1 count=3\
630 conv=notrunc status=noxfer ||
631 error "could not modify test file"
632 check_hsm_flags $f.modify "0x00000003"
634 # Open O_TRUNC sets dirty
635 cp -p /etc/passwd $f.trunc || error "could not create file"
636 $LFS hsm_set --exists $f.trunc ||
637 error "user could not change hsm flags"
638 cp /etc/group $f.trunc || error "could not override a file"
639 check_hsm_flags $f.trunc "0x00000003"
641 # Mmapped a file sets dirty
642 cp -p /etc/passwd $f.mmap || error "could not create file"
643 $LFS hsm_set --exists $f.mmap ||
644 error "user could not change hsm flags"
645 multiop $f.mmap OSMWUc || error "could not mmap a file"
646 check_hsm_flags $f.mmap "0x00000003"
648 run_test 3 "Check file dirtyness when opening for write"
652 local f=$DIR/$tdir/$tfile
653 local fid=$(make_small $f)
656 local st=$(get_request_state $fid CANCEL)
657 [[ -z "$st" ]] || error "hsm_cancel must not be registered (state=$st)"
659 run_test 4 "Useless cancel must not be registered"
662 # test needs a running copytool
666 local f=$DIR/$tdir/$tfile
667 local fid=$(copy_file /etc/passwd $f)
669 wait_request_state $fid ARCHIVE SUCCEED
671 check_hsm_flags $f "0x00000009"
675 run_test 8 "Test default archive number"
679 local f=$DIR/$tdir/$tfile
680 local fid=$(copy_file /etc/passwd $f)
681 # we do not use the default one to be sure
682 local new_an=$((HSM_ARCHIVE_NUMBER + 1))
684 copytool_setup $SINGLEAGT $MOUNT $new_an
685 $LFS hsm_archive --archive $new_an $f
686 wait_request_state $fid ARCHIVE SUCCEED
688 check_hsm_flags $f "0x00000009"
692 run_test 9 "Use of explict archive number, with dedicated copytool"
695 needclients 3 || return 0
701 copytool_cleanup $(comma_list $(agts_nodes))
703 # start all of the copytools
704 for n in $(seq $AGTCOUNT); do
708 trap "copytool_cleanup $(comma_list $(agts_nodes))" EXIT
711 for n in $(seq $AGTCOUNT); do
712 file=$DIR/$tdir/$tfile.$n
713 fid=$(make_small $file)
715 $LFS hsm_archive $file || error "could not archive file $file"
716 wait_request_state $fid ARCHIVE SUCCEED
717 check_hsm_flags $file "0x00000009"
721 copytool_cleanup $(comma_list $(agts_nodes))
723 run_test 9a "Multiple remote agents"
726 # test needs a running copytool
729 mkdir -p $DIR/$tdir/d1
730 local f=$DIR/$tdir/$tfile
731 local fid=$(copy_file /etc/hosts $f)
732 $LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f ||
733 error "hsm_archive failed"
734 wait_request_state $fid ARCHIVE SUCCEED
736 local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
737 error "fid $fid not in archive $HSM_ARCHIVE"
738 echo "Verifying content"
739 do_facet $SINGLEAGT diff $f $AFILE || error "archived file differs"
740 echo "Verifying hsm state "
741 check_hsm_flags $f "0x00000009"
743 echo "Verifying archive number is $HSM_ARCHIVE_NUMBER"
744 local st=$(get_hsm_archive_id $f)
745 [[ $st == $HSM_ARCHIVE_NUMBER ]] ||
746 error "Wrong archive number, $st != $HSM_ARCHIVE_NUMBER"
751 run_test 10a "Archive a file"
754 # test needs a running copytool
758 local f=$DIR/$tdir/$tfile
759 local fid=$(copy_file /etc/hosts $f)
760 $LFS hsm_archive $f || error "archive request failed"
761 wait_request_state $fid ARCHIVE SUCCEED
763 $LFS hsm_archive $f || error "archive of non dirty file failed"
764 local cnt=$(get_request_count $fid ARCHIVE)
765 [[ "$cnt" == "1" ]] ||
766 error "archive of non dirty file must not make a request"
770 run_test 10b "Archive of non dirty file must work without doing request"
773 # test needs a running copytool
777 local f=$DIR/$tdir/$tfile
778 local fid=$(copy_file /etc/hosts $f)
779 $LFS hsm_set --noarchive $f
780 $LFS hsm_archive $f && error "archive a noarchive file must fail"
784 run_test 10c "Check forbidden archive"
787 # test needs a running copytool
791 local f=$DIR/$tdir/$tfile
792 local fid=$(copy_file /etc/hosts $f)
793 $LFS hsm_archive $f || error "cannot archive $f"
794 wait_request_state $fid ARCHIVE SUCCEED
796 local ar=$(get_hsm_archive_id $f)
797 local dflt=$(get_hsm_param default_archive_id)
798 [[ $ar == $dflt ]] ||
799 error "archived file is not on default archive: $ar != $dflt"
803 run_test 10d "Archive a file on the default archive id"
807 copy2archive /etc/hosts $tdir/$tfile
808 local f=$DIR/$tdir/$tfile
810 import_file $tdir/$tfile $f
811 echo -n "Verifying released state: "
812 check_hsm_flags $f "0x0000000d"
814 local LSZ=$(stat -c "%s" $f)
815 local ASZ=$(do_facet $SINGLEAGT stat -c "%s" $HSM_ARCHIVE/$tdir/$tfile)
817 echo "Verifying imported size $LSZ=$ASZ"
818 [[ $LSZ -eq $ASZ ]] || error "Incorrect size $LSZ != $ASZ"
819 echo -n "Verifying released pattern: "
820 local PTRN=$($GETSTRIPE -L $f)
822 [[ $PTRN == 80000001 ]] || error "Is not released"
823 local fid=$(path2fid $f)
824 echo "Verifying new fid $fid in archive"
826 local AFILE=$(do_facet $SINGLEAGT ls $HSM_ARCHIVE'/*/*/*/*/*/*/'$fid) ||
827 error "fid $fid not in archive $HSM_ARCHIVE"
829 run_test 11 "Import a file"
832 # test needs a running copytool
836 copy2archive /etc/hosts $tdir/$tfile
838 local f=$DIR/$tdir/$tfile
839 import_file $tdir/$tfile $f
840 local f=$DIR2/$tdir/$tfile
841 echo "Verifying released state: "
842 check_hsm_flags $f "0x0000000d"
844 local fid=$(path2fid $f)
846 wait_request_state $fid RESTORE SUCCEED
848 echo "Verifying file state: "
849 check_hsm_flags $f "0x00000009"
851 do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
853 [[ $? -eq 0 ]] || error "Restored file differs"
857 run_test 12a "Restore an imported file explicitly"
860 # test needs a running copytool
864 copy2archive /etc/hosts $tdir/$tfile
866 local f=$DIR/$tdir/$tfile
867 import_file $tdir/$tfile $f
868 echo "Verifying released state: "
869 check_hsm_flags $f "0x0000000d"
871 cat $f > /dev/null || error "File read failed"
873 echo "Verifying file state after restore: "
874 check_hsm_flags $f "0x00000009"
876 do_facet $SINGLEAGT diff -q $HSM_ARCHIVE/$tdir/$tfile $f
878 [[ $? -eq 0 ]] || error "Restored file differs"
882 run_test 12b "Restore an imported file implicitly"
885 [ "$OSTCOUNT" -lt "2" ] && skip_env "skipping 2-stripe test" && return
887 # test needs a running copytool
891 local f=$DIR/$tdir/$tfile
892 $LFS setstripe -c 2 $f
893 local fid=$(make_large_for_striping $f)
894 local FILE_CRC=$(md5sum $f)
896 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
897 wait_request_state $fid ARCHIVE SUCCEED
898 $LFS hsm_release $f || error "release $f failed"
900 echo "$FILE_CRC" | md5sum -c
902 [[ $? -eq 0 ]] || error "Restored file differs"
906 run_test 12c "Restore a file with stripe of 2"
909 # test needs a running copytool
914 local f=$DIR/$tdir/$tfile
915 local fid=$(copy_file /etc/hosts $f)
916 $LFS hsm_restore $f || error "restore of non archived file failed"
917 local cnt=$(get_request_count $fid RESTORE)
918 [[ "$cnt" == "0" ]] ||
919 error "restore non archived must not make a request"
920 $LFS hsm_archive $f ||
921 error "archive request failed"
922 wait_request_state $fid ARCHIVE SUCCEED
923 $LFS hsm_restore $f ||
924 error "restore of non released file failed"
925 local cnt=$(get_request_count $fid RESTORE)
926 [[ "$cnt" == "0" ]] ||
927 error "restore a non dirty file must not make a request"
931 run_test 12d "Restore of a non archived, non released file must work"\
932 " without doing request"
935 # test needs a running copytool
938 mkdir -p $DIR/$tdir $HSM_ARCHIVE/$tdir
939 local f=$DIR/$tdir/$tfile
940 local fid=$(copy_file /etc/hosts $f)
941 $LFS hsm_archive $f || error "archive request failed"
942 wait_request_state $fid ARCHIVE SUCCEED
949 $LFS hsm_restore $f && error "restore a dirty file must fail"
953 run_test 12e "Check forbidden restore"
956 # test needs a running copytool
960 local f=$DIR/$tdir/$tfile
961 local fid=$(copy_file /etc/hosts $f)
963 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
964 wait_request_state $fid ARCHIVE SUCCEED
965 $LFS hsm_release $f || error "release of $f failed"
967 wait_request_state $fid RESTORE SUCCEED
969 echo -n "Verifying file state: "
970 check_hsm_flags $f "0x00000009"
972 diff -q /etc/hosts $f
974 [[ $? -eq 0 ]] || error "Restored file differs"
978 run_test 12f "Restore a released file explicitly"
981 # test needs a running copytool
985 local f=$DIR/$tdir/$tfile
986 local fid=$(copy_file /etc/hosts $f)
988 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
989 wait_request_state $fid ARCHIVE SUCCEED
990 $LFS hsm_release $f || error "release of $f failed"
992 diff -q /etc/hosts $f
995 # we check we had a restore done
996 wait_request_state $fid RESTORE SUCCEED
998 [[ $st -eq 0 ]] || error "Restored file differs"
1002 run_test 12g "Restore a released file implicitly"
1005 needclients 2 || return 0
1007 # test needs a running copytool
1011 local f=$DIR/$tdir/$tfile
1012 local fid=$(copy_file /etc/hosts $f)
1014 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1015 wait_request_state $fid ARCHIVE SUCCEED
1016 $LFS hsm_release $f || error "release of $f failed"
1018 do_node $CLIENT2 diff -q /etc/hosts $f
1021 # we check we had a restore done
1022 wait_request_state $fid RESTORE SUCCEED
1024 [[ $st -eq 0 ]] || error "Restored file differs"
1028 run_test 12h "Restore a released file implicitly from a second node"
1031 # test needs a running copytool
1035 local f=$DIR/$tdir/$tfile
1036 local fid=$(copy_file /etc/passwd $f)
1037 $LFS hsm_archive $f || error "archive of $f failed"
1038 wait_request_state $fid ARCHIVE SUCCEED
1040 $LFS hsm_release $f || error "release of $f failed"
1044 [[ $? -eq 0 ]] || error "Restored file differs"
1048 run_test 12m "Archive/release/implicit restore"
1051 # test needs a running copytool
1055 copy2archive /etc/hosts $tdir/$tfile
1057 local f=$DIR/$tdir/$tfile
1058 import_file $tdir/$tfile $f
1060 do_facet $SINGLEAGT cmp /etc/hosts $f ||
1061 error "Restored file differs"
1063 $LFS hsm_release $f || error "release of $f failed"
1067 run_test 12n "Import/implicit restore/release"
1070 # test needs a running copytool
1073 local ARC_SUBDIR="import.orig"
1077 # populate directory to be imported
1078 for d in $(seq 1 10); do
1079 local CURR_DIR="$HSM_ARCHIVE/$ARC_SUBDIR/dir.$d"
1080 do_facet $SINGLEAGT mkdir -p "$CURR_DIR"
1081 for f in $(seq 1 10); do
1082 CURR_FILE="$CURR_DIR/$tfile.$f"
1083 # write file-specific data
1084 do_facet $SINGLEAGT \
1085 "echo d=$d, f=$f, dir=$CURR_DIR, "\
1086 "file=$CURR_FILE > $CURR_FILE"
1090 import_file "$ARC_SUBDIR" $DIR/$tdir
1091 # diff lustre content and origin (triggers file restoration)
1092 # there must be 10x10 identical files, and no difference
1093 local cnt_ok=$(do_facet $SINGLEAGT diff -rs $HSM_ARCHIVE/$ARC_SUBDIR \
1094 $DIR/$tdir/$ARC_SUBDIR | grep identical | wc -l)
1095 local cnt_diff=$(do_facet $SINGLEAGT diff -r $HSM_ARCHIVE/$ARC_SUBDIR \
1096 $DIR/$tdir/$ARC_SUBDIR | wc -l)
1098 [ $cnt_diff -eq 0 ] ||
1099 error "$cnt_diff imported files differ from read data"
1100 [ $cnt_ok -eq 100 ] ||
1101 error "not enough identical files ($cnt_ok != 100)"
1105 run_test 13 "Recursively import and restore a directory"
1108 # test needs a running copytool
1113 local f=$DIR/$tdir/$tfile
1114 local fid=$(make_small $f)
1115 local sum=$(md5sum $f | awk '{print $1}')
1116 $LFS hsm_archive $f || error "could not archive file"
1117 wait_request_state $fid ARCHIVE SUCCEED
1121 # create released file (simulate llapi_hsm_import call)
1123 local fid2=$(path2fid $f)
1124 $LFS hsm_set --archived --exists $f || error "could not force hsm flags"
1125 $LFS hsm_release $f || error "could not release file"
1127 # rebind the archive to the newly created file
1128 echo "rebind $fid to $fid2"
1130 do_facet $SINGLEAGT \
1131 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
1132 --rebind $fid $fid2 $DIR" || error "could not rebind file"
1134 # restore file and compare md5sum
1135 local sum2=$(md5sum $f | awk '{print $1}')
1137 [[ $sum == $sum2 ]] || error "md5sum mismatch after restore"
1141 run_test 14 "Rebind archived file to a new fid"
1144 # test needs a running copytool
1149 local f=$DIR/$tdir/$tfile
1151 local tmpfile=$SHARED_DIRECTORY/tmp.$$
1155 for i in $(seq 1 $count); do
1156 fids[$i]=$(make_small $f.$i)
1157 sums[$i]=$(md5sum $f.$i | awk '{print $1}')
1158 $LFS hsm_archive $f.$i || error "could not archive file"
1160 wait_all_done $(($count*60))
1164 for i in $(seq 1 $count); do
1167 local fid2=$(path2fid $f.$i)
1168 # add the rebind operation to the list
1169 echo ${fids[$i]} $fid2 >> $tmpfile
1171 # set it released (simulate llapi_hsm_import call)
1172 $LFS hsm_set --archived --exists $f.$i ||
1173 error "could not force hsm flags"
1174 $LFS hsm_release $f.$i || error "could not release file"
1176 nl=$(wc -l < $tmpfile)
1177 [[ $nl == $count ]] || error "$nl files in list, $count expected"
1179 echo "rebind list of files"
1180 do_facet $SINGLEAGT \
1181 "$HSMTOOL --archive $HSM_ARCHIVE_NUMBER --hsm-root $HSM_ARCHIVE\
1182 --rebind $tmpfile $DIR" || error "could not rebind file list"
1184 # restore files and compare md5sum
1185 for i in $(seq 1 $count); do
1186 local sum2=$(md5sum $f.$i | awk '{print $1}')
1187 [[ $sum2 == ${sums[$i]} ]] ||
1188 error "md5sum mismatch after restore ($sum2 != ${sums[$i]})"
1194 run_test 15 "Rebind a list of files"
1197 # test needs a running copytool
1201 # create a known size file so we can verify transfer speed
1204 dd if=/dev/zero of=$ref bs=1M count=20
1207 local f=$DIR/$tdir/$tfile
1208 local fid=$(copy_file $ref $f)
1210 local start=$(date +%s)
1212 wait_request_state $fid ARCHIVE SUCCEED
1213 local end=$(date +%s)
1214 local duration=$((end - start))
1216 [[ $duration -ge $goal ]] ||
1217 error "Transfer is too fast $duration < $goal"
1221 run_test 16 "Test CT bandwith control option"
1226 local f=$DIR/$tdir/$tfile
1227 touch $f || error "touch $f failed"
1229 # Could not release a non-archived file
1230 $LFS hsm_release $f && error "release should not succeed"
1232 # For following tests, we must test them with HS_ARCHIVED set
1233 $LFS hsm_set --exists --archived $f || error "could not add flag"
1235 # Could not release a file if no-release is set
1236 $LFS hsm_set --norelease $f || error "could not add flag"
1237 $LFS hsm_release $f && error "release should not succeed"
1238 $LFS hsm_clear --norelease $f || error "could not remove flag"
1240 # Could not release a file if lost
1241 $LFS hsm_set --lost $f || error "could not add flag"
1242 $LFS hsm_release $f && error "release should not succeed"
1243 $LFS hsm_clear --lost $f || error "could not remove flag"
1245 # Could not release a file if dirty
1246 $LFS hsm_set --dirty $f || error "could not add flag"
1247 $LFS hsm_release $f && error "release should not succeed"
1248 $LFS hsm_clear --dirty $f || error "could not remove flag"
1250 run_test 20 "Release is not permitted"
1253 # test needs a running copytool
1257 local f=$DIR/$tdir/test_release
1259 # Create a file and check its states
1260 local fid=$(make_small $f)
1261 check_hsm_flags $f "0x00000000"
1263 $LFS hsm_archive $f || error "could not archive file"
1264 wait_request_state $fid ARCHIVE SUCCEED
1266 [ $(stat -c "%b" $f) -ne "1" ] || error "wrong block number"
1267 local sz=$(stat -c "%s" $f)
1268 [ $sz -ne "0" ] || error "file size should not be zero"
1270 # Release and check states
1271 $LFS hsm_release $f || error "could not release file"
1272 check_hsm_flags $f "0x0000000d"
1274 [ $(stat -c "%b" $f) -eq "1" ] || error "wrong block number"
1275 [ $(stat -c "%s" $f) -eq $sz ] || error "wrong file size"
1277 # Check we can release an file without stripe info
1281 check_hsm_flags $f "0x00000000"
1282 $LFS hsm_archive $f || error "could not archive file"
1283 wait_request_state $fid ARCHIVE SUCCEED
1285 # Release and check states
1286 $LFS hsm_release $f || error "could not release file"
1287 check_hsm_flags $f "0x0000000d"
1289 # Release again a file that is already released is OK
1290 $LFS hsm_release $f || fail "second release should succeed"
1291 check_hsm_flags $f "0x0000000d"
1295 run_test 21 "Simple release tests"
1298 # test needs a running copytool
1303 local f=$DIR/$tdir/test_release
1304 local swap=$DIR/$tdir/test_swap
1306 # Create a file and check its states
1307 local fid=$(make_small $f)
1308 check_hsm_flags $f "0x00000000"
1310 $LFS hsm_archive $f || error "could not archive file"
1311 wait_request_state $fid ARCHIVE SUCCEED
1313 # Release and check states
1314 $LFS hsm_release $f || error "could not release file"
1315 check_hsm_flags $f "0x0000000d"
1318 $LFS swap_layouts $swap $f && error "swap_layouts should failed"
1323 run_test 22 "Could not swap a release file"
1326 # test needs a running copytool
1331 local f=$DIR/$tdir/test_mtime
1333 # Create a file and check its states
1334 local fid=$(make_small $f)
1335 check_hsm_flags $f "0x00000000"
1337 $LFS hsm_archive $f || error "could not archive file"
1338 wait_request_state $fid ARCHIVE SUCCEED
1340 # Set modification time in the past
1341 touch -m -a -d @978261179 $f
1343 # Release and check states
1344 $LFS hsm_release $f || error "could not release file"
1345 check_hsm_flags $f "0x0000000d"
1347 local MTIME=$(stat -c "%Y" $f)
1348 local ATIME=$(stat -c "%X" $f)
1349 [ $MTIME -eq "978261179" ] || fail "bad mtime: $MTIME"
1350 [ $ATIME -eq "978261179" ] || fail "bad atime: $ATIME"
1354 run_test 23 "Release does not change a/mtime (utime)"
1357 local file=$DIR/$tdir/$tfile
1366 # test needs a running copytool
1371 fid=$(make_small $file)
1373 # Create a file and check its states
1374 check_hsm_flags $file "0x00000000"
1376 # Ensure atime is less than mtime and ctime.
1380 atime0=$(stat -c "%X" $file)
1381 mtime0=$(stat -c "%Y" $file)
1382 ctime0=$(stat -c "%Z" $file)
1384 [ $atime0 -lt $mtime0 ] ||
1385 error "atime $atime0 is not less than mtime $mtime0"
1387 [ $atime0 -lt $ctime0 ] ||
1388 error "atime $atime0 is not less than ctime $ctime0"
1390 # Archive should not change any timestamps.
1391 $LFS hsm_archive $file || error "cannot archive '$file'"
1392 wait_request_state $fid ARCHIVE SUCCEED
1394 atime1=$(stat -c "%X" $file)
1395 mtime1=$(stat -c "%Y" $file)
1396 ctime1=$(stat -c "%Z" $file)
1398 [ $atime0 -eq $atime1 ] ||
1399 error "archive changed atime from $atime0 to $atime1"
1401 [ $mtime0 -eq $mtime1 ] ||
1402 error "archive changed mtime from $mtime0 to $mtime1"
1404 [ $ctime0 -eq $ctime1 ] ||
1405 error "archive changed ctime from $ctime0 to $ctime1"
1407 # Release should not change any timestamps.
1408 $LFS hsm_release $file || error "cannot release '$file'"
1409 check_hsm_flags $file "0x0000000d"
1411 atime1=$(stat -c "%X" $file)
1412 mtime1=$(stat -c "%Y" $file)
1413 ctime1=$(stat -c "%Z" $file)
1415 [ $atime0 -eq $atime1 ] ||
1416 error "release changed atime from $atime0 to $atime1"
1418 [ $mtime0 -eq $mtime1 ] ||
1419 error "release changed mtime from $mtime0 to $mtime1"
1421 [ $ctime0 -eq $ctime1 ] ||
1422 error "release changed ctime from $ctime0 to $ctime1"
1424 # Restore should not change atime or mtime and should not
1426 $LFS hsm_restore $file
1427 wait_request_state $fid RESTORE SUCCEED
1429 atime1=$(stat -c "%X" $file)
1430 mtime1=$(stat -c "%Y" $file)
1431 ctime1=$(stat -c "%Z" $file)
1433 [ $atime0 -eq $atime1 ] ||
1434 error "restore changed atime from $atime0 to $atime1"
1436 [ $mtime0 -eq $mtime1 ] ||
1437 error "restore changed mtime from $mtime0 to $mtime1"
1439 [ $ctime0 -le $ctime1 ] ||
1440 error "restore changed ctime from $ctime0 to $ctime1"
1444 # Once more, after unmount and mount.
1445 umount_client $MOUNT || error "cannot unmount '$MOUNT'"
1446 mount_client $MOUNT || error "cannot mount '$MOUNT'"
1448 atime1=$(stat -c "%X" $file)
1449 mtime1=$(stat -c "%Y" $file)
1450 ctime1=$(stat -c "%Z" $file)
1452 [ $atime0 -eq $atime1 ] ||
1453 error "remount changed atime from $atime0 to $atime1"
1455 [ $mtime0 -eq $mtime1 ] ||
1456 error "remount changed mtime from $mtime0 to $mtime1"
1458 [ $ctime0 -le $ctime1 ] ||
1459 error "remount changed ctime from $ctime0 to $ctime1"
1461 run_test 24a "Archive, release, and restore does not change a/mtime (i/o)"
1464 local file=$DIR/$tdir/$tfile
1470 # Test needs a running copytool.
1474 # Check that root can do HSM actions on a ordinary user's file.
1476 fid=$(make_small $file)
1477 sum0=$(md5sum $file)
1479 chown $RUNAS_ID:$RUNAS_GID $file ||
1480 error "cannot chown '$file' to '$RUNAS_ID'"
1482 chmod ugo-w $DIR/$tdir ||
1483 error "cannot chmod '$DIR/$tdir'"
1485 $LFS hsm_archive $file
1486 wait_request_state $fid ARCHIVE SUCCEED
1488 $LFS hsm_release $file
1489 check_hsm_flags $file "0x0000000d"
1491 $LFS hsm_restore $file
1492 wait_request_state $fid RESTORE SUCCEED
1494 # Check that ordinary user can get HSM state.
1495 $RUNAS $LFS hsm_state $file ||
1496 error "user '$RUNAS_ID' cannot get HSM state of '$file'"
1498 $LFS hsm_release $file
1499 check_hsm_flags $file "0x0000000d"
1501 # Check that ordinary user can accessed released file.
1502 sum1=$($RUNAS md5sum $file) ||
1503 error "user '$RUNAS_ID' cannot read '$file'"
1505 [ "$sum0" == "$sum1" ] ||
1506 error "md5sum mismatch for '$file'"
1510 run_test 24b "root can archive, release, and restore user files"
1512 cleanup_test_24c() {
1514 set_hsm_param user_request_mask RESTORE
1515 set_hsm_param group_request_mask RESTORE
1516 set_hsm_param other_request_mask RESTORE
1520 local file=$DIR/$tdir/$tfile
1521 local action=archive
1526 # test needs a running copytool
1531 # Save the default masks and check that cleanup_24c will
1532 # restore the request masks correctly.
1533 user_save=$(get_hsm_param user_request_mask)
1534 group_save=$(get_hsm_param group_request_mask)
1535 other_save=$(get_hsm_param other_request_mask)
1537 [ "$user_save" == RESTORE ] ||
1538 error "user_request_mask is '$user_save' expected 'RESTORE'"
1539 [ "$group_save" == RESTORE ] ||
1540 error "group_request_mask is '$group_save' expected 'RESTORE'"
1541 [ "$other_save" == RESTORE ] ||
1542 error "other_request_mask is '$other_save' expected 'RESTORE'"
1544 trap cleanup_test_24c EXIT
1549 chown $RUNAS_ID:nobody $file ||
1550 error "cannot chown '$file' to '$RUNAS_ID:nobody'"
1552 set_hsm_param user_request_mask ""
1553 $RUNAS $LFS hsm_$action $file &&
1554 error "$action by user should fail"
1556 set_hsm_param user_request_mask $action
1557 $RUNAS $LFS hsm_$action $file ||
1558 error "$action by user should succeed"
1563 chown nobody:$RUNAS_GID $file ||
1564 error "cannot chown '$file' to 'nobody:$RUNAS_GID'"
1566 set_hsm_param group_request_mask ""
1567 $RUNAS $LFS hsm_$action $file &&
1568 error "$action by group should fail"
1570 set_hsm_param group_request_mask $action
1571 $RUNAS $LFS hsm_$action $file ||
1572 error "$action by group should succeed"
1577 chown nobody:nobody $file ||
1578 error "cannot chown '$file' to 'nobody:nobody'"
1580 set_hsm_param other_request_mask ""
1581 $RUNAS $LFS hsm_$action $file &&
1582 error "$action by other should fail"
1584 set_hsm_param other_request_mask $action
1585 $RUNAS $LFS hsm_$action $file ||
1586 error "$action by other should succeed"
1591 run_test 24c "check that user,group,other request masks work"
1593 cleanup_test_24d() {
1595 mount -o remount,rw $MOUNT2
1599 local file1=$DIR/$tdir/$tfile
1600 local file2=$DIR2/$tdir/$tfile
1608 fid1=$(make_small $file1)
1610 trap cleanup_test_24d EXIT
1612 mount -o remount,ro $MOUNT2
1614 fid2=$(path2fid $file2)
1615 [ "$fid1" == "$fid2" ] ||
1616 error "FID mismatch '$fid1' != '$fid2'"
1618 $LFS hsm_archive $file2 &&
1619 error "archive should fail on read-only mount"
1620 check_hsm_flags $file1 "0x00000000"
1622 $LFS hsm_archive $file1
1623 wait_request_state $fid1 ARCHIVE SUCCEED
1625 $LFS hsm_release $file1
1626 $LFS hsm_restore $file2
1627 wait_request_state $fid1 RESTORE SUCCEED
1629 $LFS hsm_release $file1 || error "cannot release '$file1'"
1630 dd if=$file2 of=/dev/null bs=1M || "cannot read '$file2'"
1632 $LFS hsm_release $file2 &&
1633 error "release should fail on read-only mount"
1638 run_test 24d "check that read-only mounts are respected"
1641 # test needs a running copytool
1645 copy2archive /etc/hosts $tdir/$tfile
1647 local f=$DIR/$tdir/$tfile
1649 import_file $tdir/$tfile $f
1651 $LFS hsm_set --lost $f
1656 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
1660 run_test 25a "Restore lost file (HS_LOST flag) from import"\
1661 " (Operation not permitted)"
1664 # test needs a running copytool
1669 local f=$DIR/$tdir/$tfile
1670 local fid=$(copy_file /etc/passwd $f)
1672 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1673 wait_request_state $fid ARCHIVE SUCCEED
1676 $LFS hsm_set --lost $f
1680 [[ $st == 1 ]] || error "lost file access should failed (returns $st)"
1684 run_test 25b "Restore lost file (HS_LOST flag) after release"\
1685 " (Operation not permitted)"
1688 # test needs a running copytool
1692 local f=$DIR/$tdir/$tfile
1693 local fid=$(make_large_for_progress $f)
1694 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1695 wait_request_state $fid ARCHIVE SUCCEED
1698 wait_request_state $fid REMOVE SUCCEED
1700 check_hsm_flags $f "0x00000000"
1704 run_test 26 "Remove the archive of a valid file"
1707 # test needs a running copytool
1711 make_archive $tdir/$tfile
1712 local f=$DIR/$tdir/$tfile
1713 import_file $tdir/$tfile $f
1714 local fid=$(path2fid $f)
1718 [[ $? != 0 ]] || error "Remove of a released file should fail"
1722 run_test 27a "Remove the archive of an imported file (Operation not permitted)"
1725 # test needs a running copytool
1729 local f=$DIR/$tdir/$tfile
1730 local fid=$(make_large_for_progress $f)
1731 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1732 wait_request_state $fid ARCHIVE SUCCEED
1737 [[ $? != 0 ]] || error "Remove of a released file should fail"
1741 run_test 27b "Remove the archive of a relased file (Operation not permitted)"
1744 # test needs a running copytool
1748 local f=$DIR/$tdir/$tfile
1749 local fid=$(make_large_for_progress $f)
1750 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1751 wait_request_state $fid ARCHIVE SUCCEED
1760 wait_request_state $fid REMOVE SUCCEED
1764 run_test 28 "Concurrent archive/file remove"
1767 # restore at exec cannot work on agent node (because of Linux kernel
1768 # protection of executables)
1769 needclients 2 || return 0
1771 # test needs a running copytool
1775 copy2archive /bin/true $tdir/$tfile
1777 local f=$DIR/$tdir/true
1778 import_file $tdir/$tfile $f
1780 local fid=$(path2fid $f)
1782 # set no retry action mode
1788 # remove no try action mode
1792 [[ $st == 0 ]] || error "Failed to exec a released file"
1796 run_test 30a "Restore at exec (import case)"
1799 # restore at exec cannot work on agent node (because of Linux kernel
1800 # protection of executables)
1801 needclients 2 || return 0
1803 # test needs a running copytool
1807 local f=$DIR/$tdir/true
1808 local fid=$(copy_file /bin/true $f)
1810 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1811 wait_request_state $fid ARCHIVE SUCCEED
1814 # set no retry action mode
1820 # remove no try action mode
1824 [[ $st == 0 ]] || error "Failed to exec a released file"
1828 run_test 30b "Restore at exec (release case)"
1831 needclients 2 || return 0
1833 # test needs a running copytool
1837 local f=$DIR/$tdir/SLEEP
1838 local fid=$(copy_file /bin/sleep $f)
1840 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1841 wait_request_state $fid ARCHIVE SUCCEED
1843 check_hsm_flags $f "0x0000000d"
1844 # set no retry action mode
1846 do_node $CLIENT2 "$f 10" &
1850 [[ $? == 0 ]] && error "Update during exec of released file must fail"
1852 [[ $? == 0 ]] || error "Execution failed during run"
1854 [[ $? == 0 ]] || error "Binary overwritten during exec"
1857 # remove no try action mode
1859 check_hsm_flags $f "0x00000009"
1863 run_test 30c "Update during exec of released file must fail"
1865 restore_and_check_size() {
1868 local s=$(stat -c "%s" $f)
1870 local st=$(get_hsm_flags $f)
1874 while [[ "$st" != "0x00000009" && $cpt -le 10 ]]
1876 n=$(stat -c "%s" $f)
1877 # we echo in both cases to show stat is not
1879 if [[ $n != $s ]]; then
1880 echo "size seen is $n != $s"
1883 echo "size seen is right: $n == $s"
1885 st=$(get_hsm_flags $f)
1889 if [[ $cpt -lt 10 ]]; then
1890 echo " restore is too long"
1894 wait_request_state $fid RESTORE SUCCEED
1899 # test needs a running copytool
1904 make_archive $tdir/$tfile
1905 local f=$DIR/$tdir/$tfile
1906 import_file $tdir/$tfile $f
1907 local fid=$($LFS path2fid $f)
1908 HSM_ARCHIVE_PURGE=false copytool_setup
1910 restore_and_check_size $f $fid
1913 [[ $err -eq 0 ]] || error "File size changed during restore"
1917 run_test 31a "Import a large file and check size during restore"
1921 # test needs a running copytool
1926 local f=$DIR/$tdir/$tfile
1927 local fid=$(make_large_for_progress $f)
1928 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1929 wait_request_state $fid ARCHIVE SUCCEED
1932 restore_and_check_size $f $fid
1935 [[ $err -eq 0 ]] || error "File size changed during restore"
1939 run_test 31b "Restore a large unaligned file and check size during restore"
1942 # test needs a running copytool
1947 local f=$DIR/$tdir/$tfile
1948 local fid=$(make_large_for_progress_aligned $f)
1949 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1950 wait_request_state $fid ARCHIVE SUCCEED
1953 restore_and_check_size $f $fid
1956 [[ $err -eq 0 ]] || error "File size changed during restore"
1960 run_test 31c "Restore a large aligned file and check size during restore"
1963 # test needs a running copytool
1968 local f=$DIR/$tdir/$tfile
1969 local fid=$(make_large_for_progress $f)
1970 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
1971 wait_request_state $fid ARCHIVE SUCCEED
1974 md5sum $f >/dev/null &
1976 wait_request_state $fid RESTORE STARTED
1981 # Check restore trigger process was killed
1982 local killed=$(ps -o pid,comm hp $pid >/dev/null)
1986 wait_request_state $fid RESTORE CANCELED
1987 wait_request_state $fid CANCEL SUCCEED
1990 error "Cannot kill process waiting for restore ($killed)"
1994 run_test 33 "Kill a restore waiting process"
1997 # test needs a running copytool
2002 local f=$DIR/$tdir/$tfile
2003 local fid=$(make_large_for_progress $f)
2004 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2005 wait_request_state $fid ARCHIVE SUCCEED
2008 md5sum $f >/dev/null &
2010 wait_request_state $fid RESTORE STARTED
2012 rm $f || error "rm $f failed"
2013 # rm must not block during restore
2014 wait_request_state $fid RESTORE STARTED
2016 wait_request_state $fid RESTORE SUCCEED
2017 # check md5sum pgm finished
2018 local there=$(ps -o pid,comm hp $pid >/dev/null)
2019 [[ -z $there ]] || error "Restore initiator does not exit"
2021 local rc=$(wait $pid)
2022 [[ $rc -eq 0 ]] || error "Restore initiator failed with $rc"
2026 run_test 34 "Remove file during restore"
2029 # test needs a running copytool
2034 local f=$DIR/$tdir/$tfile
2035 local f1=$DIR/$tdir/$tfile-1
2036 local fid=$(make_large_for_progress $f)
2037 local fid1=$(copy_file /etc/passwd $f1)
2038 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2039 wait_request_state $fid ARCHIVE SUCCEED
2042 md5sum $f >/dev/null &
2044 wait_request_state $fid RESTORE STARTED
2046 mv $f1 $f || error "mv $f1 $f failed"
2047 # mv must not block during restore
2048 wait_request_state $fid RESTORE STARTED
2050 wait_request_state $fid RESTORE SUCCEED
2051 # check md5sum pgm finished
2052 local there=$(ps -o pid,comm hp $pid >/dev/null)
2053 [[ -z $there ]] || error "Restore initiator does not exit"
2055 local rc=$(wait $pid)
2056 [[ $rc -eq 0 ]] || error "Restore initiator failed with $rc"
2059 [[ $fid2 == $fid1 ]] || error "Wrong fid after mv $fid2 != $fid1"
2063 run_test 35 "Overwrite file during restore"
2066 # test needs a running copytool
2071 local f=$DIR/$tdir/$tfile
2072 local fid=$(make_large_for_progress $f)
2073 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2074 wait_request_state $fid ARCHIVE SUCCEED
2077 md5sum $f >/dev/null &
2079 wait_request_state $fid RESTORE STARTED
2082 # rm must not block during restore
2083 wait_request_state $fid RESTORE STARTED
2085 wait_request_state $fid RESTORE SUCCEED
2086 # check md5sum pgm finished
2087 local there=$(ps -o pid,comm hp $pid >/dev/null)
2089 error "Restore initiator does not exit"
2091 local rc=$(wait $pid)
2093 error "Restore initiator failed with $rc"
2097 run_test 36 "Move file during restore"
2104 for n in $(seq 1 $count); do
2105 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $prefix.$n
2107 echo "$count archive requests submitted"
2111 local stream_count=4
2112 local file_count=100
2114 local f=$DIR/$tdir/$tfile
2119 for i in $(seq 1 $file_count); do
2120 for p in $(seq 1 $stream_count); do
2121 fid=$(copy_file /etc/hosts $f.$p.$i)
2125 # to be sure wait_all_done will not be mislead by previous tests
2127 wait_for_grace_delay
2129 # start archive streams in background (archive files in parallel)
2130 for p in $(seq 1 $stream_count); do
2131 multi_archive $f.$p $file_count &
2134 echo -n "Wait for all requests being enqueued..."
2140 run_test 40 "Parallel archive requests"
2143 # test needs a running copytool
2147 local f=$DIR/$tdir/$tfile
2148 local fid=$(copy_file /etc/motd $f 1)
2150 $LFS hsm_archive $f || error "could not archive file"
2151 wait_request_state $fid ARCHIVE SUCCEED
2152 check_hsm_flags $f "0x00000009"
2154 multiop_bg_pause $f O_c || error "multiop failed"
2158 client_up || client_up || true
2160 kill -USR1 $MULTIPID
2161 wait $MULTIPID || error "multiop close failed"
2163 check_hsm_flags $f "0x0000000b"
2167 run_test 52 "Opened for write file on an evicted client should be set dirty"
2170 # test needs a running copytool
2174 local f=$DIR/$tdir/$tfile
2175 local fid=$(copy_file /etc/motd $f 1)
2177 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2178 error "could not archive file"
2179 wait_request_state $fid ARCHIVE SUCCEED
2180 check_hsm_flags $f "0x00000009"
2182 multiop_bg_pause $f o_c || error "multiop failed"
2186 client_up || client_up || true
2188 kill -USR1 $MULTIPID
2189 wait $MULTIPID || error "multiop close failed"
2191 check_hsm_flags $f "0x00000009"
2195 run_test 53 "Opened for read file on an evicted client should not be set dirty"
2198 # test needs a running copytool
2202 local f=$DIR/$tdir/$tfile
2203 local fid=$(make_small $f)
2205 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2206 error "could not archive file"
2207 wait_request_state $fid ARCHIVE STARTED
2209 check_hsm_flags $f "0x00000001"
2211 # Avoid coordinator resending this request as soon it has failed.
2216 wait_request_state $fid ARCHIVE FAILED
2218 check_hsm_flags $f "0x00000003"
2223 run_test 54 "Write during an archive cancels it"
2226 # test needs a running copytool
2230 local f=$DIR/$tdir/$tfile
2231 local fid=$(make_small $f)
2233 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2234 error "could not archive file"
2235 wait_request_state $fid ARCHIVE STARTED
2237 check_hsm_flags $f "0x00000001"
2239 # Avoid coordinator resending this request as soon it has failed.
2242 $TRUNCATE $f 1024 || error "truncate failed"
2244 wait_request_state $fid ARCHIVE FAILED
2246 check_hsm_flags $f "0x00000003"
2251 run_test 55 "Truncate during an archive cancels it"
2254 # test needs a running copytool
2258 local f=$DIR/$tdir/$tfile
2259 local fid=$(make_large_for_progress $f)
2261 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2262 error "could not archive file"
2263 wait_request_state $fid ARCHIVE STARTED
2265 check_hsm_flags $f "0x00000001"
2267 # Change metadata and sync to be sure we are not changing only
2272 wait_request_state $fid ARCHIVE SUCCEED
2274 check_hsm_flags $f "0x00000009"
2278 run_test 56 "Setattr during an archive is ok"
2281 # Need one client for I/O, one for request
2282 needclients 2 || return 0
2284 # test needs a running copytool
2288 local f=$DIR/$tdir/test_archive_remote
2289 # Create a file on a remote node
2290 do_node $CLIENT2 "dd if=/dev/urandom of=$f bs=1M "\
2291 "count=2 conv=fsync"
2294 do_node $CLIENT2 "$LFS hsm_archive -a $HSM_ARCHIVE_NUMBER $f" ||
2295 error "hsm_archive failed"
2296 local fid=$(path2fid $f)
2297 wait_request_state $fid ARCHIVE SUCCEED
2299 # Release and implicit restore it
2300 do_node $CLIENT2 "$LFS hsm_release $f" ||
2301 error "hsm_release failed"
2302 do_node $CLIENT2 "md5sum $f" ||
2303 error "hsm_restore failed"
2305 wait_request_state $fid RESTORE SUCCEED
2309 run_test 57 "Archive a file with dirty cache on another node"
2311 truncate_released_file() {
2315 local sz=$(stat -c %s $src_file)
2316 local f=$DIR/$tdir/$tfile
2317 local fid=$(copy_file $1 $f)
2321 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f ||
2322 error "could not archive file"
2323 wait_request_state $fid ARCHIVE SUCCEED
2325 $LFS hsm_release $f || error "could not release file"
2327 $TRUNCATE $f $trunc_to || error "truncate failed"
2330 local sz1=$(stat -c %s $f)
2331 [[ $sz1 == $trunc_to ]] ||
2332 error "size after trunc: $sz1 expect $trunc_to, original $sz"
2335 check_hsm_flags $f "0x0000000b"
2337 local state=$(get_request_state $fid RESTORE)
2338 [[ "$state" == "SUCCEED" ]] ||
2339 error "truncate $sz does not trig restore, state = $state"
2341 $TRUNCATE $ref $trunc_to
2342 cmp $ref $f || error "file data wrong after truncate"
2348 # test needs a running copytool
2353 local sz=$(stat -c %s /etc/passwd)
2355 echo "truncate up from $sz to $((sz*2))"
2356 truncate_released_file /etc/passwd $((sz*2))
2358 echo "truncate down from $sz to $((sz/2))"
2359 truncate_released_file /etc/passwd $((sz/2))
2361 echo "truncate to 0"
2362 truncate_released_file /etc/passwd 0
2366 run_test 58 "Truncate a released file will trigger restore"
2371 local f=$DIR/$tdir/$tfile
2372 local FILELIST=/tmp/filelist.txt
2376 for i in $(seq 1 $file_count); do
2377 fid=$(copy_file /etc/hosts $f.$i)
2378 echo $f.$i >> $FILELIST
2381 # to be sure wait_all_done will not be mislead by previous tests
2383 wait_for_grace_delay
2384 $LFS hsm_archive --filelist $FILELIST ||
2385 error "cannot archive a file list"
2387 $LFS hsm_release --filelist $FILELIST ||
2388 error "cannot release a file list"
2389 $LFS hsm_restore --filelist $FILELIST ||
2390 error "cannot restore a file list"
2394 run_test 90 "Archive/restore a file list"
2396 double_verify_reset_hsm_param() {
2398 echo "Testing $HSM_PARAM.$p"
2399 local val=$(get_hsm_param $p)
2401 local val2=$(($val * 2))
2402 set_hsm_param $p $val2
2403 val=$(get_hsm_param $p)
2404 [[ $val == $val2 ]] ||
2405 error "$HSM_PARAM.$p: $val != $val2 should be (2 * $save)"
2406 echo "Set $p to 0 must failed"
2410 set_hsm_param $p $save
2412 if [[ $rc == 0 ]]; then
2413 error "we must not be able to set $HSM_PARAM.$p to 0"
2418 double_verify_reset_hsm_param loop_period
2419 double_verify_reset_hsm_param grace_delay
2420 double_verify_reset_hsm_param active_request_timeout
2421 double_verify_reset_hsm_param max_requests
2422 double_verify_reset_hsm_param default_archive_id
2424 run_test 100 "Set coordinator /proc tunables"
2431 run_test 102 "Verify coordinator control"
2434 # test needs a running copytool
2441 for i in $(seq 1 20); do
2442 fid=$(copy_file /etc/passwd $DIR/$tdir/$i)
2444 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/*
2448 echo "Current requests"
2449 local res=$(do_facet $SINGLEMDS "$LCTL get_param -n\
2450 $HSM_PARAM.actions |\
2451 grep -v CANCELED | grep -v SUCCEED | grep -v FAILED")
2453 [[ -z "$res" ]] || error "Some request have not been canceled"
2457 run_test 103 "Purge all requests"
2462 # test needs a running copytool
2466 local f=$DIR/$tdir/$tfile
2467 local fid=$(make_large_for_progress $f)
2468 # if cdt is on, it can serve too quickly the request
2470 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER --data $DATA $f
2471 local data1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
2472 $HSM_PARAM.actions |\
2473 grep $fid | cut -f16 -d=")
2476 [[ "$data1" == "$DATAHEX" ]] ||
2477 error "Data field in records is ($data1) and not ($DATAHEX)"
2481 run_test 104 "Copy tool data field"
2488 for i in $(seq -w 1 10); do
2489 cp /etc/passwd $DIR/$tdir/$i
2490 $LFS hsm_archive $DIR/$tdir/$i
2492 local reqcnt1=$(do_facet $SINGLEMDS "$LCTL get_param -n\
2493 $HSM_PARAM.actions |\
2494 grep WAITING | wc -l")
2497 local reqcnt2=$(do_facet $SINGLEMDS "$LCTL get_param -n\
2498 $HSM_PARAM.actions |\
2499 grep WAITING | wc -l")
2502 [[ "$reqcnt1" == "$reqcnt2" ]] ||
2503 error "Requests count after shutdown $reqcnt2 != "\
2504 "before shutdown $reqcnt1"
2506 run_test 105 "Restart of coordinator"
2509 # test needs a running copytool
2512 local uuid=$(do_rpc_nodes $(facet_active_host $SINGLEAGT) \
2513 get_client_uuid $MOUNT | cut -d' ' -f2)
2514 local agent=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.agents |
2517 [[ ! -z "$agent" ]] || error "My uuid $uuid not found in agent list"
2518 local agent=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.agents |
2520 [[ -z "$agent" ]] ||
2521 error "My uuid $uuid still found in agent list,"\
2522 " after copytool shutdown"
2524 local agent=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.agents |
2527 [[ ! -z "$agent" ]] ||
2528 error "My uuid $uuid not found in agent list after"\
2531 run_test 106 "Copytool register/unregister"
2534 # test needs a running copytool
2536 # create and archive file
2538 local f1=$DIR/$tdir/$tfile
2539 local fid=$(copy_file /etc/passwd $f1)
2540 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
2541 wait_request_state $fid ARCHIVE SUCCEED
2542 # shutdown and restart MDS
2544 # check the copytool still gets messages from MDT
2545 local f2=$DIR/$tdir/2
2546 local fid=$(copy_file /etc/passwd $f2)
2547 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f2
2548 # main check of this sanity: this request MUST succeed
2549 wait_request_state $fid ARCHIVE SUCCEED
2552 run_test 107 "Copytool re-register after MDS restart"
2554 policy_set_and_test()
2558 do_facet $SINGLEMDS $LCTL set_param "$HSM_PARAM.policy=\\\"$change\\\""
2559 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
2560 [[ "$policy" == "$target" ]] ||
2561 error "Wrong policy after '$change': '$policy' != '$target'"
2565 # to force default policy setting if error
2566 CDT_POLICY_HAD_CHANGED=true
2568 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
2569 local default="NonBlockingRestore [NoRetryAction]"
2570 [[ "$policy" == "$default" ]] ||
2571 error "default policy has changed,"\
2572 " '$policy' != '$default' update the test"
2573 policy_set_and_test "+NBR" "[NonBlockingRestore] [NoRetryAction]"
2574 policy_set_and_test "+NRA" "[NonBlockingRestore] [NoRetryAction]"
2575 policy_set_and_test "-NBR" "NonBlockingRestore [NoRetryAction]"
2576 policy_set_and_test "-NRA" "NonBlockingRestore NoRetryAction"
2577 policy_set_and_test "NRA NBR" "[NonBlockingRestore] [NoRetryAction]"
2578 # useless bacause we know but safer for futur changes to use real value
2579 local policy=$(do_facet $SINGLEMDS $LCTL get_param -n $HSM_PARAM.policy)
2580 echo "Next set_param must failed"
2581 policy_set_and_test "wrong" "$policy"
2584 echo "Back to default policy"
2585 cdt_set_sanity_policy
2587 run_test 109 "Policy display/change"
2590 # test needs a running copytool
2595 copy2archive /etc/passwd $tdir/$tfile
2597 local f=$DIR/$tdir/$tfile
2598 import_file $tdir/$tfile $f
2599 local fid=$(path2fid $f)
2601 cdt_set_non_blocking_restore
2606 wait_request_state $fid RESTORE SUCCEED
2607 cdt_clear_non_blocking_restore
2611 error "md5sum returns $st != 1, "\
2612 "should also perror ENODATA (No data available)"
2616 run_test 110a "Non blocking restore policy (import case)"
2619 # test needs a running copytool
2623 local f=$DIR/$tdir/$tfile
2624 local fid=$(copy_file /etc/passwd $f)
2625 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2626 wait_request_state $fid ARCHIVE SUCCEED
2629 cdt_set_non_blocking_restore
2634 wait_request_state $fid RESTORE SUCCEED
2635 cdt_clear_non_blocking_restore
2639 error "md5sum returns $st != 1, "\
2640 "should also perror ENODATA (No data available)"
2644 run_test 110b "Non blocking restore policy (release case)"
2647 # test needs a running copytool
2651 copy2archive /etc/passwd $tdir/$tfile
2653 local f=$DIR/$tdir/$tfile
2655 import_file $tdir/$tfile $f
2656 local fid=$(path2fid $f)
2660 copytool_remove_backend $fid
2663 wait_request_state $fid RESTORE FAILED
2670 [[ $st == 0 ]] || error "Restore does not failed"
2674 run_test 111a "No retry policy (import case), restore will error"\
2675 " (No such file or directory)"
2678 # test needs a running copytool
2682 local f=$DIR/$tdir/$tfile
2683 local fid=$(copy_file /etc/passwd $f)
2685 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2686 wait_request_state $fid ARCHIVE SUCCEED
2689 copytool_remove_backend $fid
2692 wait_request_state $fid RESTORE FAILED
2699 [[ $st == 0 ]] || error "Restore does not failed"
2703 run_test 111b "No retry policy (release case), restore will error"\
2704 " (No such file or directory)"
2707 # test needs a running copytool
2711 local f=$DIR/$tdir/$tfile
2712 local fid=$(copy_file /etc/passwd $f)
2714 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2715 local l=$($LFS hsm_action $f)
2717 local res=$(echo $l | cut -f 2- -d" " | grep ARCHIVE)
2721 wait_request_state $fid ARCHIVE SUCCEED
2724 [[ ! -z "$res" ]] || error "action is $l which is not an ARCHIVE"
2728 run_test 112 "State of recorded request"
2731 # test needs a running copytool
2735 local f=$DIR/$tdir/$tfile
2736 local fid=$(make_large_for_cancel $f)
2737 # test with cdt on is made in test_221
2739 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2742 wait_request_state $fid ARCHIVE CANCELED
2743 wait_request_state $fid CANCEL SUCCEED
2747 run_test 200 "Register/Cancel archive"
2750 # test needs a running copytool
2754 local f=$DIR/$tdir/$tfile
2755 make_archive $tdir/$tfile
2756 import_file $tdir/$tfile $f
2757 local fid=$(path2fid $f)
2759 # test with cdt on is made in test_222
2764 wait_request_state $fid RESTORE CANCELED
2765 wait_request_state $fid CANCEL SUCCEED
2769 run_test 201 "Register/Cancel restore"
2772 # test needs a running copytool
2776 local f=$DIR/$tdir/$tfile
2777 local fid=$(make_large_for_progress $f)
2778 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2779 wait_request_state $fid ARCHIVE SUCCEED
2785 wait_request_state $fid REMOVE CANCELED
2789 run_test 202 "Register/Cancel remove"
2792 # test needs a running copytool
2797 local f=$DIR/$tdir/$tfile
2798 local fid=$(copy_file /etc/passwd $f)
2802 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2803 wait_request_state $fid ARCHIVE SUCCEED
2805 local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
2809 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
2813 run_test 220 "Changelog for archive"
2816 # test needs a running copytool
2821 local f=$DIR/$tdir/$tfile
2822 local fid=$(make_large_for_cancel $f)
2826 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2827 wait_request_state $fid ARCHIVE STARTED
2829 wait_request_state $fid ARCHIVE CANCELED
2830 wait_request_state $fid CANCEL SUCCEED
2832 local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
2835 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
2839 run_test 221 "Changelog for archive canceled"
2842 # test needs a running copytool
2846 copy2archive /etc/passwd $tdir/$tfile
2848 local f=$DIR/$tdir/$tfile
2849 import_file $tdir/$tfile $f
2850 local fid=$(path2fid $f)
2855 wait_request_state $fid RESTORE SUCCEED
2857 local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
2860 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
2864 run_test 222a "Changelog for explicit restore"
2867 # test needs a running copytool
2871 local f=$DIR/$tdir/$tfile
2872 local fid=$(copy_file /etc/passwd $f)
2875 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2876 wait_request_state $fid ARCHIVE SUCCEED
2881 wait_request_state $fid RESTORE SUCCEED
2883 local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
2886 [[ $flags == $target ]] || error "Changelog flag is $flags not $target"
2890 run_test 222b "Changelog for implicit restore"
2893 # test needs a running copytool
2898 local f=$DIR/$tdir/$tfile
2899 make_archive $tdir/$tfile
2903 import_file $tdir/$tfile $f
2904 local fid=$(path2fid $f)
2907 wait_request_state $fid RESTORE STARTED
2909 wait_request_state $fid RESTORE CANCELED
2910 wait_request_state $fid CANCEL SUCCEED
2912 local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
2915 [[ $flags == $target ]] ||
2916 error "Changelog flag is $flags not $target"
2920 run_test 223a "Changelog for restore canceled (import case)"
2923 # test needs a running copytool
2928 local f=$DIR/$tdir/$tfile
2929 local fid=$(make_large_for_progress $f)
2932 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2933 wait_request_state $fid ARCHIVE SUCCEED
2936 wait_request_state $fid RESTORE STARTED
2938 wait_request_state $fid RESTORE CANCELED
2939 wait_request_state $fid CANCEL SUCCEED
2941 local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
2944 [[ $flags == $target ]] ||
2945 error "Changelog flag is $flags not $target"
2949 run_test 223b "Changelog for restore canceled (release case)"
2952 # test needs a running copytool
2957 local f=$DIR/$tdir/$tfile
2958 local fid=$(copy_file /etc/passwd $f)
2961 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2962 wait_request_state $fid ARCHIVE SUCCEED
2965 wait_request_state $fid REMOVE SUCCEED
2967 local flags=$(changelog_get_flags $MDT0 HSM $fid | tail -1)
2970 [[ $flags == $target ]] ||
2971 error "Changelog flag is $flags not $target"
2975 run_test 224 "Changelog for remove"
2978 # test needs a running copytool
2981 # test is not usable because remove request is too fast
2982 # so it is always finished before cancel can be done ...
2983 echo "Test disabled"
2988 local f=$DIR/$tdir/$tfile
2989 local fid=$(make_large_for_progress $f)
2992 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
2993 wait_request_state $fid ARCHIVE SUCCEED
2995 # if cdt is on, it can serve too quickly the request
3000 wait_request_state $fid REMOVE CANCELED
3001 wait_request_state $fid CANCEL SUCCEED
3003 flags=$(changelog_get_flags $MDT0 RENME $fid2)
3004 local flags=$($LFS changelog $MDT0 | grep HSM | grep $fid | tail -1 |
3008 [[ $flags == $target ]] ||
3009 error "Changelog flag is $flags not $target"
3013 run_test 225 "Changelog for remove canceled"
3016 # test needs a running copytool
3021 local f1=$DIR/$tdir/$tfile-1
3022 local f2=$DIR/$tdir/$tfile-2
3023 local f3=$DIR/$tdir/$tfile-3
3024 local fid1=$(copy_file /etc/passwd $f1)
3025 local fid2=$(copy_file /etc/passwd $f2)
3026 copy_file /etc/passwd $f3
3029 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f1
3030 wait_request_state $fid1 ARCHIVE SUCCEED
3032 $LFS hsm_archive $f2
3033 wait_request_state $fid2 ARCHIVE SUCCEED
3035 rm $f1 || error "rm $f1 failed"
3037 local flags=$(changelog_get_flags $MDT0 UNLNK $fid1)
3040 [[ $flags == $target ]] ||
3041 error "Changelog flag is $flags not $target"
3043 mv $f3 $f2 || error "mv $f3 $f2 failed"
3045 flags=$(changelog_get_flags $MDT0 RENME $fid2)
3048 [[ $flags == $target ]] ||
3049 error "Changelog flag is $flags not $target"
3053 run_test 226 "changelog for last rm/mv with exiting archive"
3055 check_flags_changes() {
3063 $LFS hsm_set --$hsm_flag $f ||
3064 error "Cannot set $hsm_flag on $f"
3065 local flags=($(changelog_get_flags $MDT0 HSM $fid))
3066 local seen=${#flags[*]}
3068 [[ $seen == $cnt ]] ||
3069 error "set $hsm_flag: Changelog events $seen != $cnt"
3070 [[ ${flags[$((cnt - 1))]} == $target ]] ||
3071 error "set $hsm_flag: Changelog flags are "\
3072 "${flags[$((cnt - 1))]} not $target"
3074 $LFS hsm_clear --$hsm_flag $f ||
3075 error "Cannot clear $hsm_flag on $f"
3076 flags=($(changelog_get_flags $MDT0 HSM $fid))
3079 [[ $cnt == $seen ]] ||
3080 error "clear $hsm_flag: Changelog events $seen != $cnt"
3082 [[ ${flags[$((cnt - 1))]} == $target ]] ||
3083 error "clear $hsm_flag: Changelog flag is "\
3084 "${flags[$((cnt - 1))]} not $target"
3088 # test needs a running copytool
3095 for i in norelease noarchive exists archived
3097 local f=$DIR/$tdir/$tfile-$i
3098 local fid=$(copy_file /etc/passwd $f)
3099 check_flags_changes $f $fid $i 0 1
3102 f=$DIR/$tdir/$tfile---lost
3103 fid=$(copy_file /etc/passwd $f)
3104 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3105 wait_request_state $fid ARCHIVE SUCCEED
3106 check_flags_changes $f $fid lost 3 1
3110 run_test 227 "changelog when explicit setting of HSM flags"
3113 # test needs a running copytool
3116 dd if=/dev/urandom of=$DIR/$tfile bs=1M count=1 conv=sync ||
3117 error "creating $DIR/$tfile"
3118 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tfile
3119 wait_request_state $(path2fid $DIR/$tfile) ARCHIVE SUCCEED
3121 $LFS hsm_release $DIR/$tfile
3122 check_hsm_flags $DIR/$tfile "0x0000000d"
3124 filefrag $DIR/$tfile | grep " 1 extent found" ||
3125 error "filefrag on released file must return only one extent"
3127 # only newer versions of cp detect sparse files by stat/FIEMAP
3129 cp --sparse=auto $DIR/$tfile $DIR/$tfile.2 ||
3130 error "copying $DIR/$tfile"
3131 cmp $DIR/$tfile $DIR/$tfile.2 || error "comparing copied $DIR/$tfile"
3133 $LFS hsm_release $DIR/$tfile
3134 check_hsm_flags $DIR/$tfile "0x0000000d"
3138 tar cf - --sparse $DIR/$tfile | tar xvf - -C $DIR/$tdir ||
3140 cmp $DIR/$tfile $DIR/$tdir/$DIR/$tfile ||
3141 error "comparing untarred $DIR/$tfile"
3145 run_test 228 "On released file, return extend to FIEMAP. For [cp,tar] --sparse"
3148 # test needs a running copytool
3152 local maxrequest=$(get_hsm_param max_requests)
3153 local rqcnt=$(($maxrequest * 3))
3157 for i in $(seq -w 1 $rqcnt); do
3159 dd if=/dev/urandom of=$DIR/$tdir/$i bs=1M count=10 conv=fsync
3161 # we do it in 2 steps, so all requests arrive at the same time
3162 for i in $(seq -w 1 $rqcnt); do
3163 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $DIR/$tdir/$i
3168 while [[ $cnt != 0 || $wt != 0 ]]; do
3170 cnt=$(do_facet $SINGLEMDS "$LCTL get_param -n\
3171 $HSM_PARAM.actions |\
3172 grep STARTED | grep -v CANCEL | wc -l")
3173 [[ $cnt -le $maxrequest ]] ||
3174 error "$cnt > $maxrequest too many started requests"
3175 wt=$(do_facet $SINGLEMDS "$LCTL get_param\
3176 $HSM_PARAM.actions |\
3177 grep WAITING | wc -l")
3178 echo "max=$maxrequest started=$cnt waiting=$wt"
3183 run_test 250 "Coordinator max request"
3186 # test needs a running copytool
3190 local f=$DIR/$tdir/$tfile
3191 local fid=$(make_large_for_cancel $f)
3194 # to have a short test
3195 local old_to=$(get_hsm_param active_request_timeout)
3196 set_hsm_param active_request_timeout 4
3197 # to be sure the cdt will wake up frequently so
3198 # it will be able to cancel the "old" request
3199 local old_loop=$(get_hsm_param loop_period)
3200 set_hsm_param loop_period 2
3203 $LFS hsm_archive --archive $HSM_ARCHIVE_NUMBER $f
3204 wait_request_state $fid ARCHIVE STARTED
3206 wait_request_state $fid ARCHIVE CANCELED
3208 set_hsm_param active_request_timeout $old_to
3209 set_hsm_param loop_period $old_loop
3213 run_test 251 "Coordinator request timeout"
3216 # the only way to test ondisk conf is to restart MDS ...
3217 echo "Stop coordinator and remove coordinator state at mount"
3220 # clean on disk conf set by default
3221 cdt_clear_mount_state
3222 cdt_check_state stopped
3224 # check cdt still off after umount/remount
3226 cdt_check_state stopped
3228 echo "Set coordinator start at mount, and start coordinator"
3229 cdt_set_mount_state enabled
3232 cdt_check_state enabled
3234 # check cdt still on after umount/remount
3236 cdt_check_state enabled
3238 # we are back to original state (cdt started at mount)
3240 run_test 300 "On disk coordinator state kept between MDT umount/mount"
3243 local ai=$(get_hsm_param default_archive_id)
3244 local new=$((ai + 1))
3246 set_hsm_param default_archive_id $new -P
3248 local res=$(get_hsm_param default_archive_id)
3251 set_hsm_param default_archive_id "" "-P -d"
3253 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
3255 run_test 301 "HSM tunnable are persistent"
3258 local ai=$(get_hsm_param default_archive_id)
3259 local new=$((ai + 1))
3264 set_hsm_param default_archive_id $new -P
3268 cdt_check_state enabled
3270 local res=$(get_hsm_param default_archive_id)
3273 set_hsm_param default_archive_id "" "-P -d"
3275 [[ $new == $res ]] || error "Value after MDS restart is $res != $new"
3277 run_test 302 "HSM tunnable are persistent when CDT is off"
3282 check_and_cleanup_lustre