3 # Run select tests by setting ONLY, or as arguments to the script.
4 # Skip specific tests by setting EXCEPT.
11 #Bug number for excepting test
12 ALWAYS_EXCEPT="$SANITY_LFSCK_EXCEPT"
14 [ "$SLOW" = "no" ] && EXCEPT_SLOW=""
15 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
17 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
18 . $LUSTRE/tests/test-framework.sh
20 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
23 require_dsh_mds || exit 0
27 if ! check_versions; then
28 skip "It is NOT necessary to test lfsck under interoperation mode"
32 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.3.60) ]] &&
33 skip "Need MDS version at least 2.3.60" && exit 0
37 SAVED_MDSSIZE=${MDSSIZE}
38 SAVED_OSTSIZE=${OSTSIZE}
39 SAVED_OSTCOUNT=${OSTCOUNT}
40 # use small MDS + OST size to speed formatting time
41 # do not use too small MDSSIZE/OSTSIZE, which affect the default journal size
44 # no need too many OSTs, to reduce the format/start/stop overhead
45 [ $OSTCOUNT -gt 4 ] && OSTCOUNT=4
47 # build up a clean test environment.
51 [[ $(lustre_version_code $SINGLEMDS) -le $(version_code 2.4.90) ]] &&
52 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2c"
54 [[ $(lustre_version_code ost1) -lt $(version_code 2.5.55) ]] &&
55 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 12 13 14 15 16 17 18 19 20 21"
57 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.50) ]] &&
58 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2d 2e 3 22 23 24 25 26 27 28 29 30 31"
60 # DNE does not support striped directory on zfs-based backend yet.
61 [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
62 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 31"
66 MDT_DEV="${FSNAME}-MDT0000"
67 OST_DEV="${FSNAME}-OST0000"
68 MDT_DEVNAME=$(mdsdevname ${SINGLEMDS//mds/})
69 START_NAMESPACE="do_facet $SINGLEMDS \
70 $LCTL lfsck_start -M ${MDT_DEV} -t namespace"
71 START_LAYOUT="do_facet $SINGLEMDS \
72 $LCTL lfsck_start -M ${MDT_DEV} -t layout"
73 START_LAYOUT_ON_OST="do_facet ost1 $LCTL lfsck_start -M ${OST_DEV} -t layout"
74 STOP_LFSCK="do_facet $SINGLEMDS $LCTL lfsck_stop -M ${MDT_DEV}"
75 SHOW_NAMESPACE="do_facet $SINGLEMDS \
76 $LCTL get_param -n mdd.${MDT_DEV}.lfsck_namespace"
77 SHOW_LAYOUT="do_facet $SINGLEMDS \
78 $LCTL get_param -n mdd.${MDT_DEV}.lfsck_layout"
79 SHOW_LAYOUT_ON_OST="do_facet ost1 \
80 $LCTL get_param -n obdfilter.${OST_DEV}.lfsck_layout"
81 MOUNT_OPTS_SCRUB="-o user_xattr"
82 MOUNT_OPTS_NOSCRUB="-o user_xattr,noscrub"
83 MOUNT_OPTS_SKIP_LFSCK="-o user_xattr,skip_lfsck"
92 echo "preparing... $nfiles * $ndirs files will be created $(date)."
93 if [ ! -z $igif ]; then
94 #define OBD_FAIL_FID_IGIF 0x1504
95 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1504
98 cp $LUSTRE/tests/*.sh $DIR/$tdir/
99 if [ $ndirs -gt 0 ]; then
100 createmany -d $DIR/$tdir/d $ndirs
101 createmany -m $DIR/$tdir/f $ndirs
102 if [ $nfiles -gt 0 ]; then
103 for ((i = 0; i < $ndirs; i++)); do
104 createmany -m $DIR/$tdir/d${i}/f $nfiles > \
105 /dev/null || error "createmany $nfiles"
108 createmany -d $DIR/$tdir/e $ndirs
111 if [ ! -z $igif ]; then
112 touch $DIR/$tdir/dummy
113 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
116 echo "prepared $(date)."
119 run_e2fsck_on_mdt0() {
120 [ $(facet_fstype $SINGLEMDS) != ldiskfs ] && return
122 stop $SINGLEMDS > /dev/null || error "(0) Fail to the stop MDT0"
123 run_e2fsck $(facet_active_host $SINGLEMDS) $(mdsdevname 1) "-n" |
125 run_e2fsck $(facet_active_host $SINGLEMDS) $(mdsdevname 1) "-n"
126 error "(2) Detected inconsistency on MDT0"
128 start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
129 error "(3) Fail to start MDT0"
132 wait_all_targets_blocked() {
137 local count=$(do_facet mds1 \
138 "$LCTL lfsck_query -t $com -M ${FSNAME}-MDT0000 -w |
139 awk '/^${com}_mdts_${status}/ { print \\\$2 }'")
140 [[ $count -eq $MDSCOUNT ]] || {
141 do_facet mds1 "$LCTL lfsck_query -t $com -M ${FSNAME}-MDT0000"
142 error "($err) only $count of $MDSCOUNT MDTs are in ${status}"
151 wait_update_facet mds1 "$LCTL lfsck_query -t $com -M ${FSNAME}-MDT0000 |
152 awk '/^${com}_mdts_${status}/ { print \\\$2 }'" \
153 "$MDSCOUNT" $LTIME || {
154 do_facet mds1 "$LCTL lfsck_query -t $com -M ${FSNAME}-MDT0000"
155 error "($err) some MDTs are not in ${status}"
162 #define OBD_FAIL_LFSCK_DELAY1 0x1600
163 do_facet $SINGLEMDS $LCTL set_param fail_val=3 fail_loc=0x1600
164 $START_NAMESPACE -r || error "(2) Fail to start LFSCK for namespace!"
166 $SHOW_NAMESPACE || error "Fail to monitor LFSCK (3)"
168 local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
169 [ "$STATUS" == "scanning-phase1" ] ||
170 error "(4) Expect 'scanning-phase1', but got '$STATUS'"
172 $STOP_LFSCK || error "(5) Fail to stop LFSCK!"
174 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
175 [ "$STATUS" == "stopped" ] ||
176 error "(6) Expect 'stopped', but got '$STATUS'"
178 $START_NAMESPACE || error "(7) Fail to start LFSCK for namespace!"
180 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
181 [ "$STATUS" == "scanning-phase1" ] ||
182 error "(8) Expect 'scanning-phase1', but got '$STATUS'"
184 do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
185 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
186 mdd.${MDT_DEV}.lfsck_namespace |
187 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
189 error "(9) unexpected status"
192 local repaired=$($SHOW_NAMESPACE |
193 awk '/^updated_phase1/ { print $2 }')
194 [ $repaired -eq 0 ] ||
195 error "(10) Expect nothing to be repaired, but got: $repaired"
197 local scanned1=$($SHOW_NAMESPACE | awk '/^success_count/ { print $2 }')
198 $START_NAMESPACE -r || error "(11) Fail to reset LFSCK!"
199 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
200 mdd.${MDT_DEV}.lfsck_namespace |
201 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
203 error "(12) unexpected status"
206 local scanned2=$($SHOW_NAMESPACE | awk '/^success_count/ { print $2 }')
207 [ $((scanned1 + 1)) -eq $scanned2 ] ||
208 error "(13) Expect success $((scanned1 + 1)), but got $scanned2"
210 echo "stopall, should NOT crash LU-3649"
211 stopall || error "(14) Fail to stopall"
213 run_test 0 "Control LFSCK manually"
216 [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
217 skip "OI Scrub not implemented for ZFS" && return
221 #define OBD_FAIL_FID_INDIR 0x1501
222 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1501
223 touch $DIR/$tdir/dummy
225 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
227 $START_NAMESPACE -r || error "(3) Fail to start LFSCK for namespace!"
228 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
229 mdd.${MDT_DEV}.lfsck_namespace |
230 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
232 error "(4) unexpected status"
235 local repaired=$($SHOW_NAMESPACE |
236 awk '/^dirent_repaired/ { print $2 }')
237 # for interop with old server
238 [ -z "$repaired" ] &&
239 repaired=$($SHOW_NAMESPACE |
240 awk '/^updated_phase1/ { print $2 }')
242 [ $repaired -eq 1 ] ||
243 error "(5) Fail to repair crashed FID-in-dirent: $repaired"
247 mount_client $MOUNT || error "(6) Fail to start client!"
249 #define OBD_FAIL_FID_LOOKUP 0x1505
250 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1505
251 ls $DIR/$tdir/ > /dev/null || error "(7) no FID-in-dirent."
253 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
255 run_test 1a "LFSCK can find out and repair crashed FID-in-dirent"
259 [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
260 skip "OI Scrub not implemented for ZFS" && return
264 #define OBD_FAIL_FID_INLMA 0x1502
265 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1502
266 touch $DIR/$tdir/dummy
268 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
270 #define OBD_FAIL_FID_NOLMA 0x1506
271 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1506
272 $START_NAMESPACE -r || error "(3) Fail to start LFSCK for namespace!"
273 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
274 mdd.${MDT_DEV}.lfsck_namespace |
275 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
277 error "(4) unexpected status"
280 local repaired=$($SHOW_NAMESPACE |
281 awk '/^dirent_repaired/ { print $2 }')
282 # for interop with old server
283 [ -z "$repaired" ] &&
284 repaired=$($SHOW_NAMESPACE |
285 awk '/^updated_phase1/ { print $2 }')
287 [ $repaired -eq 1 ] ||
288 error "(5) Fail to repair the missing FID-in-LMA: $repaired"
290 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
293 mount_client $MOUNT || error "(6) Fail to start client!"
295 #define OBD_FAIL_FID_LOOKUP 0x1505
296 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1505
297 stat $DIR/$tdir/dummy > /dev/null || error "(7) no FID-in-LMA."
299 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
301 run_test 1b "LFSCK can find out and repair the missing FID-in-LMA"
306 #define OBD_FAIL_LFSCK_LINKEA_CRASH 0x1603
307 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1603
308 touch $DIR/$tdir/dummy
310 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
312 $START_NAMESPACE -r || error "(3) Fail to start LFSCK for namespace!"
313 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
314 mdd.${MDT_DEV}.lfsck_namespace |
315 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
317 error "(4) unexpected status"
320 local repaired=$($SHOW_NAMESPACE |
321 awk '/^linkea_repaired/ { print $2 }')
322 # for interop with old server
323 [ -z "$repaired" ] &&
324 repaired=$($SHOW_NAMESPACE |
325 awk '/^updated_phase2/ { print $2 }')
327 [ $repaired -eq 1 ] ||
328 error "(5) Fail to repair crashed linkEA: $repaired"
332 mount_client $MOUNT || error "(6) Fail to start client!"
334 stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
335 error "(7) Fail to stat $DIR/$tdir/dummy"
337 local dummyfid=$($LFS path2fid $DIR/$tdir/dummy)
338 local dummyname=$($LFS fid2path $DIR $dummyfid)
339 [ "$dummyname" == "$DIR/$tdir/dummy" ] ||
340 error "(8) Fail to repair linkEA: $dummyfid $dummyname"
342 run_test 2a "LFSCK can find out and repair crashed linkEA entry"
348 #define OBD_FAIL_LFSCK_LINKEA_MORE 0x1604
349 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1604
350 touch $DIR/$tdir/dummy
352 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
354 $START_NAMESPACE -r || error "(3) Fail to start LFSCK for namespace!"
355 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
356 mdd.${MDT_DEV}.lfsck_namespace |
357 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
359 error "(4) unexpected status"
362 local repaired=$($SHOW_NAMESPACE |
363 awk '/^updated_phase2/ { print $2 }')
364 [ $repaired -eq 1 ] ||
365 error "(5) Fail to repair crashed linkEA: $repaired"
369 mount_client $MOUNT || error "(6) Fail to start client!"
371 stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
372 error "(7) Fail to stat $DIR/$tdir/dummy"
374 local dummyfid=$($LFS path2fid $DIR/$tdir/dummy)
375 local dummyname=$($LFS fid2path $DIR $dummyfid)
376 [ "$dummyname" == "$DIR/$tdir/dummy" ] ||
377 error "(8) Fail to repair linkEA: $dummyfid $dummyname"
379 run_test 2b "LFSCK can find out and remove invalid linkEA entry"
385 #define OBD_FAIL_LFSCK_LINKEA_MORE2 0x1605
386 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1605
387 touch $DIR/$tdir/dummy
389 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
391 $START_NAMESPACE -r || error "(3) Fail to start LFSCK for namespace!"
392 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
393 mdd.${MDT_DEV}.lfsck_namespace |
394 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
396 error "(4) unexpected status"
399 local repaired=$($SHOW_NAMESPACE |
400 awk '/^updated_phase2/ { print $2 }')
401 [ $repaired -eq 1 ] ||
402 error "(5) Fail to repair crashed linkEA: $repaired"
406 mount_client $MOUNT || error "(6) Fail to start client!"
408 stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
409 error "(7) Fail to stat $DIR/$tdir/dummy"
411 local dummyfid=$($LFS path2fid $DIR/$tdir/dummy)
412 local dummyname=$($LFS fid2path $DIR $dummyfid)
413 [ "$dummyname" == "$DIR/$tdir/dummy" ] ||
414 error "(8) Fail to repair linkEA: $dummyfid $dummyname"
416 run_test 2c "LFSCK can find out and remove repeated linkEA entry"
422 #define OBD_FAIL_LFSCK_NO_LINKEA 0x161d
423 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x161d
424 touch $DIR/$tdir/dummy
426 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
428 $START_NAMESPACE -r || error "(3) Fail to start LFSCK for namespace!"
429 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
430 mdd.${MDT_DEV}.lfsck_namespace |
431 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
433 error "(4) unexpected status"
436 local repaired=$($SHOW_NAMESPACE |
437 awk '/^linkea_repaired/ { print $2 }')
438 [ $repaired -eq 1 ] ||
439 error "(5) Fail to repair crashed linkEA: $repaired"
443 mount_client $MOUNT || error "(6) Fail to start client!"
445 stat $DIR/$tdir/dummy | grep "Links: 1" > /dev/null ||
446 error "(7) Fail to stat $DIR/$tdir/dummy"
448 local dummyfid=$($LFS path2fid $DIR/$tdir/dummy)
449 local dummyname=$($LFS fid2path $DIR $dummyfid)
450 [ "$dummyname" == "$DIR/$tdir/dummy" ] ||
451 error "(8) Fail to repair linkEA: $dummyfid $dummyname"
453 run_test 2d "LFSCK can recover the missing linkEA entry"
457 [ $MDSCOUNT -lt 2 ] &&
458 skip "We need at least 2 MDSes for this test" && return
462 $LFS mkdir -i 1 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0 on MDT1"
464 #define OBD_FAIL_LFSCK_LINKEA_CRASH 0x1603
465 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1603
466 $LFS mkdir -i 0 $DIR/$tdir/d0/d1 || error "(2) Fail to mkdir d1 on MDT0"
467 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
469 $START_NAMESPACE -r -A || error "(3) Fail to start LFSCK for namespace!"
471 wait_all_targets_blocked namespace completed 4
473 local repaired=$($SHOW_NAMESPACE |
474 awk '/^linkea_repaired/ { print $2 }')
475 [ $repaired -eq 1 ] ||
476 error "(5) Fail to repair crashed linkEA: $repaired"
478 local fid=$($LFS path2fid $DIR/$tdir/d0/d1)
479 local name=$($LFS fid2path $DIR $fid)
480 [ "$name" == "$DIR/$tdir/d0/d1" ] ||
481 error "(6) Fail to repair linkEA: $fid $name"
483 run_test 2e "namespace LFSCK can verify remote object linkEA"
489 mkdir $DIR/$tdir/dummy || error "(1) Fail to mkdir"
490 ln $DIR/$tdir/d0/f0 $DIR/$tdir/dummy/f0 || error "(2) Fail to hardlink"
491 ln $DIR/$tdir/d0/f1 $DIR/$tdir/dummy/f1 || error "(3) Fail to hardlink"
493 $LFS mkdir -i 0 $DIR/$tdir/edir || error "(4) Fail to mkdir"
494 touch $DIR/$tdir/edir/f0 || error "(5) Fail to touch"
495 touch $DIR/$tdir/edir/f1 || error "(6) Fail to touch"
497 #define OBD_FAIL_LFSCK_LINKEA_CRASH 0x1603
498 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1603
499 ln $DIR/$tdir/edir/f0 $DIR/$tdir/edir/w0 || error "(7) Fail to hardlink"
501 #define OBD_FAIL_LFSCK_LINKEA_MORE 0x1604
502 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1604
503 ln $DIR/$tdir/edir/f1 $DIR/$tdir/edir/w1 || error "(8) Fail to hardlink"
505 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
507 $START_NAMESPACE -r || error "(9) Fail to start LFSCK for namespace!"
508 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
509 mdd.${MDT_DEV}.lfsck_namespace |
510 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
512 error "(10) unexpected status"
515 local checked=$($SHOW_NAMESPACE |
516 awk '/^checked_phase2/ { print $2 }')
517 [ $checked -ge 4 ] ||
518 error "(11) Fail to check multiple-linked object: $checked"
520 local repaired=$($SHOW_NAMESPACE |
521 awk '/^multiple_linked_repaired/ { print $2 }')
522 [ $repaired -ge 2 ] ||
523 error "(12) Fail to repair multiple-linked object: $repaired"
525 run_test 3 "LFSCK can verify multiple-linked objects"
529 [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
530 skip "OI Scrub not implemented for ZFS" && return
533 cleanup_mount $MOUNT || error "(0.1) Fail to stop client!"
534 stop $SINGLEMDS > /dev/null || error "(0.2) Fail to stop MDS!"
536 mds_backup_restore $SINGLEMDS || error "(1) Fail to backup/restore!"
537 echo "start $SINGLEMDS with disabling OI scrub"
538 start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
539 error "(2) Fail to start MDS!"
541 #define OBD_FAIL_LFSCK_DELAY2 0x1601
542 do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
543 $START_NAMESPACE -r || error "(4) Fail to start LFSCK for namespace!"
544 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
545 mdd.${MDT_DEV}.lfsck_namespace |
546 awk '/^flags/ { print \\\$2 }'" "inconsistent" 32 || {
548 error "(5) unexpected status"
551 local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
552 [ "$STATUS" == "scanning-phase1" ] ||
553 error "(6) Expect 'scanning-phase1', but got '$STATUS'"
555 do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
556 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
557 mdd.${MDT_DEV}.lfsck_namespace |
558 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
560 error "(7) unexpected status"
563 FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
564 [ -z "$FLAGS" ] || error "(8) Expect empty flags, but got '$FLAGS'"
566 local repaired=$($SHOW_NAMESPACE |
567 awk '/^dirent_repaired/ { print $2 }')
568 # for interop with old server
569 [ -z "$repaired" ] &&
570 repaired=$($SHOW_NAMESPACE |
571 awk '/^updated_phase1/ { print $2 }')
573 [ $repaired -ge 9 ] ||
574 error "(9) Fail to re-generate FID-in-dirent: $repaired"
578 mount_client $MOUNT || error "(10) Fail to start client!"
580 #define OBD_FAIL_FID_LOOKUP 0x1505
581 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1505
582 ls $DIR/$tdir/ > /dev/null || error "(11) no FID-in-dirent."
583 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
585 run_test 4 "FID-in-dirent can be rebuilt after MDT file-level backup/restore"
589 [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
590 skip "OI Scrub not implemented for ZFS" && return
593 cleanup_mount $MOUNT || error "(0.1) Fail to stop client!"
594 stop $SINGLEMDS > /dev/null || error "(0.2) Fail to stop MDS!"
596 mds_backup_restore $SINGLEMDS 1 || error "(1) Fail to backup/restore!"
597 echo "start $SINGLEMDS with disabling OI scrub"
598 start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
599 error "(2) Fail to start MDS!"
601 #define OBD_FAIL_LFSCK_DELAY2 0x1601
602 do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
603 $START_NAMESPACE -r || error "(4) Fail to start LFSCK for namespace!"
604 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
605 mdd.${MDT_DEV}.lfsck_namespace |
606 awk '/^flags/ { print \\\$2 }'" "inconsistent,upgrade" 32 || {
608 error "(5) unexpected status"
611 local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
612 [ "$STATUS" == "scanning-phase1" ] ||
613 error "(6) Expect 'scanning-phase1', but got '$STATUS'"
615 do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
616 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
617 mdd.${MDT_DEV}.lfsck_namespace |
618 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
620 error "(7) unexpected status"
623 FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
624 [ -z "$FLAGS" ] || error "(8) Expect empty flags, but got '$FLAGS'"
626 local repaired=$($SHOW_NAMESPACE |
627 awk '/^dirent_repaired/ { print $2 }')
628 # for interop with old server
629 [ -z "$repaired" ] &&
630 repaired=$($SHOW_NAMESPACE |
631 awk '/^updated_phase1/ { print $2 }')
633 [ $repaired -ge 2 ] ||
634 error "(9) Fail to generate FID-in-dirent for IGIF: $repaired"
638 mount_client $MOUNT || error "(10) Fail to start client!"
640 #define OBD_FAIL_FID_LOOKUP 0x1505
641 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1505
642 stat $DIR/$tdir/dummy > /dev/null || error "(11) no FID-in-LMA."
644 ls $DIR/$tdir/ > /dev/null || error "(12) no FID-in-dirent."
646 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
647 local dummyfid=$($LFS path2fid $DIR/$tdir/dummy)
648 local dummyname=$($LFS fid2path $DIR $dummyfid)
649 [ "$dummyname" == "$DIR/$tdir/dummy" ] ||
650 error "(13) Fail to generate linkEA: $dummyfid $dummyname"
652 run_test 5 "LFSCK can handle IGIF object upgrading"
657 #define OBD_FAIL_LFSCK_DELAY1 0x1600
658 do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1600
659 $START_NAMESPACE -r || error "(2) Fail to start LFSCK for namespace!"
661 local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
662 [ "$STATUS" == "scanning-phase1" ] ||
663 error "(3) Expect 'scanning-phase1', but got '$STATUS'"
665 # Sleep 3 sec to guarantee at least one object processed by LFSCK
667 # Fail the LFSCK to guarantee there is at least one checkpoint
668 #define OBD_FAIL_LFSCK_FATAL1 0x1608
669 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x80001608
670 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
671 mdd.${MDT_DEV}.lfsck_namespace |
672 awk '/^status/ { print \\\$2 }'" "failed" 32 || {
674 error "(4) unexpected status"
677 local POS0=$($SHOW_NAMESPACE |
678 awk '/^last_checkpoint_position/ { print $2 }' |
681 #define OBD_FAIL_LFSCK_DELAY1 0x1600
682 do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1600
683 $START_NAMESPACE || error "(5) Fail to start LFSCK for namespace!"
685 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
686 [ "$STATUS" == "scanning-phase1" ] ||
687 error "(6) Expect 'scanning-phase1', but got '$STATUS'"
689 local POS1=$($SHOW_NAMESPACE |
690 awk '/^latest_start_position/ { print $2 }' |
692 [[ $POS0 -lt $POS1 ]] ||
693 error "(7) Expect larger than: $POS0, but got $POS1"
695 do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
696 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
697 mdd.${MDT_DEV}.lfsck_namespace |
698 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
700 error "(8) unexpected status"
703 run_test 6a "LFSCK resumes from last checkpoint (1)"
708 #define OBD_FAIL_LFSCK_DELAY2 0x1601
709 do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
710 $START_NAMESPACE -r || error "(2) Fail to start LFSCK for namespace!"
712 local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
713 [ "$STATUS" == "scanning-phase1" ] ||
714 error "(3) Expect 'scanning-phase1', but got '$STATUS'"
716 # Sleep 5 sec to guarantee that we are in the directory scanning
718 # Fail the LFSCK to guarantee there is at least one checkpoint
719 #define OBD_FAIL_LFSCK_FATAL2 0x1609
720 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x80001609
721 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
722 mdd.${MDT_DEV}.lfsck_namespace |
723 awk '/^status/ { print \\\$2 }'" "failed" 32 || {
725 error "(4) unexpected status"
728 local O_POS0=$($SHOW_NAMESPACE |
729 awk '/^last_checkpoint_position/ { print $2 }' |
732 local D_POS0=$($SHOW_NAMESPACE |
733 awk '/^last_checkpoint_position/ { print $4 }')
735 #define OBD_FAIL_LFSCK_DELAY2 0x1601
736 do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
737 $START_NAMESPACE || error "(5) Fail to start LFSCK for namespace!"
739 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
740 [ "$STATUS" == "scanning-phase1" ] ||
741 error "(6) Expect 'scanning-phase1', but got '$STATUS'"
743 local O_POS1=$($SHOW_NAMESPACE |
744 awk '/^latest_start_position/ { print $2 }' |
746 local D_POS1=$($SHOW_NAMESPACE |
747 awk '/^latest_start_position/ { print $4 }')
749 if [ "$D_POS0" == "N/A" -o "$D_POS1" == "N/A" ]; then
750 [[ $O_POS0 -lt $O_POS1 ]] ||
751 error "(7.1) $O_POS1 is not larger than $O_POS0"
753 [[ $D_POS0 -lt $D_POS1 ]] ||
754 error "(7.2) $D_POS1 is not larger than $D_POS0"
757 do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
758 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
759 mdd.${MDT_DEV}.lfsck_namespace |
760 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
762 error "(8) unexpected status"
765 run_test 6b "LFSCK resumes from last checkpoint (2)"
772 #define OBD_FAIL_LFSCK_DELAY2 0x1601
773 do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1601
774 $START_NAMESPACE -r || error "(2) Fail to start LFSCK for namespace!"
776 local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
777 [ "$STATUS" == "scanning-phase1" ] ||
778 error "(3) Expect 'scanning-phase1', but got '$STATUS'"
780 # Sleep 3 sec to guarantee at least one object processed by LFSCK
782 echo "stop $SINGLEMDS"
783 stop $SINGLEMDS > /dev/null || error "(4) Fail to stop MDS!"
785 do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
786 echo "start $SINGLEMDS"
787 start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
788 error "(5) Fail to start MDS!"
790 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
791 mdd.${MDT_DEV}.lfsck_namespace |
792 awk '/^status/ { print \\\$2 }'" "completed" 30 || {
794 error "(6) unexpected status"
797 run_test 7a "non-stopped LFSCK should auto restarts after MDS remount (1)"
803 #define OBD_FAIL_LFSCK_LINKEA_MORE 0x1604
804 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1604
805 for ((i = 0; i < 20; i++)); do
806 touch $DIR/$tdir/dummy${i}
809 #define OBD_FAIL_LFSCK_DELAY3 0x1602
810 do_facet $SINGLEMDS $LCTL set_param fail_val=1 fail_loc=0x1602
811 $START_NAMESPACE -r || error "(3) Fail to start LFSCK for namespace!"
812 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
813 mdd.${MDT_DEV}.lfsck_namespace |
814 awk '/^status/ { print \\\$2 }'" "scanning-phase2" 32 || {
816 error "(4) unexpected status"
820 echo "stop $SINGLEMDS"
821 stop $SINGLEMDS > /dev/null || error "(5) Fail to stop MDS!"
823 do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
824 echo "start $SINGLEMDS"
825 start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
826 error "(6) Fail to start MDS!"
828 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
829 mdd.${MDT_DEV}.lfsck_namespace |
830 awk '/^status/ { print \\\$2 }'" "completed" 30 || {
832 error "(7) unexpected status"
835 run_test 7b "non-stopped LFSCK should auto restarts after MDS remount (2)"
840 formatall > /dev/null
846 local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
847 [ "$STATUS" == "init" ] ||
848 error "(2) Expect 'init', but got '$STATUS'"
850 #define OBD_FAIL_LFSCK_LINKEA_CRASH 0x1603
851 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1603
852 mkdir $DIR/$tdir/crashed
854 #define OBD_FAIL_LFSCK_LINKEA_MORE 0x1604
855 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1604
856 for ((i = 0; i < 5; i++)); do
857 touch $DIR/$tdir/dummy${i}
860 umount_client $MOUNT || error "(3) Fail to stop client!"
862 #define OBD_FAIL_LFSCK_DELAY2 0x1601
863 do_facet $SINGLEMDS $LCTL set_param fail_val=2 fail_loc=0x1601
864 $START_NAMESPACE || error "(4) Fail to start LFSCK for namespace!"
866 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
867 [ "$STATUS" == "scanning-phase1" ] ||
868 error "(5) Expect 'scanning-phase1', but got '$STATUS'"
870 $STOP_LFSCK || error "(6) Fail to stop LFSCK!"
872 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
873 [ "$STATUS" == "stopped" ] ||
874 error "(7) Expect 'stopped', but got '$STATUS'"
876 $START_NAMESPACE || error "(8) Fail to start LFSCK for namespace!"
878 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
879 [ "$STATUS" == "scanning-phase1" ] ||
880 error "(9) Expect 'scanning-phase1', but got '$STATUS'"
882 #define OBD_FAIL_LFSCK_FATAL2 0x1609
883 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x80001609
884 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
885 mdd.${MDT_DEV}.lfsck_namespace |
886 awk '/^status/ { print \\\$2 }'" "failed" 32 || {
888 error "(10) unexpected status"
891 #define OBD_FAIL_LFSCK_DELAY1 0x1600
892 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1600
893 $START_NAMESPACE || error "(11) Fail to start LFSCK for namespace!"
895 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
896 [ "$STATUS" == "scanning-phase1" ] ||
897 error "(12) Expect 'scanning-phase1', but got '$STATUS'"
899 #define OBD_FAIL_LFSCK_CRASH 0x160a
900 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160a
903 echo "stop $SINGLEMDS"
904 stop $SINGLEMDS > /dev/null || error "(13) Fail to stop MDS!"
906 #define OBD_FAIL_LFSCK_NO_AUTO 0x160b
907 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160b
909 echo "start $SINGLEMDS"
910 start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
911 error "(14) Fail to start MDS!"
913 local timeout=$(max_recovery_time)
916 while [ $timer -lt $timeout ]; do
917 STATUS=$(do_facet $SINGLEMDS "$LCTL get_param -n \
918 mdt.${MDT_DEV}.recovery_status |
919 awk '/^status/ { print \\\$2 }'")
920 [ "$STATUS" != "RECOVERING" ] && break;
925 [ $timer != $timeout ] ||
926 error "(14.1) recovery timeout"
928 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
929 [ "$STATUS" == "crashed" ] ||
930 error "(15) Expect 'crashed', but got '$STATUS'"
932 #define OBD_FAIL_LFSCK_DELAY2 0x1601
933 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1601
934 $START_NAMESPACE || error "(16) Fail to start LFSCK for namespace!"
936 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
937 [ "$STATUS" == "scanning-phase1" ] ||
938 error "(17) Expect 'scanning-phase1', but got '$STATUS'"
940 echo "stop $SINGLEMDS"
941 stop $SINGLEMDS > /dev/null || error "(18) Fail to stop MDS!"
943 #define OBD_FAIL_LFSCK_NO_AUTO 0x160b
944 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160b
946 echo "start $SINGLEMDS"
947 start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SCRUB > /dev/null ||
948 error "(19) Fail to start MDS!"
951 while [ $timer -lt $timeout ]; do
952 STATUS=$(do_facet $SINGLEMDS "$LCTL get_param -n \
953 mdt.${MDT_DEV}.recovery_status |
954 awk '/^status/ { print \\\$2 }'")
955 [ "$STATUS" != "RECOVERING" ] && break;
960 [ $timer != $timeout ] ||
961 error "(19.1) recovery timeout"
963 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
964 [ "$STATUS" == "paused" ] ||
965 error "(20) Expect 'paused', but got '$STATUS'"
967 echo "stop $SINGLEMDS"
968 stop $SINGLEMDS > /dev/null || error "(20.1) Fail to stop MDS!"
970 echo "start $SINGLEMDS without resume LFSCK"
971 start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_SKIP_LFSCK > /dev/null ||
972 error "(20.2) Fail to start MDS!"
975 while [ $timer -lt $timeout ]; do
976 STATUS=$(do_facet $SINGLEMDS "$LCTL get_param -n \
977 mdt.${MDT_DEV}.recovery_status |
978 awk '/^status/ { print \\\$2 }'")
979 [ "$STATUS" != "RECOVERING" ] && break;
984 [ $timer != $timeout ] ||
985 error "(20.3) recovery timeout"
987 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
988 [ "$STATUS" == "paused" ] ||
989 error "(20.4) Expect 'paused', but got '$STATUS'"
991 #define OBD_FAIL_LFSCK_DELAY3 0x1602
992 do_facet $SINGLEMDS $LCTL set_param fail_val=2 fail_loc=0x1602
994 $START_NAMESPACE || error "(21) Fail to start LFSCK for namespace!"
995 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
996 mdd.${MDT_DEV}.lfsck_namespace |
997 awk '/^status/ { print \\\$2 }'" "scanning-phase2" 32 || {
999 error "(22) unexpected status"
1002 local FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
1003 [ "$FLAGS" == "scanned-once,inconsistent" ] ||
1004 error "(23) Expect 'scanned-once,inconsistent',but got '$FLAGS'"
1006 do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
1007 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1008 mdd.${MDT_DEV}.lfsck_namespace |
1009 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1011 error "(24) unexpected status"
1014 FLAGS=$($SHOW_NAMESPACE | awk '/^flags/ { print $2 }')
1015 [ -z "$FLAGS" ] || error "(25) Expect empty flags, but got '$FLAGS'"
1017 run_test 8 "LFSCK state machine"
1020 if [ -z "$(grep "processor.*: 1" /proc/cpuinfo)" ]; then
1021 skip "Testing on UP system, the speed may be inaccurate."
1025 [[ $server_version -ge $(version_code 2.7.50) ]] ||
1026 { skip "Need MDS version >= 2.7.50"; return; }
1028 check_mount_and_prep
1029 $LFS mkdir -i 0 $DIR/$tdir/lfsck || error "(1) Fail to mkdir lfsck"
1030 $LFS setstripe -c 1 -i -1 $DIR/$tdir/lfsck
1031 createmany -o $DIR/$tdir/lfsck/f 5000
1033 local BASE_SPEED1=100
1035 $START_LAYOUT -r -s $BASE_SPEED1 || error "(2) Fail to start LFSCK!"
1038 STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
1039 [ "$STATUS" == "scanning-phase1" ] ||
1040 error "(3) Expect 'scanning-phase1', but got '$STATUS'"
1042 local SPEED=$($SHOW_LAYOUT |
1043 awk '/^average_speed_phase1/ { print $2 }')
1045 # There may be time error, normally it should be less than 2 seconds.
1046 # We allow another 20% schedule error.
1048 # MAX_MARGIN = 1.2 = 12 / 10
1049 local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) / \
1050 RUN_TIME1 * 12 / 10))
1051 [ $SPEED -lt $MAX_SPEED ] ||
1052 error "(4) Got speed $SPEED, expected less than $MAX_SPEED"
1054 # adjust speed limit
1055 local BASE_SPEED2=300
1057 do_facet $SINGLEMDS \
1058 $LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit $BASE_SPEED2
1061 SPEED=$($SHOW_LAYOUT | awk '/^average_speed_phase1/ { print $2 }')
1062 # MIN_MARGIN = 0.8 = 8 / 10
1063 local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
1064 BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
1065 (RUN_TIME1 + RUN_TIME2) * 8 / 10))
1066 [ $SPEED -gt $MIN_SPEED ] || {
1067 if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
1068 error_ignore LU-5624 \
1069 "(5.1) Got speed $SPEED, expected more than $MIN_SPEED"
1072 "(5.2) Got speed $SPEED, expected more than $MIN_SPEED"
1076 # MAX_MARGIN = 1.2 = 12 / 10
1077 MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
1078 BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
1079 (RUN_TIME1 + RUN_TIME2) * 12 / 10))
1080 [ $SPEED -lt $MAX_SPEED ] ||
1081 error "(6) Got speed $SPEED, expected less than $MAX_SPEED"
1083 do_facet $SINGLEMDS \
1084 $LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
1086 wait_update_facet $SINGLEMDS \
1087 "$LCTL get_param -n mdd.${MDT_DEV}.lfsck_layout |
1088 awk '/^status/ { print \\\$2 }'" "completed" 30 ||
1089 error "(7) Failed to get expected 'completed'"
1091 run_test 9a "LFSCK speed control (1)"
1094 if [ -z "$(grep "processor.*: 1" /proc/cpuinfo)" ]; then
1095 skip "Testing on UP system, the speed may be inaccurate."
1099 [[ $server_version -ge $(version_code 2.7.50) ]] ||
1100 { skip "Need MDS version >= 2.7.50"; return; }
1104 echo "Preparing another 50 * 50 files (with error) at $(date)."
1105 #define OBD_FAIL_LFSCK_LINKEA_MORE 0x1604
1106 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1604
1107 createmany -d $DIR/$tdir/d 50
1108 createmany -m $DIR/$tdir/f 50
1109 for ((i = 0; i < 50; i++)); do
1110 createmany -m $DIR/$tdir/d${i}/f 50 > /dev/null
1113 #define OBD_FAIL_LFSCK_NO_DOUBLESCAN 0x160c
1114 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160c
1115 $START_NAMESPACE -r || error "(4) Fail to start LFSCK!"
1116 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1117 mdd.${MDT_DEV}.lfsck_namespace |
1118 awk '/^status/ { print \\\$2 }'" "stopped" 10 || {
1120 error "(5) unexpected status"
1123 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
1124 echo "Prepared at $(date)."
1126 local BASE_SPEED1=50
1128 $START_NAMESPACE -s $BASE_SPEED1 || error "(6) Fail to start LFSCK!"
1131 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
1132 [ "$STATUS" == "scanning-phase2" ] ||
1133 error "(7) Expect 'scanning-phase2', but got '$STATUS'"
1135 local SPEED=$($SHOW_NAMESPACE |
1136 awk '/^average_speed_phase2/ { print $2 }')
1137 # There may be time error, normally it should be less than 2 seconds.
1138 # We allow another 20% schedule error.
1140 # MAX_MARGIN = 1.2 = 12 / 10
1141 local MAX_SPEED=$((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) / \
1142 RUN_TIME1 * 12 / 10))
1143 [ $SPEED -lt $MAX_SPEED ] ||
1144 error "(8) Got speed $SPEED, expected less than $MAX_SPEED"
1146 # adjust speed limit
1147 local BASE_SPEED2=150
1149 do_facet $SINGLEMDS \
1150 $LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit $BASE_SPEED2
1153 SPEED=$($SHOW_NAMESPACE | awk '/^average_speed_phase2/ { print $2 }')
1154 # MIN_MARGIN = 0.8 = 8 / 10
1155 local MIN_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 - TIME_DIFF) + \
1156 BASE_SPEED2 * (RUN_TIME2 - TIME_DIFF)) / \
1157 (RUN_TIME1 + RUN_TIME2) * 8 / 10))
1158 [ $SPEED -gt $MIN_SPEED ] || {
1159 if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
1160 error_ignore LU-5624 \
1161 "(9.1) Got speed $SPEED, expected more than $MIN_SPEED"
1164 "(9.2) Got speed $SPEED, expected more than $MIN_SPEED"
1168 # MAX_MARGIN = 1.2 = 12 / 10
1169 MAX_SPEED=$(((BASE_SPEED1 * (RUN_TIME1 + TIME_DIFF) + \
1170 BASE_SPEED2 * (RUN_TIME2 + TIME_DIFF)) / \
1171 (RUN_TIME1 + RUN_TIME2) * 12 / 10))
1172 [ $SPEED -lt $MAX_SPEED ] ||
1173 error "(10) Got speed $SPEED, expected less than $MAX_SPEED"
1175 do_facet $SINGLEMDS \
1176 $LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
1177 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1178 mdd.${MDT_DEV}.lfsck_namespace |
1179 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1181 error "(11) unexpected status"
1184 run_test 9b "LFSCK speed control (2)"
1188 [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
1189 skip "lookup(..)/linkea on ZFS issue" && return
1193 echo "Preparing more files with error at $(date)."
1194 #define OBD_FAIL_LFSCK_LINKEA_CRASH 0x1603
1195 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1603
1197 for ((i = 0; i < 1000; i = $((i+2)))); do
1198 mkdir -p $DIR/$tdir/d${i}
1199 touch $DIR/$tdir/f${i}
1200 createmany -m $DIR/$tdir/d${i}/f 5 > /dev/null
1203 #define OBD_FAIL_LFSCK_LINKEA_MORE 0x1604
1204 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1604
1206 for ((i = 1; i < 1000; i = $((i+2)))); do
1207 mkdir -p $DIR/$tdir/d${i}
1208 touch $DIR/$tdir/f${i}
1209 createmany -m $DIR/$tdir/d${i}/f 5 > /dev/null
1212 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
1213 echo "Prepared at $(date)."
1215 ln $DIR/$tdir/f200 $DIR/$tdir/d200/dummy
1217 umount_client $MOUNT
1218 mount_client $MOUNT || error "(3) Fail to start client!"
1220 $START_NAMESPACE -r -s 100 || error "(5) Fail to start LFSCK!"
1223 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
1224 [ "$STATUS" == "scanning-phase1" ] ||
1225 error "(6) Expect 'scanning-phase1', but got '$STATUS'"
1227 ls -ailR $MOUNT > /dev/null || error "(7) Fail to ls!"
1229 touch $DIR/$tdir/d198/a0 || error "(8) Fail to touch!"
1231 mkdir $DIR/$tdir/d199/a1 || error "(9) Fail to mkdir!"
1233 unlink $DIR/$tdir/f200 || error "(10) Fail to unlink!"
1235 rm -rf $DIR/$tdir/d201 || error "(11) Fail to rmdir!"
1237 mv $DIR/$tdir/f202 $DIR/$tdir/d203/ || error "(12) Fail to rename!"
1239 ln $DIR/$tdir/f204 $DIR/$tdir/d205/a3 || error "(13) Fail to hardlink!"
1241 ln -s $DIR/$tdir/d206 $DIR/$tdir/d207/a4 ||
1242 error "(14) Fail to softlink!"
1244 STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
1245 [ "$STATUS" == "scanning-phase1" ] ||
1246 error "(15) Expect 'scanning-phase1', but got '$STATUS'"
1248 do_facet $SINGLEMDS \
1249 $LCTL set_param -n mdd.${MDT_DEV}.lfsck_speed_limit 0
1250 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1251 mdd.${MDT_DEV}.lfsck_namespace |
1252 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1254 error "(16) unexpected status"
1257 run_test 10 "System is available during LFSCK scanning"
1260 ost_remove_lastid() {
1263 local rcmd="do_facet ost${ost}"
1265 echo "remove LAST_ID on ost${ost}: idx=${idx}"
1267 # step 1: local mount
1268 mount_fstype ost${ost} || return 1
1269 # step 2: remove the specified LAST_ID
1270 ${rcmd} rm -fv $(facet_mntpt ost${ost})/O/${idx}/{LAST_ID,d0/0}
1272 unmount_fstype ost${ost} || return 2
1276 check_mount_and_prep
1277 $SETSTRIPE -c 1 -i 0 $DIR/$tdir
1278 createmany -o $DIR/$tdir/f 64 || error "(0) Fail to create 64 files."
1283 ost_remove_lastid 1 0 || error "(1) Fail to remove LAST_ID"
1285 start ost1 $(ostdevname 1) $MOUNT_OPTS_NOSCRUB > /dev/null ||
1286 error "(2) Fail to start ost1"
1288 #define OBD_FAIL_LFSCK_DELAY4 0x160e
1289 do_facet ost1 $LCTL set_param fail_val=3 fail_loc=0x160e
1291 echo "trigger LFSCK for layout on ost1 to rebuild the LAST_ID(s)"
1292 $START_LAYOUT_ON_OST -r || error "(4) Fail to start LFSCK on OST!"
1294 wait_update_facet ost1 "$LCTL get_param -n \
1295 obdfilter.${OST_DEV}.lfsck_layout |
1296 awk '/^flags/ { print \\\$2 }'" "crashed_lastid" 60 || {
1298 error "(5) unexpected status"
1301 do_facet ost1 $LCTL set_param fail_val=0 fail_loc=0
1303 wait_update_facet ost1 "$LCTL get_param -n \
1304 obdfilter.${OST_DEV}.lfsck_layout |
1305 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1307 error "(6) unexpected status"
1310 echo "the LAST_ID(s) should have been rebuilt"
1311 FLAGS=$($SHOW_LAYOUT_ON_OST | awk '/^flags/ { print $2 }')
1312 [ -z "$FLAGS" ] || error "(7) Expect empty flags, but got '$FLAGS'"
1314 run_test 11a "LFSCK can rebuild lost last_id"
1317 check_mount_and_prep
1318 $SETSTRIPE -c 1 -i 0 $DIR/$tdir
1320 echo "set fail_loc=0x160d to skip the updating LAST_ID on-disk"
1321 #define OBD_FAIL_LFSCK_SKIP_LASTID 0x160d
1322 do_facet ost1 $LCTL set_param fail_loc=0x160d
1324 local count=$(precreated_ost_obj_count 0 0)
1326 createmany -o $DIR/$tdir/f $((count + 32))
1328 local proc_path="${FSNAME}-OST0000-osc-MDT0000"
1329 local seq=$(do_facet mds1 $LCTL get_param -n \
1330 osp.${proc_path}.prealloc_last_seq)
1331 local lastid1=$(do_facet ost1 "lctl get_param -n \
1332 obdfilter.${ost1_svc}.last_id" | grep $seq |
1333 awk -F: '{ print $2 }')
1335 umount_client $MOUNT
1336 stop ost1 || error "(1) Fail to stop ost1"
1338 #define OBD_FAIL_OST_ENOSPC 0x215
1339 do_facet ost1 $LCTL set_param fail_loc=0x215
1341 start ost1 $(ostdevname 1) $OST_MOUNT_OPTS ||
1342 error "(2) Fail to start ost1"
1344 for ((i = 0; i < 60; i++)); do
1345 lastid2=$(do_facet ost1 "lctl get_param -n \
1346 obdfilter.${ost1_svc}.last_id" | grep $seq |
1347 awk -F: '{ print $2 }')
1348 [ ! -z $lastid2 ] && break;
1352 echo "the on-disk LAST_ID should be smaller than the expected one"
1353 [ $lastid1 -gt $lastid2 ] ||
1354 error "(4) expect lastid1 [ $lastid1 ] > lastid2 [ $lastid2 ]"
1356 echo "trigger LFSCK for layout on ost1 to rebuild the on-disk LAST_ID"
1357 $START_LAYOUT_ON_OST -r || error "(5) Fail to start LFSCK on OST!"
1359 wait_update_facet ost1 "$LCTL get_param -n \
1360 obdfilter.${OST_DEV}.lfsck_layout |
1361 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1363 error "(6) unexpected status"
1366 stop ost1 || error "(7) Fail to stop ost1"
1368 start ost1 $(ostdevname 1) $OST_MOUNT_OPTS ||
1369 error "(8) Fail to start ost1"
1371 echo "the on-disk LAST_ID should have been rebuilt"
1372 wait_update_facet ost1 "$LCTL get_param -n \
1373 obdfilter.${ost1_svc}.last_id | grep $seq |
1374 awk -F: '{ print \\\$2 }'" "$lastid1" 60 || {
1375 do_facet ost1 $LCTL get_param -n \
1376 obdfilter.${ost1_svc}.last_id
1377 error "(9) expect lastid1 $seq:$lastid1"
1380 do_facet ost1 $LCTL set_param fail_loc=0
1381 stopall || error "(10) Fail to stopall"
1383 run_test 11b "LFSCK can rebuild crashed last_id"
1386 [ $MDSCOUNT -lt 2 ] &&
1387 skip "We need at least 2 MDSes for test_12a" && return
1389 check_mount_and_prep
1390 for k in $(seq $MDSCOUNT); do
1391 $LFS mkdir -i $((k - 1)) $DIR/$tdir/${k}
1392 createmany -o $DIR/$tdir/${k}/f 100 ||
1393 error "(0) Fail to create 100 files."
1396 echo "Start namespace LFSCK on all targets by single command (-s 1)."
1397 do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -t namespace -A \
1398 -s 1 -r || error "(2) Fail to start LFSCK on all devices!"
1400 echo "All the LFSCK targets should be in 'scanning-phase1' status."
1401 wait_all_targets namespace scanning-phase1 3
1403 echo "Stop namespace LFSCK on all targets by single lctl command."
1404 do_facet mds1 $LCTL lfsck_stop -M ${FSNAME}-MDT0000 -A ||
1405 error "(4) Fail to stop LFSCK on all devices!"
1407 echo "All the LFSCK targets should be in 'stopped' status."
1408 wait_all_targets_blocked namespace stopped 5
1410 echo "Re-start namespace LFSCK on all targets by single command (-s 0)."
1411 do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -t namespace -A \
1412 -s 0 -r || error "(6) Fail to start LFSCK on all devices!"
1414 echo "All the LFSCK targets should be in 'completed' status."
1415 wait_all_targets_blocked namespace completed 7
1417 start_full_debug_logging
1419 echo "Start layout LFSCK on all targets by single command (-s 1)."
1420 do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -t layout -A \
1421 -s 1 -r || error "(8) Fail to start LFSCK on all devices!"
1423 echo "All the LFSCK targets should be in 'scanning-phase1' status."
1424 wait_all_targets layout scanning-phase1 9
1426 echo "Stop layout LFSCK on all targets by single lctl command."
1427 do_facet mds1 $LCTL lfsck_stop -M ${FSNAME}-MDT0000 -A ||
1428 error "(10) Fail to stop LFSCK on all devices!"
1430 echo "All the LFSCK targets should be in 'stopped' status."
1431 wait_all_targets_blocked layout stopped 11
1433 for k in $(seq $OSTCOUNT); do
1434 local STATUS=$(do_facet ost${k} $LCTL get_param -n \
1435 obdfilter.$(facet_svc ost${k}).lfsck_layout |
1436 awk '/^status/ { print $2 }')
1437 [ "$STATUS" == "stopped" ] ||
1438 error "(12) OST${k} Expect 'stopped', but got '$STATUS'"
1441 echo "Re-start layout LFSCK on all targets by single command (-s 0)."
1442 do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -t layout -A \
1443 -s 0 -r || error "(13) Fail to start LFSCK on all devices!"
1445 echo "All the LFSCK targets should be in 'completed' status."
1446 wait_all_targets_blocked layout completed 14
1448 stop_full_debug_logging
1450 run_test 12a "single command to trigger LFSCK on all devices"
1453 check_mount_and_prep
1455 echo "Start LFSCK without '-M' specified."
1456 do_facet mds1 $LCTL lfsck_start -A -r ||
1457 error "(0) Fail to start LFSCK without '-M'"
1459 wait_all_targets_blocked namespace completed 1
1460 wait_all_targets_blocked layout completed 2
1462 local count=$(do_facet mds1 $LCTL dl |
1463 awk '{ print $3 }' | grep mdt | wc -l)
1464 if [ $count -gt 1 ]; then
1466 echo "Start layout LFSCK on the node with multipe targets,"
1467 echo "but not specify '-M'/'-A' option. Should get failure."
1469 do_facet mds1 $LCTL lfsck_start -t layout -r &&
1470 error "(3) Start layout LFSCK should fail" || true
1473 run_test 12b "auto detect Lustre device"
1477 echo "The lmm_oi in layout EA should be consistent with the MDT-object"
1478 echo "FID; otherwise, the LFSCK should re-generate the lmm_oi from the"
1479 echo "MDT-object FID."
1482 check_mount_and_prep
1484 echo "Inject failure stub to simulate bad lmm_oi"
1485 #define OBD_FAIL_LFSCK_BAD_LMMOI 0x160f
1486 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x160f
1487 createmany -o $DIR/$tdir/f 32
1488 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
1490 echo "Trigger layout LFSCK to find out the bad lmm_oi and fix them"
1491 $START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
1493 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1494 mdd.${MDT_DEV}.lfsck_layout |
1495 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1497 error "(2) unexpected status"
1500 local repaired=$($SHOW_LAYOUT |
1501 awk '/^repaired_others/ { print $2 }')
1502 [ $repaired -eq 32 ] ||
1503 error "(3) Fail to repair crashed lmm_oi: $repaired"
1505 run_test 13 "LFSCK can repair crashed lmm_oi"
1509 echo "The OST-object referenced by the MDT-object should be there;"
1510 echo "otherwise, the LFSCK should re-create the missing OST-object."
1513 check_mount_and_prep
1514 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1516 echo "Inject failure stub to simulate dangling referenced MDT-object"
1517 #define OBD_FAIL_LFSCK_DANGLING 0x1610
1518 do_facet ost1 $LCTL set_param fail_loc=0x1610
1519 local count=$(precreated_ost_obj_count 0 0)
1521 createmany -o $DIR/$tdir/f $((count + 31))
1522 touch $DIR/$tdir/guard
1523 do_facet ost1 $LCTL set_param fail_loc=0
1525 start_full_debug_logging
1527 # exhaust other pre-created dangling cases
1528 count=$(precreated_ost_obj_count 0 0)
1529 createmany -o $DIR/$tdir/a $count ||
1530 error "(0) Fail to create $count files."
1532 echo "'ls' should fail because of dangling referenced MDT-object"
1533 ls -ail $DIR/$tdir > /dev/null 2>&1 && error "(1) ls should fail."
1535 echo "Trigger layout LFSCK to find out dangling reference"
1536 $START_LAYOUT -r || error "(2) Fail to start LFSCK for layout!"
1538 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1539 mdd.${MDT_DEV}.lfsck_layout |
1540 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1542 error "(3) unexpected status"
1545 local repaired=$($SHOW_LAYOUT |
1546 awk '/^repaired_dangling/ { print $2 }')
1547 [ $repaired -ge 32 ] ||
1548 error "(4) Fail to repair dangling reference: $repaired"
1550 echo "'stat' should fail because of not repair dangling by default"
1551 stat $DIR/$tdir/guard > /dev/null 2>&1 && error "(5) stat should fail"
1553 echo "Trigger layout LFSCK to repair dangling reference"
1554 $START_LAYOUT -r -c || error "(6) Fail to start LFSCK for layout!"
1556 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1557 mdd.${MDT_DEV}.lfsck_layout |
1558 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1560 error "(7) unexpected status"
1563 # There may be some async LFSCK updates in processing, wait for
1564 # a while until the target reparation has been done. LU-4970.
1566 echo "'stat' should success after layout LFSCK repairing"
1567 wait_update_facet client "stat $DIR/$tdir/guard |
1568 awk '/Size/ { print \\\$2 }'" "0" 32 || {
1569 stat $DIR/$tdir/guard
1571 error "(8) unexpected size"
1574 repaired=$($SHOW_LAYOUT |
1575 awk '/^repaired_dangling/ { print $2 }')
1576 [ $repaired -ge 32 ] ||
1577 error "(9) Fail to repair dangling reference: $repaired"
1579 stop_full_debug_logging
1581 run_test 14 "LFSCK can repair MDT-object with dangling reference"
1585 echo "If the OST-object referenced by the MDT-object back points"
1586 echo "to some non-exist MDT-object, then the LFSCK should repair"
1587 echo "the OST-object to back point to the right MDT-object."
1590 check_mount_and_prep
1591 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1593 echo "Inject failure stub to make the OST-object to back point to"
1594 echo "non-exist MDT-object."
1595 #define OBD_FAIL_LFSCK_UNMATCHED_PAIR1 0x1611
1597 do_facet ost1 $LCTL set_param fail_loc=0x1611
1598 dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1
1599 cancel_lru_locks osc
1600 do_facet ost1 $LCTL set_param fail_loc=0
1602 echo "Trigger layout LFSCK to find out unmatched pairs and fix them"
1603 $START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
1605 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1606 mdd.${MDT_DEV}.lfsck_layout |
1607 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1609 error "(2) unexpected status"
1612 local repaired=$($SHOW_LAYOUT |
1613 awk '/^repaired_unmatched_pair/ { print $2 }')
1614 [ $repaired -eq 1 ] ||
1615 error "(3) Fail to repair unmatched pair: $repaired"
1617 run_test 15a "LFSCK can repair unmatched MDT-object/OST-object pairs (1)"
1621 echo "If the OST-object referenced by the MDT-object back points"
1622 echo "to other MDT-object that doesn't recognize the OST-object,"
1623 echo "then the LFSCK should repair it to back point to the right"
1624 echo "MDT-object (the first one)."
1627 check_mount_and_prep
1628 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1629 dd if=/dev/zero of=$DIR/$tdir/guard bs=1M count=1
1630 cancel_lru_locks osc
1632 echo "Inject failure stub to make the OST-object to back point to"
1633 echo "other MDT-object"
1635 #define OBD_FAIL_LFSCK_UNMATCHED_PAIR2 0x1612
1636 do_facet ost1 $LCTL set_param fail_loc=0x1612
1637 dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1
1638 cancel_lru_locks osc
1639 do_facet ost1 $LCTL set_param fail_loc=0
1641 echo "Trigger layout LFSCK to find out unmatched pairs and fix them"
1642 $START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
1644 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1645 mdd.${MDT_DEV}.lfsck_layout |
1646 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1648 error "(2) unexpected status"
1651 local repaired=$($SHOW_LAYOUT |
1652 awk '/^repaired_unmatched_pair/ { print $2 }')
1653 [ $repaired -eq 1 ] ||
1654 error "(3) Fail to repair unmatched pair: $repaired"
1656 run_test 15b "LFSCK can repair unmatched MDT-object/OST-object pairs (2)"
1659 [ $MDSCOUNT -lt 2 ] &&
1660 skip "We need at least 2 MDSes for this test" && return
1662 [ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.55) ] &&
1663 skip "Skip the test after 2.7.55 see LU-6437" && return
1666 echo "According to current metadata migration implementation,"
1667 echo "before the old MDT-object is removed, both the new MDT-object"
1668 echo "and old MDT-object will reference the same LOV layout. Then if"
1669 echo "the layout LFSCK finds the new MDT-object by race, it will"
1670 echo "regard related OST-object(s) as multiple referenced case, and"
1671 echo "will try to create new OST-object(s) for the new MDT-object."
1672 echo "To avoid such trouble, the layout LFSCK needs to lock the old"
1673 echo "MDT-object before confirm the multiple referenced case."
1676 check_mount_and_prep
1677 $LFS mkdir -i 1 $DIR/$tdir/a1
1678 $LFS setstripe -c 1 -i 0 $DIR/$tdir/a1
1679 dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=1
1680 cancel_lru_locks osc
1682 echo "Inject failure stub on MDT1 to delay the migration"
1684 #define OBD_FAIL_MIGRATE_DELAY 0x1803
1685 do_facet mds2 $LCTL set_param fail_val=5 fail_loc=0x1803
1686 echo "Migrate $DIR/$tdir/a1 from MDT1 to MDT0 with delay"
1687 $LFS migrate -m 0 $DIR/$tdir/a1 &
1690 echo "Trigger layout LFSCK to race with the migration"
1691 $START_LAYOUT -A -r || error "(1) Fail to start layout LFSCK!"
1693 wait_all_targets_blocked layout completed 2
1695 do_facet mds2 $LCTL set_param fail_loc=0 fail_val=0
1696 local repaired=$($SHOW_LAYOUT |
1697 awk '/^repaired_unmatched_pair/ { print $2 }')
1698 [ $repaired -eq 1 ] ||
1699 error "(3) Fail to repair unmatched pair: $repaired"
1701 repaired=$($SHOW_LAYOUT |
1702 awk '/^repaired_multiple_referenced/ { print $2 }')
1703 [ $repaired -eq 0 ] ||
1704 error "(4) Unexpectedly repaird multiple references: $repaired"
1706 run_test 15c "LFSCK can repair unmatched MDT-object/OST-object pairs (3)"
1710 echo "If the OST-object's owner information does not match the owner"
1711 echo "information stored in the MDT-object, then the LFSCK trust the"
1712 echo "MDT-object and update the OST-object's owner information."
1715 check_mount_and_prep
1716 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1717 dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=1
1718 cancel_lru_locks osc
1720 echo "Inject failure stub to skip OST-object owner changing"
1721 #define OBD_FAIL_LFSCK_BAD_OWNER 0x1613
1722 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1613
1723 chown 1.1 $DIR/$tdir/f0
1724 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
1726 echo "Trigger layout LFSCK to find out inconsistent OST-object owner"
1729 $START_LAYOUT -r || error "(1) Fail to start LFSCK for layout!"
1731 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1732 mdd.${MDT_DEV}.lfsck_layout |
1733 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1735 error "(2) unexpected status"
1738 local repaired=$($SHOW_LAYOUT |
1739 awk '/^repaired_inconsistent_owner/ { print $2 }')
1740 [ $repaired -eq 1 ] ||
1741 error "(3) Fail to repair inconsistent owner: $repaired"
1743 run_test 16 "LFSCK can repair inconsistent MDT-object/OST-object owner"
1747 echo "If more than one MDT-objects reference the same OST-object,"
1748 echo "and the OST-object only recognizes one MDT-object, then the"
1749 echo "LFSCK should create new OST-objects for such non-recognized"
1753 check_mount_and_prep
1754 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1756 echo "Inject failure stub to make two MDT-objects to refernce"
1757 echo "the OST-object"
1759 #define OBD_FAIL_LFSCK_MULTIPLE_REF 0x1614
1760 do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0x1614
1762 dd if=/dev/zero of=$DIR/$tdir/guard bs=1M count=1
1763 cancel_lru_locks osc
1765 createmany -o $DIR/$tdir/f 1
1767 do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
1769 cancel_lru_locks mdc
1770 cancel_lru_locks osc
1772 echo "$DIR/$tdir/f0 and $DIR/$tdir/guard use the same OST-objects"
1773 local size=$(ls -l $DIR/$tdir/f0 | awk '{ print $5 }')
1774 [ $size -eq 1048576 ] ||
1775 error "(1) f0 (wrong) size should be 1048576, but got $size"
1777 echo "Trigger layout LFSCK to find out multiple refenced MDT-objects"
1780 $START_LAYOUT -r || error "(2) Fail to start LFSCK for layout!"
1782 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
1783 mdd.${MDT_DEV}.lfsck_layout |
1784 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
1786 error "(3) unexpected status"
1789 local repaired=$($SHOW_LAYOUT |
1790 awk '/^repaired_multiple_referenced/ { print $2 }')
1791 [ $repaired -eq 1 ] ||
1792 error "(4) Fail to repair multiple references: $repaired"
1794 echo "$DIR/$tdir/f0 and $DIR/$tdir/guard should use diff OST-objects"
1795 dd if=/dev/zero of=$DIR/$tdir/f0 bs=1M count=2 ||
1796 error "(5) Fail to write f0."
1797 size=$(ls -l $DIR/$tdir/guard | awk '{ print $5 }')
1798 [ $size -eq 1048576 ] ||
1799 error "(6) guard size should be 1048576, but got $size"
1801 run_test 17 "LFSCK can repair multiple references"
1803 $LCTL set_param debug=+cache > /dev/null
1807 echo "The target MDT-object is there, but related stripe information"
1808 echo "is lost or partly lost. The LFSCK should regenerate the missing"
1809 echo "layout EA entries."
1812 check_mount_and_prep
1813 $LFS mkdir -i 0 $DIR/$tdir/a1
1814 $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
1815 dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
1817 local saved_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
1819 $LFS path2fid $DIR/$tdir/a1/f1
1820 $LFS getstripe $DIR/$tdir/a1/f1
1822 if [ $MDSCOUNT -ge 2 ]; then
1823 $LFS mkdir -i 1 $DIR/$tdir/a2
1824 $LFS setstripe -c 2 -i 1 -S 1M $DIR/$tdir/a2
1825 dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
1826 $LFS path2fid $DIR/$tdir/a2/f2
1827 $LFS getstripe $DIR/$tdir/a2/f2
1830 cancel_lru_locks osc
1832 echo "Inject failure, to make the MDT-object lost its layout EA"
1833 #define OBD_FAIL_LFSCK_LOST_STRIPE 0x1615
1834 do_facet mds1 $LCTL set_param fail_loc=0x1615
1835 chown 1.1 $DIR/$tdir/a1/f1
1837 if [ $MDSCOUNT -ge 2 ]; then
1838 do_facet mds2 $LCTL set_param fail_loc=0x1615
1839 chown 1.1 $DIR/$tdir/a2/f2
1845 do_facet mds1 $LCTL set_param fail_loc=0
1846 if [ $MDSCOUNT -ge 2 ]; then
1847 do_facet mds2 $LCTL set_param fail_loc=0
1850 cancel_lru_locks mdc
1851 cancel_lru_locks osc
1853 echo "The file size should be incorrect since layout EA is lost"
1854 local cur_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
1855 [ "$cur_size" != "$saved_size" ] ||
1856 error "(1) Expect incorrect file1 size"
1858 if [ $MDSCOUNT -ge 2 ]; then
1859 cur_size=$(ls -il $DIR/$tdir/a2/f2 | awk '{ print $6 }')
1860 [ "$cur_size" != "$saved_size" ] ||
1861 error "(2) Expect incorrect file2 size"
1864 echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
1865 $START_LAYOUT -r -o || error "(3) Fail to start LFSCK for layout!"
1867 for k in $(seq $MDSCOUNT); do
1868 # The LFSCK status query internal is 30 seconds. For the case
1869 # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
1870 # time to guarantee the status sync up.
1871 wait_update_facet mds${k} "$LCTL get_param -n \
1872 mdd.$(facet_svc mds${k}).lfsck_layout |
1873 awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
1874 error "(4) MDS${k} is not the expected 'completed'"
1877 for k in $(seq $OSTCOUNT); do
1878 local cur_status=$(do_facet ost${k} $LCTL get_param -n \
1879 obdfilter.$(facet_svc ost${k}).lfsck_layout |
1880 awk '/^status/ { print $2 }')
1881 [ "$cur_status" == "completed" ] ||
1882 error "(5) OST${k} Expect 'completed', but got '$cur_status'"
1885 local repaired=$(do_facet mds1 $LCTL get_param -n \
1886 mdd.$(facet_svc mds1).lfsck_layout |
1887 awk '/^repaired_orphan/ { print $2 }')
1888 [ $repaired -eq 1 ] ||
1889 error "(6.1) Expect 1 fixed on mds1, but got: $repaired"
1891 if [ $MDSCOUNT -ge 2 ]; then
1892 repaired=$(do_facet mds2 $LCTL get_param -n \
1893 mdd.$(facet_svc mds2).lfsck_layout |
1894 awk '/^repaired_orphan/ { print $2 }')
1895 [ $repaired -eq 2 ] ||
1896 error "(6.2) Expect 2 fixed on mds2, but got: $repaired"
1899 $LFS path2fid $DIR/$tdir/a1/f1
1900 $LFS getstripe $DIR/$tdir/a1/f1
1902 if [ $MDSCOUNT -ge 2 ]; then
1903 $LFS path2fid $DIR/$tdir/a2/f2
1904 $LFS getstripe $DIR/$tdir/a2/f2
1907 echo "The file size should be correct after layout LFSCK scanning"
1908 cur_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
1909 [ "$cur_size" == "$saved_size" ] ||
1910 error "(7) Expect file1 size $saved_size, but got $cur_size"
1912 if [ $MDSCOUNT -ge 2 ]; then
1913 cur_size=$(ls -il $DIR/$tdir/a2/f2 | awk '{ print $6 }')
1914 [ "$cur_size" == "$saved_size" ] ||
1915 error "(8) Expect file2 size $saved_size, but got $cur_size"
1918 run_test 18a "Find out orphan OST-object and repair it (1)"
1922 echo "The target MDT-object is lost. The LFSCK should re-create the"
1923 echo "MDT-object under .lustre/lost+found/MDTxxxx. The admin should"
1924 echo "can move it back to normal namespace manually."
1927 check_mount_and_prep
1928 $LFS mkdir -i 0 $DIR/$tdir/a1
1929 $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
1930 dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
1931 local saved_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
1932 local fid1=$($LFS path2fid $DIR/$tdir/a1/f1)
1934 $LFS getstripe $DIR/$tdir/a1/f1
1936 if [ $MDSCOUNT -ge 2 ]; then
1937 $LFS mkdir -i 1 $DIR/$tdir/a2
1938 $LFS setstripe -c 2 -i 1 -S 1M $DIR/$tdir/a2
1939 dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
1940 fid2=$($LFS path2fid $DIR/$tdir/a2/f2)
1942 $LFS getstripe $DIR/$tdir/a2/f2
1945 cancel_lru_locks osc
1947 echo "Inject failure, to simulate the case of missing the MDT-object"
1948 #define OBD_FAIL_LFSCK_LOST_MDTOBJ 0x1616
1949 do_facet mds1 $LCTL set_param fail_loc=0x1616
1950 rm -f $DIR/$tdir/a1/f1
1952 if [ $MDSCOUNT -ge 2 ]; then
1953 do_facet mds2 $LCTL set_param fail_loc=0x1616
1954 rm -f $DIR/$tdir/a2/f2
1960 do_facet mds1 $LCTL set_param fail_loc=0
1961 if [ $MDSCOUNT -ge 2 ]; then
1962 do_facet mds2 $LCTL set_param fail_loc=0
1965 cancel_lru_locks mdc
1966 cancel_lru_locks osc
1968 echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
1969 $START_LAYOUT -r -o || error "(1) Fail to start LFSCK for layout!"
1971 for k in $(seq $MDSCOUNT); do
1972 # The LFSCK status query internal is 30 seconds. For the case
1973 # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
1974 # time to guarantee the status sync up.
1975 wait_update_facet mds${k} "$LCTL get_param -n \
1976 mdd.$(facet_svc mds${k}).lfsck_layout |
1977 awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
1978 error "(2) MDS${k} is not the expected 'completed'"
1981 for k in $(seq $OSTCOUNT); do
1982 local cur_status=$(do_facet ost${k} $LCTL get_param -n \
1983 obdfilter.$(facet_svc ost${k}).lfsck_layout |
1984 awk '/^status/ { print $2 }')
1985 [ "$cur_status" == "completed" ] ||
1986 error "(3) OST${k} Expect 'completed', but got '$cur_status'"
1989 local repaired=$(do_facet mds1 $LCTL get_param -n \
1990 mdd.$(facet_svc mds1).lfsck_layout |
1991 awk '/^repaired_orphan/ { print $2 }')
1992 [ $repaired -eq 1 ] ||
1993 error "(4.1) Expect 1 fixed on mds1, but got: $repaired"
1995 if [ $MDSCOUNT -ge 2 ]; then
1996 repaired=$(do_facet mds2 $LCTL get_param -n \
1997 mdd.$(facet_svc mds2).lfsck_layout |
1998 awk '/^repaired_orphan/ { print $2 }')
1999 [ $repaired -eq 2 ] ||
2000 error "(4.2) Expect 2 fixed on mds2, but got: $repaired"
2003 echo "Move the files from ./lustre/lost+found/MDTxxxx to namespace"
2004 mv $MOUNT/.lustre/lost+found/MDT0000/${fid1}-R-0 $DIR/$tdir/a1/f1 ||
2005 error "(5) Fail to move $MOUNT/.lustre/lost+found/MDT0000/${fid1}-R-0"
2007 if [ $MDSCOUNT -ge 2 ]; then
2008 local name=$MOUNT/.lustre/lost+found/MDT0001/${fid2}-R-0
2009 mv $name $DIR/$tdir/a2/f2 || error "(6) Fail to move $name"
2012 $LFS path2fid $DIR/$tdir/a1/f1
2013 $LFS getstripe $DIR/$tdir/a1/f1
2015 if [ $MDSCOUNT -ge 2 ]; then
2016 $LFS path2fid $DIR/$tdir/a2/f2
2017 $LFS getstripe $DIR/$tdir/a2/f2
2020 echo "The file size should be correct after layout LFSCK scanning"
2021 local cur_size=$(ls -il $DIR/$tdir/a1/f1 | awk '{ print $6 }')
2022 [ "$cur_size" == "$saved_size" ] ||
2023 error "(7) Expect file1 size $saved_size, but got $cur_size"
2025 if [ $MDSCOUNT -ge 2 ]; then
2026 cur_size=$(ls -il $DIR/$tdir/a2/f2 | awk '{ print $6 }')
2027 [ "$cur_size" == "$saved_size" ] ||
2028 error "(8) Expect file2 size $saved_size, but got $cur_size"
2031 run_test 18b "Find out orphan OST-object and repair it (2)"
2035 echo "The target MDT-object is lost, and the OST-object FID is missing."
2036 echo "The LFSCK should re-create the MDT-object with new FID under the "
2037 echo "directory .lustre/lost+found/MDTxxxx."
2040 check_mount_and_prep
2041 $LFS mkdir -i 0 $DIR/$tdir/a1
2042 $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
2044 echo "Inject failure, to simulate the case of missing parent FID"
2045 #define OBD_FAIL_LFSCK_NOPFID 0x1617
2046 do_facet ost1 $LCTL set_param fail_loc=0x1617
2048 dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
2049 $LFS getstripe $DIR/$tdir/a1/f1
2051 if [ $MDSCOUNT -ge 2 ]; then
2052 $LFS mkdir -i 1 $DIR/$tdir/a2
2053 $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a2
2054 dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
2055 $LFS getstripe $DIR/$tdir/a2/f2
2058 cancel_lru_locks osc
2060 echo "Inject failure, to simulate the case of missing the MDT-object"
2061 #define OBD_FAIL_LFSCK_LOST_MDTOBJ 0x1616
2062 do_facet mds1 $LCTL set_param fail_loc=0x1616
2063 rm -f $DIR/$tdir/a1/f1
2065 if [ $MDSCOUNT -ge 2 ]; then
2066 do_facet mds2 $LCTL set_param fail_loc=0x1616
2067 rm -f $DIR/$tdir/a2/f2
2073 do_facet mds1 $LCTL set_param fail_loc=0
2074 if [ $MDSCOUNT -ge 2 ]; then
2075 do_facet mds2 $LCTL set_param fail_loc=0
2078 cancel_lru_locks mdc
2079 cancel_lru_locks osc
2081 echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
2082 $START_LAYOUT -r -o || error "(1) Fail to start LFSCK for layout!"
2084 for k in $(seq $MDSCOUNT); do
2085 # The LFSCK status query internal is 30 seconds. For the case
2086 # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
2087 # time to guarantee the status sync up.
2088 wait_update_facet mds${k} "$LCTL get_param -n \
2089 mdd.$(facet_svc mds${k}).lfsck_layout |
2090 awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
2091 error "(2) MDS${k} is not the expected 'completed'"
2094 for k in $(seq $OSTCOUNT); do
2095 local cur_status=$(do_facet ost${k} $LCTL get_param -n \
2096 obdfilter.$(facet_svc ost${k}).lfsck_layout |
2097 awk '/^status/ { print $2 }')
2098 [ "$cur_status" == "completed" ] ||
2099 error "(3) OST${k} Expect 'completed', but got '$cur_status'"
2102 if [ $MDSCOUNT -ge 2 ]; then
2108 local repaired=$(do_facet mds1 $LCTL get_param -n \
2109 mdd.$(facet_svc mds1).lfsck_layout |
2110 awk '/^repaired_orphan/ { print $2 }')
2111 [ $repaired -eq $expected ] ||
2112 error "(4) Expect $expected fixed on mds1, but got: $repaired"
2114 if [ $MDSCOUNT -ge 2 ]; then
2115 repaired=$(do_facet mds2 $LCTL get_param -n \
2116 mdd.$(facet_svc mds2).lfsck_layout |
2117 awk '/^repaired_orphan/ { print $2 }')
2118 [ $repaired -eq 0 ] ||
2119 error "(5) Expect 0 fixed on mds2, but got: $repaired"
2122 ls -ail $MOUNT/.lustre/lost+found/
2124 echo "There should NOT be some stub under .lustre/lost+found/MDT0001/"
2125 if [ -d $MOUNT/.lustre/lost+found/MDT0001 ]; then
2126 cname=$(find $MOUNT/.lustre/lost+found/MDT0001/ -name *-N-*)
2128 error "(6) .lustre/lost+found/MDT0001/ should be empty"
2131 echo "There should be some stub under .lustre/lost+found/MDT0000/"
2132 [ -d $MOUNT/.lustre/lost+found/MDT0000 ] ||
2133 error "(7) $MOUNT/.lustre/lost+found/MDT0000/ should be there"
2135 cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-N-*)
2136 [ ! -z "$cname" ] ||
2137 error "(8) .lustre/lost+found/MDT0000/ should not be empty"
2139 run_test 18c "Find out orphan OST-object and repair it (3)"
2143 echo "The target MDT-object layout EA slot is occpuied by some new"
2144 echo "created OST-object when repair dangling reference case. Such"
2145 echo "conflict OST-object has never been modified. Then when found"
2146 echo "the orphan OST-object, LFSCK will replace it with the orphan"
2150 check_mount_and_prep
2152 $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
2153 echo "guard" > $DIR/$tdir/a1/f1
2154 echo "foo" > $DIR/$tdir/a1/f2
2155 local saved_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
2156 $LFS path2fid $DIR/$tdir/a1/f1
2157 $LFS getstripe $DIR/$tdir/a1/f1
2158 $LFS path2fid $DIR/$tdir/a1/f2
2159 $LFS getstripe $DIR/$tdir/a1/f2
2160 cancel_lru_locks osc
2162 echo "Inject failure to make $DIR/$tdir/a1/f1 and $DIR/$tdir/a1/f2"
2163 echo "to reference the same OST-object (which is f1's OST-obejct)."
2164 echo "Then drop $DIR/$tdir/a1/f1 and its OST-object, so f2 becomes"
2165 echo "dangling reference case, but f2's old OST-object is there."
2168 #define OBD_FAIL_LFSCK_CHANGE_STRIPE 0x1618
2169 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1618
2170 chown 1.1 $DIR/$tdir/a1/f2
2171 rm -f $DIR/$tdir/a1/f1
2174 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
2176 echo "stopall to cleanup object cache"
2179 setupall > /dev/null
2181 echo "The file size should be incorrect since dangling referenced"
2182 local cur_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
2183 [ "$cur_size" != "$saved_size" ] ||
2184 error "(1) Expect incorrect file2 size"
2186 echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
2187 $START_LAYOUT -r -o -c || error "(2) Fail to start LFSCK for layout!"
2189 for k in $(seq $MDSCOUNT); do
2190 # The LFSCK status query internal is 30 seconds. For the case
2191 # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
2192 # time to guarantee the status sync up.
2193 wait_update_facet mds${k} "$LCTL get_param -n \
2194 mdd.$(facet_svc mds${k}).lfsck_layout |
2195 awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
2196 error "(3) MDS${k} is not the expected 'completed'"
2199 for k in $(seq $OSTCOUNT); do
2200 local cur_status=$(do_facet ost${k} $LCTL get_param -n \
2201 obdfilter.$(facet_svc ost${k}).lfsck_layout |
2202 awk '/^status/ { print $2 }')
2203 [ "$cur_status" == "completed" ] ||
2204 error "(4) OST${k} Expect 'completed', but got '$cur_status'"
2207 local repaired=$(do_facet $SINGLEMDS $LCTL get_param -n \
2208 mdd.$(facet_svc $SINGLEMDS).lfsck_layout |
2209 awk '/^repaired_orphan/ { print $2 }')
2210 [ $repaired -eq 1 ] ||
2211 error "(5) Expect 1 orphan has been fixed, but got: $repaired"
2213 echo "The file size should be correct after layout LFSCK scanning"
2214 cur_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
2215 [ "$cur_size" == "$saved_size" ] ||
2216 error "(6) Expect file2 size $saved_size, but got $cur_size"
2218 echo "The LFSCK should find back the original data."
2219 cat $DIR/$tdir/a1/f2
2220 $LFS path2fid $DIR/$tdir/a1/f2
2221 $LFS getstripe $DIR/$tdir/a1/f2
2223 run_test 18d "Find out orphan OST-object and repair it (4)"
2227 echo "The target MDT-object layout EA slot is occpuied by some new"
2228 echo "created OST-object when repair dangling reference case. Such"
2229 echo "conflict OST-object has been modified by others. To keep the"
2230 echo "new data, the LFSCK will create a new file to refernece this"
2231 echo "old orphan OST-object."
2234 check_mount_and_prep
2236 $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
2237 echo "guard" > $DIR/$tdir/a1/f1
2238 echo "foo" > $DIR/$tdir/a1/f2
2239 local saved_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
2240 $LFS path2fid $DIR/$tdir/a1/f1
2241 $LFS getstripe $DIR/$tdir/a1/f1
2242 $LFS path2fid $DIR/$tdir/a1/f2
2243 $LFS getstripe $DIR/$tdir/a1/f2
2244 cancel_lru_locks osc
2246 echo "Inject failure to make $DIR/$tdir/a1/f1 and $DIR/$tdir/a1/f2"
2247 echo "to reference the same OST-object (which is f1's OST-obejct)."
2248 echo "Then drop $DIR/$tdir/a1/f1 and its OST-object, so f2 becomes"
2249 echo "dangling reference case, but f2's old OST-object is there."
2252 #define OBD_FAIL_LFSCK_CHANGE_STRIPE 0x1618
2253 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1618
2254 chown 1.1 $DIR/$tdir/a1/f2
2255 rm -f $DIR/$tdir/a1/f1
2258 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
2260 echo "stopall to cleanup object cache"
2263 setupall > /dev/null
2265 echo "The file size should be incorrect since dangling referenced"
2266 local cur_size=$(ls -il $DIR/$tdir/a1/f2 | awk '{ print $6 }')
2267 [ "$cur_size" != "$saved_size" ] ||
2268 error "(1) Expect incorrect file2 size"
2270 #define OBD_FAIL_LFSCK_DELAY3 0x1602
2271 do_facet $SINGLEMDS $LCTL set_param fail_val=10 fail_loc=0x1602
2273 start_full_debug_logging
2275 echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
2276 $START_LAYOUT -r -o -c || error "(2) Fail to start LFSCK for layout!"
2278 wait_update_facet mds1 "$LCTL get_param -n \
2279 mdd.$(facet_svc mds1).lfsck_layout |
2280 awk '/^status/ { print \\\$2 }'" "scanning-phase2" $LTIME ||
2281 error "(3) MDS1 is not the expected 'scanning-phase2'"
2283 # to guarantee all updates are synced.
2287 echo "Write new data to f2 to modify the new created OST-object."
2288 echo "dummy" >> $DIR/$tdir/a1/f2
2290 do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
2292 for k in $(seq $MDSCOUNT); do
2293 # The LFSCK status query internal is 30 seconds. For the case
2294 # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
2295 # time to guarantee the status sync up.
2296 wait_update_facet mds${k} "$LCTL get_param -n \
2297 mdd.$(facet_svc mds${k}).lfsck_layout |
2298 awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
2299 error "(4) MDS${k} is not the expected 'completed'"
2302 for k in $(seq $OSTCOUNT); do
2303 local cur_status=$(do_facet ost${k} $LCTL get_param -n \
2304 obdfilter.$(facet_svc ost${k}).lfsck_layout |
2305 awk '/^status/ { print $2 }')
2306 [ "$cur_status" == "completed" ] ||
2307 error "(5) OST${k} Expect 'completed', but got '$cur_status'"
2310 stop_full_debug_logging
2312 local repaired=$(do_facet $SINGLEMDS $LCTL get_param -n \
2313 mdd.$(facet_svc $SINGLEMDS).lfsck_layout |
2314 awk '/^repaired_orphan/ { print $2 }')
2315 [ $repaired -eq 1 ] ||
2316 error "(6) Expect 1 orphan has been fixed, but got: $repaired"
2318 echo "There should be stub file under .lustre/lost+found/MDT0000/"
2319 [ -d $MOUNT/.lustre/lost+found/MDT0000 ] ||
2320 error "(7) $MOUNT/.lustre/lost+found/MDT0000/ should be there"
2322 cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-C-*)
2323 [ ! -z "$cname" ] ||
2324 error "(8) .lustre/lost+found/MDT0000/ should not be empty"
2326 echo "The stub file should keep the original f2 data"
2327 cur_size=$(ls -il $cname | awk '{ print $6 }')
2328 [ "$cur_size" == "$saved_size" ] ||
2329 error "(9) Expect file2 size $saved_size, but got $cur_size"
2332 $LFS path2fid $cname
2333 $LFS getstripe $cname
2335 echo "The f2 should contains new data."
2336 cat $DIR/$tdir/a1/f2
2337 $LFS path2fid $DIR/$tdir/a1/f2
2338 $LFS getstripe $DIR/$tdir/a1/f2
2340 run_test 18e "Find out orphan OST-object and repair it (5)"
2343 [ $OSTCOUNT -lt 2 ] &&
2344 skip "The test needs at least 2 OSTs" && return
2347 echo "The target MDT-object is lost. The LFSCK should re-create the"
2348 echo "MDT-object under .lustre/lost+found/MDTxxxx. If some OST fail"
2349 echo "to verify some OST-object(s) during the first stage-scanning,"
2350 echo "the LFSCK should skip orphan OST-objects for such OST. Others"
2351 echo "should not be affected."
2354 check_mount_and_prep
2355 $LFS mkdir -i 0 $DIR/$tdir/a1
2356 $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a1
2357 dd if=/dev/zero of=$DIR/$tdir/a1/guard bs=1M count=2
2358 dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=2
2359 $LFS mkdir -i 0 $DIR/$tdir/a2
2360 $LFS setstripe -c 2 -i 0 -S 1M $DIR/$tdir/a2
2361 dd if=/dev/zero of=$DIR/$tdir/a2/f2 bs=1M count=2
2362 $LFS getstripe $DIR/$tdir/a1/f1
2363 $LFS getstripe $DIR/$tdir/a2/f2
2365 if [ $MDSCOUNT -ge 2 ]; then
2366 $LFS mkdir -i 1 $DIR/$tdir/a3
2367 $LFS setstripe -c 1 -i 0 -S 1M $DIR/$tdir/a3
2368 dd if=/dev/zero of=$DIR/$tdir/a3/guard bs=1M count=2
2369 dd if=/dev/zero of=$DIR/$tdir/a3/f3 bs=1M count=2
2370 $LFS mkdir -i 1 $DIR/$tdir/a4
2371 $LFS setstripe -c 2 -i 0 -S 1M $DIR/$tdir/a4
2372 dd if=/dev/zero of=$DIR/$tdir/a4/f4 bs=1M count=2
2373 $LFS getstripe $DIR/$tdir/a3/f3
2374 $LFS getstripe $DIR/$tdir/a4/f4
2377 cancel_lru_locks osc
2379 echo "Inject failure, to simulate the case of missing the MDT-object"
2380 #define OBD_FAIL_LFSCK_LOST_MDTOBJ 0x1616
2381 do_facet mds1 $LCTL set_param fail_loc=0x1616
2382 rm -f $DIR/$tdir/a1/f1
2383 rm -f $DIR/$tdir/a2/f2
2385 if [ $MDSCOUNT -ge 2 ]; then
2386 do_facet mds2 $LCTL set_param fail_loc=0x1616
2387 rm -f $DIR/$tdir/a3/f3
2388 rm -f $DIR/$tdir/a4/f4
2394 do_facet mds1 $LCTL set_param fail_loc=0
2395 if [ $MDSCOUNT -ge 2 ]; then
2396 do_facet mds2 $LCTL set_param fail_loc=0
2399 cancel_lru_locks mdc
2400 cancel_lru_locks osc
2402 echo "Inject failure, to simulate the OST0 fail to handle"
2403 echo "MDT0 LFSCK request during the first-stage scanning."
2404 #define OBD_FAIL_LFSCK_BAD_NETWORK 0x161c
2405 do_facet mds1 $LCTL set_param fail_loc=0x161c fail_val=0
2407 echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
2408 $START_LAYOUT -r -o || error "(1) Fail to start LFSCK for layout!"
2410 for k in $(seq $MDSCOUNT); do
2411 # The LFSCK status query internal is 30 seconds. For the case
2412 # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
2413 # time to guarantee the status sync up.
2414 wait_update_facet mds${k} "$LCTL get_param -n \
2415 mdd.$(facet_svc mds${k}).lfsck_layout |
2416 awk '/^status/ { print \\\$2 }'" "partial" $LTIME ||
2417 error "(2) MDS${k} is not the expected 'partial'"
2420 wait_update_facet ost1 "$LCTL get_param -n \
2421 obdfilter.$(facet_svc ost1).lfsck_layout |
2422 awk '/^status/ { print \\\$2 }'" "partial" $LTIME || {
2423 error "(3) OST1 is not the expected 'partial'"
2426 wait_update_facet ost2 "$LCTL get_param -n \
2427 obdfilter.$(facet_svc ost2).lfsck_layout |
2428 awk '/^status/ { print \\\$2 }'" "completed" $LTIME || {
2429 error "(4) OST2 is not the expected 'completed'"
2432 do_facet mds1 $LCTL set_param fail_loc=0 fail_val=0
2434 local repaired=$(do_facet mds1 $LCTL get_param -n \
2435 mdd.$(facet_svc mds1).lfsck_layout |
2436 awk '/^repaired_orphan/ { print $2 }')
2437 [ $repaired -eq 1 ] ||
2438 error "(5) Expect 1 fixed on mds{1}, but got: $repaired"
2440 if [ $MDSCOUNT -ge 2 ]; then
2441 repaired=$(do_facet mds2 $LCTL get_param -n \
2442 mdd.$(facet_svc mds2).lfsck_layout |
2443 awk '/^repaired_orphan/ { print $2 }')
2444 [ $repaired -eq 1 ] ||
2445 error "(6) Expect 1 fixed on mds{2}, but got: $repaired"
2448 echo "Trigger layout LFSCK on all devices again to cleanup"
2449 $START_LAYOUT -r -o || error "(7) Fail to start LFSCK for layout!"
2451 for k in $(seq $MDSCOUNT); do
2452 # The LFSCK status query internal is 30 seconds. For the case
2453 # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
2454 # time to guarantee the status sync up.
2455 wait_update_facet mds${k} "$LCTL get_param -n \
2456 mdd.$(facet_svc mds${k}).lfsck_layout |
2457 awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
2458 error "(8) MDS${k} is not the expected 'completed'"
2461 for k in $(seq $OSTCOUNT); do
2462 cur_status=$(do_facet ost${k} $LCTL get_param -n \
2463 obdfilter.$(facet_svc ost${k}).lfsck_layout |
2464 awk '/^status/ { print $2 }')
2465 [ "$cur_status" == "completed" ] ||
2466 error "(9) OST${k} Expect 'completed', but got '$cur_status'"
2470 local repaired=$(do_facet mds1 $LCTL get_param -n \
2471 mdd.$(facet_svc mds1).lfsck_layout |
2472 awk '/^repaired_orphan/ { print $2 }')
2473 [ $repaired -eq 2 ] ||
2474 error "(10) Expect 2 fixed on mds{1}, but got: $repaired"
2476 if [ $MDSCOUNT -ge 2 ]; then
2477 repaired=$(do_facet mds2 $LCTL get_param -n \
2478 mdd.$(facet_svc mds2).lfsck_layout |
2479 awk '/^repaired_orphan/ { print $2 }')
2480 [ $repaired -eq 2 ] ||
2481 error "(11) Expect 2 fixed on mds{2}, but got: $repaired"
2484 run_test 18f "Skip the failed OST(s) when handle orphan OST-objects"
2488 echo "The target MDT-object is lost, but related OI mapping is there"
2489 echo "The LFSCK should recreate the lost MDT-object without affected"
2490 echo "by the stale OI mapping."
2493 check_mount_and_prep
2494 $LFS mkdir -i 0 $DIR/$tdir/a1
2495 $LFS setstripe -c -1 -i 0 -S 1M $DIR/$tdir/a1
2496 dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=1M count=$OSTCOUNT
2497 local fid1=$($LFS path2fid $DIR/$tdir/a1/f1)
2499 $LFS getstripe $DIR/$tdir/a1/f1
2500 cancel_lru_locks osc
2502 echo "Inject failure to simulate lost MDT-object but keep OI mapping"
2503 #define OBD_FAIL_LFSCK_LOST_MDTOBJ2 0x162e
2504 do_facet mds1 $LCTL set_param fail_loc=0x162e
2505 rm -f $DIR/$tdir/a1/f1
2507 do_facet mds1 $LCTL set_param fail_loc=0
2508 cancel_lru_locks mdc
2509 cancel_lru_locks osc
2511 echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
2512 $START_LAYOUT -r -o || error "(1) Fail to start LFSCK for layout!"
2514 for k in $(seq $MDSCOUNT); do
2515 # The LFSCK status query internal is 30 seconds. For the case
2516 # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
2517 # time to guarantee the status sync up.
2518 wait_update_facet mds${k} "$LCTL get_param -n \
2519 mdd.$(facet_svc mds${k}).lfsck_layout |
2520 awk '/^status/ { print \\\$2 }'" "completed" $LTIME ||
2521 error "(2) MDS${k} is not the expected 'completed'"
2524 for k in $(seq $OSTCOUNT); do
2525 local cur_status=$(do_facet ost${k} $LCTL get_param -n \
2526 obdfilter.$(facet_svc ost${k}).lfsck_layout |
2527 awk '/^status/ { print $2 }')
2528 [ "$cur_status" == "completed" ] ||
2529 error "(3) OST${k} Expect 'completed', but got '$cur_status'"
2532 local repaired=$(do_facet mds1 $LCTL get_param -n \
2533 mdd.$(facet_svc mds1).lfsck_layout |
2534 awk '/^repaired_orphan/ { print $2 }')
2535 [ $repaired -eq $OSTCOUNT ] ||
2536 error "(4) Expect $OSTCOUNT fixed, but got: $repaired"
2538 echo "Move the files from ./lustre/lost+found/MDTxxxx to namespace"
2539 mv $MOUNT/.lustre/lost+found/MDT0000/${fid1}-R-0 $DIR/$tdir/a1/f1 ||
2540 error "(5) Fail to move $MOUNT/.lustre/lost+found/MDT0000/${fid1}-R-0"
2542 $LFS path2fid $DIR/$tdir/a1/f1
2543 $LFS getstripe $DIR/$tdir/a1/f1
2545 run_test 18g "Find out orphan OST-object and repair it (7)"
2547 $LCTL set_param debug=-cache > /dev/null
2550 check_mount_and_prep
2551 $LFS setstripe -c 1 -i 0 $DIR/$tdir
2553 echo "foo" > $DIR/$tdir/a0
2554 echo "guard" > $DIR/$tdir/a1
2555 cancel_lru_locks osc
2557 echo "Inject failure, then client will offer wrong parent FID when read"
2558 do_facet ost1 $LCTL set_param -n \
2559 obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid 1
2560 #define OBD_FAIL_LFSCK_INVALID_PFID 0x1619
2561 $LCTL set_param fail_loc=0x1619
2563 echo "Read RPC with wrong parent FID should be denied"
2564 cat $DIR/$tdir/a0 && error "(3) Read should be denied!"
2565 $LCTL set_param fail_loc=0
2567 run_test 19a "OST-object inconsistency self detect"
2570 check_mount_and_prep
2571 $LFS setstripe -c 1 -i 0 $DIR/$tdir
2573 echo "Inject failure stub to make the OST-object to back point to"
2574 echo "non-exist MDT-object"
2576 #define OBD_FAIL_LFSCK_UNMATCHED_PAIR1 0x1611
2577 do_facet ost1 $LCTL set_param fail_loc=0x1611
2578 echo "foo" > $DIR/$tdir/f0
2579 cancel_lru_locks osc
2580 do_facet ost1 $LCTL set_param fail_loc=0
2582 echo "Nothing should be fixed since self detect and repair is disabled"
2583 local repaired=$(do_facet ost1 $LCTL get_param -n \
2584 obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid |
2585 awk '/^repaired/ { print $2 }')
2586 [ $repaired -eq 0 ] ||
2587 error "(1) Expected 0 repaired, but got $repaired"
2589 echo "Read RPC with right parent FID should be accepted,"
2590 echo "and cause parent FID on OST to be fixed"
2592 do_facet ost1 $LCTL set_param -n \
2593 obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid 1
2594 cat $DIR/$tdir/f0 || error "(2) Read should not be denied!"
2596 repaired=$(do_facet ost1 $LCTL get_param -n \
2597 obdfilter.${FSNAME}-OST0000.lfsck_verify_pfid |
2598 awk '/^repaired/ { print $2 }')
2599 [ $repaired -eq 1 ] ||
2600 error "(3) Expected 1 repaired, but got $repaired"
2602 run_test 19b "OST-object inconsistency self repair"
2605 [ $OSTCOUNT -lt 2 ] &&
2606 skip "The test needs at least 2 OSTs" && return
2609 echo "The target MDT-object and some of its OST-object are lost."
2610 echo "The LFSCK should find out the left OST-objects and re-create"
2611 echo "the MDT-object under the direcotry .lustre/lost+found/MDTxxxx/"
2612 echo "with the partial OST-objects (LOV EA hole)."
2614 echo "New client can access the file with LOV EA hole via normal"
2615 echo "system tools or commands without crash the system."
2617 echo "For old client, even though it cannot access the file with"
2618 echo "LOV EA hole, it should not cause the system crash."
2621 check_mount_and_prep
2622 $LFS mkdir -i 0 $DIR/$tdir/a1
2623 if [ $OSTCOUNT -gt 2 ]; then
2624 $LFS setstripe -c 3 -i 0 -S 1M $DIR/$tdir/a1
2627 $LFS setstripe -c 2 -i 0 -S 1M $DIR/$tdir/a1
2631 # 256 blocks on the stripe0.
2632 # 1 block on the stripe1 for 2 OSTs case.
2633 # 256 blocks on the stripe1 for other cases.
2634 # 1 block on the stripe2 if OSTs > 2
2635 dd if=/dev/zero of=$DIR/$tdir/a1/f0 bs=4096 count=$bcount
2636 dd if=/dev/zero of=$DIR/$tdir/a1/f1 bs=4096 count=$bcount
2637 dd if=/dev/zero of=$DIR/$tdir/a1/f2 bs=4096 count=$bcount
2639 local fid0=$($LFS path2fid $DIR/$tdir/a1/f0)
2640 local fid1=$($LFS path2fid $DIR/$tdir/a1/f1)
2641 local fid2=$($LFS path2fid $DIR/$tdir/a1/f2)
2644 $LFS getstripe $DIR/$tdir/a1/f0
2646 $LFS getstripe $DIR/$tdir/a1/f1
2648 $LFS getstripe $DIR/$tdir/a1/f2
2650 if [ $OSTCOUNT -gt 2 ]; then
2651 dd if=/dev/zero of=$DIR/$tdir/a1/f3 bs=4096 count=$bcount
2652 fid3=$($LFS path2fid $DIR/$tdir/a1/f3)
2654 $LFS getstripe $DIR/$tdir/a1/f3
2657 cancel_lru_locks osc
2659 echo "Inject failure..."
2660 echo "To simulate f0 lost MDT-object"
2661 #define OBD_FAIL_LFSCK_LOST_MDTOBJ 0x1616
2662 do_facet mds1 $LCTL set_param fail_loc=0x1616
2663 rm -f $DIR/$tdir/a1/f0
2665 echo "To simulate f1 lost MDT-object and OST-object0"
2666 #define OBD_FAIL_LFSCK_LOST_SPEOBJ 0x161a
2667 do_facet mds1 $LCTL set_param fail_loc=0x161a
2668 rm -f $DIR/$tdir/a1/f1
2670 echo "To simulate f2 lost MDT-object and OST-object1"
2671 do_facet mds1 $LCTL set_param fail_val=1
2672 rm -f $DIR/$tdir/a1/f2
2674 if [ $OSTCOUNT -gt 2 ]; then
2675 echo "To simulate f3 lost MDT-object and OST-object2"
2676 do_facet mds1 $LCTL set_param fail_val=2
2677 rm -f $DIR/$tdir/a1/f3
2680 umount_client $MOUNT
2683 do_facet mds1 $LCTL set_param fail_loc=0 fail_val=0
2685 echo "Inject failure to slow down the LFSCK on OST0"
2686 #define OBD_FAIL_LFSCK_DELAY5 0x161b
2687 do_facet ost1 $LCTL set_param fail_loc=0x161b
2689 echo "Trigger layout LFSCK on all devices to find out orphan OST-object"
2690 $START_LAYOUT -r -o || error "(1) Fail to start LFSCK for layout!"
2693 do_facet ost1 $LCTL set_param fail_loc=0
2695 for k in $(seq $MDSCOUNT); do
2696 # The LFSCK status query internal is 30 seconds. For the case
2697 # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
2698 # time to guarantee the status sync up.
2699 wait_update_facet mds${k} "$LCTL get_param -n \
2700 mdd.$(facet_svc mds${k}).lfsck_layout |
2701 awk '/^status/ { print \\\$2 }'" "completed" 32 ||
2702 error "(2) MDS${k} is not the expected 'completed'"
2705 for k in $(seq $OSTCOUNT); do
2706 local cur_status=$(do_facet ost${k} $LCTL get_param -n \
2707 obdfilter.$(facet_svc ost${k}).lfsck_layout |
2708 awk '/^status/ { print $2 }')
2709 [ "$cur_status" == "completed" ] ||
2710 error "(3) OST${k} Expect 'completed', but got '$cur_status'"
2713 local repaired=$(do_facet mds1 $LCTL get_param -n \
2714 mdd.$(facet_svc mds1).lfsck_layout |
2715 awk '/^repaired_orphan/ { print $2 }')
2716 if [ $OSTCOUNT -gt 2 ]; then
2717 [ $repaired -eq 9 ] ||
2718 error "(4.1) Expect 9 fixed on mds1, but got: $repaired"
2720 [ $repaired -eq 4 ] ||
2721 error "(4.2) Expect 4 fixed on mds1, but got: $repaired"
2724 mount_client $MOUNT || error "(5.0) Fail to start client!"
2726 LOV_PATTERN_F_HOLE=0x40000000
2729 # ${fid0}-R-0 is the old f0
2731 local name="$MOUNT/.lustre/lost+found/MDT0000/${fid0}-R-0"
2732 echo "Check $name, which is the old f0"
2734 $LFS getstripe -v $name || error "(5.1) cannot getstripe on $name"
2736 local pattern=0x$($LFS getstripe -L $name)
2737 [[ $((pattern & LOV_PATTERN_F_HOLE)) -eq 0 ]] ||
2738 error "(5.2) NOT expect pattern flag hole, but got $pattern"
2740 local stripes=$($LFS getstripe -c $name)
2741 if [ $OSTCOUNT -gt 2 ]; then
2742 [ $stripes -eq 3 ] ||
2743 error "(5.3.1) expect the stripe count is 3, but got $stripes"
2745 [ $stripes -eq 2 ] ||
2746 error "(5.3.2) expect the stripe count is 2, but got $stripes"
2749 local size=$(stat $name | awk '/Size:/ { print $2 }')
2750 [ $size -eq $((4096 * $bcount)) ] ||
2751 error "(5.4) expect the size $((4096 * $bcount)), but got $size"
2753 cat $name > /dev/null || error "(5.5) cannot read $name"
2755 echo "dummy" >> $name || error "(5.6) cannot write $name"
2757 chown $RUNAS_ID:$RUNAS_GID $name || error "(5.7) cannot chown on $name"
2759 touch $name || error "(5.8) cannot touch $name"
2761 rm -f $name || error "(5.9) cannot unlink $name"
2764 # ${fid1}-R-0 contains the old f1's stripe1 (and stripe2 if OSTs > 2)
2766 name="$MOUNT/.lustre/lost+found/MDT0000/${fid1}-R-0"
2767 if [ $OSTCOUNT -gt 2 ]; then
2768 echo "Check $name, it contains the old f1's stripe1 and stripe2"
2770 echo "Check $name, it contains the old f1's stripe1"
2773 $LFS getstripe -v $name || error "(6.1) cannot getstripe on $name"
2775 pattern=0x$($LFS getstripe -L $name)
2776 [[ $((pattern & LOV_PATTERN_F_HOLE)) -ne 0 ]] ||
2777 error "(6.2) expect pattern flag hole, but got $pattern"
2779 stripes=$($LFS getstripe -c $name)
2780 if [ $OSTCOUNT -gt 2 ]; then
2781 [ $stripes -eq 3 ] ||
2782 error "(6.3.1) expect the stripe count is 3, but got $stripes"
2784 [ $stripes -eq 2 ] ||
2785 error "(6.3.2) expect the stripe count is 2, but got $stripes"
2788 size=$(stat $name | awk '/Size:/ { print $2 }')
2789 [ $size -eq $((4096 * $bcount)) ] ||
2790 error "(6.4) expect the size $((4096 * $bcount)), but got $size"
2792 cat $name > /dev/null && error "(6.5) normal read $name should fail"
2794 local failures=$(dd if=$name of=$DIR/$tdir/dump conv=sync,noerror \
2795 bs=4096 2>&1 | grep "Input/output error" | wc -l)
2798 [ $failures -eq 256 ] ||
2799 error "(6.6) expect 256 IO failures, but get $failures"
2801 size=$(stat $DIR/$tdir/dump | awk '/Size:/ { print $2 }')
2802 [ $size -eq $((4096 * $bcount)) ] ||
2803 error "(6.7) expect the size $((4096 * $bcount)), but got $size"
2805 dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 &&
2806 error "(6.8) write to the LOV EA hole should fail"
2808 dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 seek=300 ||
2809 error "(6.9) write to normal stripe should NOT fail"
2811 echo "foo" >> $name && error "(6.10) append write $name should fail"
2813 chown $RUNAS_ID:$RUNAS_GID $name || error "(6.11) cannot chown on $name"
2815 touch $name || error "(6.12) cannot touch $name"
2817 rm -f $name || error "(6.13) cannot unlink $name"
2820 # ${fid2}-R-0 it contains the old f2's stripe0 (and stripe2 if OSTs > 2)
2822 name="$MOUNT/.lustre/lost+found/MDT0000/${fid2}-R-0"
2823 if [ $OSTCOUNT -gt 2 ]; then
2824 echo "Check $name, it contains the old f2's stripe0 and stripe2"
2826 echo "Check $name, it contains the old f2's stripe0"
2829 $LFS getstripe -v $name || error "(7.1) cannot getstripe on $name"
2831 pattern=0x$($LFS getstripe -L $name)
2832 stripes=$($LFS getstripe -c $name)
2833 size=$(stat $name | awk '/Size:/ { print $2 }')
2834 if [ $OSTCOUNT -gt 2 ]; then
2835 [[ $((pattern & LOV_PATTERN_F_HOLE)) -ne 0 ]] ||
2836 error "(7.2.1) expect pattern flag hole, but got $pattern"
2838 [ $stripes -eq 3 ] ||
2839 error "(7.3.1) expect the stripe count is 3, but got $stripes"
2841 [ $size -eq $((4096 * $bcount)) ] ||
2842 error "(7.4.1) expect size $((4096 * $bcount)), but got $size"
2844 cat $name > /dev/null &&
2845 error "(7.5.1) normal read $name should fail"
2847 failures=$(dd if=$name of=$DIR/$tdir/dump conv=sync,noerror \
2848 bs=4096 2>&1 | grep "Input/output error" | wc -l)
2850 [ $failures -eq 256 ] ||
2851 error "(7.6) expect 256 IO failures, but get $failures"
2853 size=$(stat $DIR/$tdir/dump | awk '/Size:/ { print $2 }')
2854 [ $size -eq $((4096 * $bcount)) ] ||
2855 error "(7.7) expect the size $((4096 * $bcount)), but got $size"
2857 dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 \
2858 seek=300 && error "(7.8.0) write to the LOV EA hole should fail"
2860 dd if=/dev/zero of=$name conv=sync,notrunc bs=4096 count=1 ||
2861 error "(7.8.1) write to normal stripe should NOT fail"
2863 echo "foo" >> $name &&
2864 error "(7.8.3) append write $name should fail"
2866 chown $RUNAS_ID:$RUNAS_GID $name ||
2867 error "(7.9.1) cannot chown on $name"
2869 touch $name || error "(7.10.1) cannot touch $name"
2871 [[ $((pattern & LOV_PATTERN_F_HOLE)) -eq 0 ]] ||
2872 error "(7.2.2) NOT expect pattern flag hole, but got $pattern"
2874 [ $stripes -eq 1 ] ||
2875 error "(7.3.2) expect the stripe count is 1, but got $stripes"
2878 [ $size -eq $((4096 * (256 + 0))) ] ||
2879 error "(7.4.2) expect the size $((4096 * 256)), but got $size"
2881 cat $name > /dev/null || error "(7.5.2) cannot read $name"
2883 echo "dummy" >> $name || error "(7.8.2) cannot write $name"
2885 chown $RUNAS_ID:$RUNAS_GID $name ||
2886 error "(7.9.2) cannot chown on $name"
2888 touch $name || error "(7.10.2) cannot touch $name"
2891 rm -f $name || error "(7.11) cannot unlink $name"
2893 [ $OSTCOUNT -le 2 ] && return
2896 # ${fid3}-R-0 should contains the old f3's stripe0 and stripe1
2898 name="$MOUNT/.lustre/lost+found/MDT0000/${fid3}-R-0"
2899 echo "Check $name, which contains the old f3's stripe0 and stripe1"
2901 $LFS getstripe -v $name || error "(8.1) cannot getstripe on $name"
2903 pattern=0x$($LFS getstripe -L $name)
2904 [[ $((pattern & LOV_PATTERN_F_HOLE)) -eq 0 ]] ||
2905 error "(8.2) NOT expect pattern flag hole, but got $pattern"
2907 stripes=$($LFS getstripe -c $name)
2908 # LFSCK does not know the old f3 had 3 stripes.
2909 # It only tries to find as much as possible.
2910 # The stripe count depends on the last stripe's offset.
2911 [ $stripes -eq 2 ] ||
2912 error "(8.3) expect the stripe count is 2, but got $stripes"
2914 size=$(stat $name | awk '/Size:/ { print $2 }')
2916 [ $size -eq $((4096 * (256 + 256 + 0))) ] ||
2917 error "(8.4) expect the size $((4096 * 512)), but got $size"
2919 cat $name > /dev/null || error "(8.5) cannot read $name"
2921 echo "dummy" >> $name || error "(8.6) cannot write $name"
2923 chown $RUNAS_ID:$RUNAS_GID $name ||
2924 error "(8.7) cannot chown on $name"
2926 touch $name || error "(8.8) cannot touch $name"
2928 rm -f $name || error "(8.9) cannot unlink $name"
2930 run_test 20 "Handle the orphan with dummy LOV EA slot properly"
2933 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.5.59) ]] &&
2934 skip "ignore the test if MDS is older than 2.5.59" && return
2936 check_mount_and_prep
2937 createmany -o $DIR/$tdir/f 100 || error "(0) Fail to create 100 files"
2939 echo "Start all LFSCK components by default (-s 1)"
2940 do_facet mds1 $LCTL lfsck_start -M ${FSNAME}-MDT0000 -s 1 -r ||
2941 error "Fail to start LFSCK"
2943 echo "namespace LFSCK should be in 'scanning-phase1' status"
2944 local STATUS=$($SHOW_NAMESPACE | awk '/^status/ { print $2 }')
2945 [ "$STATUS" == "scanning-phase1" ] ||
2946 error "Expect namespace 'scanning-phase1', but got '$STATUS'"
2948 echo "layout LFSCK should be in 'scanning-phase1' status"
2949 STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
2950 [ "$STATUS" == "scanning-phase1" ] ||
2951 error "Expect layout 'scanning-phase1', but got '$STATUS'"
2953 echo "Stop all LFSCK components by default"
2954 do_facet mds1 $LCTL lfsck_stop -M ${FSNAME}-MDT0000 ||
2955 error "Fail to stop LFSCK"
2957 run_test 21 "run all LFSCK components by default"
2960 [ $MDSCOUNT -lt 2 ] &&
2961 skip "We need at least 2 MDSes for this test" && return
2964 echo "The parent_A references the child directory via some name entry,"
2965 echo "but the child directory back references another parent_B via its"
2966 echo "".." name entry. The parent_B does not exist. Then the namespace"
2967 echo "LFSCK will repair the child directory's ".." name entry."
2970 check_mount_and_prep
2972 $LFS mkdir -i 1 $DIR/$tdir/guard || error "(1) Fail to mkdir on MDT1"
2973 $LFS mkdir -i 1 $DIR/$tdir/foo || error "(2) Fail to mkdir on MDT1"
2975 echo "Inject failure stub on MDT0 to simulate bad dotdot name entry"
2976 echo "The dummy's dotdot name entry references the guard."
2977 #define OBD_FAIL_LFSCK_BAD_PARENT 0x161e
2978 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x161e
2979 $LFS mkdir -i 0 $DIR/$tdir/foo/dummy ||
2980 error "(3) Fail to mkdir on MDT0"
2981 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
2983 rmdir $DIR/$tdir/guard || error "(4) Fail to rmdir $DIR/$tdir/guard"
2985 echo "Trigger namespace LFSCK to repair unmatched pairs"
2986 $START_NAMESPACE -A -r ||
2987 error "(5) Fail to start LFSCK for namespace"
2989 wait_all_targets_blocked namespace completed 6
2991 local repaired=$($SHOW_NAMESPACE |
2992 awk '/^unmatched_pairs_repaired/ { print $2 }')
2993 [ $repaired -eq 1 ] ||
2994 error "(7) Fail to repair unmatched pairs: $repaired"
2996 echo "'ls' should success after namespace LFSCK repairing"
2997 ls -ail $DIR/$tdir/foo/dummy > /dev/null ||
2998 error "(8) ls should success."
3000 run_test 22a "LFSCK can repair unmatched pairs (1)"
3003 [ $MDSCOUNT -lt 2 ] &&
3004 skip "We need at least 2 MDSes for this test" && return
3007 echo "The parent_A references the child directory via the name entry_B,"
3008 echo "but the child directory back references another parent_C via its"
3009 echo "".." name entry. The parent_C exists, but there is no the name"
3010 echo "entry_B under the parent_C. Then the namespace LFSCK will repair"
3011 echo "the child directory's ".." name entry and its linkEA."
3014 check_mount_and_prep
3016 $LFS mkdir -i 1 $DIR/$tdir/guard || error "(1) Fail to mkdir on MDT1"
3017 $LFS mkdir -i 1 $DIR/$tdir/foo || error "(2) Fail to mkdir on MDT1"
3019 echo "Inject failure stub on MDT0 to simulate bad dotdot name entry"
3020 echo "and bad linkEA. The dummy's dotdot name entry references the"
3021 echo "guard. The dummy's linkEA references n non-exist name entry."
3022 #define OBD_FAIL_LFSCK_BAD_PARENT 0x161e
3023 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x161e
3024 $LFS mkdir -i 0 $DIR/$tdir/foo/dummy ||
3025 error "(3) Fail to mkdir on MDT0"
3026 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3028 local dummyfid=$($LFS path2fid $DIR/$tdir/foo/dummy)
3029 echo "fid2path should NOT work on the dummy's FID $dummyfid"
3030 local dummyname=$($LFS fid2path $DIR $dummyfid)
3031 [ "$dummyname" != "$DIR/$tdir/foo/dummy" ] ||
3032 error "(4) fid2path works unexpectedly."
3034 echo "Trigger namespace LFSCK to repair unmatched pairs"
3035 $START_NAMESPACE -A -r ||
3036 error "(5) Fail to start LFSCK for namespace"
3038 wait_all_targets_blocked namespace completed 6
3040 local repaired=$($SHOW_NAMESPACE |
3041 awk '/^unmatched_pairs_repaired/ { print $2 }')
3042 [ $repaired -eq 1 ] ||
3043 error "(7) Fail to repair unmatched pairs: $repaired"
3045 echo "fid2path should work on the dummy's FID $dummyfid after LFSCK"
3046 local dummyname=$($LFS fid2path $DIR $dummyfid)
3047 [ "$dummyname" == "$DIR/$tdir/foo/dummy" ] ||
3048 error "(8) fid2path does not work"
3050 run_test 22b "LFSCK can repair unmatched pairs (2)"
3053 [ $MDSCOUNT -lt 2 ] &&
3054 skip "We need at least 2 MDSes for this test" && return
3057 echo "The name entry is there, but the MDT-object for such name "
3058 echo "entry does not exist. The namespace LFSCK should find out "
3059 echo "and repair the inconsistency as required."
3062 check_mount_and_prep
3064 $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0 on MDT0"
3065 $LFS mkdir -i 1 $DIR/$tdir/d0/d1 || error "(2) Fail to mkdir d1 on MDT1"
3067 echo "Inject failure stub on MDT1 to simulate dangling name entry"
3068 #define OBD_FAIL_LFSCK_DANGLING2 0x1620
3069 do_facet mds2 $LCTL set_param fail_loc=0x1620
3070 rmdir $DIR/$tdir/d0/d1 || error "(3) Fail to rmdir d1"
3071 do_facet mds2 $LCTL set_param fail_loc=0
3073 echo "'ls' should fail because of dangling name entry"
3074 ls -ail $DIR/$tdir/d0/d1 > /dev/null 2>&1 && error "(4) ls should fail."
3076 echo "Trigger namespace LFSCK to find out dangling name entry"
3077 $START_NAMESPACE -A -r ||
3078 error "(5) Fail to start LFSCK for namespace"
3080 wait_all_targets_blocked namespace completed 6
3082 local repaired=$($SHOW_NAMESPACE |
3083 awk '/^dangling_repaired/ { print $2 }')
3084 [ $repaired -eq 1 ] ||
3085 error "(7) Fail to repair dangling name entry: $repaired"
3087 echo "'ls' should fail because not re-create MDT-object by default"
3088 ls -ail $DIR/$tdir/d0/d1 > /dev/null 2>&1 && error "(8) ls should fail."
3090 echo "Trigger namespace LFSCK again to repair dangling name entry"
3091 $START_NAMESPACE -A -r -C ||
3092 error "(9) Fail to start LFSCK for namespace"
3094 wait_all_targets_blocked namespace completed 10
3096 repaired=$($SHOW_NAMESPACE |
3097 awk '/^dangling_repaired/ { print $2 }')
3098 [ $repaired -eq 1 ] ||
3099 error "(11) Fail to repair dangling name entry: $repaired"
3101 echo "'ls' should success after namespace LFSCK repairing"
3102 ls -ail $DIR/$tdir/d0/d1 > /dev/null || error "(12) ls should success."
3104 run_test 23a "LFSCK can repair dangling name entry (1)"
3108 echo "The objectA has multiple hard links, one of them corresponding"
3109 echo "to the name entry_B. But there is something wrong for the name"
3110 echo "entry_B and cause entry_B to references non-exist object_C."
3111 echo "In the first-stage scanning, the LFSCK will think the entry_B"
3112 echo "as dangling, and re-create the lost object_C. When the LFSCK"
3113 echo "comes to the second-stage scanning, it will find that the"
3114 echo "former re-creating object_C is not proper, and will try to"
3115 echo "replace the object_C with the real object_A."
3118 check_mount_and_prep
3120 [[ -d $MOUNT/.lustre/lost+found/MDT0000 ]] || {
3121 # Trigger LFSCK firstly, that will generate the
3122 # .lustre/lost+found/MDTxxxx in advance to avoid
3123 # reusing the local object for the dangling name
3125 $START_NAMESPACE -r ||
3126 error "(0) Fail to start LFSCK for namespace"
3128 wait_all_targets_blocked namespace completed 0.1
3131 $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0 on MDT0"
3132 $LFS path2fid $DIR/$tdir/d0
3134 echo "dummy" > $DIR/$tdir/d0/f0 || error "(2) Fail to touch on MDT0"
3135 $LFS path2fid $DIR/$tdir/d0/f0
3137 echo "dead" > $DIR/$tdir/d0/f1 || error "(3) Fail to touch on MDT0"
3138 $LFS path2fid $DIR/$tdir/d0/f1
3140 local OID=$($LFS path2fid $DIR/$tdir/d0/f1 | awk -F':' '{print $2}')
3141 OID=$(printf %d $OID)
3143 if [ $OID -eq 1 ]; then
3144 # To guarantee that the f0 and f1 are in the same FID seq
3145 rm -f $DIR/$tdir/d0/f0 ||
3146 error "(3.1) Fail to unlink $DIR/$tdir/d0/f0"
3147 echo "dummy" > $DIR/$tdir/d0/f0 ||
3148 error "(3.2) Fail to touch on MDT0"
3149 $LFS path2fid $DIR/$tdir/d0/f0
3152 echo "Inject failure stub on MDT0 to simulate dangling name entry"
3153 #define OBD_FAIL_LFSCK_DANGLING3 0x1621
3154 do_facet $SINGLEMDS $LCTL set_param fail_val=$OID fail_loc=0x1621
3155 ln $DIR/$tdir/d0/f0 $DIR/$tdir/d0/foo || error "(4) Fail to hard link"
3156 do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
3158 rm -f $DIR/$tdir/d0/f1 || error "(5) Fail to unlink $DIR/$tdir/d0/f1"
3160 echo "'ls' should fail because of dangling name entry"
3161 ls -ail $DIR/$tdir/d0/foo > /dev/null 2>&1 &&
3162 error "(6) ls should fail."
3164 echo "Trigger namespace LFSCK to find out dangling name entry"
3165 $START_NAMESPACE -r -C ||
3166 error "(7) Fail to start LFSCK for namespace"
3168 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
3169 mdd.${MDT_DEV}.lfsck_namespace |
3170 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
3172 error "(8) unexpected status"
3175 local repaired=$($SHOW_NAMESPACE |
3176 awk '/^dangling_repaired/ { print $2 }')
3177 [ $repaired -eq 1 ] ||
3178 error "(9) Fail to repair dangling name entry: $repaired"
3180 repaired=$($SHOW_NAMESPACE |
3181 awk '/^multiple_linked_repaired/ { print $2 }')
3182 [ $repaired -eq 1 ] ||
3183 error "(10) Fail to drop the former created object: $repaired"
3185 local data=$(cat $DIR/$tdir/d0/foo)
3186 [ "$data" == "dummy" ] ||
3187 error "(11) The $DIR/$tdir/d0/foo is not recovered: $data"
3189 run_test 23b "LFSCK can repair dangling name entry (2)"
3193 echo "The objectA has multiple hard links, one of them corresponding"
3194 echo "to the name entry_B. But there is something wrong for the name"
3195 echo "entry_B and cause entry_B to references non-exist object_C."
3196 echo "In the first-stage scanning, the LFSCK will think the entry_B"
3197 echo "as dangling, and re-create the lost object_C. And then others"
3198 echo "modified the re-created object_C. When the LFSCK comes to the"
3199 echo "second-stage scanning, it will find that the former re-creating"
3200 echo "object_C maybe wrong and try to replace the object_C with the"
3201 echo "real object_A. But because object_C has been modified, so the"
3202 echo "LFSCK cannot replace it."
3205 start_full_debug_logging
3207 check_mount_and_prep
3209 [[ -d $MOUNT/.lustre/lost+found/MDT0000 ]] || {
3210 # Trigger LFSCK firstly, that will generate the
3211 # .lustre/lost+found/MDTxxxx in advance to avoid
3212 # reusing the local object for the dangling name
3214 $START_NAMESPACE -r ||
3215 error "(0) Fail to start LFSCK for namespace"
3217 wait_all_targets_blocked namespace completed 0.1
3220 $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0 on MDT0"
3221 $LFS path2fid $DIR/$tdir/d0
3223 echo "dummy" > $DIR/$tdir/d0/f0 || error "(2) Fail to touch on MDT0"
3224 $LFS path2fid $DIR/$tdir/d0/f0
3226 echo "dead" > $DIR/$tdir/d0/f1 || error "(3) Fail to touch on MDT0"
3227 $LFS path2fid $DIR/$tdir/d0/f1
3229 local OID=$($LFS path2fid $DIR/$tdir/d0/f1 | awk -F':' '{print $2}')
3230 OID=$(printf %d $OID)
3232 if [ $OID -eq 1 ]; then
3233 # To guarantee that the f0 and f1 are in the same FID seq
3234 rm -f $DIR/$tdir/d0/f0 ||
3235 error "(3.1) Fail to unlink $DIR/$tdir/d0/f0"
3236 echo "dummy" > $DIR/$tdir/d0/f0 ||
3237 error "(3.2) Fail to touch on MDT0"
3238 $LFS path2fid $DIR/$tdir/d0/f0
3241 echo "Inject failure stub on MDT0 to simulate dangling name entry"
3242 #define OBD_FAIL_LFSCK_DANGLING3 0x1621
3243 do_facet $SINGLEMDS $LCTL set_param fail_val=$OID fail_loc=0x1621
3244 ln $DIR/$tdir/d0/f0 $DIR/$tdir/d0/foo || error "(4) Fail to hard link"
3245 do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
3247 rm -f $DIR/$tdir/d0/f1 || error "(5) Fail to unlink $DIR/$tdir/d0/f1"
3249 echo "'ls' should fail because of dangling name entry"
3250 ls -ail $DIR/$tdir/d0/foo > /dev/null 2>&1 &&
3251 error "(6) ls should fail."
3253 #define OBD_FAIL_LFSCK_DELAY3 0x1602
3254 do_facet $SINGLEMDS $LCTL set_param fail_val=10 fail_loc=0x1602
3256 echo "Trigger namespace LFSCK to find out dangling name entry"
3257 $START_NAMESPACE -r -C ||
3258 error "(7) Fail to start LFSCK for namespace"
3260 wait_update_facet client "stat $DIR/$tdir/d0/foo |
3261 awk '/Size/ { print \\\$2 }'" "0" $LTIME || {
3262 stat $DIR/$tdir/guard
3264 error "(8) unexpected size"
3267 echo "data" >> $DIR/$tdir/d0/foo || error "(9) Fail to write"
3268 cancel_lru_locks osc
3270 do_facet $SINGLEMDS $LCTL set_param fail_val=0 fail_loc=0
3271 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
3272 mdd.${MDT_DEV}.lfsck_namespace |
3273 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
3275 error "(10) unexpected status"
3278 stop_full_debug_logging
3280 local repaired=$($SHOW_NAMESPACE |
3281 awk '/^dangling_repaired/ { print $2 }')
3282 [ $repaired -eq 1 ] ||
3283 error "(11) Fail to repair dangling name entry: $repaired"
3285 local data=$(cat $DIR/$tdir/d0/foo)
3286 [ "$data" != "dummy" ] ||
3287 error "(12) The $DIR/$tdir/d0/foo should not be recovered"
3289 run_test 23c "LFSCK can repair dangling name entry (3)"
3292 [ $MDSCOUNT -lt 2 ] &&
3293 skip "We need at least 2 MDSes for this test" && return
3296 echo "Two MDT-objects back reference the same name entry via their"
3297 echo "each own linkEA entry, but the name entry only references one"
3298 echo "MDT-object. The namespace LFSCK will remove the linkEA entry"
3299 echo "for the MDT-object that is not recognized. If such MDT-object"
3300 echo "has no other linkEA entry after the removing, then the LFSCK"
3301 echo "will add it as orphan under the .lustre/lost+found/MDTxxxx/."
3304 check_mount_and_prep
3306 $LFS mkdir -i 1 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
3308 mkdir $DIR/$tdir/d0/guard || error "(1) Fail to mkdir guard"
3309 $LFS path2fid $DIR/$tdir/d0/guard
3311 mkdir $DIR/$tdir/d0/dummy || error "(2) Fail to mkdir dummy"
3312 $LFS path2fid $DIR/$tdir/d0/dummy
3315 if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
3316 pfid=$($LFS path2fid $DIR/$tdir/d0/guard)
3318 pfid=$($LFS path2fid $DIR/$tdir/d0/dummy)
3321 touch $DIR/$tdir/d0/guard/foo ||
3322 error "(3) Fail to touch $DIR/$tdir/d0/guard/foo"
3324 echo "Inject failure stub on MDT0 to simulate the case that"
3325 echo "the $DIR/$tdir/d0/dummy/foo has the 'bad' linkEA entry"
3326 echo "that references $DIR/$tdir/d0/guard/foo."
3327 echo "Then remove the name entry $DIR/$tdir/d0/dummy/foo."
3328 echo "So the MDT-object $DIR/$tdir/d0/dummy/foo will be left"
3329 echo "there with the same linkEA entry as another MDT-object"
3330 echo "$DIR/$tdir/d0/guard/foo has"
3332 #define OBD_FAIL_LFSCK_MUL_REF 0x1622
3333 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1622
3334 $LFS mkdir -i 0 $DIR/$tdir/d0/dummy/foo ||
3335 error "(4) Fail to mkdir $DIR/$tdir/d0/dummy/foo"
3336 $LFS path2fid $DIR/$tdir/d0/dummy/foo
3337 local cfid=$($LFS path2fid $DIR/$tdir/d0/dummy/foo)
3338 rmdir $DIR/$tdir/d0/dummy/foo ||
3339 error "(5) Fail to remove $DIR/$tdir/d0/dummy/foo name entry"
3340 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3342 echo "stat $DIR/$tdir/d0/dummy/foo should fail"
3343 stat $DIR/$tdir/d0/dummy/foo > /dev/null 2>&1 &&
3344 error "(6) stat successfully unexpectedly"
3346 echo "Trigger namespace LFSCK to repair multiple-referenced name entry"
3347 $START_NAMESPACE -A -r ||
3348 error "(7) Fail to start LFSCK for namespace"
3350 wait_all_targets_blocked namespace completed 8
3352 local repaired=$($SHOW_NAMESPACE |
3353 awk '/^multiple_referenced_repaired/ { print $2 }')
3354 [ $repaired -eq 1 ] ||
3355 error "(9) Fail to repair multiple referenced name entry: $repaired"
3357 echo "There should be an orphan under .lustre/lost+found/MDT0000/"
3358 [ -d $MOUNT/.lustre/lost+found/MDT0000 ] ||
3359 error "(10) $MOUNT/.lustre/lost+found/MDT0000/ should be there"
3361 local cname="$cfid-$pfid-D-0"
3362 ls -ail $MOUNT/.lustre/lost+found/MDT0000/$cname ||
3363 error "(11) .lustre/lost+found/MDT0000/ should not be empty"
3365 run_test 24 "LFSCK can repair multiple-referenced name entry"
3368 [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
3369 skip "Only support to inject failure on ldiskfs" && return
3372 echo "The file type in the name entry does not match the file type"
3373 echo "claimed by the referenced object. Then the LFSCK will update"
3374 echo "the file type in the name entry."
3377 check_mount_and_prep
3379 $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
3381 echo "Inject failure stub on MDT0 to simulate the case that"
3382 echo "the file type stored in the name entry is wrong."
3384 #define OBD_FAIL_LFSCK_BAD_TYPE 0x1623
3385 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1623
3386 touch $DIR/$tdir/d0/foo || error "(2) Fail to touch $DIR/$tdir/d0/foo"
3387 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3389 echo "Trigger namespace LFSCK to repair bad file type in the name entry"
3390 $START_NAMESPACE -r || error "(3) Fail to start LFSCK for namespace"
3392 wait_update_facet $SINGLEMDS "$LCTL get_param -n \
3393 mdd.${MDT_DEV}.lfsck_namespace |
3394 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
3396 error "(4) unexpected status"
3399 local repaired=$($SHOW_NAMESPACE |
3400 awk '/^bad_file_type_repaired/ { print $2 }')
3401 [ $repaired -eq 1 ] ||
3402 error "(5) Fail to repair bad file type in name entry: $repaired"
3404 ls -ail $DIR/$tdir/d0 || error "(6) Fail to 'ls' the $DIR/$tdir/d0"
3406 run_test 25 "LFSCK can repair bad file type in the name entry"
3410 echo "The local name entry back referenced by the MDT-object is lost."
3411 echo "The namespace LFSCK will add the missing local name entry back"
3412 echo "to the normal namespace."
3415 check_mount_and_prep
3417 $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
3418 touch $DIR/$tdir/d0/foo || error "(2) Fail to create foo"
3419 local foofid=$($LFS path2fid $DIR/$tdir/d0/foo)
3421 ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/dummy ||
3422 error "(3) Fail to hard link to $DIR/$tdir/d0/foo"
3424 echo "Inject failure stub on MDT0 to simulate the case that"
3425 echo "foo's name entry will be removed, but the foo's object"
3426 echo "and its linkEA are kept in the system."
3428 #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
3429 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1624
3430 rm -f $DIR/$tdir/d0/foo || error "(4) Fail to unlink $DIR/$tdir/d0/foo"
3431 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3433 ls -ail $DIR/$tdir/d0/foo > /dev/null 2>&1 &&
3434 error "(5) 'ls' should fail"
3436 echo "Trigger namespace LFSCK to repair the missing remote name entry"
3437 $START_NAMESPACE -r -A ||
3438 error "(6) Fail to start LFSCK for namespace"
3440 wait_all_targets_blocked namespace completed 7
3442 local repaired=$($SHOW_NAMESPACE |
3443 awk '/^lost_dirent_repaired/ { print $2 }')
3444 [ $repaired -eq 1 ] ||
3445 error "(8) Fail to repair lost dirent: $repaired"
3447 ls -ail $DIR/$tdir/d0/foo ||
3448 error "(9) Fail to 'ls' $DIR/$tdir/d0/foo"
3450 local foofid2=$($LFS path2fid $DIR/$tdir/d0/foo)
3451 [ "$foofid" == "$foofid2" ] ||
3452 error "(10) foo's FID changed: $foofid, $foofid2"
3454 run_test 26a "LFSCK can add the missing local name entry back to the namespace"
3457 [ $MDSCOUNT -lt 2 ] &&
3458 skip "We need at least 2 MDSes for this test" && return
3461 echo "The remote name entry back referenced by the MDT-object is lost."
3462 echo "The namespace LFSCK will add the missing remote name entry back"
3463 echo "to the normal namespace."
3466 check_mount_and_prep
3468 $LFS mkdir -i 1 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
3469 $LFS mkdir -i 0 $DIR/$tdir/d0/foo || error "(2) Fail to mkdir foo"
3470 local foofid=$($LFS path2fid $DIR/$tdir/d0/foo)
3472 echo "Inject failure stub on MDT0 to simulate the case that"
3473 echo "foo's name entry will be removed, but the foo's object"
3474 echo "and its linkEA are kept in the system."
3476 #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
3477 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1624
3478 rmdir $DIR/$tdir/d0/foo || error "(3) Fail to rmdir $DIR/$tdir/d0/foo"
3479 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3481 ls -ail $DIR/$tdir/d0/foo > /dev/null 2>&1 &&
3482 error "(4) 'ls' should fail"
3484 echo "Trigger namespace LFSCK to repair the missing remote name entry"
3485 $START_NAMESPACE -r -A ||
3486 error "(5) Fail to start LFSCK for namespace"
3488 wait_all_targets_blocked namespace completed 6
3490 local repaired=$($SHOW_NAMESPACE |
3491 awk '/^lost_dirent_repaired/ { print $2 }')
3492 [ $repaired -eq 1 ] ||
3493 error "(7) Fail to repair lost dirent: $repaired"
3495 ls -ail $DIR/$tdir/d0/foo ||
3496 error "(8) Fail to 'ls' $DIR/$tdir/d0/foo"
3498 local foofid2=$($LFS path2fid $DIR/$tdir/d0/foo)
3499 [ "$foofid" == "$foofid2" ] ||
3500 error "(9) foo's FID changed: $foofid, $foofid2"
3502 run_test 26b "LFSCK can add the missing remote name entry back to the namespace"
3506 echo "The local parent referenced by the MDT-object linkEA is lost."
3507 echo "The namespace LFSCK will re-create the lost parent as orphan."
3510 check_mount_and_prep
3512 $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
3513 touch $DIR/$tdir/d0/foo || error "(2) Fail to create foo"
3514 ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/dummy ||
3515 error "(3) Fail to hard link to $DIR/$tdir/d0/foo"
3517 echo "Inject failure stub on MDT0 to simulate the case that"
3518 echo "foo's name entry will be removed, but the foo's object"
3519 echo "and its linkEA are kept in the system. And then remove"
3520 echo "another hard link and the parent directory."
3522 #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
3523 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1624
3524 rm -f $DIR/$tdir/d0/foo ||
3525 error "(4) Fail to unlink $DIR/$tdir/d0/foo"
3526 rm -f $DIR/$tdir/d0/dummy ||
3527 error "(5) Fail to unlink $DIR/$tdir/d0/dummy"
3528 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3530 rm -rf $DIR/$tdir/d0 || error "(5) Fail to unlink the dir d0"
3531 ls -ail $DIR/$tdir/d0 > /dev/null 2>&1 && error "(6) 'ls' should fail"
3533 echo "Trigger namespace LFSCK to repair the lost parent"
3534 $START_NAMESPACE -r -A ||
3535 error "(6) Fail to start LFSCK for namespace"
3537 wait_all_targets_blocked namespace completed 7
3539 local repaired=$($SHOW_NAMESPACE |
3540 awk '/^lost_dirent_repaired/ { print $2 }')
3541 [ $repaired -eq 1 ] ||
3542 error "(8) Fail to repair lost dirent: $repaired"
3544 echo "There should be an orphan under .lustre/lost+found/MDT0000/"
3545 [ -d $MOUNT/.lustre/lost+found/MDT0000 ] ||
3546 error "(9) $MOUNT/.lustre/lost+found/MDT0000/ should be there"
3548 ls -ail $MOUNT/.lustre/lost+found/MDT0000/
3550 cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-P-*)
3551 [ ! -z "$cname" ] ||
3552 error "(10) .lustre/lost+found/MDT0000/ should not be empty"
3554 run_test 27a "LFSCK can recreate the lost local parent directory as orphan"
3557 [ $MDSCOUNT -lt 2 ] &&
3558 skip "We need at least 2 MDSes for this test" && return
3561 echo "The remote parent referenced by the MDT-object linkEA is lost."
3562 echo "The namespace LFSCK will re-create the lost parent as orphan."
3565 check_mount_and_prep
3567 $LFS mkdir -i 1 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
3568 $LFS mkdir -i 0 $DIR/$tdir/d0/foo || error "(2) Fail to mkdir foo"
3570 $LFS path2fid $DIR/$tdir/d0
3572 echo "Inject failure stub on MDT0 to simulate the case that"
3573 echo "foo's name entry will be removed, but the foo's object"
3574 echo "and its linkEA are kept in the system. And then remove"
3575 echo "the parent directory."
3577 #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
3578 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1624
3579 rmdir $DIR/$tdir/d0/foo || error "(3) Fail to rmdir $DIR/$tdir/d0/foo"
3580 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3582 rmdir $DIR/$tdir/d0 || error "(4) Fail to unlink the dir d0"
3583 ls -ail $DIR/$tdir/d0 > /dev/null 2>&1 && error "(5) 'ls' should fail"
3585 echo "Trigger namespace LFSCK to repair the missing remote name entry"
3586 $START_NAMESPACE -r -A ||
3587 error "(6) Fail to start LFSCK for namespace"
3589 wait_all_targets_blocked namespace completed 7
3591 local repaired=$($SHOW_NAMESPACE |
3592 awk '/^lost_dirent_repaired/ { print $2 }')
3593 [ $repaired -eq 1 ] ||
3594 error "(8) Fail to repair lost dirent: $repaired"
3596 ls -ail $MOUNT/.lustre/lost+found/
3598 echo "There should be an orphan under .lustre/lost+found/MDT0001/"
3599 [ -d $MOUNT/.lustre/lost+found/MDT0001 ] ||
3600 error "(9) $MOUNT/.lustre/lost+found/MDT0001/ should be there"
3602 ls -ail $MOUNT/.lustre/lost+found/MDT0001/
3604 cname=$(find $MOUNT/.lustre/lost+found/MDT0001/ -name *-P-*)
3605 [ ! -z "$cname" ] ||
3606 error "(10) .lustre/lost+found/MDT0001/ should not be empty"
3608 run_test 27b "LFSCK can recreate the lost remote parent directory as orphan"
3611 [ $MDSCOUNT -lt 2 ] &&
3612 skip "The test needs at least 2 MDTs" && return
3615 echo "The target name entry is lost. The LFSCK should insert the"
3616 echo "orphan MDT-object under .lustre/lost+found/MDTxxxx. But if"
3617 echo "the MDT (on which the orphan MDT-object resides) has ever"
3618 echo "failed to respond some name entry verification during the"
3619 echo "first stage-scanning, then the LFSCK should skip to handle"
3620 echo "orphan MDT-object on this MDT. But other MDTs should not"
3624 check_mount_and_prep
3625 $LFS mkdir -i 0 $DIR/$tdir/d1
3626 $LFS mkdir -i 1 $DIR/$tdir/d1/a1
3627 $LFS mkdir -i 1 $DIR/$tdir/d1/a2
3629 $LFS mkdir -i 1 $DIR/$tdir/d2
3630 $LFS mkdir -i 0 $DIR/$tdir/d2/a1
3631 $LFS mkdir -i 0 $DIR/$tdir/d2/a2
3633 echo "Inject failure stub on MDT0 to simulate the case that"
3634 echo "d1/a1's name entry will be removed, but the d1/a1's object"
3635 echo "and its linkEA are kept in the system. And the case that"
3636 echo "d2/a2's name entry will be removed, but the d2/a2's object"
3637 echo "and its linkEA are kept in the system."
3639 #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
3640 do_facet mds1 $LCTL set_param fail_loc=0x1624
3641 do_facet mds2 $LCTL set_param fail_loc=0x1624
3642 rmdir $DIR/$tdir/d1/a1 || error "(1) Fail to rmdir $DIR/$tdir/d1/a1"
3643 rmdir $DIR/$tdir/d2/a2 || error "(2) Fail to rmdir $DIR/$tdir/d2/a2"
3644 do_facet mds1 $LCTL set_param fail_loc=0
3645 do_facet mds2 $LCTL set_param fail_loc=0
3647 cancel_lru_locks mdc
3648 cancel_lru_locks osc
3650 echo "Inject failure, to simulate the MDT0 fail to handle"
3651 echo "MDT1 LFSCK request during the first-stage scanning."
3652 #define OBD_FAIL_LFSCK_BAD_NETWORK 0x161c
3653 do_facet mds2 $LCTL set_param fail_loc=0x161c fail_val=0
3655 echo "Trigger namespace LFSCK on all devices to find out orphan object"
3656 $START_NAMESPACE -r -A ||
3657 error "(3) Fail to start LFSCK for namespace"
3659 wait_update_facet mds1 "$LCTL get_param -n \
3660 mdd.$(facet_svc mds1).lfsck_namespace |
3661 awk '/^status/ { print \\\$2 }'" "partial" 32 || {
3662 error "(4) mds1 is not the expected 'partial'"
3665 wait_update_facet mds2 "$LCTL get_param -n \
3666 mdd.$(facet_svc mds2).lfsck_namespace |
3667 awk '/^status/ { print \\\$2 }'" "completed" 32 || {
3668 error "(5) mds2 is not the expected 'completed'"
3671 do_facet mds2 $LCTL set_param fail_loc=0 fail_val=0
3673 local repaired=$(do_facet mds1 $LCTL get_param -n \
3674 mdd.$(facet_svc mds1).lfsck_namespace |
3675 awk '/^lost_dirent_repaired/ { print $2 }')
3676 [ $repaired -eq 0 ] ||
3677 error "(6) Expect 0 fixed on mds1, but got: $repaired"
3679 repaired=$(do_facet mds2 $LCTL get_param -n \
3680 mdd.$(facet_svc mds2).lfsck_namespace |
3681 awk '/^lost_dirent_repaired/ { print $2 }')
3682 [ $repaired -eq 1 ] ||
3683 error "(7) Expect 1 fixed on mds2, but got: $repaired"
3685 echo "Trigger namespace LFSCK on all devices again to cleanup"
3686 $START_NAMESPACE -r -A ||
3687 error "(8) Fail to start LFSCK for namespace"
3689 wait_all_targets_blocked namespace completed 9
3691 local repaired=$(do_facet mds1 $LCTL get_param -n \
3692 mdd.$(facet_svc mds1).lfsck_namespace |
3693 awk '/^lost_dirent_repaired/ { print $2 }')
3694 [ $repaired -eq 1 ] ||
3695 error "(10) Expect 1 fixed on mds1, but got: $repaired"
3697 repaired=$(do_facet mds2 $LCTL get_param -n \
3698 mdd.$(facet_svc mds2).lfsck_namespace |
3699 awk '/^lost_dirent_repaired/ { print $2 }')
3700 [ $repaired -eq 0 ] ||
3701 error "(11) Expect 0 fixed on mds2, but got: $repaired"
3703 run_test 28 "Skip the failed MDT(s) when handle orphan MDT-objects"
3707 echo "The object's nlink attribute is larger than the object's known"
3708 echo "name entries count. The LFSCK will repair the object's nlink"
3709 echo "attribute to match the known name entries count"
3712 check_mount_and_prep
3714 $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
3715 touch $DIR/$tdir/d0/foo || error "(2) Fail to create foo"
3717 echo "Inject failure stub on MDT0 to simulate the case that foo's"
3718 echo "nlink attribute is larger than its name entries count."
3720 #define OBD_FAIL_LFSCK_MORE_NLINK 0x1625
3721 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1625
3722 ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/h1 ||
3723 error "(3) Fail to hard link to $DIR/$tdir/d0/foo"
3724 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3726 cancel_lru_locks mdc
3727 local count=$(stat --format=%h $DIR/$tdir/d0/foo)
3728 [ $count -eq 3 ] || error "(4) Cannot inject error: $count"
3730 echo "Trigger namespace LFSCK to repair the nlink count"
3731 $START_NAMESPACE -r -A ||
3732 error "(5) Fail to start LFSCK for namespace"
3734 wait_all_targets_blocked namespace completed 6
3736 local repaired=$($SHOW_NAMESPACE |
3737 awk '/^nlinks_repaired/ { print $2 }')
3738 [ $repaired -eq 1 ] ||
3739 error "(7) Fail to repair nlink count: $repaired"
3741 cancel_lru_locks mdc
3742 count=$(stat --format=%h $DIR/$tdir/d0/foo)
3743 [ $count -eq 2 ] || error "(8) Fail to repair nlink count: $count"
3745 run_test 29a "LFSCK can repair bad nlink count (1)"
3749 echo "The object's nlink attribute is smaller than the object's known"
3750 echo "name entries count. The LFSCK will repair the object's nlink"
3751 echo "attribute to match the known name entries count"
3754 check_mount_and_prep
3756 $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
3757 touch $DIR/$tdir/d0/foo || error "(2) Fail to create foo"
3759 echo "Inject failure stub on MDT0 to simulate the case that foo's"
3760 echo "nlink attribute is smaller than its name entries count."
3762 #define OBD_FAIL_LFSCK_LESS_NLINK 0x1626
3763 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1626
3764 ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/h1 ||
3765 error "(3) Fail to hard link to $DIR/$tdir/d0/foo"
3766 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3768 cancel_lru_locks mdc
3769 local count=$(stat --format=%h $DIR/$tdir/d0/foo)
3770 [ $count -eq 1 ] || error "(4) Cannot inject error: $count"
3772 echo "Trigger namespace LFSCK to repair the nlink count"
3773 $START_NAMESPACE -r -A ||
3774 error "(5) Fail to start LFSCK for namespace"
3776 wait_all_targets_blocked namespace completed 6
3778 local repaired=$($SHOW_NAMESPACE |
3779 awk '/^nlinks_repaired/ { print $2 }')
3780 [ $repaired -eq 1 ] ||
3781 error "(7) Fail to repair nlink count: $repaired"
3783 cancel_lru_locks mdc
3784 count=$(stat --format=%h $DIR/$tdir/d0/foo)
3785 [ $count -eq 2 ] || error "(8) Fail to repair nlink count: $count"
3787 run_test 29b "LFSCK can repair bad nlink count (2)"
3791 echo "There are too many hard links to the object, and exceeds the"
3792 echo "object's linkEA limitation, as to NOT all the known name entries"
3793 echo "will be recorded in the linkEA. Under such case, LFSCK should"
3794 echo "skip the nlink verification for this object."
3797 check_mount_and_prep
3799 $LFS mkdir -i 0 $DIR/$tdir/d0 || error "(1) Fail to mkdir d0"
3800 touch $DIR/$tdir/d0/foo || error "(2) Fail to create foo"
3801 ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/h1 ||
3802 error "(3) Fail to hard link to $DIR/$tdir/d0/foo"
3804 echo "Inject failure stub on MDT0 to simulate the case that"
3805 echo "foo's hard links exceed the object's linkEA limitation."
3807 #define OBD_FAIL_LFSCK_LINKEA_OVERFLOW 0x1627
3808 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1627
3809 ln $DIR/$tdir/d0/foo $DIR/$tdir/d0/h2 ||
3810 error "(4) Fail to hard link to $DIR/$tdir/d0/foo"
3812 cancel_lru_locks mdc
3814 local count1=$(stat --format=%h $DIR/$tdir/d0/foo)
3815 [ $count1 -eq 3 ] || error "(5) Stat failure: $count1"
3817 local foofid=$($LFS path2fid $DIR/$tdir/d0/foo)
3818 $LFS fid2path $DIR $foofid
3819 local count2=$($LFS fid2path $DIR $foofid | wc -l)
3820 [ $count2 -eq 2 ] || error "(6) Fail to inject error: $count2"
3822 echo "Trigger namespace LFSCK to repair the nlink count"
3823 $START_NAMESPACE -r -A ||
3824 error "(7) Fail to start LFSCK for namespace"
3826 wait_all_targets_blocked namespace completed 8
3828 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3829 local repaired=$($SHOW_NAMESPACE |
3830 awk '/^nlinks_repaired/ { print $2 }')
3831 [ $repaired -eq 0 ] ||
3832 error "(9) Repair nlink count unexpcetedly: $repaired"
3834 cancel_lru_locks mdc
3836 count1=$(stat --format=%h $DIR/$tdir/d0/foo)
3837 [ $count1 -eq 3 ] || error "(10) Stat failure: $count1"
3839 count2=$($LFS fid2path $DIR $foofid | wc -l)
3840 [ $count2 -eq 2 ] ||
3841 error "(11) Repaired something unexpectedly: $count2"
3843 run_test 29c "Not verify nlink attr if hark links exceed linkEA limitation"
3846 [ $(facet_fstype $SINGLEMDS) != ldiskfs ] &&
3847 skip "Only support backend /lost+found for ldiskfs" && return
3850 echo "The namespace LFSCK will move the orphans from backend"
3851 echo "/lost+found directory to normal client visible namespace"
3852 echo "or to global visible ./lustre/lost+found/MDTxxxx/ directory"
3855 check_mount_and_prep
3857 $LFS mkdir -i 0 $DIR/$tdir/foo || error "(1) Fail to mkdir foo"
3858 touch $DIR/$tdir/foo/f0 || error "(2) Fail to touch f1"
3860 echo "Inject failure stub on MDT0 to simulate the case that"
3861 echo "directory d0 has no linkEA entry, then the LFSCK will"
3862 echo "move it into .lustre/lost+found/MDTxxxx/ later."
3864 #define OBD_FAIL_LFSCK_NO_LINKEA 0x161d
3865 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x161d
3866 mkdir $DIR/$tdir/foo/d0 || error "(3) Fail to mkdir d0"
3867 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3869 touch $DIR/$tdir/foo/d0/f1 || error "(4) Fail to touch f1"
3870 mkdir $DIR/$tdir/foo/d0/d1 || error "(5) Fail to mkdir d1"
3872 echo "Inject failure stub on MDT0 to simulate the case that the"
3873 echo "object's name entry will be removed, but not destroy the"
3874 echo "object. Then backend e2fsck will handle it as orphan and"
3875 echo "add them into the backend /lost+found directory."
3877 #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
3878 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1624
3879 rmdir $DIR/$tdir/foo/d0/d1 || error "(6) Fail to rmdir d1"
3880 rm -f $DIR/$tdir/foo/d0/f1 || error "(7) Fail to unlink f1"
3881 rmdir $DIR/$tdir/foo/d0 || error "(8) Fail to rmdir d0"
3882 rm -f $DIR/$tdir/foo/f0 || error "(9) Fail to unlink f0"
3883 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
3885 umount_client $MOUNT || error "(10) Fail to stop client!"
3887 stop $SINGLEMDS || error "(11) Fail to stop MDT0"
3890 run_e2fsck $(facet_host $SINGLEMDS) $MDT_DEVNAME "-y" ||
3891 error "(12) Fail to run e2fsck"
3893 start $SINGLEMDS $MDT_DEVNAME $MOUNT_OPTS_NOSCRUB > /dev/null ||
3894 error "(13) Fail to start MDT0"
3896 echo "Trigger namespace LFSCK to recover backend orphans"
3897 $START_NAMESPACE -r -A ||
3898 error "(14) Fail to start LFSCK for namespace"
3900 wait_all_targets_blocked namespace completed 15
3902 local repaired=$($SHOW_NAMESPACE |
3903 awk '/^local_lost_found_moved/ { print $2 }')
3904 [ $repaired -ge 4 ] ||
3905 error "(16) Fail to recover backend orphans: $repaired"
3907 mount_client $MOUNT || error "(17) Fail to start client!"
3909 stat $DIR/$tdir/foo/f0 || error "(18) f0 is not recovered"
3911 ls -ail $MOUNT/.lustre/lost+found/
3913 echo "d0 should become orphan under .lustre/lost+found/MDT0000/"
3914 [ -d $MOUNT/.lustre/lost+found/MDT0000 ] ||
3915 error "(19) $MOUNT/.lustre/lost+found/MDT0000/ should be there"
3917 ls -ail $MOUNT/.lustre/lost+found/MDT0000/
3919 cname=$(find $MOUNT/.lustre/lost+found/MDT0000/ -name *-*-D-*)
3920 [ ! -z "$cname" ] || error "(20) d0 is not recovered"
3922 stat ${cname}/d1 || error "(21) d0 is not recovered"
3923 stat ${cname}/f1 || error "(22) f1 is not recovered"
3925 run_test 30 "LFSCK can recover the orphans from backend /lost+found"
3928 [ $MDSCOUNT -lt 2 ] &&
3929 skip "The test needs at least 2 MDTs" && return
3932 echo "For the name entry under a striped directory, if the name"
3933 echo "hash does not match the shard, then the LFSCK will repair"
3934 echo "the bad name entry"
3937 check_mount_and_prep
3939 $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
3940 error "(1) Fail to create striped directory"
3942 echo "Inject failure stub on client to simulate the case that"
3943 echo "some name entry should be inserted into other non-first"
3944 echo "shard, but inserted into the first shard by wrong"
3946 #define OBD_FAIL_LFSCK_BAD_NAME_HASH 0x1628
3947 $LCTL set_param fail_loc=0x1628 fail_val=0
3948 createmany -d $DIR/$tdir/striped_dir/d $MDSCOUNT ||
3949 error "(2) Fail to create file under striped directory"
3950 $LCTL set_param fail_loc=0 fail_val=0
3952 echo "Trigger namespace LFSCK to repair bad name hash"
3953 $START_NAMESPACE -r -A ||
3954 error "(3) Fail to start LFSCK for namespace"
3956 wait_all_targets_blocked namespace completed 4
3958 local repaired=$($SHOW_NAMESPACE |
3959 awk '/^name_hash_repaired/ { print $2 }')
3960 [ $repaired -ge 1 ] ||
3961 error "(5) Fail to repair bad name hash: $repaired"
3963 umount_client $MOUNT || error "(6) umount failed"
3964 mount_client $MOUNT || error "(7) mount failed"
3966 for ((i = 0; i < $MDSCOUNT; i++)); do
3967 stat $DIR/$tdir/striped_dir/d$i ||
3968 error "(8) Fail to stat d$i after LFSCK"
3969 rmdir $DIR/$tdir/striped_dir/d$i ||
3970 error "(9) Fail to unlink d$i after LFSCK"
3973 rmdir $DIR/$tdir/striped_dir ||
3974 error "(10) Fail to remove the striped directory after LFSCK"
3976 run_test 31a "The LFSCK can find/repair the name entry with bad name hash (1)"
3979 [ $MDSCOUNT -lt 2 ] &&
3980 skip "The test needs at least 2 MDTs" && return
3983 echo "For the name entry under a striped directory, if the name"
3984 echo "hash does not match the shard, then the LFSCK will repair"
3985 echo "the bad name entry"
3988 check_mount_and_prep
3990 $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
3991 error "(1) Fail to create striped directory"
3993 echo "Inject failure stub on client to simulate the case that"
3994 echo "some name entry should be inserted into other non-second"
3995 echo "shard, but inserted into the secod shard by wrong"
3997 #define OBD_FAIL_LFSCK_BAD_NAME_HASH 0x1628
3998 $LCTL set_param fail_loc=0x1628 fail_val=1
3999 createmany -d $DIR/$tdir/striped_dir/d $MDSCOUNT ||
4000 error "(2) Fail to create file under striped directory"
4001 $LCTL set_param fail_loc=0 fail_val=0
4003 echo "Trigger namespace LFSCK to repair bad name hash"
4004 $START_NAMESPACE -r -A ||
4005 error "(3) Fail to start LFSCK for namespace"
4007 wait_all_targets_blocked namespace completed 4
4009 local repaired=$(do_facet mds2 $LCTL get_param -n \
4010 mdd.$(facet_svc mds2).lfsck_namespace |
4011 awk '/^name_hash_repaired/ { print $2 }')
4012 [ $repaired -ge 1 ] ||
4013 error "(5) Fail to repair bad name hash: $repaired"
4015 umount_client $MOUNT || error "(6) umount failed"
4016 mount_client $MOUNT || error "(7) mount failed"
4018 for ((i = 0; i < $MDSCOUNT; i++)); do
4019 stat $DIR/$tdir/striped_dir/d$i ||
4020 error "(8) Fail to stat d$i after LFSCK"
4021 rmdir $DIR/$tdir/striped_dir/d$i ||
4022 error "(9) Fail to unlink d$i after LFSCK"
4025 rmdir $DIR/$tdir/striped_dir ||
4026 error "(10) Fail to remove the striped directory after LFSCK"
4028 run_test 31b "The LFSCK can find/repair the name entry with bad name hash (2)"
4031 [ $MDSCOUNT -lt 2 ] &&
4032 skip "The test needs at least 2 MDTs" && return
4035 echo "For some reason, the master MDT-object of the striped directory"
4036 echo "may lost its master LMV EA. If nobody created files under the"
4037 echo "master directly after the master LMV EA lost, then the LFSCK"
4038 echo "should re-generate the master LMV EA."
4041 check_mount_and_prep
4043 echo "Inject failure stub on MDT0 to simulate the case that the"
4044 echo "master MDT-object of the striped directory lost the LMV EA."
4046 #define OBD_FAIL_LFSCK_LOST_MASTER_LMV 0x1629
4047 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1629
4048 $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
4049 error "(1) Fail to create striped directory"
4050 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
4052 echo "Trigger namespace LFSCK to re-generate master LMV EA"
4053 $START_NAMESPACE -r -A ||
4054 error "(2) Fail to start LFSCK for namespace"
4056 wait_all_targets_blocked namespace completed 3
4058 local repaired=$($SHOW_NAMESPACE |
4059 awk '/^striped_dirs_repaired/ { print $2 }')
4060 [ $repaired -eq 1 ] ||
4061 error "(4) Fail to re-generate master LMV EA: $repaired"
4063 umount_client $MOUNT || error "(5) umount failed"
4064 mount_client $MOUNT || error "(6) mount failed"
4066 local empty=$(ls $DIR/$tdir/striped_dir/)
4067 [ -z "$empty" ] || error "(7) The master LMV EA is not repaired: $empty"
4069 rmdir $DIR/$tdir/striped_dir ||
4070 error "(8) Fail to remove the striped directory after LFSCK"
4072 run_test 31c "Re-generate the lost master LMV EA for striped directory"
4075 [ $MDSCOUNT -lt 2 ] &&
4076 skip "The test needs at least 2 MDTs" && return
4079 echo "For some reason, the master MDT-object of the striped directory"
4080 echo "may lost its master LMV EA. If somebody created files under the"
4081 echo "master directly after the master LMV EA lost, then the LFSCK"
4082 echo "should NOT re-generate the master LMV EA, instead, it should"
4083 echo "change the broken striped dirctory as read-only to prevent"
4084 echo "further damage"
4087 check_mount_and_prep
4089 echo "Inject failure stub on MDT0 to simulate the case that the"
4090 echo "master MDT-object of the striped directory lost the LMV EA."
4092 #define OBD_FAIL_LFSCK_LOST_MASTER_LMV 0x1629
4093 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x1629
4094 $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
4095 error "(1) Fail to create striped directory"
4096 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0
4098 umount_client $MOUNT || error "(2) umount failed"
4099 mount_client $MOUNT || error "(3) mount failed"
4101 touch $DIR/$tdir/striped_dir/dummy ||
4102 error "(4) Fail to touch under broken striped directory"
4104 echo "Trigger namespace LFSCK to find out the inconsistency"
4105 $START_NAMESPACE -r -A ||
4106 error "(5) Fail to start LFSCK for namespace"
4108 wait_all_targets_blocked namespace completed 6
4110 local repaired=$($SHOW_NAMESPACE |
4111 awk '/^striped_dirs_repaired/ { print $2 }')
4112 [ $repaired -eq 0 ] ||
4113 error "(7) Re-generate master LMV EA unexpected: $repaired"
4115 stat $DIR/$tdir/striped_dir/dummy ||
4116 error "(8) Fail to stat $DIR/$tdir/striped_dir/dummy"
4118 touch $DIR/$tdir/striped_dir/foo &&
4119 error "(9) The broken striped directory should be read-only"
4121 chattr -i $DIR/$tdir/striped_dir ||
4122 error "(10) Fail to chattr on the broken striped directory"
4124 rmdir $DIR/$tdir/striped_dir ||
4125 error "(11) Fail to remove the striped directory after LFSCK"
4127 run_test 31d "Set broken striped directory (modified after broken) as read-only"
4130 [ $MDSCOUNT -lt 2 ] &&
4131 skip "The test needs at least 2 MDTs" && return
4134 echo "For some reason, the slave MDT-object of the striped directory"
4135 echo "may lost its slave LMV EA. The LFSCK should re-generate the"
4136 echo "slave LMV EA."
4139 check_mount_and_prep
4141 echo "Inject failure stub on MDT0 to simulate the case that the"
4142 echo "slave MDT-object (that resides on the same MDT as the master"
4143 echo "MDT-object resides on) lost the LMV EA."
4145 #define OBD_FAIL_LFSCK_LOST_SLAVE_LMV 0x162a
4146 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x162a fail_val=0
4147 $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
4148 error "(1) Fail to create striped directory"
4149 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0 fail_val=0
4151 echo "Trigger namespace LFSCK to re-generate slave LMV EA"
4152 $START_NAMESPACE -r -A ||
4153 error "(2) Fail to start LFSCK for namespace"
4155 wait_all_targets_blocked namespace completed 3
4157 local repaired=$($SHOW_NAMESPACE |
4158 awk '/^striped_shards_repaired/ { print $2 }')
4159 [ $repaired -eq 1 ] ||
4160 error "(4) Fail to re-generate slave LMV EA: $repaired"
4162 rmdir $DIR/$tdir/striped_dir ||
4163 error "(5) Fail to remove the striped directory after LFSCK"
4165 run_test 31e "Re-generate the lost slave LMV EA for striped directory (1)"
4168 [ $MDSCOUNT -lt 2 ] &&
4169 skip "The test needs at least 2 MDTs" && return
4172 echo "For some reason, the slave MDT-object of the striped directory"
4173 echo "may lost its slave LMV EA. The LFSCK should re-generate the"
4174 echo "slave LMV EA."
4177 check_mount_and_prep
4179 echo "Inject failure stub on MDT0 to simulate the case that the"
4180 echo "slave MDT-object (that resides on different MDT as the master"
4181 echo "MDT-object resides on) lost the LMV EA."
4183 #define OBD_FAIL_LFSCK_LOST_SLAVE_LMV 0x162a
4184 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x162a fail_val=1
4185 $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
4186 error "(1) Fail to create striped directory"
4187 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0 fail_val=0
4189 echo "Trigger namespace LFSCK to re-generate slave LMV EA"
4190 $START_NAMESPACE -r -A ||
4191 error "(2) Fail to start LFSCK for namespace"
4193 wait_all_targets_blocked namespace completed 3
4195 local repaired=$(do_facet mds2 $LCTL get_param -n \
4196 mdd.$(facet_svc mds2).lfsck_namespace |
4197 awk '/^striped_shards_repaired/ { print $2 }')
4198 [ $repaired -eq 1 ] ||
4199 error "(4) Fail to re-generate slave LMV EA: $repaired"
4201 rmdir $DIR/$tdir/striped_dir ||
4202 error "(5) Fail to remove the striped directory after LFSCK"
4204 run_test 31f "Re-generate the lost slave LMV EA for striped directory (2)"
4207 [ $MDSCOUNT -lt 2 ] &&
4208 skip "The test needs at least 2 MDTs" && return
4211 echo "For some reason, the stripe index in the slave LMV EA is"
4212 echo "corrupted. The LFSCK should repair the slave LMV EA."
4215 check_mount_and_prep
4217 echo "Inject failure stub on MDT0 to simulate the case that the"
4218 echo "slave LMV EA on the first shard of the striped directory"
4219 echo "claims the same index as the second shard claims"
4221 #define OBD_FAIL_LFSCK_BAD_SLAVE_LMV 0x162b
4222 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x162b fail_val=0
4223 $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
4224 error "(1) Fail to create striped directory"
4225 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0 fail_val=0
4227 echo "Trigger namespace LFSCK to repair the slave LMV EA"
4228 $START_NAMESPACE -r -A ||
4229 error "(2) Fail to start LFSCK for namespace"
4231 wait_all_targets_blocked namespace completed 3
4233 local repaired=$($SHOW_NAMESPACE |
4234 awk '/^striped_shards_repaired/ { print $2 }')
4235 [ $repaired -eq 1 ] ||
4236 error "(4) Fail to repair slave LMV EA: $repaired"
4238 umount_client $MOUNT || error "(5) umount failed"
4239 mount_client $MOUNT || error "(6) mount failed"
4241 touch $DIR/$tdir/striped_dir/foo ||
4242 error "(7) Fail to touch file after the LFSCK"
4244 rm -f $DIR/$tdir/striped_dir/foo ||
4245 error "(8) Fail to unlink file after the LFSCK"
4247 rmdir $DIR/$tdir/striped_dir ||
4248 error "(9) Fail to remove the striped directory after LFSCK"
4250 run_test 31g "Repair the corrupted slave LMV EA"
4253 [ $MDSCOUNT -lt 2 ] &&
4254 skip "The test needs at least 2 MDTs" && return
4257 echo "For some reason, the shard's name entry in the striped"
4258 echo "directory may be corrupted. The LFSCK should repair the"
4259 echo "bad shard's name entry."
4262 check_mount_and_prep
4264 echo "Inject failure stub on MDT0 to simulate the case that the"
4265 echo "first shard's name entry in the striped directory claims"
4266 echo "the same index as the second shard's name entry claims."
4268 #define OBD_FAIL_LFSCK_BAD_SLAVE_NAME 0x162c
4269 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x162c fail_val=0
4270 $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir/striped_dir ||
4271 error "(1) Fail to create striped directory"
4272 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x0 fail_val=0
4274 echo "Trigger namespace LFSCK to repair the shard's name entry"
4275 $START_NAMESPACE -r -A ||
4276 error "(2) Fail to start LFSCK for namespace"
4278 wait_all_targets_blocked namespace completed 3
4280 local repaired=$($SHOW_NAMESPACE |
4281 awk '/^dirent_repaired/ { print $2 }')
4282 [ $repaired -eq 1 ] ||
4283 error "(4) Fail to repair shard's name entry: $repaired"
4285 umount_client $MOUNT || error "(5) umount failed"
4286 mount_client $MOUNT || error "(6) mount failed"
4288 touch $DIR/$tdir/striped_dir/foo ||
4289 error "(7) Fail to touch file after the LFSCK"
4291 rm -f $DIR/$tdir/striped_dir/foo ||
4292 error "(8) Fail to unlink file after the LFSCK"
4294 rmdir $DIR/$tdir/striped_dir ||
4295 error "(9) Fail to remove the striped directory after LFSCK"
4297 run_test 31h "Repair the corrupted shard's name entry"
4302 umount_client $MOUNT
4304 #define OBD_FAIL_LFSCK_ASSISTANT_DIRECT 0x162d
4305 do_facet $SINGLEMDS $LCTL set_param fail_val=3 fail_loc=0x162d
4306 $START_LAYOUT -r || error "(2) Fail to start LFSCK for layout!"
4308 local STATUS=$($SHOW_LAYOUT | awk '/^status/ { print $2 }')
4309 [ "$STATUS" == "scanning-phase1" ] ||
4310 error "(3) Expect 'scanning-phase1', but got '$STATUS'"
4313 stop ost1 > /dev/null || error "(4) Fail to stop OST1!"
4315 do_facet $SINGLEMDS $LCTL set_param fail_loc=0 fail_val=0
4319 $STOP_LFSCK || error "(5) Fail to stop LFSCK!"
4321 run_test 32 "stop LFSCK when some OST failed"
4323 # restore MDS/OST size
4324 MDSSIZE=${SAVED_MDSSIZE}
4325 OSTSIZE=${SAVED_OSTSIZE}
4326 OSTCOUNT=${SAVED_OSTCOUNT}
4328 # cleanup the system at last