7 # This test needs to be run on the client
10 LUSTRE=${LUSTRE:-`dirname $0`/..}
13 . $LUSTRE/tests/test-framework.sh
15 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
16 CHECK_GRANT=${CHECK_GRANT:-"yes"}
17 GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
21 # bug number: 2766 4176 11404
22 ALWAYS_EXCEPT="0b 39 56 $REPLAY_SINGLE_EXCEPT"
24 # 63 min 7 min AT AT AT AT"
25 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 6b 12 16 44 44b 65 66 67 68"
29 cleanup_and_setup_lustre
33 rm -rf $DIR/[df][0-9]*
38 replay_barrier $SINGLEMDS
42 run_test 0 "empty replay"
45 # this test attempts to trigger a race in the precreation code,
46 # and must run before any other objects are created on the filesystem
48 createmany -o $DIR/$tfile 20 || return 1
49 unlinkmany $DIR/$tfile 20 || return 2
51 run_test 0b "ensure object created after recover exists. (3284)"
57 lctl set_param -n seq.cli-srv-$mds-mdc-*.width=$width
63 lctl get_param -n seq.cli-srv-$mds-mdc-*.width
66 # This test should pass for single-mds and multi-mds configs.
67 # But for different configurations it tests different things.
71 # (1) fld_create replay should happen;
73 # (2) fld_create replay should not return -EEXISTS, if it does
74 # this means sequence manager recovery code is buggy and allocated
75 # same sequence two times after recovery.
79 # (1) fld_create replay may not happen, because its home MDS is
80 # MDS2 which is not involved to revovery;
82 # (2) as fld_create does not happen on MDS1, it does not make any
85 local label=`mdsdevlabel 1`
86 [ -z "$label" ] && echo "No label for mds1" && return 1
88 replay_barrier $SINGLEMDS
89 local sw=`seq_get_width $label`
91 # make seq manager switch to next sequence each
92 # time as new fid is needed.
93 seq_set_width $label 1
95 # make sure that fld has created at least one new
97 touch $DIR/$tfile || return 2
98 seq_set_width $label $sw
100 # fail $SINGLEMDS and start recovery, replay RPCs, etc.
103 # wait for recovery finish
107 # flush fld cache and dentry cache to make it lookup
108 # created entry instead of revalidating existent one
110 zconf_mount `hostname` $MOUNT
112 # issue lookup which should call fld lookup which
113 # should fail if client did not replay fld create
114 # correctly and server has no fld entry
115 touch $DIR/$tfile || return 3
116 rm $DIR/$tfile || return 4
118 run_test 0c "fld create"
121 replay_barrier $SINGLEMDS
124 $CHECKSTAT -t file $DIR/$tfile || return 1
127 run_test 1 "simple create"
130 replay_barrier $SINGLEMDS
133 $CHECKSTAT -t file $DIR/$tfile || return 1
140 replay_barrier $SINGLEMDS
143 $CHECKSTAT -t file $DIR/$tfile || return 1
149 replay_barrier $SINGLEMDS
151 o_directory $DIR/$tfile
153 $CHECKSTAT -t file $DIR/$tfile || return 2
156 run_test 3a "replay failed open(O_DIRECTORY)"
159 replay_barrier $SINGLEMDS
160 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
161 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000114"
163 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
165 $CHECKSTAT -t file $DIR/$tfile && return 2
168 run_test 3b "replay failed open -ENOMEM"
171 replay_barrier $SINGLEMDS
172 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
173 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000128"
175 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
178 $CHECKSTAT -t file $DIR/$tfile && return 2
181 run_test 3c "replay failed open -ENOMEM"
184 replay_barrier $SINGLEMDS
185 for i in `seq 10`; do
186 echo "tag-$i" > $DIR/$tfile-$i
189 for i in `seq 10`; do
190 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
193 run_test 4 "|x| 10 open(O_CREAT)s"
196 replay_barrier $SINGLEMDS
199 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
201 run_test 4b "|x| rm 10 files"
203 # The idea is to get past the first block of precreated files on both
204 # osts, and then replay.
206 replay_barrier $SINGLEMDS
207 for i in `seq 220`; do
208 echo "tag-$i" > $DIR/$tfile-$i
211 for i in `seq 220`; do
212 grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
216 # waiting for commitment of removal
218 run_test 5 "|x| 220 open(O_CREAT)"
222 replay_barrier $SINGLEMDS
223 mcreate $DIR/$tdir/$tfile
225 $CHECKSTAT -t dir $DIR/$tdir || return 1
226 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
228 # waiting for log process thread
230 run_test 6 "mkdir + contained create"
233 replay_barrier $SINGLEMDS
236 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
238 run_test 6b "|X| rmdir"
241 replay_barrier $SINGLEMDS
242 mcreate $DIR/$tdir/$tfile
244 $CHECKSTAT -t dir $DIR/$tdir || return 1
245 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
248 run_test 7 "mkdir |X| contained create"
251 # make sure no side-effect from previous test.
253 replay_barrier $SINGLEMDS
254 multiop_bg_pause $DIR/$tfile mo_c || return 4
258 $CHECKSTAT -t file $DIR/$tfile || return 1
259 kill -USR1 $MULTIPID || return 2
260 wait $MULTIPID || return 3
263 run_test 8 "creat open |X| close"
266 replay_barrier $SINGLEMDS
268 local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
270 local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
272 echo " old_inum == $old_inum, new_inum == $new_inum"
273 if [ $old_inum -eq $new_inum ] ;
275 echo " old_inum and new_inum match"
277 echo "!!!! old_inum and new_inum NOT match"
282 run_test 9 "|X| create (same inum/gen)"
286 replay_barrier $SINGLEMDS
287 mv $DIR/$tfile $DIR/$tfile-2
290 $CHECKSTAT $DIR/$tfile && return 1
291 $CHECKSTAT $DIR/$tfile-2 ||return 2
295 run_test 10 "create |X| rename unlink"
299 echo "old" > $DIR/$tfile
300 mv $DIR/$tfile $DIR/$tfile-2
301 replay_barrier $SINGLEMDS
302 echo "new" > $DIR/$tfile
304 grep old $DIR/$tfile-2
306 grep new $DIR/$tfile || return 1
307 grep old $DIR/$tfile-2 || return 2
309 run_test 11 "create open write rename |X| create-old-name read"
313 multiop_bg_pause $DIR/$tfile o_tSc || return 3
316 replay_barrier $SINGLEMDS
318 wait $pid || return 1
321 [ -e $DIR/$tfile ] && return 2
324 run_test 12 "open, unlink |X| close"
327 # 1777 - replay open after committed chmod that would make
328 # a regular open a failure
331 multiop_bg_pause $DIR/$tfile O_wc || return 3
334 $CHECKSTAT -p 0 $DIR/$tfile
335 replay_barrier $SINGLEMDS
338 wait $pid || return 1
340 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
343 run_test 13 "open chmod 0 |x| write close"
346 multiop_bg_pause $DIR/$tfile O_tSc || return 4
349 replay_barrier $SINGLEMDS
350 kill -USR1 $pid || return 1
351 wait $pid || return 2
354 [ -e $DIR/$tfile ] && return 3
357 run_test 14 "open(O_CREAT), unlink |X| close"
360 multiop_bg_pause $DIR/$tfile O_tSc || return 5
363 replay_barrier $SINGLEMDS
364 touch $DIR/g11 || return 1
366 wait $pid || return 2
369 [ -e $DIR/$tfile ] && return 3
370 touch $DIR/h11 || return 4
373 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
377 replay_barrier $SINGLEMDS
380 mcreate $DIR/$tfile-2
382 [ -e $DIR/$tfile ] && return 1
383 [ -e $DIR/$tfile-2 ] || return 2
384 munlink $DIR/$tfile-2 || return 3
386 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
389 replay_barrier $SINGLEMDS
390 multiop_bg_pause $DIR/$tfile O_c || return 4
393 kill -USR1 $pid || return 1
394 wait $pid || return 2
395 $CHECKSTAT -t file $DIR/$tfile || return 3
398 run_test 17 "|X| open(O_CREAT), |replay| close"
401 replay_barrier $SINGLEMDS
402 multiop_bg_pause $DIR/$tfile O_tSc || return 8
405 touch $DIR/$tfile-2 || return 1
406 echo "pid: $pid will close"
408 wait $pid || return 2
411 [ -e $DIR/$tfile ] && return 3
412 [ -e $DIR/$tfile-2 ] || return 4
413 # this touch frequently fails
414 touch $DIR/$tfile-3 || return 5
415 munlink $DIR/$tfile-2 || return 6
416 munlink $DIR/$tfile-3 || return 7
419 run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
421 # bug 1855 (a simpler form of test_11 above)
423 replay_barrier $SINGLEMDS
425 echo "old" > $DIR/$tfile
426 mv $DIR/$tfile $DIR/$tfile-2
427 grep old $DIR/$tfile-2
429 grep old $DIR/$tfile-2 || return 2
431 run_test 19 "|X| mcreate, open, write, rename "
434 replay_barrier $SINGLEMDS
435 multiop_bg_pause $DIR/$tfile O_tSc || return 3
441 wait $pid || return 1
442 [ -e $DIR/$tfile ] && return 2
445 run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
447 test_20b() { # bug 10480
448 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
450 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
452 while [ ! -e $DIR/$tfile ] ; do
453 usleep 60 # give dd a chance to start
456 lfs getstripe $DIR/$tfile || return 1
457 rm -f $DIR/$tfile || return 2 # make it an orphan
459 df -P $DIR || df -P $DIR || true # reconnect
461 fail $SINGLEMDS # start orphan recovery
462 df -P $DIR || df -P $DIR || true # reconnect
463 wait_mds_recovery_done || error "MDS recovery not done"
465 # FIXME just because recovery is done doesn't mean we've finished
466 # orphan cleanup. Fake it with a sleep for now...
468 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
469 log "before $BEFOREUSED, after $AFTERUSED"
470 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
471 error "after $AFTERUSED > before $BEFOREUSED"
474 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
476 test_20c() { # bug 10480
477 multiop_bg_pause $DIR/$tfile Ow_c || return 1
484 df -P $DIR || df -P $DIR || true # reconnect
487 test -s $DIR/$tfile || error "File was truncated"
491 run_test 20c "check that client eviction does not affect file content"
494 replay_barrier $SINGLEMDS
495 multiop_bg_pause $DIR/$tfile O_tSc || return 5
498 touch $DIR/g11 || return 1
502 wait $pid || return 2
503 [ -e $DIR/$tfile ] && return 3
504 touch $DIR/h11 || return 4
507 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
510 multiop_bg_pause $DIR/$tfile O_tSc || return 3
513 replay_barrier $SINGLEMDS
518 wait $pid || return 1
519 [ -e $DIR/$tfile ] && return 2
522 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
525 multiop_bg_pause $DIR/$tfile O_tSc || return 5
528 replay_barrier $SINGLEMDS
530 touch $DIR/g11 || return 1
534 wait $pid || return 2
535 [ -e $DIR/$tfile ] && return 3
536 touch $DIR/h11 || return 4
539 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
542 multiop_bg_pause $DIR/$tfile O_tSc || return 3
545 replay_barrier $SINGLEMDS
549 wait $pid || return 1
550 [ -e $DIR/$tfile ] && return 2
553 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
556 multiop_bg_pause $DIR/$tfile O_tSc || return 3
560 replay_barrier $SINGLEMDS
563 wait $pid || return 1
564 [ -e $DIR/$tfile ] && return 2
567 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
570 replay_barrier $SINGLEMDS
571 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
573 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
578 wait $pid2 || return 1
582 wait $pid1 || return 2
583 [ -e $DIR/$tfile-1 ] && return 3
584 [ -e $DIR/$tfile-2 ] && return 4
587 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
590 replay_barrier $SINGLEMDS
591 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
593 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
600 wait $pid1 || return 1
602 wait $pid2 || return 2
603 [ -e $DIR/$tfile-1 ] && return 3
604 [ -e $DIR/$tfile-2 ] && return 4
607 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
610 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
612 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
614 replay_barrier $SINGLEMDS
618 wait $pid2 || return 1
622 wait $pid1 || return 2
623 [ -e $DIR/$tfile-1 ] && return 3
624 [ -e $DIR/$tfile-2 ] && return 4
627 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
630 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
632 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
634 replay_barrier $SINGLEMDS
640 wait $pid1 || return 1
642 wait $pid2 || return 2
643 [ -e $DIR/$tfile-1 ] && return 3
644 [ -e $DIR/$tfile-2 ] && return 4
647 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
650 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
652 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
657 replay_barrier $SINGLEMDS
660 wait $pid1 || return 1
662 wait $pid2 || return 2
663 [ -e $DIR/$tfile-1 ] && return 3
664 [ -e $DIR/$tfile-2 ] && return 4
667 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
670 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
672 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
676 replay_barrier $SINGLEMDS
680 wait $pid1 || return 1
682 wait $pid2 || return 2
683 [ -e $DIR/$tfile-1 ] && return 3
684 [ -e $DIR/$tfile-2 ] && return 4
687 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
689 # tests for bug 2104; completion without crashing is success. The close is
690 # stale, but we always return 0 for close, so the app never sees it.
692 multiop_bg_pause $DIR/$tfile O_c || return 2
694 multiop_bg_pause $DIR/$tfile O_c || return 3
697 df $MOUNT || sleep 1 && df $MOUNT || return 1
703 run_test 32 "close() notices client eviction; close() after client eviction"
705 # Abort recovery before client complete
707 replay_barrier $SINGLEMDS
708 createmany -o $DIR/$tfile-%d 100
709 fail_abort $SINGLEMDS
710 # this file should be gone, because the replay was aborted
711 $CHECKSTAT -t file $DIR/$tfile-* && return 3
712 unlinkmany $DIR/$tfile-%d 0 100
715 run_test 33 "abort recovery before client does replay"
719 replay_barrier $SINGLEMDS
720 createmany -o $DIR/$tfile-%d 10
721 fail_abort $SINGLEMDS
722 unlinkmany $DIR/$tfile-%d 0 10
723 # recreate shouldn't fail
724 createmany -o $DIR/$tfile-%d 10 || return 3
725 unlinkmany $DIR/$tfile-%d 0 10
728 run_test 33a "fid shouldn't be reused after abort recovery"
731 multiop_bg_pause $DIR/$tfile O_c || return 2
735 replay_barrier $SINGLEMDS
736 fail_abort $SINGLEMDS
738 [ -e $DIR/$tfile ] && return 1
742 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
744 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
748 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
749 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
754 # give a chance to remove from MDS
755 fail_abort $SINGLEMDS
756 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
758 run_test 35 "test recovery from llog for unlink op"
760 # b=2432 resent cancel after replay uses wrong cookie,
761 # so don't resend cancels
763 replay_barrier $SINGLEMDS
765 checkstat $DIR/$tfile
766 facet_failover $SINGLEMDS
768 if dmesg | grep "unknown lock cookie"; then
769 echo "cancel after replay failed"
773 run_test 36 "don't resend cancel"
776 # directory orphans can't be unlinked from PENDING directory
778 rmdir $DIR/$tfile 2>/dev/null
779 multiop_bg_pause $DIR/$tfile dD_c || return 2
783 replay_barrier $SINGLEMDS
784 # clear the dmesg buffer so we only see errors from this recovery
786 fail_abort $SINGLEMDS
788 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
792 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
795 createmany -o $DIR/$tfile-%d 800
796 unlinkmany $DIR/$tfile-%d 0 400
797 replay_barrier $SINGLEMDS
799 unlinkmany $DIR/$tfile-%d 400 400
801 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
803 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
805 test_39() { # bug 4176
806 createmany -o $DIR/$tfile-%d 800
807 replay_barrier $SINGLEMDS
808 unlinkmany $DIR/$tfile-%d 0 400
810 unlinkmany $DIR/$tfile-%d 400 400
812 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
814 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
817 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
822 $LCTL mark multiop $MOUNT/$tfile OS_c
823 multiop $MOUNT/$tfile OS_c &
825 writeme -s $MOUNT/${tfile}-2 &
828 facet_failover $SINGLEMDS
829 #define OBD_FAIL_MDS_CONNECT_NET 0x117
830 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000117"
832 stat1=`count_ost_writes`
834 stat2=`count_ost_writes`
835 echo "$stat1, $stat2"
836 if [ $stat1 -lt $stat2 ]; then
837 echo "writes continuing during recovery"
840 echo "writes not continuing during recovery, bug 2477"
843 echo "waiting for writeme $WRITE_PID"
847 echo "waiting for multiop $PID"
848 wait $PID || return 2
849 do_facet client munlink $MOUNT/$tfile || return 3
850 do_facet client munlink $MOUNT/${tfile}-2 || return 3
853 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
857 # make sure that a read to one osc doesn't try to double-unlock its page just
858 # because another osc is invalid. trigger_group_io used to mistakenly return
859 # an error if any oscs were invalid even after having successfully put rpcs
860 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
861 # the page, guarnateeing that the unlock from the RPC completion would
862 # assert on trying to unlock the unlocked page.
864 [ $OSTCOUNT -lt 2 ] && \
865 skip "skipping test 41: we don't have a second OST to test with" && \
868 local f=$MOUNT/$tfile
869 # make sure the start of the file is ost1
870 lfs setstripe $f -s $((128 * 1024)) -i 0
871 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
873 # fail ost2 and read from ost1
874 local osc2dev=`do_facet mds "lctl get_param -n devices | grep ${ost2_svc}-osc-MDT0000" | awk '{print $1}'`
875 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices && return 4
876 do_facet mds $LCTL --device $osc2dev deactivate || return 1
877 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
878 do_facet mds $LCTL --device $osc2dev activate || return 2
881 run_test 41 "read from a valid osc while other oscs are invalid"
883 # test MDS recovery after ost failure
885 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
886 createmany -o $DIR/$tfile-%d 800
888 unlinkmany $DIR/$tfile-%d 0 400
890 sysctl -w lnet.debug=-1
893 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
894 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
895 #[ $blocks_after -lt $blocks ] || return 1
896 echo wait for MDS to timeout and recover
897 sleep $((TIMEOUT * 2))
899 unlinkmany $DIR/$tfile-%d 400 400
900 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
902 run_test 42 "recovery after ost failure"
904 # timeout in MDS/OST recovery RPC will LBUG MDS
905 test_43() { # bug 2530
906 replay_barrier $SINGLEMDS
908 # OBD_FAIL_OST_CREATE_NET 0x204
909 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
912 do_facet ost1 "sysctl -w lustre.fail_loc=0"
916 run_test 43 "mds osc import failure during recovery; don't LBUG"
919 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
920 [ "$mdcdev" ] || exit 2
921 for i in `seq 1 10`; do
922 #define OBD_FAIL_TGT_CONN_RACE 0x701
923 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000701"
924 $LCTL --device $mdcdev recover
927 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
930 run_test 44 "race in target handle connect"
933 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
934 [ "$mdcdev" ] || exit 2
935 for i in `seq 1 10`; do
936 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
937 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000704"
938 $LCTL --device $mdcdev recover
941 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
944 run_test 44b "race in target handle connect"
946 # Handle failed close
948 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
949 [ "$mdcdev" ] || exit 2
950 $LCTL --device $mdcdev recover
952 multiop_bg_pause $DIR/$tfile O_c || return 1
955 # This will cause the CLOSE to fail before even
956 # allocating a reply buffer
957 $LCTL --device $mdcdev deactivate || return 4
961 wait $pid || return 1
963 $LCTL --device $mdcdev activate || return 5
966 $CHECKSTAT -t file $DIR/$tfile || return 2
969 run_test 45 "Handle failed close"
973 drop_reply "touch $DIR/$tfile"
975 # ironically, the previous test, 45, will cause a real forced close,
976 # so just look for one for this test
977 dmesg | grep -i "force closing client file handle for $tfile" && return 1
980 run_test 46 "Don't leak file handle after open resend (3325)"
982 test_47() { # bug 2824
983 # create some files to make sure precreate has been done on all
984 # OSTs. (just in case this test is run independently)
985 createmany -o $DIR/$tfile 20 || return 1
987 # OBD_FAIL_OST_CREATE_NET 0x204
989 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
990 df $MOUNT || return 2
992 # let the MDS discover the OST failure, attempt to recover, fail
994 sleep $((3 * TIMEOUT))
996 # Without 2824, this createmany would hang
997 createmany -o $DIR/$tfile 20 || return 3
998 unlinkmany $DIR/$tfile 20 || return 4
1000 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1003 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1006 replay_barrier $SINGLEMDS
1007 createmany -o $DIR/$tfile 20 || return 1
1008 # OBD_FAIL_OST_EROFS 0x216
1010 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000216"
1011 df $MOUNT || return 2
1013 createmany -o $DIR/$tfile 20 20 || return 2
1014 unlinkmany $DIR/$tfile 40 || return 3
1016 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1019 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1022 local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
1023 [ "$oscdev" ] || return 1
1024 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1025 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1026 # give the mds_lov_sync threads a chance to run
1029 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1031 # b3764 timed out lock replay
1034 cancel_lru_locks mdc
1036 multiop $DIR/$tfile s || return 1
1037 replay_barrier $SINGLEMDS
1038 #define OBD_FAIL_LDLM_REPLY 0x30c
1039 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000030c"
1040 fail $SINGLEMDS || return 2
1041 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1043 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1045 run_test 52 "time out lock replay (3764)"
1047 # bug 3462 - simultaneous MDC requests
1049 mkdir -p $DIR/${tdir}-1
1050 mkdir -p $DIR/${tdir}-2
1051 multiop $DIR/${tdir}-1/f O_c &
1053 # give multiop a change to open
1056 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1057 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
1058 kill -USR1 $close_pid
1059 cancel_lru_locks mdc # force the close
1060 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1062 mcreate $DIR/${tdir}-2/f || return 1
1064 # close should still be here
1065 [ -d /proc/$close_pid ] || return 2
1067 replay_barrier_nodf $SINGLEMDS
1069 wait $close_pid || return 3
1071 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1072 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1073 rm -rf $DIR/${tdir}-*
1075 run_test 53a "|X| close request while two MDC requests in flight"
1078 mkdir -p $DIR/${tdir}-1
1079 mkdir -p $DIR/${tdir}-2
1080 multiop $DIR/${tdir}-1/f O_c &
1083 #define OBD_FAIL_MDS_REINT_NET 0x107
1084 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
1085 mcreate $DIR/${tdir}-2/f &
1089 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1090 kill -USR1 $close_pid
1091 cancel_lru_locks mdc # force the close
1092 wait $close_pid || return 1
1093 # open should still be here
1094 [ -d /proc/$open_pid ] || return 2
1096 replay_barrier_nodf $SINGLEMDS
1098 wait $open_pid || return 3
1100 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1101 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1102 rm -rf $DIR/${tdir}-*
1104 run_test 53b "|X| open request while two MDC requests in flight"
1107 mkdir -p $DIR/${tdir}-1
1108 mkdir -p $DIR/${tdir}-2
1109 multiop $DIR/${tdir}-1/f O_c &
1112 #define OBD_FAIL_MDS_REINT_NET 0x107
1113 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
1114 mcreate $DIR/${tdir}-2/f &
1118 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1119 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
1120 kill -USR1 $close_pid
1121 cancel_lru_locks mdc # force the close
1123 replay_barrier_nodf $SINGLEMDS
1124 fail_nodf $SINGLEMDS
1125 wait $open_pid || return 1
1127 # close should be gone
1128 [ -d /proc/$close_pid ] && return 2
1129 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1131 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1132 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1133 rm -rf $DIR/${tdir}-*
1135 run_test 53c "|X| open request and close request while two MDC requests in flight"
1138 mkdir -p $DIR/${tdir}-1
1139 mkdir -p $DIR/${tdir}-2
1140 multiop $DIR/${tdir}-1/f O_c &
1142 # give multiop a chance to open
1145 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1146 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000013f"
1147 kill -USR1 $close_pid
1148 cancel_lru_locks mdc # force the close
1149 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1150 mcreate $DIR/${tdir}-2/f || return 1
1152 # close should still be here
1153 [ -d /proc/$close_pid ] || return 2
1155 wait $close_pid || return 3
1157 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1158 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1159 rm -rf $DIR/${tdir}-*
1161 run_test 53d "|X| close reply while two MDC requests in flight"
1164 mkdir -p $DIR/${tdir}-1
1165 mkdir -p $DIR/${tdir}-2
1166 multiop $DIR/${tdir}-1/f O_c &
1169 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1170 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x119"
1171 mcreate $DIR/${tdir}-2/f &
1175 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1176 kill -USR1 $close_pid
1177 cancel_lru_locks mdc # force the close
1178 wait $close_pid || return 1
1179 # open should still be here
1180 [ -d /proc/$open_pid ] || return 2
1182 replay_barrier_nodf $SINGLEMDS
1184 wait $open_pid || return 3
1186 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1187 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1188 rm -rf $DIR/${tdir}-*
1190 run_test 53e "|X| open reply while two MDC requests in flight"
1193 mkdir -p $DIR/${tdir}-1
1194 mkdir -p $DIR/${tdir}-2
1195 multiop $DIR/${tdir}-1/f O_c &
1198 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1199 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x119"
1200 mcreate $DIR/${tdir}-2/f &
1204 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1205 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000013f"
1206 kill -USR1 $close_pid
1207 cancel_lru_locks mdc # force the close
1209 replay_barrier_nodf $SINGLEMDS
1210 fail_nodf $SINGLEMDS
1211 wait $open_pid || return 1
1213 # close should be gone
1214 [ -d /proc/$close_pid ] && return 2
1215 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1217 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1218 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1219 rm -rf $DIR/${tdir}-*
1221 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1224 mkdir -p $DIR/${tdir}-1
1225 mkdir -p $DIR/${tdir}-2
1226 multiop $DIR/${tdir}-1/f O_c &
1229 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1230 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x119"
1231 mcreate $DIR/${tdir}-2/f &
1235 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1236 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
1237 kill -USR1 $close_pid
1238 cancel_lru_locks mdc # force the close
1240 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1241 replay_barrier_nodf $SINGLEMDS
1242 fail_nodf $SINGLEMDS
1243 wait $open_pid || return 1
1245 # close should be gone
1246 [ -d /proc/$close_pid ] && return 2
1248 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1249 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1250 rm -rf $DIR/${tdir}-*
1252 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1255 mkdir -p $DIR/${tdir}-1
1256 mkdir -p $DIR/${tdir}-2
1257 multiop $DIR/${tdir}-1/f O_c &
1260 #define OBD_FAIL_MDS_REINT_NET 0x107
1261 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
1262 mcreate $DIR/${tdir}-2/f &
1266 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1267 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000013f"
1268 kill -USR1 $close_pid
1269 cancel_lru_locks mdc # force the close
1272 replay_barrier_nodf $SINGLEMDS
1273 fail_nodf $SINGLEMDS
1274 wait $open_pid || return 1
1276 # close should be gone
1277 [ -d /proc/$close_pid ] && return 2
1278 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1280 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1281 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1282 rm -rf $DIR/${tdir}-*
1284 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1286 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1288 #b3761 ASSERTION(hash != 0) failed
1290 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1291 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012b"
1293 # give touch a chance to run
1295 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1299 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1301 #b3440 ASSERTION(rec->ur_fid2->id) failed
1303 ln -s foo $DIR/$tfile
1304 replay_barrier $SINGLEMDS
1305 #drop_reply "cat $DIR/$tfile"
1309 run_test 56 "don't replay a symlink open request (3440)"
1311 #recovery one mds-ost setattr from llog
1313 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1314 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1316 replay_barrier $SINGLEMDS
1319 $CHECKSTAT -t file $DIR/$tfile || return 1
1320 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1323 run_test 57 "test recovery from llog for setattr op"
1325 #recovery many mds-ost setattr from llog
1327 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1328 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1329 createmany -o $DIR/$tdir/$tfile-%d 2500
1330 replay_barrier $SINGLEMDS
1333 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1334 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1335 unlinkmany $DIR/$tdir/$tfile-%d 2500
1338 run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
1340 # log_commit_thread vs filter_destroy race used to lead to import use after free
1343 createmany -o $DIR/$tdir/$tfile-%d 200
1345 unlinkmany $DIR/$tdir/$tfile-%d 200
1346 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1347 do_facet ost1 "sysctl -w lustre.fail_loc=0x507"
1350 do_facet ost1 "sysctl -w lustre.fail_loc=0x0"
1354 run_test 59 "test log_commit_thread vs filter_destroy race"
1356 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1357 # bug 12086: should no oops and No ctxt error for this test
1359 createmany -o $DIR/$tdir/$tfile-%d 200
1360 replay_barrier $SINGLEMDS
1361 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1363 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1364 local no_ctxt=`dmesg | grep "No ctxt"`
1365 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1367 run_test 60 "test llog post recovery init vs llog unlink"
1369 #test race llog recovery thread vs llog cleanup
1372 createmany -o $DIR/$tdir/$tfile-%d 800
1374 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1375 unlinkmany $DIR/$tdir/$tfile-%d 800
1376 set_nodes_failloc "$(osts_nodes)" 0x80000221
1381 set_nodes_failloc "$(osts_nodes)" 0x0
1383 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1386 run_test 61 "test race llog recovery vs llog cleanup"
1388 #test race mds llog sync vs llog cleanup
1390 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x140
1391 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000140"
1392 facet_failover $SINGLEMDS
1395 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1397 run_test 61b "test race mds llog sync vs llog cleanup"
1399 #test race cancel cookie cb vs llog cleanup
1401 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1403 set_nodes_failloc "$(osts_nodes)" 0x80000222
1407 set_nodes_failloc "$(osts_nodes)" 0x0
1409 run_test 61c "test race mds llog sync vs llog cleanup"
1411 # start multi-client tests
1413 [ -z "$CLIENTS" ] && \
1414 { skip "Need two or more clients." && return; }
1415 [ $CLIENTCOUNT -lt 2 ] && \
1416 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1418 echo "mount clients $CLIENTS ..."
1419 zconf_mount_clients $CLIENTS $DIR
1421 local clients=${CLIENTS//,/ }
1422 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1423 for CLIENT in $clients; do
1424 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1425 of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
1426 error "dd failed on $CLIENT"
1429 local prev_client=$(echo $clients | sed 's/^.* \(\w\+\)$/\1/')
1430 for C in ${CLIENTS//,/ }; do
1431 do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
1432 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
1438 zconf_umount_clients $CLIENTS $DIR
1440 run_test 70a "check multi client t-f"
1443 [ -z "$CLIENTS" ] && \
1444 { skip "Need two or more clients." && return; }
1445 [ $CLIENTCOUNT -lt 2 ] && \
1446 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1448 zconf_mount_clients $CLIENTS $DIR
1450 local duration="-t 60"
1451 local cmd="rundbench 1 $duration "
1453 for CLIENT in ${CLIENTS//,/ }; do
1454 $PDSH $CLIENT "set -x; PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:${DBENCH_LIB} DBENCH_LIB=${DBENCH_LIB} $cmd" &
1456 echo $PID >pid.$CLIENT
1457 echo "Started load PID=`cat pid.$CLIENT`"
1460 replay_barrier $SINGLEMDS
1461 sleep 3 # give clients a time to do operations
1463 log "$TESTNAME fail mds 1"
1466 # wait for client to reconnect to MDS
1469 for CLIENT in ${CLIENTS//,/ }; do
1470 PID=`cat pid.$CLIENT`
1473 echo "load on ${CLIENT} returned $rc"
1476 zconf_umount_clients $CLIENTS $DIR
1478 run_test 70b "mds recovery; $CLIENTCOUNT clients"
1479 # end multi-client tests
1481 equals_msg `basename $0`: test complete, cleaning up
1482 check_and_cleanup_lustre
1483 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true