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
22 ALWAYS_EXCEPT="0b 39 $REPLAY_SINGLE_EXCEPT"
26 cleanup_and_setup_lustre
33 replay_barrier $SINGLEMDS
37 run_test 0 "empty replay"
40 # this test attempts to trigger a race in the precreation code,
41 # and must run before any other objects are created on the filesystem
43 createmany -o $DIR/$tfile 20 || return 1
44 unlinkmany $DIR/$tfile 20 || return 2
46 run_test 0b "ensure object created after recover exists. (3284)"
52 local file=`ls /proc/fs/lustre/seq/cli-srv-$mds-mdc-*/width`
59 local file=`ls /proc/fs/lustre/seq/cli-srv-$mds-mdc-*/width`
63 # This test should pass for single-mds and multi-mds configs.
64 # But for different configurations it tests different things.
68 # (1) fld_create replay should happen;
70 # (2) fld_create replay should not return -EEXISTS, if it does
71 # this means sequence manager recovery code is buggy and allocated
72 # same sequence two times after recovery.
76 # (1) fld_create replay may not happen, because its home MDS is
77 # MDS2 which is not involved to revovery;
79 # (2) as fld_create does not happen on MDS1, it does not make any
82 local label=`mdsdevlabel 1`
83 [ -z "$label" ] && echo "No label for mds1" && return 1
85 replay_barrier $SINGLEMDS
86 local sw=`seq_get_width $label`
88 # make seq manager switch to next sequence each
89 # time as new fid is needed.
90 seq_set_width $label 1
92 # make sure that fld has created at least one new
94 touch $DIR/$tfile || return 2
95 seq_set_width $label $sw
97 # fail $SINGLEMDS and start recovery, replay RPCs, etc.
100 # wait for recovery finish
104 # flush fld cache and dentry cache to make it lookup
105 # created entry instead of revalidating existent one
107 zconf_mount `hostname` $MOUNT
109 # issue lookup which should call fld lookup which
110 # should fail if client did not replay fld create
111 # correctly and server has no fld entry
112 touch $DIR/$tfile || return 3
113 rm $DIR/$tfile || return 4
115 run_test 0c "fld create"
118 replay_barrier $SINGLEMDS
121 $CHECKSTAT -t file $DIR/$tfile || return 1
124 run_test 1 "simple create"
127 replay_barrier $SINGLEMDS
130 $CHECKSTAT -t file $DIR/$tfile || return 1
137 replay_barrier $SINGLEMDS
140 $CHECKSTAT -t file $DIR/$tfile || return 1
146 replay_barrier $SINGLEMDS
148 o_directory $DIR/$tfile
150 $CHECKSTAT -t file $DIR/$tfile || return 2
153 run_test 3a "replay failed open(O_DIRECTORY)"
156 replay_barrier $SINGLEMDS
157 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
158 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000114"
160 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
162 $CHECKSTAT -t file $DIR/$tfile && return 2
165 run_test 3b "replay failed open -ENOMEM"
168 replay_barrier $SINGLEMDS
169 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
170 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000128"
172 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
175 $CHECKSTAT -t file $DIR/$tfile && return 2
178 run_test 3c "replay failed open -ENOMEM"
181 replay_barrier $SINGLEMDS
182 for i in `seq 10`; do
183 echo "tag-$i" > $DIR/$tfile-$i
186 for i in `seq 10`; do
187 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
190 run_test 4 "|x| 10 open(O_CREAT)s"
193 replay_barrier $SINGLEMDS
196 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
198 run_test 4b "|x| rm 10 files"
200 # The idea is to get past the first block of precreated files on both
201 # osts, and then replay.
203 replay_barrier $SINGLEMDS
204 for i in `seq 220`; do
205 echo "tag-$i" > $DIR/$tfile-$i
208 for i in `seq 220`; do
209 grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
213 # waiting for commitment of removal
215 run_test 5 "|x| 220 open(O_CREAT)"
219 replay_barrier $SINGLEMDS
221 mcreate $DIR/$tdir/$tfile
223 $CHECKSTAT -t dir $DIR/$tdir || return 1
224 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
226 # waiting for log process thread
228 run_test 6 "mkdir + contained create"
231 replay_barrier $SINGLEMDS
234 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
236 run_test 6b "|X| rmdir"
240 replay_barrier $SINGLEMDS
241 mcreate $DIR/$tdir/$tfile
243 $CHECKSTAT -t dir $DIR/$tdir || return 1
244 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
247 run_test 7 "mkdir |X| contained create"
250 # make sure no side-effect from previous test.
252 replay_barrier $SINGLEMDS
253 multiop $DIR/$tfile mo_c &
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 $DIR/$tfile o_tSc &
315 # give multiop a chance to open
318 replay_barrier $SINGLEMDS
320 wait $pid || return 1
323 [ -e $DIR/$tfile ] && return 2
326 run_test 12 "open, unlink |X| close"
329 # 1777 - replay open after committed chmod that would make
330 # a regular open a failure
333 multiop $DIR/$tfile O_wc &
335 # give multiop a chance to open
338 $CHECKSTAT -p 0 $DIR/$tfile
339 replay_barrier $SINGLEMDS
342 wait $pid || return 1
344 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
347 run_test 13 "open chmod 0 |x| write close"
350 multiop $DIR/$tfile O_tSc &
352 # give multiop a chance to open
355 replay_barrier $SINGLEMDS
356 kill -USR1 $pid || return 1
357 wait $pid || return 2
360 [ -e $DIR/$tfile ] && return 3
363 run_test 14 "open(O_CREAT), unlink |X| close"
366 multiop $DIR/$tfile O_tSc &
368 # give multiop a chance to open
371 replay_barrier $SINGLEMDS
372 touch $DIR/g11 || return 1
374 wait $pid || return 2
377 [ -e $DIR/$tfile ] && return 3
378 touch $DIR/h11 || return 4
381 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
385 replay_barrier $SINGLEMDS
388 mcreate $DIR/$tfile-2
390 [ -e $DIR/$tfile ] && return 1
391 [ -e $DIR/$tfile-2 ] || return 2
392 munlink $DIR/$tfile-2 || return 3
394 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
397 replay_barrier $SINGLEMDS
398 multiop $DIR/$tfile O_c &
400 # give multiop a chance to open
403 kill -USR1 $pid || return 1
404 wait $pid || return 2
405 $CHECKSTAT -t file $DIR/$tfile || return 3
408 run_test 17 "|X| open(O_CREAT), |replay| close"
411 replay_barrier $SINGLEMDS
412 multiop $DIR/$tfile O_tSc &
414 # give multiop a chance to open
417 touch $DIR/$tfile-2 || return 1
418 echo "pid: $pid will close"
420 wait $pid || return 2
423 [ -e $DIR/$tfile ] && return 3
424 [ -e $DIR/$tfile-2 ] || return 4
425 # this touch frequently fails
426 touch $DIR/$tfile-3 || return 5
427 munlink $DIR/$tfile-2 || return 6
428 munlink $DIR/$tfile-3 || return 7
431 run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
433 # bug 1855 (a simpler form of test_11 above)
435 replay_barrier $SINGLEMDS
437 echo "old" > $DIR/$tfile
438 mv $DIR/$tfile $DIR/$tfile-2
439 grep old $DIR/$tfile-2
441 grep old $DIR/$tfile-2 || return 2
443 run_test 19 "|X| mcreate, open, write, rename "
446 replay_barrier $SINGLEMDS
447 multiop $DIR/$tfile O_tSc &
449 # give multiop a chance to open
455 wait $pid || return 1
456 [ -e $DIR/$tfile ] && return 2
459 run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
461 test_20b() { # bug 10480
462 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
464 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
466 while [ ! -e $DIR/$tfile ] ; do
467 usleep 60 # give dd a chance to start
470 lfs getstripe $DIR/$tfile || return 1
471 rm -f $DIR/$tfile || return 2 # make it an orphan
473 df -P $DIR || df -P $DIR || true # reconnect
475 fail $SINGLEMDS # start orphan recovery
476 df -P $DIR || df -P $DIR || true # reconnect
477 wait_mds_recovery_done || error "MDS recovery not done"
479 # FIXME just because recovery is done doesn't mean we've finished
480 # orphan cleanup. Fake it with a sleep for now...
482 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
483 log "before $BEFOREUSED, after $AFTERUSED"
484 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
485 error "after $AFTERUSED > before $BEFOREUSED"
488 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
490 test_20c() { # bug 10480
491 multiop $DIR/$tfile Ow_c &
493 # give multiop a chance to open
500 df -P $DIR || df -P $DIR || true # reconnect
503 test -s $DIR/$tfile || error "File was truncated"
507 run_test 20c "check that client eviction does not affect file content"
510 replay_barrier $SINGLEMDS
511 multiop $DIR/$tfile O_tSc &
513 # give multiop a chance to open
516 touch $DIR/g11 || return 1
520 wait $pid || return 2
521 [ -e $DIR/$tfile ] && return 3
522 touch $DIR/h11 || return 4
525 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
528 multiop $DIR/$tfile O_tSc &
530 # give multiop a chance to open
533 replay_barrier $SINGLEMDS
538 wait $pid || return 1
539 [ -e $DIR/$tfile ] && return 2
542 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
545 multiop $DIR/$tfile O_tSc &
547 # give multiop a chance to open
550 replay_barrier $SINGLEMDS
552 touch $DIR/g11 || return 1
556 wait $pid || return 2
557 [ -e $DIR/$tfile ] && return 3
558 touch $DIR/h11 || return 4
561 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
564 multiop $DIR/$tfile O_tSc &
566 # give multiop a chance to open
569 replay_barrier $SINGLEMDS
573 wait $pid || return 1
574 [ -e $DIR/$tfile ] && return 2
577 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
580 multiop $DIR/$tfile O_tSc &
582 # give multiop a chance to open
586 replay_barrier $SINGLEMDS
589 wait $pid || return 1
590 [ -e $DIR/$tfile ] && return 2
593 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
596 replay_barrier $SINGLEMDS
597 multiop $DIR/$tfile-1 O_tSc &
599 multiop $DIR/$tfile-2 O_tSc &
601 # give multiop a chance to open
606 wait $pid2 || return 1
610 wait $pid1 || return 2
611 [ -e $DIR/$tfile-1 ] && return 3
612 [ -e $DIR/$tfile-2 ] && return 4
615 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
618 replay_barrier $SINGLEMDS
619 multiop $DIR/$tfile-1 O_tSc &
621 multiop $DIR/$tfile-2 O_tSc &
623 # give multiop a chance to open
630 wait $pid1 || return 1
632 wait $pid2 || return 2
633 [ -e $DIR/$tfile-1 ] && return 3
634 [ -e $DIR/$tfile-2 ] && return 4
637 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
640 multiop $DIR/$tfile-1 O_tSc &
642 multiop $DIR/$tfile-2 O_tSc &
644 # give multiop a chance to open
646 replay_barrier $SINGLEMDS
650 wait $pid2 || return 1
654 wait $pid1 || return 2
655 [ -e $DIR/$tfile-1 ] && return 3
656 [ -e $DIR/$tfile-2 ] && return 4
659 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
662 multiop $DIR/$tfile-1 O_tSc &
664 multiop $DIR/$tfile-2 O_tSc &
666 # give multiop a chance to open
668 replay_barrier $SINGLEMDS
674 wait $pid1 || return 1
676 wait $pid2 || return 2
677 [ -e $DIR/$tfile-1 ] && return 3
678 [ -e $DIR/$tfile-2 ] && return 4
681 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
684 multiop $DIR/$tfile-1 O_tSc &
686 multiop $DIR/$tfile-2 O_tSc &
688 # give multiop a chance to open
693 replay_barrier $SINGLEMDS
696 wait $pid1 || return 1
698 wait $pid2 || return 2
699 [ -e $DIR/$tfile-1 ] && return 3
700 [ -e $DIR/$tfile-2 ] && return 4
703 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
706 multiop $DIR/$tfile-1 O_tSc &
708 multiop $DIR/$tfile-2 O_tSc &
710 # give multiop a chance to open
714 replay_barrier $SINGLEMDS
718 wait $pid1 || return 1
720 wait $pid2 || return 2
721 [ -e $DIR/$tfile-1 ] && return 3
722 [ -e $DIR/$tfile-2 ] && return 4
725 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
727 # tests for bug 2104; completion without crashing is success. The close is
728 # stale, but we always return 0 for close, so the app never sees it.
730 multiop $DIR/$tfile O_c &
732 multiop $DIR/$tfile O_c &
734 # give multiop a chance to open
737 df $MOUNT || sleep 1 && df $MOUNT || return 1
743 run_test 32 "close() notices client eviction; close() after client eviction"
745 # Abort recovery before client complete
747 replay_barrier $SINGLEMDS
748 createmany -o $DIR/$tfile-%d 100
749 fail_abort $SINGLEMDS
750 # this file should be gone, because the replay was aborted
751 $CHECKSTAT -t file $DIR/$tfile-* && return 3
752 unlinkmany $DIR/$tfile-%d 0 100
755 run_test 33 "abort recovery before client does replay"
759 replay_barrier $SINGLEMDS
760 createmany -o $DIR/$tfile-%d 10
761 fail_abort $SINGLEMDS
762 unlinkmany $DIR/$tfile-%d 0 10
763 # recreate shouldn't fail
764 createmany -o $DIR/$tfile-%d 10 || return 3
765 unlinkmany $DIR/$tfile-%d 0 10
768 run_test 33a "fid shouldn't be reused after abort recovery"
771 multiop $DIR/$tfile O_c &
773 # give multiop a chance to open
777 replay_barrier $SINGLEMDS
778 fail_abort $SINGLEMDS
780 [ -e $DIR/$tfile ] && return 1
784 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
786 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
790 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
791 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
796 # give a chance to remove from MDS
797 fail_abort $SINGLEMDS
798 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
800 run_test 35 "test recovery from llog for unlink op"
802 # b=2432 resent cancel after replay uses wrong cookie,
803 # so don't resend cancels
805 replay_barrier $SINGLEMDS
807 checkstat $DIR/$tfile
808 facet_failover $SINGLEMDS
810 if dmesg | grep "unknown lock cookie"; then
811 echo "cancel after replay failed"
815 run_test 36 "don't resend cancel"
818 # directory orphans can't be unlinked from PENDING directory
820 rmdir $DIR/$tfile 2>/dev/null
821 multiop $DIR/$tfile dD_c &
823 # give multiop a chance to open
827 replay_barrier $SINGLEMDS
828 # clear the dmesg buffer so we only see errors from this recovery
830 fail_abort $SINGLEMDS
832 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
836 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
839 createmany -o $DIR/$tfile-%d 800
840 unlinkmany $DIR/$tfile-%d 0 400
841 replay_barrier $SINGLEMDS
843 unlinkmany $DIR/$tfile-%d 400 400
845 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
847 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
849 test_39() { # bug 4176
850 createmany -o $DIR/$tfile-%d 800
851 replay_barrier $SINGLEMDS
852 unlinkmany $DIR/$tfile-%d 0 400
854 unlinkmany $DIR/$tfile-%d 400 400
856 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
858 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
861 awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' $LPROC/osc/*/stats
866 $LCTL mark multiop $MOUNT/$tfile OS_c
867 multiop $MOUNT/$tfile OS_c &
869 writeme -s $MOUNT/${tfile}-2 &
872 facet_failover $SINGLEMDS
873 #define OBD_FAIL_MDS_CONNECT_NET 0x117
874 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000117"
876 stat1=`count_ost_writes`
878 stat2=`count_ost_writes`
879 echo "$stat1, $stat2"
880 if [ $stat1 -lt $stat2 ]; then
881 echo "writes continuing during recovery"
884 echo "writes not continuing during recovery, bug 2477"
887 echo "waiting for writeme $WRITE_PID"
891 echo "waiting for multiop $PID"
892 wait $PID || return 2
893 do_facet client munlink $MOUNT/$tfile || return 3
894 do_facet client munlink $MOUNT/${tfile}-2 || return 3
897 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
901 # make sure that a read to one osc doesn't try to double-unlock its page just
902 # because another osc is invalid. trigger_group_io used to mistakenly return
903 # an error if any oscs were invalid even after having successfully put rpcs
904 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
905 # the page, guarnateeing that the unlock from the RPC completion would
906 # assert on trying to unlock the unlocked page.
908 [ $OSTCOUNT -lt 2 ] && \
909 skip "skipping test 41: we don't have a second OST to test with" && \
912 local f=$MOUNT/$tfile
913 # make sure the start of the file is ost1
914 lfs setstripe $f $((128 * 1024)) 0 0
915 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
917 # fail ost2 and read from ost1
918 local osc2dev=`do_facet mds "grep ${ost2_svc}-osc-MDT0000 $LPROC/devices" | awk '{print $1}'`
919 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && cat $LPROC/devices && return 4
920 do_facet mds $LCTL --device $osc2dev deactivate || return 1
921 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
922 do_facet mds $LCTL --device $osc2dev activate || return 2
925 run_test 41 "read from a valid osc while other oscs are invalid"
927 # test MDS recovery after ost failure
929 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
930 createmany -o $DIR/$tfile-%d 800
932 unlinkmany $DIR/$tfile-%d 0 400
934 sysctl -w lnet.debug=-1
937 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
938 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
939 #[ $blocks_after -lt $blocks ] || return 1
940 echo wait for MDS to timeout and recover
941 sleep $((TIMEOUT * 2))
943 unlinkmany $DIR/$tfile-%d 400 400
944 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
946 run_test 42 "recovery after ost failure"
948 # timeout in MDS/OST recovery RPC will LBUG MDS
949 test_43() { # bug 2530
950 replay_barrier $SINGLEMDS
952 # OBD_FAIL_OST_CREATE_NET 0x204
953 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
956 do_facet ost1 "sysctl -w lustre.fail_loc=0"
960 run_test 43 "mds osc import failure during recovery; don't LBUG"
963 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
964 [ "$mdcdev" ] || exit 2
965 for i in `seq 1 10`; do
966 #define OBD_FAIL_TGT_CONN_RACE 0x701
967 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000701"
968 $LCTL --device $mdcdev recover
971 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
974 run_test 44 "race in target handle connect"
977 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
978 [ "$mdcdev" ] || exit 2
979 for i in `seq 1 10`; do
980 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
981 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000704"
982 $LCTL --device $mdcdev recover
985 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
988 run_test 44b "race in target handle connect"
990 # Handle failed close
992 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
993 [ "$mdcdev" ] || exit 2
994 $LCTL --device $mdcdev recover
996 multiop $DIR/$tfile O_c &
1000 # This will cause the CLOSE to fail before even
1001 # allocating a reply buffer
1002 $LCTL --device $mdcdev deactivate || return 4
1006 wait $pid || return 1
1008 $LCTL --device $mdcdev activate || return 5
1011 $CHECKSTAT -t file $DIR/$tfile || return 2
1014 run_test 45 "Handle failed close"
1018 drop_reply "touch $DIR/$tfile"
1020 # ironically, the previous test, 45, will cause a real forced close,
1021 # so just look for one for this test
1022 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1025 run_test 46 "Don't leak file handle after open resend (3325)"
1027 test_47() { # bug 2824
1028 # create some files to make sure precreate has been done on all
1029 # OSTs. (just in case this test is run independently)
1030 createmany -o $DIR/$tfile 20 || return 1
1032 # OBD_FAIL_OST_CREATE_NET 0x204
1034 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
1035 df $MOUNT || return 2
1037 # let the MDS discover the OST failure, attempt to recover, fail
1038 # and recover again.
1039 sleep $((3 * TIMEOUT))
1041 # Without 2824, this createmany would hang
1042 createmany -o $DIR/$tfile 20 || return 3
1043 unlinkmany $DIR/$tfile 20 || return 4
1045 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1048 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1051 replay_barrier $SINGLEMDS
1052 createmany -o $DIR/$tfile 20 || return 1
1053 # OBD_FAIL_OST_EROFS 0x216
1055 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000216"
1056 df $MOUNT || return 2
1058 createmany -o $DIR/$tfile 20 20 || return 2
1059 unlinkmany $DIR/$tfile 40 || return 3
1061 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1064 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1067 local oscdev=`do_facet $SINGLEMDS grep ${ost1_svc}-osc-MDT0000 $LPROC/devices | awk '{print $1}'`
1068 [ "$oscdev" ] || return 1
1069 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1070 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1071 # give the mds_lov_sync threads a chance to run
1074 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1076 # b3764 timed out lock replay
1079 cancel_lru_locks mdc
1081 multiop $DIR/$tfile s || return 1
1082 replay_barrier $SINGLEMDS
1083 #define OBD_FAIL_LDLM_REPLY 0x30c
1084 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000030c"
1085 fail $SINGLEMDS || return 2
1086 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1088 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1090 run_test 52 "time out lock replay (3764)"
1092 #b_cray 53 "|X| open request and close reply while two MDC requests in flight"
1093 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1095 #b3761 ASSERTION(hash != 0) failed
1097 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1098 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012b"
1100 # give touch a chance to run
1102 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1106 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1108 #b3440 ASSERTION(rec->ur_fid2->id) failed
1110 ln -s foo $DIR/$tfile
1111 replay_barrier $SINGLEMDS
1112 #drop_reply "cat $DIR/$tfile"
1116 run_test 56 "don't replay a symlink open request (3440)"
1118 #recovery one mds-ost setattr from llog
1120 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1121 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1123 replay_barrier $SINGLEMDS
1126 $CHECKSTAT -t file $DIR/$tfile || return 1
1127 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1130 run_test 57 "test recovery from llog for setattr op"
1132 #recovery many mds-ost setattr from llog
1134 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1135 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1137 createmany -o $DIR/$tdir/$tfile-%d 2500
1138 replay_barrier $SINGLEMDS
1141 $CHECKSTAT -t file $DIR/$tdir/$tfile-* || return 1
1142 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1143 unlinkmany $DIR/$tdir/$tfile-%d 2500
1146 run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
1148 # log_commit_thread vs filter_destroy race used to lead to import use after free
1152 createmany -o $DIR/$tdir/$tfile-%d 200
1154 unlinkmany $DIR/$tdir/$tfile-%d 200
1155 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1156 do_facet ost1 "sysctl -w lustre.fail_loc=0x507"
1159 do_facet ost1 "sysctl -w lustre.fail_loc=0x0"
1163 run_test 59 "test log_commit_thread vs filter_destroy race"
1165 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1166 # bug 12086: should no oops and No ctxt error for this test
1169 createmany -o $DIR/$tdir/$tfile-%d 200
1170 replay_barrier $SINGLEMDS
1171 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1173 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1174 local no_ctxt=`dmesg | grep "No ctxt"`
1175 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1177 run_test 60 "test llog post recovery init vs llog unlink"
1179 equals_msg `basename $0`: test complete, cleaning up
1180 check_and_cleanup_lustre
1181 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true