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 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
480 log "before $BEFOREUSED, after $AFTERUSED"
481 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
482 error "after $AFTERUSED > before $BEFOREUSED" && return 5
485 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
487 test_20c() { # bug 10480
488 multiop $DIR/$tfile Ow_c &
490 # give multiop a chance to open
497 df -P $DIR || df -P $DIR || true # reconnect
500 test -s $DIR/$tfile || error "File was truncated"
504 run_test 20c "check that client eviction does not affect file content"
507 replay_barrier $SINGLEMDS
508 multiop $DIR/$tfile O_tSc &
510 # give multiop a chance to open
513 touch $DIR/g11 || return 1
517 wait $pid || return 2
518 [ -e $DIR/$tfile ] && return 3
519 touch $DIR/h11 || return 4
522 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
525 multiop $DIR/$tfile O_tSc &
527 # give multiop a chance to open
530 replay_barrier $SINGLEMDS
535 wait $pid || return 1
536 [ -e $DIR/$tfile ] && return 2
539 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
542 multiop $DIR/$tfile O_tSc &
544 # give multiop a chance to open
547 replay_barrier $SINGLEMDS
549 touch $DIR/g11 || return 1
553 wait $pid || return 2
554 [ -e $DIR/$tfile ] && return 3
555 touch $DIR/h11 || return 4
558 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
561 multiop $DIR/$tfile O_tSc &
563 # give multiop a chance to open
566 replay_barrier $SINGLEMDS
570 wait $pid || return 1
571 [ -e $DIR/$tfile ] && return 2
574 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
577 multiop $DIR/$tfile O_tSc &
579 # give multiop a chance to open
583 replay_barrier $SINGLEMDS
586 wait $pid || return 1
587 [ -e $DIR/$tfile ] && return 2
590 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
593 replay_barrier $SINGLEMDS
594 multiop $DIR/$tfile-1 O_tSc &
596 multiop $DIR/$tfile-2 O_tSc &
598 # give multiop a chance to open
603 wait $pid2 || return 1
607 wait $pid1 || return 2
608 [ -e $DIR/$tfile-1 ] && return 3
609 [ -e $DIR/$tfile-2 ] && return 4
612 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
615 replay_barrier $SINGLEMDS
616 multiop $DIR/$tfile-1 O_tSc &
618 multiop $DIR/$tfile-2 O_tSc &
620 # give multiop a chance to open
627 wait $pid1 || return 1
629 wait $pid2 || return 2
630 [ -e $DIR/$tfile-1 ] && return 3
631 [ -e $DIR/$tfile-2 ] && return 4
634 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
637 multiop $DIR/$tfile-1 O_tSc &
639 multiop $DIR/$tfile-2 O_tSc &
641 # give multiop a chance to open
643 replay_barrier $SINGLEMDS
647 wait $pid2 || return 1
651 wait $pid1 || return 2
652 [ -e $DIR/$tfile-1 ] && return 3
653 [ -e $DIR/$tfile-2 ] && return 4
656 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
659 multiop $DIR/$tfile-1 O_tSc &
661 multiop $DIR/$tfile-2 O_tSc &
663 # give multiop a chance to open
665 replay_barrier $SINGLEMDS
671 wait $pid1 || return 1
673 wait $pid2 || return 2
674 [ -e $DIR/$tfile-1 ] && return 3
675 [ -e $DIR/$tfile-2 ] && return 4
678 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
681 multiop $DIR/$tfile-1 O_tSc &
683 multiop $DIR/$tfile-2 O_tSc &
685 # give multiop a chance to open
690 replay_barrier $SINGLEMDS
693 wait $pid1 || return 1
695 wait $pid2 || return 2
696 [ -e $DIR/$tfile-1 ] && return 3
697 [ -e $DIR/$tfile-2 ] && return 4
700 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
703 multiop $DIR/$tfile-1 O_tSc &
705 multiop $DIR/$tfile-2 O_tSc &
707 # give multiop a chance to open
711 replay_barrier $SINGLEMDS
715 wait $pid1 || return 1
717 wait $pid2 || return 2
718 [ -e $DIR/$tfile-1 ] && return 3
719 [ -e $DIR/$tfile-2 ] && return 4
722 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
724 # tests for bug 2104; completion without crashing is success. The close is
725 # stale, but we always return 0 for close, so the app never sees it.
727 multiop $DIR/$tfile O_c &
729 multiop $DIR/$tfile O_c &
731 # give multiop a chance to open
734 df $MOUNT || sleep 1 && df $MOUNT || return 1
740 run_test 32 "close() notices client eviction; close() after client eviction"
742 # Abort recovery before client complete
744 replay_barrier $SINGLEMDS
745 createmany -o $DIR/$tfile-%d 100
746 fail_abort $SINGLEMDS
747 # this file should be gone, because the replay was aborted
748 $CHECKSTAT -t file $DIR/$tfile-* && return 3
749 unlinkmany $DIR/$tfile-%d 0 100
752 run_test 33 "abort recovery before client does replay"
756 replay_barrier $SINGLEMDS
757 createmany -o $DIR/$tfile-%d 10
758 fail_abort $SINGLEMDS
759 unlinkmany $DIR/$tfile-%d 0 10
760 # recreate shouldn't fail
761 createmany -o $DIR/$tfile-%d 10 || return 3
762 unlinkmany $DIR/$tfile-%d 0 10
765 run_test 33a "fid shouldn't be reused after abort recovery"
768 multiop $DIR/$tfile O_c &
770 # give multiop a chance to open
774 replay_barrier $SINGLEMDS
775 fail_abort $SINGLEMDS
777 [ -e $DIR/$tfile ] && return 1
781 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
783 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
787 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
788 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
793 # give a chance to remove from MDS
794 fail_abort $SINGLEMDS
795 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
797 run_test 35 "test recovery from llog for unlink op"
799 # b=2432 resent cancel after replay uses wrong cookie,
800 # so don't resend cancels
802 replay_barrier $SINGLEMDS
804 checkstat $DIR/$tfile
805 facet_failover $SINGLEMDS
807 if dmesg | grep "unknown lock cookie"; then
808 echo "cancel after replay failed"
812 run_test 36 "don't resend cancel"
815 # directory orphans can't be unlinked from PENDING directory
817 rmdir $DIR/$tfile 2>/dev/null
818 multiop $DIR/$tfile dD_c &
820 # give multiop a chance to open
824 replay_barrier $SINGLEMDS
825 # clear the dmesg buffer so we only see errors from this recovery
827 fail_abort $SINGLEMDS
829 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
833 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
836 createmany -o $DIR/$tfile-%d 800
837 unlinkmany $DIR/$tfile-%d 0 400
838 replay_barrier $SINGLEMDS
840 unlinkmany $DIR/$tfile-%d 400 400
842 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
844 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
846 test_39() { # bug 4176
847 createmany -o $DIR/$tfile-%d 800
848 replay_barrier $SINGLEMDS
849 unlinkmany $DIR/$tfile-%d 0 400
851 unlinkmany $DIR/$tfile-%d 400 400
853 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
855 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
858 awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' $LPROC/osc/*/stats
863 $LCTL mark multiop $MOUNT/$tfile OS_c
864 multiop $MOUNT/$tfile OS_c &
866 writeme -s $MOUNT/${tfile}-2 &
869 facet_failover $SINGLEMDS
870 #define OBD_FAIL_MDS_CONNECT_NET 0x117
871 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000117"
873 stat1=`count_ost_writes`
875 stat2=`count_ost_writes`
876 echo "$stat1, $stat2"
877 if [ $stat1 -lt $stat2 ]; then
878 echo "writes continuing during recovery"
881 echo "writes not continuing during recovery, bug 2477"
884 echo "waiting for writeme $WRITE_PID"
888 echo "waiting for multiop $PID"
889 wait $PID || return 2
890 do_facet client munlink $MOUNT/$tfile || return 3
891 do_facet client munlink $MOUNT/${tfile}-2 || return 3
894 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
898 # make sure that a read to one osc doesn't try to double-unlock its page just
899 # because another osc is invalid. trigger_group_io used to mistakenly return
900 # an error if any oscs were invalid even after having successfully put rpcs
901 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
902 # the page, guarnateeing that the unlock from the RPC completion would
903 # assert on trying to unlock the unlocked page.
905 [ $OSTCOUNT -lt 2 ] && \
906 skip "skipping test 41: we don't have a second OST to test with" && \
909 local f=$MOUNT/$tfile
910 # make sure the start of the file is ost1
911 lfs setstripe $f $((128 * 1024)) 0 0
912 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
914 # fail ost2 and read from ost1
915 local osc2dev=`do_facet mds "grep ${ost2_svc}-osc-MDT0000 $LPROC/devices" | awk '{print $1}'`
916 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && cat $LPROC/devices && return 4
917 do_facet mds $LCTL --device $osc2dev deactivate || return 1
918 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
919 do_facet mds $LCTL --device $osc2dev activate || return 2
922 run_test 41 "read from a valid osc while other oscs are invalid"
924 # test MDS recovery after ost failure
926 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
927 createmany -o $DIR/$tfile-%d 800
929 unlinkmany $DIR/$tfile-%d 0 400
931 sysctl -w lnet.debug=-1
934 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
935 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
936 #[ $blocks_after -lt $blocks ] || return 1
937 echo wait for MDS to timeout and recover
938 sleep $((TIMEOUT * 2))
940 unlinkmany $DIR/$tfile-%d 400 400
941 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
943 run_test 42 "recovery after ost failure"
945 # timeout in MDS/OST recovery RPC will LBUG MDS
946 test_43() { # bug 2530
947 replay_barrier $SINGLEMDS
949 # OBD_FAIL_OST_CREATE_NET 0x204
950 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
953 do_facet ost1 "sysctl -w lustre.fail_loc=0"
957 run_test 43 "mds osc import failure during recovery; don't LBUG"
960 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
961 [ "$mdcdev" ] || exit 2
962 for i in `seq 1 10`; do
963 #define OBD_FAIL_TGT_CONN_RACE 0x701
964 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000701"
965 $LCTL --device $mdcdev recover
968 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
971 run_test 44 "race in target handle connect"
974 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
975 [ "$mdcdev" ] || exit 2
976 for i in `seq 1 10`; do
977 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
978 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000704"
979 $LCTL --device $mdcdev recover
982 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
985 run_test 44b "race in target handle connect"
987 # Handle failed close
989 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
990 [ "$mdcdev" ] || exit 2
991 $LCTL --device $mdcdev recover
993 multiop $DIR/$tfile O_c &
997 # This will cause the CLOSE to fail before even
998 # allocating a reply buffer
999 $LCTL --device $mdcdev deactivate || return 4
1003 wait $pid || return 1
1005 $LCTL --device $mdcdev activate || return 5
1008 $CHECKSTAT -t file $DIR/$tfile || return 2
1011 run_test 45 "Handle failed close"
1015 drop_reply "touch $DIR/$tfile"
1017 # ironically, the previous test, 45, will cause a real forced close,
1018 # so just look for one for this test
1019 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1022 run_test 46 "Don't leak file handle after open resend (3325)"
1024 test_47() { # bug 2824
1025 # create some files to make sure precreate has been done on all
1026 # OSTs. (just in case this test is run independently)
1027 createmany -o $DIR/$tfile 20 || return 1
1029 # OBD_FAIL_OST_CREATE_NET 0x204
1031 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
1032 df $MOUNT || return 2
1034 # let the MDS discover the OST failure, attempt to recover, fail
1035 # and recover again.
1036 sleep $((3 * TIMEOUT))
1038 # Without 2824, this createmany would hang
1039 createmany -o $DIR/$tfile 20 || return 3
1040 unlinkmany $DIR/$tfile 20 || return 4
1042 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1045 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1048 replay_barrier $SINGLEMDS
1049 createmany -o $DIR/$tfile 20 || return 1
1050 # OBD_FAIL_OST_EROFS 0x216
1052 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000216"
1053 df $MOUNT || return 2
1055 createmany -o $DIR/$tfile 20 20 || return 2
1056 unlinkmany $DIR/$tfile 40 || return 3
1058 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1061 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1064 local oscdev=`do_facet $SINGLEMDS grep ${ost1_svc}-osc-MDT0000 $LPROC/devices | awk '{print $1}'`
1065 [ "$oscdev" ] || return 1
1066 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1067 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1068 # give the mds_lov_sync threads a chance to run
1071 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1073 # b3764 timed out lock replay
1076 cancel_lru_locks mdc
1078 multiop $DIR/$tfile s || return 1
1079 replay_barrier $SINGLEMDS
1080 #define OBD_FAIL_LDLM_REPLY 0x30c
1081 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000030c"
1082 fail $SINGLEMDS || return 2
1083 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1085 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1087 run_test 52 "time out lock replay (3764)"
1089 #b_cray 53 "|X| open request and close reply while two MDC requests in flight"
1090 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1092 #b3761 ASSERTION(hash != 0) failed
1094 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1095 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012b"
1097 # give touch a chance to run
1099 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1103 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1105 #b3440 ASSERTION(rec->ur_fid2->id) failed
1107 ln -s foo $DIR/$tfile
1108 replay_barrier $SINGLEMDS
1109 #drop_reply "cat $DIR/$tfile"
1113 run_test 56 "don't replay a symlink open request (3440)"
1115 #recovery one mds-ost setattr from llog
1117 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1118 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1120 replay_barrier $SINGLEMDS
1123 $CHECKSTAT -t file $DIR/$tfile || return 1
1124 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1127 run_test 57 "test recovery from llog for setattr op"
1129 #recovery many mds-ost setattr from llog
1131 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1132 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1134 createmany -o $DIR/$tdir/$tfile-%d 2500
1135 replay_barrier $SINGLEMDS
1138 $CHECKSTAT -t file $DIR/$tdir/$tfile-* || return 1
1139 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1140 unlinkmany $DIR/$tdir/$tfile-%d 2500
1143 run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
1145 # log_commit_thread vs filter_destroy race used to lead to import use after free
1149 createmany -o $DIR/$tdir/$tfile-%d 200
1151 unlinkmany $DIR/$tdir/$tfile-%d 200
1152 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1153 do_facet ost1 "sysctl -w lustre.fail_loc=0x507"
1156 do_facet ost1 "sysctl -w lustre.fail_loc=0x0"
1160 run_test 59 "test log_commit_thread vs filter_destroy race"
1162 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1163 # bug 12086: should no oops and No ctxt error for this test
1166 createmany -o $DIR/$tdir/$tfile-%d 200
1167 replay_barrier $SINGLEMDS
1168 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1170 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1171 local no_ctxt=`dmesg | grep "No ctxt"`
1172 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1174 run_test 60 "test llog post recovery init vs llog unlink"
1176 equals_msg `basename $0`: test complete, cleaning up
1177 check_and_cleanup_lustre
1178 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true