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 $REPLAY_SINGLE_EXCEPT"
24 # 63 min 7 min AT AT AT AT"
25 [ "SLOW" = "no" ] && EXCEPT="$EXCEPT 1 2 3 4 6 6b 12 16 44 44b 65 66 67 68"
29 cleanup_and_setup_lustre
33 rm -rf $DIR/${TESTSUITE}/[df][0-9]* # bug 13798 new t-f tdir staff
34 rm -rf $DIR/[df][0-9]*
39 replay_barrier $SINGLEMDS
43 run_test 0 "empty replay"
46 # this test attempts to trigger a race in the precreation code,
47 # and must run before any other objects are created on the filesystem
49 createmany -o $DIR/$tfile 20 || return 1
50 unlinkmany $DIR/$tfile 20 || return 2
52 run_test 0b "ensure object created after recover exists. (3284)"
58 local file=`ls /proc/fs/lustre/seq/cli-srv-$mds-mdc-*/width`
65 local file=`ls /proc/fs/lustre/seq/cli-srv-$mds-mdc-*/width`
69 # This test should pass for single-mds and multi-mds configs.
70 # But for different configurations it tests different things.
74 # (1) fld_create replay should happen;
76 # (2) fld_create replay should not return -EEXISTS, if it does
77 # this means sequence manager recovery code is buggy and allocated
78 # same sequence two times after recovery.
82 # (1) fld_create replay may not happen, because its home MDS is
83 # MDS2 which is not involved to revovery;
85 # (2) as fld_create does not happen on MDS1, it does not make any
88 local label=`mdsdevlabel 1`
89 [ -z "$label" ] && echo "No label for mds1" && return 1
91 replay_barrier $SINGLEMDS
92 local sw=`seq_get_width $label`
94 # make seq manager switch to next sequence each
95 # time as new fid is needed.
96 seq_set_width $label 1
98 # make sure that fld has created at least one new
100 touch $DIR/$tfile || return 2
101 seq_set_width $label $sw
103 # fail $SINGLEMDS and start recovery, replay RPCs, etc.
106 # wait for recovery finish
110 # flush fld cache and dentry cache to make it lookup
111 # created entry instead of revalidating existent one
113 zconf_mount `hostname` $MOUNT
115 # issue lookup which should call fld lookup which
116 # should fail if client did not replay fld create
117 # correctly and server has no fld entry
118 touch $DIR/$tfile || return 3
119 rm $DIR/$tfile || return 4
121 run_test 0c "fld create"
124 replay_barrier $SINGLEMDS
127 $CHECKSTAT -t file $DIR/$tfile || return 1
130 run_test 1 "simple create"
133 replay_barrier $SINGLEMDS
136 $CHECKSTAT -t file $DIR/$tfile || return 1
143 replay_barrier $SINGLEMDS
146 $CHECKSTAT -t file $DIR/$tfile || return 1
152 replay_barrier $SINGLEMDS
154 o_directory $DIR/$tfile
156 $CHECKSTAT -t file $DIR/$tfile || return 2
159 run_test 3a "replay failed open(O_DIRECTORY)"
162 replay_barrier $SINGLEMDS
163 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
164 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000114"
166 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
168 $CHECKSTAT -t file $DIR/$tfile && return 2
171 run_test 3b "replay failed open -ENOMEM"
174 replay_barrier $SINGLEMDS
175 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
176 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000128"
178 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
181 $CHECKSTAT -t file $DIR/$tfile && return 2
184 run_test 3c "replay failed open -ENOMEM"
187 replay_barrier $SINGLEMDS
188 for i in `seq 10`; do
189 echo "tag-$i" > $DIR/$tfile-$i
192 for i in `seq 10`; do
193 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
196 run_test 4 "|x| 10 open(O_CREAT)s"
199 replay_barrier $SINGLEMDS
202 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
204 run_test 4b "|x| rm 10 files"
206 # The idea is to get past the first block of precreated files on both
207 # osts, and then replay.
209 replay_barrier $SINGLEMDS
210 for i in `seq 220`; do
211 echo "tag-$i" > $DIR/$tfile-$i
214 for i in `seq 220`; do
215 grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
219 # waiting for commitment of removal
221 run_test 5 "|x| 220 open(O_CREAT)"
225 replay_barrier $SINGLEMDS
226 mcreate $DIR/$tdir/$tfile
228 $CHECKSTAT -t dir $DIR/$tdir || return 1
229 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
231 # waiting for log process thread
233 run_test 6 "mkdir + contained create"
236 replay_barrier $SINGLEMDS
239 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
241 run_test 6b "|X| rmdir"
244 replay_barrier $SINGLEMDS
245 mcreate $DIR/$tdir/$tfile
247 $CHECKSTAT -t dir $DIR/$tdir || return 1
248 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
251 run_test 7 "mkdir |X| contained create"
254 # make sure no side-effect from previous test.
256 replay_barrier $SINGLEMDS
257 multiop $DIR/$tfile mo_c &
262 $CHECKSTAT -t file $DIR/$tfile || return 1
263 kill -USR1 $MULTIPID || return 2
264 wait $MULTIPID || return 3
267 run_test 8 "creat open |X| close"
270 replay_barrier $SINGLEMDS
272 local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
274 local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
276 echo " old_inum == $old_inum, new_inum == $new_inum"
277 if [ $old_inum -eq $new_inum ] ;
279 echo " old_inum and new_inum match"
281 echo "!!!! old_inum and new_inum NOT match"
286 run_test 9 "|X| create (same inum/gen)"
290 replay_barrier $SINGLEMDS
291 mv $DIR/$tfile $DIR/$tfile-2
294 $CHECKSTAT $DIR/$tfile && return 1
295 $CHECKSTAT $DIR/$tfile-2 ||return 2
299 run_test 10 "create |X| rename unlink"
303 echo "old" > $DIR/$tfile
304 mv $DIR/$tfile $DIR/$tfile-2
305 replay_barrier $SINGLEMDS
306 echo "new" > $DIR/$tfile
308 grep old $DIR/$tfile-2
310 grep new $DIR/$tfile || return 1
311 grep old $DIR/$tfile-2 || return 2
313 run_test 11 "create open write rename |X| create-old-name read"
317 multiop $DIR/$tfile o_tSc &
319 # give multiop a chance to open
322 replay_barrier $SINGLEMDS
324 wait $pid || return 1
327 [ -e $DIR/$tfile ] && return 2
330 run_test 12 "open, unlink |X| close"
333 # 1777 - replay open after committed chmod that would make
334 # a regular open a failure
337 multiop $DIR/$tfile O_wc &
339 # give multiop a chance to open
342 $CHECKSTAT -p 0 $DIR/$tfile
343 replay_barrier $SINGLEMDS
346 wait $pid || return 1
348 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
351 run_test 13 "open chmod 0 |x| write close"
354 multiop $DIR/$tfile O_tSc &
356 # give multiop a chance to open
359 replay_barrier $SINGLEMDS
360 kill -USR1 $pid || return 1
361 wait $pid || return 2
364 [ -e $DIR/$tfile ] && return 3
367 run_test 14 "open(O_CREAT), unlink |X| close"
370 multiop $DIR/$tfile O_tSc &
372 # give multiop a chance to open
375 replay_barrier $SINGLEMDS
376 touch $DIR/g11 || return 1
378 wait $pid || return 2
381 [ -e $DIR/$tfile ] && return 3
382 touch $DIR/h11 || return 4
385 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
389 replay_barrier $SINGLEMDS
392 mcreate $DIR/$tfile-2
394 [ -e $DIR/$tfile ] && return 1
395 [ -e $DIR/$tfile-2 ] || return 2
396 munlink $DIR/$tfile-2 || return 3
398 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
401 replay_barrier $SINGLEMDS
402 multiop $DIR/$tfile O_c &
404 # give multiop a chance to open
407 kill -USR1 $pid || return 1
408 wait $pid || return 2
409 $CHECKSTAT -t file $DIR/$tfile || return 3
412 run_test 17 "|X| open(O_CREAT), |replay| close"
415 replay_barrier $SINGLEMDS
416 multiop $DIR/$tfile O_tSc &
418 # give multiop a chance to open
421 touch $DIR/$tfile-2 || return 1
422 echo "pid: $pid will close"
424 wait $pid || return 2
427 [ -e $DIR/$tfile ] && return 3
428 [ -e $DIR/$tfile-2 ] || return 4
429 # this touch frequently fails
430 touch $DIR/$tfile-3 || return 5
431 munlink $DIR/$tfile-2 || return 6
432 munlink $DIR/$tfile-3 || return 7
435 run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
437 # bug 1855 (a simpler form of test_11 above)
439 replay_barrier $SINGLEMDS
441 echo "old" > $DIR/$tfile
442 mv $DIR/$tfile $DIR/$tfile-2
443 grep old $DIR/$tfile-2
445 grep old $DIR/$tfile-2 || return 2
447 run_test 19 "|X| mcreate, open, write, rename "
450 replay_barrier $SINGLEMDS
451 multiop $DIR/$tfile O_tSc &
453 # give multiop a chance to open
459 wait $pid || return 1
460 [ -e $DIR/$tfile ] && return 2
463 run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
465 test_20b() { # bug 10480
466 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
468 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
470 while [ ! -e $DIR/$tfile ] ; do
471 usleep 60 # give dd a chance to start
474 lfs getstripe $DIR/$tfile || return 1
475 rm -f $DIR/$tfile || return 2 # make it an orphan
477 df -P $DIR || df -P $DIR || true # reconnect
479 fail $SINGLEMDS # start orphan recovery
480 df -P $DIR || df -P $DIR || true # reconnect
481 wait_mds_recovery_done || error "MDS recovery not done"
483 # FIXME just because recovery is done doesn't mean we've finished
484 # orphan cleanup. Fake it with a sleep for now...
486 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
487 log "before $BEFOREUSED, after $AFTERUSED"
488 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
489 error "after $AFTERUSED > before $BEFOREUSED"
492 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
494 test_20c() { # bug 10480
495 multiop $DIR/$tfile Ow_c &
497 # give multiop a chance to open
504 df -P $DIR || df -P $DIR || true # reconnect
507 test -s $DIR/$tfile || error "File was truncated"
511 run_test 20c "check that client eviction does not affect file content"
514 replay_barrier $SINGLEMDS
515 multiop $DIR/$tfile O_tSc &
517 # give multiop a chance to open
520 touch $DIR/g11 || return 1
524 wait $pid || return 2
525 [ -e $DIR/$tfile ] && return 3
526 touch $DIR/h11 || return 4
529 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
532 multiop $DIR/$tfile O_tSc &
534 # give multiop a chance to open
537 replay_barrier $SINGLEMDS
542 wait $pid || return 1
543 [ -e $DIR/$tfile ] && return 2
546 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
549 multiop $DIR/$tfile O_tSc &
551 # give multiop a chance to open
554 replay_barrier $SINGLEMDS
556 touch $DIR/g11 || return 1
560 wait $pid || return 2
561 [ -e $DIR/$tfile ] && return 3
562 touch $DIR/h11 || return 4
565 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
568 multiop $DIR/$tfile O_tSc &
570 # give multiop a chance to open
573 replay_barrier $SINGLEMDS
577 wait $pid || return 1
578 [ -e $DIR/$tfile ] && return 2
581 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
584 multiop $DIR/$tfile O_tSc &
586 # give multiop a chance to open
590 replay_barrier $SINGLEMDS
593 wait $pid || return 1
594 [ -e $DIR/$tfile ] && return 2
597 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
600 replay_barrier $SINGLEMDS
601 multiop $DIR/$tfile-1 O_tSc &
603 multiop $DIR/$tfile-2 O_tSc &
605 # give multiop a chance to open
610 wait $pid2 || return 1
614 wait $pid1 || return 2
615 [ -e $DIR/$tfile-1 ] && return 3
616 [ -e $DIR/$tfile-2 ] && return 4
619 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
622 replay_barrier $SINGLEMDS
623 multiop $DIR/$tfile-1 O_tSc &
625 multiop $DIR/$tfile-2 O_tSc &
627 # give multiop a chance to open
634 wait $pid1 || return 1
636 wait $pid2 || return 2
637 [ -e $DIR/$tfile-1 ] && return 3
638 [ -e $DIR/$tfile-2 ] && return 4
641 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
644 multiop $DIR/$tfile-1 O_tSc &
646 multiop $DIR/$tfile-2 O_tSc &
648 # give multiop a chance to open
650 replay_barrier $SINGLEMDS
654 wait $pid2 || return 1
658 wait $pid1 || return 2
659 [ -e $DIR/$tfile-1 ] && return 3
660 [ -e $DIR/$tfile-2 ] && return 4
663 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
666 multiop $DIR/$tfile-1 O_tSc &
668 multiop $DIR/$tfile-2 O_tSc &
670 # give multiop a chance to open
672 replay_barrier $SINGLEMDS
678 wait $pid1 || return 1
680 wait $pid2 || return 2
681 [ -e $DIR/$tfile-1 ] && return 3
682 [ -e $DIR/$tfile-2 ] && return 4
685 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
688 multiop $DIR/$tfile-1 O_tSc &
690 multiop $DIR/$tfile-2 O_tSc &
692 # give multiop a chance to open
697 replay_barrier $SINGLEMDS
700 wait $pid1 || return 1
702 wait $pid2 || return 2
703 [ -e $DIR/$tfile-1 ] && return 3
704 [ -e $DIR/$tfile-2 ] && return 4
707 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
710 multiop $DIR/$tfile-1 O_tSc &
712 multiop $DIR/$tfile-2 O_tSc &
714 # give multiop a chance to open
718 replay_barrier $SINGLEMDS
722 wait $pid1 || return 1
724 wait $pid2 || return 2
725 [ -e $DIR/$tfile-1 ] && return 3
726 [ -e $DIR/$tfile-2 ] && return 4
729 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
731 # tests for bug 2104; completion without crashing is success. The close is
732 # stale, but we always return 0 for close, so the app never sees it.
734 multiop $DIR/$tfile O_c &
736 multiop $DIR/$tfile O_c &
738 # give multiop a chance to open
741 df $MOUNT || sleep 1 && df $MOUNT || return 1
747 run_test 32 "close() notices client eviction; close() after client eviction"
749 # Abort recovery before client complete
751 replay_barrier $SINGLEMDS
752 createmany -o $DIR/$tfile-%d 100
753 fail_abort $SINGLEMDS
754 # this file should be gone, because the replay was aborted
755 $CHECKSTAT -t file $DIR/$tfile-* && return 3
756 unlinkmany $DIR/$tfile-%d 0 100
759 run_test 33 "abort recovery before client does replay"
763 replay_barrier $SINGLEMDS
764 createmany -o $DIR/$tfile-%d 10
765 fail_abort $SINGLEMDS
766 unlinkmany $DIR/$tfile-%d 0 10
767 # recreate shouldn't fail
768 createmany -o $DIR/$tfile-%d 10 || return 3
769 unlinkmany $DIR/$tfile-%d 0 10
772 run_test 33a "fid shouldn't be reused after abort recovery"
775 multiop $DIR/$tfile O_c &
777 # give multiop a chance to open
781 replay_barrier $SINGLEMDS
782 fail_abort $SINGLEMDS
784 [ -e $DIR/$tfile ] && return 1
788 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
790 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
794 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
795 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
800 # give a chance to remove from MDS
801 fail_abort $SINGLEMDS
802 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
804 run_test 35 "test recovery from llog for unlink op"
806 # b=2432 resent cancel after replay uses wrong cookie,
807 # so don't resend cancels
809 replay_barrier $SINGLEMDS
811 checkstat $DIR/$tfile
812 facet_failover $SINGLEMDS
814 if dmesg | grep "unknown lock cookie"; then
815 echo "cancel after replay failed"
819 run_test 36 "don't resend cancel"
822 # directory orphans can't be unlinked from PENDING directory
824 rmdir $DIR/$tfile 2>/dev/null
825 multiop $DIR/$tfile dD_c &
827 # give multiop a chance to open
831 replay_barrier $SINGLEMDS
832 # clear the dmesg buffer so we only see errors from this recovery
834 fail_abort $SINGLEMDS
836 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
840 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
843 createmany -o $DIR/$tfile-%d 800
844 unlinkmany $DIR/$tfile-%d 0 400
845 replay_barrier $SINGLEMDS
847 unlinkmany $DIR/$tfile-%d 400 400
849 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
851 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
853 test_39() { # bug 4176
854 createmany -o $DIR/$tfile-%d 800
855 replay_barrier $SINGLEMDS
856 unlinkmany $DIR/$tfile-%d 0 400
858 unlinkmany $DIR/$tfile-%d 400 400
860 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
862 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
865 awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' $LPROC/osc/*/stats
870 $LCTL mark multiop $MOUNT/$tfile OS_c
871 multiop $MOUNT/$tfile OS_c &
873 writeme -s $MOUNT/${tfile}-2 &
876 facet_failover $SINGLEMDS
877 #define OBD_FAIL_MDS_CONNECT_NET 0x117
878 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000117"
880 stat1=`count_ost_writes`
882 stat2=`count_ost_writes`
883 echo "$stat1, $stat2"
884 if [ $stat1 -lt $stat2 ]; then
885 echo "writes continuing during recovery"
888 echo "writes not continuing during recovery, bug 2477"
891 echo "waiting for writeme $WRITE_PID"
895 echo "waiting for multiop $PID"
896 wait $PID || return 2
897 do_facet client munlink $MOUNT/$tfile || return 3
898 do_facet client munlink $MOUNT/${tfile}-2 || return 3
901 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
905 # make sure that a read to one osc doesn't try to double-unlock its page just
906 # because another osc is invalid. trigger_group_io used to mistakenly return
907 # an error if any oscs were invalid even after having successfully put rpcs
908 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
909 # the page, guarnateeing that the unlock from the RPC completion would
910 # assert on trying to unlock the unlocked page.
912 [ $OSTCOUNT -lt 2 ] && \
913 skip "skipping test 41: we don't have a second OST to test with" && \
916 local f=$MOUNT/$tfile
917 # make sure the start of the file is ost1
918 lfs setstripe $f $((128 * 1024)) 0 0
919 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
921 # fail ost2 and read from ost1
922 local osc2dev=`do_facet mds "grep ${ost2_svc}-osc-MDT0000 $LPROC/devices" | awk '{print $1}'`
923 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && cat $LPROC/devices && return 4
924 do_facet mds $LCTL --device $osc2dev deactivate || return 1
925 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
926 do_facet mds $LCTL --device $osc2dev activate || return 2
929 run_test 41 "read from a valid osc while other oscs are invalid"
931 # test MDS recovery after ost failure
933 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
934 createmany -o $DIR/$tfile-%d 800
936 unlinkmany $DIR/$tfile-%d 0 400
938 sysctl -w lnet.debug=-1
941 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
942 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
943 #[ $blocks_after -lt $blocks ] || return 1
944 echo wait for MDS to timeout and recover
945 sleep $((TIMEOUT * 2))
947 unlinkmany $DIR/$tfile-%d 400 400
948 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
950 run_test 42 "recovery after ost failure"
952 # timeout in MDS/OST recovery RPC will LBUG MDS
953 test_43() { # bug 2530
954 replay_barrier $SINGLEMDS
956 # OBD_FAIL_OST_CREATE_NET 0x204
957 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
960 do_facet ost1 "sysctl -w lustre.fail_loc=0"
964 run_test 43 "mds osc import failure during recovery; don't LBUG"
967 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
968 [ "$mdcdev" ] || exit 2
969 for i in `seq 1 10`; do
970 #define OBD_FAIL_TGT_CONN_RACE 0x701
971 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000701"
972 $LCTL --device $mdcdev recover
975 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
978 run_test 44 "race in target handle connect"
981 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
982 [ "$mdcdev" ] || exit 2
983 for i in `seq 1 10`; do
984 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
985 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000704"
986 $LCTL --device $mdcdev recover
989 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
992 run_test 44b "race in target handle connect"
994 # Handle failed close
996 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
997 [ "$mdcdev" ] || exit 2
998 $LCTL --device $mdcdev recover
1000 multiop $DIR/$tfile O_c &
1004 # This will cause the CLOSE to fail before even
1005 # allocating a reply buffer
1006 $LCTL --device $mdcdev deactivate || return 4
1010 wait $pid || return 1
1012 $LCTL --device $mdcdev activate || return 5
1015 $CHECKSTAT -t file $DIR/$tfile || return 2
1018 run_test 45 "Handle failed close"
1022 drop_reply "touch $DIR/$tfile"
1024 # ironically, the previous test, 45, will cause a real forced close,
1025 # so just look for one for this test
1026 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1029 run_test 46 "Don't leak file handle after open resend (3325)"
1031 test_47() { # bug 2824
1032 # create some files to make sure precreate has been done on all
1033 # OSTs. (just in case this test is run independently)
1034 createmany -o $DIR/$tfile 20 || return 1
1036 # OBD_FAIL_OST_CREATE_NET 0x204
1038 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
1039 df $MOUNT || return 2
1041 # let the MDS discover the OST failure, attempt to recover, fail
1042 # and recover again.
1043 sleep $((3 * TIMEOUT))
1045 # Without 2824, this createmany would hang
1046 createmany -o $DIR/$tfile 20 || return 3
1047 unlinkmany $DIR/$tfile 20 || return 4
1049 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1052 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1055 replay_barrier $SINGLEMDS
1056 createmany -o $DIR/$tfile 20 || return 1
1057 # OBD_FAIL_OST_EROFS 0x216
1059 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000216"
1060 df $MOUNT || return 2
1062 createmany -o $DIR/$tfile 20 20 || return 2
1063 unlinkmany $DIR/$tfile 40 || return 3
1065 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1068 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1071 local oscdev=`do_facet $SINGLEMDS grep ${ost1_svc}-osc-MDT0000 $LPROC/devices | awk '{print $1}'`
1072 [ "$oscdev" ] || return 1
1073 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1074 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1075 # give the mds_lov_sync threads a chance to run
1078 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1080 # b3764 timed out lock replay
1083 cancel_lru_locks mdc
1085 multiop $DIR/$tfile s || return 1
1086 replay_barrier $SINGLEMDS
1087 #define OBD_FAIL_LDLM_REPLY 0x30c
1088 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000030c"
1089 fail $SINGLEMDS || return 2
1090 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1092 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1094 run_test 52 "time out lock replay (3764)"
1096 #b_cray 53 "|X| open request and close reply while two MDC requests in flight"
1097 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1099 #b3761 ASSERTION(hash != 0) failed
1101 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1102 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012b"
1104 # give touch a chance to run
1106 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1110 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1112 #b3440 ASSERTION(rec->ur_fid2->id) failed
1114 ln -s foo $DIR/$tfile
1115 replay_barrier $SINGLEMDS
1116 #drop_reply "cat $DIR/$tfile"
1120 run_test 56 "don't replay a symlink open request (3440)"
1122 #recovery one mds-ost setattr from llog
1124 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1125 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1127 replay_barrier $SINGLEMDS
1130 $CHECKSTAT -t file $DIR/$tfile || return 1
1131 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1134 run_test 57 "test recovery from llog for setattr op"
1136 #recovery many mds-ost setattr from llog
1138 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1139 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1140 createmany -o $DIR/$tdir/$tfile-%d 2500
1141 replay_barrier $SINGLEMDS
1144 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1145 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1146 unlinkmany $DIR/$tdir/$tfile-%d 2500
1149 run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
1151 # log_commit_thread vs filter_destroy race used to lead to import use after free
1154 createmany -o $DIR/$tdir/$tfile-%d 200
1156 unlinkmany $DIR/$tdir/$tfile-%d 200
1157 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1158 do_facet ost1 "sysctl -w lustre.fail_loc=0x507"
1161 do_facet ost1 "sysctl -w lustre.fail_loc=0x0"
1165 run_test 59 "test log_commit_thread vs filter_destroy race"
1167 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1168 # bug 12086: should no oops and No ctxt error for this test
1170 createmany -o $DIR/$tdir/$tfile-%d 200
1171 replay_barrier $SINGLEMDS
1172 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1174 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1175 local no_ctxt=`dmesg | grep "No ctxt"`
1176 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1178 run_test 60 "test llog post recovery init vs llog unlink"
1180 equals_msg `basename $0`: test complete, cleaning up
1181 check_and_cleanup_lustre
1182 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true