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
227 mcreate $DIR/$tdir/$tfile
229 $CHECKSTAT -t dir $DIR/$tdir || return 1
230 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
232 # waiting for log process thread
234 run_test 6 "mkdir + contained create"
237 replay_barrier $SINGLEMDS
240 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
242 run_test 6b "|X| rmdir"
246 replay_barrier $SINGLEMDS
247 mcreate $DIR/$tdir/$tfile
249 $CHECKSTAT -t dir $DIR/$tdir || return 1
250 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
253 run_test 7 "mkdir |X| contained create"
256 # make sure no side-effect from previous test.
258 replay_barrier $SINGLEMDS
259 multiop $DIR/$tfile mo_c &
264 $CHECKSTAT -t file $DIR/$tfile || return 1
265 kill -USR1 $MULTIPID || return 2
266 wait $MULTIPID || return 3
269 run_test 8 "creat open |X| close"
272 replay_barrier $SINGLEMDS
274 local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
276 local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
278 echo " old_inum == $old_inum, new_inum == $new_inum"
279 if [ $old_inum -eq $new_inum ] ;
281 echo " old_inum and new_inum match"
283 echo "!!!! old_inum and new_inum NOT match"
288 run_test 9 "|X| create (same inum/gen)"
292 replay_barrier $SINGLEMDS
293 mv $DIR/$tfile $DIR/$tfile-2
296 $CHECKSTAT $DIR/$tfile && return 1
297 $CHECKSTAT $DIR/$tfile-2 ||return 2
301 run_test 10 "create |X| rename unlink"
305 echo "old" > $DIR/$tfile
306 mv $DIR/$tfile $DIR/$tfile-2
307 replay_barrier $SINGLEMDS
308 echo "new" > $DIR/$tfile
310 grep old $DIR/$tfile-2
312 grep new $DIR/$tfile || return 1
313 grep old $DIR/$tfile-2 || return 2
315 run_test 11 "create open write rename |X| create-old-name read"
319 multiop $DIR/$tfile o_tSc &
321 # give multiop a chance to open
324 replay_barrier $SINGLEMDS
326 wait $pid || return 1
329 [ -e $DIR/$tfile ] && return 2
332 run_test 12 "open, unlink |X| close"
335 # 1777 - replay open after committed chmod that would make
336 # a regular open a failure
339 multiop $DIR/$tfile O_wc &
341 # give multiop a chance to open
344 $CHECKSTAT -p 0 $DIR/$tfile
345 replay_barrier $SINGLEMDS
348 wait $pid || return 1
350 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
353 run_test 13 "open chmod 0 |x| write close"
356 multiop $DIR/$tfile O_tSc &
358 # give multiop a chance to open
361 replay_barrier $SINGLEMDS
362 kill -USR1 $pid || return 1
363 wait $pid || return 2
366 [ -e $DIR/$tfile ] && return 3
369 run_test 14 "open(O_CREAT), unlink |X| close"
372 multiop $DIR/$tfile O_tSc &
374 # give multiop a chance to open
377 replay_barrier $SINGLEMDS
378 touch $DIR/g11 || return 1
380 wait $pid || return 2
383 [ -e $DIR/$tfile ] && return 3
384 touch $DIR/h11 || return 4
387 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
391 replay_barrier $SINGLEMDS
394 mcreate $DIR/$tfile-2
396 [ -e $DIR/$tfile ] && return 1
397 [ -e $DIR/$tfile-2 ] || return 2
398 munlink $DIR/$tfile-2 || return 3
400 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
403 replay_barrier $SINGLEMDS
404 multiop $DIR/$tfile O_c &
406 # give multiop a chance to open
409 kill -USR1 $pid || return 1
410 wait $pid || return 2
411 $CHECKSTAT -t file $DIR/$tfile || return 3
414 run_test 17 "|X| open(O_CREAT), |replay| close"
417 replay_barrier $SINGLEMDS
418 multiop $DIR/$tfile O_tSc &
420 # give multiop a chance to open
423 touch $DIR/$tfile-2 || return 1
424 echo "pid: $pid will close"
426 wait $pid || return 2
429 [ -e $DIR/$tfile ] && return 3
430 [ -e $DIR/$tfile-2 ] || return 4
431 # this touch frequently fails
432 touch $DIR/$tfile-3 || return 5
433 munlink $DIR/$tfile-2 || return 6
434 munlink $DIR/$tfile-3 || return 7
437 run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
439 # bug 1855 (a simpler form of test_11 above)
441 replay_barrier $SINGLEMDS
443 echo "old" > $DIR/$tfile
444 mv $DIR/$tfile $DIR/$tfile-2
445 grep old $DIR/$tfile-2
447 grep old $DIR/$tfile-2 || return 2
449 run_test 19 "|X| mcreate, open, write, rename "
452 replay_barrier $SINGLEMDS
453 multiop $DIR/$tfile O_tSc &
455 # give multiop a chance to open
461 wait $pid || return 1
462 [ -e $DIR/$tfile ] && return 2
465 run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
467 test_20b() { # bug 10480
468 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
470 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
472 while [ ! -e $DIR/$tfile ] ; do
473 usleep 60 # give dd a chance to start
476 lfs getstripe $DIR/$tfile || return 1
477 rm -f $DIR/$tfile || return 2 # make it an orphan
479 df -P $DIR || df -P $DIR || true # reconnect
481 fail $SINGLEMDS # start orphan recovery
482 df -P $DIR || df -P $DIR || true # reconnect
483 wait_mds_recovery_done || error "MDS recovery not done"
485 # FIXME just because recovery is done doesn't mean we've finished
486 # orphan cleanup. Fake it with a sleep for now...
488 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
489 log "before $BEFOREUSED, after $AFTERUSED"
490 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
491 error "after $AFTERUSED > before $BEFOREUSED"
494 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
496 test_20c() { # bug 10480
497 multiop $DIR/$tfile Ow_c &
499 # give multiop a chance to open
506 df -P $DIR || df -P $DIR || true # reconnect
509 test -s $DIR/$tfile || error "File was truncated"
513 run_test 20c "check that client eviction does not affect file content"
516 replay_barrier $SINGLEMDS
517 multiop $DIR/$tfile O_tSc &
519 # give multiop a chance to open
522 touch $DIR/g11 || return 1
526 wait $pid || return 2
527 [ -e $DIR/$tfile ] && return 3
528 touch $DIR/h11 || return 4
531 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
534 multiop $DIR/$tfile O_tSc &
536 # give multiop a chance to open
539 replay_barrier $SINGLEMDS
544 wait $pid || return 1
545 [ -e $DIR/$tfile ] && return 2
548 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
551 multiop $DIR/$tfile O_tSc &
553 # give multiop a chance to open
556 replay_barrier $SINGLEMDS
558 touch $DIR/g11 || return 1
562 wait $pid || return 2
563 [ -e $DIR/$tfile ] && return 3
564 touch $DIR/h11 || return 4
567 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
570 multiop $DIR/$tfile O_tSc &
572 # give multiop a chance to open
575 replay_barrier $SINGLEMDS
579 wait $pid || return 1
580 [ -e $DIR/$tfile ] && return 2
583 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
586 multiop $DIR/$tfile O_tSc &
588 # give multiop a chance to open
592 replay_barrier $SINGLEMDS
595 wait $pid || return 1
596 [ -e $DIR/$tfile ] && return 2
599 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
602 replay_barrier $SINGLEMDS
603 multiop $DIR/$tfile-1 O_tSc &
605 multiop $DIR/$tfile-2 O_tSc &
607 # give multiop a chance to open
612 wait $pid2 || return 1
616 wait $pid1 || return 2
617 [ -e $DIR/$tfile-1 ] && return 3
618 [ -e $DIR/$tfile-2 ] && return 4
621 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
624 replay_barrier $SINGLEMDS
625 multiop $DIR/$tfile-1 O_tSc &
627 multiop $DIR/$tfile-2 O_tSc &
629 # give multiop a chance to open
636 wait $pid1 || return 1
638 wait $pid2 || return 2
639 [ -e $DIR/$tfile-1 ] && return 3
640 [ -e $DIR/$tfile-2 ] && return 4
643 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
646 multiop $DIR/$tfile-1 O_tSc &
648 multiop $DIR/$tfile-2 O_tSc &
650 # give multiop a chance to open
652 replay_barrier $SINGLEMDS
656 wait $pid2 || return 1
660 wait $pid1 || return 2
661 [ -e $DIR/$tfile-1 ] && return 3
662 [ -e $DIR/$tfile-2 ] && return 4
665 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
668 multiop $DIR/$tfile-1 O_tSc &
670 multiop $DIR/$tfile-2 O_tSc &
672 # give multiop a chance to open
674 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 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
690 multiop $DIR/$tfile-1 O_tSc &
692 multiop $DIR/$tfile-2 O_tSc &
694 # give multiop a chance to open
699 replay_barrier $SINGLEMDS
702 wait $pid1 || return 1
704 wait $pid2 || return 2
705 [ -e $DIR/$tfile-1 ] && return 3
706 [ -e $DIR/$tfile-2 ] && return 4
709 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
712 multiop $DIR/$tfile-1 O_tSc &
714 multiop $DIR/$tfile-2 O_tSc &
716 # give multiop a chance to open
720 replay_barrier $SINGLEMDS
724 wait $pid1 || return 1
726 wait $pid2 || return 2
727 [ -e $DIR/$tfile-1 ] && return 3
728 [ -e $DIR/$tfile-2 ] && return 4
731 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
733 # tests for bug 2104; completion without crashing is success. The close is
734 # stale, but we always return 0 for close, so the app never sees it.
736 multiop $DIR/$tfile O_c &
738 multiop $DIR/$tfile O_c &
740 # give multiop a chance to open
743 df $MOUNT || sleep 1 && df $MOUNT || return 1
749 run_test 32 "close() notices client eviction; close() after client eviction"
751 # Abort recovery before client complete
753 replay_barrier $SINGLEMDS
754 createmany -o $DIR/$tfile-%d 100
755 fail_abort $SINGLEMDS
756 # this file should be gone, because the replay was aborted
757 $CHECKSTAT -t file $DIR/$tfile-* && return 3
758 unlinkmany $DIR/$tfile-%d 0 100
761 run_test 33 "abort recovery before client does replay"
765 replay_barrier $SINGLEMDS
766 createmany -o $DIR/$tfile-%d 10
767 fail_abort $SINGLEMDS
768 unlinkmany $DIR/$tfile-%d 0 10
769 # recreate shouldn't fail
770 createmany -o $DIR/$tfile-%d 10 || return 3
771 unlinkmany $DIR/$tfile-%d 0 10
774 run_test 33a "fid shouldn't be reused after abort recovery"
777 multiop $DIR/$tfile O_c &
779 # give multiop a chance to open
783 replay_barrier $SINGLEMDS
784 fail_abort $SINGLEMDS
786 [ -e $DIR/$tfile ] && return 1
790 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
792 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
796 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
797 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
802 # give a chance to remove from MDS
803 fail_abort $SINGLEMDS
804 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
806 run_test 35 "test recovery from llog for unlink op"
808 # b=2432 resent cancel after replay uses wrong cookie,
809 # so don't resend cancels
811 replay_barrier $SINGLEMDS
813 checkstat $DIR/$tfile
814 facet_failover $SINGLEMDS
816 if dmesg | grep "unknown lock cookie"; then
817 echo "cancel after replay failed"
821 run_test 36 "don't resend cancel"
824 # directory orphans can't be unlinked from PENDING directory
826 rmdir $DIR/$tfile 2>/dev/null
827 multiop $DIR/$tfile dD_c &
829 # give multiop a chance to open
833 replay_barrier $SINGLEMDS
834 # clear the dmesg buffer so we only see errors from this recovery
836 fail_abort $SINGLEMDS
838 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
842 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
845 createmany -o $DIR/$tfile-%d 800
846 unlinkmany $DIR/$tfile-%d 0 400
847 replay_barrier $SINGLEMDS
849 unlinkmany $DIR/$tfile-%d 400 400
851 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
853 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
855 test_39() { # bug 4176
856 createmany -o $DIR/$tfile-%d 800
857 replay_barrier $SINGLEMDS
858 unlinkmany $DIR/$tfile-%d 0 400
860 unlinkmany $DIR/$tfile-%d 400 400
862 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
864 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
867 awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' $LPROC/osc/*/stats
872 $LCTL mark multiop $MOUNT/$tfile OS_c
873 multiop $MOUNT/$tfile OS_c &
875 writeme -s $MOUNT/${tfile}-2 &
878 facet_failover $SINGLEMDS
879 #define OBD_FAIL_MDS_CONNECT_NET 0x117
880 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000117"
882 stat1=`count_ost_writes`
884 stat2=`count_ost_writes`
885 echo "$stat1, $stat2"
886 if [ $stat1 -lt $stat2 ]; then
887 echo "writes continuing during recovery"
890 echo "writes not continuing during recovery, bug 2477"
893 echo "waiting for writeme $WRITE_PID"
897 echo "waiting for multiop $PID"
898 wait $PID || return 2
899 do_facet client munlink $MOUNT/$tfile || return 3
900 do_facet client munlink $MOUNT/${tfile}-2 || return 3
903 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
907 # make sure that a read to one osc doesn't try to double-unlock its page just
908 # because another osc is invalid. trigger_group_io used to mistakenly return
909 # an error if any oscs were invalid even after having successfully put rpcs
910 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
911 # the page, guarnateeing that the unlock from the RPC completion would
912 # assert on trying to unlock the unlocked page.
914 [ $OSTCOUNT -lt 2 ] && \
915 skip "skipping test 41: we don't have a second OST to test with" && \
918 local f=$MOUNT/$tfile
919 # make sure the start of the file is ost1
920 lfs setstripe $f $((128 * 1024)) 0 0
921 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
923 # fail ost2 and read from ost1
924 local osc2dev=`do_facet mds "grep ${ost2_svc}-osc-MDT0000 $LPROC/devices" | awk '{print $1}'`
925 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && cat $LPROC/devices && return 4
926 do_facet mds $LCTL --device $osc2dev deactivate || return 1
927 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
928 do_facet mds $LCTL --device $osc2dev activate || return 2
931 run_test 41 "read from a valid osc while other oscs are invalid"
933 # test MDS recovery after ost failure
935 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
936 createmany -o $DIR/$tfile-%d 800
938 unlinkmany $DIR/$tfile-%d 0 400
940 sysctl -w lnet.debug=-1
943 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
944 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
945 #[ $blocks_after -lt $blocks ] || return 1
946 echo wait for MDS to timeout and recover
947 sleep $((TIMEOUT * 2))
949 unlinkmany $DIR/$tfile-%d 400 400
950 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
952 run_test 42 "recovery after ost failure"
954 # timeout in MDS/OST recovery RPC will LBUG MDS
955 test_43() { # bug 2530
956 replay_barrier $SINGLEMDS
958 # OBD_FAIL_OST_CREATE_NET 0x204
959 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
962 do_facet ost1 "sysctl -w lustre.fail_loc=0"
966 run_test 43 "mds osc import failure during recovery; don't LBUG"
969 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
970 [ "$mdcdev" ] || exit 2
971 for i in `seq 1 10`; do
972 #define OBD_FAIL_TGT_CONN_RACE 0x701
973 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000701"
974 $LCTL --device $mdcdev recover
977 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
980 run_test 44 "race in target handle connect"
983 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
984 [ "$mdcdev" ] || exit 2
985 for i in `seq 1 10`; do
986 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
987 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000704"
988 $LCTL --device $mdcdev recover
991 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
994 run_test 44b "race in target handle connect"
996 # Handle failed close
998 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
999 [ "$mdcdev" ] || exit 2
1000 $LCTL --device $mdcdev recover
1002 multiop $DIR/$tfile O_c &
1006 # This will cause the CLOSE to fail before even
1007 # allocating a reply buffer
1008 $LCTL --device $mdcdev deactivate || return 4
1012 wait $pid || return 1
1014 $LCTL --device $mdcdev activate || return 5
1017 $CHECKSTAT -t file $DIR/$tfile || return 2
1020 run_test 45 "Handle failed close"
1024 drop_reply "touch $DIR/$tfile"
1026 # ironically, the previous test, 45, will cause a real forced close,
1027 # so just look for one for this test
1028 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1031 run_test 46 "Don't leak file handle after open resend (3325)"
1033 test_47() { # bug 2824
1034 # create some files to make sure precreate has been done on all
1035 # OSTs. (just in case this test is run independently)
1036 createmany -o $DIR/$tfile 20 || return 1
1038 # OBD_FAIL_OST_CREATE_NET 0x204
1040 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
1041 df $MOUNT || return 2
1043 # let the MDS discover the OST failure, attempt to recover, fail
1044 # and recover again.
1045 sleep $((3 * TIMEOUT))
1047 # Without 2824, this createmany would hang
1048 createmany -o $DIR/$tfile 20 || return 3
1049 unlinkmany $DIR/$tfile 20 || return 4
1051 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1054 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1057 replay_barrier $SINGLEMDS
1058 createmany -o $DIR/$tfile 20 || return 1
1059 # OBD_FAIL_OST_EROFS 0x216
1061 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000216"
1062 df $MOUNT || return 2
1064 createmany -o $DIR/$tfile 20 20 || return 2
1065 unlinkmany $DIR/$tfile 40 || return 3
1067 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1070 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1073 local oscdev=`do_facet $SINGLEMDS grep ${ost1_svc}-osc-MDT0000 $LPROC/devices | awk '{print $1}'`
1074 [ "$oscdev" ] || return 1
1075 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1076 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1077 # give the mds_lov_sync threads a chance to run
1080 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1082 # b3764 timed out lock replay
1085 cancel_lru_locks mdc
1087 multiop $DIR/$tfile s || return 1
1088 replay_barrier $SINGLEMDS
1089 #define OBD_FAIL_LDLM_REPLY 0x30c
1090 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000030c"
1091 fail $SINGLEMDS || return 2
1092 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1094 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1096 run_test 52 "time out lock replay (3764)"
1098 #b_cray 53 "|X| open request and close reply while two MDC requests in flight"
1099 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1101 #b3761 ASSERTION(hash != 0) failed
1103 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1104 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012b"
1106 # give touch a chance to run
1108 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1112 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1114 #b3440 ASSERTION(rec->ur_fid2->id) failed
1116 ln -s foo $DIR/$tfile
1117 replay_barrier $SINGLEMDS
1118 #drop_reply "cat $DIR/$tfile"
1122 run_test 56 "don't replay a symlink open request (3440)"
1124 #recovery one mds-ost setattr from llog
1126 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1127 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1129 replay_barrier $SINGLEMDS
1132 $CHECKSTAT -t file $DIR/$tfile || return 1
1133 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1136 run_test 57 "test recovery from llog for setattr op"
1138 #recovery many mds-ost setattr from llog
1140 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1141 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1143 createmany -o $DIR/$tdir/$tfile-%d 2500
1144 replay_barrier $SINGLEMDS
1147 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1148 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1149 unlinkmany $DIR/$tdir/$tfile-%d 2500
1152 run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
1154 # log_commit_thread vs filter_destroy race used to lead to import use after free
1158 createmany -o $DIR/$tdir/$tfile-%d 200
1160 unlinkmany $DIR/$tdir/$tfile-%d 200
1161 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1162 do_facet ost1 "sysctl -w lustre.fail_loc=0x507"
1165 do_facet ost1 "sysctl -w lustre.fail_loc=0x0"
1169 run_test 59 "test log_commit_thread vs filter_destroy race"
1171 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1172 # bug 12086: should no oops and No ctxt error for this test
1175 createmany -o $DIR/$tdir/$tfile-%d 200
1176 replay_barrier $SINGLEMDS
1177 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1179 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1180 local no_ctxt=`dmesg | grep "No ctxt"`
1181 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1183 run_test 60 "test llog post recovery init vs llog unlink"
1185 equals_msg `basename $0`: test complete, cleaning up
1186 check_and_cleanup_lustre
1187 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true