7 # This test needs to be run on the client
10 LUSTRE=${LUSTRE:-`dirname $0`/..}
13 . $LUSTRE/tests/test-framework.sh
15 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
16 CHECK_GRANT=${CHECK_GRANT:-"yes"}
17 GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
21 # bug number: 2766 4176 11404 13821
22 ALWAYS_EXCEPT="0b 39 56 59 $REPLAY_SINGLE_EXCEPT"
24 # 63 min 7 min AT AT AT AT"
25 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 6b 12 16 44 44b 65 66 67 68"
29 cleanup_and_setup_lustre
33 rm -rf $DIR/[df][0-9]*
38 replay_barrier $SINGLEMDS
42 run_test 0 "empty replay"
45 # this test attempts to trigger a race in the precreation code,
46 # and must run before any other objects are created on the filesystem
48 createmany -o $DIR/$tfile 20 || return 1
49 unlinkmany $DIR/$tfile 20 || return 2
51 run_test 0b "ensure object created after recover exists. (3284)"
57 local file=`ls /proc/fs/lustre/seq/cli-srv-$mds-mdc-*/width`
64 local file=`ls /proc/fs/lustre/seq/cli-srv-$mds-mdc-*/width`
68 # This test should pass for single-mds and multi-mds configs.
69 # But for different configurations it tests different things.
73 # (1) fld_create replay should happen;
75 # (2) fld_create replay should not return -EEXISTS, if it does
76 # this means sequence manager recovery code is buggy and allocated
77 # same sequence two times after recovery.
81 # (1) fld_create replay may not happen, because its home MDS is
82 # MDS2 which is not involved to revovery;
84 # (2) as fld_create does not happen on MDS1, it does not make any
87 local label=`mdsdevlabel 1`
88 [ -z "$label" ] && echo "No label for mds1" && return 1
90 replay_barrier $SINGLEMDS
91 local sw=`seq_get_width $label`
93 # make seq manager switch to next sequence each
94 # time as new fid is needed.
95 seq_set_width $label 1
97 # make sure that fld has created at least one new
99 touch $DIR/$tfile || return 2
100 seq_set_width $label $sw
102 # fail $SINGLEMDS and start recovery, replay RPCs, etc.
105 # wait for recovery finish
109 # flush fld cache and dentry cache to make it lookup
110 # created entry instead of revalidating existent one
112 zconf_mount `hostname` $MOUNT
114 # issue lookup which should call fld lookup which
115 # should fail if client did not replay fld create
116 # correctly and server has no fld entry
117 touch $DIR/$tfile || return 3
118 rm $DIR/$tfile || return 4
120 run_test 0c "fld create"
123 replay_barrier $SINGLEMDS
126 $CHECKSTAT -t file $DIR/$tfile || return 1
129 run_test 1 "simple create"
132 replay_barrier $SINGLEMDS
135 $CHECKSTAT -t file $DIR/$tfile || return 1
142 replay_barrier $SINGLEMDS
145 $CHECKSTAT -t file $DIR/$tfile || return 1
151 replay_barrier $SINGLEMDS
153 o_directory $DIR/$tfile
155 $CHECKSTAT -t file $DIR/$tfile || return 2
158 run_test 3a "replay failed open(O_DIRECTORY)"
161 replay_barrier $SINGLEMDS
162 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
163 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000114"
165 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
167 $CHECKSTAT -t file $DIR/$tfile && return 2
170 run_test 3b "replay failed open -ENOMEM"
173 replay_barrier $SINGLEMDS
174 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
175 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000128"
177 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
180 $CHECKSTAT -t file $DIR/$tfile && return 2
183 run_test 3c "replay failed open -ENOMEM"
186 replay_barrier $SINGLEMDS
187 for i in `seq 10`; do
188 echo "tag-$i" > $DIR/$tfile-$i
191 for i in `seq 10`; do
192 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
195 run_test 4 "|x| 10 open(O_CREAT)s"
198 replay_barrier $SINGLEMDS
201 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
203 run_test 4b "|x| rm 10 files"
205 # The idea is to get past the first block of precreated files on both
206 # osts, and then replay.
208 replay_barrier $SINGLEMDS
209 for i in `seq 220`; do
210 echo "tag-$i" > $DIR/$tfile-$i
213 for i in `seq 220`; do
214 grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
218 # waiting for commitment of removal
220 run_test 5 "|x| 220 open(O_CREAT)"
224 replay_barrier $SINGLEMDS
225 mcreate $DIR/$tdir/$tfile
227 $CHECKSTAT -t dir $DIR/$tdir || return 1
228 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
230 # waiting for log process thread
232 run_test 6 "mkdir + contained create"
235 replay_barrier $SINGLEMDS
238 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
240 run_test 6b "|X| rmdir"
243 replay_barrier $SINGLEMDS
244 mcreate $DIR/$tdir/$tfile
246 $CHECKSTAT -t dir $DIR/$tdir || return 1
247 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
250 run_test 7 "mkdir |X| contained create"
253 # make sure no side-effect from previous test.
255 replay_barrier $SINGLEMDS
256 multiop $DIR/$tfile mo_c &
261 $CHECKSTAT -t file $DIR/$tfile || return 1
262 kill -USR1 $MULTIPID || return 2
263 wait $MULTIPID || return 3
266 run_test 8 "creat open |X| close"
269 replay_barrier $SINGLEMDS
271 local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
273 local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
275 echo " old_inum == $old_inum, new_inum == $new_inum"
276 if [ $old_inum -eq $new_inum ] ;
278 echo " old_inum and new_inum match"
280 echo "!!!! old_inum and new_inum NOT match"
285 run_test 9 "|X| create (same inum/gen)"
289 replay_barrier $SINGLEMDS
290 mv $DIR/$tfile $DIR/$tfile-2
293 $CHECKSTAT $DIR/$tfile && return 1
294 $CHECKSTAT $DIR/$tfile-2 ||return 2
298 run_test 10 "create |X| rename unlink"
302 echo "old" > $DIR/$tfile
303 mv $DIR/$tfile $DIR/$tfile-2
304 replay_barrier $SINGLEMDS
305 echo "new" > $DIR/$tfile
307 grep old $DIR/$tfile-2
309 grep new $DIR/$tfile || return 1
310 grep old $DIR/$tfile-2 || return 2
312 run_test 11 "create open write rename |X| create-old-name read"
316 multiop $DIR/$tfile o_tSc &
318 # give multiop a chance to open
321 replay_barrier $SINGLEMDS
323 wait $pid || return 1
326 [ -e $DIR/$tfile ] && return 2
329 run_test 12 "open, unlink |X| close"
332 # 1777 - replay open after committed chmod that would make
333 # a regular open a failure
336 multiop $DIR/$tfile O_wc &
338 # give multiop a chance to open
341 $CHECKSTAT -p 0 $DIR/$tfile
342 replay_barrier $SINGLEMDS
345 wait $pid || return 1
347 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
350 run_test 13 "open chmod 0 |x| write close"
353 multiop $DIR/$tfile O_tSc &
355 # give multiop a chance to open
358 replay_barrier $SINGLEMDS
359 kill -USR1 $pid || return 1
360 wait $pid || return 2
363 [ -e $DIR/$tfile ] && return 3
366 run_test 14 "open(O_CREAT), unlink |X| close"
369 multiop $DIR/$tfile O_tSc &
371 # give multiop a chance to open
374 replay_barrier $SINGLEMDS
375 touch $DIR/g11 || return 1
377 wait $pid || return 2
380 [ -e $DIR/$tfile ] && return 3
381 touch $DIR/h11 || return 4
384 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
388 replay_barrier $SINGLEMDS
391 mcreate $DIR/$tfile-2
393 [ -e $DIR/$tfile ] && return 1
394 [ -e $DIR/$tfile-2 ] || return 2
395 munlink $DIR/$tfile-2 || return 3
397 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
400 replay_barrier $SINGLEMDS
401 multiop $DIR/$tfile O_c &
403 # give multiop a chance to open
406 kill -USR1 $pid || return 1
407 wait $pid || return 2
408 $CHECKSTAT -t file $DIR/$tfile || return 3
411 run_test 17 "|X| open(O_CREAT), |replay| close"
414 replay_barrier $SINGLEMDS
415 multiop $DIR/$tfile O_tSc &
417 # give multiop a chance to open
420 touch $DIR/$tfile-2 || return 1
421 echo "pid: $pid will close"
423 wait $pid || return 2
426 [ -e $DIR/$tfile ] && return 3
427 [ -e $DIR/$tfile-2 ] || return 4
428 # this touch frequently fails
429 touch $DIR/$tfile-3 || return 5
430 munlink $DIR/$tfile-2 || return 6
431 munlink $DIR/$tfile-3 || return 7
434 run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
436 # bug 1855 (a simpler form of test_11 above)
438 replay_barrier $SINGLEMDS
440 echo "old" > $DIR/$tfile
441 mv $DIR/$tfile $DIR/$tfile-2
442 grep old $DIR/$tfile-2
444 grep old $DIR/$tfile-2 || return 2
446 run_test 19 "|X| mcreate, open, write, rename "
449 replay_barrier $SINGLEMDS
450 multiop $DIR/$tfile O_tSc &
452 # give multiop a chance to open
458 wait $pid || return 1
459 [ -e $DIR/$tfile ] && return 2
462 run_test 20 "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
464 test_20b() { # bug 10480
465 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
467 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
469 while [ ! -e $DIR/$tfile ] ; do
470 usleep 60 # give dd a chance to start
473 lfs getstripe $DIR/$tfile || return 1
474 rm -f $DIR/$tfile || return 2 # make it an orphan
476 df -P $DIR || df -P $DIR || true # reconnect
478 fail $SINGLEMDS # start orphan recovery
479 df -P $DIR || df -P $DIR || true # reconnect
480 wait_mds_recovery_done || error "MDS recovery not done"
482 # FIXME just because recovery is done doesn't mean we've finished
483 # orphan cleanup. Fake it with a sleep for now...
485 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
486 log "before $BEFOREUSED, after $AFTERUSED"
487 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
488 error "after $AFTERUSED > before $BEFOREUSED"
491 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
493 test_20c() { # bug 10480
494 multiop $DIR/$tfile Ow_c &
496 # give multiop a chance to open
503 df -P $DIR || df -P $DIR || true # reconnect
506 test -s $DIR/$tfile || error "File was truncated"
510 run_test 20c "check that client eviction does not affect file content"
513 replay_barrier $SINGLEMDS
514 multiop $DIR/$tfile O_tSc &
516 # give multiop a chance to open
519 touch $DIR/g11 || return 1
523 wait $pid || return 2
524 [ -e $DIR/$tfile ] && return 3
525 touch $DIR/h11 || return 4
528 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
531 multiop $DIR/$tfile O_tSc &
533 # give multiop a chance to open
536 replay_barrier $SINGLEMDS
541 wait $pid || return 1
542 [ -e $DIR/$tfile ] && return 2
545 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
548 multiop $DIR/$tfile O_tSc &
550 # give multiop a chance to open
553 replay_barrier $SINGLEMDS
555 touch $DIR/g11 || return 1
559 wait $pid || return 2
560 [ -e $DIR/$tfile ] && return 3
561 touch $DIR/h11 || return 4
564 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
567 multiop $DIR/$tfile O_tSc &
569 # give multiop a chance to open
572 replay_barrier $SINGLEMDS
576 wait $pid || return 1
577 [ -e $DIR/$tfile ] && return 2
580 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
583 multiop $DIR/$tfile O_tSc &
585 # give multiop a chance to open
589 replay_barrier $SINGLEMDS
592 wait $pid || return 1
593 [ -e $DIR/$tfile ] && return 2
596 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
599 replay_barrier $SINGLEMDS
600 multiop $DIR/$tfile-1 O_tSc &
602 multiop $DIR/$tfile-2 O_tSc &
604 # give multiop a chance to open
609 wait $pid2 || return 1
613 wait $pid1 || return 2
614 [ -e $DIR/$tfile-1 ] && return 3
615 [ -e $DIR/$tfile-2 ] && return 4
618 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
621 replay_barrier $SINGLEMDS
622 multiop $DIR/$tfile-1 O_tSc &
624 multiop $DIR/$tfile-2 O_tSc &
626 # give multiop a chance to open
633 wait $pid1 || return 1
635 wait $pid2 || return 2
636 [ -e $DIR/$tfile-1 ] && return 3
637 [ -e $DIR/$tfile-2 ] && return 4
640 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
643 multiop $DIR/$tfile-1 O_tSc &
645 multiop $DIR/$tfile-2 O_tSc &
647 # give multiop a chance to open
649 replay_barrier $SINGLEMDS
653 wait $pid2 || return 1
657 wait $pid1 || return 2
658 [ -e $DIR/$tfile-1 ] && return 3
659 [ -e $DIR/$tfile-2 ] && return 4
662 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
665 multiop $DIR/$tfile-1 O_tSc &
667 multiop $DIR/$tfile-2 O_tSc &
669 # give multiop a chance to open
671 replay_barrier $SINGLEMDS
677 wait $pid1 || return 1
679 wait $pid2 || return 2
680 [ -e $DIR/$tfile-1 ] && return 3
681 [ -e $DIR/$tfile-2 ] && return 4
684 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
687 multiop $DIR/$tfile-1 O_tSc &
689 multiop $DIR/$tfile-2 O_tSc &
691 # give multiop a chance to open
696 replay_barrier $SINGLEMDS
699 wait $pid1 || return 1
701 wait $pid2 || return 2
702 [ -e $DIR/$tfile-1 ] && return 3
703 [ -e $DIR/$tfile-2 ] && return 4
706 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
709 multiop $DIR/$tfile-1 O_tSc &
711 multiop $DIR/$tfile-2 O_tSc &
713 # give multiop a chance to open
717 replay_barrier $SINGLEMDS
721 wait $pid1 || return 1
723 wait $pid2 || return 2
724 [ -e $DIR/$tfile-1 ] && return 3
725 [ -e $DIR/$tfile-2 ] && return 4
728 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
730 # tests for bug 2104; completion without crashing is success. The close is
731 # stale, but we always return 0 for close, so the app never sees it.
733 multiop $DIR/$tfile O_c &
735 multiop $DIR/$tfile O_c &
737 # give multiop a chance to open
740 df $MOUNT || sleep 1 && df $MOUNT || return 1
746 run_test 32 "close() notices client eviction; close() after client eviction"
748 # Abort recovery before client complete
750 replay_barrier $SINGLEMDS
751 createmany -o $DIR/$tfile-%d 100
752 fail_abort $SINGLEMDS
753 # this file should be gone, because the replay was aborted
754 $CHECKSTAT -t file $DIR/$tfile-* && return 3
755 unlinkmany $DIR/$tfile-%d 0 100
758 run_test 33 "abort recovery before client does replay"
762 replay_barrier $SINGLEMDS
763 createmany -o $DIR/$tfile-%d 10
764 fail_abort $SINGLEMDS
765 unlinkmany $DIR/$tfile-%d 0 10
766 # recreate shouldn't fail
767 createmany -o $DIR/$tfile-%d 10 || return 3
768 unlinkmany $DIR/$tfile-%d 0 10
771 run_test 33a "fid shouldn't be reused after abort recovery"
774 multiop $DIR/$tfile O_c &
776 # give multiop a chance to open
780 replay_barrier $SINGLEMDS
781 fail_abort $SINGLEMDS
783 [ -e $DIR/$tfile ] && return 1
787 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
789 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
793 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
794 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000119"
799 # give a chance to remove from MDS
800 fail_abort $SINGLEMDS
801 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
803 run_test 35 "test recovery from llog for unlink op"
805 # b=2432 resent cancel after replay uses wrong cookie,
806 # so don't resend cancels
808 replay_barrier $SINGLEMDS
810 checkstat $DIR/$tfile
811 facet_failover $SINGLEMDS
813 if dmesg | grep "unknown lock cookie"; then
814 echo "cancel after replay failed"
818 run_test 36 "don't resend cancel"
821 # directory orphans can't be unlinked from PENDING directory
823 rmdir $DIR/$tfile 2>/dev/null
824 multiop $DIR/$tfile dD_c &
826 # give multiop a chance to open
830 replay_barrier $SINGLEMDS
831 # clear the dmesg buffer so we only see errors from this recovery
833 fail_abort $SINGLEMDS
835 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
839 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
842 createmany -o $DIR/$tfile-%d 800
843 unlinkmany $DIR/$tfile-%d 0 400
844 replay_barrier $SINGLEMDS
846 unlinkmany $DIR/$tfile-%d 400 400
848 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
850 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
852 test_39() { # bug 4176
853 createmany -o $DIR/$tfile-%d 800
854 replay_barrier $SINGLEMDS
855 unlinkmany $DIR/$tfile-%d 0 400
857 unlinkmany $DIR/$tfile-%d 400 400
859 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
861 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
864 awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }' $LPROC/osc/*/stats
869 $LCTL mark multiop $MOUNT/$tfile OS_c
870 multiop $MOUNT/$tfile OS_c &
872 writeme -s $MOUNT/${tfile}-2 &
875 facet_failover $SINGLEMDS
876 #define OBD_FAIL_MDS_CONNECT_NET 0x117
877 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000117"
879 stat1=`count_ost_writes`
881 stat2=`count_ost_writes`
882 echo "$stat1, $stat2"
883 if [ $stat1 -lt $stat2 ]; then
884 echo "writes continuing during recovery"
887 echo "writes not continuing during recovery, bug 2477"
890 echo "waiting for writeme $WRITE_PID"
894 echo "waiting for multiop $PID"
895 wait $PID || return 2
896 do_facet client munlink $MOUNT/$tfile || return 3
897 do_facet client munlink $MOUNT/${tfile}-2 || return 3
900 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
904 # make sure that a read to one osc doesn't try to double-unlock its page just
905 # because another osc is invalid. trigger_group_io used to mistakenly return
906 # an error if any oscs were invalid even after having successfully put rpcs
907 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
908 # the page, guarnateeing that the unlock from the RPC completion would
909 # assert on trying to unlock the unlocked page.
911 [ $OSTCOUNT -lt 2 ] && \
912 skip "skipping test 41: we don't have a second OST to test with" && \
915 local f=$MOUNT/$tfile
916 # make sure the start of the file is ost1
917 lfs setstripe $f -s $((128 * 1024)) -i 0
918 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
920 # fail ost2 and read from ost1
921 local osc2dev=`do_facet mds "grep ${ost2_svc}-osc-MDT0000 $LPROC/devices" | awk '{print $1}'`
922 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && cat $LPROC/devices && return 4
923 do_facet mds $LCTL --device $osc2dev deactivate || return 1
924 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
925 do_facet mds $LCTL --device $osc2dev activate || return 2
928 run_test 41 "read from a valid osc while other oscs are invalid"
930 # test MDS recovery after ost failure
932 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
933 createmany -o $DIR/$tfile-%d 800
935 unlinkmany $DIR/$tfile-%d 0 400
937 sysctl -w lnet.debug=-1
940 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
941 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
942 #[ $blocks_after -lt $blocks ] || return 1
943 echo wait for MDS to timeout and recover
944 sleep $((TIMEOUT * 2))
946 unlinkmany $DIR/$tfile-%d 400 400
947 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
949 run_test 42 "recovery after ost failure"
951 # timeout in MDS/OST recovery RPC will LBUG MDS
952 test_43() { # bug 2530
953 replay_barrier $SINGLEMDS
955 # OBD_FAIL_OST_CREATE_NET 0x204
956 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
959 do_facet ost1 "sysctl -w lustre.fail_loc=0"
963 run_test 43 "mds osc import failure during recovery; don't LBUG"
966 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
967 [ "$mdcdev" ] || exit 2
968 for i in `seq 1 10`; do
969 #define OBD_FAIL_TGT_CONN_RACE 0x701
970 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000701"
971 $LCTL --device $mdcdev recover
974 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
977 run_test 44 "race in target handle connect"
980 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
981 [ "$mdcdev" ] || exit 2
982 for i in `seq 1 10`; do
983 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
984 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000704"
985 $LCTL --device $mdcdev recover
988 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
991 run_test 44b "race in target handle connect"
993 # Handle failed close
995 mdcdev=`awk '/MDT0000-mdc-/ {print $1}' $LPROC/devices`
996 [ "$mdcdev" ] || exit 2
997 $LCTL --device $mdcdev recover
999 multiop $DIR/$tfile O_c &
1003 # This will cause the CLOSE to fail before even
1004 # allocating a reply buffer
1005 $LCTL --device $mdcdev deactivate || return 4
1009 wait $pid || return 1
1011 $LCTL --device $mdcdev activate || return 5
1014 $CHECKSTAT -t file $DIR/$tfile || return 2
1017 run_test 45 "Handle failed close"
1021 drop_reply "touch $DIR/$tfile"
1023 # ironically, the previous test, 45, will cause a real forced close,
1024 # so just look for one for this test
1025 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1028 run_test 46 "Don't leak file handle after open resend (3325)"
1030 test_47() { # bug 2824
1031 # create some files to make sure precreate has been done on all
1032 # OSTs. (just in case this test is run independently)
1033 createmany -o $DIR/$tfile 20 || return 1
1035 # OBD_FAIL_OST_CREATE_NET 0x204
1037 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000204"
1038 df $MOUNT || return 2
1040 # let the MDS discover the OST failure, attempt to recover, fail
1041 # and recover again.
1042 sleep $((3 * TIMEOUT))
1044 # Without 2824, this createmany would hang
1045 createmany -o $DIR/$tfile 20 || return 3
1046 unlinkmany $DIR/$tfile 20 || return 4
1048 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1051 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1054 replay_barrier $SINGLEMDS
1055 createmany -o $DIR/$tfile 20 || return 1
1056 # OBD_FAIL_OST_EROFS 0x216
1058 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000216"
1059 df $MOUNT || return 2
1061 createmany -o $DIR/$tfile 20 20 || return 2
1062 unlinkmany $DIR/$tfile 40 || return 3
1064 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1067 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1070 local oscdev=`do_facet $SINGLEMDS grep ${ost1_svc}-osc-MDT0000 $LPROC/devices | awk '{print $1}'`
1071 [ "$oscdev" ] || return 1
1072 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1073 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1074 # give the mds_lov_sync threads a chance to run
1077 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1079 # b3764 timed out lock replay
1082 cancel_lru_locks mdc
1084 multiop $DIR/$tfile s || return 1
1085 replay_barrier $SINGLEMDS
1086 #define OBD_FAIL_LDLM_REPLY 0x30c
1087 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000030c"
1088 fail $SINGLEMDS || return 2
1089 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1091 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1093 run_test 52 "time out lock replay (3764)"
1095 # bug 3462 - simultaneous MDC requests
1097 mkdir -p $DIR/${tdir}-1
1098 mkdir -p $DIR/${tdir}-2
1099 multiop $DIR/${tdir}-1/f O_c &
1101 # give multiop a change to open
1104 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1105 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
1106 kill -USR1 $close_pid
1107 cancel_lru_locks mdc # force the close
1108 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1110 mcreate $DIR/${tdir}-2/f || return 1
1112 # close should still be here
1113 [ -d /proc/$close_pid ] || return 2
1115 replay_barrier_nodf $SINGLEMDS
1117 wait $close_pid || return 3
1119 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1120 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1121 rm -rf $DIR/${tdir}-*
1123 run_test 53a "|X| close request while two MDC requests in flight"
1126 mkdir -p $DIR/${tdir}-1
1127 mkdir -p $DIR/${tdir}-2
1128 multiop $DIR/${tdir}-1/f O_c &
1131 #define OBD_FAIL_MDS_REINT_NET 0x107
1132 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
1133 mcreate $DIR/${tdir}-2/f &
1137 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1138 kill -USR1 $close_pid
1139 cancel_lru_locks mdc # force the close
1140 wait $close_pid || return 1
1141 # open should still be here
1142 [ -d /proc/$open_pid ] || return 2
1144 replay_barrier_nodf $SINGLEMDS
1146 wait $open_pid || return 3
1148 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1149 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1150 rm -rf $DIR/${tdir}-*
1152 run_test 53b "|X| open request while two MDC requests in flight"
1155 mkdir -p $DIR/${tdir}-1
1156 mkdir -p $DIR/${tdir}-2
1157 multiop $DIR/${tdir}-1/f O_c &
1160 #define OBD_FAIL_MDS_REINT_NET 0x107
1161 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
1162 mcreate $DIR/${tdir}-2/f &
1166 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1167 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
1168 kill -USR1 $close_pid
1169 cancel_lru_locks mdc # force the close
1171 replay_barrier_nodf $SINGLEMDS
1172 fail_nodf $SINGLEMDS
1173 wait $open_pid || return 1
1175 # close should be gone
1176 [ -d /proc/$close_pid ] && return 2
1177 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1179 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1180 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1181 rm -rf $DIR/${tdir}-*
1183 run_test 53c "|X| open request and close request while two MDC requests in flight"
1186 mkdir -p $DIR/${tdir}-1
1187 mkdir -p $DIR/${tdir}-2
1188 multiop $DIR/${tdir}-1/f O_c &
1190 # give multiop a chance to open
1193 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1194 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000013f"
1195 kill -USR1 $close_pid
1196 cancel_lru_locks mdc # force the close
1197 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1198 mcreate $DIR/${tdir}-2/f || return 1
1200 # close should still be here
1201 [ -d /proc/$close_pid ] || return 2
1203 wait $close_pid || return 3
1205 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1206 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1207 rm -rf $DIR/${tdir}-*
1209 run_test 53d "|X| close reply while two MDC requests in flight"
1212 mkdir -p $DIR/${tdir}-1
1213 mkdir -p $DIR/${tdir}-2
1214 multiop $DIR/${tdir}-1/f O_c &
1217 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1218 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x119"
1219 mcreate $DIR/${tdir}-2/f &
1223 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1224 kill -USR1 $close_pid
1225 cancel_lru_locks mdc # force the close
1226 wait $close_pid || return 1
1227 # open should still be here
1228 [ -d /proc/$open_pid ] || return 2
1230 replay_barrier_nodf $SINGLEMDS
1232 wait $open_pid || return 3
1234 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1235 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1236 rm -rf $DIR/${tdir}-*
1238 run_test 53e "|X| open reply while two MDC requests in flight"
1241 mkdir -p $DIR/${tdir}-1
1242 mkdir -p $DIR/${tdir}-2
1243 multiop $DIR/${tdir}-1/f O_c &
1246 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1247 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x119"
1248 mcreate $DIR/${tdir}-2/f &
1252 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1253 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000013f"
1254 kill -USR1 $close_pid
1255 cancel_lru_locks mdc # force the close
1257 replay_barrier_nodf $SINGLEMDS
1258 fail_nodf $SINGLEMDS
1259 wait $open_pid || return 1
1261 # close should be gone
1262 [ -d /proc/$close_pid ] && return 2
1263 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1265 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1266 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1267 rm -rf $DIR/${tdir}-*
1269 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1272 mkdir -p $DIR/${tdir}-1
1273 mkdir -p $DIR/${tdir}-2
1274 multiop $DIR/${tdir}-1/f O_c &
1277 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1278 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x119"
1279 mcreate $DIR/${tdir}-2/f &
1283 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1284 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000115"
1285 kill -USR1 $close_pid
1286 cancel_lru_locks mdc # force the close
1288 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1289 replay_barrier_nodf $SINGLEMDS
1290 fail_nodf $SINGLEMDS
1291 wait $open_pid || return 1
1293 # close should be gone
1294 [ -d /proc/$close_pid ] && return 2
1296 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1297 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1298 rm -rf $DIR/${tdir}-*
1300 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1303 mkdir -p $DIR/${tdir}-1
1304 mkdir -p $DIR/${tdir}-2
1305 multiop $DIR/${tdir}-1/f O_c &
1308 #define OBD_FAIL_MDS_REINT_NET 0x107
1309 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000107"
1310 mcreate $DIR/${tdir}-2/f &
1314 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
1315 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000013f"
1316 kill -USR1 $close_pid
1317 cancel_lru_locks mdc # force the close
1320 replay_barrier_nodf $SINGLEMDS
1321 fail_nodf $SINGLEMDS
1322 wait $open_pid || return 1
1324 # close should be gone
1325 [ -d /proc/$close_pid ] && return 2
1326 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0"
1328 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1329 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1330 rm -rf $DIR/${tdir}-*
1332 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1334 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1336 #b3761 ASSERTION(hash != 0) failed
1338 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1339 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012b"
1341 # give touch a chance to run
1343 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1347 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1349 #b3440 ASSERTION(rec->ur_fid2->id) failed
1351 ln -s foo $DIR/$tfile
1352 replay_barrier $SINGLEMDS
1353 #drop_reply "cat $DIR/$tfile"
1357 run_test 56 "don't replay a symlink open request (3440)"
1359 #recovery one mds-ost setattr from llog
1361 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1362 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1364 replay_barrier $SINGLEMDS
1367 $CHECKSTAT -t file $DIR/$tfile || return 1
1368 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1371 run_test 57 "test recovery from llog for setattr op"
1373 #recovery many mds-ost setattr from llog
1375 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1376 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x8000012c"
1377 createmany -o $DIR/$tdir/$tfile-%d 2500
1378 replay_barrier $SINGLEMDS
1381 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1382 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x0"
1383 unlinkmany $DIR/$tdir/$tfile-%d 2500
1386 run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
1388 # log_commit_thread vs filter_destroy race used to lead to import use after free
1391 createmany -o $DIR/$tdir/$tfile-%d 200
1393 unlinkmany $DIR/$tdir/$tfile-%d 200
1394 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1395 do_facet ost1 "sysctl -w lustre.fail_loc=0x507"
1398 do_facet ost1 "sysctl -w lustre.fail_loc=0x0"
1402 run_test 59 "test log_commit_thread vs filter_destroy race"
1404 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1405 # bug 12086: should no oops and No ctxt error for this test
1407 createmany -o $DIR/$tdir/$tfile-%d 200
1408 replay_barrier $SINGLEMDS
1409 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1411 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1412 local no_ctxt=`dmesg | grep "No ctxt"`
1413 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1415 run_test 60 "test llog post recovery init vs llog unlink"
1417 #test race llog recovery thread vs llog cleanup
1420 createmany -o $DIR/$tdir/$tfile-%d 800
1422 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1423 unlinkmany $DIR/$tdir/$tfile-%d 800
1424 do_facet ost "sysctl -w lustre.fail_loc=0x80000221"
1429 do_facet ost "sysctl -w lustre.fail_loc=0x0"
1430 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1433 run_test 61 "test race llog recovery vs llog cleanup"
1435 #test race mds llog sync vs llog cleanup
1437 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x140
1438 do_facet $SINGLEMDS "sysctl -w lustre.fail_loc=0x80000140"
1439 facet_failover $SINGLEMDS
1442 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1444 run_test 61b "test race mds llog sync vs llog cleanup"
1446 #test race cancel cookie cb vs llog cleanup
1448 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1450 do_facet ost "sysctl -w lustre.fail_loc=0x80000222"
1455 run_test 61c "test race mds llog sync vs llog cleanup"
1457 equals_msg `basename $0`: test complete, cleaning up
1458 check_and_cleanup_lustre
1459 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true