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=${GRANT_CHECK_LIST:-""}
21 ALWAYS_EXCEPT="$REPLAY_SINGLE_EXCEPT"
23 # 63 min 7 min AT AT AT AT"
24 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a 44b 65 66 67 68"
28 cleanup_and_setup_lustre
32 rm -rf $DIR/[df][0-9]*
34 test_0a() { # was test_0
38 run_test 0a "empty replay"
41 # this test attempts to trigger a race in the precreation code,
42 # and must run before any other objects are created on the filesystem
44 createmany -o $DIR/$tfile 20 || return 1
45 unlinkmany $DIR/$tfile 20 || return 2
47 run_test 0b "ensure object created after recover exists. (3284)"
53 $CHECKSTAT -t file $DIR/$tfile || return 1
56 run_test 1 "simple create"
62 $CHECKSTAT -t file $DIR/$tfile || return 1
72 $CHECKSTAT -t file $DIR/$tfile || return 1
80 o_directory $DIR/$tfile
82 $CHECKSTAT -t file $DIR/$tfile || return 2
85 run_test 3a "replay failed open(O_DIRECTORY)"
89 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
90 do_facet mds "lctl set_param fail_loc=0x80000114"
92 do_facet mds "lctl set_param fail_loc=0"
94 $CHECKSTAT -t file $DIR/$tfile && return 2
97 run_test 3b "replay failed open -ENOMEM"
101 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
102 do_facet mds "lctl set_param fail_loc=0x80000128"
104 do_facet mds "lctl set_param fail_loc=0"
107 $CHECKSTAT -t file $DIR/$tfile && return 2
110 run_test 3c "replay failed open -ENOMEM"
112 test_4a() { # was test_4
114 for i in `seq 10`; do
115 echo "tag-$i" > $DIR/$tfile-$i
118 for i in `seq 10`; do
119 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
122 run_test 4a "|x| 10 open(O_CREAT)s"
128 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
130 run_test 4b "|x| rm 10 files"
132 # The idea is to get past the first block of precreated files on both
133 # osts, and then replay.
136 for i in `seq 220`; do
137 echo "tag-$i" > $DIR/$tfile-$i
140 for i in `seq 220`; do
141 grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
145 # waiting for commitment of removal
147 run_test 5 "|x| 220 open(O_CREAT)"
150 test_6a() { # was test_6a
152 mcreate $DIR/$tdir/$tfile
154 $CHECKSTAT -t dir $DIR/$tdir || return 1
155 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
157 # waiting for log process thread
159 run_test 6a "mkdir + contained create"
165 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
167 run_test 6b "|X| rmdir"
171 mcreate $DIR/$tdir/$tfile
173 $CHECKSTAT -t dir $DIR/$tdir || return 1
174 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
177 run_test 7 "mkdir |X| contained create"
181 multiop_bg_pause $DIR/$tfile mo_c || return 4
185 $CHECKSTAT -t file $DIR/$tfile || return 1
186 kill -USR1 $MULTIPID || return 2
187 wait $MULTIPID || return 3
190 run_test 8 "creat open |X| close"
195 local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
197 local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
199 echo " old_inum == $old_inum, new_inum == $new_inum"
200 if [ $old_inum -eq $new_inum ] ;
202 echo " old_inum and new_inum match"
204 echo "!!!! old_inum and new_inum NOT match"
209 run_test 9 "|X| create (same inum/gen)"
214 mv $DIR/$tfile $DIR/$tfile-2
217 $CHECKSTAT $DIR/$tfile && return 1
218 $CHECKSTAT $DIR/$tfile-2 ||return 2
222 run_test 10 "create |X| rename unlink"
226 echo "old" > $DIR/$tfile
227 mv $DIR/$tfile $DIR/$tfile-2
229 echo "new" > $DIR/$tfile
231 grep old $DIR/$tfile-2
233 grep new $DIR/$tfile || return 1
234 grep old $DIR/$tfile-2 || return 2
236 run_test 11 "create open write rename |X| create-old-name read"
240 multiop_bg_pause $DIR/$tfile o_tSc || return 3
245 wait $pid || return 1
248 [ -e $DIR/$tfile ] && return 2
251 run_test 12 "open, unlink |X| close"
254 # 1777 - replay open after committed chmod that would make
255 # a regular open a failure
258 multiop_bg_pause $DIR/$tfile O_wc || return 3
261 $CHECKSTAT -p 0 $DIR/$tfile
265 wait $pid || return 1
267 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
270 run_test 13 "open chmod 0 |x| write close"
273 multiop_bg_pause $DIR/$tfile O_tSc || return 4
277 kill -USR1 $pid || return 1
278 wait $pid || return 2
281 [ -e $DIR/$tfile ] && return 3
284 run_test 14 "open(O_CREAT), unlink |X| close"
287 multiop_bg_pause $DIR/$tfile O_tSc || return 5
291 touch $DIR/g11 || return 1
293 wait $pid || return 2
296 [ -e $DIR/$tfile ] && return 3
297 touch $DIR/h11 || return 4
300 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
307 mcreate $DIR/$tfile-2
309 [ -e $DIR/$tfile ] && return 1
310 [ -e $DIR/$tfile-2 ] || return 2
311 munlink $DIR/$tfile-2 || return 3
313 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
317 multiop_bg_pause $DIR/$tfile O_c || return 4
320 kill -USR1 $pid || return 1
321 wait $pid || return 2
322 $CHECKSTAT -t file $DIR/$tfile || return 3
325 run_test 17 "|X| open(O_CREAT), |replay| close"
329 multiop_bg_pause $DIR/$tfile O_tSc || return 8
332 touch $DIR/$tfile-2 || return 1
333 echo "pid: $pid will close"
335 wait $pid || return 2
338 [ -e $DIR/$tfile ] && return 3
339 [ -e $DIR/$tfile-2 ] || return 4
340 # this touch frequently fails
341 touch $DIR/$tfile-3 || return 5
342 munlink $DIR/$tfile-2 || return 6
343 munlink $DIR/$tfile-3 || return 7
346 run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
348 # bug 1855 (a simpler form of test_11 above)
352 echo "old" > $DIR/$tfile
353 mv $DIR/$tfile $DIR/$tfile-2
354 grep old $DIR/$tfile-2
356 grep old $DIR/$tfile-2 || return 2
358 run_test 19 "|X| mcreate, open, write, rename "
360 test_20a() { # was test_20
362 multiop_bg_pause $DIR/$tfile O_tSc || return 3
368 wait $pid || return 1
369 [ -e $DIR/$tfile ] && return 2
372 run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
374 test_20b() { # bug 10480
375 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
377 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
379 while [ ! -e $DIR/$tfile ] ; do
380 sleep 0.060s # give dd a chance to start
383 lfs getstripe $DIR/$tfile || return 1
384 rm -f $DIR/$tfile || return 2 # make it an orphan
386 df -P $DIR || df -P $DIR || true # reconnect
388 fail mds # start orphan recovery
389 df -P $DIR || df -P $DIR || true # reconnect
390 wait_mds_recovery_done || error "MDS recovery not done"
392 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
393 log "before $BEFOREUSED, after $AFTERUSED"
394 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
395 error "after $AFTERUSED > before $BEFOREUSED" && return 5
398 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
400 test_20c() { # bug 10480
401 multiop $DIR/$tfile Ow_c &
403 # give multiop a chance to open
410 df -P $DIR || df -P $DIR || true # reconnect
413 test -s $DIR/$tfile || error "File was truncated"
417 run_test 20c "check that client eviction does not affect file content"
421 multiop_bg_pause $DIR/$tfile O_tSc || return 5
424 touch $DIR/g11 || return 1
428 wait $pid || return 2
429 [ -e $DIR/$tfile ] && return 3
430 touch $DIR/h11 || return 4
433 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
436 multiop_bg_pause $DIR/$tfile O_tSc || return 3
444 wait $pid || return 1
445 [ -e $DIR/$tfile ] && return 2
448 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
451 multiop_bg_pause $DIR/$tfile O_tSc || return 5
456 touch $DIR/g11 || return 1
460 wait $pid || return 2
461 [ -e $DIR/$tfile ] && return 3
462 touch $DIR/h11 || return 4
465 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
468 multiop_bg_pause $DIR/$tfile O_tSc || return 3
475 wait $pid || return 1
476 [ -e $DIR/$tfile ] && return 2
479 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
482 multiop_bg_pause $DIR/$tfile O_tSc || return 3
489 wait $pid || return 1
490 [ -e $DIR/$tfile ] && return 2
493 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
497 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
499 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
504 wait $pid2 || return 1
508 wait $pid1 || return 2
509 [ -e $DIR/$tfile-1 ] && return 3
510 [ -e $DIR/$tfile-2 ] && return 4
513 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
517 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
519 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
526 wait $pid1 || return 1
528 wait $pid2 || return 2
529 [ -e $DIR/$tfile-1 ] && return 3
530 [ -e $DIR/$tfile-2 ] && return 4
533 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
536 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
538 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
544 wait $pid2 || return 1
548 wait $pid1 || return 2
549 [ -e $DIR/$tfile-1 ] && return 3
550 [ -e $DIR/$tfile-2 ] && return 4
553 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
556 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
558 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
566 wait $pid1 || return 1
568 wait $pid2 || return 2
569 [ -e $DIR/$tfile-1 ] && return 3
570 [ -e $DIR/$tfile-2 ] && return 4
573 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
576 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
578 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
586 wait $pid1 || return 1
588 wait $pid2 || return 2
589 [ -e $DIR/$tfile-1 ] && return 3
590 [ -e $DIR/$tfile-2 ] && return 4
593 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
596 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
598 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
606 wait $pid1 || return 1
608 wait $pid2 || return 2
609 [ -e $DIR/$tfile-1 ] && return 3
610 [ -e $DIR/$tfile-2 ] && return 4
613 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
615 # tests for bug 2104; completion without crashing is success. The close is
616 # stale, but we always return 0 for close, so the app never sees it.
618 multiop_bg_pause $DIR/$tfile O_c || return 2
620 multiop_bg_pause $DIR/$tfile O_c || return 3
623 df $MOUNT || sleep 1 && df $MOUNT || return 1
629 run_test 32 "close() notices client eviction; close() after client eviction"
631 # Abort recovery before client complete
632 test_33a() { # was test_33
634 createmany -o $DIR/$tfile-%d 100
636 # this file should be gone, because the replay was aborted
637 $CHECKSTAT -t file $DIR/$tfile-* && return 3
638 unlinkmany $DIR/$tfile-%d 0 100
641 run_test 33a "abort recovery before client does replay"
644 multiop_bg_pause $DIR/$tfile O_c || return 2
651 [ -e $DIR/$tfile ] && return 1
655 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
657 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
661 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
662 do_facet mds "lctl set_param fail_loc=0x80000119"
667 # give a chance to remove from MDS
669 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
671 run_test 35 "test recovery from llog for unlink op"
673 # b=2432 resent cancel after replay uses wrong cookie,
674 # so don't resend cancels
678 checkstat $DIR/$tfile
681 if dmesg | grep "unknown lock cookie"; then
682 echo "cancel after replay failed"
686 run_test 36 "don't resend cancel"
689 # directory orphans can't be unlinked from PENDING directory
691 rmdir $DIR/$tfile 2>/dev/null
692 multiop_bg_pause $DIR/$tfile dD_c || return 2
697 # clear the dmesg buffer so we only see errors from this recovery
701 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
705 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
708 createmany -o $DIR/$tfile-%d 800
709 unlinkmany $DIR/$tfile-%d 0 400
712 unlinkmany $DIR/$tfile-%d 400 400
714 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
716 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
718 test_39() { # bug 4176
719 createmany -o $DIR/$tfile-%d 800
721 unlinkmany $DIR/$tfile-%d 0 400
723 unlinkmany $DIR/$tfile-%d 400 400
726 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
728 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
731 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
736 $LCTL mark multiop $MOUNT/$tfile OS_c
737 multiop $MOUNT/$tfile OS_c &
739 writeme -s $MOUNT/${tfile}-2 &
743 #define OBD_FAIL_MDS_CONNECT_NET 0x117
744 do_facet mds "lctl set_param fail_loc=0x80000117"
746 stat1=`count_ost_writes`
748 stat2=`count_ost_writes`
749 echo "$stat1, $stat2"
750 if [ $stat1 -lt $stat2 ]; then
751 echo "writes continuing during recovery"
754 echo "writes not continuing during recovery, bug 2477"
757 echo "waiting for writeme $WRITE_PID"
761 echo "waiting for multiop $PID"
762 wait $PID || return 2
763 do_facet client munlink $MOUNT/$tfile || return 3
764 do_facet client munlink $MOUNT/${tfile}-2 || return 3
767 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
771 # make sure that a read to one osc doesn't try to double-unlock its page just
772 # because another osc is invalid. trigger_group_io used to mistakenly return
773 # an error if any oscs were invalid even after having successfully put rpcs
774 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
775 # the page, guarnateeing that the unlock from the RPC completion would
776 # assert on trying to unlock the unlocked page.
778 [ $OSTCOUNT -lt 2 ] && \
779 skip "skipping test 41: we don't have a second OST to test with" && \
782 local f=$MOUNT/$tfile
783 # make sure the start of the file is ost1
784 lfs setstripe $f -s $((128 * 1024)) -i 0
785 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
787 # fail ost2 and read from ost1
788 local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | awk '{print $1}'`
789 [ "$osc2dev" ] || return 4
790 $LCTL --device $osc2dev deactivate || return 1
791 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
792 $LCTL --device $osc2dev activate || return 2
795 run_test 41 "read from a valid osc while other oscs are invalid"
797 # test MDS recovery after ost failure
799 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
800 createmany -o $DIR/$tfile-%d 800
802 unlinkmany $DIR/$tfile-%d 0 400
804 lctl set_param debug=-1
807 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
808 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
809 #[ $blocks_after -lt $blocks ] || return 1
810 echo wait for MDS to timeout and recover
811 sleep $((TIMEOUT * 2))
813 unlinkmany $DIR/$tfile-%d 400 400
814 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
816 run_test 42 "recovery after ost failure"
818 # timeout in MDS/OST recovery RPC will LBUG MDS
819 test_43() { # bug 2530
822 # OBD_FAIL_OST_CREATE_NET 0x204
823 do_facet ost1 "lctl set_param fail_loc=0x80000204"
826 do_facet ost1 "lctl set_param fail_loc=0"
830 run_test 43 "mds osc import failure during recovery; don't LBUG"
832 test_44a() { # was test_44
835 mdcdev=`lctl get_param -n devices | awk '/-mdc-/ {print $1}'`
836 [ "$mdcdev" ] || exit 2
838 # adaptive timeouts slow this way down
839 if at_is_valid && at_is_enabled; then
840 at_max_saved=$(at_max_get mds)
844 for i in `seq 1 10`; do
845 echo "$i of 10 ($(date +%s))"
846 do_facet mds "lctl get_param -n mdt.MDS.mds.timeouts | grep service"
847 #define OBD_FAIL_TGT_CONN_RACE 0x701
848 do_facet mds "lctl set_param fail_loc=0x80000701"
849 $LCTL --device $mdcdev recover
853 do_facet mds "lctl set_param fail_loc=0"
854 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
857 run_test 44a "race in target handle connect"
860 mdcdev=`lctl get_param -n devices | awk '/-mdc-/ {print $1}'`
861 [ "$mdcdev" ] || exit 2
862 for i in `seq 1 10`; do
863 echo "$i of 10 ($(date +%s))"
864 do_facet mds "lctl get_param -n mdt.MDS.mds.timeouts | grep service"
865 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
866 do_facet mds "lctl set_param fail_loc=0x80000704"
867 $LCTL --device $mdcdev recover
870 do_facet mds "lctl set_param fail_loc=0"
873 run_test 44b "race in target handle connect"
875 # Handle failed close
877 mdcdev=`lctl get_param -n devices | awk '/-mdc-/ {print $1}'`
878 [ "$mdcdev" ] || exit 2
879 $LCTL --device $mdcdev recover
881 multiop_bg_pause $DIR/$tfile O_c || return 1
884 # This will cause the CLOSE to fail before even
885 # allocating a reply buffer
886 $LCTL --device $mdcdev deactivate || return 4
890 wait $pid || return 1
892 $LCTL --device $mdcdev activate || return 5
895 $CHECKSTAT -t file $DIR/$tfile || return 2
898 run_test 45 "Handle failed close"
902 drop_reply "touch $DIR/$tfile"
904 # ironically, the previous test, 45, will cause a real forced close,
905 # so just look for one for this test
906 dmesg | grep -i "force closing client file handle for $tfile" && return 1
909 run_test 46 "Don't leak file handle after open resend (3325)"
911 test_47() { # bug 2824
912 # create some files to make sure precreate has been done on all
913 # OSTs. (just in case this test is run independently)
914 createmany -o $DIR/$tfile 20 || return 1
916 # OBD_FAIL_OST_CREATE_NET 0x204
918 do_facet ost1 "lctl set_param fail_loc=0x80000204"
919 df $MOUNT || return 2
921 # let the MDS discover the OST failure, attempt to recover, fail
923 sleep $((3 * TIMEOUT))
925 # Without 2824, this createmany would hang
926 createmany -o $DIR/$tfile 20 || return 3
927 unlinkmany $DIR/$tfile 20 || return 4
929 do_facet ost1 "lctl set_param fail_loc=0"
932 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
936 createmany -o $DIR/$tfile 20 || return 1
937 # OBD_FAIL_OST_EROFS 0x216
939 do_facet ost1 "lctl set_param fail_loc=0x80000216"
940 df $MOUNT || return 2
942 createmany -o $DIR/$tfile 20 20 || return 2
943 unlinkmany $DIR/$tfile 40 || return 3
945 do_facet ost1 "lctl set_param fail_loc=0"
948 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
951 local oscdev=`do_facet mds lctl get_param -n devices | grep ${ost1_svc}-osc | awk '{print $1}' | head -1`
952 [ "$oscdev" ] || return 1
953 do_facet mds $LCTL --device $oscdev recover || return 2
954 do_facet mds $LCTL --device $oscdev recover || return 3
955 # give the mds_lov_sync threads a chance to run
958 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
960 # b3764 timed out lock replay
965 multiop $DIR/$tfile s || return 1
967 #define OBD_FAIL_LDLM_REPLY 0x30c
968 do_facet mds "lctl set_param fail_loc=0x8000030c"
970 do_facet mds "lctl set_param fail_loc=0x0"
972 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
974 run_test 52 "time out lock replay (3764)"
976 # bug 3462 - simultaneous MDC requests
978 mkdir -p $DIR/${tdir}-1
979 mkdir -p $DIR/${tdir}-2
980 multiop $DIR/${tdir}-1/f O_c &
982 # give multiop a chance to open
985 #define OBD_FAIL_MDS_CLOSE_NET 0x115
986 do_facet mds "lctl set_param fail_loc=0x80000115"
987 kill -USR1 $close_pid
988 cancel_lru_locks MDC # force the close
989 do_facet mds "lctl set_param fail_loc=0"
990 mcreate $DIR/${tdir}-2/f || return 1
992 # close should still be here
993 [ -d /proc/$close_pid ] || return 2
994 replay_barrier_nodf mds
996 wait $close_pid || return 3
998 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
999 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1000 rm -rf $DIR/${tdir}-*
1002 run_test 53a "|X| close request while two MDC requests in flight"
1005 mkdir -p $DIR/$tdir-1
1006 mkdir -p $DIR/$tdir-2
1007 multiop $DIR/$tdir-1/f O_c &
1010 #define OBD_FAIL_MDS_REINT_NET 0x107
1011 do_facet mds "lctl set_param fail_loc=0x80000107"
1012 mcreate $DIR/${tdir}-2/f &
1016 do_facet mds "lctl set_param fail_loc=0"
1017 kill -USR1 $close_pid
1018 cancel_lru_locks MDC # force the close
1019 wait $close_pid || return 1
1020 # open should still be here
1021 [ -d /proc/$open_pid ] || return 2
1023 replay_barrier_nodf mds
1025 wait $open_pid || return 3
1027 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1028 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1029 rm -rf $DIR/${tdir}-*
1031 run_test 53b "|X| open request while two MDC requests in flight"
1034 mkdir -p $DIR/${tdir}-1
1035 mkdir -p $DIR/${tdir}-2
1036 multiop $DIR/${tdir}-1/f O_c &
1039 do_facet mds "lctl set_param fail_loc=0x80000107"
1040 mcreate $DIR/${tdir}-2/f &
1044 do_facet mds "lctl set_param fail_loc=0x80000115"
1045 kill -USR1 $close_pid
1046 cancel_lru_locks MDC # force the close
1048 replay_barrier_nodf mds
1050 wait $open_pid || return 1
1052 # close should be gone
1053 [ -d /proc/$close_pid ] && return 2
1054 do_facet mds "lctl set_param fail_loc=0"
1056 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1057 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1058 rm -rf $DIR/${tdir}-*
1060 run_test 53c "|X| open request and close request while two MDC requests in flight"
1063 mkdir -p $DIR/${tdir}-1
1064 mkdir -p $DIR/${tdir}-2
1065 multiop $DIR/${tdir}-1/f O_c &
1067 # give multiop a chance to open
1070 # define OBD_FAIL_MDS_CLOSE_NET_REP 0X138
1071 do_facet mds "lctl set_param fail_loc=0x8000013b"
1072 kill -USR1 $close_pid
1073 cancel_lru_locks MDC # force the close
1074 do_facet mds "lctl set_param fail_loc=0"
1075 mcreate $DIR/${tdir}-2/f || return 1
1077 # close should still be here
1078 [ -d /proc/$close_pid ] || return 2
1079 replay_barrier_nodf mds
1081 wait $close_pid || return 3
1083 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1084 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1085 rm -rf $DIR/${tdir}-*
1087 run_test 53d "|X| close reply while two MDC requests in flight"
1090 mkdir -p $DIR/$tdir-1
1091 mkdir -p $DIR/$tdir-2
1092 multiop $DIR/$tdir-1/f O_c &
1095 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1096 do_facet mds "lctl set_param fail_loc=0x80000119"
1097 mcreate $DIR/${tdir}-2/f &
1101 do_facet mds "lctl set_param fail_loc=0"
1102 kill -USR1 $close_pid
1103 cancel_lru_locks MDC # force the close
1104 wait $close_pid || return 1
1105 # open should still be here
1106 [ -d /proc/$open_pid ] || return 2
1108 replay_barrier_nodf mds
1110 wait $open_pid || return 3
1112 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1113 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1114 rm -rf $DIR/${tdir}-*
1116 run_test 53e "|X| open reply while two MDC requests in flight"
1119 mkdir -p $DIR/${tdir}-1
1120 mkdir -p $DIR/${tdir}-2
1121 multiop $DIR/${tdir}-1/f O_c &
1124 do_facet mds "lctl set_param fail_loc=0x80000119"
1125 mcreate $DIR/${tdir}-2/f &
1129 do_facet mds "lctl set_param fail_loc=0x8000013b"
1130 kill -USR1 $close_pid
1131 cancel_lru_locks MDC
1133 replay_barrier_nodf mds
1135 wait $open_pid || return 1
1137 #close should be gone
1138 [ -d /proc/$close_pid ] && return 2
1139 do_facet mds "lctl set_param fail_loc=0"
1141 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1142 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1143 rm -rf $DIR/${tdir}-*
1145 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1148 mkdir -p $DIR/${tdir}-1
1149 mkdir -p $DIR/${tdir}-2
1150 multiop $DIR/${tdir}-1/f O_c &
1153 do_facet mds "lctl set_param fail_loc=0x80000119"
1154 mcreate $DIR/${tdir}-2/f &
1158 do_facet mds "lctl set_param fail_loc=0x80000115"
1159 kill -USR1 $close_pid
1160 cancel_lru_locks MDC # force the close
1162 do_facet mds "lctl set_param fail_loc=0"
1163 replay_barrier_nodf mds
1165 wait $open_pid || return 1
1167 # close should be gone
1168 [ -d /proc/$close_pid ] && return 2
1170 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1171 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1172 rm -rf $DIR/${tdir}-*
1174 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1177 mkdir -p $DIR/${tdir}-1
1178 mkdir -p $DIR/${tdir}-2
1179 multiop $DIR/${tdir}-1/f O_c &
1182 do_facet mds "lctl set_param fail_loc=0x80000107"
1183 mcreate $DIR/${tdir}-2/f &
1187 do_facet mds "lctl set_param fail_loc=0x8000013b"
1188 kill -USR1 $close_pid
1189 cancel_lru_locks MDC # force the close
1192 replay_barrier_nodf mds
1194 wait $open_pid || return 1
1196 # close should be gone
1197 [ -d /proc/$close_pid ] && return 2
1198 do_facet mds "lctl set_param fail_loc=0"
1200 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1201 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1202 rm -rf $DIR/${tdir}-*
1204 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1206 #b3761 ASSERTION(hash != 0) failed
1208 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1209 do_facet mds "lctl set_param fail_loc=0x8000012b"
1211 # give touch a chance to run
1213 do_facet mds "lctl set_param fail_loc=0x0"
1217 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1219 #b3440 ASSERTION(rec->ur_fid2->id) failed
1221 ln -s foo $DIR/$tfile
1223 #drop_reply "cat $DIR/$tfile"
1227 run_test 56 "don't replay a symlink open request (3440)"
1229 #recovery one mds-ost setattr from llog
1231 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1232 do_facet mds "lctl set_param fail_loc=0x8000012c"
1237 $CHECKSTAT -t file $DIR/$tfile || return 1
1238 do_facet mds "lctl set_param fail_loc=0x0"
1241 run_test 57 "test recovery from llog for setattr op"
1243 #recovery many mds-ost setattr from llog
1245 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1246 do_facet mds "lctl set_param fail_loc=0x8000012c"
1247 createmany -o $DIR/$tdir/$tfile-%d 2500
1251 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1252 do_facet mds "lctl set_param fail_loc=0x0"
1253 unlinkmany $DIR/$tdir/$tfile-%d 2500
1256 run_test 58 "test recovery from llog for setattr op (test llog_gen_rec)"
1258 # log_commit_thread vs filter_destroy race used to lead to import use after free
1261 createmany -o $DIR/$tdir/$tfile-%d 200
1263 unlinkmany $DIR/$tdir/$tfile-%d 200
1264 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1265 do_facet ost1 "lctl set_param fail_loc=0x507"
1268 do_facet ost1 "lctl set_param fail_loc=0x0"
1272 run_test 59 "test log_commit_thread vs filter_destroy race"
1274 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1275 # bug 12086: should no oops and No ctxt error for this test
1277 createmany -o $DIR/$tdir/$tfile-%d 200
1279 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1281 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1282 local no_ctxt=`dmesg | grep "No ctxt"`
1283 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1285 run_test 60 "test llog post recovery init vs llog unlink"
1287 #test race llog recovery thread vs llog cleanup
1289 createmany -o $DIR/$tdir/$tfile-%d 800
1291 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1292 unlinkmany $DIR/$tdir/$tfile-%d 800
1293 do_facet ost "lctl set_param fail_loc=0x80000221"
1298 do_facet ost "lctl set_param fail_loc=0x0"
1299 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1302 run_test 61a "test race llog recovery vs llog cleanup"
1304 #test race mds llog sync vs llog cleanup
1306 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1307 do_facet mds "lctl set_param fail_loc=0x8000013a"
1311 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1313 run_test 61b "test race mds llog sync vs llog cleanup"
1315 #test race cancel cookie cb vs llog cleanup
1317 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1319 do_facet ost "lctl set_param fail_loc=0x80000222"
1324 run_test 61c "test race mds llog sync vs llog cleanup"
1326 test_62() { # Bug 15756 - don't mis-drop resent replay
1328 createmany -o $DIR/$tdir/$tfile- 25
1329 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1330 do_facet mds "lctl set_param fail_loc=0x80000707"
1332 df $MOUNT || return 1
1333 do_facet mds "lctl set_param fail_loc=0"
1334 unlinkmany $DIR/$tdir/$tfile- 25 || return 2
1337 run_test 62 "don't mis-drop resent replay"
1339 #Adaptive Timeouts (bug 3055)
1344 if ! at_is_valid; then
1345 skip "AT env is invalid"
1349 if ! at_is_enabled; then
1350 echo "AT is disabled, enable it by force temporarily"
1351 at_max_set 600 mds ost client
1355 if [ -z "$ATOLDBASE" ]; then
1356 local at_history=$(do_facet mds "find /sys/ -name at_history")
1357 [ -z "$at_history" ] && skip "missing /sys/.../at_history " && return 1
1358 ATOLDBASE=$(do_facet mds "cat $at_history")
1359 # speed up the timebase so we can check decreasing AT
1360 do_facet mds "echo 8 >> $at_history"
1361 do_facet ost1 "echo 8 >> $at_history"
1365 test_65a() #bug 3055
1367 at_start || return 0
1368 $LCTL dk > /dev/null
1370 lctl set_param debug="+other"
1371 # slow down a request
1372 do_facet mds lctl set_param fail_val=30000
1373 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1374 do_facet mds lctl set_param fail_loc=0x8000050a
1375 createmany -o $DIR/$tfile 10 > /dev/null
1376 unlinkmany $DIR/$tfile 10 > /dev/null
1377 # check for log message
1378 $LCTL dk | grep "Early reply #" || error "No early reply"
1379 # client should show 30s estimates
1380 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1382 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1384 run_test 65a "AT: verify early replies"
1386 test_65b() #bug 3055
1388 at_start || return 0
1391 lctl set_param debug="+other"
1392 $LCTL dk > /dev/null
1393 # slow down bulk i/o
1394 do_facet ost1 lctl set_param fail_val=30
1395 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1396 do_facet ost1 lctl set_param fail_loc=0x224
1399 lfs setstripe $DIR/$tfile --index=0 --count=1
1400 # force some real bulk transfer
1401 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1403 do_facet ost1 lctl set_param fail_loc=0
1404 # check for log message
1405 $LCTL dk | grep "Early reply #" || error "No early reply"
1407 # client should show 30s estimates
1408 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1410 run_test 65b "AT: verify early replies on packed reply / bulk"
1412 test_66a() #bug 3055
1414 at_start || return 0
1415 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1416 # adjust 5s at a time so no early reply is sent (within deadline)
1417 do_facet mds "lctl set_param fail_val=5000"
1418 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1419 do_facet mds "lctl set_param fail_loc=0x8000050a"
1420 createmany -o $DIR/$tfile 20 > /dev/null
1421 unlinkmany $DIR/$tfile 20 > /dev/null
1422 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1423 do_facet mds "lctl set_param fail_val=10000"
1424 do_facet mds "lctl set_param fail_loc=0x8000050a"
1425 createmany -o $DIR/$tfile 20 > /dev/null
1426 unlinkmany $DIR/$tfile 20 > /dev/null
1427 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1428 do_facet mds "lctl set_param fail_loc=0"
1430 createmany -o $DIR/$tfile 20 > /dev/null
1431 unlinkmany $DIR/$tfile 20 > /dev/null
1432 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal | grep "portal 12"
1433 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1434 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1435 echo "Current MDT timeout $CUR, worst $WORST"
1436 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1438 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1440 test_66b() #bug 3055
1442 at_start || return 0
1443 ORIG=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1444 lctl set_param fail_val=$(($ORIG + 5))
1445 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1446 lctl set_param fail_loc=0x50c
1447 ls $DIR/$tfile > /dev/null 2>&1
1448 lctl set_param fail_loc=0
1449 CUR=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1450 WORST=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $6}')
1451 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1452 [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
1454 run_test 66b "AT: verify net latency adjusts"
1456 test_67a() #bug 3055
1458 at_start || return 0
1459 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1460 # sleeping threads may drive values above this
1461 do_facet ost1 "lctl set_param fail_val=400"
1462 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1463 do_facet ost1 "lctl set_param fail_loc=0x50a"
1464 createmany -o $DIR/$tfile 20 > /dev/null
1465 unlinkmany $DIR/$tfile 20 > /dev/null
1466 do_facet ost1 "lctl set_param fail_loc=0"
1467 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1468 ATTEMPTS=$(($CONN2 - $CONN1))
1469 echo "$ATTEMPTS osc reconnect attemps on gradual slow"
1470 [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
1473 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1475 test_67b() #bug 3055
1477 at_start || return 0
1478 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1479 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1480 do_facet ost1 "lctl set_param fail_val=20000"
1481 do_facet ost1 "lctl set_param fail_loc=0x80000223"
1482 cp /etc/profile $DIR/$tfile || error "cp failed"
1484 lctl get_param -n ost.OSS.ost_create.timeouts
1486 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1487 ATTEMPTS=$(($CONN2 - $CONN1))
1488 echo "$ATTEMPTS osc reconnect attemps on instant slow"
1489 # do it again; should not timeout
1490 do_facet ost1 "lctl set_param fail_loc=0x80000223"
1491 cp /etc/profile $DIR/$tfile || error "cp failed"
1492 do_facet ost1 "lctl set_param fail_loc=0"
1494 lctl get_param -n ost.OSS.ost_create.timeouts
1495 CONN3=$(`lctl get_param -n osc.*.stats` | awk '/_connect/ {total+=$2} END {print total}')
1496 ATTEMPTS=$(($CONN3 - $CONN2))
1497 echo "$ATTEMPTS osc reconnect attemps on 2nd slow"
1498 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1501 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1503 test_68 () #bug 13813
1505 at_start || return 0
1506 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1507 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1508 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1509 echo $TIMEOUT >> $ldlm_enqueue_min
1510 rm -f $DIR/${tfile}_[1-2]
1511 lfs setstripe $DIR/$tfile --index=0 --count=1
1512 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1513 lctl set_param fail_val=$(($TIMEOUT - 1))
1514 lctl set_param fail_loc=0x80000312
1515 cp /etc/profile $DIR/${tfile}_1 || error "1st cp failed $?"
1516 lctl set_param fail_val=$((TIMEOUT * 3 / 2))
1517 lctl set_param fail_loc=0x80000312
1518 cp /etc/profile $DIR/${tfile}_2 || error "2nd cp failed $?"
1519 lctl set_param fail_loc=0
1520 echo $ENQ_MIN >> $ldlm_enqueue_min
1523 run_test 68 "AT: verify slowing locks"
1525 if [ -n "$ATOLDBASE" ]; then
1526 at_history=$(do_facet mds "find /sys/ -name at_history")
1527 do_facet mds "echo $ATOLDBASE >> $at_history" || true
1528 do_facet ost1 "echo $ATOLDBASE >> $at_history" || true
1531 if [ $AT_MAX_SET -ne 0 ]; then
1532 echo "restore AT status to be disabled"
1533 at_max_set 0 mds ost client
1536 # end of AT tests includes above lines
1538 # start multi-client tests
1540 [ -z "$CLIENTS" ] && \
1541 { skip "Need two or more clients." && return; }
1542 [ $CLIENTCOUNT -lt 2 ] && \
1543 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1545 echo "mount clients $CLIENTS ..."
1546 zconf_mount_clients $CLIENTS $DIR
1548 local clients=${CLIENTS//,/ }
1549 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1550 for CLIENT in $clients; do
1551 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1552 of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
1553 error "dd failed on $CLIENT"
1556 local prev_client=$(echo $clients | sed 's/^.* \(\w\+\)$/\1/')
1557 for C in ${CLIENTS//,/ }; do
1558 do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
1559 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
1565 zconf_umount_clients $CLIENTS $DIR
1567 run_test 70a "check multi client t-f"
1570 [ -z "$CLIENTS" ] && \
1571 { skip "Need two or more clients." && return; }
1572 [ $CLIENTCOUNT -lt 2 ] && \
1573 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1575 zconf_mount_clients $CLIENTS $DIR
1577 local duration="-t 60"
1578 local cmd="rundbench 1 $duration "
1580 for CLIENT in ${CLIENTS//,/ }; do
1581 $PDSH $CLIENT "set -x; PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:${DBENCH_LIB} DBENCH_LIB=${DBENCH_LIB} $cmd" &
1583 echo $PID >pid.$CLIENT
1584 echo "Started load PID=`cat pid.$CLIENT`"
1588 sleep 3 # give clients a time to do operations
1590 log "$TESTNAME fail mds 1"
1593 # wait for client to reconnect to MDS
1596 for CLIENT in ${CLIENTS//,/ }; do
1597 PID=`cat pid.$CLIENT`
1600 echo "load on ${CLIENT} returned $rc"
1603 zconf_umount_clients $CLIENTS $DIR
1605 run_test 70b "mds recovery; $CLIENTCOUNT clients"
1606 # end multi-client tests
1608 equals_msg `basename $0`: test complete, cleaning up
1609 check_and_cleanup_lustre
1610 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true