7 # This test needs to be run on the client
10 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
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:-""}
19 remote_mds_nodsh && log "SKIP: remote MDS with nodsh" && exit 0
22 # bug number: 17466 15962
23 ALWAYS_EXCEPT="61d 33b $REPLAY_SINGLE_EXCEPT"
25 if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
26 CONFIG_EXCEPTIONS="0b 42 47 61a 61c"
27 echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
28 echo "Except the tests: $CONFIG_EXCEPTIONS"
29 ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
32 # 63 min 7 min AT AT AT AT"
33 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a 44b 65 66 67 68"
37 check_and_setup_lustre
42 rm -rf $DIR/[df][0-9]*
44 test_0a() { # was test_0
47 replay_barrier $SINGLEMDS
51 run_test 0a "empty replay"
54 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
56 # this test attempts to trigger a race in the precreation code,
57 # and must run before any other objects are created on the filesystem
59 createmany -o $DIR/$tfile 20 || return 1
60 unlinkmany $DIR/$tfile 20 || return 2
62 run_test 0b "ensure object created after recover exists. (3284)"
68 lctl set_param -n seq.cli-srv-$mds-mdc-*.width=$width
74 lctl get_param -n seq.cli-srv-$mds-mdc-*.width
77 # This test should pass for single-mds and multi-mds configs.
78 # But for different configurations it tests different things.
82 # (1) fld_create replay should happen;
84 # (2) fld_create replay should not return -EEXISTS, if it does
85 # this means sequence manager recovery code is buggy and allocated
86 # same sequence two times after recovery.
90 # (1) fld_create replay may not happen, because its home MDS is
91 # MDS2 which is not involved to revovery;
93 # (2) as fld_create does not happen on MDS1, it does not make any
96 local label=`mdsdevlabel 1`
97 [ -z "$label" ] && echo "No label for mds1" && return 1
99 replay_barrier $SINGLEMDS
100 local sw=`seq_get_width $label`
102 # make seq manager switch to next sequence each
103 # time as new fid is needed.
104 seq_set_width $label 1
106 # make sure that fld has created at least one new
108 touch $DIR/$tfile || return 2
109 seq_set_width $label $sw
111 # fail $SINGLEMDS and start recovery, replay RPCs, etc.
114 # wait for recovery finish
118 # flush fld cache and dentry cache to make it lookup
119 # created entry instead of revalidating existent one
121 zconf_mount `hostname` $MOUNT
123 # issue lookup which should call fld lookup which
124 # should fail if client did not replay fld create
125 # correctly and server has no fld entry
126 touch $DIR/$tfile || return 3
127 rm $DIR/$tfile || return 4
129 run_test 0c "fld create"
132 replay_barrier $SINGLEMDS
135 $CHECKSTAT -t file $DIR/$tfile || return 1
138 run_test 1 "simple create"
141 replay_barrier $SINGLEMDS
144 $CHECKSTAT -t file $DIR/$tfile || return 1
151 replay_barrier $SINGLEMDS
154 $CHECKSTAT -t file $DIR/$tfile || return 1
160 local file=$DIR/$tfile
161 replay_barrier $SINGLEMDS
163 openfile -f O_DIRECTORY $file
165 $CHECKSTAT -t file $file || return 2
168 run_test 3a "replay failed open(O_DIRECTORY)"
171 replay_barrier $SINGLEMDS
172 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
173 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000114"
175 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
177 $CHECKSTAT -t file $DIR/$tfile && return 2
180 run_test 3b "replay failed open -ENOMEM"
183 replay_barrier $SINGLEMDS
184 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
185 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
187 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
190 $CHECKSTAT -t file $DIR/$tfile && return 2
193 run_test 3c "replay failed open -ENOMEM"
195 test_4a() { # was test_4
196 replay_barrier $SINGLEMDS
197 for i in `seq 10`; do
198 echo "tag-$i" > $DIR/$tfile-$i
201 for i in `seq 10`; do
202 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
205 run_test 4a "|x| 10 open(O_CREAT)s"
208 replay_barrier $SINGLEMDS
211 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
213 run_test 4b "|x| rm 10 files"
215 # The idea is to get past the first block of precreated files on both
216 # osts, and then replay.
218 replay_barrier $SINGLEMDS
219 for i in `seq 220`; do
220 echo "tag-$i" > $DIR/$tfile-$i
223 for i in `seq 220`; do
224 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
228 # waiting for commitment of removal
230 run_test 5 "|x| 220 open(O_CREAT)"
233 test_6a() { # was test_6
235 replay_barrier $SINGLEMDS
236 mcreate $DIR/$tdir/$tfile
238 $CHECKSTAT -t dir $DIR/$tdir || return 1
239 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
241 # waiting for log process thread
243 run_test 6a "mkdir + contained create"
247 replay_barrier $SINGLEMDS
250 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
252 run_test 6b "|X| rmdir"
256 replay_barrier $SINGLEMDS
257 mcreate $DIR/$tdir/$tfile
259 $CHECKSTAT -t dir $DIR/$tdir || return 1
260 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
263 run_test 7 "mkdir |X| contained create"
266 # make sure no side-effect from previous test.
268 replay_barrier $SINGLEMDS
269 multiop_bg_pause $DIR/$tfile mo_c || return 4
273 $CHECKSTAT -t file $DIR/$tfile || return 1
274 kill -USR1 $MULTIPID || return 2
275 wait $MULTIPID || return 3
278 run_test 8 "creat open |X| close"
281 replay_barrier $SINGLEMDS
283 local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
285 local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
287 echo " old_inum == $old_inum, new_inum == $new_inum"
288 if [ $old_inum -eq $new_inum ] ;
290 echo " old_inum and new_inum match"
292 echo "!!!! old_inum and new_inum NOT match"
297 run_test 9 "|X| create (same inum/gen)"
301 replay_barrier $SINGLEMDS
302 mv $DIR/$tfile $DIR/$tfile-2
305 $CHECKSTAT $DIR/$tfile && return 1
306 $CHECKSTAT $DIR/$tfile-2 ||return 2
310 run_test 10 "create |X| rename unlink"
314 echo "old" > $DIR/$tfile
315 mv $DIR/$tfile $DIR/$tfile-2
316 replay_barrier $SINGLEMDS
317 echo "new" > $DIR/$tfile
319 grep old $DIR/$tfile-2
321 grep new $DIR/$tfile || return 1
322 grep old $DIR/$tfile-2 || return 2
324 run_test 11 "create open write rename |X| create-old-name read"
328 multiop_bg_pause $DIR/$tfile o_tSc || return 3
331 replay_barrier $SINGLEMDS
333 wait $pid || return 1
336 [ -e $DIR/$tfile ] && return 2
339 run_test 12 "open, unlink |X| close"
342 # 1777 - replay open after committed chmod that would make
343 # a regular open a failure
346 multiop_bg_pause $DIR/$tfile O_wc || return 3
349 $CHECKSTAT -p 0 $DIR/$tfile
350 replay_barrier $SINGLEMDS
353 wait $pid || return 1
355 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
358 run_test 13 "open chmod 0 |x| write close"
361 multiop_bg_pause $DIR/$tfile O_tSc || return 4
364 replay_barrier $SINGLEMDS
365 kill -USR1 $pid || return 1
366 wait $pid || return 2
369 [ -e $DIR/$tfile ] && return 3
372 run_test 14 "open(O_CREAT), unlink |X| close"
375 multiop_bg_pause $DIR/$tfile O_tSc || return 5
378 replay_barrier $SINGLEMDS
379 touch $DIR/g11 || return 1
381 wait $pid || return 2
384 [ -e $DIR/$tfile ] && return 3
385 touch $DIR/h11 || return 4
388 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
392 replay_barrier $SINGLEMDS
395 mcreate $DIR/$tfile-2
397 [ -e $DIR/$tfile ] && return 1
398 [ -e $DIR/$tfile-2 ] || return 2
399 munlink $DIR/$tfile-2 || return 3
401 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
404 replay_barrier $SINGLEMDS
405 multiop_bg_pause $DIR/$tfile O_c || return 4
408 kill -USR1 $pid || return 1
409 wait $pid || return 2
410 $CHECKSTAT -t file $DIR/$tfile || return 3
413 run_test 17 "|X| open(O_CREAT), |replay| close"
416 replay_barrier $SINGLEMDS
417 multiop_bg_pause $DIR/$tfile O_tSc || return 8
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 "
448 test_20a() { # was test_20
449 replay_barrier $SINGLEMDS
450 multiop_bg_pause $DIR/$tfile O_tSc || return 3
456 wait $pid || return 1
457 [ -e $DIR/$tfile ] && return 2
460 run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
462 test_20b() { # bug 10480
463 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
465 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
467 while [ ! -e $DIR/$tfile ] ; do
468 usleep 60 # give dd a chance to start
471 lfs getstripe $DIR/$tfile || return 1
472 rm -f $DIR/$tfile || return 2 # make it an orphan
474 df -P $DIR || df -P $DIR || true # reconnect
476 fail $SINGLEMDS # start orphan recovery
477 df -P $DIR || df -P $DIR || true # reconnect
478 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
480 # just because recovery is done doesn't mean we've finished
481 # orphan cleanup. Wait for llogs to get synchronized.
482 echo waiting for orphan cleanup...
484 local -a sync=($(do_facet ost "$LCTL get_param obdfilter.*.mds_sync" | awk -F= ' {print $2}'))
486 for ((i=0; i<${#sync[@]}; i++)); do
487 [ ${sync[$i]} -eq 0 ] && continue
488 # there is a not finished MDS-OST synchronization
492 [ ${con} -eq 1 ] && break
495 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
496 log "before $BEFOREUSED, after $AFTERUSED"
497 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
498 error "after $AFTERUSED > before $BEFOREUSED"
501 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
503 test_20c() { # bug 10480
504 multiop_bg_pause $DIR/$tfile Ow_c || return 1
511 df -P $DIR || df -P $DIR || true # reconnect
514 test -s $DIR/$tfile || error "File was truncated"
516 wait $pid || return 1
519 run_test 20c "check that client eviction does not affect file content"
522 replay_barrier $SINGLEMDS
523 multiop_bg_pause $DIR/$tfile O_tSc || return 5
526 touch $DIR/g11 || return 1
530 wait $pid || return 2
531 [ -e $DIR/$tfile ] && return 3
532 touch $DIR/h11 || return 4
535 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
538 multiop_bg_pause $DIR/$tfile O_tSc || return 3
541 replay_barrier $SINGLEMDS
546 wait $pid || return 1
547 [ -e $DIR/$tfile ] && return 2
550 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
553 multiop_bg_pause $DIR/$tfile O_tSc || return 5
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_bg_pause $DIR/$tfile O_tSc || return 3
573 replay_barrier $SINGLEMDS
577 wait $pid || return 1
578 [ -e $DIR/$tfile ] && return 2
581 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
584 multiop_bg_pause $DIR/$tfile O_tSc || return 3
588 replay_barrier $SINGLEMDS
591 wait $pid || return 1
592 [ -e $DIR/$tfile ] && return 2
595 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
598 replay_barrier $SINGLEMDS
599 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
601 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
606 wait $pid2 || return 1
610 wait $pid1 || return 2
611 [ -e $DIR/$tfile-1 ] && return 3
612 [ -e $DIR/$tfile-2 ] && return 4
615 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
618 replay_barrier $SINGLEMDS
619 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
621 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
628 wait $pid1 || return 1
630 wait $pid2 || return 2
631 [ -e $DIR/$tfile-1 ] && return 3
632 [ -e $DIR/$tfile-2 ] && return 4
635 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
638 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
640 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
642 replay_barrier $SINGLEMDS
646 wait $pid2 || return 1
650 wait $pid1 || return 2
651 [ -e $DIR/$tfile-1 ] && return 3
652 [ -e $DIR/$tfile-2 ] && return 4
655 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
658 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
660 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
662 replay_barrier $SINGLEMDS
668 wait $pid1 || return 1
670 wait $pid2 || return 2
671 [ -e $DIR/$tfile-1 ] && return 3
672 [ -e $DIR/$tfile-2 ] && return 4
675 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
678 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
680 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
685 replay_barrier $SINGLEMDS
688 wait $pid1 || return 1
690 wait $pid2 || return 2
691 [ -e $DIR/$tfile-1 ] && return 3
692 [ -e $DIR/$tfile-2 ] && return 4
695 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
698 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
700 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
704 replay_barrier $SINGLEMDS
708 wait $pid1 || return 1
710 wait $pid2 || return 2
711 [ -e $DIR/$tfile-1 ] && return 3
712 [ -e $DIR/$tfile-2 ] && return 4
715 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
717 # tests for bug 2104; completion without crashing is success. The close is
718 # stale, but we always return 0 for close, so the app never sees it.
720 multiop_bg_pause $DIR/$tfile O_c || return 2
722 multiop_bg_pause $DIR/$tfile O_c || return 3
725 df $MOUNT || sleep 1 && df $MOUNT || return 1
728 wait $pid1 || return 4
729 wait $pid2 || return 5
732 run_test 32 "close() notices client eviction; close() after client eviction"
734 # Abort recovery before client complete
735 test_33a() { # was test_33
736 replay_barrier $SINGLEMDS
737 createmany -o $DIR/$tfile-%d 100
738 fail_abort $SINGLEMDS
739 # this file should be gone, because the replay was aborted
740 $CHECKSTAT -t file $DIR/$tfile-* && return 3
741 unlinkmany $DIR/$tfile-%d 0 100
744 run_test 33a "abort recovery before client does replay"
746 # Stale FID sequence bug 15962
747 test_33b() { # was test_33a
748 replay_barrier $SINGLEMDS
749 createmany -o $DIR/$tfile-%d 10
750 fail_abort $SINGLEMDS
751 unlinkmany $DIR/$tfile-%d 0 10
752 # recreate shouldn't fail
753 createmany -o $DIR/$tfile-%d 10 || return 3
754 unlinkmany $DIR/$tfile-%d 0 10
757 run_test 33b "fid shouldn't be reused after abort recovery"
760 multiop_bg_pause $DIR/$tfile O_c || return 2
764 replay_barrier $SINGLEMDS
765 fail_abort $SINGLEMDS
767 wait $pid || return 3
768 [ -e $DIR/$tfile ] && return 1
772 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
774 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
778 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
779 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
784 # give a chance to remove from MDS
785 fail_abort $SINGLEMDS
786 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
788 run_test 35 "test recovery from llog for unlink op"
790 # b=2432 resent cancel after replay uses wrong cookie,
791 # so don't resend cancels
793 replay_barrier $SINGLEMDS
795 checkstat $DIR/$tfile
796 facet_failover $SINGLEMDS
798 if dmesg | grep "unknown lock cookie"; then
799 echo "cancel after replay failed"
803 run_test 36 "don't resend cancel"
806 # directory orphans can't be unlinked from PENDING directory
808 rmdir $DIR/$tfile 2>/dev/null
809 multiop_bg_pause $DIR/$tfile dD_c || return 2
813 replay_barrier $SINGLEMDS
814 # clear the dmesg buffer so we only see errors from this recovery
816 fail_abort $SINGLEMDS
818 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
819 wait $pid || return 3
823 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
826 createmany -o $DIR/$tfile-%d 800
827 unlinkmany $DIR/$tfile-%d 0 400
828 replay_barrier $SINGLEMDS
830 unlinkmany $DIR/$tfile-%d 400 400
832 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
834 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
836 test_39() { # bug 4176
837 createmany -o $DIR/$tfile-%d 800
838 replay_barrier $SINGLEMDS
839 unlinkmany $DIR/$tfile-%d 0 400
841 unlinkmany $DIR/$tfile-%d 400 400
843 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
845 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
848 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
853 $LCTL mark multiop $MOUNT/$tfile OS_c
854 multiop $MOUNT/$tfile OS_c &
856 writeme -s $MOUNT/${tfile}-2 &
859 facet_failover $SINGLEMDS
860 #define OBD_FAIL_MDS_CONNECT_NET 0x117
861 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
863 stat1=`count_ost_writes`
865 stat2=`count_ost_writes`
866 echo "$stat1, $stat2"
867 if [ $stat1 -lt $stat2 ]; then
868 echo "writes continuing during recovery"
871 echo "writes not continuing during recovery, bug 2477"
874 echo "waiting for writeme $WRITE_PID"
878 echo "waiting for multiop $PID"
879 wait $PID || return 2
880 do_facet client munlink $MOUNT/$tfile || return 3
881 do_facet client munlink $MOUNT/${tfile}-2 || return 3
884 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
888 # make sure that a read to one osc doesn't try to double-unlock its page just
889 # because another osc is invalid. trigger_group_io used to mistakenly return
890 # an error if any oscs were invalid even after having successfully put rpcs
891 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
892 # the page, guarnateeing that the unlock from the RPC completion would
893 # assert on trying to unlock the unlocked page.
895 [ $OSTCOUNT -lt 2 ] && \
896 skip "skipping test 41: we don't have a second OST to test with" && \
899 local f=$MOUNT/$tfile
900 # make sure the start of the file is ost1
901 lfs setstripe $f -s $((128 * 1024)) -i 0
902 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
904 # fail ost2 and read from ost1
905 local osc2dev=`do_facet $SINGLEMDS "lctl get_param -n devices | grep ${ost2_svc}-osc-MDT0000" | awk '{print $1}'`
906 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices && return 4
907 do_facet $SINGLEMDS $LCTL --device $osc2dev deactivate || return 1
908 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
909 do_facet $SINGLEMDS $LCTL --device $osc2dev activate || return 2
912 run_test 41 "read from a valid osc while other oscs are invalid"
914 # test MDS recovery after ost failure
916 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
917 createmany -o $DIR/$tfile-%d 800
919 unlinkmany $DIR/$tfile-%d 0 400
921 lctl set_param debug=-1
924 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
925 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
926 #[ $blocks_after -lt $blocks ] || return 1
927 echo wait for MDS to timeout and recover
928 sleep $((TIMEOUT * 2))
930 unlinkmany $DIR/$tfile-%d 400 400
931 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
933 run_test 42 "recovery after ost failure"
935 # timeout in MDS/OST recovery RPC will LBUG MDS
936 test_43() { # bug 2530
937 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
939 replay_barrier $SINGLEMDS
941 # OBD_FAIL_OST_CREATE_NET 0x204
942 do_facet ost1 "lctl set_param fail_loc=0x80000204"
945 do_facet ost1 "lctl set_param fail_loc=0"
949 run_test 43 "mds osc import failure during recovery; don't LBUG"
951 test_44a() { # was test_44
954 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
955 [ "$mdcdev" ] || exit 2
957 # adaptive timeouts slow this way down
958 if at_is_enabled; then
959 at_max_saved=$(at_max_get mds)
963 for i in `seq 1 10`; do
964 echo "$i of 10 ($(date +%s))"
965 do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
966 #define OBD_FAIL_TGT_CONN_RACE 0x701
967 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
968 $LCTL --device $mdcdev recover
971 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
972 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
975 run_test 44a "race in target handle connect"
978 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
979 [ "$mdcdev" ] || exit 2
980 for i in `seq 1 10`; do
981 echo "$i of 10 ($(date +%s))"
982 do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
983 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
984 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
985 $LCTL --device $mdcdev recover
988 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
991 run_test 44b "race in target handle connect"
993 # Handle failed close
995 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
996 [ "$mdcdev" ] || exit 2
997 $LCTL --device $mdcdev recover
999 multiop_bg_pause $DIR/$tfile O_c || return 1
1002 # This will cause the CLOSE to fail before even
1003 # allocating a reply buffer
1004 $LCTL --device $mdcdev deactivate || return 4
1008 wait $pid || return 1
1010 $LCTL --device $mdcdev activate || return 5
1013 $CHECKSTAT -t file $DIR/$tfile || return 2
1016 run_test 45 "Handle failed close"
1020 drop_reply "touch $DIR/$tfile"
1022 # ironically, the previous test, 45, will cause a real forced close,
1023 # so just look for one for this test
1024 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1027 run_test 46 "Don't leak file handle after open resend (3325)"
1029 test_47() { # bug 2824
1030 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1032 # create some files to make sure precreate has been done on all
1033 # OSTs. (just in case this test is run independently)
1034 createmany -o $DIR/$tfile 20 || return 1
1036 # OBD_FAIL_OST_CREATE_NET 0x204
1038 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1039 df $MOUNT || return 2
1041 # let the MDS discover the OST failure, attempt to recover, fail
1042 # and recover again.
1043 sleep $((3 * TIMEOUT))
1045 # Without 2824, this createmany would hang
1046 createmany -o $DIR/$tfile 20 || return 3
1047 unlinkmany $DIR/$tfile 20 || return 4
1049 do_facet ost1 "lctl set_param fail_loc=0"
1052 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1055 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1056 [ "$OSTCOUNT" -lt "2" ] && skip "$OSTCOUNT < 2 OSTs -- skipping" && return
1058 replay_barrier $SINGLEMDS
1059 createmany -o $DIR/$tfile 20 || return 1
1060 # OBD_FAIL_OST_EROFS 0x216
1061 facet_failover $SINGLEMDS
1062 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1063 df $MOUNT || return 2
1065 createmany -o $DIR/$tfile 20 20 || return 2
1066 unlinkmany $DIR/$tfile 40 || return 3
1069 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1072 local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
1073 [ "$oscdev" ] || return 1
1074 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1075 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1076 # give the mds_lov_sync threads a chance to run
1079 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1081 # b3764 timed out lock replay
1084 cancel_lru_locks mdc
1086 multiop $DIR/$tfile s || return 1
1087 replay_barrier $SINGLEMDS
1088 #define OBD_FAIL_LDLM_REPLY 0x30c
1089 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1090 fail $SINGLEMDS || return 2
1091 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1093 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1095 run_test 52 "time out lock replay (3764)"
1097 # bug 3462 - simultaneous MDC requests
1099 mkdir -p $DIR/${tdir}-1
1100 mkdir -p $DIR/${tdir}-2
1101 multiop $DIR/${tdir}-1/f O_c &
1103 # give multiop a change to open
1106 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1107 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1108 kill -USR1 $close_pid
1109 cancel_lru_locks mdc # force the close
1110 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1112 mcreate $DIR/${tdir}-2/f || return 1
1114 # close should still be here
1115 [ -d /proc/$close_pid ] || return 2
1117 replay_barrier_nodf $SINGLEMDS
1119 wait $close_pid || return 3
1121 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1122 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1123 rm -rf $DIR/${tdir}-*
1125 run_test 53a "|X| close request while two MDC requests in flight"
1128 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1130 mkdir -p $DIR/${tdir}-1
1131 mkdir -p $DIR/${tdir}-2
1132 multiop $DIR/${tdir}-1/f O_c &
1135 #define OBD_FAIL_MDS_REINT_NET 0x107
1136 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1137 mcreate $DIR/${tdir}-2/f &
1141 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1142 kill -USR1 $close_pid
1143 cancel_lru_locks mdc # force the close
1144 wait $close_pid || return 1
1145 # open should still be here
1146 [ -d /proc/$open_pid ] || return 2
1148 replay_barrier_nodf $SINGLEMDS
1150 wait $open_pid || return 3
1152 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1153 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1154 rm -rf $DIR/${tdir}-*
1156 run_test 53b "|X| open request while two MDC requests in flight"
1159 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1161 mkdir -p $DIR/${tdir}-1
1162 mkdir -p $DIR/${tdir}-2
1163 multiop $DIR/${tdir}-1/f O_c &
1166 #define OBD_FAIL_MDS_REINT_NET 0x107
1167 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1168 mcreate $DIR/${tdir}-2/f &
1172 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1173 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1174 kill -USR1 $close_pid
1175 cancel_lru_locks mdc # force the close
1177 replay_barrier_nodf $SINGLEMDS
1178 fail_nodf $SINGLEMDS
1179 wait $open_pid || return 1
1181 # close should be gone
1182 [ -d /proc/$close_pid ] && return 2
1183 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1185 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1186 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1187 rm -rf $DIR/${tdir}-*
1189 run_test 53c "|X| open request and close request while two MDC requests in flight"
1192 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1194 mkdir -p $DIR/${tdir}-1
1195 mkdir -p $DIR/${tdir}-2
1196 multiop $DIR/${tdir}-1/f O_c &
1198 # give multiop a chance to open
1201 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1202 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1203 kill -USR1 $close_pid
1204 cancel_lru_locks mdc # force the close
1205 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1206 mcreate $DIR/${tdir}-2/f || return 1
1208 # close should still be here
1209 [ -d /proc/$close_pid ] || return 2
1211 wait $close_pid || return 3
1213 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1214 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1215 rm -rf $DIR/${tdir}-*
1217 run_test 53d "|X| close reply while two MDC requests in flight"
1220 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1222 mkdir -p $DIR/${tdir}-1
1223 mkdir -p $DIR/${tdir}-2
1224 multiop $DIR/${tdir}-1/f O_c &
1227 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1228 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1229 mcreate $DIR/${tdir}-2/f &
1233 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1234 kill -USR1 $close_pid
1235 cancel_lru_locks mdc # force the close
1236 wait $close_pid || return 1
1237 # open should still be here
1238 [ -d /proc/$open_pid ] || return 2
1240 replay_barrier_nodf $SINGLEMDS
1242 wait $open_pid || return 3
1244 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1245 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1246 rm -rf $DIR/${tdir}-*
1248 run_test 53e "|X| open reply while two MDC requests in flight"
1251 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1253 mkdir -p $DIR/${tdir}-1
1254 mkdir -p $DIR/${tdir}-2
1255 multiop $DIR/${tdir}-1/f O_c &
1258 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1259 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1260 mcreate $DIR/${tdir}-2/f &
1264 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1265 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1266 kill -USR1 $close_pid
1267 cancel_lru_locks mdc # force the close
1269 replay_barrier_nodf $SINGLEMDS
1270 fail_nodf $SINGLEMDS
1271 wait $open_pid || return 1
1273 # close should be gone
1274 [ -d /proc/$close_pid ] && return 2
1275 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1277 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1278 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1279 rm -rf $DIR/${tdir}-*
1281 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1284 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1286 mkdir -p $DIR/${tdir}-1
1287 mkdir -p $DIR/${tdir}-2
1288 multiop $DIR/${tdir}-1/f O_c &
1291 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1292 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1293 mcreate $DIR/${tdir}-2/f &
1297 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1298 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1299 kill -USR1 $close_pid
1300 cancel_lru_locks mdc # force the close
1302 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1303 replay_barrier_nodf $SINGLEMDS
1304 fail_nodf $SINGLEMDS
1305 wait $open_pid || return 1
1307 # close should be gone
1308 [ -d /proc/$close_pid ] && return 2
1310 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1311 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1312 rm -rf $DIR/${tdir}-*
1314 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1317 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1319 mkdir -p $DIR/${tdir}-1
1320 mkdir -p $DIR/${tdir}-2
1321 multiop $DIR/${tdir}-1/f O_c &
1324 #define OBD_FAIL_MDS_REINT_NET 0x107
1325 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1326 mcreate $DIR/${tdir}-2/f &
1330 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1331 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1332 kill -USR1 $close_pid
1333 cancel_lru_locks mdc # force the close
1336 replay_barrier_nodf $SINGLEMDS
1337 fail_nodf $SINGLEMDS
1338 wait $open_pid || return 1
1340 # close should be gone
1341 [ -d /proc/$close_pid ] && return 2
1342 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1344 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1345 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1346 rm -rf $DIR/${tdir}-*
1348 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1350 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1352 #b3761 ASSERTION(hash != 0) failed
1354 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1355 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1357 # give touch a chance to run
1359 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1363 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1365 #b3440 ASSERTION(rec->ur_fid2->id) failed
1367 ln -s foo $DIR/$tfile
1368 replay_barrier $SINGLEMDS
1369 #drop_reply "cat $DIR/$tfile"
1373 run_test 56 "don't replay a symlink open request (3440)"
1375 #recovery one mds-ost setattr from llog
1377 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1378 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1380 replay_barrier $SINGLEMDS
1383 $CHECKSTAT -t file $DIR/$tfile || return 1
1384 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1387 run_test 57 "test recovery from llog for setattr op"
1389 #recovery many mds-ost setattr from llog
1392 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1393 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1394 createmany -o $DIR/$tdir/$tfile-%d 2500
1395 replay_barrier $SINGLEMDS
1398 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1399 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1400 unlinkmany $DIR/$tdir/$tfile-%d 2500
1403 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
1406 mount_client $MOUNT2
1408 touch $DIR/$tdir/$tfile
1409 replay_barrier $SINGLEMDS
1410 setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile
1412 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1413 [ x$VAL = x"bar" ] || return 1
1414 rm -f $DIR/$tdir/$tfile
1416 zconf_umount `hostname` $MOUNT2
1418 run_test 58b "test replay of setxattr op"
1420 test_58c() { # bug 16570
1421 mount_client $MOUNT2
1423 touch $DIR/$tdir/$tfile
1424 drop_request "setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile" || \
1426 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1427 [ x$VAL = x"bar" ] || return 2
1428 drop_reint_reply "setfattr -n trusted.foo1 -v bar1 $DIR/$tdir/$tfile" || \
1430 VAL=`getfattr --absolute-names --only-value -n trusted.foo1 $MOUNT2/$tdir/$tfile`
1431 [ x$VAL = x"bar1" ] || return 4
1432 rm -f $DIR/$tdir/$tfile
1434 zconf_umount `hostname` $MOUNT2
1436 run_test 58c "resend/reconstruct setxattr op"
1438 # log_commit_thread vs filter_destroy race used to lead to import use after free
1441 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1444 createmany -o $DIR/$tdir/$tfile-%d 200
1446 unlinkmany $DIR/$tdir/$tfile-%d 200
1447 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1448 do_facet ost1 "lctl set_param fail_loc=0x507"
1451 do_facet ost1 "lctl set_param fail_loc=0x0"
1455 run_test 59 "test log_commit_thread vs filter_destroy race"
1457 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1458 # bug 12086: should no oops and No ctxt error for this test
1461 createmany -o $DIR/$tdir/$tfile-%d 200
1462 replay_barrier $SINGLEMDS
1463 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1465 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1466 local no_ctxt=`dmesg | grep "No ctxt"`
1467 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1469 run_test 60 "test llog post recovery init vs llog unlink"
1471 #test race llog recovery thread vs llog cleanup
1472 test_61a() { # was test_61
1473 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1476 createmany -o $DIR/$tdir/$tfile-%d 800
1478 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1479 unlinkmany $DIR/$tdir/$tfile-%d 800
1480 set_nodes_failloc "$(osts_nodes)" 0x80000221
1485 set_nodes_failloc "$(osts_nodes)" 0x0
1487 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1490 run_test 61a "test race llog recovery vs llog cleanup"
1492 #test race mds llog sync vs llog cleanup
1494 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1495 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
1496 facet_failover $SINGLEMDS
1499 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1501 run_test 61b "test race mds llog sync vs llog cleanup"
1503 #test race cancel cookie cb vs llog cleanup
1505 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1507 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1509 set_nodes_failloc "$(osts_nodes)" 0x80000222
1513 set_nodes_failloc "$(osts_nodes)" 0x0
1515 run_test 61c "test race mds llog sync vs llog cleanup"
1517 test_61d() { # bug 16002 # bug 17466
1518 shutdown_facet $SINGLEMDS
1519 #define OBD_FAIL_OBD_LLOG_SETUP 0x605
1520 do_facet $SINGLEMDS "lctl set_param fail_loc=0x605"
1521 start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS && error "mds start should have failed"
1522 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1523 start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS || error "cannot restart mds"
1525 run_test 61d "error in llog_setup should cleanup the llog context correctly"
1527 test_62() { # Bug 15756 - don't mis-drop resent replay
1529 replay_barrier $SINGLEMDS
1530 createmany -o $DIR/$tdir/$tfile- 25
1531 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1532 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1533 facet_failover $SINGLEMDS
1534 df $MOUNT || return 1
1535 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1536 unlinkmany $DIR/$tdir/$tfile- 25 || return 2
1539 run_test 62 "don't mis-drop resent replay"
1541 #Adaptive Timeouts (bug 3055)
1549 echo "Cleaning up AT ..."
1550 if [ -n "$ATOLDBASE" ]; then
1551 local at_history=$($LCTL get_param -n at_history)
1552 do_facet mds "lctl set_param at_history=$at_history" || true
1553 do_facet ost1 "lctl set_param at_history=$at_history" || true
1556 if [ $AT_MAX_SET -ne 0 ]; then
1557 for facet in mds client ost; do
1558 var=AT_MAX_SAVE_${facet}
1559 echo restore AT on $facet to saved value ${!var}
1560 at_max_set ${!var} $facet
1561 at_new=$(at_max_get $facet)
1562 echo Restored AT value on $facet $at_new
1563 [ $at_new -eq ${!var} ] || \
1564 error "$facet : AT value was not restored SAVED ${!var} NEW $at_new"
1571 local at_max_new=600
1573 # Save at_max original values
1575 if [ $AT_MAX_SET -eq 0 ]; then
1576 # Suppose that all osts have the same at_max
1577 for facet in mds client ost; do
1578 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1582 for facet in mds client ost; do
1583 at_max=$(at_max_get $facet)
1584 if [ $at_max -ne $at_max_new ]; then
1585 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1586 at_max_set $at_max_new $facet
1591 if [ -z "$ATOLDBASE" ]; then
1592 ATOLDBASE=$(do_facet mds "lctl get_param -n at_history")
1593 # speed up the timebase so we can check decreasing AT
1594 do_facet mds "lctl set_param at_history=8" || true
1595 do_facet ost1 "lctl set_param at_history=8" || true
1597 # sleep for a while to cool down, should be > 8s and also allow
1598 # at least one ping to be sent. simply use TIMEOUT to be safe.
1603 test_65a() #bug 3055
1605 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1607 at_start || return 0
1608 $LCTL dk > /dev/null
1610 sysctl -w lnet.debug="+other"
1611 # Slow down a request to the current service time, this is critical
1612 # because previous tests may have caused this value to increase.
1613 REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
1614 awk '/portal 12/ {print $5}'`
1615 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1617 do_facet mds lctl set_param fail_val=$((${REQ_DELAY} * 1000))
1618 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1619 do_facet mds sysctl -w lustre.fail_loc=0x8000050a
1620 createmany -o $DIR/$tfile 10 > /dev/null
1621 unlinkmany $DIR/$tfile 10 > /dev/null
1622 # check for log message
1623 $LCTL dk | grep "Early reply #" || error "No early reply"
1625 # client should show REQ_DELAY estimates
1626 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1628 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1630 run_test 65a "AT: verify early replies"
1632 test_65b() #bug 3055
1634 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1636 at_start || return 0
1639 sysctl -w lnet.debug="other trace"
1640 $LCTL dk > /dev/null
1641 # Slow down a request to the current service time, this is critical
1642 # because previous tests may have caused this value to increase.
1643 REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
1644 awk '/portal 6/ {print $5}'`
1645 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1647 do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
1648 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1649 do_facet ost1 sysctl -w lustre.fail_loc=0x224
1652 lfs setstripe $DIR/$tfile --index=0 --count=1
1653 # force some real bulk transfer
1654 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1656 do_facet ost1 sysctl -w lustre.fail_loc=0
1657 # check for log message
1658 $LCTL dk | grep "Early reply #" || error "No early reply"
1660 # client should show REQ_DELAY estimates
1661 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1663 run_test 65b "AT: verify early replies on packed reply / bulk"
1665 test_66a() #bug 3055
1667 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1669 at_start || return 0
1670 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1671 # adjust 5s at a time so no early reply is sent (within deadline)
1672 do_facet mds "sysctl -w lustre.fail_val=5000"
1673 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1674 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1675 createmany -o $DIR/$tfile 20 > /dev/null
1676 unlinkmany $DIR/$tfile 20 > /dev/null
1677 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1678 do_facet mds "sysctl -w lustre.fail_val=10000"
1679 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1680 createmany -o $DIR/$tfile 20 > /dev/null
1681 unlinkmany $DIR/$tfile 20 > /dev/null
1682 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1683 do_facet mds "sysctl -w lustre.fail_loc=0"
1685 createmany -o $DIR/$tfile 20 > /dev/null
1686 unlinkmany $DIR/$tfile 20 > /dev/null
1687 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1688 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1689 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1690 echo "Current MDT timeout $CUR, worst $WORST"
1691 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1693 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1695 test_66b() #bug 3055
1697 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1699 at_start || return 0
1700 ORIG=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1701 sysctl -w lustre.fail_val=$(($ORIG + 5))
1702 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1703 sysctl -w lustre.fail_loc=0x50c
1704 ls $DIR/$tfile > /dev/null 2>&1
1705 sysctl -w lustre.fail_loc=0
1706 CUR=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1707 WORST=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $6}')
1708 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1709 [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
1711 run_test 66b "AT: verify net latency adjusts"
1713 test_67a() #bug 3055
1715 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1717 at_start || return 0
1718 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1719 # sleeping threads may drive values above this
1720 do_facet ost1 "sysctl -w lustre.fail_val=400"
1721 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1722 do_facet ost1 "sysctl -w lustre.fail_loc=0x50a"
1723 createmany -o $DIR/$tfile 20 > /dev/null
1724 unlinkmany $DIR/$tfile 20 > /dev/null
1725 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1726 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1727 ATTEMPTS=$(($CONN2 - $CONN1))
1728 echo "$ATTEMPTS osc reconnect attempts on gradual slow"
1729 [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
1732 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1734 test_67b() #bug 3055
1736 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1738 at_start || return 0
1739 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1740 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1741 do_facet ost1 "sysctl -w lustre.fail_val=20000"
1742 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1743 cp /etc/profile $DIR/$tfile || error "cp failed"
1745 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1747 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1748 ATTEMPTS=$(($CONN2 - $CONN1))
1749 echo "$ATTEMPTS osc reconnect attempts on instant slow"
1750 # do it again; should not timeout
1751 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1752 cp /etc/profile $DIR/$tfile || error "cp failed"
1753 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1755 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1756 CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1757 ATTEMPTS=$(($CONN3 - $CONN2))
1758 echo "$ATTEMPTS osc reconnect attempts on 2nd slow"
1759 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1762 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1764 test_68 () #bug 13813
1766 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1768 at_start || return 0
1769 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1770 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1771 local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
1772 [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
1773 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1774 local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
1775 echo $TIMEOUT >> $ldlm_enqueue_min
1776 do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
1780 lfs setstripe $DIR/$tdir --index=0 --count=1
1781 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1782 sysctl -w lustre.fail_val=$(($TIMEOUT - 1))
1783 sysctl -w lustre.fail_loc=0x80000312
1784 cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
1785 sysctl -w lustre.fail_val=$((TIMEOUT * 5 / 4))
1786 sysctl -w lustre.fail_loc=0x80000312
1787 cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
1788 sysctl -w lustre.fail_loc=0
1790 echo $ENQ_MIN >> $ldlm_enqueue_min
1791 do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
1795 run_test 68 "AT: verify slowing locks"
1798 # end of AT tests includes above lines
1801 # start multi-client tests
1803 [ -z "$CLIENTS" ] && \
1804 { skip "Need two or more clients." && return; }
1805 [ $CLIENTCOUNT -lt 2 ] && \
1806 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1808 echo "mount clients $CLIENTS ..."
1809 zconf_mount_clients $CLIENTS $DIR
1811 local clients=${CLIENTS//,/ }
1812 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1813 for CLIENT in $clients; do
1814 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1815 of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
1816 error "dd failed on $CLIENT"
1819 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
1820 for C in ${CLIENTS//,/ }; do
1821 do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
1822 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
1828 run_test 70a "check multi client t-f"
1831 local clients=${CLIENTS:-$HOSTNAME}
1833 zconf_mount_clients $clients $DIR
1836 [ "$SLOW" = "no" ] && duration=60
1837 local cmd="rundbench 1 -t $duration"
1839 do_nodes $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
1840 PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
1841 DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
1844 log "Started rundbench load PID=$PID ..."
1847 START_TS=$(date +%s)
1848 CURRENT_TS=$START_TS
1849 while [ $ELAPSED -lt $duration ]; do
1851 replay_barrier $SINGLEMDS
1852 sleep 1 # give clients a time to do operations
1853 # Increment the number of failovers
1854 NUM_FAILOVERS=$((NUM_FAILOVERS+1))
1855 log "$TESTNAME fail mds1 $NUM_FAILOVERS times"
1856 facet_failover $SINGLEMDS
1857 CURRENT_TS=$(date +%s)
1858 ELAPSED=$((CURRENT_TS - START_TS))
1860 wait $PID || error "rundbench load on $CLIENTS failed!"
1862 run_test 70b "mds recovery; $CLIENTCOUNT clients"
1863 # end multi-client tests
1866 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1870 replay_barrier $SINGLEMDS
1871 #define OBD_FAIL_LDLM_ENQUEUE 0x302
1872 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
1875 wait $pid || return 1
1876 [ -e $DIR/$tfile ] && return 2
1879 run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay , close"
1882 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1886 replay_barrier $SINGLEMDS
1887 #define OBD_FAIL_LDLM_REPLY 0x30c
1888 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1891 wait $pid || return 1
1892 [ -e $DIR/$tfile ] && return 2
1895 run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
1898 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1902 replay_barrier $SINGLEMDS
1903 #define OBD_FAIL_TGT_LAST_REPLAY 0x710
1904 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000710"
1907 wait $pid || return 1
1908 [ -e $DIR/$tfile ] && return 2
1911 run_test 73c "open(O_CREAT), unlink, replay, reconnect at last_replay, close"
1916 zconf_umount $(hostname) $MOUNT
1918 zconf_mount $(hostname) $MOUNT
1920 touch $DIR/$tfile || return 1
1921 rm $DIR/$tfile || return 2
1922 df $MOUNT || error "df failed: $?"
1925 run_test 74 "Ensure applications don't fail waiting for OST reocvery"
1928 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1932 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
1933 rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
1935 stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
1938 run_test 80a "CMD: unlink cross-node dir (fail mds with inode)"
1941 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1945 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
1946 rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
1948 stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
1951 run_test 80b "CMD: unlink cross-node dir (fail mds with name)"
1954 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1957 createmany -o $DIR/$tdir/f 3000 || error "createmany failed"
1959 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir failed"
1960 $CHECKSTAT -t file $DIR/$tdir/f1002 || error "$CHECKSTAT -t file failed"
1962 rm $DIR/$tdir/f1002 || error "rm $DIR/$tdir/f1002 failed"
1964 stat $DIR/$tdir/f1002
1966 run_test 81a "CMD: unlink cross-node file (fail mds with name)"
1969 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1973 mkdir $dir || error "mkdir $dir failed"
1977 $CHECKSTAT -t dir $dir || error "$CHECKSTAT -t dir $dir failed"
1979 run_test 82a "CMD: mkdir cross-node dir (fail mds with inode)"
1982 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1986 mkdir $dir || error "mkdir $dir failed"
1990 $CHECKSTAT -t dir $dir || error "$CHECKSTAT -t dir $dir failed"
1992 run_test 82b "CMD: mkdir cross-node dir (fail mds with name)"
1995 #define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x143
1996 do_facet mds "lctl set_param fail_loc=0x80000143"
1997 createmany -o $DIR/$tfile- 1 &
2001 df -P $DIR || df -P $DIR || true # reconnect
2003 run_test 84 "stale open during export disconnect"
2005 equals_msg `basename $0`: test complete, cleaning up
2006 check_and_cleanup_lustre
2007 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true