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
496 # let the statfs cache to get old enough.
499 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
500 log "before $BEFOREUSED, after $AFTERUSED"
501 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
502 error "after $AFTERUSED > before $BEFOREUSED"
505 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
507 test_20c() { # bug 10480
508 multiop_bg_pause $DIR/$tfile Ow_c || return 1
515 df -P $DIR || df -P $DIR || true # reconnect
518 test -s $DIR/$tfile || error "File was truncated"
520 wait $pid || return 1
523 run_test 20c "check that client eviction does not affect file content"
526 replay_barrier $SINGLEMDS
527 multiop_bg_pause $DIR/$tfile O_tSc || return 5
530 touch $DIR/g11 || return 1
534 wait $pid || return 2
535 [ -e $DIR/$tfile ] && return 3
536 touch $DIR/h11 || return 4
539 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
542 multiop_bg_pause $DIR/$tfile O_tSc || return 3
545 replay_barrier $SINGLEMDS
550 wait $pid || return 1
551 [ -e $DIR/$tfile ] && return 2
554 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
557 multiop_bg_pause $DIR/$tfile O_tSc || return 5
560 replay_barrier $SINGLEMDS
562 touch $DIR/g11 || return 1
566 wait $pid || return 2
567 [ -e $DIR/$tfile ] && return 3
568 touch $DIR/h11 || return 4
571 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
574 multiop_bg_pause $DIR/$tfile O_tSc || return 3
577 replay_barrier $SINGLEMDS
581 wait $pid || return 1
582 [ -e $DIR/$tfile ] && return 2
585 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
588 multiop_bg_pause $DIR/$tfile O_tSc || return 3
592 replay_barrier $SINGLEMDS
595 wait $pid || return 1
596 [ -e $DIR/$tfile ] && return 2
599 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
602 replay_barrier $SINGLEMDS
603 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
605 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
610 wait $pid2 || return 1
614 wait $pid1 || return 2
615 [ -e $DIR/$tfile-1 ] && return 3
616 [ -e $DIR/$tfile-2 ] && return 4
619 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
622 replay_barrier $SINGLEMDS
623 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
625 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
632 wait $pid1 || return 1
634 wait $pid2 || return 2
635 [ -e $DIR/$tfile-1 ] && return 3
636 [ -e $DIR/$tfile-2 ] && return 4
639 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
642 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
644 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
646 replay_barrier $SINGLEMDS
650 wait $pid2 || return 1
654 wait $pid1 || return 2
655 [ -e $DIR/$tfile-1 ] && return 3
656 [ -e $DIR/$tfile-2 ] && return 4
659 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
662 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
664 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
666 replay_barrier $SINGLEMDS
672 wait $pid1 || return 1
674 wait $pid2 || return 2
675 [ -e $DIR/$tfile-1 ] && return 3
676 [ -e $DIR/$tfile-2 ] && return 4
679 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
682 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
684 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
689 replay_barrier $SINGLEMDS
692 wait $pid1 || return 1
694 wait $pid2 || return 2
695 [ -e $DIR/$tfile-1 ] && return 3
696 [ -e $DIR/$tfile-2 ] && return 4
699 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
702 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
704 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
708 replay_barrier $SINGLEMDS
712 wait $pid1 || return 1
714 wait $pid2 || return 2
715 [ -e $DIR/$tfile-1 ] && return 3
716 [ -e $DIR/$tfile-2 ] && return 4
719 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
721 # tests for bug 2104; completion without crashing is success. The close is
722 # stale, but we always return 0 for close, so the app never sees it.
724 multiop_bg_pause $DIR/$tfile O_c || return 2
726 multiop_bg_pause $DIR/$tfile O_c || return 3
729 df $MOUNT || sleep 1 && df $MOUNT || return 1
732 wait $pid1 || return 4
733 wait $pid2 || return 5
736 run_test 32 "close() notices client eviction; close() after client eviction"
738 # Abort recovery before client complete
739 test_33a() { # was test_33
740 replay_barrier $SINGLEMDS
741 createmany -o $DIR/$tfile-%d 100
742 fail_abort $SINGLEMDS
743 # this file should be gone, because the replay was aborted
744 $CHECKSTAT -t file $DIR/$tfile-* && return 3
745 unlinkmany $DIR/$tfile-%d 0 100
748 run_test 33a "abort recovery before client does replay"
750 # Stale FID sequence bug 15962
751 test_33b() { # was test_33a
752 replay_barrier $SINGLEMDS
753 createmany -o $DIR/$tfile-%d 10
754 fail_abort $SINGLEMDS
755 unlinkmany $DIR/$tfile-%d 0 10
756 # recreate shouldn't fail
757 createmany -o $DIR/$tfile-%d 10 || return 3
758 unlinkmany $DIR/$tfile-%d 0 10
761 run_test 33b "fid shouldn't be reused after abort recovery"
764 multiop_bg_pause $DIR/$tfile O_c || return 2
768 replay_barrier $SINGLEMDS
769 fail_abort $SINGLEMDS
771 wait $pid || return 3
772 [ -e $DIR/$tfile ] && return 1
776 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
778 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
782 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
783 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
788 # give a chance to remove from MDS
789 fail_abort $SINGLEMDS
790 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
792 run_test 35 "test recovery from llog for unlink op"
794 # b=2432 resent cancel after replay uses wrong cookie,
795 # so don't resend cancels
797 replay_barrier $SINGLEMDS
799 checkstat $DIR/$tfile
800 facet_failover $SINGLEMDS
802 if dmesg | grep "unknown lock cookie"; then
803 echo "cancel after replay failed"
807 run_test 36 "don't resend cancel"
810 # directory orphans can't be unlinked from PENDING directory
812 rmdir $DIR/$tfile 2>/dev/null
813 multiop_bg_pause $DIR/$tfile dD_c || return 2
817 replay_barrier $SINGLEMDS
818 # clear the dmesg buffer so we only see errors from this recovery
820 fail_abort $SINGLEMDS
822 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
823 wait $pid || return 3
827 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
830 createmany -o $DIR/$tfile-%d 800
831 unlinkmany $DIR/$tfile-%d 0 400
832 replay_barrier $SINGLEMDS
834 unlinkmany $DIR/$tfile-%d 400 400
836 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
838 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
840 test_39() { # bug 4176
841 createmany -o $DIR/$tfile-%d 800
842 replay_barrier $SINGLEMDS
843 unlinkmany $DIR/$tfile-%d 0 400
845 unlinkmany $DIR/$tfile-%d 400 400
847 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
849 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
852 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
857 $LCTL mark multiop $MOUNT/$tfile OS_c
858 multiop $MOUNT/$tfile OS_c &
860 writeme -s $MOUNT/${tfile}-2 &
863 facet_failover $SINGLEMDS
864 #define OBD_FAIL_MDS_CONNECT_NET 0x117
865 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
867 stat1=`count_ost_writes`
869 stat2=`count_ost_writes`
870 echo "$stat1, $stat2"
871 if [ $stat1 -lt $stat2 ]; then
872 echo "writes continuing during recovery"
875 echo "writes not continuing during recovery, bug 2477"
878 echo "waiting for writeme $WRITE_PID"
882 echo "waiting for multiop $PID"
883 wait $PID || return 2
884 do_facet client munlink $MOUNT/$tfile || return 3
885 do_facet client munlink $MOUNT/${tfile}-2 || return 3
888 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
892 # make sure that a read to one osc doesn't try to double-unlock its page just
893 # because another osc is invalid. trigger_group_io used to mistakenly return
894 # an error if any oscs were invalid even after having successfully put rpcs
895 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
896 # the page, guarnateeing that the unlock from the RPC completion would
897 # assert on trying to unlock the unlocked page.
899 [ $OSTCOUNT -lt 2 ] && \
900 skip "skipping test 41: we don't have a second OST to test with" && \
903 local f=$MOUNT/$tfile
904 # make sure the start of the file is ost1
905 lfs setstripe $f -s $((128 * 1024)) -i 0
906 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
908 # fail ost2 and read from ost1
909 local osc2dev=`do_facet $SINGLEMDS "lctl get_param -n devices | grep ${ost2_svc}-osc-MDT0000" | awk '{print $1}'`
910 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices && return 4
911 do_facet $SINGLEMDS $LCTL --device $osc2dev deactivate || return 1
912 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
913 do_facet $SINGLEMDS $LCTL --device $osc2dev activate || return 2
916 run_test 41 "read from a valid osc while other oscs are invalid"
918 # test MDS recovery after ost failure
920 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
921 createmany -o $DIR/$tfile-%d 800
923 unlinkmany $DIR/$tfile-%d 0 400
925 lctl set_param debug=-1
928 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
929 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
930 #[ $blocks_after -lt $blocks ] || return 1
931 echo wait for MDS to timeout and recover
932 sleep $((TIMEOUT * 2))
934 unlinkmany $DIR/$tfile-%d 400 400
935 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
937 run_test 42 "recovery after ost failure"
939 # timeout in MDS/OST recovery RPC will LBUG MDS
940 test_43() { # bug 2530
941 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
943 replay_barrier $SINGLEMDS
945 # OBD_FAIL_OST_CREATE_NET 0x204
946 do_facet ost1 "lctl set_param fail_loc=0x80000204"
949 do_facet ost1 "lctl set_param fail_loc=0"
953 run_test 43 "mds osc import failure during recovery; don't LBUG"
955 test_44a() { # was test_44
958 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
959 [ "$mdcdev" ] || exit 2
961 # adaptive timeouts slow this way down
962 if at_is_enabled; then
963 at_max_saved=$(at_max_get mds)
967 for i in `seq 1 10`; do
968 echo "$i of 10 ($(date +%s))"
969 do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
970 #define OBD_FAIL_TGT_CONN_RACE 0x701
971 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
972 $LCTL --device $mdcdev recover
975 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
976 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
979 run_test 44a "race in target handle connect"
982 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
983 [ "$mdcdev" ] || exit 2
984 for i in `seq 1 10`; do
985 echo "$i of 10 ($(date +%s))"
986 do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
987 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
988 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
989 $LCTL --device $mdcdev recover
992 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
995 run_test 44b "race in target handle connect"
997 # Handle failed close
999 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
1000 [ "$mdcdev" ] || exit 2
1001 $LCTL --device $mdcdev recover
1003 multiop_bg_pause $DIR/$tfile O_c || return 1
1006 # This will cause the CLOSE to fail before even
1007 # allocating a reply buffer
1008 $LCTL --device $mdcdev deactivate || return 4
1012 wait $pid || return 1
1014 $LCTL --device $mdcdev activate || return 5
1017 $CHECKSTAT -t file $DIR/$tfile || return 2
1020 run_test 45 "Handle failed close"
1024 drop_reply "touch $DIR/$tfile"
1026 # ironically, the previous test, 45, will cause a real forced close,
1027 # so just look for one for this test
1028 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1031 run_test 46 "Don't leak file handle after open resend (3325)"
1033 test_47() { # bug 2824
1034 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1036 # create some files to make sure precreate has been done on all
1037 # OSTs. (just in case this test is run independently)
1038 createmany -o $DIR/$tfile 20 || return 1
1040 # OBD_FAIL_OST_CREATE_NET 0x204
1042 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1043 df $MOUNT || return 2
1045 # let the MDS discover the OST failure, attempt to recover, fail
1046 # and recover again.
1047 sleep $((3 * TIMEOUT))
1049 # Without 2824, this createmany would hang
1050 createmany -o $DIR/$tfile 20 || return 3
1051 unlinkmany $DIR/$tfile 20 || return 4
1053 do_facet ost1 "lctl set_param fail_loc=0"
1056 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1059 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1060 [ "$OSTCOUNT" -lt "2" ] && skip "$OSTCOUNT < 2 OSTs -- skipping" && return
1062 replay_barrier $SINGLEMDS
1063 createmany -o $DIR/$tfile 20 || return 1
1064 # OBD_FAIL_OST_EROFS 0x216
1065 facet_failover $SINGLEMDS
1066 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1067 df $MOUNT || return 2
1069 createmany -o $DIR/$tfile 20 20 || return 2
1070 unlinkmany $DIR/$tfile 40 || return 3
1073 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1076 local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
1077 [ "$oscdev" ] || return 1
1078 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1079 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1080 # give the mds_lov_sync threads a chance to run
1083 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1085 # b3764 timed out lock replay
1088 cancel_lru_locks mdc
1090 multiop $DIR/$tfile s || return 1
1091 replay_barrier $SINGLEMDS
1092 #define OBD_FAIL_LDLM_REPLY 0x30c
1093 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1094 fail $SINGLEMDS || return 2
1095 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1097 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1099 run_test 52 "time out lock replay (3764)"
1101 # bug 3462 - simultaneous MDC requests
1103 mkdir -p $DIR/${tdir}-1
1104 mkdir -p $DIR/${tdir}-2
1105 multiop $DIR/${tdir}-1/f O_c &
1107 # give multiop a change to open
1110 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1111 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1112 kill -USR1 $close_pid
1113 cancel_lru_locks mdc # force the close
1114 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1116 mcreate $DIR/${tdir}-2/f || return 1
1118 # close should still be here
1119 [ -d /proc/$close_pid ] || return 2
1121 replay_barrier_nodf $SINGLEMDS
1123 wait $close_pid || return 3
1125 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1126 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1127 rm -rf $DIR/${tdir}-*
1129 run_test 53a "|X| close request while two MDC requests in flight"
1132 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1134 mkdir -p $DIR/${tdir}-1
1135 mkdir -p $DIR/${tdir}-2
1136 multiop $DIR/${tdir}-1/f O_c &
1139 #define OBD_FAIL_MDS_REINT_NET 0x107
1140 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1141 mcreate $DIR/${tdir}-2/f &
1145 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1146 kill -USR1 $close_pid
1147 cancel_lru_locks mdc # force the close
1148 wait $close_pid || return 1
1149 # open should still be here
1150 [ -d /proc/$open_pid ] || return 2
1152 replay_barrier_nodf $SINGLEMDS
1154 wait $open_pid || return 3
1156 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1157 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1158 rm -rf $DIR/${tdir}-*
1160 run_test 53b "|X| open request while two MDC requests in flight"
1163 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1165 mkdir -p $DIR/${tdir}-1
1166 mkdir -p $DIR/${tdir}-2
1167 multiop $DIR/${tdir}-1/f O_c &
1170 #define OBD_FAIL_MDS_REINT_NET 0x107
1171 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1172 mcreate $DIR/${tdir}-2/f &
1176 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1177 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1178 kill -USR1 $close_pid
1179 cancel_lru_locks mdc # force the close
1181 replay_barrier_nodf $SINGLEMDS
1182 fail_nodf $SINGLEMDS
1183 wait $open_pid || return 1
1185 # close should be gone
1186 [ -d /proc/$close_pid ] && return 2
1187 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1189 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1190 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1191 rm -rf $DIR/${tdir}-*
1193 run_test 53c "|X| open request and close request while two MDC requests in flight"
1196 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1198 mkdir -p $DIR/${tdir}-1
1199 mkdir -p $DIR/${tdir}-2
1200 multiop $DIR/${tdir}-1/f O_c &
1202 # give multiop a chance to open
1205 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1206 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1207 kill -USR1 $close_pid
1208 cancel_lru_locks mdc # force the close
1209 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1210 mcreate $DIR/${tdir}-2/f || return 1
1212 # close should still be here
1213 [ -d /proc/$close_pid ] || return 2
1215 wait $close_pid || return 3
1217 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1218 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1219 rm -rf $DIR/${tdir}-*
1221 run_test 53d "|X| close reply while two MDC requests in flight"
1224 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1226 mkdir -p $DIR/${tdir}-1
1227 mkdir -p $DIR/${tdir}-2
1228 multiop $DIR/${tdir}-1/f O_c &
1231 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1232 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1233 mcreate $DIR/${tdir}-2/f &
1237 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1238 kill -USR1 $close_pid
1239 cancel_lru_locks mdc # force the close
1240 wait $close_pid || return 1
1241 # open should still be here
1242 [ -d /proc/$open_pid ] || return 2
1244 replay_barrier_nodf $SINGLEMDS
1246 wait $open_pid || return 3
1248 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1249 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1250 rm -rf $DIR/${tdir}-*
1252 run_test 53e "|X| open reply while two MDC requests in flight"
1255 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1257 mkdir -p $DIR/${tdir}-1
1258 mkdir -p $DIR/${tdir}-2
1259 multiop $DIR/${tdir}-1/f O_c &
1262 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1263 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1264 mcreate $DIR/${tdir}-2/f &
1268 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1269 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1270 kill -USR1 $close_pid
1271 cancel_lru_locks mdc # force the close
1273 replay_barrier_nodf $SINGLEMDS
1274 fail_nodf $SINGLEMDS
1275 wait $open_pid || return 1
1277 # close should be gone
1278 [ -d /proc/$close_pid ] && return 2
1279 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1281 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1282 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1283 rm -rf $DIR/${tdir}-*
1285 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1288 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1290 mkdir -p $DIR/${tdir}-1
1291 mkdir -p $DIR/${tdir}-2
1292 multiop $DIR/${tdir}-1/f O_c &
1295 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1296 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1297 mcreate $DIR/${tdir}-2/f &
1301 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1302 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1303 kill -USR1 $close_pid
1304 cancel_lru_locks mdc # force the close
1306 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1307 replay_barrier_nodf $SINGLEMDS
1308 fail_nodf $SINGLEMDS
1309 wait $open_pid || return 1
1311 # close should be gone
1312 [ -d /proc/$close_pid ] && return 2
1314 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1315 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1316 rm -rf $DIR/${tdir}-*
1318 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1321 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1323 mkdir -p $DIR/${tdir}-1
1324 mkdir -p $DIR/${tdir}-2
1325 multiop $DIR/${tdir}-1/f O_c &
1328 #define OBD_FAIL_MDS_REINT_NET 0x107
1329 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1330 mcreate $DIR/${tdir}-2/f &
1334 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1335 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1336 kill -USR1 $close_pid
1337 cancel_lru_locks mdc # force the close
1340 replay_barrier_nodf $SINGLEMDS
1341 fail_nodf $SINGLEMDS
1342 wait $open_pid || return 1
1344 # close should be gone
1345 [ -d /proc/$close_pid ] && return 2
1346 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1348 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1349 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1350 rm -rf $DIR/${tdir}-*
1352 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1354 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1356 #b3761 ASSERTION(hash != 0) failed
1358 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1359 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1361 # give touch a chance to run
1363 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1367 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1369 #b3440 ASSERTION(rec->ur_fid2->id) failed
1371 ln -s foo $DIR/$tfile
1372 replay_barrier $SINGLEMDS
1373 #drop_reply "cat $DIR/$tfile"
1377 run_test 56 "don't replay a symlink open request (3440)"
1379 #recovery one mds-ost setattr from llog
1381 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1382 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1384 replay_barrier $SINGLEMDS
1387 $CHECKSTAT -t file $DIR/$tfile || return 1
1388 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1391 run_test 57 "test recovery from llog for setattr op"
1393 #recovery many mds-ost setattr from llog
1396 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1397 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1398 createmany -o $DIR/$tdir/$tfile-%d 2500
1399 replay_barrier $SINGLEMDS
1402 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1403 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1404 unlinkmany $DIR/$tdir/$tfile-%d 2500
1407 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
1410 mount_client $MOUNT2
1412 touch $DIR/$tdir/$tfile
1413 replay_barrier $SINGLEMDS
1414 setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile
1416 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1417 [ x$VAL = x"bar" ] || return 1
1418 rm -f $DIR/$tdir/$tfile
1420 zconf_umount `hostname` $MOUNT2
1422 run_test 58b "test replay of setxattr op"
1424 test_58c() { # bug 16570
1425 mount_client $MOUNT2
1427 touch $DIR/$tdir/$tfile
1428 drop_request "setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile" || \
1430 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1431 [ x$VAL = x"bar" ] || return 2
1432 drop_reint_reply "setfattr -n trusted.foo1 -v bar1 $DIR/$tdir/$tfile" || \
1434 VAL=`getfattr --absolute-names --only-value -n trusted.foo1 $MOUNT2/$tdir/$tfile`
1435 [ x$VAL = x"bar1" ] || return 4
1436 rm -f $DIR/$tdir/$tfile
1438 zconf_umount `hostname` $MOUNT2
1440 run_test 58c "resend/reconstruct setxattr op"
1442 # log_commit_thread vs filter_destroy race used to lead to import use after free
1445 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1448 createmany -o $DIR/$tdir/$tfile-%d 200
1450 unlinkmany $DIR/$tdir/$tfile-%d 200
1451 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1452 do_facet ost1 "lctl set_param fail_loc=0x507"
1455 do_facet ost1 "lctl set_param fail_loc=0x0"
1459 run_test 59 "test log_commit_thread vs filter_destroy race"
1461 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1462 # bug 12086: should no oops and No ctxt error for this test
1465 createmany -o $DIR/$tdir/$tfile-%d 200
1466 replay_barrier $SINGLEMDS
1467 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1469 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1470 local no_ctxt=`dmesg | grep "No ctxt"`
1471 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1473 run_test 60 "test llog post recovery init vs llog unlink"
1475 #test race llog recovery thread vs llog cleanup
1476 test_61a() { # was test_61
1477 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1480 createmany -o $DIR/$tdir/$tfile-%d 800
1482 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1483 unlinkmany $DIR/$tdir/$tfile-%d 800
1484 set_nodes_failloc "$(osts_nodes)" 0x80000221
1489 set_nodes_failloc "$(osts_nodes)" 0x0
1491 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1494 run_test 61a "test race llog recovery vs llog cleanup"
1496 #test race mds llog sync vs llog cleanup
1498 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1499 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
1500 facet_failover $SINGLEMDS
1503 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1505 run_test 61b "test race mds llog sync vs llog cleanup"
1507 #test race cancel cookie cb vs llog cleanup
1509 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1511 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1513 set_nodes_failloc "$(osts_nodes)" 0x80000222
1517 set_nodes_failloc "$(osts_nodes)" 0x0
1519 run_test 61c "test race mds llog sync vs llog cleanup"
1521 test_61d() { # bug 16002 # bug 17466
1522 shutdown_facet $SINGLEMDS
1523 #define OBD_FAIL_OBD_LLOG_SETUP 0x605
1524 do_facet $SINGLEMDS "lctl set_param fail_loc=0x605"
1525 start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS && error "mds start should have failed"
1526 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1527 start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS || error "cannot restart mds"
1529 run_test 61d "error in llog_setup should cleanup the llog context correctly"
1531 test_62() { # Bug 15756 - don't mis-drop resent replay
1533 replay_barrier $SINGLEMDS
1534 createmany -o $DIR/$tdir/$tfile- 25
1535 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1536 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1537 facet_failover $SINGLEMDS
1538 df $MOUNT || return 1
1539 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1540 unlinkmany $DIR/$tdir/$tfile- 25 || return 2
1543 run_test 62 "don't mis-drop resent replay"
1545 #Adaptive Timeouts (bug 3055)
1553 echo "Cleaning up AT ..."
1554 if [ -n "$ATOLDBASE" ]; then
1555 local at_history=$($LCTL get_param -n at_history)
1556 do_facet mds "lctl set_param at_history=$at_history" || true
1557 do_facet ost1 "lctl set_param at_history=$at_history" || true
1560 if [ $AT_MAX_SET -ne 0 ]; then
1561 for facet in mds client ost; do
1562 var=AT_MAX_SAVE_${facet}
1563 echo restore AT on $facet to saved value ${!var}
1564 at_max_set ${!var} $facet
1565 at_new=$(at_max_get $facet)
1566 echo Restored AT value on $facet $at_new
1567 [ $at_new -eq ${!var} ] || \
1568 error "$facet : AT value was not restored SAVED ${!var} NEW $at_new"
1575 local at_max_new=600
1577 # Save at_max original values
1579 if [ $AT_MAX_SET -eq 0 ]; then
1580 # Suppose that all osts have the same at_max
1581 for facet in mds client ost; do
1582 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1586 for facet in mds client ost; do
1587 at_max=$(at_max_get $facet)
1588 if [ $at_max -ne $at_max_new ]; then
1589 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1590 at_max_set $at_max_new $facet
1595 if [ -z "$ATOLDBASE" ]; then
1596 ATOLDBASE=$(do_facet mds "lctl get_param -n at_history")
1597 # speed up the timebase so we can check decreasing AT
1598 do_facet mds "lctl set_param at_history=8" || true
1599 do_facet ost1 "lctl set_param at_history=8" || true
1601 # sleep for a while to cool down, should be > 8s and also allow
1602 # at least one ping to be sent. simply use TIMEOUT to be safe.
1607 test_65a() #bug 3055
1609 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1611 at_start || return 0
1612 $LCTL dk > /dev/null
1614 sysctl -w lnet.debug="+other"
1615 # Slow down a request to the current service time, this is critical
1616 # because previous tests may have caused this value to increase.
1617 REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
1618 awk '/portal 12/ {print $5}'`
1619 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1621 do_facet mds lctl set_param fail_val=$((${REQ_DELAY} * 1000))
1622 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1623 do_facet mds sysctl -w lustre.fail_loc=0x8000050a
1624 createmany -o $DIR/$tfile 10 > /dev/null
1625 unlinkmany $DIR/$tfile 10 > /dev/null
1626 # check for log message
1627 $LCTL dk | grep "Early reply #" || error "No early reply"
1629 # client should show REQ_DELAY estimates
1630 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1632 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1634 run_test 65a "AT: verify early replies"
1636 test_65b() #bug 3055
1638 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1640 at_start || return 0
1643 sysctl -w lnet.debug="other trace"
1644 $LCTL dk > /dev/null
1645 # Slow down a request to the current service time, this is critical
1646 # because previous tests may have caused this value to increase.
1647 REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
1648 awk '/portal 6/ {print $5}'`
1649 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1651 do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
1652 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1653 do_facet ost1 sysctl -w lustre.fail_loc=0x224
1656 lfs setstripe $DIR/$tfile --index=0 --count=1
1657 # force some real bulk transfer
1658 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1660 do_facet ost1 sysctl -w lustre.fail_loc=0
1661 # check for log message
1662 $LCTL dk | grep "Early reply #" || error "No early reply"
1664 # client should show REQ_DELAY estimates
1665 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1667 run_test 65b "AT: verify early replies on packed reply / bulk"
1669 test_66a() #bug 3055
1671 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1673 at_start || return 0
1674 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1675 # adjust 5s at a time so no early reply is sent (within deadline)
1676 do_facet mds "sysctl -w lustre.fail_val=5000"
1677 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1678 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1679 createmany -o $DIR/$tfile 20 > /dev/null
1680 unlinkmany $DIR/$tfile 20 > /dev/null
1681 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1682 do_facet mds "sysctl -w lustre.fail_val=10000"
1683 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1684 createmany -o $DIR/$tfile 20 > /dev/null
1685 unlinkmany $DIR/$tfile 20 > /dev/null
1686 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1687 do_facet mds "sysctl -w lustre.fail_loc=0"
1689 createmany -o $DIR/$tfile 20 > /dev/null
1690 unlinkmany $DIR/$tfile 20 > /dev/null
1691 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1692 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1693 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1694 echo "Current MDT timeout $CUR, worst $WORST"
1695 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1697 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1699 test_66b() #bug 3055
1701 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1703 at_start || return 0
1704 ORIG=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1705 sysctl -w lustre.fail_val=$(($ORIG + 5))
1706 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1707 sysctl -w lustre.fail_loc=0x50c
1708 ls $DIR/$tfile > /dev/null 2>&1
1709 sysctl -w lustre.fail_loc=0
1710 CUR=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1711 WORST=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $6}')
1712 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1713 [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
1715 run_test 66b "AT: verify net latency adjusts"
1717 test_67a() #bug 3055
1719 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1721 at_start || return 0
1722 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1723 # sleeping threads may drive values above this
1724 do_facet ost1 "sysctl -w lustre.fail_val=400"
1725 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1726 do_facet ost1 "sysctl -w lustre.fail_loc=0x50a"
1727 createmany -o $DIR/$tfile 20 > /dev/null
1728 unlinkmany $DIR/$tfile 20 > /dev/null
1729 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1730 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1731 ATTEMPTS=$(($CONN2 - $CONN1))
1732 echo "$ATTEMPTS osc reconnect attempts on gradual slow"
1733 [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
1736 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1738 test_67b() #bug 3055
1740 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1742 at_start || return 0
1743 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1744 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1745 do_facet ost1 "sysctl -w lustre.fail_val=20000"
1746 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1747 cp /etc/profile $DIR/$tfile || error "cp failed"
1749 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1751 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1752 ATTEMPTS=$(($CONN2 - $CONN1))
1753 echo "$ATTEMPTS osc reconnect attempts on instant slow"
1754 # do it again; should not timeout
1755 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1756 cp /etc/profile $DIR/$tfile || error "cp failed"
1757 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1759 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1760 CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1761 ATTEMPTS=$(($CONN3 - $CONN2))
1762 echo "$ATTEMPTS osc reconnect attempts on 2nd slow"
1763 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1766 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1768 test_68 () #bug 13813
1770 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1772 at_start || return 0
1773 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1774 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1775 local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
1776 [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
1777 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1778 local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
1779 echo $TIMEOUT >> $ldlm_enqueue_min
1780 do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
1784 lfs setstripe $DIR/$tdir --index=0 --count=1
1785 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1786 sysctl -w lustre.fail_val=$(($TIMEOUT - 1))
1787 sysctl -w lustre.fail_loc=0x80000312
1788 cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
1789 sysctl -w lustre.fail_val=$((TIMEOUT * 5 / 4))
1790 sysctl -w lustre.fail_loc=0x80000312
1791 cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
1792 sysctl -w lustre.fail_loc=0
1794 echo $ENQ_MIN >> $ldlm_enqueue_min
1795 do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
1799 run_test 68 "AT: verify slowing locks"
1802 # end of AT tests includes above lines
1805 # start multi-client tests
1807 [ -z "$CLIENTS" ] && \
1808 { skip "Need two or more clients." && return; }
1809 [ $CLIENTCOUNT -lt 2 ] && \
1810 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1812 echo "mount clients $CLIENTS ..."
1813 zconf_mount_clients $CLIENTS $DIR
1815 local clients=${CLIENTS//,/ }
1816 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1817 for CLIENT in $clients; do
1818 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1819 of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
1820 error "dd failed on $CLIENT"
1823 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
1824 for C in ${CLIENTS//,/ }; do
1825 do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
1826 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
1832 run_test 70a "check multi client t-f"
1835 local clients=${CLIENTS:-$HOSTNAME}
1837 zconf_mount_clients $clients $DIR
1840 [ "$SLOW" = "no" ] && duration=60
1841 local cmd="rundbench 1 -t $duration"
1843 do_nodes $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
1844 PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
1845 DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
1848 log "Started rundbench load PID=$PID ..."
1851 START_TS=$(date +%s)
1852 CURRENT_TS=$START_TS
1853 while [ $ELAPSED -lt $duration ]; do
1855 replay_barrier $SINGLEMDS
1856 sleep 1 # give clients a time to do operations
1857 # Increment the number of failovers
1858 NUM_FAILOVERS=$((NUM_FAILOVERS+1))
1859 log "$TESTNAME fail mds1 $NUM_FAILOVERS times"
1861 CURRENT_TS=$(date +%s)
1862 ELAPSED=$((CURRENT_TS - START_TS))
1864 wait $PID || error "rundbench load on $CLIENTS failed!"
1866 run_test 70b "mds recovery; $CLIENTCOUNT clients"
1867 # end multi-client tests
1870 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1874 replay_barrier $SINGLEMDS
1875 #define OBD_FAIL_LDLM_ENQUEUE 0x302
1876 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
1879 wait $pid || return 1
1880 [ -e $DIR/$tfile ] && return 2
1883 run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay , close"
1886 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1890 replay_barrier $SINGLEMDS
1891 #define OBD_FAIL_LDLM_REPLY 0x30c
1892 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1895 wait $pid || return 1
1896 [ -e $DIR/$tfile ] && return 2
1899 run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
1902 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1906 replay_barrier $SINGLEMDS
1907 #define OBD_FAIL_TGT_LAST_REPLAY 0x710
1908 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000710"
1911 wait $pid || return 1
1912 [ -e $DIR/$tfile ] && return 2
1915 run_test 73c "open(O_CREAT), unlink, replay, reconnect at last_replay, close"
1919 local clients=${CLIENTS:-$HOSTNAME}
1922 zconf_umount_clients $clients $MOUNT
1923 facet_failover $SINGLEMDS
1924 zconf_mount_clients $clients $MOUNT
1926 touch $DIR/$tfile || return 1
1927 rm $DIR/$tfile || return 2
1928 client_df || error "df failed: $?"
1931 run_test 74 "Ensure applications don't fail waiting for OST recovery"
1934 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1938 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
1939 rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
1941 stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
1944 run_test 80a "CMD: unlink cross-node dir (fail mds with inode)"
1947 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1951 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
1952 rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
1954 stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
1957 run_test 80b "CMD: unlink cross-node dir (fail mds with name)"
1960 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1963 createmany -o $DIR/$tdir/f 3000 || error "createmany failed"
1965 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir failed"
1966 $CHECKSTAT -t file $DIR/$tdir/f1002 || error "$CHECKSTAT -t file failed"
1968 rm $DIR/$tdir/f1002 || error "rm $DIR/$tdir/f1002 failed"
1970 stat $DIR/$tdir/f1002
1972 run_test 81a "CMD: unlink cross-node file (fail mds with name)"
1975 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1979 mkdir $dir || error "mkdir $dir failed"
1983 $CHECKSTAT -t dir $dir || error "$CHECKSTAT -t dir $dir failed"
1985 run_test 82a "CMD: mkdir cross-node dir (fail mds with inode)"
1988 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1992 mkdir $dir || error "mkdir $dir failed"
1996 $CHECKSTAT -t dir $dir || error "$CHECKSTAT -t dir $dir failed"
1998 run_test 82b "CMD: mkdir cross-node dir (fail mds with name)"
2001 #define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x143
2002 do_facet mds "lctl set_param fail_loc=0x80000143"
2003 createmany -o $DIR/$tfile- 1 &
2007 df -P $DIR || df -P $DIR || true # reconnect
2009 run_test 84 "stale open during export disconnect"
2011 equals_msg `basename $0`: test complete, cleaning up
2012 check_and_cleanup_lustre
2013 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true