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 $REPLAY_SINGLE_EXCEPT"
25 if [ "$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 start_full_debug_logging
130 run_test 0c "fld create"
131 stop_full_debug_logging
134 replay_barrier $SINGLEMDS
137 $CHECKSTAT -t file $DIR/$tfile || return 1
140 run_test 1 "simple create"
143 replay_barrier $SINGLEMDS
146 $CHECKSTAT -t file $DIR/$tfile || return 1
153 replay_barrier $SINGLEMDS
156 $CHECKSTAT -t file $DIR/$tfile || return 1
162 local file=$DIR/$tfile
163 replay_barrier $SINGLEMDS
165 openfile -f O_DIRECTORY $file
167 $CHECKSTAT -t file $file || return 2
170 run_test 3a "replay failed open(O_DIRECTORY)"
173 replay_barrier $SINGLEMDS
174 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
175 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000114"
177 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
179 $CHECKSTAT -t file $DIR/$tfile && return 2
182 run_test 3b "replay failed open -ENOMEM"
185 replay_barrier $SINGLEMDS
186 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
187 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
189 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
192 $CHECKSTAT -t file $DIR/$tfile && return 2
195 run_test 3c "replay failed open -ENOMEM"
197 test_4a() { # was test_4
198 replay_barrier $SINGLEMDS
199 for i in `seq 10`; do
200 echo "tag-$i" > $DIR/$tfile-$i
203 for i in `seq 10`; do
204 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
207 run_test 4a "|x| 10 open(O_CREAT)s"
210 replay_barrier $SINGLEMDS
213 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
215 run_test 4b "|x| rm 10 files"
217 # The idea is to get past the first block of precreated files on both
218 # osts, and then replay.
220 replay_barrier $SINGLEMDS
221 for i in `seq 220`; do
222 echo "tag-$i" > $DIR/$tfile-$i
225 for i in `seq 220`; do
226 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
230 # waiting for commitment of removal
232 run_test 5 "|x| 220 open(O_CREAT)"
235 test_6a() { # was test_6
237 replay_barrier $SINGLEMDS
238 mcreate $DIR/$tdir/$tfile
240 $CHECKSTAT -t dir $DIR/$tdir || return 1
241 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
243 # waiting for log process thread
245 run_test 6a "mkdir + contained create"
249 replay_barrier $SINGLEMDS
252 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
254 run_test 6b "|X| rmdir"
258 replay_barrier $SINGLEMDS
259 mcreate $DIR/$tdir/$tfile
261 $CHECKSTAT -t dir $DIR/$tdir || return 1
262 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
265 run_test 7 "mkdir |X| contained create"
268 # make sure no side-effect from previous test.
270 replay_barrier $SINGLEMDS
271 multiop_bg_pause $DIR/$tfile mo_c || return 4
275 $CHECKSTAT -t file $DIR/$tfile || return 1
276 kill -USR1 $MULTIPID || return 2
277 wait $MULTIPID || return 3
280 run_test 8 "creat open |X| close"
283 replay_barrier $SINGLEMDS
285 local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
287 local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
289 echo " old_inum == $old_inum, new_inum == $new_inum"
290 if [ $old_inum -eq $new_inum ] ;
292 echo " old_inum and new_inum match"
294 echo "!!!! old_inum and new_inum NOT match"
299 run_test 9 "|X| create (same inum/gen)"
303 replay_barrier $SINGLEMDS
304 mv $DIR/$tfile $DIR/$tfile-2
307 $CHECKSTAT $DIR/$tfile && return 1
308 $CHECKSTAT $DIR/$tfile-2 ||return 2
312 run_test 10 "create |X| rename unlink"
316 echo "old" > $DIR/$tfile
317 mv $DIR/$tfile $DIR/$tfile-2
318 replay_barrier $SINGLEMDS
319 echo "new" > $DIR/$tfile
321 grep old $DIR/$tfile-2
323 grep new $DIR/$tfile || return 1
324 grep old $DIR/$tfile-2 || return 2
326 run_test 11 "create open write rename |X| create-old-name read"
330 multiop_bg_pause $DIR/$tfile o_tSc || return 3
333 replay_barrier $SINGLEMDS
335 wait $pid || return 1
338 [ -e $DIR/$tfile ] && return 2
341 run_test 12 "open, unlink |X| close"
344 # 1777 - replay open after committed chmod that would make
345 # a regular open a failure
348 multiop_bg_pause $DIR/$tfile O_wc || return 3
351 $CHECKSTAT -p 0 $DIR/$tfile
352 replay_barrier $SINGLEMDS
355 wait $pid || return 1
357 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
358 rm $DIR/$tfile || return 4
361 run_test 13 "open chmod 0 |x| write close"
364 multiop_bg_pause $DIR/$tfile O_tSc || return 4
367 replay_barrier $SINGLEMDS
368 kill -USR1 $pid || return 1
369 wait $pid || return 2
372 [ -e $DIR/$tfile ] && return 3
375 run_test 14 "open(O_CREAT), unlink |X| close"
378 multiop_bg_pause $DIR/$tfile O_tSc || return 5
381 replay_barrier $SINGLEMDS
382 touch $DIR/g11 || return 1
384 wait $pid || return 2
387 [ -e $DIR/$tfile ] && return 3
388 touch $DIR/h11 || return 4
391 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
395 replay_barrier $SINGLEMDS
398 mcreate $DIR/$tfile-2
400 [ -e $DIR/$tfile ] && return 1
401 [ -e $DIR/$tfile-2 ] || return 2
402 munlink $DIR/$tfile-2 || return 3
404 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
407 replay_barrier $SINGLEMDS
408 multiop_bg_pause $DIR/$tfile O_c || return 4
411 kill -USR1 $pid || return 1
412 wait $pid || return 2
413 $CHECKSTAT -t file $DIR/$tfile || return 3
416 run_test 17 "|X| open(O_CREAT), |replay| close"
419 replay_barrier $SINGLEMDS
420 multiop_bg_pause $DIR/$tfile O_tSc || return 8
423 touch $DIR/$tfile-2 || return 1
424 echo "pid: $pid will close"
426 wait $pid || return 2
429 [ -e $DIR/$tfile ] && return 3
430 [ -e $DIR/$tfile-2 ] || return 4
431 # this touch frequently fails
432 touch $DIR/$tfile-3 || return 5
433 munlink $DIR/$tfile-2 || return 6
434 munlink $DIR/$tfile-3 || return 7
437 run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
439 # bug 1855 (a simpler form of test_11 above)
441 replay_barrier $SINGLEMDS
443 echo "old" > $DIR/$tfile
444 mv $DIR/$tfile $DIR/$tfile-2
445 grep old $DIR/$tfile-2
447 grep old $DIR/$tfile-2 || return 2
449 run_test 19 "|X| mcreate, open, write, rename "
451 test_20a() { # was test_20
452 replay_barrier $SINGLEMDS
453 multiop_bg_pause $DIR/$tfile O_tSc || return 3
459 wait $pid || return 1
460 [ -e $DIR/$tfile ] && return 2
463 run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
465 test_20b() { # bug 10480
466 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
468 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
470 while [ ! -e $DIR/$tfile ] ; do
471 usleep 60 # give dd a chance to start
474 lfs getstripe $DIR/$tfile || return 1
475 rm -f $DIR/$tfile || return 2 # make it an orphan
477 df -P $DIR || df -P $DIR || true # reconnect
479 fail $SINGLEMDS # start orphan recovery
480 df -P $DIR || df -P $DIR || true # reconnect
481 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
483 # just because recovery is done doesn't mean we've finished
484 # orphan cleanup. Wait for llogs to get synchronized.
485 echo waiting for orphan cleanup...
487 local -a sync=($(do_nodes $(comma_list $(osts_nodes)) \
488 "$LCTL get_param obdfilter.*.mds_sync" | awk -F= ' {print $2}'))
490 for ((i=0; i<${#sync[@]}; i++)); do
491 [ ${sync[$i]} -eq 0 ] && continue
492 # there is a not finished MDS-OST synchronization
496 [ ${con} -eq 1 ] && break
500 # let the statfs cache to get old enough.
503 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
504 log "before $BEFOREUSED, after $AFTERUSED"
505 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
506 error "after $AFTERUSED > before $BEFOREUSED"
509 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
511 test_20c() { # bug 10480
512 multiop_bg_pause $DIR/$tfile Ow_c || return 1
519 df -P $DIR || df -P $DIR || true # reconnect
522 test -s $DIR/$tfile || error "File was truncated"
524 wait $pid || return 1
527 run_test 20c "check that client eviction does not affect file content"
530 replay_barrier $SINGLEMDS
531 multiop_bg_pause $DIR/$tfile O_tSc || return 5
534 touch $DIR/g11 || return 1
538 wait $pid || return 2
539 [ -e $DIR/$tfile ] && return 3
540 touch $DIR/h11 || return 4
543 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
546 multiop_bg_pause $DIR/$tfile O_tSc || return 3
549 replay_barrier $SINGLEMDS
554 wait $pid || return 1
555 [ -e $DIR/$tfile ] && return 2
558 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
561 multiop_bg_pause $DIR/$tfile O_tSc || return 5
564 replay_barrier $SINGLEMDS
566 touch $DIR/g11 || return 1
570 wait $pid || return 2
571 [ -e $DIR/$tfile ] && return 3
572 touch $DIR/h11 || return 4
575 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
578 multiop_bg_pause $DIR/$tfile O_tSc || return 3
581 replay_barrier $SINGLEMDS
585 wait $pid || return 1
586 [ -e $DIR/$tfile ] && return 2
589 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
592 multiop_bg_pause $DIR/$tfile O_tSc || return 3
596 replay_barrier $SINGLEMDS
599 wait $pid || return 1
600 [ -e $DIR/$tfile ] && return 2
603 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
606 replay_barrier $SINGLEMDS
607 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
609 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
614 wait $pid2 || return 1
618 wait $pid1 || return 2
619 [ -e $DIR/$tfile-1 ] && return 3
620 [ -e $DIR/$tfile-2 ] && return 4
623 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
626 replay_barrier $SINGLEMDS
627 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
629 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
636 wait $pid1 || return 1
638 wait $pid2 || return 2
639 [ -e $DIR/$tfile-1 ] && return 3
640 [ -e $DIR/$tfile-2 ] && return 4
643 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
646 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
648 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
650 replay_barrier $SINGLEMDS
654 wait $pid2 || return 1
658 wait $pid1 || return 2
659 [ -e $DIR/$tfile-1 ] && return 3
660 [ -e $DIR/$tfile-2 ] && return 4
663 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
666 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
668 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
670 replay_barrier $SINGLEMDS
676 wait $pid1 || return 1
678 wait $pid2 || return 2
679 [ -e $DIR/$tfile-1 ] && return 3
680 [ -e $DIR/$tfile-2 ] && return 4
683 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
686 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
688 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
693 replay_barrier $SINGLEMDS
696 wait $pid1 || return 1
698 wait $pid2 || return 2
699 [ -e $DIR/$tfile-1 ] && return 3
700 [ -e $DIR/$tfile-2 ] && return 4
703 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
706 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
708 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
712 replay_barrier $SINGLEMDS
716 wait $pid1 || return 1
718 wait $pid2 || return 2
719 [ -e $DIR/$tfile-1 ] && return 3
720 [ -e $DIR/$tfile-2 ] && return 4
723 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
725 # tests for bug 2104; completion without crashing is success. The close is
726 # stale, but we always return 0 for close, so the app never sees it.
728 multiop_bg_pause $DIR/$tfile O_c || return 2
730 multiop_bg_pause $DIR/$tfile O_c || return 3
733 df $MOUNT || sleep 1 && df $MOUNT || return 1
736 wait $pid1 || return 4
737 wait $pid2 || return 5
740 run_test 32 "close() notices client eviction; close() after client eviction"
742 # Abort recovery before client complete
743 test_33a() { # was test_33
744 replay_barrier $SINGLEMDS
745 createmany -o $DIR/$tfile-%d 100
746 fail_abort $SINGLEMDS
747 # this file should be gone, because the replay was aborted
748 $CHECKSTAT -t file $DIR/$tfile-* && return 3
749 unlinkmany $DIR/$tfile-%d 0 100
752 run_test 33a "abort recovery before client does replay"
754 # Stale FID sequence bug 15962
755 test_33b() { # was test_33a
756 replay_barrier $SINGLEMDS
757 createmany -o $DIR/$tfile-%d 10
758 fail_abort $SINGLEMDS
759 unlinkmany $DIR/$tfile-%d 0 10
760 # recreate shouldn't fail
761 createmany -o $DIR/$tfile-%d 10 || return 3
762 unlinkmany $DIR/$tfile-%d 0 10
765 run_test 33b "fid shouldn't be reused after abort recovery"
768 multiop_bg_pause $DIR/$tfile O_c || return 2
772 replay_barrier $SINGLEMDS
773 fail_abort $SINGLEMDS
775 wait $pid || return 3
776 [ -e $DIR/$tfile ] && return 1
780 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
782 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
786 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
787 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
792 # give a chance to remove from MDS
793 fail_abort $SINGLEMDS
794 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
796 run_test 35 "test recovery from llog for unlink op"
798 # b=2432 resent cancel after replay uses wrong cookie,
799 # so don't resend cancels
801 replay_barrier $SINGLEMDS
803 checkstat $DIR/$tfile
804 facet_failover $SINGLEMDS
806 if dmesg | grep "unknown lock cookie"; then
807 echo "cancel after replay failed"
811 run_test 36 "don't resend cancel"
814 # directory orphans can't be unlinked from PENDING directory
816 rmdir $DIR/$tfile 2>/dev/null
817 multiop_bg_pause $DIR/$tfile dD_c || return 2
821 replay_barrier $SINGLEMDS
822 # clear the dmesg buffer so we only see errors from this recovery
824 fail_abort $SINGLEMDS
826 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
827 wait $pid || return 3
831 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
834 createmany -o $DIR/$tfile-%d 800
835 unlinkmany $DIR/$tfile-%d 0 400
836 replay_barrier $SINGLEMDS
838 unlinkmany $DIR/$tfile-%d 400 400
840 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
842 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
844 test_39() { # bug 4176
845 createmany -o $DIR/$tfile-%d 800
846 replay_barrier $SINGLEMDS
847 unlinkmany $DIR/$tfile-%d 0 400
849 unlinkmany $DIR/$tfile-%d 400 400
851 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
853 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
856 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
861 $LCTL mark multiop $MOUNT/$tfile OS_c
862 multiop $MOUNT/$tfile OS_c &
864 writeme -s $MOUNT/${tfile}-2 &
867 facet_failover $SINGLEMDS
868 #define OBD_FAIL_MDS_CONNECT_NET 0x117
869 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
871 stat1=`count_ost_writes`
873 stat2=`count_ost_writes`
874 echo "$stat1, $stat2"
875 if [ $stat1 -lt $stat2 ]; then
876 echo "writes continuing during recovery"
879 echo "writes not continuing during recovery, bug 2477"
882 echo "waiting for writeme $WRITE_PID"
886 echo "waiting for multiop $PID"
887 wait $PID || return 2
888 do_facet client munlink $MOUNT/$tfile || return 3
889 do_facet client munlink $MOUNT/${tfile}-2 || return 3
892 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
896 # make sure that a read to one osc doesn't try to double-unlock its page just
897 # because another osc is invalid. trigger_group_io used to mistakenly return
898 # an error if any oscs were invalid even after having successfully put rpcs
899 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
900 # the page, guarnateeing that the unlock from the RPC completion would
901 # assert on trying to unlock the unlocked page.
903 [ $OSTCOUNT -lt 2 ] && \
904 skip_env "skipping test 41: we don't have a second OST to test with" && \
907 local f=$MOUNT/$tfile
908 # make sure the start of the file is ost1
909 lfs setstripe $f -s $((128 * 1024)) -i 0
910 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
912 # fail ost2 and read from ost1
913 local osc2dev=`do_facet $SINGLEMDS "lctl get_param -n devices | grep ${ost2_svc}-osc-MDT0000" | awk '{print $1}'`
914 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices && return 4
915 do_facet $SINGLEMDS $LCTL --device $osc2dev deactivate || return 1
916 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
917 do_facet $SINGLEMDS $LCTL --device $osc2dev activate || return 2
920 run_test 41 "read from a valid osc while other oscs are invalid"
922 # test MDS recovery after ost failure
924 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
925 createmany -o $DIR/$tfile-%d 800
927 unlinkmany $DIR/$tfile-%d 0 400
929 lctl set_param debug=-1
932 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
933 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
934 #[ $blocks_after -lt $blocks ] || return 1
935 echo wait for MDS to timeout and recover
936 sleep $((TIMEOUT * 2))
938 unlinkmany $DIR/$tfile-%d 400 400
939 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
941 run_test 42 "recovery after ost failure"
943 # timeout in MDS/OST recovery RPC will LBUG MDS
944 test_43() { # bug 2530
945 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
947 replay_barrier $SINGLEMDS
949 # OBD_FAIL_OST_CREATE_NET 0x204
950 do_facet ost1 "lctl set_param fail_loc=0x80000204"
953 do_facet ost1 "lctl set_param fail_loc=0"
957 run_test 43 "mds osc import failure during recovery; don't LBUG"
959 test_44a() { # was test_44
962 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
963 [ "$mdcdev" ] || exit 2
965 # adaptive timeouts slow this way down
966 if at_is_enabled; then
967 at_max_saved=$(at_max_get mds)
971 for i in `seq 1 10`; do
972 echo "$i of 10 ($(date +%s))"
973 do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
974 #define OBD_FAIL_TGT_CONN_RACE 0x701
975 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
976 $LCTL --device $mdcdev recover
979 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
980 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
983 run_test 44a "race in target handle connect"
986 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
987 [ "$mdcdev" ] || exit 2
988 for i in `seq 1 10`; do
989 echo "$i of 10 ($(date +%s))"
990 do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
991 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
992 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
993 $LCTL --device $mdcdev recover
996 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
999 run_test 44b "race in target handle connect"
1001 # Handle failed close
1003 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
1004 [ "$mdcdev" ] || exit 2
1005 $LCTL --device $mdcdev recover
1007 multiop_bg_pause $DIR/$tfile O_c || return 1
1010 # This will cause the CLOSE to fail before even
1011 # allocating a reply buffer
1012 $LCTL --device $mdcdev deactivate || return 4
1016 wait $pid || return 1
1018 $LCTL --device $mdcdev activate || return 5
1021 $CHECKSTAT -t file $DIR/$tfile || return 2
1024 run_test 45 "Handle failed close"
1028 drop_reply "touch $DIR/$tfile"
1030 # ironically, the previous test, 45, will cause a real forced close,
1031 # so just look for one for this test
1032 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1035 run_test 46 "Don't leak file handle after open resend (3325)"
1037 test_47() { # bug 2824
1038 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1040 # create some files to make sure precreate has been done on all
1041 # OSTs. (just in case this test is run independently)
1042 createmany -o $DIR/$tfile 20 || return 1
1044 # OBD_FAIL_OST_CREATE_NET 0x204
1046 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1047 df $MOUNT || return 2
1049 # let the MDS discover the OST failure, attempt to recover, fail
1050 # and recover again.
1051 sleep $((3 * TIMEOUT))
1053 # Without 2824, this createmany would hang
1054 createmany -o $DIR/$tfile 20 || return 3
1055 unlinkmany $DIR/$tfile 20 || return 4
1057 do_facet ost1 "lctl set_param fail_loc=0"
1060 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1063 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1064 [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2 OSTs -- skipping" && return
1066 replay_barrier $SINGLEMDS
1067 createmany -o $DIR/$tfile 20 || return 1
1068 # OBD_FAIL_OST_EROFS 0x216
1069 facet_failover $SINGLEMDS
1070 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1071 df $MOUNT || return 2
1073 createmany -o $DIR/$tfile 20 20 || return 2
1074 unlinkmany $DIR/$tfile 40 || return 3
1077 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1080 local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
1081 [ "$oscdev" ] || return 1
1082 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1083 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1084 # give the mds_lov_sync threads a chance to run
1087 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1089 # b3764 timed out lock replay
1092 cancel_lru_locks mdc
1094 multiop $DIR/$tfile s || return 1
1095 replay_barrier $SINGLEMDS
1096 #define OBD_FAIL_LDLM_REPLY 0x30c
1097 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1098 fail $SINGLEMDS || return 2
1099 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1101 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1103 run_test 52 "time out lock replay (3764)"
1105 # bug 3462 - simultaneous MDC requests
1107 mkdir -p $DIR/${tdir}-1
1108 mkdir -p $DIR/${tdir}-2
1109 multiop $DIR/${tdir}-1/f O_c &
1111 # give multiop a change to open
1114 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1115 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1116 kill -USR1 $close_pid
1117 cancel_lru_locks mdc # force the close
1118 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1120 mcreate $DIR/${tdir}-2/f || return 1
1122 # close should still be here
1123 [ -d /proc/$close_pid ] || return 2
1125 replay_barrier_nodf $SINGLEMDS
1127 wait $close_pid || return 3
1129 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1130 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1131 rm -rf $DIR/${tdir}-*
1133 run_test 53a "|X| close request while two MDC requests in flight"
1136 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1138 mkdir -p $DIR/${tdir}-1
1139 mkdir -p $DIR/${tdir}-2
1140 multiop_bg_pause $DIR/${tdir}-1/f O_c || return 6
1143 #define OBD_FAIL_MDS_REINT_NET 0x107
1144 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1145 mcreate $DIR/${tdir}-2/f &
1149 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1150 kill -USR1 $close_pid
1151 cancel_lru_locks mdc # force the close
1152 wait $close_pid || return 1
1153 # open should still be here
1154 [ -d /proc/$open_pid ] || return 2
1156 replay_barrier_nodf $SINGLEMDS
1158 wait $open_pid || return 3
1160 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1161 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1162 rm -rf $DIR/${tdir}-*
1164 run_test 53b "|X| open request while two MDC requests in flight"
1167 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1169 mkdir -p $DIR/${tdir}-1
1170 mkdir -p $DIR/${tdir}-2
1171 multiop $DIR/${tdir}-1/f O_c &
1174 #define OBD_FAIL_MDS_REINT_NET 0x107
1175 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1176 mcreate $DIR/${tdir}-2/f &
1180 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1181 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1182 kill -USR1 $close_pid
1183 cancel_lru_locks mdc # force the close
1185 replay_barrier_nodf $SINGLEMDS
1186 fail_nodf $SINGLEMDS
1187 wait $open_pid || return 1
1189 # close should be gone
1190 [ -d /proc/$close_pid ] && return 2
1191 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1193 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1194 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1195 rm -rf $DIR/${tdir}-*
1197 run_test 53c "|X| open request and close request while two MDC requests in flight"
1200 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1202 mkdir -p $DIR/${tdir}-1
1203 mkdir -p $DIR/${tdir}-2
1204 multiop $DIR/${tdir}-1/f O_c &
1206 # give multiop a chance to open
1209 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1210 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1211 kill -USR1 $close_pid
1212 cancel_lru_locks mdc # force the close
1213 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1214 mcreate $DIR/${tdir}-2/f || return 1
1216 # close should still be here
1217 [ -d /proc/$close_pid ] || return 2
1219 wait $close_pid || return 3
1221 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1222 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1223 rm -rf $DIR/${tdir}-*
1225 run_test 53d "|X| close reply while two MDC requests in flight"
1228 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1230 mkdir -p $DIR/${tdir}-1
1231 mkdir -p $DIR/${tdir}-2
1232 multiop $DIR/${tdir}-1/f O_c &
1235 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1236 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1237 mcreate $DIR/${tdir}-2/f &
1241 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1242 kill -USR1 $close_pid
1243 cancel_lru_locks mdc # force the close
1244 wait $close_pid || return 1
1245 # open should still be here
1246 [ -d /proc/$open_pid ] || return 2
1248 replay_barrier_nodf $SINGLEMDS
1250 wait $open_pid || return 3
1252 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1253 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1254 rm -rf $DIR/${tdir}-*
1256 run_test 53e "|X| open reply while two MDC requests in flight"
1259 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1261 mkdir -p $DIR/${tdir}-1
1262 mkdir -p $DIR/${tdir}-2
1263 multiop $DIR/${tdir}-1/f O_c &
1266 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1267 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1268 mcreate $DIR/${tdir}-2/f &
1272 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1273 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1274 kill -USR1 $close_pid
1275 cancel_lru_locks mdc # force the close
1277 replay_barrier_nodf $SINGLEMDS
1278 fail_nodf $SINGLEMDS
1279 wait $open_pid || return 1
1281 # close should be gone
1282 [ -d /proc/$close_pid ] && return 2
1283 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1285 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1286 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1287 rm -rf $DIR/${tdir}-*
1289 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1292 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1294 mkdir -p $DIR/${tdir}-1
1295 mkdir -p $DIR/${tdir}-2
1296 multiop $DIR/${tdir}-1/f O_c &
1299 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1300 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1301 mcreate $DIR/${tdir}-2/f &
1305 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1306 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1307 kill -USR1 $close_pid
1308 cancel_lru_locks mdc # force the close
1310 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1311 replay_barrier_nodf $SINGLEMDS
1312 fail_nodf $SINGLEMDS
1313 wait $open_pid || return 1
1315 # close should be gone
1316 [ -d /proc/$close_pid ] && return 2
1318 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1319 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1320 rm -rf $DIR/${tdir}-*
1322 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1325 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1327 mkdir -p $DIR/${tdir}-1
1328 mkdir -p $DIR/${tdir}-2
1329 multiop $DIR/${tdir}-1/f O_c &
1332 #define OBD_FAIL_MDS_REINT_NET 0x107
1333 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1334 mcreate $DIR/${tdir}-2/f &
1338 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1339 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1340 kill -USR1 $close_pid
1341 cancel_lru_locks mdc # force the close
1344 replay_barrier_nodf $SINGLEMDS
1345 fail_nodf $SINGLEMDS
1346 wait $open_pid || return 1
1348 # close should be gone
1349 [ -d /proc/$close_pid ] && return 2
1350 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1352 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1353 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1354 rm -rf $DIR/${tdir}-*
1356 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1358 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1360 #b3761 ASSERTION(hash != 0) failed
1362 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1363 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1365 # give touch a chance to run
1367 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1371 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1373 #b3440 ASSERTION(rec->ur_fid2->id) failed
1375 ln -s foo $DIR/$tfile
1376 replay_barrier $SINGLEMDS
1377 #drop_reply "cat $DIR/$tfile"
1381 run_test 56 "don't replay a symlink open request (3440)"
1383 #recovery one mds-ost setattr from llog
1385 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1386 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1388 replay_barrier $SINGLEMDS
1391 $CHECKSTAT -t file $DIR/$tfile || return 1
1392 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1395 run_test 57 "test recovery from llog for setattr op"
1397 #recovery many mds-ost setattr from llog
1400 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1401 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1402 createmany -o $DIR/$tdir/$tfile-%d 2500
1403 replay_barrier $SINGLEMDS
1406 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1407 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1408 unlinkmany $DIR/$tdir/$tfile-%d 2500
1411 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
1414 mount_client $MOUNT2
1416 touch $DIR/$tdir/$tfile
1417 replay_barrier $SINGLEMDS
1418 setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile
1420 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1421 [ x$VAL = x"bar" ] || return 1
1422 rm -f $DIR/$tdir/$tfile
1424 zconf_umount `hostname` $MOUNT2
1426 run_test 58b "test replay of setxattr op"
1428 test_58c() { # bug 16570
1429 mount_client $MOUNT2
1431 touch $DIR/$tdir/$tfile
1432 drop_request "setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile" || \
1434 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1435 [ x$VAL = x"bar" ] || return 2
1436 drop_reint_reply "setfattr -n trusted.foo1 -v bar1 $DIR/$tdir/$tfile" || \
1438 VAL=`getfattr --absolute-names --only-value -n trusted.foo1 $MOUNT2/$tdir/$tfile`
1439 [ x$VAL = x"bar1" ] || return 4
1440 rm -f $DIR/$tdir/$tfile
1442 zconf_umount `hostname` $MOUNT2
1444 run_test 58c "resend/reconstruct setxattr op"
1446 # log_commit_thread vs filter_destroy race used to lead to import use after free
1449 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1452 createmany -o $DIR/$tdir/$tfile-%d 200
1454 unlinkmany $DIR/$tdir/$tfile-%d 200
1455 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1456 do_facet ost1 "lctl set_param fail_loc=0x507"
1459 do_facet ost1 "lctl set_param fail_loc=0x0"
1463 run_test 59 "test log_commit_thread vs filter_destroy race"
1465 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1466 # bug 12086: should no oops and No ctxt error for this test
1469 createmany -o $DIR/$tdir/$tfile-%d 200
1470 replay_barrier $SINGLEMDS
1471 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1473 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1474 local no_ctxt=`dmesg | grep "No ctxt"`
1475 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1477 run_test 60 "test llog post recovery init vs llog unlink"
1479 #test race llog recovery thread vs llog cleanup
1480 test_61a() { # was test_61
1481 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1484 createmany -o $DIR/$tdir/$tfile-%d 800
1486 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1487 unlinkmany $DIR/$tdir/$tfile-%d 800
1488 set_nodes_failloc "$(osts_nodes)" 0x80000221
1493 set_nodes_failloc "$(osts_nodes)" 0x0
1495 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1498 run_test 61a "test race llog recovery vs llog cleanup"
1500 #test race mds llog sync vs llog cleanup
1502 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1503 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
1504 facet_failover $SINGLEMDS
1507 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1509 run_test 61b "test race mds llog sync vs llog cleanup"
1511 #test race cancel cookie cb vs llog cleanup
1513 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1515 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1517 set_nodes_failloc "$(osts_nodes)" 0x80000222
1521 set_nodes_failloc "$(osts_nodes)" 0x0
1523 run_test 61c "test race mds llog sync vs llog cleanup"
1525 test_61d() { # bug 16002 # bug 17466
1526 shutdown_facet $SINGLEMDS
1527 #define OBD_FAIL_OBD_LLOG_SETUP 0x605
1528 do_facet $SINGLEMDS "lctl set_param fail_loc=0x605"
1529 start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS && error "mds start should have failed"
1530 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1531 start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS || error "cannot restart mds"
1533 run_test 61d "error in llog_setup should cleanup the llog context correctly"
1535 test_62() { # Bug 15756 - don't mis-drop resent replay
1537 replay_barrier $SINGLEMDS
1538 createmany -o $DIR/$tdir/$tfile- 25
1539 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1540 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1541 facet_failover $SINGLEMDS
1542 df $MOUNT || return 1
1543 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1544 unlinkmany $DIR/$tdir/$tfile- 25 || return 2
1547 run_test 62 "don't mis-drop resent replay"
1549 #Adaptive Timeouts (bug 3055)
1557 echo "Cleaning up AT ..."
1558 if [ -n "$ATOLDBASE" ]; then
1559 local at_history=$($LCTL get_param -n at_history)
1560 do_facet mds "lctl set_param at_history=$at_history" || true
1561 do_facet ost1 "lctl set_param at_history=$at_history" || true
1564 if [ $AT_MAX_SET -ne 0 ]; then
1565 for facet in mds client ost; do
1566 var=AT_MAX_SAVE_${facet}
1567 echo restore AT on $facet to saved value ${!var}
1568 at_max_set ${!var} $facet
1569 at_new=$(at_max_get $facet)
1570 echo Restored AT value on $facet $at_new
1571 [ $at_new -eq ${!var} ] || \
1572 error "$facet : AT value was not restored SAVED ${!var} NEW $at_new"
1579 local at_max_new=600
1581 # Save at_max original values
1583 if [ $AT_MAX_SET -eq 0 ]; then
1584 # Suppose that all osts have the same at_max
1585 for facet in mds client ost; do
1586 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1590 for facet in mds client ost; do
1591 at_max=$(at_max_get $facet)
1592 if [ $at_max -ne $at_max_new ]; then
1593 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1594 at_max_set $at_max_new $facet
1599 if [ -z "$ATOLDBASE" ]; then
1600 ATOLDBASE=$(do_facet mds "lctl get_param -n at_history")
1601 # speed up the timebase so we can check decreasing AT
1602 do_facet mds "lctl set_param at_history=8" || true
1603 do_facet ost1 "lctl set_param at_history=8" || true
1605 # sleep for a while to cool down, should be > 8s and also allow
1606 # at least one ping to be sent. simply use TIMEOUT to be safe.
1611 test_65a() #bug 3055
1613 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1615 at_start || return 0
1616 $LCTL dk > /dev/null
1618 sysctl -w lnet.debug="+other"
1619 # Slow down a request to the current service time, this is critical
1620 # because previous tests may have caused this value to increase.
1621 REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
1622 awk '/portal 12/ {print $5}'`
1623 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1625 do_facet mds lctl set_param fail_val=$((${REQ_DELAY} * 1000))
1626 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1627 do_facet mds sysctl -w lustre.fail_loc=0x8000050a
1628 createmany -o $DIR/$tfile 10 > /dev/null
1629 unlinkmany $DIR/$tfile 10 > /dev/null
1630 # check for log message
1631 $LCTL dk | grep "Early reply #" || error "No early reply"
1633 # client should show REQ_DELAY estimates
1634 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1636 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1638 run_test 65a "AT: verify early replies"
1640 test_65b() #bug 3055
1642 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1644 at_start || return 0
1647 sysctl -w lnet.debug="other trace"
1648 $LCTL dk > /dev/null
1649 # Slow down a request to the current service time, this is critical
1650 # because previous tests may have caused this value to increase.
1651 REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
1652 awk '/portal 6/ {print $5}'`
1653 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1655 do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
1656 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1657 do_facet ost1 sysctl -w lustre.fail_loc=0x224
1660 lfs setstripe $DIR/$tfile --index=0 --count=1
1661 # force some real bulk transfer
1662 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1664 do_facet ost1 sysctl -w lustre.fail_loc=0
1665 # check for log message
1666 $LCTL dk | grep "Early reply #" || error "No early reply"
1668 # client should show REQ_DELAY estimates
1669 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1671 run_test 65b "AT: verify early replies on packed reply / bulk"
1673 test_66a() #bug 3055
1675 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1677 at_start || return 0
1678 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1679 # adjust 5s at a time so no early reply is sent (within deadline)
1680 do_facet mds "sysctl -w lustre.fail_val=5000"
1681 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1682 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1683 createmany -o $DIR/$tfile 20 > /dev/null
1684 unlinkmany $DIR/$tfile 20 > /dev/null
1685 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1686 do_facet mds "sysctl -w lustre.fail_val=10000"
1687 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1688 createmany -o $DIR/$tfile 20 > /dev/null
1689 unlinkmany $DIR/$tfile 20 > /dev/null
1690 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1691 do_facet mds "sysctl -w lustre.fail_loc=0"
1693 createmany -o $DIR/$tfile 20 > /dev/null
1694 unlinkmany $DIR/$tfile 20 > /dev/null
1695 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1696 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1697 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1698 echo "Current MDT timeout $CUR, worst $WORST"
1699 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1701 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1703 test_66b() #bug 3055
1705 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1707 at_start || return 0
1708 ORIG=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1709 sysctl -w lustre.fail_val=$(($ORIG + 5))
1710 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1711 sysctl -w lustre.fail_loc=0x50c
1712 ls $DIR/$tfile > /dev/null 2>&1
1713 sysctl -w lustre.fail_loc=0
1714 CUR=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1715 WORST=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $6}')
1716 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1717 [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
1719 run_test 66b "AT: verify net latency adjusts"
1721 test_67a() #bug 3055
1723 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1725 at_start || return 0
1726 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1727 # sleeping threads may drive values above this
1728 do_facet ost1 "sysctl -w lustre.fail_val=400"
1729 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1730 do_facet ost1 "sysctl -w lustre.fail_loc=0x50a"
1731 createmany -o $DIR/$tfile 20 > /dev/null
1732 unlinkmany $DIR/$tfile 20 > /dev/null
1733 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1734 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1735 ATTEMPTS=$(($CONN2 - $CONN1))
1736 echo "$ATTEMPTS osc reconnect attempts on gradual slow"
1737 [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
1740 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1742 test_67b() #bug 3055
1744 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1746 at_start || return 0
1747 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1748 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1749 do_facet ost1 "sysctl -w lustre.fail_val=20000"
1750 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1751 cp /etc/profile $DIR/$tfile || error "cp failed"
1753 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1755 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1756 ATTEMPTS=$(($CONN2 - $CONN1))
1757 echo "$ATTEMPTS osc reconnect attempts on instant slow"
1758 # do it again; should not timeout
1759 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1760 cp /etc/profile $DIR/$tfile || error "cp failed"
1761 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1763 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1764 CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1765 ATTEMPTS=$(($CONN3 - $CONN2))
1766 echo "$ATTEMPTS osc reconnect attempts on 2nd slow"
1767 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1770 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1772 test_68 () #bug 13813
1774 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1776 at_start || return 0
1777 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1778 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1779 local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
1780 [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
1781 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1782 local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
1783 echo $TIMEOUT >> $ldlm_enqueue_min
1784 do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
1788 lfs setstripe $DIR/$tdir --index=0 --count=1
1789 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1790 sysctl -w lustre.fail_val=$(($TIMEOUT - 1))
1791 sysctl -w lustre.fail_loc=0x80000312
1792 cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
1793 sysctl -w lustre.fail_val=$((TIMEOUT * 5 / 4))
1794 sysctl -w lustre.fail_loc=0x80000312
1795 cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
1796 sysctl -w lustre.fail_loc=0
1798 echo $ENQ_MIN >> $ldlm_enqueue_min
1799 do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
1803 run_test 68 "AT: verify slowing locks"
1806 # end of AT tests includes above lines
1809 # start multi-client tests
1811 [ -z "$CLIENTS" ] && \
1812 { skip "Need two or more clients." && return; }
1813 [ $CLIENTCOUNT -lt 2 ] && \
1814 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1816 echo "mount clients $CLIENTS ..."
1817 zconf_mount_clients $CLIENTS $DIR
1819 local clients=${CLIENTS//,/ }
1820 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1821 for CLIENT in $clients; do
1822 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1823 of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
1824 error "dd failed on $CLIENT"
1827 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
1828 for C in ${CLIENTS//,/ }; do
1829 do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
1830 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
1836 run_test 70a "check multi client t-f"
1839 local clients=${CLIENTS:-$HOSTNAME}
1841 zconf_mount_clients $clients $DIR
1844 [ "$SLOW" = "no" ] && duration=60
1845 local cmd="rundbench 1 -t $duration"
1847 do_nodes --verbose $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
1848 PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
1849 DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
1852 log "Started rundbench load PID=$PID ..."
1855 START_TS=$(date +%s)
1856 CURRENT_TS=$START_TS
1857 while [ $ELAPSED -lt $duration ]; do
1859 replay_barrier $SINGLEMDS
1860 sleep 1 # give clients a time to do operations
1861 # Increment the number of failovers
1862 NUM_FAILOVERS=$((NUM_FAILOVERS+1))
1863 log "$TESTNAME fail mds1 $NUM_FAILOVERS times"
1865 CURRENT_TS=$(date +%s)
1866 ELAPSED=$((CURRENT_TS - START_TS))
1868 wait $PID || error "rundbench load on $CLIENTS failed!"
1870 run_test 70b "mds recovery; $CLIENTCOUNT clients"
1871 # end multi-client tests
1874 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1878 replay_barrier $SINGLEMDS
1879 #define OBD_FAIL_LDLM_ENQUEUE 0x302
1880 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
1883 wait $pid || return 1
1884 [ -e $DIR/$tfile ] && return 2
1887 run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay , close"
1890 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1894 replay_barrier $SINGLEMDS
1895 #define OBD_FAIL_LDLM_REPLY 0x30c
1896 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1899 wait $pid || return 1
1900 [ -e $DIR/$tfile ] && return 2
1903 run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
1906 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1910 replay_barrier $SINGLEMDS
1911 #define OBD_FAIL_TGT_LAST_REPLAY 0x710
1912 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000710"
1915 wait $pid || return 1
1916 [ -e $DIR/$tfile ] && return 2
1919 run_test 73c "open(O_CREAT), unlink, replay, reconnect at last_replay, close"
1923 local clients=${CLIENTS:-$HOSTNAME}
1926 zconf_umount_clients $clients $MOUNT
1927 facet_failover $SINGLEMDS
1928 zconf_mount_clients $clients $MOUNT
1930 touch $DIR/$tfile || return 1
1931 rm $DIR/$tfile || return 2
1932 client_df || error "df failed: $?"
1935 run_test 74 "Ensure applications don't fail waiting for OST recovery"
1938 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1942 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
1943 rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
1945 stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
1948 run_test 80a "CMD: unlink cross-node dir (fail mds with inode)"
1951 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1955 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
1956 rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
1958 stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
1961 run_test 80b "CMD: unlink cross-node dir (fail mds with name)"
1964 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1967 createmany -o $DIR/$tdir/f 3000 || error "createmany failed"
1969 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir failed"
1970 $CHECKSTAT -t file $DIR/$tdir/f1002 || error "$CHECKSTAT -t file failed"
1972 rm $DIR/$tdir/f1002 || error "rm $DIR/$tdir/f1002 failed"
1974 stat $DIR/$tdir/f1002
1976 run_test 81a "CMD: unlink cross-node file (fail mds with name)"
1979 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1983 mkdir $dir || error "mkdir $dir failed"
1987 $CHECKSTAT -t dir $dir || error "$CHECKSTAT -t dir $dir failed"
1989 run_test 82a "CMD: mkdir cross-node dir (fail mds with inode)"
1992 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1996 mkdir $dir || error "mkdir $dir failed"
2000 $CHECKSTAT -t dir $dir || error "$CHECKSTAT -t dir $dir failed"
2002 run_test 82b "CMD: mkdir cross-node dir (fail mds with name)"
2006 createmany -o $DIR/$tdir/$tfile- 10 || return 1
2007 #define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD 0x140
2008 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
2009 unlinkmany $DIR/$tdir/$tfile- 10 || return 2
2011 run_test 83a "fail log_add during unlink recovery"
2015 createmany -o $DIR/$tdir/$tfile- 10 || return 1
2016 replay_barrier $SINGLEMDS
2017 unlinkmany $DIR/$tdir/$tfile- 10 || return 2
2018 #define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD 0x140
2019 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
2022 run_test 83b "fail log_add during unlink recovery"
2025 #define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x143
2026 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000143"
2027 createmany -o $DIR/$tfile- 1 &
2031 df -P $DIR || df -P $DIR || true # reconnect
2033 run_test 84a "stale open during export disconnect"
2035 equals_msg `basename $0`: test complete, cleaning up
2036 check_and_cleanup_lustre
2037 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true