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_facet ost "$LCTL get_param obdfilter.*.mds_sync" | awk -F= ' {print $2}'))
489 for ((i=0; i<${#sync[@]}; i++)); do
490 [ ${sync[$i]} -eq 0 ] && continue
491 # there is a not finished MDS-OST synchronization
495 [ ${con} -eq 1 ] && break
499 # let the statfs cache to get old enough.
502 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
503 log "before $BEFOREUSED, after $AFTERUSED"
504 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
505 error "after $AFTERUSED > before $BEFOREUSED"
508 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
510 test_20c() { # bug 10480
511 multiop_bg_pause $DIR/$tfile Ow_c || return 1
518 df -P $DIR || df -P $DIR || true # reconnect
521 test -s $DIR/$tfile || error "File was truncated"
523 wait $pid || return 1
526 run_test 20c "check that client eviction does not affect file content"
529 replay_barrier $SINGLEMDS
530 multiop_bg_pause $DIR/$tfile O_tSc || return 5
533 touch $DIR/g11 || return 1
537 wait $pid || return 2
538 [ -e $DIR/$tfile ] && return 3
539 touch $DIR/h11 || return 4
542 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
545 multiop_bg_pause $DIR/$tfile O_tSc || return 3
548 replay_barrier $SINGLEMDS
553 wait $pid || return 1
554 [ -e $DIR/$tfile ] && return 2
557 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
560 multiop_bg_pause $DIR/$tfile O_tSc || return 5
563 replay_barrier $SINGLEMDS
565 touch $DIR/g11 || return 1
569 wait $pid || return 2
570 [ -e $DIR/$tfile ] && return 3
571 touch $DIR/h11 || return 4
574 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
577 multiop_bg_pause $DIR/$tfile O_tSc || return 3
580 replay_barrier $SINGLEMDS
584 wait $pid || return 1
585 [ -e $DIR/$tfile ] && return 2
588 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
591 multiop_bg_pause $DIR/$tfile O_tSc || return 3
595 replay_barrier $SINGLEMDS
598 wait $pid || return 1
599 [ -e $DIR/$tfile ] && return 2
602 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
605 replay_barrier $SINGLEMDS
606 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
608 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
613 wait $pid2 || return 1
617 wait $pid1 || return 2
618 [ -e $DIR/$tfile-1 ] && return 3
619 [ -e $DIR/$tfile-2 ] && return 4
622 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
625 replay_barrier $SINGLEMDS
626 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
628 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
635 wait $pid1 || return 1
637 wait $pid2 || return 2
638 [ -e $DIR/$tfile-1 ] && return 3
639 [ -e $DIR/$tfile-2 ] && return 4
642 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
645 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
647 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
649 replay_barrier $SINGLEMDS
653 wait $pid2 || return 1
657 wait $pid1 || return 2
658 [ -e $DIR/$tfile-1 ] && return 3
659 [ -e $DIR/$tfile-2 ] && return 4
662 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
665 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
667 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
669 replay_barrier $SINGLEMDS
675 wait $pid1 || return 1
677 wait $pid2 || return 2
678 [ -e $DIR/$tfile-1 ] && return 3
679 [ -e $DIR/$tfile-2 ] && return 4
682 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
685 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
687 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
692 replay_barrier $SINGLEMDS
695 wait $pid1 || return 1
697 wait $pid2 || return 2
698 [ -e $DIR/$tfile-1 ] && return 3
699 [ -e $DIR/$tfile-2 ] && return 4
702 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
705 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
707 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
711 replay_barrier $SINGLEMDS
715 wait $pid1 || return 1
717 wait $pid2 || return 2
718 [ -e $DIR/$tfile-1 ] && return 3
719 [ -e $DIR/$tfile-2 ] && return 4
722 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
724 # tests for bug 2104; completion without crashing is success. The close is
725 # stale, but we always return 0 for close, so the app never sees it.
727 multiop_bg_pause $DIR/$tfile O_c || return 2
729 multiop_bg_pause $DIR/$tfile O_c || return 3
732 df $MOUNT || sleep 1 && df $MOUNT || return 1
735 wait $pid1 || return 4
736 wait $pid2 || return 5
739 run_test 32 "close() notices client eviction; close() after client eviction"
741 # Abort recovery before client complete
742 test_33a() { # was test_33
743 replay_barrier $SINGLEMDS
744 createmany -o $DIR/$tfile-%d 100
745 fail_abort $SINGLEMDS
746 # this file should be gone, because the replay was aborted
747 $CHECKSTAT -t file $DIR/$tfile-* && return 3
748 unlinkmany $DIR/$tfile-%d 0 100
751 run_test 33a "abort recovery before client does replay"
753 # Stale FID sequence bug 15962
754 test_33b() { # was test_33a
755 replay_barrier $SINGLEMDS
756 createmany -o $DIR/$tfile-%d 10
757 fail_abort $SINGLEMDS
758 unlinkmany $DIR/$tfile-%d 0 10
759 # recreate shouldn't fail
760 createmany -o $DIR/$tfile-%d 10 || return 3
761 unlinkmany $DIR/$tfile-%d 0 10
764 run_test 33b "fid shouldn't be reused after abort recovery"
767 multiop_bg_pause $DIR/$tfile O_c || return 2
771 replay_barrier $SINGLEMDS
772 fail_abort $SINGLEMDS
774 wait $pid || return 3
775 [ -e $DIR/$tfile ] && return 1
779 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
781 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
785 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
786 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
791 # give a chance to remove from MDS
792 fail_abort $SINGLEMDS
793 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
795 run_test 35 "test recovery from llog for unlink op"
797 # b=2432 resent cancel after replay uses wrong cookie,
798 # so don't resend cancels
800 replay_barrier $SINGLEMDS
802 checkstat $DIR/$tfile
803 facet_failover $SINGLEMDS
805 if dmesg | grep "unknown lock cookie"; then
806 echo "cancel after replay failed"
810 run_test 36 "don't resend cancel"
813 # directory orphans can't be unlinked from PENDING directory
815 rmdir $DIR/$tfile 2>/dev/null
816 multiop_bg_pause $DIR/$tfile dD_c || return 2
820 replay_barrier $SINGLEMDS
821 # clear the dmesg buffer so we only see errors from this recovery
823 fail_abort $SINGLEMDS
825 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
826 wait $pid || return 3
830 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
833 createmany -o $DIR/$tfile-%d 800
834 unlinkmany $DIR/$tfile-%d 0 400
835 replay_barrier $SINGLEMDS
837 unlinkmany $DIR/$tfile-%d 400 400
839 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
841 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
843 test_39() { # bug 4176
844 createmany -o $DIR/$tfile-%d 800
845 replay_barrier $SINGLEMDS
846 unlinkmany $DIR/$tfile-%d 0 400
848 unlinkmany $DIR/$tfile-%d 400 400
850 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
852 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
855 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
860 $LCTL mark multiop $MOUNT/$tfile OS_c
861 multiop $MOUNT/$tfile OS_c &
863 writeme -s $MOUNT/${tfile}-2 &
866 facet_failover $SINGLEMDS
867 #define OBD_FAIL_MDS_CONNECT_NET 0x117
868 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
870 stat1=`count_ost_writes`
872 stat2=`count_ost_writes`
873 echo "$stat1, $stat2"
874 if [ $stat1 -lt $stat2 ]; then
875 echo "writes continuing during recovery"
878 echo "writes not continuing during recovery, bug 2477"
881 echo "waiting for writeme $WRITE_PID"
885 echo "waiting for multiop $PID"
886 wait $PID || return 2
887 do_facet client munlink $MOUNT/$tfile || return 3
888 do_facet client munlink $MOUNT/${tfile}-2 || return 3
891 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
895 # make sure that a read to one osc doesn't try to double-unlock its page just
896 # because another osc is invalid. trigger_group_io used to mistakenly return
897 # an error if any oscs were invalid even after having successfully put rpcs
898 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
899 # the page, guarnateeing that the unlock from the RPC completion would
900 # assert on trying to unlock the unlocked page.
902 [ $OSTCOUNT -lt 2 ] && \
903 skip "skipping test 41: we don't have a second OST to test with" && \
906 local f=$MOUNT/$tfile
907 # make sure the start of the file is ost1
908 lfs setstripe $f -s $((128 * 1024)) -i 0
909 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
911 # fail ost2 and read from ost1
912 local osc2dev=`do_facet $SINGLEMDS "lctl get_param -n devices | grep ${ost2_svc}-osc-MDT0000" | awk '{print $1}'`
913 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices && return 4
914 do_facet $SINGLEMDS $LCTL --device $osc2dev deactivate || return 1
915 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
916 do_facet $SINGLEMDS $LCTL --device $osc2dev activate || return 2
919 run_test 41 "read from a valid osc while other oscs are invalid"
921 # test MDS recovery after ost failure
923 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
924 createmany -o $DIR/$tfile-%d 800
926 unlinkmany $DIR/$tfile-%d 0 400
928 lctl set_param debug=-1
931 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
932 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
933 #[ $blocks_after -lt $blocks ] || return 1
934 echo wait for MDS to timeout and recover
935 sleep $((TIMEOUT * 2))
937 unlinkmany $DIR/$tfile-%d 400 400
938 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
940 run_test 42 "recovery after ost failure"
942 # timeout in MDS/OST recovery RPC will LBUG MDS
943 test_43() { # bug 2530
944 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
946 replay_barrier $SINGLEMDS
948 # OBD_FAIL_OST_CREATE_NET 0x204
949 do_facet ost1 "lctl set_param fail_loc=0x80000204"
952 do_facet ost1 "lctl set_param fail_loc=0"
956 run_test 43 "mds osc import failure during recovery; don't LBUG"
958 test_44a() { # was test_44
961 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
962 [ "$mdcdev" ] || exit 2
964 # adaptive timeouts slow this way down
965 if at_is_enabled; then
966 at_max_saved=$(at_max_get mds)
970 for i in `seq 1 10`; do
971 echo "$i of 10 ($(date +%s))"
972 do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
973 #define OBD_FAIL_TGT_CONN_RACE 0x701
974 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
975 $LCTL --device $mdcdev recover
978 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
979 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
982 run_test 44a "race in target handle connect"
985 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
986 [ "$mdcdev" ] || exit 2
987 for i in `seq 1 10`; do
988 echo "$i of 10 ($(date +%s))"
989 do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
990 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
991 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
992 $LCTL --device $mdcdev recover
995 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
998 run_test 44b "race in target handle connect"
1000 # Handle failed close
1002 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
1003 [ "$mdcdev" ] || exit 2
1004 $LCTL --device $mdcdev recover
1006 multiop_bg_pause $DIR/$tfile O_c || return 1
1009 # This will cause the CLOSE to fail before even
1010 # allocating a reply buffer
1011 $LCTL --device $mdcdev deactivate || return 4
1015 wait $pid || return 1
1017 $LCTL --device $mdcdev activate || return 5
1020 $CHECKSTAT -t file $DIR/$tfile || return 2
1023 run_test 45 "Handle failed close"
1027 drop_reply "touch $DIR/$tfile"
1029 # ironically, the previous test, 45, will cause a real forced close,
1030 # so just look for one for this test
1031 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1034 run_test 46 "Don't leak file handle after open resend (3325)"
1036 test_47() { # bug 2824
1037 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1039 # create some files to make sure precreate has been done on all
1040 # OSTs. (just in case this test is run independently)
1041 createmany -o $DIR/$tfile 20 || return 1
1043 # OBD_FAIL_OST_CREATE_NET 0x204
1045 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1046 df $MOUNT || return 2
1048 # let the MDS discover the OST failure, attempt to recover, fail
1049 # and recover again.
1050 sleep $((3 * TIMEOUT))
1052 # Without 2824, this createmany would hang
1053 createmany -o $DIR/$tfile 20 || return 3
1054 unlinkmany $DIR/$tfile 20 || return 4
1056 do_facet ost1 "lctl set_param fail_loc=0"
1059 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1062 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1063 [ "$OSTCOUNT" -lt "2" ] && skip "$OSTCOUNT < 2 OSTs -- skipping" && return
1065 replay_barrier $SINGLEMDS
1066 createmany -o $DIR/$tfile 20 || return 1
1067 # OBD_FAIL_OST_EROFS 0x216
1068 facet_failover $SINGLEMDS
1069 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1070 df $MOUNT || return 2
1072 createmany -o $DIR/$tfile 20 20 || return 2
1073 unlinkmany $DIR/$tfile 40 || return 3
1076 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1079 local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
1080 [ "$oscdev" ] || return 1
1081 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1082 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1083 # give the mds_lov_sync threads a chance to run
1086 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1088 # b3764 timed out lock replay
1091 cancel_lru_locks mdc
1093 multiop $DIR/$tfile s || return 1
1094 replay_barrier $SINGLEMDS
1095 #define OBD_FAIL_LDLM_REPLY 0x30c
1096 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1097 fail $SINGLEMDS || return 2
1098 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1100 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1102 run_test 52 "time out lock replay (3764)"
1104 # bug 3462 - simultaneous MDC requests
1106 mkdir -p $DIR/${tdir}-1
1107 mkdir -p $DIR/${tdir}-2
1108 multiop $DIR/${tdir}-1/f O_c &
1110 # give multiop a change to open
1113 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1114 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1115 kill -USR1 $close_pid
1116 cancel_lru_locks mdc # force the close
1117 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1119 mcreate $DIR/${tdir}-2/f || return 1
1121 # close should still be here
1122 [ -d /proc/$close_pid ] || return 2
1124 replay_barrier_nodf $SINGLEMDS
1126 wait $close_pid || return 3
1128 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1129 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1130 rm -rf $DIR/${tdir}-*
1132 run_test 53a "|X| close request while two MDC requests in flight"
1135 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1137 mkdir -p $DIR/${tdir}-1
1138 mkdir -p $DIR/${tdir}-2
1139 multiop_bg_pause $DIR/${tdir}-1/f O_c || return 6
1142 #define OBD_FAIL_MDS_REINT_NET 0x107
1143 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1144 mcreate $DIR/${tdir}-2/f &
1148 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1149 kill -USR1 $close_pid
1150 cancel_lru_locks mdc # force the close
1151 wait $close_pid || return 1
1152 # open should still be here
1153 [ -d /proc/$open_pid ] || return 2
1155 replay_barrier_nodf $SINGLEMDS
1157 wait $open_pid || return 3
1159 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1160 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1161 rm -rf $DIR/${tdir}-*
1163 run_test 53b "|X| open request while two MDC requests in flight"
1166 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1168 mkdir -p $DIR/${tdir}-1
1169 mkdir -p $DIR/${tdir}-2
1170 multiop $DIR/${tdir}-1/f O_c &
1173 #define OBD_FAIL_MDS_REINT_NET 0x107
1174 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1175 mcreate $DIR/${tdir}-2/f &
1179 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1180 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1181 kill -USR1 $close_pid
1182 cancel_lru_locks mdc # force the close
1184 replay_barrier_nodf $SINGLEMDS
1185 fail_nodf $SINGLEMDS
1186 wait $open_pid || return 1
1188 # close should be gone
1189 [ -d /proc/$close_pid ] && return 2
1190 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1192 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1193 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1194 rm -rf $DIR/${tdir}-*
1196 run_test 53c "|X| open request and close request while two MDC requests in flight"
1199 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1201 mkdir -p $DIR/${tdir}-1
1202 mkdir -p $DIR/${tdir}-2
1203 multiop $DIR/${tdir}-1/f O_c &
1205 # give multiop a chance to open
1208 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1209 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1210 kill -USR1 $close_pid
1211 cancel_lru_locks mdc # force the close
1212 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1213 mcreate $DIR/${tdir}-2/f || return 1
1215 # close should still be here
1216 [ -d /proc/$close_pid ] || return 2
1218 wait $close_pid || return 3
1220 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1221 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1222 rm -rf $DIR/${tdir}-*
1224 run_test 53d "|X| close reply while two MDC requests in flight"
1227 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1229 mkdir -p $DIR/${tdir}-1
1230 mkdir -p $DIR/${tdir}-2
1231 multiop $DIR/${tdir}-1/f O_c &
1234 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1235 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1236 mcreate $DIR/${tdir}-2/f &
1240 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1241 kill -USR1 $close_pid
1242 cancel_lru_locks mdc # force the close
1243 wait $close_pid || return 1
1244 # open should still be here
1245 [ -d /proc/$open_pid ] || return 2
1247 replay_barrier_nodf $SINGLEMDS
1249 wait $open_pid || return 3
1251 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1252 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1253 rm -rf $DIR/${tdir}-*
1255 run_test 53e "|X| open reply while two MDC requests in flight"
1258 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1260 mkdir -p $DIR/${tdir}-1
1261 mkdir -p $DIR/${tdir}-2
1262 multiop $DIR/${tdir}-1/f O_c &
1265 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1266 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1267 mcreate $DIR/${tdir}-2/f &
1271 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1272 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1273 kill -USR1 $close_pid
1274 cancel_lru_locks mdc # force the close
1276 replay_barrier_nodf $SINGLEMDS
1277 fail_nodf $SINGLEMDS
1278 wait $open_pid || return 1
1280 # close should be gone
1281 [ -d /proc/$close_pid ] && return 2
1282 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1284 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1285 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1286 rm -rf $DIR/${tdir}-*
1288 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1291 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1293 mkdir -p $DIR/${tdir}-1
1294 mkdir -p $DIR/${tdir}-2
1295 multiop $DIR/${tdir}-1/f O_c &
1298 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1299 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1300 mcreate $DIR/${tdir}-2/f &
1304 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1305 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1306 kill -USR1 $close_pid
1307 cancel_lru_locks mdc # force the close
1309 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1310 replay_barrier_nodf $SINGLEMDS
1311 fail_nodf $SINGLEMDS
1312 wait $open_pid || return 1
1314 # close should be gone
1315 [ -d /proc/$close_pid ] && return 2
1317 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1318 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1319 rm -rf $DIR/${tdir}-*
1321 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1324 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1326 mkdir -p $DIR/${tdir}-1
1327 mkdir -p $DIR/${tdir}-2
1328 multiop $DIR/${tdir}-1/f O_c &
1331 #define OBD_FAIL_MDS_REINT_NET 0x107
1332 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1333 mcreate $DIR/${tdir}-2/f &
1337 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1338 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1339 kill -USR1 $close_pid
1340 cancel_lru_locks mdc # force the close
1343 replay_barrier_nodf $SINGLEMDS
1344 fail_nodf $SINGLEMDS
1345 wait $open_pid || return 1
1347 # close should be gone
1348 [ -d /proc/$close_pid ] && return 2
1349 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1351 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1352 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1353 rm -rf $DIR/${tdir}-*
1355 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1357 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1359 #b3761 ASSERTION(hash != 0) failed
1361 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1362 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1364 # give touch a chance to run
1366 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1370 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1372 #b3440 ASSERTION(rec->ur_fid2->id) failed
1374 ln -s foo $DIR/$tfile
1375 replay_barrier $SINGLEMDS
1376 #drop_reply "cat $DIR/$tfile"
1380 run_test 56 "don't replay a symlink open request (3440)"
1382 #recovery one mds-ost setattr from llog
1384 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1385 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1387 replay_barrier $SINGLEMDS
1390 $CHECKSTAT -t file $DIR/$tfile || return 1
1391 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1394 run_test 57 "test recovery from llog for setattr op"
1396 #recovery many mds-ost setattr from llog
1399 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1400 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1401 createmany -o $DIR/$tdir/$tfile-%d 2500
1402 replay_barrier $SINGLEMDS
1405 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1406 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1407 unlinkmany $DIR/$tdir/$tfile-%d 2500
1410 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
1413 mount_client $MOUNT2
1415 touch $DIR/$tdir/$tfile
1416 replay_barrier $SINGLEMDS
1417 setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile
1419 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1420 [ x$VAL = x"bar" ] || return 1
1421 rm -f $DIR/$tdir/$tfile
1423 zconf_umount `hostname` $MOUNT2
1425 run_test 58b "test replay of setxattr op"
1427 test_58c() { # bug 16570
1428 mount_client $MOUNT2
1430 touch $DIR/$tdir/$tfile
1431 drop_request "setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile" || \
1433 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1434 [ x$VAL = x"bar" ] || return 2
1435 drop_reint_reply "setfattr -n trusted.foo1 -v bar1 $DIR/$tdir/$tfile" || \
1437 VAL=`getfattr --absolute-names --only-value -n trusted.foo1 $MOUNT2/$tdir/$tfile`
1438 [ x$VAL = x"bar1" ] || return 4
1439 rm -f $DIR/$tdir/$tfile
1441 zconf_umount `hostname` $MOUNT2
1443 run_test 58c "resend/reconstruct setxattr op"
1445 # log_commit_thread vs filter_destroy race used to lead to import use after free
1448 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1451 createmany -o $DIR/$tdir/$tfile-%d 200
1453 unlinkmany $DIR/$tdir/$tfile-%d 200
1454 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1455 do_facet ost1 "lctl set_param fail_loc=0x507"
1458 do_facet ost1 "lctl set_param fail_loc=0x0"
1462 run_test 59 "test log_commit_thread vs filter_destroy race"
1464 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1465 # bug 12086: should no oops and No ctxt error for this test
1468 createmany -o $DIR/$tdir/$tfile-%d 200
1469 replay_barrier $SINGLEMDS
1470 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1472 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1473 local no_ctxt=`dmesg | grep "No ctxt"`
1474 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1476 run_test 60 "test llog post recovery init vs llog unlink"
1478 #test race llog recovery thread vs llog cleanup
1479 test_61a() { # was test_61
1480 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1483 createmany -o $DIR/$tdir/$tfile-%d 800
1485 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1486 unlinkmany $DIR/$tdir/$tfile-%d 800
1487 set_nodes_failloc "$(osts_nodes)" 0x80000221
1492 set_nodes_failloc "$(osts_nodes)" 0x0
1494 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1497 run_test 61a "test race llog recovery vs llog cleanup"
1499 #test race mds llog sync vs llog cleanup
1501 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1502 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
1503 facet_failover $SINGLEMDS
1506 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1508 run_test 61b "test race mds llog sync vs llog cleanup"
1510 #test race cancel cookie cb vs llog cleanup
1512 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1514 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1516 set_nodes_failloc "$(osts_nodes)" 0x80000222
1520 set_nodes_failloc "$(osts_nodes)" 0x0
1522 run_test 61c "test race mds llog sync vs llog cleanup"
1524 test_61d() { # bug 16002 # bug 17466
1525 shutdown_facet $SINGLEMDS
1526 #define OBD_FAIL_OBD_LLOG_SETUP 0x605
1527 do_facet $SINGLEMDS "lctl set_param fail_loc=0x605"
1528 start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS && error "mds start should have failed"
1529 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1530 start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS || error "cannot restart mds"
1532 run_test 61d "error in llog_setup should cleanup the llog context correctly"
1534 test_62() { # Bug 15756 - don't mis-drop resent replay
1536 replay_barrier $SINGLEMDS
1537 createmany -o $DIR/$tdir/$tfile- 25
1538 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1539 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1540 facet_failover $SINGLEMDS
1541 df $MOUNT || return 1
1542 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1543 unlinkmany $DIR/$tdir/$tfile- 25 || return 2
1546 run_test 62 "don't mis-drop resent replay"
1548 #Adaptive Timeouts (bug 3055)
1556 echo "Cleaning up AT ..."
1557 if [ -n "$ATOLDBASE" ]; then
1558 local at_history=$($LCTL get_param -n at_history)
1559 do_facet mds "lctl set_param at_history=$at_history" || true
1560 do_facet ost1 "lctl set_param at_history=$at_history" || true
1563 if [ $AT_MAX_SET -ne 0 ]; then
1564 for facet in mds client ost; do
1565 var=AT_MAX_SAVE_${facet}
1566 echo restore AT on $facet to saved value ${!var}
1567 at_max_set ${!var} $facet
1568 at_new=$(at_max_get $facet)
1569 echo Restored AT value on $facet $at_new
1570 [ $at_new -eq ${!var} ] || \
1571 error "$facet : AT value was not restored SAVED ${!var} NEW $at_new"
1578 local at_max_new=600
1580 # Save at_max original values
1582 if [ $AT_MAX_SET -eq 0 ]; then
1583 # Suppose that all osts have the same at_max
1584 for facet in mds client ost; do
1585 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1589 for facet in mds client ost; do
1590 at_max=$(at_max_get $facet)
1591 if [ $at_max -ne $at_max_new ]; then
1592 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1593 at_max_set $at_max_new $facet
1598 if [ -z "$ATOLDBASE" ]; then
1599 ATOLDBASE=$(do_facet mds "lctl get_param -n at_history")
1600 # speed up the timebase so we can check decreasing AT
1601 do_facet mds "lctl set_param at_history=8" || true
1602 do_facet ost1 "lctl set_param at_history=8" || true
1604 # sleep for a while to cool down, should be > 8s and also allow
1605 # at least one ping to be sent. simply use TIMEOUT to be safe.
1610 test_65a() #bug 3055
1612 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1614 at_start || return 0
1615 $LCTL dk > /dev/null
1617 sysctl -w lnet.debug="+other"
1618 # Slow down a request to the current service time, this is critical
1619 # because previous tests may have caused this value to increase.
1620 REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
1621 awk '/portal 12/ {print $5}'`
1622 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1624 do_facet mds lctl set_param fail_val=$((${REQ_DELAY} * 1000))
1625 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1626 do_facet mds sysctl -w lustre.fail_loc=0x8000050a
1627 createmany -o $DIR/$tfile 10 > /dev/null
1628 unlinkmany $DIR/$tfile 10 > /dev/null
1629 # check for log message
1630 $LCTL dk | grep "Early reply #" || error "No early reply"
1632 # client should show REQ_DELAY estimates
1633 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1635 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1637 run_test 65a "AT: verify early replies"
1639 test_65b() #bug 3055
1641 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1643 at_start || return 0
1646 sysctl -w lnet.debug="other trace"
1647 $LCTL dk > /dev/null
1648 # Slow down a request to the current service time, this is critical
1649 # because previous tests may have caused this value to increase.
1650 REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
1651 awk '/portal 6/ {print $5}'`
1652 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1654 do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
1655 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1656 do_facet ost1 sysctl -w lustre.fail_loc=0x224
1659 lfs setstripe $DIR/$tfile --index=0 --count=1
1660 # force some real bulk transfer
1661 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1663 do_facet ost1 sysctl -w lustre.fail_loc=0
1664 # check for log message
1665 $LCTL dk | grep "Early reply #" || error "No early reply"
1667 # client should show REQ_DELAY estimates
1668 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1670 run_test 65b "AT: verify early replies on packed reply / bulk"
1672 test_66a() #bug 3055
1674 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1676 at_start || return 0
1677 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1678 # adjust 5s at a time so no early reply is sent (within deadline)
1679 do_facet mds "sysctl -w lustre.fail_val=5000"
1680 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1681 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1682 createmany -o $DIR/$tfile 20 > /dev/null
1683 unlinkmany $DIR/$tfile 20 > /dev/null
1684 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1685 do_facet mds "sysctl -w lustre.fail_val=10000"
1686 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1687 createmany -o $DIR/$tfile 20 > /dev/null
1688 unlinkmany $DIR/$tfile 20 > /dev/null
1689 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1690 do_facet mds "sysctl -w lustre.fail_loc=0"
1692 createmany -o $DIR/$tfile 20 > /dev/null
1693 unlinkmany $DIR/$tfile 20 > /dev/null
1694 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1695 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1696 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1697 echo "Current MDT timeout $CUR, worst $WORST"
1698 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1700 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1702 test_66b() #bug 3055
1704 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1706 at_start || return 0
1707 ORIG=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1708 sysctl -w lustre.fail_val=$(($ORIG + 5))
1709 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1710 sysctl -w lustre.fail_loc=0x50c
1711 ls $DIR/$tfile > /dev/null 2>&1
1712 sysctl -w lustre.fail_loc=0
1713 CUR=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1714 WORST=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $6}')
1715 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1716 [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
1718 run_test 66b "AT: verify net latency adjusts"
1720 test_67a() #bug 3055
1722 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1724 at_start || return 0
1725 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1726 # sleeping threads may drive values above this
1727 do_facet ost1 "sysctl -w lustre.fail_val=400"
1728 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1729 do_facet ost1 "sysctl -w lustre.fail_loc=0x50a"
1730 createmany -o $DIR/$tfile 20 > /dev/null
1731 unlinkmany $DIR/$tfile 20 > /dev/null
1732 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1733 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1734 ATTEMPTS=$(($CONN2 - $CONN1))
1735 echo "$ATTEMPTS osc reconnect attempts on gradual slow"
1736 [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
1739 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1741 test_67b() #bug 3055
1743 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1745 at_start || return 0
1746 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1747 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1748 do_facet ost1 "sysctl -w lustre.fail_val=20000"
1749 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1750 cp /etc/profile $DIR/$tfile || error "cp failed"
1752 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1754 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1755 ATTEMPTS=$(($CONN2 - $CONN1))
1756 echo "$ATTEMPTS osc reconnect attempts on instant slow"
1757 # do it again; should not timeout
1758 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1759 cp /etc/profile $DIR/$tfile || error "cp failed"
1760 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1762 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1763 CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1764 ATTEMPTS=$(($CONN3 - $CONN2))
1765 echo "$ATTEMPTS osc reconnect attempts on 2nd slow"
1766 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1769 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1771 test_68 () #bug 13813
1773 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1775 at_start || return 0
1776 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1777 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1778 local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
1779 [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
1780 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1781 local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
1782 echo $TIMEOUT >> $ldlm_enqueue_min
1783 do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
1787 lfs setstripe $DIR/$tdir --index=0 --count=1
1788 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1789 sysctl -w lustre.fail_val=$(($TIMEOUT - 1))
1790 sysctl -w lustre.fail_loc=0x80000312
1791 cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
1792 sysctl -w lustre.fail_val=$((TIMEOUT * 5 / 4))
1793 sysctl -w lustre.fail_loc=0x80000312
1794 cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
1795 sysctl -w lustre.fail_loc=0
1797 echo $ENQ_MIN >> $ldlm_enqueue_min
1798 do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
1802 run_test 68 "AT: verify slowing locks"
1805 # end of AT tests includes above lines
1808 # start multi-client tests
1810 [ -z "$CLIENTS" ] && \
1811 { skip "Need two or more clients." && return; }
1812 [ $CLIENTCOUNT -lt 2 ] && \
1813 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1815 echo "mount clients $CLIENTS ..."
1816 zconf_mount_clients $CLIENTS $DIR
1818 local clients=${CLIENTS//,/ }
1819 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1820 for CLIENT in $clients; do
1821 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1822 of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
1823 error "dd failed on $CLIENT"
1826 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
1827 for C in ${CLIENTS//,/ }; do
1828 do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
1829 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
1835 run_test 70a "check multi client t-f"
1838 local clients=${CLIENTS:-$HOSTNAME}
1840 zconf_mount_clients $clients $DIR
1843 [ "$SLOW" = "no" ] && duration=60
1844 local cmd="rundbench 1 -t $duration"
1846 do_nodes --verbose $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
1847 PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
1848 DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
1851 log "Started rundbench load PID=$PID ..."
1854 START_TS=$(date +%s)
1855 CURRENT_TS=$START_TS
1856 while [ $ELAPSED -lt $duration ]; do
1858 replay_barrier $SINGLEMDS
1859 sleep 1 # give clients a time to do operations
1860 # Increment the number of failovers
1861 NUM_FAILOVERS=$((NUM_FAILOVERS+1))
1862 log "$TESTNAME fail mds1 $NUM_FAILOVERS times"
1864 CURRENT_TS=$(date +%s)
1865 ELAPSED=$((CURRENT_TS - START_TS))
1867 wait $PID || error "rundbench load on $CLIENTS failed!"
1869 run_test 70b "mds recovery; $CLIENTCOUNT clients"
1870 # end multi-client tests
1873 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1877 replay_barrier $SINGLEMDS
1878 #define OBD_FAIL_LDLM_ENQUEUE 0x302
1879 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
1882 wait $pid || return 1
1883 [ -e $DIR/$tfile ] && return 2
1886 run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay , close"
1889 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1893 replay_barrier $SINGLEMDS
1894 #define OBD_FAIL_LDLM_REPLY 0x30c
1895 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1898 wait $pid || return 1
1899 [ -e $DIR/$tfile ] && return 2
1902 run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
1905 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1909 replay_barrier $SINGLEMDS
1910 #define OBD_FAIL_TGT_LAST_REPLAY 0x710
1911 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000710"
1914 wait $pid || return 1
1915 [ -e $DIR/$tfile ] && return 2
1918 run_test 73c "open(O_CREAT), unlink, replay, reconnect at last_replay, close"
1922 local clients=${CLIENTS:-$HOSTNAME}
1925 zconf_umount_clients $clients $MOUNT
1926 facet_failover $SINGLEMDS
1927 zconf_mount_clients $clients $MOUNT
1929 touch $DIR/$tfile || return 1
1930 rm $DIR/$tfile || return 2
1931 client_df || error "df failed: $?"
1934 run_test 74 "Ensure applications don't fail waiting for OST recovery"
1937 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1941 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
1942 rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
1944 stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
1947 run_test 80a "CMD: unlink cross-node dir (fail mds with inode)"
1950 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1954 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir $DIR/$tdir failed"
1955 rmdir $DIR/$tdir || error "rmdir $DIR/$tdir failed"
1957 stat $DIR/$tdir 2&>/dev/null && error "$DIR/$tdir still exist after recovery!"
1960 run_test 80b "CMD: unlink cross-node dir (fail mds with name)"
1963 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1966 createmany -o $DIR/$tdir/f 3000 || error "createmany failed"
1968 $CHECKSTAT -t dir $DIR/$tdir || error "$CHECKSTAT -t dir failed"
1969 $CHECKSTAT -t file $DIR/$tdir/f1002 || error "$CHECKSTAT -t file failed"
1971 rm $DIR/$tdir/f1002 || error "rm $DIR/$tdir/f1002 failed"
1973 stat $DIR/$tdir/f1002
1975 run_test 81a "CMD: unlink cross-node file (fail mds with name)"
1978 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1982 mkdir $dir || error "mkdir $dir failed"
1986 $CHECKSTAT -t dir $dir || error "$CHECKSTAT -t dir $dir failed"
1988 run_test 82a "CMD: mkdir cross-node dir (fail mds with inode)"
1991 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1995 mkdir $dir || error "mkdir $dir failed"
1999 $CHECKSTAT -t dir $dir || error "$CHECKSTAT -t dir $dir failed"
2001 run_test 82b "CMD: mkdir cross-node dir (fail mds with name)"
2005 createmany -o $DIR/$tdir/$tfile- 10 || return 1
2006 #define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD 0x140
2007 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
2008 unlinkmany $DIR/$tdir/$tfile- 10 || return 2
2010 run_test 83a "fail log_add during unlink recovery"
2014 createmany -o $DIR/$tdir/$tfile- 10 || return 1
2015 replay_barrier $SINGLEMDS
2016 unlinkmany $DIR/$tdir/$tfile- 10 || return 2
2017 #define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD 0x140
2018 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
2021 run_test 83b "fail log_add during unlink recovery"
2024 #define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x143
2025 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000143"
2026 createmany -o $DIR/$tfile- 1 &
2030 df -P $DIR || df -P $DIR || true # reconnect
2032 run_test 84a "stale open during export disconnect"
2034 equals_msg `basename $0`: test complete, cleaning up
2035 check_and_cleanup_lustre
2036 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true