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/$NAME.sh}
17 CHECK_GRANT=${CHECK_GRANT:-"yes"}
18 GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
20 require_dsh_mds || exit 0
23 # bug number: 17466 18857
24 ALWAYS_EXCEPT="61d 33a 33b $REPLAY_SINGLE_EXCEPT"
26 if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
27 CONFIG_EXCEPTIONS="0b 42 47 61a 61c"
28 echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
29 echo "Except the tests: $CONFIG_EXCEPTIONS"
30 ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
33 # 63 min 7 min AT AT AT AT"
34 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a 44b 65 66 67 68"
38 check_and_setup_lustre
43 rm -rf $DIR/[df][0-9]*
45 test_0a() { # was test_0
48 replay_barrier $SINGLEMDS
52 run_test 0a "empty replay"
55 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
57 # this test attempts to trigger a race in the precreation code,
58 # and must run before any other objects are created on the filesystem
60 createmany -o $DIR/$tfile 20 || return 1
61 unlinkmany $DIR/$tfile 20 || return 2
63 run_test 0b "ensure object created after recover exists. (3284)"
69 lctl set_param -n seq.cli-srv-$mds-mdc-*.width=$width
75 lctl get_param -n seq.cli-srv-$mds-mdc-*.width
78 # This test should pass for single-mds and multi-mds configs.
79 # But for different configurations it tests different things.
83 # (1) fld_create replay should happen;
85 # (2) fld_create replay should not return -EEXISTS, if it does
86 # this means sequence manager recovery code is buggy and allocated
87 # same sequence two times after recovery.
91 # (1) fld_create replay may not happen, because its home MDS is
92 # MDS2 which is not involved to revovery;
94 # (2) as fld_create does not happen on MDS1, it does not make any
97 local label=`mdsdevlabel 1`
98 [ -z "$label" ] && echo "No label for mds1" && return 1
100 replay_barrier $SINGLEMDS
101 local sw=`seq_get_width $label`
103 # make seq manager switch to next sequence each
104 # time as new fid is needed.
105 seq_set_width $label 1
107 # make sure that fld has created at least one new
109 touch $DIR/$tfile || return 2
110 seq_set_width $label $sw
112 # fail $SINGLEMDS and start recovery, replay RPCs, etc.
115 # wait for recovery finish
119 # flush fld cache and dentry cache to make it lookup
120 # created entry instead of revalidating existent one
122 zconf_mount `hostname` $MOUNT
124 # issue lookup which should call fld lookup which
125 # should fail if client did not replay fld create
126 # correctly and server has no fld entry
127 touch $DIR/$tfile || return 3
128 rm $DIR/$tfile || return 4
130 start_full_debug_logging
131 run_test 0c "fld create"
132 stop_full_debug_logging
135 replay_barrier $SINGLEMDS
138 $CHECKSTAT -t file $DIR/$tfile || return 1
141 run_test 1 "simple create"
144 replay_barrier $SINGLEMDS
147 $CHECKSTAT -t file $DIR/$tfile || return 1
154 replay_barrier $SINGLEMDS
157 $CHECKSTAT -t file $DIR/$tfile || return 1
163 local file=$DIR/$tfile
164 replay_barrier $SINGLEMDS
166 openfile -f O_DIRECTORY $file
168 $CHECKSTAT -t file $file || return 2
171 run_test 3a "replay failed open(O_DIRECTORY)"
174 replay_barrier $SINGLEMDS
175 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
176 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000114"
178 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
180 $CHECKSTAT -t file $DIR/$tfile && return 2
183 run_test 3b "replay failed open -ENOMEM"
186 replay_barrier $SINGLEMDS
187 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
188 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
190 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
193 $CHECKSTAT -t file $DIR/$tfile && return 2
196 run_test 3c "replay failed open -ENOMEM"
198 test_4a() { # was test_4
199 replay_barrier $SINGLEMDS
200 for i in `seq 10`; do
201 echo "tag-$i" > $DIR/$tfile-$i
204 for i in `seq 10`; do
205 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
208 run_test 4a "|x| 10 open(O_CREAT)s"
211 replay_barrier $SINGLEMDS
214 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
216 run_test 4b "|x| rm 10 files"
218 # The idea is to get past the first block of precreated files on both
219 # osts, and then replay.
221 replay_barrier $SINGLEMDS
222 for i in `seq 220`; do
223 echo "tag-$i" > $DIR/$tfile-$i
226 for i in `seq 220`; do
227 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
231 # waiting for commitment of removal
233 run_test 5 "|x| 220 open(O_CREAT)"
236 test_6a() { # was test_6
238 replay_barrier $SINGLEMDS
239 mcreate $DIR/$tdir/$tfile
241 $CHECKSTAT -t dir $DIR/$tdir || return 1
242 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
244 # waiting for log process thread
246 run_test 6a "mkdir + contained create"
250 replay_barrier $SINGLEMDS
253 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
255 run_test 6b "|X| rmdir"
259 replay_barrier $SINGLEMDS
260 mcreate $DIR/$tdir/$tfile
262 $CHECKSTAT -t dir $DIR/$tdir || return 1
263 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
266 run_test 7 "mkdir |X| contained create"
269 # make sure no side-effect from previous test.
271 replay_barrier $SINGLEMDS
272 multiop_bg_pause $DIR/$tfile mo_c || return 4
276 $CHECKSTAT -t file $DIR/$tfile || return 1
277 kill -USR1 $MULTIPID || return 2
278 wait $MULTIPID || return 3
281 run_test 8 "creat open |X| close"
284 replay_barrier $SINGLEMDS
286 local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
288 local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
290 echo " old_inum == $old_inum, new_inum == $new_inum"
291 if [ $old_inum -eq $new_inum ] ;
293 echo " old_inum and new_inum match"
295 echo "!!!! old_inum and new_inum NOT match"
300 run_test 9 "|X| create (same inum/gen)"
304 replay_barrier $SINGLEMDS
305 mv $DIR/$tfile $DIR/$tfile-2
308 $CHECKSTAT $DIR/$tfile && return 1
309 $CHECKSTAT $DIR/$tfile-2 ||return 2
313 run_test 10 "create |X| rename unlink"
317 echo "old" > $DIR/$tfile
318 mv $DIR/$tfile $DIR/$tfile-2
319 replay_barrier $SINGLEMDS
320 echo "new" > $DIR/$tfile
322 grep old $DIR/$tfile-2
324 grep new $DIR/$tfile || return 1
325 grep old $DIR/$tfile-2 || return 2
327 run_test 11 "create open write rename |X| create-old-name read"
331 multiop_bg_pause $DIR/$tfile o_tSc || return 3
334 replay_barrier $SINGLEMDS
336 wait $pid || return 1
339 [ -e $DIR/$tfile ] && return 2
342 run_test 12 "open, unlink |X| close"
345 # 1777 - replay open after committed chmod that would make
346 # a regular open a failure
349 multiop_bg_pause $DIR/$tfile O_wc || return 3
352 $CHECKSTAT -p 0 $DIR/$tfile
353 replay_barrier $SINGLEMDS
356 wait $pid || return 1
358 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
359 rm $DIR/$tfile || return 4
362 run_test 13 "open chmod 0 |x| write close"
365 multiop_bg_pause $DIR/$tfile O_tSc || return 4
368 replay_barrier $SINGLEMDS
369 kill -USR1 $pid || return 1
370 wait $pid || return 2
373 [ -e $DIR/$tfile ] && return 3
376 run_test 14 "open(O_CREAT), unlink |X| close"
379 multiop_bg_pause $DIR/$tfile O_tSc || return 5
382 replay_barrier $SINGLEMDS
383 touch $DIR/g11 || return 1
385 wait $pid || return 2
388 [ -e $DIR/$tfile ] && return 3
389 touch $DIR/h11 || return 4
392 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
396 replay_barrier $SINGLEMDS
399 mcreate $DIR/$tfile-2
401 [ -e $DIR/$tfile ] && return 1
402 [ -e $DIR/$tfile-2 ] || return 2
403 munlink $DIR/$tfile-2 || return 3
405 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
408 replay_barrier $SINGLEMDS
409 multiop_bg_pause $DIR/$tfile O_c || return 4
412 kill -USR1 $pid || return 1
413 wait $pid || return 2
414 $CHECKSTAT -t file $DIR/$tfile || return 3
417 run_test 17 "|X| open(O_CREAT), |replay| close"
420 replay_barrier $SINGLEMDS
421 multiop_bg_pause $DIR/$tfile O_tSc || return 8
424 touch $DIR/$tfile-2 || return 1
425 echo "pid: $pid will close"
427 wait $pid || return 2
430 [ -e $DIR/$tfile ] && return 3
431 [ -e $DIR/$tfile-2 ] || return 4
432 # this touch frequently fails
433 touch $DIR/$tfile-3 || return 5
434 munlink $DIR/$tfile-2 || return 6
435 munlink $DIR/$tfile-3 || return 7
438 run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
440 # bug 1855 (a simpler form of test_11 above)
442 replay_barrier $SINGLEMDS
444 echo "old" > $DIR/$tfile
445 mv $DIR/$tfile $DIR/$tfile-2
446 grep old $DIR/$tfile-2
448 grep old $DIR/$tfile-2 || return 2
450 run_test 19 "|X| mcreate, open, write, rename "
452 test_20a() { # was test_20
453 replay_barrier $SINGLEMDS
454 multiop_bg_pause $DIR/$tfile O_tSc || return 3
460 wait $pid || return 1
461 [ -e $DIR/$tfile ] && return 2
464 run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
466 test_20b() { # bug 10480
467 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
469 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
471 while [ ! -e $DIR/$tfile ] ; do
472 usleep 60 # give dd a chance to start
475 lfs getstripe $DIR/$tfile || return 1
476 rm -f $DIR/$tfile || return 2 # make it an orphan
478 client_up || client_up || true # reconnect
480 fail $SINGLEMDS # start orphan recovery
481 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
482 wait_mds_ost_sync || return 3
483 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
484 log "before $BEFOREUSED, after $AFTERUSED"
485 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
486 error "after $AFTERUSED > before $BEFOREUSED"
489 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
491 test_20c() { # bug 10480
492 multiop_bg_pause $DIR/$tfile Ow_c || return 1
498 client_up || client_up || true # reconnect
501 wait $pid || return 1
502 [ -s $DIR/$tfile ] || error "File was truncated"
506 run_test 20c "check that client eviction does not affect file content"
509 replay_barrier $SINGLEMDS
510 multiop_bg_pause $DIR/$tfile O_tSc || return 5
513 touch $DIR/g11 || return 1
517 wait $pid || return 2
518 [ -e $DIR/$tfile ] && return 3
519 touch $DIR/h11 || return 4
522 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
525 multiop_bg_pause $DIR/$tfile O_tSc || return 3
528 replay_barrier $SINGLEMDS
533 wait $pid || return 1
534 [ -e $DIR/$tfile ] && return 2
537 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
540 multiop_bg_pause $DIR/$tfile O_tSc || return 5
543 replay_barrier $SINGLEMDS
545 touch $DIR/g11 || return 1
549 wait $pid || return 2
550 [ -e $DIR/$tfile ] && return 3
551 touch $DIR/h11 || return 4
554 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
557 multiop_bg_pause $DIR/$tfile O_tSc || return 3
560 replay_barrier $SINGLEMDS
564 wait $pid || return 1
565 [ -e $DIR/$tfile ] && return 2
568 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
571 multiop_bg_pause $DIR/$tfile O_tSc || return 3
575 replay_barrier $SINGLEMDS
578 wait $pid || return 1
579 [ -e $DIR/$tfile ] && return 2
582 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
585 replay_barrier $SINGLEMDS
586 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
588 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
593 wait $pid2 || return 1
597 wait $pid1 || return 2
598 [ -e $DIR/$tfile-1 ] && return 3
599 [ -e $DIR/$tfile-2 ] && return 4
602 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (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
615 wait $pid1 || return 1
617 wait $pid2 || return 2
618 [ -e $DIR/$tfile-1 ] && return 3
619 [ -e $DIR/$tfile-2 ] && return 4
622 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
625 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
627 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
629 replay_barrier $SINGLEMDS
633 wait $pid2 || return 1
637 wait $pid1 || return 2
638 [ -e $DIR/$tfile-1 ] && return 3
639 [ -e $DIR/$tfile-2 ] && return 4
642 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (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
655 wait $pid1 || return 1
657 wait $pid2 || return 2
658 [ -e $DIR/$tfile-1 ] && return 3
659 [ -e $DIR/$tfile-2 ] && return 4
662 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (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
672 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 30 "open(O_CREAT) two, 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
691 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 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
704 # tests for bug 2104; completion without crashing is success. The close is
705 # stale, but we always return 0 for close, so the app never sees it.
707 multiop_bg_pause $DIR/$tfile O_c || return 2
709 multiop_bg_pause $DIR/$tfile O_c || return 3
712 client_up || client_up || return 1
715 wait $pid1 || return 4
716 wait $pid2 || return 5
719 run_test 32 "close() notices client eviction; close() after client eviction"
721 # Abort recovery before client complete
722 test_33a() { # was test_33
723 replay_barrier $SINGLEMDS
724 createmany -o $DIR/$tfile-%d 100
725 fail_abort $SINGLEMDS
726 # this file should be gone, because the replay was aborted
727 $CHECKSTAT -t file $DIR/$tfile-* && return 3
728 unlinkmany $DIR/$tfile-%d 0 100
731 run_test 33a "abort recovery before client does replay"
733 # Stale FID sequence bug 15962
734 test_33b() { # was test_33a
735 replay_barrier $SINGLEMDS
736 createmany -o $DIR/$tfile-%d 10
737 fail_abort $SINGLEMDS
738 unlinkmany $DIR/$tfile-%d 0 10
739 # recreate shouldn't fail
740 createmany -o $DIR/$tfile-%d 10 || return 3
741 unlinkmany $DIR/$tfile-%d 0 10
744 run_test 33b "fid shouldn't be reused after abort recovery"
747 multiop_bg_pause $DIR/$tfile O_c || return 2
751 replay_barrier $SINGLEMDS
752 fail_abort $SINGLEMDS
754 wait $pid || return 3
755 [ -e $DIR/$tfile ] && return 1
759 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
761 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
765 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
766 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
771 # give a chance to remove from MDS
772 fail_abort $SINGLEMDS
773 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
775 run_test 35 "test recovery from llog for unlink op"
777 # b=2432 resent cancel after replay uses wrong cookie,
778 # so don't resend cancels
780 replay_barrier $SINGLEMDS
782 checkstat $DIR/$tfile
783 facet_failover $SINGLEMDS
785 if dmesg | grep "unknown lock cookie"; then
786 echo "cancel after replay failed"
790 run_test 36 "don't resend cancel"
793 # directory orphans can't be unlinked from PENDING directory
795 rmdir $DIR/$tfile 2>/dev/null
796 multiop_bg_pause $DIR/$tfile dD_c || return 2
800 replay_barrier $SINGLEMDS
801 # clear the dmesg buffer so we only see errors from this recovery
803 fail_abort $SINGLEMDS
805 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
806 wait $pid || return 3
810 start_full_debug_logging
811 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
812 stop_full_debug_logging
815 createmany -o $DIR/$tfile-%d 800
816 unlinkmany $DIR/$tfile-%d 0 400
817 replay_barrier $SINGLEMDS
819 unlinkmany $DIR/$tfile-%d 400 400
821 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
823 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
825 test_39() { # bug 4176
826 createmany -o $DIR/$tfile-%d 800
827 replay_barrier $SINGLEMDS
828 unlinkmany $DIR/$tfile-%d 0 400
830 unlinkmany $DIR/$tfile-%d 400 400
832 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
834 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
837 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
842 $LCTL mark multiop $MOUNT/$tfile OS_c
843 multiop $MOUNT/$tfile OS_c &
845 writeme -s $MOUNT/${tfile}-2 &
848 facet_failover $SINGLEMDS
849 #define OBD_FAIL_MDS_CONNECT_NET 0x117
850 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
852 stat1=`count_ost_writes`
854 stat2=`count_ost_writes`
855 echo "$stat1, $stat2"
856 if [ $stat1 -lt $stat2 ]; then
857 echo "writes continuing during recovery"
860 echo "writes not continuing during recovery, bug 2477"
863 echo "waiting for writeme $WRITE_PID"
867 echo "waiting for multiop $PID"
868 wait $PID || return 2
869 do_facet client munlink $MOUNT/$tfile || return 3
870 do_facet client munlink $MOUNT/${tfile}-2 || return 3
873 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
877 # make sure that a read to one osc doesn't try to double-unlock its page just
878 # because another osc is invalid. trigger_group_io used to mistakenly return
879 # an error if any oscs were invalid even after having successfully put rpcs
880 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
881 # the page, guarnateeing that the unlock from the RPC completion would
882 # assert on trying to unlock the unlocked page.
884 [ $OSTCOUNT -lt 2 ] && \
885 skip_env "skipping test 41: we don't have a second OST to test with" && \
888 local f=$MOUNT/$tfile
889 # make sure the start of the file is ost1
890 lfs setstripe $f -s $((128 * 1024)) -i 0
891 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
893 # fail ost2 and read from ost1
894 local osc2dev=`do_facet $SINGLEMDS "lctl get_param -n devices | grep ${ost2_svc}-osc-MDT0000" | awk '{print $1}'`
895 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" && lctl get_param -n devices && return 4
896 do_facet $SINGLEMDS $LCTL --device $osc2dev deactivate || return 1
897 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
898 do_facet $SINGLEMDS $LCTL --device $osc2dev activate || return 2
901 run_test 41 "read from a valid osc while other oscs are invalid"
903 # test MDS recovery after ost failure
905 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
906 createmany -o $DIR/$tfile-%d 800
908 unlinkmany $DIR/$tfile-%d 0 400
910 lctl set_param debug=-1
913 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
914 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
915 #[ $blocks_after -lt $blocks ] || return 1
916 echo wait for MDS to timeout and recover
917 sleep $((TIMEOUT * 2))
919 unlinkmany $DIR/$tfile-%d 400 400
920 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
922 run_test 42 "recovery after ost failure"
924 # timeout in MDS/OST recovery RPC will LBUG MDS
925 test_43() { # bug 2530
926 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
928 replay_barrier $SINGLEMDS
930 # OBD_FAIL_OST_CREATE_NET 0x204
931 do_facet ost1 "lctl set_param fail_loc=0x80000204"
934 do_facet ost1 "lctl set_param fail_loc=0"
938 run_test 43 "mds osc import failure during recovery; don't LBUG"
940 test_44a() { # was test_44
943 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
944 [ "$mdcdev" ] || return 2
945 [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
947 # adaptive timeouts slow this way down
948 if at_is_enabled; then
949 at_max_saved=$(at_max_get mds)
953 for i in `seq 1 10`; do
954 echo "$i of 10 ($(date +%s))"
955 do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
956 #define OBD_FAIL_TGT_CONN_RACE 0x701
957 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
958 # lctl below may fail, it is valid case
959 $LCTL --device $mdcdev recover
962 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
963 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
966 run_test 44a "race in target handle connect"
969 local mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
970 [ "$mdcdev" ] || return 2
971 [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
973 for i in `seq 1 10`; do
974 echo "$i of 10 ($(date +%s))"
975 do_facet $SINGLEMDS "lctl get_param -n mdt.*.mdt.timeouts | grep service"
976 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
977 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
978 # lctl below may fail, it is valid case
979 $LCTL --device $mdcdev recover
982 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
985 run_test 44b "race in target handle connect"
987 # Handle failed close
989 mdcdev=`lctl get_param -n devices | awk '/MDT0000-mdc-/ {print $1}'`
990 [ "$mdcdev" ] || return 2
991 [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
993 $LCTL --device $mdcdev recover || return 6
995 multiop_bg_pause $DIR/$tfile O_c || return 1
998 # This will cause the CLOSE to fail before even
999 # allocating a reply buffer
1000 $LCTL --device $mdcdev deactivate || return 4
1004 wait $pid || return 1
1006 $LCTL --device $mdcdev activate || return 5
1009 $CHECKSTAT -t file $DIR/$tfile || return 2
1012 run_test 45 "Handle failed close"
1016 drop_reply "touch $DIR/$tfile"
1018 # ironically, the previous test, 45, will cause a real forced close,
1019 # so just look for one for this test
1020 dmesg | grep -i "force closing client file handle for $tfile" && return 1
1023 run_test 46 "Don't leak file handle after open resend (3325)"
1025 test_47() { # bug 2824
1026 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1028 # create some files to make sure precreate has been done on all
1029 # OSTs. (just in case this test is run independently)
1030 createmany -o $DIR/$tfile 20 || return 1
1032 # OBD_FAIL_OST_CREATE_NET 0x204
1034 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1035 client_up || return 2
1037 # let the MDS discover the OST failure, attempt to recover, fail
1038 # and recover again.
1039 sleep $((3 * TIMEOUT))
1041 # Without 2824, this createmany would hang
1042 createmany -o $DIR/$tfile 20 || return 3
1043 unlinkmany $DIR/$tfile 20 || return 4
1045 do_facet ost1 "lctl set_param fail_loc=0"
1048 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1051 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1052 [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2 OSTs -- skipping" && return
1054 replay_barrier $SINGLEMDS
1055 createmany -o $DIR/$tfile 20 || return 1
1056 # OBD_FAIL_OST_EROFS 0x216
1057 facet_failover $SINGLEMDS
1058 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1059 client_up || return 2
1061 createmany -o $DIR/$tfile 20 20 || return 2
1062 unlinkmany $DIR/$tfile 40 || return 3
1065 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1068 local oscdev=`do_facet $SINGLEMDS lctl get_param -n devices | grep ${ost1_svc}-osc-MDT0000 | awk '{print $1}'`
1069 [ "$oscdev" ] || return 1
1070 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 2
1071 do_facet $SINGLEMDS $LCTL --device $oscdev recover || return 3
1072 # give the mds_lov_sync threads a chance to run
1075 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1077 # b3764 timed out lock replay
1080 cancel_lru_locks mdc
1082 multiop $DIR/$tfile s || return 1
1083 replay_barrier $SINGLEMDS
1084 #define OBD_FAIL_LDLM_REPLY 0x30c
1085 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
1086 fail $SINGLEMDS || return 2
1087 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1089 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1091 run_test 52 "time out lock replay (3764)"
1093 # bug 3462 - simultaneous MDC requests
1095 mkdir -p $DIR/${tdir}-1
1096 mkdir -p $DIR/${tdir}-2
1097 multiop $DIR/${tdir}-1/f O_c &
1099 # give multiop a change to open
1102 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1103 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1104 kill -USR1 $close_pid
1105 cancel_lru_locks mdc # force the close
1106 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1108 mcreate $DIR/${tdir}-2/f || return 1
1110 # close should still be here
1111 [ -d /proc/$close_pid ] || return 2
1113 replay_barrier_nodf $SINGLEMDS
1115 wait $close_pid || return 3
1117 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1118 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1119 rm -rf $DIR/${tdir}-*
1121 run_test 53a "|X| close request while two MDC requests in flight"
1124 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1126 mkdir -p $DIR/${tdir}-1
1127 mkdir -p $DIR/${tdir}-2
1128 multiop_bg_pause $DIR/${tdir}-1/f O_c || return 6
1131 #define OBD_FAIL_MDS_REINT_NET 0x107
1132 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1133 mcreate $DIR/${tdir}-2/f &
1137 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1138 kill -USR1 $close_pid
1139 cancel_lru_locks mdc # force the close
1140 wait $close_pid || return 1
1141 # open should still be here
1142 [ -d /proc/$open_pid ] || return 2
1144 replay_barrier_nodf $SINGLEMDS
1146 wait $open_pid || return 3
1148 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1149 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1150 rm -rf $DIR/${tdir}-*
1152 run_test 53b "|X| open request while two MDC requests in flight"
1155 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1157 mkdir -p $DIR/${tdir}-1
1158 mkdir -p $DIR/${tdir}-2
1159 multiop $DIR/${tdir}-1/f O_c &
1162 #define OBD_FAIL_MDS_REINT_NET 0x107
1163 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1164 mcreate $DIR/${tdir}-2/f &
1168 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1169 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1170 kill -USR1 $close_pid
1171 cancel_lru_locks mdc # force the close
1173 replay_barrier_nodf $SINGLEMDS
1174 fail_nodf $SINGLEMDS
1175 wait $open_pid || return 1
1177 # close should be gone
1178 [ -d /proc/$close_pid ] && return 2
1179 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1181 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1182 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1183 rm -rf $DIR/${tdir}-*
1185 run_test 53c "|X| open request and close request while two MDC requests in flight"
1188 cancel_lru_locks mdc # cleanup locks from former test cases
1189 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1191 mkdir -p $DIR/${tdir}-1
1192 mkdir -p $DIR/${tdir}-2
1193 multiop $DIR/${tdir}-1/f O_c &
1195 # give multiop a chance to open
1198 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1199 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1200 kill -USR1 $close_pid
1201 cancel_lru_locks mdc # force the close
1202 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1203 mcreate $DIR/${tdir}-2/f || return 1
1205 # close should still be here
1206 [ -d /proc/$close_pid ] || return 2
1208 wait $close_pid || return 3
1210 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1211 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1212 rm -rf $DIR/${tdir}-*
1214 run_test 53d "|X| close reply while two MDC requests in flight"
1217 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1219 mkdir -p $DIR/${tdir}-1
1220 mkdir -p $DIR/${tdir}-2
1221 multiop $DIR/${tdir}-1/f O_c &
1224 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1225 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1226 mcreate $DIR/${tdir}-2/f &
1230 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1231 kill -USR1 $close_pid
1232 cancel_lru_locks mdc # force the close
1233 wait $close_pid || return 1
1234 # open should still be here
1235 [ -d /proc/$open_pid ] || return 2
1237 replay_barrier_nodf $SINGLEMDS
1239 wait $open_pid || return 3
1241 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1242 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1243 rm -rf $DIR/${tdir}-*
1245 run_test 53e "|X| open reply while two MDC requests in flight"
1248 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1250 mkdir -p $DIR/${tdir}-1
1251 mkdir -p $DIR/${tdir}-2
1252 multiop $DIR/${tdir}-1/f O_c &
1255 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1256 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1257 mcreate $DIR/${tdir}-2/f &
1261 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1262 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1263 kill -USR1 $close_pid
1264 cancel_lru_locks mdc # force the close
1266 replay_barrier_nodf $SINGLEMDS
1267 fail_nodf $SINGLEMDS
1268 wait $open_pid || return 1
1270 # close should be gone
1271 [ -d /proc/$close_pid ] && return 2
1272 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1274 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1275 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1276 rm -rf $DIR/${tdir}-*
1278 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1281 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1283 mkdir -p $DIR/${tdir}-1
1284 mkdir -p $DIR/${tdir}-2
1285 multiop $DIR/${tdir}-1/f O_c &
1288 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1289 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1290 mcreate $DIR/${tdir}-2/f &
1294 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1295 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1296 kill -USR1 $close_pid
1297 cancel_lru_locks mdc # force the close
1299 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1300 replay_barrier_nodf $SINGLEMDS
1301 fail_nodf $SINGLEMDS
1302 wait $open_pid || return 1
1304 # close should be gone
1305 [ -d /proc/$close_pid ] && return 2
1307 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1308 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1309 rm -rf $DIR/${tdir}-*
1311 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1314 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1316 mkdir -p $DIR/${tdir}-1
1317 mkdir -p $DIR/${tdir}-2
1318 multiop $DIR/${tdir}-1/f O_c &
1321 #define OBD_FAIL_MDS_REINT_NET 0x107
1322 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1323 mcreate $DIR/${tdir}-2/f &
1327 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1328 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1329 kill -USR1 $close_pid
1330 cancel_lru_locks mdc # force the close
1333 replay_barrier_nodf $SINGLEMDS
1334 fail_nodf $SINGLEMDS
1335 wait $open_pid || return 1
1337 # close should be gone
1338 [ -d /proc/$close_pid ] && return 2
1339 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1341 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1342 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1343 rm -rf $DIR/${tdir}-*
1345 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1347 #b_cray 54 "|X| open request and close reply while two MDC requests in flight"
1349 #b3761 ASSERTION(hash != 0) failed
1351 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1352 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1354 # give touch a chance to run
1356 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1360 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1362 #b3440 ASSERTION(rec->ur_fid2->id) failed
1364 ln -s foo $DIR/$tfile
1365 replay_barrier $SINGLEMDS
1366 #drop_reply "cat $DIR/$tfile"
1370 run_test 56 "don't replay a symlink open request (3440)"
1372 #recovery one mds-ost setattr from llog
1374 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1375 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1377 replay_barrier $SINGLEMDS
1380 $CHECKSTAT -t file $DIR/$tfile || return 1
1381 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1384 run_test 57 "test recovery from llog for setattr op"
1386 #recovery many mds-ost setattr from llog
1389 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1390 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1391 createmany -o $DIR/$tdir/$tfile-%d 2500
1392 replay_barrier $SINGLEMDS
1395 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1396 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1397 unlinkmany $DIR/$tdir/$tfile-%d 2500
1400 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
1403 mount_client $MOUNT2
1405 touch $DIR/$tdir/$tfile
1406 replay_barrier $SINGLEMDS
1407 setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile
1409 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1410 [ x$VAL = x"bar" ] || return 1
1411 rm -f $DIR/$tdir/$tfile
1413 zconf_umount `hostname` $MOUNT2
1415 run_test 58b "test replay of setxattr op"
1417 test_58c() { # bug 16570
1418 mount_client $MOUNT2
1420 touch $DIR/$tdir/$tfile
1421 drop_request "setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile" || \
1423 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1424 [ x$VAL = x"bar" ] || return 2
1425 drop_reint_reply "setfattr -n trusted.foo1 -v bar1 $DIR/$tdir/$tfile" || \
1427 VAL=`getfattr --absolute-names --only-value -n trusted.foo1 $MOUNT2/$tdir/$tfile`
1428 [ x$VAL = x"bar1" ] || return 4
1429 rm -f $DIR/$tdir/$tfile
1431 zconf_umount `hostname` $MOUNT2
1433 run_test 58c "resend/reconstruct setxattr op"
1435 # log_commit_thread vs filter_destroy race used to lead to import use after free
1438 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1441 createmany -o $DIR/$tdir/$tfile-%d 200
1443 unlinkmany $DIR/$tdir/$tfile-%d 200
1444 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1445 do_facet ost1 "lctl set_param fail_loc=0x507"
1448 do_facet ost1 "lctl set_param fail_loc=0x0"
1452 run_test 59 "test log_commit_thread vs filter_destroy race"
1454 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1455 # bug 12086: should no oops and No ctxt error for this test
1458 createmany -o $DIR/$tdir/$tfile-%d 200
1459 replay_barrier $SINGLEMDS
1460 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1462 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1463 local no_ctxt=`dmesg | grep "No ctxt"`
1464 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1466 run_test 60 "test llog post recovery init vs llog unlink"
1468 #test race llog recovery thread vs llog cleanup
1469 test_61a() { # was test_61
1470 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1473 createmany -o $DIR/$tdir/$tfile-%d 800
1475 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1476 unlinkmany $DIR/$tdir/$tfile-%d 800
1477 set_nodes_failloc "$(osts_nodes)" 0x80000221
1482 set_nodes_failloc "$(osts_nodes)" 0x0
1484 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1487 run_test 61a "test race llog recovery vs llog cleanup"
1489 #test race mds llog sync vs llog cleanup
1491 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1492 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
1493 facet_failover $SINGLEMDS
1496 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1498 run_test 61b "test race mds llog sync vs llog cleanup"
1500 #test race cancel cookie cb vs llog cleanup
1502 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1504 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1506 set_nodes_failloc "$(osts_nodes)" 0x80000222
1510 set_nodes_failloc "$(osts_nodes)" 0x0
1512 run_test 61c "test race mds llog sync vs llog cleanup"
1514 test_61d() { # bug 16002 # bug 17466
1515 shutdown_facet $SINGLEMDS
1516 #define OBD_FAIL_OBD_LLOG_SETUP 0x605
1517 do_facet $SINGLEMDS "lctl set_param fail_loc=0x605"
1518 start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS && error "mds start should have failed"
1519 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1520 start $SINGLEMDS `mdsdevname 1` $MDS_MOUNT_OPTS || error "cannot restart mds"
1522 run_test 61d "error in llog_setup should cleanup the llog context correctly"
1524 test_62() { # Bug 15756 - don't mis-drop resent replay
1526 replay_barrier $SINGLEMDS
1527 createmany -o $DIR/$tdir/$tfile- 25
1528 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1529 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1531 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1532 unlinkmany $DIR/$tdir/$tfile- 25 || return 2
1535 run_test 62 "don't mis-drop resent replay"
1537 #Adaptive Timeouts (bug 3055)
1545 echo "Cleaning up AT ..."
1546 if [ -n "$ATOLDBASE" ]; then
1547 local at_history=$($LCTL get_param -n at_history)
1548 do_facet mds "lctl set_param at_history=$at_history" || true
1549 do_facet ost1 "lctl set_param at_history=$at_history" || true
1552 if [ $AT_MAX_SET -ne 0 ]; then
1553 for facet in mds client ost; do
1554 var=AT_MAX_SAVE_${facet}
1555 echo restore AT on $facet to saved value ${!var}
1556 at_max_set ${!var} $facet
1557 at_new=$(at_max_get $facet)
1558 echo Restored AT value on $facet $at_new
1559 [ $at_new -eq ${!var} ] || \
1560 error "$facet : AT value was not restored SAVED ${!var} NEW $at_new"
1567 local at_max_new=600
1569 # Save at_max original values
1571 if [ $AT_MAX_SET -eq 0 ]; then
1572 # Suppose that all osts have the same at_max
1573 for facet in mds client ost; do
1574 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1578 for facet in mds client ost; do
1579 at_max=$(at_max_get $facet)
1580 if [ $at_max -ne $at_max_new ]; then
1581 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1582 at_max_set $at_max_new $facet
1587 if [ -z "$ATOLDBASE" ]; then
1588 ATOLDBASE=$(do_facet mds "lctl get_param -n at_history")
1589 # speed up the timebase so we can check decreasing AT
1590 do_facet mds "lctl set_param at_history=8" || true
1591 do_facet ost1 "lctl set_param at_history=8" || true
1593 # sleep for a while to cool down, should be > 8s and also allow
1594 # at least one ping to be sent. simply use TIMEOUT to be safe.
1599 test_65a() #bug 3055
1601 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1603 at_start || return 0
1604 $LCTL dk > /dev/null
1606 sysctl -w lnet.debug="+other"
1607 # Slow down a request to the current service time, this is critical
1608 # because previous tests may have caused this value to increase.
1609 REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
1610 awk '/portal 12/ {print $5}'`
1611 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1613 do_facet mds lctl set_param fail_val=$((${REQ_DELAY} * 1000))
1614 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1615 do_facet mds sysctl -w lustre.fail_loc=0x8000050a
1616 createmany -o $DIR/$tfile 10 > /dev/null
1617 unlinkmany $DIR/$tfile 10 > /dev/null
1618 # check for log message
1619 $LCTL dk | grep "Early reply #" || error "No early reply"
1621 # client should show REQ_DELAY estimates
1622 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1624 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1626 run_test 65a "AT: verify early replies"
1628 test_65b() #bug 3055
1630 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1632 at_start || return 0
1635 sysctl -w lnet.debug="other trace"
1636 $LCTL dk > /dev/null
1637 # Slow down a request to the current service time, this is critical
1638 # because previous tests may have caused this value to increase.
1639 REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
1640 awk '/portal 6/ {print $5}'`
1641 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1643 do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
1644 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1645 do_facet ost1 sysctl -w lustre.fail_loc=0x224
1648 lfs setstripe $DIR/$tfile --index=0 --count=1
1649 # force some real bulk transfer
1650 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1652 do_facet ost1 sysctl -w lustre.fail_loc=0
1653 # check for log message
1654 $LCTL dk | grep "Early reply #" || error "No early reply"
1656 # client should show REQ_DELAY estimates
1657 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1659 run_test 65b "AT: verify early replies on packed reply / bulk"
1661 test_66a() #bug 3055
1663 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1665 at_start || return 0
1666 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1667 # adjust 5s at a time so no early reply is sent (within deadline)
1668 do_facet mds "sysctl -w lustre.fail_val=5000"
1669 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1670 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1671 createmany -o $DIR/$tfile 20 > /dev/null
1672 unlinkmany $DIR/$tfile 20 > /dev/null
1673 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1674 do_facet mds "sysctl -w lustre.fail_val=10000"
1675 do_facet mds "sysctl -w lustre.fail_loc=0x8000050a"
1676 createmany -o $DIR/$tfile 20 > /dev/null
1677 unlinkmany $DIR/$tfile 20 > /dev/null
1678 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1679 do_facet mds "sysctl -w lustre.fail_loc=0"
1681 createmany -o $DIR/$tfile 20 > /dev/null
1682 unlinkmany $DIR/$tfile 20 > /dev/null
1683 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1684 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1685 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1686 echo "Current MDT timeout $CUR, worst $WORST"
1687 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1689 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1691 test_66b() #bug 3055
1693 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1695 at_start || return 0
1696 ORIG=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1697 sysctl -w lustre.fail_val=$(($ORIG + 5))
1698 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1699 sysctl -w lustre.fail_loc=0x50c
1700 ls $DIR/$tfile > /dev/null 2>&1
1701 sysctl -w lustre.fail_loc=0
1702 CUR=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1703 WORST=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $6}')
1704 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1705 [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
1707 run_test 66b "AT: verify net latency adjusts"
1709 test_67a() #bug 3055
1711 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1713 at_start || return 0
1714 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1715 # sleeping threads may drive values above this
1716 do_facet ost1 "sysctl -w lustre.fail_val=400"
1717 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1718 do_facet ost1 "sysctl -w lustre.fail_loc=0x50a"
1719 createmany -o $DIR/$tfile 20 > /dev/null
1720 unlinkmany $DIR/$tfile 20 > /dev/null
1721 do_facet ost1 "sysctl -w lustre.fail_loc=0"
1722 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1723 ATTEMPTS=$(($CONN2 - $CONN1))
1724 echo "$ATTEMPTS osc reconnect attempts on gradual slow"
1725 [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
1728 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1730 test_67b() #bug 3055
1732 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1734 at_start || return 0
1735 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1737 # exhaust precreations on ost1
1738 local OST=$(lfs osts | grep 0": " | awk '{print $2}' | sed -e 's/_UUID$//')
1739 local mdtosc=$(get_mdtosc_proc_path $OST)
1740 local last_id=$(do_facet mds lctl get_param -n osc.$mdtosc.prealloc_last_id)
1741 local next_id=$(do_facet mds lctl get_param -n osc.$mdtosc.prealloc_next_id)
1743 mkdir -p $DIR/$tdir/${OST}
1744 lfs setstripe $DIR/$tdir/${OST} -o 0 -c 1 || error "setstripe"
1745 echo "Creating to objid $last_id on ost $OST..."
1746 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1747 do_facet ost1 "sysctl -w lustre.fail_val=20000"
1748 do_facet ost1 "sysctl -w lustre.fail_loc=0x80000223"
1749 createmany -o $DIR/$tdir/${OST}/f $next_id $((last_id - next_id + 2))
1752 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1754 CONN2=$(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_nodesv $clients "set -x; MISSING_DBENCH_OK=$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 clients_up || error "client evicted: $?"
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 client_up || client_up || 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