6 ALWAYS_EXCEPT="24 27 52 $RECOVERY_SMALL_EXCEPT"
8 PTLDEBUG=${PTLDEBUG:--1}
9 LUSTRE=${LUSTRE:-`dirname $0`/..}
10 . $LUSTRE/tests/test-framework.sh
12 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
14 # also long tests: 19, 21a, 21e, 21f, 23, 27
15 # 1 2.5 2.5 4 4 (min)"
16 [ "$SLOW" = "no" ] && EXCEPT_SLOW="17 26a 26b 50 51 57"
20 # Allow us to override the setup if we already have a mounted system by
21 # setting SETUP=" " and CLEANUP=" "
23 CLEANUP=${CLEANUP:-""}
25 cleanup_and_setup_lustre
27 rm -rf $DIR/[df][0-9]*
29 SAMPLE_NAME=recovery-small.junk
30 SAMPLE_FILE=$TMP/$SAMPLE_NAME
31 # make this big, else test 9 doesn't wait for bulk -- bz 5595
32 dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=4
35 drop_request "mcreate $MOUNT/1" || return 1
36 drop_reint_reply "mcreate $MOUNT/2" || return 2
38 run_test 1 "mcreate: drop req, drop rep"
41 drop_request "tchmod 111 $MOUNT/2" || return 1
42 drop_reint_reply "tchmod 666 $MOUNT/2" || return 2
44 run_test 2 "chmod: drop req, drop rep"
47 drop_request "statone $MOUNT/2" || return 1
48 drop_reply "statone $MOUNT/2" || return 2
50 run_test 3 "stat: drop req, drop rep"
53 do_facet client "cp /etc/inittab $MOUNT/inittab" || return 1
54 drop_request "cat $MOUNT/inittab > /dev/null" || return 2
55 drop_reply "cat $MOUNT/inittab > /dev/null" || return 3
57 run_test 4 "open: drop req, drop rep"
60 drop_request "mv $MOUNT/inittab $MOUNT/renamed" || return 1
61 drop_reint_reply "mv $MOUNT/renamed $MOUNT/renamed-again" || return 2
62 do_facet client "checkstat -v $MOUNT/renamed-again" || return 3
64 run_test 5 "rename: drop req, drop rep"
66 [ ! -e $MOUNT/renamed-again ] && cp /etc/inittab $MOUNT/renamed-again
68 drop_request "mlink $MOUNT/renamed-again $MOUNT/link1" || return 1
69 drop_reint_reply "mlink $MOUNT/renamed-again $MOUNT/link2" || return 2
71 run_test 6 "link: drop req, drop rep"
73 [ ! -e $MOUNT/link1 ] && mlink $MOUNT/renamed-again $MOUNT/link1
74 [ ! -e $MOUNT/link2 ] && mlink $MOUNT/renamed-again $MOUNT/link2
76 drop_request "munlink $MOUNT/link1" || return 1
77 drop_reint_reply "munlink $MOUNT/link2" || return 2
79 run_test 7 "unlink: drop req, drop rep"
83 drop_reint_reply "touch $MOUNT/$tfile" || return 1
85 run_test 8 "touch: drop rep (bug 1423)"
89 pause_bulk "cp /etc/profile $MOUNT/$tfile" || return 1
90 do_facet client "cp ${SAMPLE_FILE} $MOUNT/${tfile}.2" || return 2
91 do_facet client "sync"
92 do_facet client "rm $MOUNT/$tfile $MOUNT/${tfile}.2" || return 3
94 run_test 9 "pause bulk on OST (bug 1420)"
98 do_facet client mcreate $MOUNT/$tfile || return 1
99 drop_bl_callback "chmod 0777 $MOUNT/$tfile" || echo "evicted as expected"
100 # wait for the mds to evict the client
101 #echo "sleep $(($TIMEOUT*2))"
102 #sleep $(($TIMEOUT*2))
103 do_facet client touch $MOUNT/$tfile || echo "touch failed, evicted"
104 do_facet client checkstat -v -p 0777 $MOUNT/$tfile || return 3
105 do_facet client "munlink $MOUNT/$tfile"
107 run_test 10 "finish request on server after client eviction (bug 1521)"
110 # wake up a thread waiting for completion after eviction
112 do_facet client multiop $MOUNT/$tfile Ow || return 1
113 do_facet client multiop $MOUNT/$tfile or || return 2
117 do_facet client multiop $MOUNT/$tfile or || return 3
118 drop_bl_callback multiop $MOUNT/$tfile Ow || echo "evicted as expected"
120 do_facet client munlink $MOUNT/$tfile || return 4
122 run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
126 $LCTL mark multiop $MOUNT/$tfile OS_c
127 do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
128 clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
129 multiop_bg_pause $MOUNT/$tfile OS_c || return 1
131 #define OBD_FAIL_MDS_CLOSE_NET 0x115
133 echo "waiting for multiop $PID"
134 wait $PID || return 2
135 do_facet client munlink $MOUNT/$tfile || return 3
137 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
139 # Bug 113, check that readdir lost recv timeout works.
141 mkdir $MOUNT/readdir || return 1
142 touch $MOUNT/readdir/newentry || return
143 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
144 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104"
145 ls $MOUNT/readdir || return 3
146 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
147 rm -rf $MOUNT/readdir || return 4
149 run_test 13 "mdc_readpage restart test (bug 1138)"
151 # Bug 113, check that readdir lost send timeout works.
154 touch $MOUNT/readdir/newentry
155 # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
156 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000106"
157 ls $MOUNT/readdir || return 1
158 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
160 run_test 14 "mdc_readpage resend test (bug 1138)"
163 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
164 touch $DIR/$tfile && return 1
167 run_test 15 "failed open (-ENOMEM)"
169 READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1`
171 lctl set_param -n llite.*.max_read_ahead_mb 0
175 lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD
179 do_facet client cp /etc/termcap $MOUNT
183 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
184 do_facet ost1 "lctl set_param fail_loc=0x80000504"
186 # OST bulk will time out here, client resends
187 do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 1
188 do_facet ost1 lctl set_param fail_loc=0
189 # give recovery a chance to finish (shouldn't take long)
191 do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 2
194 run_test 16 "timeout bulk put, don't evict client (2732)"
199 # With adaptive timeouts, bulk_get won't expire until adaptive_timeout_max
200 if at_is_valid && at_is_enabled; then
201 at_max_saved=$(at_max_get ost1)
202 at_max_set $TIMEOUT ost1
205 # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
206 # OST bulk will time out here, client retries
207 do_facet ost1 lctl set_param fail_loc=0x80000503
208 # need to ensure we send an RPC
209 do_facet client cp /etc/termcap $DIR/$tfile
212 # with AT, client will wait adaptive_max*factor+net_latency before
213 # expiring the req, hopefully timeout*2 is enough
214 sleep $(($TIMEOUT*2))
216 do_facet ost1 lctl set_param fail_loc=0
217 do_facet client "df $DIR"
218 # expect cmp to succeed, client resent bulk
219 do_facet client "cmp /etc/termcap $DIR/$tfile" || return 3
220 do_facet client "rm $DIR/$tfile" || return 4
221 [ $at_max_saved -ne 0 ] && $(at_max_set $at_max_saved ost1)
224 run_test 17 "timeout bulk get, don't evict client (2732)"
227 [ -z ${ost2_svc} ] && skip "needs 2 osts" && return 0
229 do_facet client mkdir -p $MOUNT/$tdir
230 f=$MOUNT/$tdir/$tfile
233 pgcache_empty || return 1
236 lfs setstripe $f -s $((128 * 1024)) -i 1 -c 1
238 do_facet client cp $SAMPLE_FILE $f
240 local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | egrep -v 'MDT' | awk '{print $1}'`
241 $LCTL --device $osc2dev deactivate || return 3
242 # my understanding is that there should be nothing in the page
243 # cache after the client reconnects?
245 pgcache_empty || rc=2
246 $LCTL --device $osc2dev activate
250 run_test 18a "manual ost invalidate clears page cache immediately"
253 do_facet client mkdir -p $MOUNT/$tdir
254 f=$MOUNT/$tdir/$tfile
255 f2=$MOUNT/$tdir/${tfile}-2
258 pgcache_empty || return 1
260 # shouldn't have to set stripe size of count==1
261 lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1
262 lfs setstripe $f2 -s $((128 * 1024)) -i 0 -c 1
264 do_facet client cp $SAMPLE_FILE $f
267 # allow recovery to complete
268 sleep $((TIMEOUT + 2))
269 # my understanding is that there should be nothing in the page
270 # cache after the client reconnects?
272 pgcache_empty || rc=2
276 run_test 18b "eviction and reconnect clears page cache (2766)"
279 do_facet client mkdir -p $MOUNT/$tdir
280 f=$MOUNT/$tdir/$tfile
281 f2=$MOUNT/$tdir/${tfile}-2
284 pgcache_empty || return 1
286 # shouldn't have to set stripe size of count==1
287 lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1
288 lfs setstripe $f2 -s $((128 * 1024)) -i 0 -c 1
290 do_facet client cp $SAMPLE_FILE $f
294 # OBD_FAIL_OST_CONNECT_NET2
295 # lost reply to connect request
296 do_facet ost1 lctl set_param fail_loc=0x80000225
298 df $MOUNT > /dev/null 2>&1
300 # my understanding is that there should be nothing in the page
301 # cache after the client reconnects?
303 pgcache_empty || rc=2
307 run_test 18c "Dropped connect reply after eviction handing (14755)"
311 do_facet client mcreate $f || return 1
312 drop_ldlm_cancel "chmod 0777 $f" || echo "evicted as expected"
314 do_facet client checkstat -v -p 0777 $f || echo evicted
315 # let the client reconnect
317 do_facet client "munlink $f"
319 run_test 19a "test expired_lock_main on mds (2867)"
323 do_facet client multiop $f Ow || return 1
324 do_facet client multiop $f or || return 2
328 do_facet client multiop $f or || return 3
329 drop_ldlm_cancel multiop $f Ow || echo "client evicted, as expected"
331 do_facet client munlink $f || return 4
333 run_test 19b "test expired_lock_main on ost (2867)"
335 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
337 lfs setstripe $DIR/$tdir/${tfile} -i 0 -c 1
338 multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
341 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
342 do_facet ost1 lctl set_param fail_loc=0x80000308
343 kill -USR1 $MULTI_PID
346 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
348 run_test 20a "ldlm_handle_enqueue error (should return error)"
350 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
352 lfs setstripe $DIR/$tdir/${tfile} -i 0 -c 1
354 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
355 do_facet ost1 lctl set_param fail_loc=0x80000308
356 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
357 error "didn't fail open enqueue" || true
359 run_test 20b "ldlm_handle_enqueue error (should return error)"
362 mkdir -p $DIR/$tdir-1
363 mkdir -p $DIR/$tdir-2
364 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
367 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
368 multiop $DIR/$tdir-2/f Oc &
371 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
373 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
374 kill -USR1 $close_pid
376 wait $close_pid || return 1
377 wait $open_pid || return 2
378 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
380 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
381 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
385 run_test 21a "drop close request while close and open are both in flight"
388 mkdir -p $DIR/$tdir-1
389 mkdir -p $DIR/$tdir-2
390 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
393 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
394 mcreate $DIR/$tdir-2/f &
397 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
399 kill -USR1 $close_pid
401 wait $close_pid || return 1
402 wait $open_pid || return 3
404 $CHECKSTAT -t file $DIR/$tdir-1/f || return 4
405 $CHECKSTAT -t file $DIR/$tdir-2/f || return 5
408 run_test 21b "drop open request while close and open are both in flight"
411 mkdir -p $DIR/$tdir-1
412 mkdir -p $DIR/$tdir-2
413 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
416 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
417 mcreate $DIR/$tdir-2/f &
420 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
422 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
423 kill -USR1 $close_pid
425 wait $close_pid || return 1
426 wait $open_pid || return 2
428 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
430 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
431 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
434 run_test 21c "drop both request while close and open are both in flight"
437 mkdir -p $DIR/$tdir-1
438 mkdir -p $DIR/$tdir-2
439 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
442 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
443 multiop $DIR/$tdir-2/f Oc &
445 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
447 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
450 wait $pid || return 1
451 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
453 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
454 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
458 run_test 21d "drop close reply while close and open are both in flight"
461 mkdir -p $DIR/$tdir-1
462 mkdir -p $DIR/$tdir-2
463 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
466 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
467 touch $DIR/$tdir-2/f &
469 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
473 wait $pid || return 1
476 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
477 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
480 run_test 21e "drop open reply while close and open are both in flight"
483 mkdir -p $DIR/$tdir-1
484 mkdir -p $DIR/$tdir-2
485 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
488 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
489 touch $DIR/$tdir-2/f &
491 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
493 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
496 wait $pid || return 1
497 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
499 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
500 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
503 run_test 21f "drop both reply while close and open are both in flight"
506 mkdir -p $DIR/$tdir-1
507 mkdir -p $DIR/$tdir-2
508 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
511 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
512 touch $DIR/$tdir-2/f &
514 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
516 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
519 wait $pid || return 1
520 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
522 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
523 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
526 run_test 21g "drop open reply and close request while close and open are both in flight"
529 mkdir -p $DIR/$tdir-1
530 mkdir -p $DIR/$tdir-2
531 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
534 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
535 touch $DIR/$tdir-2/f &
538 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
540 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
543 wait $pid || return 1
544 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
546 wait $touch_pid || return 2
548 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
549 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
552 run_test 21h "drop open request and close reply while close and open are both in flight"
554 # bug 3462 - multiple MDC requests
559 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
564 multiop $f1 msu || return 1
567 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
569 wait $close_pid || return 2
570 rm -rf $f2 || return 4
572 run_test 22 "drop close request and do mknod"
575 multiop_bg_pause $DIR/$tfile O_c || return 1
577 # give a chance for open
581 drop_request "kill -USR1 $pid"
584 wait $pid || return 1
587 run_test 23 "client hang when close a file after mds crash"
589 test_24() { # bug 11710 details correct fsync() behavior
591 lfs setstripe $DIR/$tdir -s 0 -i 0 -c 1
593 multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
596 kill -USR1 $MULTI_PID
599 lctl set_param fail_loc=0x0
601 [ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true
603 run_test 24 "fsync error (should return error)"
605 test_26a() { # was test_26 bug 5921 - evict dead exports by pinger
606 # this test can only run from a client on a separate node.
607 remote_ost || { skip "local OST" && return 0; }
608 remote_mds || { skip "local MDS" && return 0; }
609 OST_FILE=obdfilter.${ost1_svc}.num_exports
610 OST_EXP="`do_facet ost1 lctl get_param -n $OST_FILE`"
611 OST_NEXP1=`echo $OST_EXP | cut -d' ' -f2`
612 echo starting with $OST_NEXP1 OST exports
613 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
614 do_facet client lctl set_param fail_loc=0x505
615 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
616 # But if there's a race to start the evictor from various obds,
617 # the loser might have to wait for the next ping.
618 echo Waiting for $(($TIMEOUT * 8)) secs
619 sleep $(($TIMEOUT * 8))
620 OST_EXP="`do_facet ost1 lctl get_param -n $OST_FILE`"
621 OST_NEXP2=`echo $OST_EXP | cut -d' ' -f2`
622 echo ending with $OST_NEXP2 OST exports
623 do_facet client lctl set_param fail_loc=0x0
624 [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted"
627 run_test 26a "evict dead exports"
629 test_26b() { # bug 10140 - evict dead exports by pinger
631 zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
632 sleep 1 # wait connections being established
633 MDS_FILE=mdt.${mds1_svc}.num_exports
634 MDS_NEXP1="`do_facet $SINGLEMDS lctl get_param -n $MDS_FILE | cut -d' ' -f2`"
635 OST_FILE=obdfilter.${ost1_svc}.num_exports
636 OST_NEXP1="`do_facet ost1 lctl get_param -n $OST_FILE | cut -d' ' -f2`"
637 echo starting with $OST_NEXP1 OST and $MDS_NEXP1 MDS exports
638 zconf_umount `hostname` $MOUNT2 -f
639 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
640 # But if there's a race to start the evictor from various obds,
641 # the loser might have to wait for the next ping.
642 echo Waiting for $(($TIMEOUT * 3)) secs
643 sleep $(($TIMEOUT * 3))
644 OST_NEXP2="`do_facet ost1 lctl get_param -n $OST_FILE | cut -d' ' -f2`"
645 MDS_NEXP2="`do_facet $SINGLEMDS lctl get_param -n $MDS_FILE | cut -d' ' -f2`"
646 echo ending with $OST_NEXP2 OST and $MDS_NEXP2 MDS exports
647 [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted from OST"
648 [ $MDS_NEXP1 -le $MDS_NEXP2 ] && error "client not evicted from MDS"
651 run_test 26b "evict dead exports"
654 remote_mds && { skip "remote MDS" && return 0; }
656 writemany -q -a $DIR/$tdir/$tfile 0 5 &
660 facet_failover $SINGLEMDS
661 #define OBD_FAIL_OSC_SHUTDOWN 0x407
662 lctl set_param fail_loc=0x80000407
663 # need to wait for reconnect
664 echo -n waiting for fail_loc
665 while [ `lctl get_param -n fail_loc` -eq -2147482617 ]; do
669 facet_failover $SINGLEMDS
671 kill -USR1 $CLIENT_PID
675 run_test 27 "fail LOV while using OSC's"
677 test_28() { # bug 6086 - error adding new clients
678 do_facet client mcreate $MOUNT/$tfile || return 1
679 drop_bl_callback "chmod 0777 $MOUNT/$tfile" ||echo "evicted as expected"
680 #define OBD_FAIL_MDS_CLIENT_ADD 0x12f
681 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
682 # fail once (evicted), reconnect fail (fail_loc), ok
683 df || (sleep 10; df) || (sleep 10; df) || error "reconnect failed"
685 fail $SINGLEMDS # verify MDS last_rcvd can be loaded
687 run_test 28 "handle error adding new clients (bug 6086)"
691 # put a load of file creates/writes/deletes
692 writemany -q $DIR/$tdir/$tfile 0 5 &
694 echo writemany pid $CLIENT_PID
698 # wait for client to reconnect to MDS
703 # client process should see no problems even though MDS went down
705 kill -USR1 $CLIENT_PID
708 echo writemany returned $rc
709 #these may fail because of eviction due to slow AST response.
710 [ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
712 run_test 50 "failover MDS under load"
716 # put a load of file creates/writes/deletes
717 writemany -q $DIR/$tdir/$tfile 0 5 &
721 facet_failover $SINGLEMDS
722 # failover at various points during recovery
723 SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
724 echo will failover at $SEQ
727 echo failover in $i sec
729 facet_failover $SINGLEMDS
731 # client process should see no problems even though MDS went down
732 # and recovery was interrupted
734 kill -USR1 $CLIENT_PID
737 echo writemany returned $rc
738 [ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
740 run_test 51 "failover MDS during recovery"
743 do_facet client "mkdir -p $DIR/$tdir"
744 do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
746 echo writemany pid $CLIENT_PID
751 wait $CLIENT_PID || rc=$?
752 # active client process should see an EIO for down OST
753 [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
754 # but timing or failover setup may allow success
755 [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
756 echo "writemany returned $rc"
764 [ $rc -ne 0 ] && { return $rc; }
765 # wait for client to reconnect to OST
769 [ $rc -ne 0 ] && { return $rc; }
776 run_test 52 "failover OST under load"
778 # test of open reconstruct
781 drop_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
784 run_test 53 "touch: drop rep"
787 zconf_mount `hostname` $MOUNT2
791 cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
794 ERROR=`dmesg | egrep "(test 54|went back in time)" | tail -n1 | grep "went back in time"`
795 [ x"$ERROR" == x ] || error "back in time occured"
797 run_test 54 "back in time"
799 # bug 11330 - liblustre application death during I/O locks up OST
801 remote_ost && { skip "remote OST" && return 0; }
805 # first dd should be finished quickly
806 dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
809 echo "step1: testing ......"
811 if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
813 if [ $count -gt 64 ]; then
814 error "dd should be finished!"
818 echo "(dd_pid=$DDPID, time=$count)successful"
820 #define OBD_FAIL_OST_DROP_REQ 0x21d
821 do_facet ost lctl set_param fail_loc=0x0000021d
822 # second dd will be never finished
823 dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 &
826 echo "step2: testing ......"
827 while [ $count -le 64 ]; do
828 dd_name="`ps x | awk '$1 == '$DDPID' { print $5 }'`"
829 if [ -z $dd_name ]; then
831 echo "debug: (dd_name=$dd_name, dd_pid=$DDPID, time=$count)"
832 error "dd shouldn't be finished!"
837 echo "(dd_pid=$DDPID, time=$count)successful"
839 #Recover fail_loc and dd will finish soon
840 do_facet ost lctl set_param fail_loc=0
842 echo "step3: testing ......"
844 if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
846 if [ $count -gt 500 ]; then
847 error "dd should be finished!"
851 echo "(dd_pid=$DDPID, time=$count)successful"
855 run_test 55 "ost_brw_read/write drops timed-out read/write request"
857 test_56() { # b=11277
858 #define OBD_FAIL_MDS_RESEND 0x136
860 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
862 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
865 run_test 56 "do not allow reconnect to busy exports"
868 # no oscs means no client or mdt
869 while lctl get_param osc.*.* > /dev/null 2>&1; do
870 : # loop until proc file is removed
874 test_57() { # bug 10866
878 #define OBD_FAIL_LPROC_REMOVE 0xB00
879 lctl set_param fail_loc=0x80000B00
880 zconf_umount `hostname` $DIR
881 lctl set_param fail_loc=0x80000B00
882 fail_abort $SINGLEMDS
884 lctl set_param fail_loc=0
886 do_facet client "df $DIR"
888 run_test 57 "read procfs entries causes kernel crash"
890 test_58() { # bug 11546
891 #define OBD_FAIL_MDC_ENQUEUE_PAUSE 0x801
894 lctl set_param fail_loc=0x80000801
895 cp $MOUNT/$tfile /dev/null &
898 lctl set_param fail_loc=0
899 drop_bl_callback rm -f $MOUNT/$tfile
901 do_facet client "df $DIR"
903 run_test 58 "Eviction in the middle of open RPC reply processing"
905 test_59() { # bug 10589
906 zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
907 echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
908 #define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311
909 lctl set_param fail_loc=0x311
910 writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
911 [ $? = 0 ] || error "dd write failed"
912 writes=$(echo $writes | awk -F '+' '/out/ {print $1}')
913 lctl set_param fail_loc=0
915 zconf_umount `hostname` $MOUNT2 -f
916 reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
917 [ $? = 0 ] || error "dd read failed"
918 reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
919 [ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
921 run_test 59 "Read cancel race on client eviction"
923 equals_msg `basename $0`: test complete, cleaning up
924 check_and_cleanup_lustre
925 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG || true