6 ALWAYS_EXCEPT="24 52 $RECOVERY_SMALL_EXCEPT"
8 PTLDEBUG=${PTLDEBUG:--1}
9 LUSTRE=${LUSTRE:-`dirname $0`/..}
10 . $LUSTRE/tests/test-framework.sh
12 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
15 require_dsh_mds || exit 0
17 # also long tests: 19, 21a, 21e, 21f, 23, 27
18 # 1 2.5 2.5 4 4 (min)"
19 [ "$SLOW" = "no" ] && EXCEPT_SLOW="17 26a 26b 50 51 57"
23 # Allow us to override the setup if we already have a mounted system by
24 # setting SETUP=" " and CLEANUP=" "
26 CLEANUP=${CLEANUP:-""}
28 check_and_setup_lustre
31 rm -rf $DIR/[df][0-9]*
34 drop_request "mcreate $DIR/f1" || return 1
35 drop_reint_reply "mcreate $DIR/f2" || return 2
37 run_test 1 "mcreate: drop req, drop rep"
40 drop_request "tchmod 111 $DIR/f2" || return 1
41 drop_reint_reply "tchmod 666 $DIR/f2" || return 2
43 run_test 2 "chmod: drop req, drop rep"
46 drop_request "statone $DIR/f2" || return 1
47 drop_reply "statone $DIR/f2" || return 2
49 run_test 3 "stat: drop req, drop rep"
51 SAMPLE_NAME=f0.recovery-small.junk
52 SAMPLE_FILE=$TMP/$SAMPLE_NAME
53 # make this big, else test 9 doesn't wait for bulk -- bz 5595
54 dd if=/dev/urandom of=$SAMPLE_FILE bs=1M count=4
57 do_facet client "cp $SAMPLE_FILE $DIR/$SAMPLE_NAME" || return 1
58 drop_request "cat $DIR/$SAMPLE_NAME > /dev/null" || return 2
59 drop_reply "cat $DIR/$SAMPLE_NAME > /dev/null" || return 3
61 run_test 4 "open: drop req, drop rep"
63 RENAMED_AGAIN=$DIR/f0.renamed-again
66 drop_request "mv $DIR/$SAMPLE_NAME $DIR/$tfile-renamed" || return 1
67 drop_reint_reply "mv $DIR/$tfile-renamed $RENAMED_AGAIN" || return 2
68 do_facet client "checkstat -v $RENAMED_AGAIN" || return 3
70 run_test 5 "rename: drop req, drop rep"
72 [ ! -e $RENAMED_AGAIN ] && cp $SAMPLE_FILE $RENAMED_AGAIN
77 drop_request "mlink $RENAMED_AGAIN $LINK1" || return 1
78 drop_reint_reply "mlink $RENAMED_AGAIN $LINK2" || return 2
80 run_test 6 "link: drop req, drop rep"
82 [ ! -e $LINK1 ] && mlink $RENAMED_AGAIN $LINK1
83 [ ! -e $LINK2 ] && mlink $RENAMED_AGAIN $LINK2
85 drop_request "munlink $LINK1" || return 1
86 drop_reint_reply "munlink $LINK2" || return 2
88 run_test 7 "unlink: drop req, drop rep"
92 drop_reint_reply "touch $DIR/$tfile" || return 1
94 run_test 8 "touch: drop rep (bug 1423)"
98 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
100 pause_bulk "cp /etc/profile $DIR/$tfile" || return 1
101 do_facet client "cp $SAMPLE_FILE $DIR/${tfile}.2" || return 2
102 do_facet client "sync"
103 do_facet client "rm $DIR/$tfile $DIR/${tfile}.2" || return 3
105 run_test 9 "pause bulk on OST (bug 1420)"
109 do_facet client mcreate $DIR/$tfile || return 1
110 drop_bl_callback "chmod 0777 $DIR/$tfile" || echo "evicted as expected"
111 # wait for the mds to evict the client
112 #echo "sleep $(($TIMEOUT*2))"
113 #sleep $(($TIMEOUT*2))
114 do_facet client touch $DIR/$tfile || echo "touch failed, evicted"
115 do_facet client checkstat -v -p 0777 $DIR/$tfile || return 3
116 do_facet client "munlink $DIR/$tfile"
118 run_test 10 "finish request on server after client eviction (bug 1521)"
121 # wake up a thread waiting for completion after eviction
123 do_facet client multiop $DIR/$tfile Ow || return 1
124 do_facet client multiop $DIR/$tfile or || return 2
128 do_facet client multiop $DIR/$tfile or || return 3
129 drop_bl_callback multiop $DIR/$tfile Ow || echo "evicted as expected"
131 do_facet client munlink $DIR/$tfile || return 4
133 run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
137 $LCTL mark multiop $DIR/$tfile OS_c
138 do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
139 clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
140 multiop_bg_pause $DIR/$tfile OS_c || return 1
142 #define OBD_FAIL_MDS_CLOSE_NET 0x115
144 echo "waiting for multiop $PID"
145 wait $PID || return 2
146 do_facet client munlink $DIR/$tfile || return 3
148 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
150 # Bug 113, check that readdir lost recv timeout works.
152 mkdir -p $DIR/$tdir || return 1
153 touch $DIR/$tdir/newentry || return
154 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
155 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104"
156 ls $DIR/$tdir || return 3
157 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
158 rm -rf $DIR/$tdir || return 4
160 run_test 13 "mdc_readpage restart test (bug 1138)"
162 # Bug 113, check that readdir lost send timeout works.
165 touch $DIR/$tdir/newentry
166 # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
167 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000106"
168 ls $DIR/$tdir || return 1
169 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
171 run_test 14 "mdc_readpage resend test (bug 1138)"
174 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
175 touch $DIR/$tfile && return 1
178 run_test 15 "failed open (-ENOMEM)"
180 READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1`
182 lctl set_param -n llite.*.max_read_ahead_mb 0
186 lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD
190 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
192 do_facet client cp $SAMPLE_FILE $DIR
196 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
197 do_facet ost1 "lctl set_param fail_loc=0x80000504"
199 # OST bulk will time out here, client resends
200 do_facet client "cmp $SAMPLE_FILE $DIR/${SAMPLE_FILE##*/}" || return 1
201 do_facet ost1 lctl set_param fail_loc=0
202 # give recovery a chance to finish (shouldn't take long)
204 do_facet client "cmp $SAMPLE_FILE $DIR/${SAMPLE_FILE##*/}" || return 2
207 run_test 16 "timeout bulk put, don't evict client (2732)"
212 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
214 # With adaptive timeouts, bulk_get won't expire until adaptive_timeout_max
215 if at_is_enabled; then
216 at_max_saved=$(at_max_get ost1)
217 at_max_set $TIMEOUT ost1
220 # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
221 # OST bulk will time out here, client retries
222 do_facet ost1 lctl set_param fail_loc=0x80000503
223 # need to ensure we send an RPC
224 do_facet client cp $SAMPLE_FILE $DIR/$tfile
227 # with AT, client will wait adaptive_max*factor+net_latency before
228 # expiring the req, hopefully timeout*2 is enough
229 sleep $(($TIMEOUT*2))
231 do_facet ost1 lctl set_param fail_loc=0
232 do_facet client "df $DIR"
233 # expect cmp to succeed, client resent bulk
234 do_facet client "cmp $SAMPLE_FILE $DIR/$tfile" || return 3
235 do_facet client "rm $DIR/$tfile" || return 4
236 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved ost1
239 run_test 17 "timeout bulk get, don't evict client (2732)"
242 [ -z ${ost2_svc} ] && skip_env "needs 2 osts" && return 0
244 do_facet client mkdir -p $DIR/$tdir
248 pgcache_empty || return 1
251 $LFS setstripe -i 1 -c 1 $f
252 stripe_index=$($LFS getstripe -i $f)
253 if [ $stripe_index -ne 1 ]; then
255 error "$f: stripe_index $stripe_index != 1" && return
258 do_facet client cp $SAMPLE_FILE $f
260 local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | egrep -v 'MDT' | awk '{print $1}'`
261 $LCTL --device $osc2dev deactivate || return 3
262 # my understanding is that there should be nothing in the page
263 # cache after the client reconnects?
265 pgcache_empty || rc=2
266 $LCTL --device $osc2dev activate
270 run_test 18a "manual ost invalidate clears page cache immediately"
273 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
275 do_facet client mkdir -p $DIR/$tdir
279 pgcache_empty || return 1
281 $LFS setstripe -i 0 -c 1 $f
282 stripe_index=$($LFS getstripe -i $f)
283 if [ $stripe_index -ne 0 ]; then
285 error "$f: stripe_index $stripe_index != 0" && return
288 do_facet client cp $SAMPLE_FILE $f
291 # allow recovery to complete
292 sleep $((TIMEOUT + 2))
293 # my understanding is that there should be nothing in the page
294 # cache after the client reconnects?
296 pgcache_empty || rc=2
300 run_test 18b "eviction and reconnect clears page cache (2766)"
303 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
305 do_facet client mkdir -p $DIR/$tdir
309 pgcache_empty || return 1
311 $LFS setstripe -i 0 -c 1 $f
312 stripe_index=$($LFS getstripe -i $f)
313 if [ $stripe_index -ne 0 ]; then
315 error "$f: stripe_index $stripe_index != 0" && return
318 do_facet client cp $SAMPLE_FILE $f
322 # OBD_FAIL_OST_CONNECT_NET2
323 # lost reply to connect request
324 do_facet ost1 lctl set_param fail_loc=0x80000225
327 df $MOUNT > /dev/null 2>&1
329 # my understanding is that there should be nothing in the page
330 # cache after the client reconnects?
332 pgcache_empty || rc=2
336 run_test 18c "Dropped connect reply after eviction handing (14755)"
340 do_facet client mcreate $f || return 1
341 drop_ldlm_cancel "chmod 0777 $f" || echo "evicted as expected"
343 do_facet client checkstat -v -p 0777 $f || echo evicted
344 # let the client reconnect
346 do_facet client "munlink $f"
348 run_test 19a "test expired_lock_main on mds (2867)"
352 do_facet client multiop $f Ow || return 1
353 do_facet client multiop $f or || return 2
357 do_facet client multiop $f or || return 3
358 drop_ldlm_cancel multiop $f Ow || echo "client evicted, as expected"
360 do_facet client munlink $f || return 4
362 run_test 19b "test expired_lock_main on ost (2867)"
364 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
365 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
368 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
369 multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
372 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
373 do_facet ost1 lctl set_param fail_loc=0x80000308
374 kill -USR1 $MULTI_PID
377 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
379 run_test 20a "ldlm_handle_enqueue error (should return error)"
381 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
382 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
385 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
387 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
388 do_facet ost1 lctl set_param fail_loc=0x80000308
389 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
390 error "didn't fail open enqueue" || true
392 run_test 20b "ldlm_handle_enqueue error (should return error)"
395 mkdir -p $DIR/$tdir-1
396 mkdir -p $DIR/$tdir-2
397 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
400 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
401 multiop $DIR/$tdir-2/f Oc &
404 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
406 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
407 kill -USR1 $close_pid
409 wait $close_pid || return 1
410 wait $open_pid || return 2
411 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
413 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
414 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
418 run_test 21a "drop close request while close and open are both in flight"
421 mkdir -p $DIR/$tdir-1
422 mkdir -p $DIR/$tdir-2
423 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
426 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
427 mcreate $DIR/$tdir-2/f &
430 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
432 kill -USR1 $close_pid
434 wait $close_pid || return 1
435 wait $open_pid || return 3
437 $CHECKSTAT -t file $DIR/$tdir-1/f || return 4
438 $CHECKSTAT -t file $DIR/$tdir-2/f || return 5
441 run_test 21b "drop open request while close and open are both in flight"
444 mkdir -p $DIR/$tdir-1
445 mkdir -p $DIR/$tdir-2
446 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
449 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
450 mcreate $DIR/$tdir-2/f &
453 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
455 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
456 kill -USR1 $close_pid
458 wait $close_pid || return 1
459 wait $open_pid || return 2
461 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
463 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
464 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
467 run_test 21c "drop both request while close and open are both in flight"
470 mkdir -p $DIR/$tdir-1
471 mkdir -p $DIR/$tdir-2
472 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
475 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
476 multiop $DIR/$tdir-2/f Oc &
478 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
480 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
483 wait $pid || return 1
484 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
486 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
487 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
491 run_test 21d "drop close reply while close and open are both in flight"
494 mkdir -p $DIR/$tdir-1
495 mkdir -p $DIR/$tdir-2
496 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
499 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
500 touch $DIR/$tdir-2/f &
502 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
506 wait $pid || return 1
509 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
510 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
513 run_test 21e "drop open reply while close and open are both in flight"
516 mkdir -p $DIR/$tdir-1
517 mkdir -p $DIR/$tdir-2
518 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
521 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
522 touch $DIR/$tdir-2/f &
524 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
526 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
529 wait $pid || return 1
530 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
532 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
533 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
536 run_test 21f "drop both reply while close and open are both in flight"
539 mkdir -p $DIR/$tdir-1
540 mkdir -p $DIR/$tdir-2
541 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
544 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
545 touch $DIR/$tdir-2/f &
547 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
549 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
552 wait $pid || return 1
553 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
555 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
556 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
559 run_test 21g "drop open reply and close request while close and open are both in flight"
562 mkdir -p $DIR/$tdir-1
563 mkdir -p $DIR/$tdir-2
564 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
567 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
568 touch $DIR/$tdir-2/f &
571 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
573 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
576 wait $pid || return 1
577 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
579 wait $touch_pid || return 2
581 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
582 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
585 run_test 21h "drop open request and close reply while close and open are both in flight"
587 # bug 3462 - multiple MDC requests
592 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
597 multiop $f1 msu || return 1
600 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
602 wait $close_pid || return 2
603 rm -rf $f2 || return 4
605 run_test 22 "drop close request and do mknod"
608 multiop_bg_pause $DIR/$tfile O_c || return 1
610 # give a chance for open
614 drop_request "kill -USR1 $pid"
617 wait $pid || return 1
620 run_test 23 "client hang when close a file after mds crash"
622 test_24() { # bug 11710 details correct fsync() behavior
623 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
626 $LFS setstripe -i 0 -c 1 $DIR/$tdir
628 multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
631 kill -USR1 $MULTI_PID
634 lctl set_param fail_loc=0x0
636 [ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true
638 run_test 24 "fsync error (should return error)"
640 wait_client_evicted () {
643 local varsvc=${facet}_svc
645 wait_update $(facet_active_host $facet) \
646 "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \
650 test_26a() { # was test_26 bug 5921 - evict dead exports by pinger
651 # this test can only run from a client on a separate node.
652 remote_ost || { skip "local OST" && return 0; }
653 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
654 remote_mds || { skip "local MDS" && return 0; }
656 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
657 skip "msg and ost1 are at the same node"
661 check_timeout || return 1
663 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
665 echo starting with $OST_NEXP OST exports
666 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
667 do_facet client lctl set_param fail_loc=0x505
668 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
669 # But if there's a race to start the evictor from various obds,
670 # the loser might have to wait for the next ping.
673 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4))
675 do_facet client lctl set_param fail_loc=0x0
676 [ $rc -eq 0 ] || error "client not evicted from OST"
678 run_test 26a "evict dead exports"
680 test_26b() { # bug 10140 - evict dead exports by pinger
681 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
683 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
684 skip "msg and ost1 are at the same node"
688 check_timeout || return 1
690 zconf_mount `hostname` $MOUNT2 ||
691 { error "Failed to mount $MOUNT2"; return 2; }
692 sleep 1 # wait connections being established
694 local MDS_NEXP=$(do_facet $SINGLEMDS lctl get_param -n mdt.${mds1_svc}.num_exports | cut -d' ' -f2)
695 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
697 echo starting with $OST_NEXP OST and $MDS_NEXP MDS exports
699 zconf_umount `hostname` $MOUNT2 -f
701 # PING_INTERVAL max(obd_timeout / 4, 1U)
702 # PING_EVICT_TIMEOUT (PING_INTERVAL * 6)
704 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
705 # But if there's a race to start the evictor from various obds,
706 # the loser might have to wait for the next ping.
707 # = 9 * PING_INTERVAL + PING_INTERVAL
708 # = 10 PING_INTERVAL = 10 obd_timeout / 4 = 2.5 obd_timeout
709 # let's wait $((TIMEOUT * 3)) # bug 19887
711 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 3)) || \
712 error "Client was not evicted by ost" rc=1
713 wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 3)) || \
714 error "Client was not evicted by mds"
716 run_test 26b "evict dead exports"
720 writemany -q -a $DIR/$tdir/$tfile 0 5 &
723 local save_FAILURE_MODE=$FAILURE_MODE
725 facet_failover $SINGLEMDS
726 #define OBD_FAIL_OSC_SHUTDOWN 0x407
727 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000407
728 # need to wait for reconnect
729 echo waiting for fail_loc
730 wait_update_facet $SINGLEMDS "lctl get_param -n fail_loc" "-2147482617"
731 facet_failover $SINGLEMDS
733 kill -USR1 $CLIENT_PID
736 FAILURE_MODE=$save_FAILURE_MODE
738 run_test 27 "fail LOV while using OSC's"
740 test_28() { # bug 6086 - error adding new clients
741 do_facet client mcreate $DIR/$tfile || return 1
742 drop_bl_callback "chmod 0777 $DIR/$tfile" ||echo "evicted as expected"
743 #define OBD_FAIL_MDS_CLIENT_ADD 0x12f
744 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
745 # fail once (evicted), reconnect fail (fail_loc), ok
746 client_up || (sleep 10; client_up) || (sleep 10; client_up) || error "reconnect failed"
748 fail $SINGLEMDS # verify MDS last_rcvd can be loaded
750 run_test 28 "handle error adding new clients (bug 6086)"
752 test_29a() { # bug 22273 - error adding new clients
753 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
754 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000711"
755 # fail abort so client will be new again
756 fail_abort $SINGLEMDS
757 client_up || error "reconnect failed"
760 run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
762 test_29b() { # bug 22273 - error adding new clients
763 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
764 do_facet ost1 "lctl set_param fail_loc=0x80000711"
765 # fail abort so client will be new again
767 client_up || error "reconnect failed"
770 run_test 29b "error adding new clients doesn't cause LBUG (bug 22273)"
774 # put a load of file creates/writes/deletes
775 writemany -q $DIR/$tdir/$tfile 0 5 &
777 echo writemany pid $CLIENT_PID
781 # wait for client to reconnect to MDS
786 # client process should see no problems even though MDS went down
788 kill -USR1 $CLIENT_PID
791 echo writemany returned $rc
792 #these may fail because of eviction due to slow AST response.
793 [ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
795 run_test 50 "failover MDS under load"
798 #define OBD_FAIL_MDS_SYNC_CAPA_SL 0x1310
799 do_facet ost1 lctl set_param fail_loc=0x00001310
802 # put a load of file creates/writes/deletes
803 writemany -q $DIR/$tdir/$tfile 0 5 &
807 facet_failover $SINGLEMDS
808 # failover at various points during recovery
809 SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
810 echo will failover at $SEQ
813 echo failover in $i sec
815 facet_failover $SINGLEMDS
817 # client process should see no problems even though MDS went down
818 # and recovery was interrupted
820 kill -USR1 $CLIENT_PID
823 echo writemany returned $rc
824 [ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
826 run_test 51 "failover MDS during recovery"
829 do_facet client "mkdir -p $DIR/$tdir"
830 do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
832 echo writemany pid $CLIENT_PID
837 wait $CLIENT_PID || rc=$?
838 # active client process should see an EIO for down OST
839 [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
840 # but timing or failover setup may allow success
841 [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
842 echo "writemany returned $rc"
847 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
852 [ $rc -ne 0 ] && { return $rc; }
853 # wait for client to reconnect to OST
857 [ $rc -ne 0 ] && { return $rc; }
864 run_test 52 "failover OST under load"
866 # test of open reconstruct
869 drop_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
872 run_test 53 "touch: drop rep"
875 zconf_mount `hostname` $MOUNT2
879 cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
882 ERROR=`dmesg | egrep "(test 54|went back in time)" | tail -n1 | grep "went back in time"`
883 [ x"$ERROR" == x ] || error "back in time occured"
885 run_test 54 "back in time"
887 # bug 11330 - liblustre application death during I/O locks up OST
889 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
893 # first dd should be finished quickly
894 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
895 dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
898 echo "step1: testing ......"
900 if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
902 if [ $count -gt 64 ]; then
903 error "dd should be finished!"
907 echo "(dd_pid=$DDPID, time=$count)successful"
909 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2
910 #define OBD_FAIL_OST_DROP_REQ 0x21d
911 do_facet ost1 lctl set_param fail_loc=0x0000021d
912 # second dd will be never finished
913 dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 &
916 echo "step2: testing ......"
917 while [ $count -le 64 ]; do
918 dd_name="`ps x | awk '$1 == '$DDPID' { print $5 }'`"
919 if [ -z $dd_name ]; then
921 echo "debug: (dd_name=$dd_name, dd_pid=$DDPID, time=$count)"
922 error "dd shouldn't be finished!"
927 echo "(dd_pid=$DDPID, time=$count)successful"
929 #Recover fail_loc and dd will finish soon
930 do_facet ost1 lctl set_param fail_loc=0
932 echo "step3: testing ......"
934 if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
936 if [ $count -gt 500 ]; then
937 error "dd should be finished!"
941 echo "(dd_pid=$DDPID, time=$count)successful"
945 run_test 55 "ost_brw_read/write drops timed-out read/write request"
947 test_56() { # b=11277
948 #define OBD_FAIL_MDS_RESEND 0x136
950 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
952 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
955 run_test 56 "do not allow reconnect to busy exports"
958 # no oscs means no client or mdt
959 while lctl get_param osc.*.* > /dev/null 2>&1; do
960 : # loop until proc file is removed
964 test_57() { # bug 10866
968 #define OBD_FAIL_LPROC_REMOVE 0xB00
969 lctl set_param fail_loc=0x80000B00
970 zconf_umount `hostname` $DIR
971 lctl set_param fail_loc=0x80000B00
972 fail_abort $SINGLEMDS
974 lctl set_param fail_loc=0
976 do_facet client "df $DIR"
978 run_test 57 "read procfs entries causes kernel crash"
980 test_58() { # bug 11546
981 #define OBD_FAIL_MDC_ENQUEUE_PAUSE 0x801
984 lctl set_param fail_loc=0x80000801
985 cp $DIR/$tfile /dev/null &
988 lctl set_param fail_loc=0
989 drop_bl_callback rm -f $DIR/$tfile
991 do_facet client "df $DIR"
993 run_test 58 "Eviction in the middle of open RPC reply processing"
995 test_59() { # bug 10589
996 zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
997 echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
998 #define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311
999 lctl set_param fail_loc=0x311
1000 writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
1001 [ $? = 0 ] || error "dd write failed"
1002 writes=$(echo $writes | awk -F '+' '/out/ {print $1}')
1003 lctl set_param fail_loc=0
1005 zconf_umount `hostname` $MOUNT2 -f
1006 reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
1007 [ $? = 0 ] || error "dd read failed"
1008 reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
1009 [ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
1011 run_test 59 "Read cancel race on client eviction"
1014 # we assume that all md changes are in the MDT0 changelog
1015 if [ $MDSCOUNT -gt 1 ]; then
1016 error_ignore 17935 $*
1023 MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \
1024 awk '{gsub(/_UUID/,""); print $1}' | head -1)
1029 # Register (and start) changelog
1030 USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n)
1031 echo "Registered as $MDT0 changelog user $USER"
1033 # Generate a large number of changelog entries
1034 createmany -o $DIR/$tdir/$tfile $NUM_FILES
1038 # Unlink files in the background
1039 unlinkmany $DIR/$tdir/$tfile $NUM_FILES &
1043 # Failover the MDS while unlinks are happening
1044 facet_failover $SINGLEMDS
1046 # Wait for unlinkmany to finish
1049 # Check if all the create/unlink events were recorded
1051 $LFS changelog $MDT0 >> $DIR/$tdir/changelog
1052 local cl_count=$(grep UNLNK $DIR/$tdir/changelog | wc -l)
1053 echo "$cl_count unlinks in $MDT0 changelog"
1055 do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $USER
1056 USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \
1057 mdd.$MDT0.changelog_users | wc -l) - 2 ))
1058 if [ $USERS -eq 0 ]; then
1059 [ $cl_count -eq $NUM_FILES ] || \
1060 err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
1061 # Also make sure we can clear large changelogs
1062 cl_count=$($LFS changelog $FSNAME | wc -l)
1063 [ $cl_count -le 2 ] || \
1064 error "Changelog not empty: $cl_count entries"
1066 # If there are other users, there may be other unlinks in the log
1067 [ $cl_count -ge $NUM_FILES ] || \
1068 err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
1069 echo "$USERS other changelog users; can't verify clear"
1072 run_test 60 "Add Changelog entries during MDS failover"
1076 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0000)
1077 mdtosc=${mdtosc/-MDT*/-MDT\*}
1078 local cflags="osc.$mdtosc.connect_flags"
1079 do_facet $SINGLEMDS "lctl get_param -n $cflags" |grep -q skip_orphan
1080 [ $? -ne 0 ] && skip "don't have skip orphan feature" && return
1082 mkdir -p $DIR/$tdir || error "mkdir dir $DIR/$tdir failed"
1083 # Set the default stripe of $DIR/$tdir to put the files to ost1
1084 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1086 replay_barrier $SINGLEMDS
1087 createmany -o $DIR/$tdir/$tfile-%d 10
1088 local oid=`do_facet ost1 "lctl get_param -n obdfilter.${ost1_svc}.last_id"`
1090 fail_abort $SINGLEMDS
1092 touch $DIR/$tdir/$tfile
1093 local id=`$LFS getstripe $DIR/$tdir/$tfile | awk '$1 == 0 { print $2 }'`
1094 [ $id -le $oid ] && error "the orphan objid was reused, failed"
1099 run_test 61 "Verify to not reuse orphan objects - bug 17025"
1101 check_cli_ir_state()
1103 local NODE=${1:-$HOSTNAME}
1105 st=$(do_node $NODE "lctl get_param mgc.*.ir_state |
1106 awk '/imperative_recovery:/ { print \\\$2}'")
1107 [ $st != ON -o $st != OFF ] ||
1108 error "Error state $st, must be ON or OFF"
1112 check_target_ir_state()
1115 local name=${target}_svc
1116 local recovery_proc=obdfilter.${!name}.recovery_status
1119 st=$(do_facet $target "lctl get_param -n $recovery_proc |
1120 awk '/IR:/{ print \\\$2}'")
1121 [ $st != ON -o $st != OFF ] ||
1122 error "Error state $st, must be ON or OFF"
1128 do_facet mgs lctl set_param -n mgs.MGS.live.$FSNAME="state=$1"
1133 local state=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1134 awk '/state:/{ print \\\$2 }'")
1138 nidtbl_version_mgs()
1140 local ver=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1141 awk '/nidtbl_version:/{ print \\\$2 }'")
1145 # nidtbl_version_client <mds1|client> [node]
1146 nidtbl_version_client()
1149 local node=${2:-$HOSTNAME}
1151 if [ X$cli = Xclient ]; then
1154 local obdtype=${cli/%[0-9]*/}
1155 [ $obdtype != mds ] && error "wrong parameters $cli"
1157 node=$(facet_active_host $cli)
1162 local vers=$(do_node $node "lctl get_param -n mgc.*.ir_state" |
1163 awk "/$cli/{print \$6}" |sort -u)
1165 # in case there are multiple mounts on the client node
1167 [ ${#arr[@]} -ne 1 ] && error "versions on client node mismatch"
1171 nidtbl_versions_match()
1173 [ $(nidtbl_version_mgs) -eq $(nidtbl_version_client ${1:-client}) ]
1176 target_instance_match()
1182 obdtype=${srv/%[0-9]*/}
1193 error "invalid target type" $srv
1198 local target=${srv}_svc
1199 local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance)
1200 local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import | \
1201 awk '/instance/{ print $2 }' |head -1)
1203 return $([ $si -eq $ci ])
1209 set_ir_status disabled
1211 local saved_FAILURE_MODE=$FAILURE_MODE
1212 [ $(facet_host mgs) = $(facet_host ost1) ] && FAILURE_MODE="SOFT"
1216 nidtbl_versions_match &&
1217 error "version must differ due to IR disabled"
1218 target_instance_match ost1 || error "instance mismatch"
1222 FAILURE_MODE=$saved_FAILURE_MODE
1224 run_test 100 "IR: Make sure normal recovery still works w/o IR"
1230 local OST1_IMP=$(get_osc_import_name client ost1)
1232 # disable pinger recovery
1233 lctl set_param -n osc.$OST1_IMP.pinger_recov=0
1237 target_instance_match ost1 || error "instance mismatch"
1238 nidtbl_versions_match || error "version must match"
1240 lctl set_param -n osc.$OST1_IMP.pinger_recov=1
1242 run_test 101 "IR: Make sure IR works w/o normal recovery"
1246 local clients=${CLIENTS:-$HOSTNAME}
1253 # let's have a new nidtbl version
1256 # sleep for a while so that clients can see the failure of ost
1257 # it must be MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC.
1258 # int mgc_request.c:
1259 # define MGC_TIMEOUT_MIN_SECONDS 5
1260 # define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 *
1261 local count=30 # 20 seconds at most
1262 while [ $count -gt 0 ]; do
1263 nidtbl_versions_match && break
1268 nidtbl_versions_match || error "nidtbl mismatch"
1271 old_version=$(nidtbl_version_client client)
1273 zconf_umount_clients $clients $MOUNT || error "Cannot umount client"
1276 combined_mgs_mds && mgsdev=mds1
1277 remount_facet $mgsdev
1280 zconf_mount_clients $clients $MOUNT || error "Cannot mount client"
1283 new_version=$(nidtbl_version_client client)
1284 [ $new_version -lt $old_version ] &&
1285 error "nidtbl version wrong after mgs restarts"
1288 run_test 102 "IR: New client gets updated nidtbl after MGS restart"
1292 combined_mgs_mds && skip "mgs and mds on the same target" && return 0
1294 # workaround solution to generate config log on the mds
1300 # We need this test because mds is like a client in IR context.
1301 start mds1 $MDSDEV1 || error "MDS should start w/o mgs"
1303 # start mgs and remount mds w/ ir
1307 # remount client so that fsdb will be created on the MGS
1308 umount_client $MOUNT || error "umount failed"
1309 mount_client $MOUNT || error "mount failed"
1311 # sleep 30 seconds so the MDS has a chance to detect MGS restarting
1313 while [ $count -gt 0 ]; do
1314 [ $(nidtbl_version_client mds1) -ne 0 ] && break
1319 # after a while, mds should be able to reconnect to mgs and fetch
1320 # up-to-date nidtbl version
1321 nidtbl_versions_match mds1 || error "mds nidtbl mismatch"
1326 run_test 103 "IR: MDS can start w/o MGS and get updated nidtbl later"
1333 start ost1 $(ostdevname 1) "$OST_MOUNT_OPTS -onoir" ||
1334 error "OST1 cannot start"
1337 local ir_state=$(check_target_ir_state ost1)
1338 [ $ir_state = "OFF" ] || error "ir status on ost1 should be OFF"
1340 run_test 104 "IR: ost can disable IR voluntarily"
1344 [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
1348 # get one of the clients from client list
1349 local rcli=$(echo $RCLIENTS |cut -d' ' -f 1)
1351 local old_MOUNTOPT=$MOUNTOPT
1352 MOUNTOPT=${MOUNTOPT},noir
1353 zconf_umount $rcli $MOUNT || error "umount failed"
1354 zconf_mount $rcli $MOUNT || error "mount failed"
1356 # make sure lustre mount at $rcli disabling IR
1357 local ir_state=$(check_cli_ir_state $rcli)
1358 [ $ir_state = OFF ] || error "IR state must be OFF at $rcli"
1360 # make sure MGS's state is Partial
1361 [ $(get_ir_status) = "partial" ] || error "MGS IR state must be partial"
1364 # make sure IR on ost1 is OFF
1365 local ir_state=$(check_target_ir_state ost1)
1366 [ $ir_state = "OFF" ] || error "IR status on ost1 should be OFF"
1369 MOUNTOPT=$old_MOUNTOPT
1370 zconf_umount $rcli $MOUNT || error "umount failed"
1371 zconf_mount $rcli $MOUNT || error "mount failed"
1373 # make sure MGS's state is full
1374 [ $(get_ir_status) = "full" ] || error "MGS IR status must be full"
1377 # make sure IR on ost1 is ON
1378 local ir_state=$(check_target_ir_state ost1)
1379 [ $ir_state = "ON" ] || error "IR status on ost1 should be OFF"
1383 run_test 105 "IR: NON IR clients support"
1385 complete $(basename $0) $SECONDS
1386 check_and_cleanup_lustre