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 $f -s $((128 * 1024)) -i 1 -c 1
252 get_stripe_info client $f
253 if [ $stripe_index -ne 1 ]; then
255 error "$f: different stripe offset ($stripe_index)" && 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 # shouldn't have to set stripe size of count==1
282 lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1
283 get_stripe_info client $f
284 if [ $stripe_index -ne 0 ]; then
286 error "$f: different stripe offset ($stripe_index)" && return
289 do_facet client cp $SAMPLE_FILE $f
292 # allow recovery to complete
293 sleep $((TIMEOUT + 2))
294 # my understanding is that there should be nothing in the page
295 # cache after the client reconnects?
297 pgcache_empty || rc=2
301 run_test 18b "eviction and reconnect clears page cache (2766)"
304 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
306 do_facet client mkdir -p $DIR/$tdir
310 pgcache_empty || return 1
312 # shouldn't have to set stripe size of count==1
313 lfs setstripe $f -s $((128 * 1024)) -i 0 -c 1
314 get_stripe_info client $f
315 if [ $stripe_index -ne 0 ]; then
317 error "$f: different stripe offset ($stripe_index)" && return
320 do_facet client cp $SAMPLE_FILE $f
324 # OBD_FAIL_OST_CONNECT_NET2
325 # lost reply to connect request
326 do_facet ost1 lctl set_param fail_loc=0x80000225
329 df $MOUNT > /dev/null 2>&1
331 # my understanding is that there should be nothing in the page
332 # cache after the client reconnects?
334 pgcache_empty || rc=2
338 run_test 18c "Dropped connect reply after eviction handing (14755)"
342 do_facet client mcreate $f || return 1
343 drop_ldlm_cancel "chmod 0777 $f" || echo "evicted as expected"
345 do_facet client checkstat -v -p 0777 $f || echo evicted
346 # let the client reconnect
348 do_facet client "munlink $f"
350 run_test 19a "test expired_lock_main on mds (2867)"
354 do_facet client multiop $f Ow || return 1
355 do_facet client multiop $f or || return 2
359 do_facet client multiop $f or || return 3
360 drop_ldlm_cancel multiop $f Ow || echo "client evicted, as expected"
362 do_facet client munlink $f || return 4
364 run_test 19b "test expired_lock_main on ost (2867)"
366 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
367 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
370 lfs setstripe $DIR/$tdir/${tfile} -i 0 -c 1
371 multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
374 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
375 do_facet ost1 lctl set_param fail_loc=0x80000308
376 kill -USR1 $MULTI_PID
379 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
381 run_test 20a "ldlm_handle_enqueue error (should return error)"
383 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
384 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
387 lfs setstripe $DIR/$tdir/${tfile} -i 0 -c 1
389 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
390 do_facet ost1 lctl set_param fail_loc=0x80000308
391 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
392 error "didn't fail open enqueue" || true
394 run_test 20b "ldlm_handle_enqueue error (should return error)"
397 mkdir -p $DIR/$tdir-1
398 mkdir -p $DIR/$tdir-2
399 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
402 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
403 multiop $DIR/$tdir-2/f Oc &
406 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
408 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
409 kill -USR1 $close_pid
411 wait $close_pid || return 1
412 wait $open_pid || return 2
413 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
415 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
416 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
420 run_test 21a "drop close request while close and open are both in flight"
423 mkdir -p $DIR/$tdir-1
424 mkdir -p $DIR/$tdir-2
425 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
428 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
429 mcreate $DIR/$tdir-2/f &
432 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
434 kill -USR1 $close_pid
436 wait $close_pid || return 1
437 wait $open_pid || return 3
439 $CHECKSTAT -t file $DIR/$tdir-1/f || return 4
440 $CHECKSTAT -t file $DIR/$tdir-2/f || return 5
443 run_test 21b "drop open request while close and open are both in flight"
446 mkdir -p $DIR/$tdir-1
447 mkdir -p $DIR/$tdir-2
448 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
451 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
452 mcreate $DIR/$tdir-2/f &
455 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
457 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
458 kill -USR1 $close_pid
460 wait $close_pid || return 1
461 wait $open_pid || return 2
463 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
465 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
466 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
469 run_test 21c "drop both request while close and open are both in flight"
472 mkdir -p $DIR/$tdir-1
473 mkdir -p $DIR/$tdir-2
474 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
477 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
478 multiop $DIR/$tdir-2/f Oc &
480 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
482 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
485 wait $pid || return 1
486 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
488 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
489 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
493 run_test 21d "drop close reply while close and open are both in flight"
496 mkdir -p $DIR/$tdir-1
497 mkdir -p $DIR/$tdir-2
498 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
501 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
502 touch $DIR/$tdir-2/f &
504 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
508 wait $pid || return 1
511 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
512 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
515 run_test 21e "drop open reply while close and open are both in flight"
518 mkdir -p $DIR/$tdir-1
519 mkdir -p $DIR/$tdir-2
520 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
523 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
524 touch $DIR/$tdir-2/f &
526 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
528 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
531 wait $pid || return 1
532 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
534 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
535 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
538 run_test 21f "drop both reply while close and open are both in flight"
541 mkdir -p $DIR/$tdir-1
542 mkdir -p $DIR/$tdir-2
543 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
546 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
547 touch $DIR/$tdir-2/f &
549 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
551 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
554 wait $pid || return 1
555 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
557 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
558 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
561 run_test 21g "drop open reply and close request while close and open are both in flight"
564 mkdir -p $DIR/$tdir-1
565 mkdir -p $DIR/$tdir-2
566 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
569 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
570 touch $DIR/$tdir-2/f &
573 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
575 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
578 wait $pid || return 1
579 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
581 wait $touch_pid || return 2
583 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
584 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
587 run_test 21h "drop open request and close reply while close and open are both in flight"
589 # bug 3462 - multiple MDC requests
594 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
599 multiop $f1 msu || return 1
602 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
604 wait $close_pid || return 2
605 rm -rf $f2 || return 4
607 run_test 22 "drop close request and do mknod"
610 multiop_bg_pause $DIR/$tfile O_c || return 1
612 # give a chance for open
616 drop_request "kill -USR1 $pid"
619 wait $pid || return 1
622 run_test 23 "client hang when close a file after mds crash"
624 test_24() { # bug 11710 details correct fsync() behavior
625 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
628 lfs setstripe $DIR/$tdir -s 0 -i 0 -c 1
630 multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
633 kill -USR1 $MULTI_PID
636 lctl set_param fail_loc=0x0
638 [ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true
640 run_test 24 "fsync error (should return error)"
642 wait_client_evicted () {
645 local varsvc=${facet}_svc
647 wait_update $(facet_active_host $facet) \
648 "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \
652 test_26a() { # was test_26 bug 5921 - evict dead exports by pinger
653 # this test can only run from a client on a separate node.
654 remote_ost || { skip "local OST" && return 0; }
655 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
656 remote_mds || { skip "local MDS" && return 0; }
658 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
659 skip "msg and ost1 are at the same node"
663 check_timeout || return 1
665 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
667 echo starting with $OST_NEXP OST exports
668 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
669 do_facet client lctl set_param fail_loc=0x505
670 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
671 # But if there's a race to start the evictor from various obds,
672 # the loser might have to wait for the next ping.
675 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4))
677 do_facet client lctl set_param fail_loc=0x0
678 [ $rc -eq 0 ] || error "client not evicted from OST"
680 run_test 26a "evict dead exports"
682 test_26b() { # bug 10140 - evict dead exports by pinger
683 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
685 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
686 skip "msg and ost1 are at the same node"
690 check_timeout || return 1
692 zconf_mount `hostname` $MOUNT2 ||
693 { error "Failed to mount $MOUNT2"; return 2; }
694 sleep 1 # wait connections being established
696 local MDS_NEXP=$(do_facet $SINGLEMDS lctl get_param -n mdt.${mds1_svc}.num_exports | cut -d' ' -f2)
697 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
699 echo starting with $OST_NEXP OST and $MDS_NEXP MDS exports
701 zconf_umount `hostname` $MOUNT2 -f
703 # PING_INTERVAL max(obd_timeout / 4, 1U)
704 # PING_EVICT_TIMEOUT (PING_INTERVAL * 6)
706 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
707 # But if there's a race to start the evictor from various obds,
708 # the loser might have to wait for the next ping.
709 # = 9 * PING_INTERVAL + PING_INTERVAL
710 # = 10 PING_INTERVAL = 10 obd_timeout / 4 = 2.5 obd_timeout
711 # let's wait $((TIMEOUT * 3)) # bug 19887
713 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 3)) || \
714 error "Client was not evicted by ost" rc=1
715 wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 3)) || \
716 error "Client was not evicted by mds"
718 run_test 26b "evict dead exports"
722 writemany -q -a $DIR/$tdir/$tfile 0 5 &
725 local save_FAILURE_MODE=$FAILURE_MODE
727 facet_failover $SINGLEMDS
728 #define OBD_FAIL_OSC_SHUTDOWN 0x407
729 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000407
730 # need to wait for reconnect
731 echo waiting for fail_loc
732 wait_update_facet $SINGLEMDS "lctl get_param -n fail_loc" "-2147482617"
733 facet_failover $SINGLEMDS
735 kill -USR1 $CLIENT_PID
738 FAILURE_MODE=$save_FAILURE_MODE
740 run_test 27 "fail LOV while using OSC's"
742 test_28() { # bug 6086 - error adding new clients
743 do_facet client mcreate $DIR/$tfile || return 1
744 drop_bl_callback "chmod 0777 $DIR/$tfile" ||echo "evicted as expected"
745 #define OBD_FAIL_MDS_CLIENT_ADD 0x12f
746 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
747 # fail once (evicted), reconnect fail (fail_loc), ok
748 client_up || (sleep 10; client_up) || (sleep 10; client_up) || error "reconnect failed"
750 fail $SINGLEMDS # verify MDS last_rcvd can be loaded
752 run_test 28 "handle error adding new clients (bug 6086)"
754 test_29a() { # bug 22273 - error adding new clients
755 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
756 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000711"
757 # fail abort so client will be new again
758 fail_abort $SINGLEMDS
759 client_up || error "reconnect failed"
762 run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
764 test_29b() { # bug 22273 - error adding new clients
765 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
766 do_facet ost1 "lctl set_param fail_loc=0x80000711"
767 # fail abort so client will be new again
769 client_up || error "reconnect failed"
772 run_test 29b "error adding new clients doesn't cause LBUG (bug 22273)"
776 # put a load of file creates/writes/deletes
777 writemany -q $DIR/$tdir/$tfile 0 5 &
779 echo writemany pid $CLIENT_PID
783 # wait for client to reconnect to MDS
788 # client process should see no problems even though MDS went down
790 kill -USR1 $CLIENT_PID
793 echo writemany returned $rc
794 #these may fail because of eviction due to slow AST response.
795 [ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
797 run_test 50 "failover MDS under load"
800 #define OBD_FAIL_MDS_SYNC_CAPA_SL 0x1310
801 do_facet ost1 lctl set_param fail_loc=0x00001310
804 # put a load of file creates/writes/deletes
805 writemany -q $DIR/$tdir/$tfile 0 5 &
809 facet_failover $SINGLEMDS
810 # failover at various points during recovery
811 SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
812 echo will failover at $SEQ
815 echo failover in $i sec
817 facet_failover $SINGLEMDS
819 # client process should see no problems even though MDS went down
820 # and recovery was interrupted
822 kill -USR1 $CLIENT_PID
825 echo writemany returned $rc
826 [ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
828 run_test 51 "failover MDS during recovery"
831 do_facet client "mkdir -p $DIR/$tdir"
832 do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
834 echo writemany pid $CLIENT_PID
839 wait $CLIENT_PID || rc=$?
840 # active client process should see an EIO for down OST
841 [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
842 # but timing or failover setup may allow success
843 [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
844 echo "writemany returned $rc"
849 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
854 [ $rc -ne 0 ] && { return $rc; }
855 # wait for client to reconnect to OST
859 [ $rc -ne 0 ] && { return $rc; }
866 run_test 52 "failover OST under load"
868 # test of open reconstruct
871 drop_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
874 run_test 53 "touch: drop rep"
877 zconf_mount `hostname` $MOUNT2
881 cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
884 ERROR=`dmesg | egrep "(test 54|went back in time)" | tail -n1 | grep "went back in time"`
885 [ x"$ERROR" == x ] || error "back in time occured"
887 run_test 54 "back in time"
889 # bug 11330 - liblustre application death during I/O locks up OST
891 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
895 # first dd should be finished quickly
896 lfs setstripe $DIR/$tdir/$tfile-1 -c 1 -i 0
897 dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
900 echo "step1: testing ......"
902 if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
904 if [ $count -gt 64 ]; then
905 error "dd should be finished!"
909 echo "(dd_pid=$DDPID, time=$count)successful"
911 lfs setstripe $DIR/$tdir/$tfile-2 -c 1 -i 0
912 #define OBD_FAIL_OST_DROP_REQ 0x21d
913 do_facet ost1 lctl set_param fail_loc=0x0000021d
914 # second dd will be never finished
915 dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 &
918 echo "step2: testing ......"
919 while [ $count -le 64 ]; do
920 dd_name="`ps x | awk '$1 == '$DDPID' { print $5 }'`"
921 if [ -z $dd_name ]; then
923 echo "debug: (dd_name=$dd_name, dd_pid=$DDPID, time=$count)"
924 error "dd shouldn't be finished!"
929 echo "(dd_pid=$DDPID, time=$count)successful"
931 #Recover fail_loc and dd will finish soon
932 do_facet ost1 lctl set_param fail_loc=0
934 echo "step3: testing ......"
936 if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
938 if [ $count -gt 500 ]; then
939 error "dd should be finished!"
943 echo "(dd_pid=$DDPID, time=$count)successful"
947 run_test 55 "ost_brw_read/write drops timed-out read/write request"
949 test_56() { # b=11277
950 #define OBD_FAIL_MDS_RESEND 0x136
952 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
954 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
957 run_test 56 "do not allow reconnect to busy exports"
960 # no oscs means no client or mdt
961 while lctl get_param osc.*.* > /dev/null 2>&1; do
962 : # loop until proc file is removed
966 test_57() { # bug 10866
970 #define OBD_FAIL_LPROC_REMOVE 0xB00
971 lctl set_param fail_loc=0x80000B00
972 zconf_umount `hostname` $DIR
973 lctl set_param fail_loc=0x80000B00
974 fail_abort $SINGLEMDS
976 lctl set_param fail_loc=0
978 do_facet client "df $DIR"
980 run_test 57 "read procfs entries causes kernel crash"
982 test_58() { # bug 11546
983 #define OBD_FAIL_MDC_ENQUEUE_PAUSE 0x801
986 lctl set_param fail_loc=0x80000801
987 cp $DIR/$tfile /dev/null &
990 lctl set_param fail_loc=0
991 drop_bl_callback rm -f $DIR/$tfile
993 do_facet client "df $DIR"
995 run_test 58 "Eviction in the middle of open RPC reply processing"
997 test_59() { # bug 10589
998 zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
999 echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
1000 #define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311
1001 lctl set_param fail_loc=0x311
1002 writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
1003 [ $? = 0 ] || error "dd write failed"
1004 writes=$(echo $writes | awk -F '+' '/out/ {print $1}')
1005 lctl set_param fail_loc=0
1007 zconf_umount `hostname` $MOUNT2 -f
1008 reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
1009 [ $? = 0 ] || error "dd read failed"
1010 reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
1011 [ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
1013 run_test 59 "Read cancel race on client eviction"
1016 # we assume that all md changes are in the MDT0 changelog
1017 if [ $MDSCOUNT -gt 1 ]; then
1018 error_ignore 17935 $*
1025 MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \
1026 awk '{gsub(/_UUID/,""); print $1}' | head -1)
1031 # Register (and start) changelog
1032 USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n)
1033 echo "Registered as $MDT0 changelog user $USER"
1035 # Generate a large number of changelog entries
1036 createmany -o $DIR/$tdir/$tfile $NUM_FILES
1040 # Unlink files in the background
1041 unlinkmany $DIR/$tdir/$tfile $NUM_FILES &
1045 # Failover the MDS while unlinks are happening
1046 facet_failover $SINGLEMDS
1048 # Wait for unlinkmany to finish
1051 # Check if all the create/unlink events were recorded
1053 $LFS changelog $MDT0 >> $DIR/$tdir/changelog
1054 local cl_count=$(grep UNLNK $DIR/$tdir/changelog | wc -l)
1055 echo "$cl_count unlinks in $MDT0 changelog"
1057 do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $USER
1058 USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \
1059 mdd.$MDT0.changelog_users | wc -l) - 2 ))
1060 if [ $USERS -eq 0 ]; then
1061 [ $cl_count -eq $NUM_FILES ] || \
1062 err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
1063 # Also make sure we can clear large changelogs
1064 cl_count=$($LFS changelog $FSNAME | wc -l)
1065 [ $cl_count -le 2 ] || \
1066 error "Changelog not empty: $cl_count entries"
1068 # If there are other users, there may be other unlinks in the log
1069 [ $cl_count -ge $NUM_FILES ] || \
1070 err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
1071 echo "$USERS other changelog users; can't verify clear"
1074 run_test 60 "Add Changelog entries during MDS failover"
1078 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0000)
1079 mdtosc=${mdtosc/-MDT*/-MDT\*}
1080 local cflags="osc.$mdtosc.connect_flags"
1081 do_facet $SINGLEMDS "lctl get_param -n $cflags" |grep -q skip_orphan
1082 [ $? -ne 0 ] && skip "don't have skip orphan feature" && return
1084 mkdir -p $DIR/$tdir || error "mkdir dir $DIR/$tdir failed"
1085 # Set the default stripe of $DIR/$tdir to put the files to ost1
1086 $LFS setstripe -c 1 --index 0 $DIR/$tdir
1088 replay_barrier $SINGLEMDS
1089 createmany -o $DIR/$tdir/$tfile-%d 10
1090 local oid=`do_facet ost1 "lctl get_param -n obdfilter.${ost1_svc}.last_id"`
1092 fail_abort $SINGLEMDS
1094 touch $DIR/$tdir/$tfile
1095 local id=`$LFS getstripe $DIR/$tdir/$tfile |awk '($1 ~ 0 && $2 ~ /^[1-9]+/) {print $2}'`
1096 [ $id -le $oid ] && error "the orphan objid was reused, failed"
1101 run_test 61 "Verify to not reuse orphan objects - bug 17025"
1103 check_cli_ir_state()
1105 local NODE=${1:-`hostname`}
1107 st=$(do_node $NODE "lctl get_param mgc.*.ir_state |
1108 awk '/imperative_recovery:/ { print \\\$2}' ")
1109 [ $st != ON -o $st != OFF ] ||
1110 error "Error state $st, must be ON or OFF"
1114 check_target_ir_state()
1117 local name=${target}_svc
1118 local recovery_proc=obdfilter.${!name}.recovery_status
1121 st=$(do_facet $target "lctl get_param -n $recovery_proc |
1122 awk '/IR:/{ print \\\$2}'")
1123 [ $st != ON -o $st != OFF ] ||
1124 error "Error state $st, must be ON or OFF"
1130 do_facet mgs lctl set_param -n mgs.MGS.live.$FSNAME="state=$1"
1135 local state=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1136 awk '/state:/{ print \\\$2 }'")
1140 nidtbl_version_mgs()
1142 local ver=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1143 awk '/nidtbl_version:/{ print \\\$2 }'")
1147 # nidtbl_version_client <mds1|client> [node]
1148 nidtbl_version_client()
1151 local node=${2:-`hostname`}
1153 if [ X$cli = Xclient ]; then
1156 local obdtype=${cli/%[0-9]*/}
1157 [ $obdtype != mds ] && error "wrong parameters $cli"
1159 node=`facet_active_host $cli`
1164 local vers=$(do_node $node "lctl get_param -n mgc.*.ir_state" |
1165 awk "/$cli/{print \$6}" |sort -u)
1167 # in case there are multiple mounts on the client node
1169 [ ${#arr[@]} -ne 1 ] && error "versions on client node mismatch"
1173 nidtbl_versions_match()
1175 [ `nidtbl_version_mgs` -eq `nidtbl_version_client ${1:-client}` ]
1178 target_instance_match()
1184 obdtype=${srv/%[0-9]*/}
1195 error "invalid target type" $srv
1200 local target=${srv}_svc
1201 local si=`do_facet $srv lctl get_param -n $obdname.${!target}.instance`
1202 local ci=`lctl get_param -n $cliname.${!target}-${cliname}-*.import | \
1203 awk '/instance/{ print $2 }' |head -1`
1205 return `[ $si -eq $ci ]`
1211 set_ir_status disabled
1213 local saved_FAILURE_MODE=$FAILURE_MODE
1214 [ `facet_host mgs` = `facet_host ost1` ] && FAILURE_MODE="SOFT"
1218 nidtbl_versions_match &&
1219 error "version must differ due to IR disabled"
1220 target_instance_match ost1 || error "instance mismatch"
1224 FAILURE_MODE=$saved_FAILURE_MODE
1226 run_test 100 "IR: Make sure normal recovery still works w/o IR"
1232 local OST1_IMP=$(get_osc_import_name client ost1)
1234 # disable pinger recovery
1235 lctl set_param -n osc.$OST1_IMP.pinger_recov=0
1239 target_instance_match ost1 || error "instance mismatch"
1240 nidtbl_versions_match || error "version must match"
1242 lctl set_param -n osc.$OST1_IMP.pinger_recov=1
1244 run_test 101 "IR: Make sure IR works w/o normal recovery"
1248 local clients=${CLIENTS:-`hostname`}
1255 # let's have a new nidtbl version
1258 # sleep for a while so that clients can see the failure of ost
1259 # it must be MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC.
1260 # int mgc_request.c:
1261 # define MGC_TIMEOUT_MIN_SECONDS 5
1262 # define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 *
1263 local count=30 # 20 seconds at most
1264 while [ $count -gt 0 ]; do
1265 nidtbl_versions_match && break
1270 nidtbl_versions_match || error "nidtbl mismatch"
1273 old_version=`nidtbl_version_client client`
1275 zconf_umount_clients $clients $MOUNT || error "Cannot umount client"
1278 combined_mgs_mds && mgsdev=mds1
1279 remount_facet $mgsdev
1282 zconf_mount_clients $clients $MOUNT || error "Cannot mount client"
1285 new_version=`nidtbl_version_client client`
1286 [ $new_version -lt $old_version ] &&
1287 error "nidtbl version wrong after mgs restarts"
1290 run_test 102 "IR: New client gets updated nidtbl after MGS restart"
1294 combined_mgs_mds && skip "mgs and mds on the same target" && return 0
1296 # workaround solution to generate config log on the mds
1302 # We need this test because mds is like a client in IR context.
1303 start mds1 $MDSDEV1 || error "MDS should start w/o mgs"
1305 # start mgs and remount mds w/ ir
1309 # remount client so that fsdb will be created on the MGS
1310 umount_client $MOUNT || error "umount failed"
1311 mount_client $MOUNT || error "mount failed"
1313 # sleep 30 seconds so the MDS has a chance to detect MGS restarting
1315 while [ $count -gt 0 ]; do
1316 [ `nidtbl_version_client mds1` -ne 0 ] && break
1321 # after a while, mds should be able to reconnect to mgs and fetch
1322 # up-to-date nidtbl version
1323 nidtbl_versions_match mds1 || error "mds nidtbl mismatch"
1328 run_test 103 "IR: MDS can start w/o MGS and get updated nidtbl later"
1335 start ost1 `ostdevname 1` "$OST_MOUNT_OPTS -onoir" ||
1336 error "OST1 cannot start"
1339 local ir_state=$(check_target_ir_state ost1)
1340 [ $ir_state = "OFF" ] || error "ir status on ost1 should be OFF"
1343 run_test 104 "IR: ost can disable IR voluntarily"
1347 [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
1351 # get one of the clients from client list
1352 rcli=`echo $RCLIENTS |cut -d' ' -f 1`
1354 local old_MOUNTOPT=$MOUNTOPT
1355 MOUNTOPT=${MOUNTOPT},noir
1356 zconf_umount $rcli $MOUNT || error "umount failed"
1357 zconf_mount $rcli $MOUNT || error "mount failed"
1359 # make sure lustre mount at $rcli disabling IR
1360 local ir_state=$(check_cli_ir_state $rcli)
1361 [ $ir_state = OFF ] || error "IR state must be OFF at $rcli"
1363 # make sure MGS's state is Partial
1364 [ `get_ir_status` = "partial" ] || error "MGS IR state must be partial"
1367 # make sure IR on ost1 is OFF
1368 local ir_state=$(check_target_ir_state ost1)
1369 [ $ir_state = "OFF" ] || error "IR status on ost1 should be OFF"
1372 MOUNTOPT=$old_MOUNTOPT
1373 zconf_umount $rcli $MOUNT || error "umount failed"
1374 zconf_mount $rcli $MOUNT || error "mount failed"
1376 # make sure MGS's state is full
1377 [ `get_ir_status` = "full" ] || error "MGS IR status must be full"
1380 # make sure IR on ost1 is ON
1381 local ir_state=$(check_target_ir_state ost1)
1382 [ $ir_state = "ON" ] || error "IR status on ost1 should be OFF"
1386 run_test 105 "IR: NON IR clients support"
1388 complete $(basename $0) $SECONDS
1389 check_and_cleanup_lustre