6 ALWAYS_EXCEPT="52 60 $RECOVERY_SMALL_EXCEPT"
8 export MULTIOP=${MULTIOP:-multiop}
9 PTLDEBUG=${PTLDEBUG:--1}
10 LUSTRE=${LUSTRE:-`dirname $0`/..}
11 . $LUSTRE/tests/test-framework.sh
13 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
16 require_dsh_mds || exit 0
18 # also long tests: 19, 21a, 21e, 21f, 23, 27
19 # 1 2.5 2.5 4 4 (min)"
20 [ "$SLOW" = "no" ] && EXCEPT_SLOW="17 26a 26b 50 51 57"
22 [ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
23 # bug number for skipped test: LU-2194
24 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 19b"
28 # Allow us to override the setup if we already have a mounted system by
29 # setting SETUP=" " and CLEANUP=" "
31 CLEANUP=${CLEANUP:-""}
33 check_and_setup_lustre
36 rm -rf $DIR/d[0-9]* $DIR/f.${TESTSUITE}*
39 local f1="$DIR/$tfile"
40 local f2="$DIR/$tfile.2"
42 drop_request "mcreate $f1" ||
43 error_noexit "create '$f1': drop req"
45 drop_reint_reply "mcreate $f2" ||
46 error_noexit "create '$f2': drop rep"
48 drop_request "tchmod 111 $f2" ||
49 error_noexit "chmod '$f2': drop req"
51 drop_reint_reply "tchmod 666 $f2" ||
52 error_noexit "chmod '$f2': drop rep"
54 drop_request "statone $f2" ||
55 error_noexit "stat '$f2': drop req"
57 drop_reply "statone $f2" ||
58 error_noexit "stat '$f2': drop rep"
60 run_test 1 "create, chmod, stat: drop req, drop rep"
64 do_facet_create_file client $t 10K ||
65 error_noexit "Create file $t"
67 drop_request "cat $t > /dev/null" ||
68 error_noexit "Open request for $t file"
70 drop_reply "cat $t > /dev/null" ||
71 error_noexit "Open replay for $t file"
73 run_test 4 "open: drop req, drop rep"
78 local RR="$T-renamed-again"
79 do_facet_create_file client $T 10K ||
80 error_noexit "Create file $T"
82 drop_request "mv $T $R" ||
83 error_noexit "Rename $T"
85 drop_reint_reply "mv $R $RR" ||
86 error_noexit "Failed rename replay on $R"
88 do_facet client "checkstat -v $RR" ||
89 error_noexit "checkstat error on $RR"
91 do_facet client "rm $RR" ||
92 error_noexit "Can't remove file $RR"
94 run_test 5 "rename: drop req, drop rep"
98 local LINK1=$DIR/$tfile.link1
99 local LINK2=$DIR/$tfile.link2
101 do_facet_create_file client $T 10K ||
102 error_noexit "Create file $T"
104 drop_request "mlink $T $LINK1" ||
105 error_noexit "mlink request for $T"
107 drop_reint_reply "mlink $T $LINK2" ||
108 error_noexit "mlink reply for $T"
110 drop_request "munlink $LINK1" ||
111 error_noexit "munlink request for $T"
113 drop_reint_reply "munlink $LINK2" ||
114 error_noexit "munlink reply for $T"
116 do_facet client "rm $T" ||
117 error_noexit "Can't remove file $T"
119 run_test 6 "link, unlink: drop req, drop rep"
123 drop_reint_reply "touch $DIR/$tfile" || return 1
125 run_test 8 "touch: drop rep (bug 1423)"
129 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
133 do_facet_random_file client $TMP/$tfile 1K ||
134 error_noexit "Create random file $TMP/$tfile"
135 # make this big, else test 9 doesn't wait for bulk -- bz 5595
136 do_facet_create_file client $TMP/$t1 4M ||
137 error_noexit "Create file $TMP/$t1"
138 do_facet client "cp $TMP/$t1 $DIR/$t1" ||
139 error_noexit "Can't copy to $DIR/$t1 file"
140 pause_bulk "cp $TMP/$tfile $DIR/$tfile" ||
141 error_noexit "Can't pause_bulk copy"
142 do_facet client "cp $TMP/$t1 $DIR/$t2" ||
143 error_noexit "Can't copy file"
144 do_facet client "sync"
145 do_facet client "rm $DIR/$tfile $DIR/$t2 $DIR/$t1" ||
146 error_noexit "Can't remove files"
147 do_facet client "rm $TMP/$t1 $TMP/$tfile"
149 run_test 9 "pause bulk on OST (bug 1420)"
153 do_facet client mcreate $DIR/$tfile ||
154 { error "mcreate failed: $?"; return 1; }
155 drop_bl_callback "chmod 0777 $DIR/$tfile" || echo "evicted as expected"
156 # wait for the mds to evict the client
157 #echo "sleep $(($TIMEOUT*2))"
158 #sleep $(($TIMEOUT*2))
159 do_facet client touch $DIR/$tfile || echo "touch failed, evicted"
160 do_facet client checkstat -v -p 0777 $DIR/$tfile ||
161 { error "client checkstat failed: $?"; return 3; }
162 do_facet client "munlink $DIR/$tfile"
163 # allow recovery to complete
164 client_up || client_up || sleep $TIMEOUT
166 run_test 10 "finish request on server after client eviction (bug 1521)"
169 # wake up a thread waiting for completion after eviction
171 do_facet client $MULTIOP $DIR/$tfile Ow ||
172 { error "multiop write failed: $?"; return 1; }
173 do_facet client $MULTIOP $DIR/$tfile or ||
174 { error "multiop read failed: $?"; return 2; }
178 do_facet client $MULTIOP $DIR/$tfile or ||
179 { error "multiop read failed: $?"; return 3; }
180 drop_bl_callback $MULTIOP $DIR/$tfile Ow || echo "evicted as expected"
182 do_facet client munlink $DIR/$tfile ||
183 { error "munlink failed: $?"; return 4; }
184 # allow recovery to complete
185 client_up || client_up || sleep $TIMEOUT
187 run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
191 $LCTL mark $MULTIOP $DIR/$tfile OS_c
192 do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
193 clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
194 multiop_bg_pause $DIR/$tfile OS_c ||
195 { error "multiop failed: $?"; return 1; }
197 #define OBD_FAIL_MDS_CLOSE_NET 0x115
199 echo "waiting for multiop $PID"
200 wait $PID || { error "wait for multiop faile: $?"; return 2; }
201 do_facet client munlink $DIR/$tfile ||
202 { error "client munlink failed: $?"; return 3; }
203 # allow recovery to complete
204 client_up || client_up || sleep $TIMEOUT
206 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
208 # Bug 113, check that readdir lost recv timeout works.
210 mkdir -p $DIR/$tdir || { error "mkdir failed: $?"; return 1; }
211 touch $DIR/$tdir/newentry || { error "touch failed: $?"; return 2; }
212 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
213 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104"
214 ls $DIR/$tdir || { error "ls failed: $?"; return 3; }
215 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
216 rm -rf $DIR/$tdir || { error "remove test dir failed: $?"; return 4; }
218 run_test 13 "mdc_readpage restart test (bug 1138)"
220 # Bug 113, check that readdir lost send timeout works.
223 touch $DIR/$tdir/newentry
224 # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
225 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000106"
226 ls $DIR/$tdir || return 1
227 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
229 run_test 14 "mdc_readpage resend test (bug 1138)"
232 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
233 touch $DIR/$tfile && return 1
236 run_test 15 "failed open (-ENOMEM)"
238 READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1`
240 lctl set_param -n llite.*.max_read_ahead_mb 0
244 lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD
248 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
250 do_facet_random_file client $TMP/$tfile 100K ||
251 { error_noexit "Create random file $TMP/$T" ; return 0; }
252 do_facet client "cp $TMP/$tfile $DIR/$tfile" ||
253 { error_noexit "Copy to $DIR/$tfile file" ; return 0; }
257 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
258 do_facet ost1 "lctl set_param fail_loc=0x80000504"
260 # OST bulk will time out here, client resends
261 do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 1
262 do_facet ost1 lctl set_param fail_loc=0
263 # give recovery a chance to finish (shouldn't take long)
265 do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 2
268 run_test 16 "timeout bulk put, don't evict client (2732)"
273 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
275 local SAMPLE_FILE=$TMP/$tfile
276 do_facet_random_file client $SAMPLE_FILE 20K ||
277 { error_noexit "Create random file $SAMPLE_FILE" ; return 0; }
279 # With adaptive timeouts, bulk_get won't expire until adaptive_timeout_max
280 if at_is_enabled; then
281 at_max_saved=$(at_max_get ost1)
282 at_max_set $TIMEOUT ost1
285 # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
286 # OST bulk will time out here, client retries
287 do_facet ost1 lctl set_param fail_loc=0x80000503
288 # need to ensure we send an RPC
289 do_facet client cp $SAMPLE_FILE $DIR/$tfile
292 # with AT, client will wait adaptive_max*factor+net_latency before
293 # expiring the req, hopefully timeout*2 is enough
294 sleep $(($TIMEOUT*2))
296 do_facet ost1 lctl set_param fail_loc=0
297 do_facet client "df $DIR"
298 # expect cmp to succeed, client resent bulk
299 do_facet client "cmp $SAMPLE_FILE $DIR/$tfile" || return 3
300 do_facet client "rm $DIR/$tfile" || return 4
301 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved ost1
304 run_test 17 "timeout bulk get, don't evict client (2732)"
307 [ -z ${ost2_svc} ] && skip_env "needs 2 osts" && return 0
309 do_facet_create_file client $TMP/$tfile 20K ||
310 { error_noexit "Create file $TMP/$tfile" ; return 0; }
312 do_facet client mkdir -p $DIR/$tdir
316 pgcache_empty || return 1
319 $LFS setstripe -i 1 -c 1 $f
320 stripe_index=$($LFS getstripe -i $f)
321 if [ $stripe_index -ne 1 ]; then
323 error "$f: stripe_index $stripe_index != 1" && return
326 do_facet client cp $TMP/$tfile $f
328 local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | egrep -v 'MDT' | awk '{print $1}'`
329 $LCTL --device $osc2dev deactivate || return 3
330 # my understanding is that there should be nothing in the page
331 # cache after the client reconnects?
333 pgcache_empty || rc=2
334 $LCTL --device $osc2dev activate
338 run_test 18a "manual ost invalidate clears page cache immediately"
341 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
343 do_facet_create_file client $TMP/$tfile 20K ||
344 { error_noexit "Create file $TMP/$tfile" ; return 0; }
346 do_facet client mkdir -p $DIR/$tdir
350 pgcache_empty || return 1
352 $LFS setstripe -i 0 -c 1 $f
353 stripe_index=$($LFS getstripe -i $f)
354 if [ $stripe_index -ne 0 ]; then
356 error "$f: stripe_index $stripe_index != 0" && return
359 do_facet client cp $TMP/$tfile $f
362 # allow recovery to complete
363 sleep $((TIMEOUT + 2))
364 # my understanding is that there should be nothing in the page
365 # cache after the client reconnects?
367 pgcache_empty || rc=2
371 run_test 18b "eviction and reconnect clears page cache (2766)"
374 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
376 do_facet_create_file client $TMP/$tfile 20K ||
377 { error_noexit "Create file $TMP/$tfile" ; return 0; }
379 do_facet client mkdir -p $DIR/$tdir
383 pgcache_empty || return 1
385 $LFS setstripe -i 0 -c 1 $f
386 stripe_index=$($LFS getstripe -i $f)
387 if [ $stripe_index -ne 0 ]; then
389 error "$f: stripe_index $stripe_index != 0" && return
392 do_facet client cp $TMP/$tfile $f
396 # OBD_FAIL_OST_CONNECT_NET2
397 # lost reply to connect request
398 do_facet ost1 lctl set_param fail_loc=0x80000225
401 df $MOUNT > /dev/null 2>&1
403 # my understanding is that there should be nothing in the page
404 # cache after the client reconnects?
406 pgcache_empty || rc=2
410 run_test 18c "Dropped connect reply after eviction handing (14755)"
413 local BEFORE=`date +%s`
416 mount_client $DIR2 || error "failed to mount $DIR2"
418 do_facet client "stat $DIR > /dev/null" ||
419 error "failed to stat $DIR: $?"
420 drop_ldlm_cancel "chmod 0777 $DIR2" ||
421 error "failed to chmod $DIR2"
425 # let the client reconnect
427 EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state | \
428 awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
430 [ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
431 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
432 error "no eviction: $EVICT before:$BEFORE")
434 run_test 19a "test expired_lock_main on mds (2867)"
437 local BEFORE=`date +%s`
440 mount_client $DIR2 || error "failed to mount $DIR2: $?"
442 do_facet client $MULTIOP $DIR/$tfile Ow ||
443 error "failed to run multiop: $?"
444 drop_ldlm_cancel $MULTIOP $DIR2/$tfile Ow ||
445 error "failed to ldlm_cancel: $?"
447 umount_client $DIR2 || error "failed to unmount $DIR2: $?"
448 do_facet client munlink $DIR/$tfile ||
449 error "failed to unlink $DIR/$tfile: $?"
451 # let the client reconnect
453 EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state | \
454 awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
456 [ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
457 (do_facet client $LCTL get_param osc.$FSNAME-OST*.state;
458 error "no eviction: $EVICT before:$BEFORE")
460 run_test 19b "test expired_lock_main on ost (2867)"
463 local BEFORE=`date +%s`
466 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=0
468 mkdir -p $DIR1/$tfile
471 #define OBD_FAIL_PTLRPC_CANCEL_RESEND 0x516
472 do_facet mds $LCTL set_param fail_loc=0x80000516
474 touch $DIR2/$tfile/file1 &
476 # let touch to get blocked on the server
480 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=1
483 # let the client reconnect
485 EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
486 awk -F"[ [,]" '/EVICTED]$/ { if (mx<$4) {mx=$4;} } END { print mx }')
488 [ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
490 run_test 19c "check reconnect and lock resend do not trigger expired_lock_main"
492 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
493 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
496 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
497 multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
500 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
501 do_facet ost1 lctl set_param fail_loc=0x80000308
502 kill -USR1 $MULTI_PID
505 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
507 run_test 20a "ldlm_handle_enqueue error (should return error)"
509 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
510 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
513 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
515 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
516 do_facet ost1 lctl set_param fail_loc=0x80000308
517 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
518 error "didn't fail open enqueue" || true
520 run_test 20b "ldlm_handle_enqueue error (should return error)"
523 mkdir -p $DIR/$tdir-1
524 mkdir -p $DIR/$tdir-2
525 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
528 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
529 $MULTIOP $DIR/$tdir-2/f Oc &
532 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
534 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
535 kill -USR1 $close_pid
537 wait $close_pid || return 1
538 wait $open_pid || return 2
539 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
541 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
542 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
546 run_test 21a "drop close request while close and open are both in flight"
549 mkdir -p $DIR/$tdir-1
550 mkdir -p $DIR/$tdir-2
551 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
554 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
555 mcreate $DIR/$tdir-2/f &
558 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
560 kill -USR1 $close_pid
562 wait $close_pid || return 1
563 wait $open_pid || return 3
565 $CHECKSTAT -t file $DIR/$tdir-1/f || return 4
566 $CHECKSTAT -t file $DIR/$tdir-2/f || return 5
569 run_test 21b "drop open request while close and open are both in flight"
572 mkdir -p $DIR/$tdir-1
573 mkdir -p $DIR/$tdir-2
574 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
577 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
578 mcreate $DIR/$tdir-2/f &
581 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
583 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
584 kill -USR1 $close_pid
586 wait $close_pid || return 1
587 wait $open_pid || return 2
589 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
591 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
592 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
595 run_test 21c "drop both request while close and open are both in flight"
598 mkdir -p $DIR/$tdir-1
599 mkdir -p $DIR/$tdir-2
600 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
603 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
604 $MULTIOP $DIR/$tdir-2/f Oc &
606 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
608 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
611 wait $pid || return 1
612 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
614 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
615 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
619 run_test 21d "drop close reply while close and open are both in flight"
622 mkdir -p $DIR/$tdir-1
623 mkdir -p $DIR/$tdir-2
624 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
627 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
628 touch $DIR/$tdir-2/f &
630 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
634 wait $pid || return 1
637 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
638 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
641 run_test 21e "drop open reply while close and open are both in flight"
644 mkdir -p $DIR/$tdir-1
645 mkdir -p $DIR/$tdir-2
646 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
649 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
650 touch $DIR/$tdir-2/f &
652 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
654 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
657 wait $pid || return 1
658 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
660 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
661 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
664 run_test 21f "drop both reply while close and open are both in flight"
667 mkdir -p $DIR/$tdir-1
668 mkdir -p $DIR/$tdir-2
669 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
672 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
673 touch $DIR/$tdir-2/f &
675 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
677 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
680 wait $pid || return 1
681 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
683 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
684 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
687 run_test 21g "drop open reply and close request while close and open are both in flight"
690 mkdir -p $DIR/$tdir-1
691 mkdir -p $DIR/$tdir-2
692 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
695 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
696 touch $DIR/$tdir-2/f &
699 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
701 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
704 wait $pid || return 1
705 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
707 wait $touch_pid || return 2
709 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
710 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
713 run_test 21h "drop open request and close reply while close and open are both in flight"
715 # bug 3462 - multiple MDC requests
720 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
725 $MULTIOP $f1 msu || return 1
728 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
730 wait $close_pid || return 2
731 rm -rf $f2 || return 4
733 run_test 22 "drop close request and do mknod"
736 multiop_bg_pause $DIR/$tfile O_c || return 1
738 # give a chance for open
742 drop_request "kill -USR1 $pid"
745 wait $pid || return 1
748 run_test 23 "client hang when close a file after mds crash"
750 test_24a() { # bug 11710 details correct fsync() behavior
751 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
754 $LFS setstripe -i 0 -c 1 $DIR/$tdir
756 multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
759 kill -USR1 $MULTI_PID
762 lctl set_param fail_loc=0x0
764 [ $rc -eq 0 ] && error_ignore 5494 "multiop didn't fail fsync: rc $rc" || true
766 run_test 24a "fsync error (should return error)"
768 wait_client_evicted () {
771 local varsvc=${facet}_svc
773 wait_update $(facet_active_host $facet) \
774 "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \
779 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
783 lfs setstripe $DIR/$tdir -s 0 -i 0 -c 1
785 multiop_bg_pause $DIR/$tdir/$tfile-1 Ow8192_yc ||
786 error "mulitop Ow8192_yc failed"
789 multiop_bg_pause $DIR/$tdir/$tfile-2 Ow8192_c ||
790 error "mulitop Ow8192_c failed"
795 kill -USR1 $MULTI_PID1
798 kill -USR1 $MULTI_PID2
801 lctl set_param fail_loc=0x0
803 [ $rc1 -eq 0 -o $rc2 -eq 0 ] &&
804 error_ignore 5494 "multiop didn't fail fsync: $rc1 or close: $rc2" ||
807 dmesg | grep "dirty page discard:" || \
808 error "no discarded dirty page found!"
810 run_test 24b "test dirty page discard due to client eviction"
812 test_26a() { # was test_26 bug 5921 - evict dead exports by pinger
813 # this test can only run from a client on a separate node.
814 remote_ost || { skip "local OST" && return 0; }
815 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
816 remote_mds || { skip "local MDS" && return 0; }
818 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
819 skip "msg and ost1 are at the same node"
823 check_timeout || return 1
825 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
827 echo starting with $OST_NEXP OST exports
828 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
829 do_facet client lctl set_param fail_loc=0x505
830 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
831 # But if there's a race to start the evictor from various obds,
832 # the loser might have to wait for the next ping.
835 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4))
837 do_facet client lctl set_param fail_loc=0x0
838 [ $rc -eq 0 ] || error "client not evicted from OST"
840 run_test 26a "evict dead exports"
842 test_26b() { # bug 10140 - evict dead exports by pinger
843 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
845 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
846 skip "msg and ost1 are at the same node"
850 check_timeout || return 1
852 zconf_mount `hostname` $MOUNT2 ||
853 { error "Failed to mount $MOUNT2"; return 2; }
854 sleep 1 # wait connections being established
856 local MDS_NEXP=$(do_facet $SINGLEMDS lctl get_param -n mdt.${mds1_svc}.num_exports | cut -d' ' -f2)
857 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
859 echo starting with $OST_NEXP OST and $MDS_NEXP MDS exports
861 zconf_umount `hostname` $MOUNT2 -f
863 # PING_INTERVAL max(obd_timeout / 4, 1U)
864 # PING_EVICT_TIMEOUT (PING_INTERVAL * 6)
866 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
867 # But if there's a race to start the evictor from various obds,
868 # the loser might have to wait for the next ping.
869 # = 9 * PING_INTERVAL + PING_INTERVAL
870 # = 10 PING_INTERVAL = 10 obd_timeout / 4 = 2.5 obd_timeout
871 # let's wait $((TIMEOUT * 3)) # bug 19887
873 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 3)) || \
874 error "Client was not evicted by ost" rc=1
875 wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 3)) || \
876 error "Client was not evicted by mds"
878 run_test 26b "evict dead exports"
882 writemany -q -a $DIR/$tdir/$tfile 0 5 &
885 local save_FAILURE_MODE=$FAILURE_MODE
887 facet_failover $SINGLEMDS
888 #define OBD_FAIL_OSC_SHUTDOWN 0x407
889 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000407
890 # need to wait for reconnect
891 echo waiting for fail_loc
892 wait_update_facet $SINGLEMDS "lctl get_param -n fail_loc" "-2147482617"
893 facet_failover $SINGLEMDS
895 kill -USR1 $CLIENT_PID
898 FAILURE_MODE=$save_FAILURE_MODE
900 run_test 27 "fail LOV while using OSC's"
902 test_28() { # bug 6086 - error adding new clients
903 do_facet client mcreate $DIR/$tfile || return 1
904 drop_bl_callback "chmod 0777 $DIR/$tfile" ||echo "evicted as expected"
905 #define OBD_FAIL_MDS_CLIENT_ADD 0x12f
906 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
907 # fail once (evicted), reconnect fail (fail_loc), ok
908 client_up || (sleep 10; client_up) || (sleep 10; client_up) || error "reconnect failed"
910 fail $SINGLEMDS # verify MDS last_rcvd can be loaded
912 run_test 28 "handle error adding new clients (bug 6086)"
914 test_29a() { # bug 22273 - error adding new clients
915 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
916 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000711"
917 # fail abort so client will be new again
918 fail_abort $SINGLEMDS
919 client_up || error "reconnect failed"
920 wait_osc_import_state mds ost FULL
923 run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
925 test_29b() { # bug 22273 - error adding new clients
926 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
927 do_facet ost1 "lctl set_param fail_loc=0x80000711"
928 # fail abort so client will be new again
930 client_up || error "reconnect failed"
933 run_test 29b "error adding new clients doesn't cause LBUG (bug 22273)"
937 # put a load of file creates/writes/deletes
938 writemany -q $DIR/$tdir/$tfile 0 5 &
940 echo writemany pid $CLIENT_PID
944 # wait for client to reconnect to MDS
949 # client process should see no problems even though MDS went down
951 kill -USR1 $CLIENT_PID
954 echo writemany returned $rc
955 #these may fail because of eviction due to slow AST response.
956 [ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
958 run_test 50 "failover MDS under load"
961 #define OBD_FAIL_MDS_SYNC_CAPA_SL 0x1310
962 do_facet ost1 lctl set_param fail_loc=0x00001310
965 # put a load of file creates/writes/deletes
966 writemany -q $DIR/$tdir/$tfile 0 5 &
970 facet_failover $SINGLEMDS
971 # failover at various points during recovery
972 SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
973 echo will failover at $SEQ
976 #echo failover in $i sec
977 log "test_$testnum: failover in $i sec"
979 facet_failover $SINGLEMDS
981 # client process should see no problems even though MDS went down
982 # and recovery was interrupted
984 kill -USR1 $CLIENT_PID
987 echo writemany returned $rc
988 [ $rc -eq 0 ] || error_ignore 13652 "writemany returned rc $rc" || true
990 run_test 51 "failover MDS during recovery"
993 do_facet client "mkdir -p $DIR/$tdir"
994 do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
996 echo writemany pid $CLIENT_PID
1001 wait $CLIENT_PID || rc=$?
1002 # active client process should see an EIO for down OST
1003 [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
1004 # but timing or failover setup may allow success
1005 [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
1006 echo "writemany returned $rc"
1011 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1016 [ $rc -ne 0 ] && { return $rc; }
1017 # wait for client to reconnect to OST
1021 [ $rc -ne 0 ] && { return $rc; }
1028 run_test 52 "failover OST under load"
1030 # test of open reconstruct
1033 drop_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
1036 run_test 53 "touch: drop rep"
1039 zconf_mount `hostname` $MOUNT2
1041 touch $DIR2/$tfile.1
1043 cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
1046 ERROR=`dmesg | egrep "(test 54|went back in time)" | tail -n1 | grep "went back in time"`
1047 [ x"$ERROR" == x ] || error "back in time occured"
1049 run_test 54 "back in time"
1051 # bug 11330 - liblustre application death during I/O locks up OST
1053 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1057 # first dd should be finished quickly
1058 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
1059 dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
1062 echo "step1: testing ......"
1064 if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
1066 if [ $count -gt 64 ]; then
1067 error "dd should be finished!"
1071 echo "(dd_pid=$DDPID, time=$count)successful"
1073 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2
1074 #define OBD_FAIL_OST_DROP_REQ 0x21d
1075 do_facet ost1 lctl set_param fail_loc=0x0000021d
1076 # second dd will be never finished
1077 dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 &
1080 echo "step2: testing ......"
1081 while [ $count -le 64 ]; do
1082 dd_name="`ps x | awk '$1 == '$DDPID' { print $5 }'`"
1083 if [ -z $dd_name ]; then
1085 echo "debug: (dd_name=$dd_name, dd_pid=$DDPID, time=$count)"
1086 error "dd shouldn't be finished!"
1091 echo "(dd_pid=$DDPID, time=$count)successful"
1093 #Recover fail_loc and dd will finish soon
1094 do_facet ost1 lctl set_param fail_loc=0
1096 echo "step3: testing ......"
1098 if [ -z `ps x | awk '$1 == '$DDPID' { print $5 }'` ]; then break; fi
1100 if [ $count -gt 500 ]; then
1101 error "dd should be finished!"
1105 echo "(dd_pid=$DDPID, time=$count)successful"
1109 run_test 55 "ost_brw_read/write drops timed-out read/write request"
1111 test_56() { # b=11277
1112 #define OBD_FAIL_MDS_RESEND 0x136
1114 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
1116 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1119 run_test 56 "do not allow reconnect to busy exports"
1122 # no oscs means no client or mdt
1123 while lctl get_param osc.*.* > /dev/null 2>&1; do
1124 : # loop until proc file is removed
1128 test_57() { # bug 10866
1132 #define OBD_FAIL_LPROC_REMOVE 0xB00
1133 lctl set_param fail_loc=0x80000B00
1134 zconf_umount `hostname` $DIR
1135 lctl set_param fail_loc=0x80000B00
1136 fail_abort $SINGLEMDS
1138 lctl set_param fail_loc=0
1140 do_facet client "df $DIR"
1142 run_test 57 "read procfs entries causes kernel crash"
1144 test_58() { # bug 11546
1145 #define OBD_FAIL_MDC_ENQUEUE_PAUSE 0x801
1148 lctl set_param fail_loc=0x80000801
1149 cp $DIR/$tfile /dev/null &
1152 lctl set_param fail_loc=0
1153 drop_bl_callback rm -f $DIR/$tfile
1155 # the first 'df' could tigger the eviction caused by
1156 # 'drop_bl_callback', and it's normal case.
1157 # but the next 'df' should return successfully.
1158 do_facet client "df $DIR" || do_facet client "df $DIR"
1160 run_test 58 "Eviction in the middle of open RPC reply processing"
1162 test_59() { # bug 10589
1163 zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
1164 echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
1165 #define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311
1166 lctl set_param fail_loc=0x311
1167 writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
1168 [ $? = 0 ] || error "dd write failed"
1169 writes=$(echo $writes | awk -F '+' '/out/ {print $1}')
1170 lctl set_param fail_loc=0
1172 zconf_umount `hostname` $MOUNT2 -f
1173 reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
1174 [ $? = 0 ] || error "dd read failed"
1175 reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
1176 [ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
1178 run_test 59 "Read cancel race on client eviction"
1181 # we assume that all md changes are in the MDT0 changelog
1182 if [ $MDSCOUNT -gt 1 ]; then
1183 error_ignore 17935 $*
1190 MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid | \
1191 awk '{gsub(/_UUID/,""); print $1}' | head -1)
1196 # Register (and start) changelog
1197 USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n)
1198 echo "Registered as $MDT0 changelog user $USER"
1200 # Generate a large number of changelog entries
1201 createmany -o $DIR/$tdir/$tfile $NUM_FILES
1205 # Unlink files in the background
1206 unlinkmany $DIR/$tdir/$tfile $NUM_FILES &
1210 # Failover the MDS while unlinks are happening
1211 facet_failover $SINGLEMDS
1213 # Wait for unlinkmany to finish
1216 # Check if all the create/unlink events were recorded
1218 $LFS changelog $MDT0 >> $DIR/$tdir/changelog
1219 local cl_count=$(grep UNLNK $DIR/$tdir/changelog | wc -l)
1220 echo "$cl_count unlinks in $MDT0 changelog"
1222 do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $USER
1223 USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \
1224 mdd.$MDT0.changelog_users | wc -l) - 2 ))
1225 if [ $USERS -eq 0 ]; then
1226 [ $cl_count -eq $NUM_FILES ] || \
1227 err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
1228 # Also make sure we can clear large changelogs
1229 cl_count=$($LFS changelog $FSNAME | wc -l)
1230 [ $cl_count -le 2 ] || \
1231 error "Changelog not empty: $cl_count entries"
1233 # If there are other users, there may be other unlinks in the log
1234 [ $cl_count -ge $NUM_FILES ] || \
1235 err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
1236 echo "$USERS other changelog users; can't verify clear"
1239 run_test 60 "Add Changelog entries during MDS failover"
1243 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0000)
1244 mdtosc=${mdtosc/-MDT*/-MDT\*}
1245 local cflags="osc.$mdtosc.connect_flags"
1246 do_facet $SINGLEMDS "lctl get_param -n $cflags" |grep -q skip_orphan
1247 [ $? -ne 0 ] && skip "don't have skip orphan feature" && return
1249 mkdir -p $DIR/$tdir || error "mkdir dir $DIR/$tdir failed"
1250 # Set the default stripe of $DIR/$tdir to put the files to ost1
1251 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1253 replay_barrier $SINGLEMDS
1254 createmany -o $DIR/$tdir/$tfile-%d 10
1255 local oid=`do_facet ost1 "lctl get_param -n obdfilter.${ost1_svc}.last_id"`
1257 fail_abort $SINGLEMDS
1259 touch $DIR/$tdir/$tfile
1260 local id=`$LFS getstripe $DIR/$tdir/$tfile | awk '$1 == 0 { print $2 }'`
1261 [ $id -le $oid ] && error "the orphan objid was reused, failed"
1266 run_test 61 "Verify to not reuse orphan objects - bug 17025"
1268 # test_62 as seen it b2_1 please do not reuse test_62
1271 # zconf_umount `hostname` $DIR
1272 # #define OBD_FAIL_PTLRPC_DELAY_IMP_FULL 0x516
1273 # lctl set_param fail_loc=0x516
1276 #run_test 62 "Verify connection flags race - bug LU-1716"
1278 check_cli_ir_state()
1280 local NODE=${1:-$HOSTNAME}
1282 st=$(do_node $NODE "lctl get_param mgc.*.ir_state |
1283 awk '/imperative_recovery:/ { print \\\$2}'")
1284 [ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
1285 error "Error state $st, must be ENABLED or DISABLED"
1289 check_target_ir_state()
1292 local name=${target}_svc
1293 local recovery_proc=obdfilter.${!name}.recovery_status
1296 st=$(do_facet $target "lctl get_param -n $recovery_proc |
1297 awk '/IR:/{ print \\\$2}'")
1298 [ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
1299 error "Error state $st, must be ENABLED or DISABLED"
1305 do_facet mgs lctl set_param -n mgs.MGS.live.$FSNAME="state=$1"
1310 local state=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1311 awk '/state:/{ print \\\$2 }'")
1315 nidtbl_version_mgs()
1317 local ver=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1318 awk '/nidtbl_version:/{ print \\\$2 }'")
1322 # nidtbl_version_client <mds1|client> [node]
1323 nidtbl_version_client()
1326 local node=${2:-$HOSTNAME}
1328 if [ X$cli = Xclient ]; then
1331 local obdtype=${cli/%[0-9]*/}
1332 [ $obdtype != mds ] && error "wrong parameters $cli"
1334 node=$(facet_active_host $cli)
1339 local vers=$(do_node $node "lctl get_param -n mgc.*.ir_state" |
1340 awk "/$cli/{print \$6}" |sort -u)
1342 # in case there are multiple mounts on the client node
1344 [ ${#arr[@]} -ne 1 ] && error "versions on client node mismatch"
1348 nidtbl_versions_match()
1350 [ $(nidtbl_version_mgs) -eq $(nidtbl_version_client ${1:-client}) ]
1353 target_instance_match()
1359 obdtype=${srv/%[0-9]*/}
1370 error "invalid target type" $srv
1375 local target=${srv}_svc
1376 local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance)
1377 local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import | \
1378 awk '/instance/{ print $2 }' |head -1)
1380 return $([ $si -eq $ci ])
1385 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1386 { skip "MGS without IR support"; return 0; }
1388 # MDT was just restarted in the previous test, make sure everything
1391 while [ $cnt -gt 0 ]; do
1392 nidtbl_versions_match && break
1398 set_ir_status disabled
1400 local prev_ver=$(nidtbl_version_client client)
1402 local saved_FAILURE_MODE=$FAILURE_MODE
1403 [ $(facet_host mgs) = $(facet_host ost1) ] && FAILURE_MODE="SOFT"
1407 [ $(nidtbl_version_client client) -eq $prev_ver ] ||
1408 error "version must not change due to IR disabled"
1409 target_instance_match ost1 || error "instance mismatch"
1413 FAILURE_MODE=$saved_FAILURE_MODE
1415 run_test 100 "IR: Make sure normal recovery still works w/o IR"
1419 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1420 { skip "MGS without IR support"; return 0; }
1424 local OST1_IMP=$(get_osc_import_name client ost1)
1426 # disable pinger recovery
1427 lctl set_param -n osc.$OST1_IMP.pinger_recov=0
1431 target_instance_match ost1 || error "instance mismatch"
1432 nidtbl_versions_match || error "version must match"
1434 lctl set_param -n osc.$OST1_IMP.pinger_recov=1
1436 run_test 101 "IR: Make sure IR works w/o normal recovery"
1440 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1441 { skip "MGS without IR support"; return 0; }
1443 local clients=${CLIENTS:-$HOSTNAME}
1450 # let's have a new nidtbl version
1453 # sleep for a while so that clients can see the failure of ost
1454 # it must be MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC.
1455 # int mgc_request.c:
1456 # define MGC_TIMEOUT_MIN_SECONDS 5
1457 # define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 *
1458 local count=30 # 20 seconds at most
1459 while [ $count -gt 0 ]; do
1460 nidtbl_versions_match && break
1465 nidtbl_versions_match || error "nidtbl mismatch"
1468 old_version=$(nidtbl_version_client client)
1470 zconf_umount_clients $clients $MOUNT || error "Cannot umount client"
1473 combined_mgs_mds && mgsdev=mds1
1474 remount_facet $mgsdev
1477 zconf_mount_clients $clients $MOUNT || error "Cannot mount client"
1480 new_version=$(nidtbl_version_client client)
1481 [ $new_version -lt $old_version ] &&
1482 error "nidtbl version wrong after mgs restarts"
1485 run_test 102 "IR: New client gets updated nidtbl after MGS restart"
1489 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1490 { skip "MGS without IR support"; return 0; }
1492 combined_mgs_mds && skip "mgs and mds on the same target" && return 0
1494 # workaround solution to generate config log on the mds
1500 # We need this test because mds is like a client in IR context.
1501 start mds1 $MDSDEV1 || error "MDS should start w/o mgs"
1503 # start mgs and remount mds w/ ir
1507 # remount client so that fsdb will be created on the MGS
1508 umount_client $MOUNT || error "umount failed"
1509 mount_client $MOUNT || error "mount failed"
1511 # sleep 30 seconds so the MDS has a chance to detect MGS restarting
1513 while [ $count -gt 0 ]; do
1514 [ $(nidtbl_version_client mds1) -ne 0 ] && break
1519 # after a while, mds should be able to reconnect to mgs and fetch
1520 # up-to-date nidtbl version
1521 nidtbl_versions_match mds1 || error "mds nidtbl mismatch"
1526 run_test 103 "IR: MDS can start w/o MGS and get updated nidtbl later"
1530 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1531 { skip "MGS without IR support"; return 0; }
1536 start ost1 $(ostdevname 1) "$OST_MOUNT_OPTS -onoir" ||
1537 error "OST1 cannot start"
1540 local ir_state=$(check_target_ir_state ost1)
1541 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1542 error "ir status on ost1 should be DISABLED"
1544 run_test 104 "IR: ost can disable IR voluntarily"
1548 [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
1549 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1550 { skip "MGS without IR support"; return 0; }
1554 # get one of the clients from client list
1555 local rcli=$(echo $RCLIENTS |cut -d' ' -f 1)
1557 local old_MOUNTOPT=$MOUNTOPT
1558 MOUNTOPT=${MOUNTOPT},noir
1559 zconf_umount $rcli $MOUNT || error "umount failed"
1560 zconf_mount $rcli $MOUNT || error "mount failed"
1562 # make sure lustre mount at $rcli disabling IR
1563 local ir_state=$(check_cli_ir_state $rcli)
1564 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1565 error "IR state must be DISABLED at $rcli"
1567 # Since the client just mounted, its last_rcvd entry is not on disk.
1568 # Send an RPC so exp_need_sync forces last_rcvd to commit this export
1569 # so the client can reconnect during OST recovery (LU-924, LU-1582)
1570 $SETSTRIPE -i 0 $DIR/$tfile
1571 dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=sync
1573 # make sure MGS's state is Partial
1574 [ $(get_ir_status) = "partial" ] || error "MGS IR state must be partial"
1577 # make sure IR on ost1 is DISABLED
1578 local ir_state=$(check_target_ir_state ost1)
1579 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1580 error "IR status on ost1 should be DISABLED"
1583 MOUNTOPT=$old_MOUNTOPT
1584 zconf_umount $rcli $MOUNT || error "umount failed"
1585 zconf_mount $rcli $MOUNT || error "mount failed"
1587 # make sure MGS's state is full
1588 [ $(get_ir_status) = "full" ] || error "MGS IR status must be full"
1591 # make sure IR on ost1 is ENABLED
1592 local ir_state=$(check_target_ir_state ost1)
1593 [ $ir_state = "ENABLED" -o $ir_state = "ON" ] ||
1594 error "IR status on ost1 should be ENABLED"
1598 run_test 105 "IR: NON IR clients support"
1606 test_106() { # LU-1789
1607 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.50) ]] ||
1608 { skip "Need MDS version at least 2.3.50"; return 0; }
1610 #define OBD_FAIL_MDC_LIGHTWEIGHT 0x805
1611 $LCTL set_param fail_loc=0x805
1614 trap cleanup_106 EXIT
1616 # enable lightweight flag on mdc connection
1619 local MDS_NEXP=$(do_facet $SINGLEMDS \
1620 lctl get_param -n mdt.${mds1_svc}.num_exports |
1622 $LCTL set_param fail_loc=0
1624 touch $DIR2/$tfile || error "failed to create empty file"
1625 replay_barrier $SINGLEMDS
1627 $LCTL set_param debug=console
1629 facet_failover $SINGLEMDS
1631 # lightweight connection must be evicted
1632 touch -c $DIR2/$tfile || true
1633 $LCTL dk $TMP/lustre-log-$TESTNAME.log
1634 evicted=`awk '/This client was evicted by .*MDT0000/ {
1636 }' $TMP/lustre-log-$TESTNAME.log`
1637 [ -z "$evicted" ] && error "lightweight client not evicted by mds"
1639 # and all operations performed by lightweight client should be
1640 # synchronous, so the file created before mds restart should be there
1641 $CHECKSTAT -t file $DIR/$tfile || error "file not present"
1646 run_test 106 "lightweight connection support"
1653 # OBD_FAIL_MDS_REINT_NET_REP 0x119
1654 do_facet $SINGLEMDS lctl set_param fail_loc=0x119
1655 multiop $DIR/$tdir D_c &
1657 mkdir $DIR/$tdir/dir_106 &
1659 do_facet $SINGLEMDS lctl set_param fail_loc=0
1662 wait $CLIENT_PID || rc=$?
1663 checkstat -t dir $DIR/$tdir/dir_106 || return 1
1665 kill -USR1 $close_pid
1666 wait $close_pid || return 2
1670 run_test 107 "drop reint reply, then restart MDT"
1673 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1674 local remote_dir=$DIR/$tdir/remote_dir
1678 drop_request "$LFS mkdir -i $MDTIDX $remote_dir" ||
1679 error "lfs mkdir failed"
1680 local diridx=$($GETSTRIPE -M $remote_dir)
1681 [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
1683 rm -rf $DIR/$tdir || error "rmdir failed"
1685 run_test 110a "create remote directory: drop client req"
1688 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1689 local remote_dir=$DIR/$tdir/remote_dir
1693 drop_reint_reply "$LFS mkdir -i $MDTIDX $remote_dir" ||
1694 error "lfs mkdir failed"
1696 diridx=$($GETSTRIPE -M $remote_dir)
1697 [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
1699 rm -rf $DIR/$tdir || error "rmdir failed"
1701 run_test 110b "create remote directory: drop Master rep"
1704 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1705 local remote_dir=$DIR/$tdir/remote_dir
1709 drop_update_reply $((MDTIDX + 1)) "$LFS mkdir -i $MDTIDX $remote_dir" ||
1710 error "lfs mkdir failed"
1712 diridx=$($GETSTRIPE -M $remote_dir)
1713 [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
1715 rm -rf $DIR/$tdir || error "rmdir failed"
1717 run_test 110c "create remote directory: drop update rep on slave MDT"
1720 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1721 local remote_dir=$DIR/$tdir/remote_dir
1725 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
1727 drop_request "rm -rf $remote_dir" || error "rm remote dir failed"
1729 rm -rf $DIR/$tdir || error "rmdir failed"
1733 run_test 110d "remove remote directory: drop client req"
1736 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1737 local remote_dir=$DIR/$tdir/remote_dir
1741 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
1742 drop_reint_reply "rm -rf $remote_dir" || error "rm remote dir failed"
1744 rm -rf $DIR/$tdir || error "rmdir failed"
1748 run_test 110e "remove remote directory: drop master rep"
1751 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1752 local remote_dir=$DIR/$tdir/remote_dir
1756 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
1757 drop_update_reply $MDTIDX "rm -rf $remote_dir" ||
1758 error "rm remote dir failed"
1760 rm -rf $DIR/$tdir || error "rmdir failed"
1762 run_test 110f "remove remote directory: drop slave rep"
1765 check_and_cleanup_lustre