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 $SINGLEMDS) = "zfs" ] &&
23 # bug number for skipped test: LU-2194 LU-2547
24 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 19b 24a 24b"
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 local before=$(date +%s)
156 do_facet client "stat $DIR > /dev/null" ||
157 error "failed to stat $DIR: $?"
158 drop_bl_callback "chmod 0777 $DIR" ||
159 error "failed to chmod $DIR: $?"
161 # let the client reconnect
163 evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
164 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
165 [ ! -z "$evict" ] && [[ $evict -gt $before ]] ||
166 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
167 error "no eviction: $evict before:$before")
169 do_facet client checkstat -v -p 0777 $DIR ||
170 error "client checkstat failed: $?"
172 run_test 10a "finish request on server after client eviction (bug 1521)"
175 local before=$(date +%s)
178 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.53) ]] &&
179 skip "Need MDS version at least 2.6.53" && return
180 do_facet client "stat $DIR > /dev/null" ||
181 error "failed to stat $DIR: $?"
182 drop_bl_callback_once "chmod 0777 $DIR" ||
183 error "failed to chmod $DIR: $?"
185 # let the client reconnect
187 evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
188 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
190 [ -z "$evict" ] || [[ $evict -le $before ]] ||
191 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
192 error "eviction happened: $evict before:$before")
194 do_facet client checkstat -v -p 0777 $DIR ||
195 error "client checkstat failed: $?"
197 run_test 10b "re-send BL AST"
200 # wake up a thread waiting for completion after eviction
202 do_facet client $MULTIOP $DIR/$tfile Ow ||
203 { error "multiop write failed: $?"; return 1; }
204 do_facet client $MULTIOP $DIR/$tfile or ||
205 { error "multiop read failed: $?"; return 2; }
209 do_facet client $MULTIOP $DIR/$tfile or ||
210 { error "multiop read failed: $?"; return 3; }
211 drop_bl_callback_once $MULTIOP $DIR/$tfile Ow ||
212 echo "evicted as expected"
214 do_facet client munlink $DIR/$tfile ||
215 { error "munlink failed: $?"; return 4; }
216 # allow recovery to complete
217 client_up || client_up || sleep $TIMEOUT
219 run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
223 $LCTL mark $MULTIOP $DIR/$tfile OS_c
224 do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
225 clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
226 multiop_bg_pause $DIR/$tfile OS_c ||
227 { error "multiop failed: $?"; return 1; }
229 #define OBD_FAIL_MDS_CLOSE_NET 0x115
231 echo "waiting for multiop $PID"
232 wait $PID || { error "wait for multiop faile: $?"; return 2; }
233 do_facet client munlink $DIR/$tfile ||
234 { error "client munlink failed: $?"; return 3; }
235 # allow recovery to complete
236 client_up || client_up || sleep $TIMEOUT
238 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
240 # Bug 113, check that readdir lost recv timeout works.
242 mkdir -p $DIR/$tdir || { error "mkdir failed: $?"; return 1; }
243 touch $DIR/$tdir/newentry || { error "touch failed: $?"; return 2; }
244 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
245 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104"
246 ls $DIR/$tdir || { error "ls failed: $?"; return 3; }
247 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
248 rm -rf $DIR/$tdir || { error "remove test dir failed: $?"; return 4; }
250 run_test 13 "mdc_readpage restart test (bug 1138)"
252 # Bug 113, check that readdir lost send timeout works.
255 touch $DIR/$tdir/newentry
256 # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
257 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000106"
258 ls $DIR/$tdir || return 1
259 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
261 run_test 14 "mdc_readpage resend test (bug 1138)"
264 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
265 touch $DIR/$tfile && return 1
268 run_test 15 "failed open (-ENOMEM)"
270 READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1`
272 lctl set_param -n llite.*.max_read_ahead_mb 0
276 lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD
280 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
282 do_facet_random_file client $TMP/$tfile 100K ||
283 { error_noexit "Create random file $TMP/$T" ; return 0; }
284 do_facet client "cp $TMP/$tfile $DIR/$tfile" ||
285 { error_noexit "Copy to $DIR/$tfile file" ; return 0; }
289 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
290 do_facet ost1 "lctl set_param fail_loc=0x80000504"
292 # OST bulk will time out here, client resends
293 do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 1
294 do_facet ost1 lctl set_param fail_loc=0
295 # give recovery a chance to finish (shouldn't take long)
297 do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 2
301 run_test 16 "timeout bulk put, don't evict client (2732)"
306 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
308 local SAMPLE_FILE=$TMP/$tfile
309 do_facet_random_file client $SAMPLE_FILE 20K ||
310 { error_noexit "Create random file $SAMPLE_FILE" ; return 0; }
312 # With adaptive timeouts, bulk_get won't expire until adaptive_timeout_max
313 if at_is_enabled; then
314 at_max_saved=$(at_max_get ost1)
315 at_max_set $TIMEOUT ost1
318 # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
319 # OST bulk will time out here, client retries
320 do_facet ost1 lctl set_param fail_loc=0x80000503
321 # need to ensure we send an RPC
322 do_facet client cp $SAMPLE_FILE $DIR/$tfile
325 # with AT, client will wait adaptive_max*factor+net_latency before
326 # expiring the req, hopefully timeout*2 is enough
327 sleep $(($TIMEOUT*2))
329 do_facet ost1 lctl set_param fail_loc=0
330 do_facet client "df $DIR"
331 # expect cmp to succeed, client resent bulk
332 do_facet client "cmp $SAMPLE_FILE $DIR/$tfile" || return 3
333 do_facet client "rm $DIR/$tfile" || return 4
334 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved ost1
337 run_test 17 "timeout bulk get, don't evict client (2732)"
340 [ -z ${ost2_svc} ] && skip_env "needs 2 osts" && return 0
342 do_facet_create_file client $TMP/$tfile 20K ||
343 { error_noexit "Create file $TMP/$tfile" ; return 0; }
345 do_facet client mkdir -p $DIR/$tdir
349 pgcache_empty || return 1
352 $LFS setstripe -i 1 -c 1 $f
353 stripe_index=$($LFS getstripe -i $f)
354 if [ $stripe_index -ne 1 ]; then
356 error "$f: stripe_index $stripe_index != 1" && return
359 do_facet client cp $TMP/$tfile $f
361 local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | egrep -v 'MDT' | awk '{print $1}'`
362 $LCTL --device $osc2dev deactivate || return 3
363 # my understanding is that there should be nothing in the page
364 # cache after the client reconnects?
366 pgcache_empty || rc=2
367 $LCTL --device $osc2dev activate
371 run_test 18a "manual ost invalidate clears page cache immediately"
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
395 # allow recovery to complete
396 sleep $((TIMEOUT + 2))
397 # my understanding is that there should be nothing in the page
398 # cache after the client reconnects?
400 pgcache_empty || rc=2
404 run_test 18b "eviction and reconnect clears page cache (2766)"
407 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
409 do_facet_create_file client $TMP/$tfile 20K ||
410 { error_noexit "Create file $TMP/$tfile" ; return 0; }
412 do_facet client mkdir -p $DIR/$tdir
416 pgcache_empty || return 1
418 $LFS setstripe -i 0 -c 1 $f
419 stripe_index=$($LFS getstripe -i $f)
420 if [ $stripe_index -ne 0 ]; then
422 error "$f: stripe_index $stripe_index != 0" && return
425 do_facet client cp $TMP/$tfile $f
429 # OBD_FAIL_OST_CONNECT_NET2
430 # lost reply to connect request
431 do_facet ost1 lctl set_param fail_loc=0x80000225
434 df $MOUNT > /dev/null 2>&1
436 # my understanding is that there should be nothing in the page
437 # cache after the client reconnects?
439 pgcache_empty || rc=2
443 run_test 18c "Dropped connect reply after eviction handing (14755)"
446 local BEFORE=`date +%s`
449 mount_client $DIR2 || error "failed to mount $DIR2"
451 # cancel cached locks from OST to avoid eviction from it
454 do_facet client "stat $DIR > /dev/null" ||
455 error "failed to stat $DIR: $?"
456 drop_ldlm_cancel "chmod 0777 $DIR2" ||
457 error "failed to chmod $DIR2"
461 # let the client reconnect
463 EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
464 awk -F"[ [,]" '/EVICTED ]$/ \
465 { if (mx<$5) {mx=$5;} } END { print mx }')
467 [ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
468 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
469 error "no eviction: $EVICT before:$BEFORE")
471 run_test 19a "test expired_lock_main on mds (2867)"
474 local BEFORE=`date +%s`
477 mount_client $DIR2 || error "failed to mount $DIR2: $?"
479 # cancel cached locks from MDT to avoid eviction from it
482 do_facet client $MULTIOP $DIR/$tfile Ow ||
483 error "failed to run multiop: $?"
484 drop_ldlm_cancel $MULTIOP $DIR2/$tfile Ow ||
485 error "failed to ldlm_cancel: $?"
487 umount_client $DIR2 || error "failed to unmount $DIR2: $?"
488 do_facet client munlink $DIR/$tfile ||
489 error "failed to unlink $DIR/$tfile: $?"
491 # let the client reconnect
493 EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state |
494 awk -F"[ [,]" '/EVICTED ]$/ \
495 { if (mx < $5) {mx = $5;} } END { print mx }')
497 [ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
498 (do_facet client $LCTL get_param osc.$FSNAME-OST*.state;
499 error "no eviction: $EVICT before:$BEFORE")
501 run_test 19b "test expired_lock_main on ost (2867)"
504 local BEFORE=`date +%s`
507 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=0
509 mkdir -p $DIR1/$tfile
512 #define OBD_FAIL_PTLRPC_CANCEL_RESEND 0x516
513 do_facet mds $LCTL set_param fail_loc=0x80000516
515 touch $DIR2/$tfile/file1 &
517 # let touch to get blocked on the server
521 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=1
524 # let the client reconnect
526 EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
527 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
529 [ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
531 run_test 19c "check reconnect and lock resend do not trigger expired_lock_main"
533 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
534 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
537 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
538 multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
541 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
542 do_facet ost1 lctl set_param fail_loc=0x80000308
543 kill -USR1 $MULTI_PID
546 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
548 run_test 20a "ldlm_handle_enqueue error (should return error)"
550 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
551 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
554 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
556 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
557 do_facet ost1 lctl set_param fail_loc=0x80000308
558 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
559 error "didn't fail open enqueue" || true
561 run_test 20b "ldlm_handle_enqueue error (should return error)"
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=0x80000129"
570 $MULTIOP $DIR/$tdir-2/f Oc &
573 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
575 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
576 kill -USR1 $close_pid
578 wait $close_pid || return 1
579 wait $open_pid || return 2
580 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
582 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
583 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
587 run_test 21a "drop close request while close and open are both in flight"
590 mkdir -p $DIR/$tdir-1
591 mkdir -p $DIR/$tdir-2
592 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
595 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
596 mcreate $DIR/$tdir-2/f &
599 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
601 kill -USR1 $close_pid
603 wait $close_pid || return 1
604 wait $open_pid || return 3
606 $CHECKSTAT -t file $DIR/$tdir-1/f || return 4
607 $CHECKSTAT -t file $DIR/$tdir-2/f || return 5
610 run_test 21b "drop open request while close and open are both in flight"
613 mkdir -p $DIR/$tdir-1
614 mkdir -p $DIR/$tdir-2
615 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
618 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
619 mcreate $DIR/$tdir-2/f &
622 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
624 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
625 kill -USR1 $close_pid
627 wait $close_pid || return 1
628 wait $open_pid || return 2
630 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
632 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
633 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
636 run_test 21c "drop both request while close and open are both in flight"
639 mkdir -p $DIR/$tdir-1
640 mkdir -p $DIR/$tdir-2
641 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
644 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
645 $MULTIOP $DIR/$tdir-2/f Oc &
647 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
649 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
652 wait $pid || return 1
653 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
655 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
656 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
660 run_test 21d "drop close reply while close and open are both in flight"
663 mkdir -p $DIR/$tdir-1
664 mkdir -p $DIR/$tdir-2
665 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
668 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
669 touch $DIR/$tdir-2/f &
671 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
675 wait $pid || return 1
678 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
679 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
682 run_test 21e "drop open reply while close and open are both in flight"
685 mkdir -p $DIR/$tdir-1
686 mkdir -p $DIR/$tdir-2
687 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
690 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
691 touch $DIR/$tdir-2/f &
693 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
695 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
698 wait $pid || return 1
699 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
701 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
702 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
705 run_test 21f "drop both reply while close and open are both in flight"
708 mkdir -p $DIR/$tdir-1
709 mkdir -p $DIR/$tdir-2
710 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
713 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
714 touch $DIR/$tdir-2/f &
716 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
718 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
721 wait $pid || return 1
722 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
724 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
725 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
728 run_test 21g "drop open reply and close request while close and open are both in flight"
731 mkdir -p $DIR/$tdir-1
732 mkdir -p $DIR/$tdir-2
733 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
736 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
737 touch $DIR/$tdir-2/f &
740 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
742 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
745 wait $pid || return 1
746 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
748 wait $touch_pid || return 2
750 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
751 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
754 run_test 21h "drop open request and close reply while close and open are both in flight"
756 # bug 3462 - multiple MDC requests
761 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
766 $MULTIOP $f1 msu || return 1
769 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
771 wait $close_pid || return 2
772 rm -rf $f2 || return 4
774 run_test 22 "drop close request and do mknod"
777 multiop_bg_pause $DIR/$tfile O_c || return 1
779 # give a chance for open
783 drop_request "kill -USR1 $pid"
786 wait $pid || return 1
789 run_test 23 "client hang when close a file after mds crash"
791 test_24a() { # bug 11710 details correct fsync() behavior
792 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
795 $LFS setstripe -i 0 -c 1 $DIR/$tdir
797 multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
800 kill -USR1 $MULTI_PID
803 lctl set_param fail_loc=0x0
806 error_ignore bz5494 "multiop didn't fail fsync: rc $rc" || true
808 run_test 24a "fsync error (should return error)"
810 wait_client_evicted () {
813 local varsvc=${facet}_svc
815 wait_update $(facet_active_host $facet) \
816 "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \
821 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
825 lfs setstripe $DIR/$tdir -s 0 -i 0 -c 1
827 multiop_bg_pause $DIR/$tdir/$tfile-1 Ow8192_yc ||
828 error "mulitop Ow8192_yc failed"
831 multiop_bg_pause $DIR/$tdir/$tfile-2 Ow8192_c ||
832 error "mulitop Ow8192_c failed"
837 kill -USR1 $MULTI_PID1
840 kill -USR1 $MULTI_PID2
843 lctl set_param fail_loc=0x0
845 [ $rc1 -eq 0 -o $rc2 -eq 0 ] &&
846 error_ignore bz5494 "multiop didn't fail fsync: $rc1 or close: $rc2" ||
849 dmesg | grep "dirty page discard:" ||
850 error "no discarded dirty page found!"
852 run_test 24b "test dirty page discard due to client eviction"
854 test_26a() { # was test_26 bug 5921 - evict dead exports by pinger
855 # this test can only run from a client on a separate node.
856 remote_ost || { skip "local OST" && return 0; }
857 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
858 remote_mds || { skip "local MDS" && return 0; }
860 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
861 skip "msg and ost1 are at the same node"
865 check_timeout || return 1
867 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
869 echo starting with $OST_NEXP OST exports
870 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
871 do_facet client lctl set_param fail_loc=0x505
872 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
873 # But if there's a race to start the evictor from various obds,
874 # the loser might have to wait for the next ping.
877 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2 + TIMEOUT * 3 / 4))
879 do_facet client lctl set_param fail_loc=0x0
880 [ $rc -eq 0 ] || error "client not evicted from OST"
882 run_test 26a "evict dead exports"
884 test_26b() { # bug 10140 - evict dead exports by pinger
885 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
887 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
888 skip "msg and ost1 are at the same node"
892 check_timeout || return 1
894 zconf_mount `hostname` $MOUNT2 ||
895 { error "Failed to mount $MOUNT2"; return 2; }
896 sleep 1 # wait connections being established
898 local MDS_NEXP=$(do_facet $SINGLEMDS lctl get_param -n mdt.${mds1_svc}.num_exports | cut -d' ' -f2)
899 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
901 echo starting with $OST_NEXP OST and $MDS_NEXP MDS exports
903 zconf_umount `hostname` $MOUNT2 -f
905 # PING_INTERVAL max(obd_timeout / 4, 1U)
906 # PING_EVICT_TIMEOUT (PING_INTERVAL * 6)
908 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
909 # But if there's a race to start the evictor from various obds,
910 # the loser might have to wait for the next ping.
911 # = 9 * PING_INTERVAL + PING_INTERVAL
912 # = 10 PING_INTERVAL = 10 obd_timeout / 4 = 2.5 obd_timeout
913 # let's wait $((TIMEOUT * 3)) # bug 19887
915 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 3)) || \
916 error "Client was not evicted by ost" rc=1
917 wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 3)) || \
918 error "Client was not evicted by mds"
920 run_test 26b "evict dead exports"
924 writemany -q -a $DIR/$tdir/$tfile 0 5 &
927 local save_FAILURE_MODE=$FAILURE_MODE
929 facet_failover $SINGLEMDS
930 #define OBD_FAIL_OSC_SHUTDOWN 0x407
931 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000407
932 # need to wait for reconnect
933 echo waiting for fail_loc
934 wait_update_facet $SINGLEMDS "lctl get_param -n fail_loc" "-2147482617"
935 facet_failover $SINGLEMDS
937 kill -USR1 $CLIENT_PID
940 FAILURE_MODE=$save_FAILURE_MODE
942 run_test 27 "fail LOV while using OSC's"
944 test_28() { # bug 6086 - error adding new clients
945 do_facet client mcreate $DIR/$tfile || return 1
946 drop_bl_callback_once "chmod 0777 $DIR/$tfile" ||
947 echo "evicted as expected"
948 #define OBD_FAIL_MDS_CLIENT_ADD 0x12f
949 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
950 # fail once (evicted), reconnect fail (fail_loc), ok
951 client_up || (sleep 10; client_up) || (sleep 10; client_up) || error "reconnect failed"
953 fail $SINGLEMDS # verify MDS last_rcvd can be loaded
955 run_test 28 "handle error adding new clients (bug 6086)"
957 test_29a() { # bug 22273 - error adding new clients
958 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
959 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000711"
960 # fail abort so client will be new again
961 fail_abort $SINGLEMDS
962 client_up || error "reconnect failed"
963 wait_osc_import_state $SINGLEMDS ost FULL
966 run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
968 test_29b() { # bug 22273 - error adding new clients
969 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
970 do_facet ost1 "lctl set_param fail_loc=0x80000711"
971 # fail abort so client will be new again
973 client_up || error "reconnect failed"
976 run_test 29b "error adding new clients doesn't cause LBUG (bug 22273)"
980 # put a load of file creates/writes/deletes
981 writemany -q $DIR/$tdir/$tfile 0 5 &
983 echo writemany pid $CLIENT_PID
987 # wait for client to reconnect to MDS
992 # client process should see no problems even though MDS went down
994 kill -USR1 $CLIENT_PID
997 echo writemany returned $rc
998 #these may fail because of eviction due to slow AST response.
1000 error_ignore bz13652 "writemany returned rc $rc" || true
1002 run_test 50 "failover MDS under load"
1005 #define OBD_FAIL_MDS_SYNC_CAPA_SL 0x1310
1006 do_facet ost1 lctl set_param fail_loc=0x00001310
1009 # put a load of file creates/writes/deletes
1010 writemany -q $DIR/$tdir/$tfile 0 5 &
1014 facet_failover $SINGLEMDS
1015 # failover at various points during recovery
1016 SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
1017 echo will failover at $SEQ
1020 #echo failover in $i sec
1021 log "test_$testnum: failover in $i sec"
1023 facet_failover $SINGLEMDS
1025 # client process should see no problems even though MDS went down
1026 # and recovery was interrupted
1028 kill -USR1 $CLIENT_PID
1031 echo writemany returned $rc
1033 error_ignore bz13652 "writemany returned rc $rc" || true
1035 run_test 51 "failover MDS during recovery"
1038 do_facet client "mkdir -p $DIR/$tdir"
1039 do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
1041 echo writemany pid $CLIENT_PID
1046 wait $CLIENT_PID || rc=$?
1047 # active client process should see an EIO for down OST
1048 [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
1049 # but timing or failover setup may allow success
1050 [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
1051 echo "writemany returned $rc"
1056 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1061 [ $rc -ne 0 ] && { return $rc; }
1062 # wait for client to reconnect to OST
1066 [ $rc -ne 0 ] && { return $rc; }
1073 run_test 52 "failover OST under load"
1075 # test of open reconstruct
1078 drop_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
1081 run_test 53 "touch: drop rep"
1084 zconf_mount `hostname` $MOUNT2
1086 touch $DIR2/$tfile.1
1088 cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
1091 ERROR=`dmesg | egrep "(test 54|went back in time)" | tail -n1 | grep "went back in time"`
1092 [ x"$ERROR" == x ] || error "back in time occured"
1094 run_test 54 "back in time"
1096 # bug 11330 - liblustre application death during I/O locks up OST
1098 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1102 # Minimum pass speed is 2MBps
1104 # LU-2887/LU-3089 - set min pass speed to 500KBps
1105 [ "$(facet_fstype ost1)" = "zfs" ] && ddtimeout=256
1107 # first dd should be finished quickly
1108 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
1109 dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
1112 echo "step1: testing ......"
1113 while kill -0 $DDPID 2> /dev/null; do
1115 if [ $count -gt $ddtimeout ]; then
1116 error "dd should be finished!"
1120 echo "(dd_pid=$DDPID, time=$count)successful"
1122 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2
1123 #define OBD_FAIL_OST_DROP_REQ 0x21d
1124 do_facet ost1 lctl set_param fail_loc=0x0000021d
1125 # second dd will be never finished
1126 dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 &
1129 echo "step2: testing ......"
1130 while [ $count -le $ddtimeout ]; do
1131 if ! kill -0 $DDPID 2> /dev/null; then
1133 error "dd shouldn't be finished! (time=$count)"
1138 echo "(dd_pid=$DDPID, time=$count)successful"
1140 #Recover fail_loc and dd will finish soon
1141 do_facet ost1 lctl set_param fail_loc=0
1143 echo "step3: testing ......"
1144 while kill -0 $DDPID 2> /dev/null; do
1146 if [ $count -gt $((ddtimeout + 440)) ]; then
1147 error "dd should be finished!"
1151 echo "(dd_pid=$DDPID, time=$count)successful"
1155 run_test 55 "ost_brw_read/write drops timed-out read/write request"
1157 test_56() { # b=11277
1158 #define OBD_FAIL_MDS_RESEND 0x136
1160 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
1161 stat $DIR/$tfile || error "stat failed"
1162 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1165 run_test 56 "do not fail on getattr resend"
1168 # no oscs means no client or mdt
1169 while lctl get_param osc.*.* > /dev/null 2>&1; do
1170 : # loop until proc file is removed
1174 test_57() { # bug 10866
1178 #define OBD_FAIL_LPROC_REMOVE 0xB00
1179 lctl set_param fail_loc=0x80000B00
1180 zconf_umount `hostname` $DIR
1181 lctl set_param fail_loc=0x80000B00
1182 fail_abort $SINGLEMDS
1184 lctl set_param fail_loc=0
1186 do_facet client "df $DIR"
1188 run_test 57 "read procfs entries causes kernel crash"
1190 test_58() { # bug 11546
1191 #define OBD_FAIL_MDC_ENQUEUE_PAUSE 0x801
1194 lctl set_param fail_loc=0x80000801
1195 cp $DIR/$tfile /dev/null &
1198 lctl set_param fail_loc=0
1199 drop_bl_callback_once rm -f $DIR/$tfile
1201 # the first 'df' could tigger the eviction caused by
1202 # 'drop_bl_callback_once', and it's normal case.
1203 # but the next 'df' should return successfully.
1204 do_facet client "df $DIR" || do_facet client "df $DIR"
1206 run_test 58 "Eviction in the middle of open RPC reply processing"
1208 test_59() { # bug 10589
1209 zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
1210 echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
1211 #define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311
1212 lctl set_param fail_loc=0x311
1213 writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
1214 [ $? = 0 ] || error "dd write failed"
1215 writes=$(echo $writes | awk -F '+' '/out/ {print $1}')
1216 lctl set_param fail_loc=0
1218 zconf_umount `hostname` $MOUNT2 -f
1219 reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
1220 [ $? = 0 ] || error "dd read failed"
1221 reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
1222 [ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
1224 run_test 59 "Read cancel race on client eviction"
1227 # we assume that all md changes are in the MDT0 changelog
1228 if [ $MDSCOUNT -gt 1 ]; then
1229 error_ignore bz17935 $*
1236 MDT0=$($LCTL get_param -n mdc.*.mds_server_uuid |
1237 awk '{ gsub(/_UUID/,""); print $1 }' | head -n1)
1242 # Register (and start) changelog
1243 USER=$(do_facet $SINGLEMDS lctl --device $MDT0 changelog_register -n)
1244 echo "Registered as $MDT0 changelog user $USER"
1246 # Generate a large number of changelog entries
1247 createmany -o $DIR/$tdir/$tfile $NUM_FILES
1251 # Unlink files in the background
1252 unlinkmany $DIR/$tdir/$tfile $NUM_FILES &
1256 # Failover the MDS while unlinks are happening
1257 facet_failover $SINGLEMDS
1259 # Wait for unlinkmany to finish
1262 # Check if all the create/unlink events were recorded
1264 $LFS changelog $MDT0 >> $DIR/$tdir/changelog
1265 local cl_count=$(grep UNLNK $DIR/$tdir/changelog | wc -l)
1266 echo "$cl_count unlinks in $MDT0 changelog"
1268 do_facet $SINGLEMDS lctl --device $MDT0 changelog_deregister $USER
1269 USERS=$(( $(do_facet $SINGLEMDS lctl get_param -n \
1270 mdd.$MDT0.changelog_users | wc -l) - 2 ))
1271 if [ $USERS -eq 0 ]; then
1272 [ $cl_count -eq $NUM_FILES ] || \
1273 err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
1274 # Also make sure we can clear large changelogs
1275 cl_count=$($LFS changelog $FSNAME | wc -l)
1276 [ $cl_count -le 2 ] || \
1277 error "Changelog not empty: $cl_count entries"
1279 # If there are other users, there may be other unlinks in the log
1280 [ $cl_count -ge $NUM_FILES ] || \
1281 err17935 "Recorded ${cl_count} unlinks out of $NUM_FILES"
1282 echo "$USERS other changelog users; can't verify clear"
1285 run_test 60 "Add Changelog entries during MDS failover"
1289 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0000)
1290 mdtosc=${mdtosc/-MDT*/-MDT\*}
1291 local cflags="osc.$mdtosc.connect_flags"
1292 do_facet $SINGLEMDS "lctl get_param -n $cflags" |grep -q skip_orphan
1293 [ $? -ne 0 ] && skip "don't have skip orphan feature" && return
1295 mkdir -p $DIR/$tdir || error "mkdir dir $DIR/$tdir failed"
1296 # Set the default stripe of $DIR/$tdir to put the files to ost1
1297 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1299 replay_barrier $SINGLEMDS
1300 createmany -o $DIR/$tdir/$tfile-%d 10
1301 local oid=`do_facet ost1 "lctl get_param -n obdfilter.${ost1_svc}.last_id"`
1303 fail_abort $SINGLEMDS
1305 touch $DIR/$tdir/$tfile
1306 local id=`$LFS getstripe $DIR/$tdir/$tfile | awk '$1 == 0 { print $2 }'`
1307 [ $id -le $oid ] && error "the orphan objid was reused, failed"
1312 run_test 61 "Verify to not reuse orphan objects - bug 17025"
1314 # test_62 as seen it b2_1 please do not reuse test_62
1317 # zconf_umount `hostname` $DIR
1318 # #define OBD_FAIL_PTLRPC_DELAY_IMP_FULL 0x516
1319 # lctl set_param fail_loc=0x516
1322 #run_test 62 "Verify connection flags race - bug LU-1716"
1324 check_cli_ir_state()
1326 local NODE=${1:-$HOSTNAME}
1328 st=$(do_node $NODE "lctl get_param mgc.*.ir_state |
1329 awk '/imperative_recovery:/ { print \\\$2}'")
1330 [ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
1331 error "Error state $st, must be ENABLED or DISABLED"
1335 check_target_ir_state()
1338 local name=${target}_svc
1339 local recovery_proc=obdfilter.${!name}.recovery_status
1342 st=$(do_facet $target "lctl get_param -n $recovery_proc |
1343 awk '/IR:/{ print \\\$2}'")
1344 [ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
1345 error "Error state $st, must be ENABLED or DISABLED"
1351 do_facet mgs lctl set_param -n mgs.MGS.live.$FSNAME="state=$1"
1356 local state=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1357 awk '/state:/{ print \\\$2 }'")
1361 nidtbl_version_mgs()
1363 local ver=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1364 awk '/nidtbl_version:/{ print \\\$2 }'")
1368 # nidtbl_version_client <mds1|client> [node]
1369 nidtbl_version_client()
1372 local node=${2:-$HOSTNAME}
1374 if [ X$cli = Xclient ]; then
1377 local obdtype=${cli/%[0-9]*/}
1378 [ $obdtype != mds ] && error "wrong parameters $cli"
1380 node=$(facet_active_host $cli)
1385 local vers=$(do_node $node "lctl get_param -n mgc.*.ir_state" |
1386 awk "/$cli/{print \$6}" |sort -u)
1388 # in case there are multiple mounts on the client node
1390 [ ${#arr[@]} -ne 1 ] && error "versions on client node mismatch"
1394 nidtbl_versions_match()
1396 [ $(nidtbl_version_mgs) -eq $(nidtbl_version_client ${1:-client}) ]
1399 target_instance_match()
1405 obdtype=${srv/%[0-9]*/}
1416 error "invalid target type" $srv
1421 local target=${srv}_svc
1422 local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance)
1423 local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import |
1424 awk '/instance/{ print $2 }' | head -n1)
1426 return $([ $si -eq $ci ])
1431 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1432 { skip "MGS without IR support"; return 0; }
1434 # MDT was just restarted in the previous test, make sure everything
1437 while [ $cnt -gt 0 ]; do
1438 nidtbl_versions_match && break
1444 set_ir_status disabled
1446 local prev_ver=$(nidtbl_version_client client)
1448 local saved_FAILURE_MODE=$FAILURE_MODE
1449 [ $(facet_host mgs) = $(facet_host ost1) ] && FAILURE_MODE="SOFT"
1453 [ $(nidtbl_version_client client) -eq $prev_ver ] ||
1454 error "version must not change due to IR disabled"
1455 target_instance_match ost1 || error "instance mismatch"
1459 FAILURE_MODE=$saved_FAILURE_MODE
1461 run_test 100 "IR: Make sure normal recovery still works w/o IR"
1465 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1466 { skip "MGS without IR support"; return 0; }
1470 local OST1_IMP=$(get_osc_import_name client ost1)
1472 # disable pinger recovery
1473 lctl set_param -n osc.$OST1_IMP.pinger_recov=0
1477 target_instance_match ost1 || error "instance mismatch"
1478 nidtbl_versions_match || error "version must match"
1480 lctl set_param -n osc.$OST1_IMP.pinger_recov=1
1482 run_test 101 "IR: Make sure IR works w/o normal recovery"
1486 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1487 { skip "MGS without IR support"; return 0; }
1489 local clients=${CLIENTS:-$HOSTNAME}
1496 # let's have a new nidtbl version
1499 # sleep for a while so that clients can see the failure of ost
1500 # it must be MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC.
1501 # int mgc_request.c:
1502 # define MGC_TIMEOUT_MIN_SECONDS 5
1503 # define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 *
1504 local count=30 # 20 seconds at most
1505 while [ $count -gt 0 ]; do
1506 nidtbl_versions_match && break
1511 nidtbl_versions_match || error "nidtbl mismatch"
1514 old_version=$(nidtbl_version_client client)
1516 zconf_umount_clients $clients $MOUNT || error "Cannot umount client"
1519 combined_mgs_mds && mgsdev=mds1
1520 remount_facet $mgsdev
1523 zconf_mount_clients $clients $MOUNT || error "Cannot mount client"
1526 new_version=$(nidtbl_version_client client)
1527 [ $new_version -lt $old_version ] &&
1528 error "nidtbl version wrong after mgs restarts"
1531 run_test 102 "IR: New client gets updated nidtbl after MGS restart"
1535 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1536 { skip "MGS without IR support"; return 0; }
1538 combined_mgs_mds && skip "mgs and mds on the same target" && return 0
1540 # workaround solution to generate config log on the mds
1546 # We need this test because mds is like a client in IR context.
1547 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS ||
1548 error "MDS should start w/o mgs"
1550 # start mgs and remount mds w/ ir
1551 start mgs $(mgsdevname) $MGS_MOUNT_OPTS
1554 # remount client so that fsdb will be created on the MGS
1555 umount_client $MOUNT || error "umount failed"
1556 mount_client $MOUNT || error "mount failed"
1558 # sleep 30 seconds so the MDS has a chance to detect MGS restarting
1560 while [ $count -gt 0 ]; do
1561 [ $(nidtbl_version_client mds1) -ne 0 ] && break
1566 # after a while, mds should be able to reconnect to mgs and fetch
1567 # up-to-date nidtbl version
1568 nidtbl_versions_match mds1 || error "mds nidtbl mismatch"
1573 run_test 103 "IR: MDS can start w/o MGS and get updated nidtbl later"
1577 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1578 { skip "MGS without IR support"; return 0; }
1583 start ost1 $(ostdevname 1) "$OST_MOUNT_OPTS -onoir" ||
1584 error "OST1 cannot start"
1587 local ir_state=$(check_target_ir_state ost1)
1588 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1589 error "ir status on ost1 should be DISABLED"
1591 run_test 104 "IR: ost can disable IR voluntarily"
1595 [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
1596 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1597 { skip "MGS without IR support"; return 0; }
1601 # get one of the clients from client list
1602 local rcli=$(echo $RCLIENTS |cut -d' ' -f 1)
1604 local mount_opts=${MOUNT_OPTS:+$MOUNT_OPTS,}noir
1605 zconf_umount $rcli $MOUNT || error "umount failed"
1606 zconf_mount $rcli $MOUNT $mount_opts || error "mount failed"
1608 # make sure lustre mount at $rcli disabling IR
1609 local ir_state=$(check_cli_ir_state $rcli)
1610 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1611 error "IR state must be DISABLED at $rcli"
1613 # Since the client just mounted, its last_rcvd entry is not on disk.
1614 # Send an RPC so exp_need_sync forces last_rcvd to commit this export
1615 # so the client can reconnect during OST recovery (LU-924, LU-1582)
1616 $SETSTRIPE -i 0 $DIR/$tfile
1617 dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=sync
1619 # make sure MGS's state is Partial
1620 [ $(get_ir_status) = "partial" ] || error "MGS IR state must be partial"
1623 # make sure IR on ost1 is DISABLED
1624 local ir_state=$(check_target_ir_state ost1)
1625 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1626 error "IR status on ost1 should be DISABLED"
1628 # remount with the default MOUNT_OPTS
1629 zconf_umount $rcli $MOUNT || error "umount failed"
1630 zconf_mount $rcli $MOUNT || error "mount failed"
1632 # make sure MGS's state is full
1633 [ $(get_ir_status) = "full" ] || error "MGS IR status must be full"
1636 # make sure IR on ost1 is ENABLED
1637 local ir_state=$(check_target_ir_state ost1)
1638 [ $ir_state = "ENABLED" -o $ir_state = "ON" ] ||
1639 error "IR status on ost1 should be ENABLED"
1643 run_test 105 "IR: NON IR clients support"
1651 test_106() { # LU-1789
1652 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.50) ]] ||
1653 { skip "Need MDS version at least 2.3.50"; return 0; }
1655 #define OBD_FAIL_MDC_LIGHTWEIGHT 0x805
1656 $LCTL set_param fail_loc=0x805
1659 trap cleanup_106 EXIT
1661 # enable lightweight flag on mdc connection
1664 local MDS_NEXP=$(do_facet $SINGLEMDS \
1665 lctl get_param -n mdt.${mds1_svc}.num_exports |
1667 $LCTL set_param fail_loc=0
1669 touch $DIR2/$tfile || error "failed to create empty file"
1670 replay_barrier $SINGLEMDS
1672 $LCTL set_param debug=console
1674 facet_failover $SINGLEMDS
1676 # lightweight connection must be evicted
1677 touch -c $DIR2/$tfile || true
1678 $LCTL dk $TMP/lustre-log-$TESTNAME.log
1679 evicted=`awk '/This client was evicted by .*MDT0000/ {
1681 }' $TMP/lustre-log-$TESTNAME.log`
1682 [ -z "$evicted" ] && error "lightweight client not evicted by mds"
1684 # and all operations performed by lightweight client should be
1685 # synchronous, so the file created before mds restart should be there
1686 $CHECKSTAT -t file $DIR/$tfile || error "file not present"
1691 run_test 106 "lightweight connection support"
1698 # OBD_FAIL_MDS_REINT_NET_REP 0x119
1699 do_facet $SINGLEMDS lctl set_param fail_loc=0x119
1700 multiop $DIR/$tdir D_c &
1702 mkdir $DIR/$tdir/dir_106 &
1704 do_facet $SINGLEMDS lctl set_param fail_loc=0
1707 wait $CLIENT_PID || rc=$?
1708 checkstat -t dir $DIR/$tdir/dir_106 || return 1
1710 kill -USR1 $close_pid
1711 wait $close_pid || return 2
1715 run_test 107 "drop reint reply, then restart MDT"
1718 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1719 local remote_dir=$DIR/$tdir/remote_dir
1723 #prepare for 110 test, which need set striped dir on remote MDT.
1724 for num in $(seq $MDSCOUNT); do
1726 lctl set_param -n mdt.${FSNAME}*.enable_remote_dir=1 \
1731 drop_request "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
1732 error "lfs mkdir failed"
1733 local diridx=$($GETSTRIPE -M $remote_dir)
1734 [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
1736 rm -rf $DIR/$tdir || error "rmdir failed"
1738 run_test 110a "create remote directory: drop client req"
1741 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1742 local remote_dir=$DIR/$tdir/remote_dir
1746 drop_reint_reply "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
1747 error "lfs mkdir failed"
1749 diridx=$($GETSTRIPE -M $remote_dir)
1750 [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
1752 rm -rf $DIR/$tdir || error "rmdir failed"
1754 run_test 110b "create remote directory: drop Master rep"
1757 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1758 local remote_dir=$DIR/$tdir/remote_dir
1762 drop_update_reply $MDTIDX "$LFS mkdir -i $MDTIDX -c2 $remote_dir" ||
1763 error "lfs mkdir failed"
1765 diridx=$($GETSTRIPE -M $remote_dir)
1766 [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
1768 rm -rf $DIR/$tdir || error "rmdir failed"
1770 run_test 110c "create remote directory: drop update rep on slave MDT"
1773 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1774 local remote_dir=$DIR/$tdir/remote_dir
1778 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
1780 drop_request "rm -rf $remote_dir" || error "rm remote dir failed"
1782 rm -rf $DIR/$tdir || error "rmdir failed"
1786 run_test 110d "remove remote directory: drop client req"
1789 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1790 local remote_dir=$DIR/$tdir/remote_dir
1794 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
1795 drop_reint_reply "rm -rf $remote_dir" || error "rm remote dir failed"
1797 rm -rf $DIR/$tdir || error "rmdir failed"
1801 run_test 110e "remove remote directory: drop master rep"
1804 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1805 local remote_dir=$DIR/$tdir/remote_dir
1809 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
1810 drop_update_reply $MDTIDX "rm -rf $remote_dir" ||
1811 error "rm remote dir failed"
1813 rm -rf $DIR/$tdir || error "rmdir failed"
1815 run_test 110f "remove remote directory: drop slave rep"
1818 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
1819 local remote_dir=$DIR/$tdir/remote_dir
1822 mkdir -p $remote_dir
1824 createmany -o $remote_dir/f 100
1826 #define OBD_FAIL_MIGRATE_NET_REP 0x1702
1827 do_facet mds$MDTIDX lctl set_param fail_loc=0x1702
1828 $LFS mv -M $MDTIDX $remote_dir || error "migrate failed"
1829 do_facet mds$MDTIDX lctl set_param fail_loc=0x0
1831 for file in $(find $remote_dir); do
1832 mdt_index=$($LFS getstripe -M $file)
1833 [ $mdt_index == $MDTIDX ] ||
1834 error "$file is not on MDT${MDTIDX}"
1837 rm -rf $DIR/$tdir || error "rmdir failed"
1839 run_test 110g "drop reply during migration"
1841 # LU-2844 mdt prepare fail should not cause umount oops
1844 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.62) ]] ||
1845 { skip "Need MDS version at least 2.3.62"; return 0; }
1847 local mdsdev=$(mdsdevname ${SINGLEMDS//mds/})
1848 #define OBD_FAIL_MDS_CHANGELOG_INIT 0x151
1849 do_facet $SINGLEMDS lctl set_param fail_loc=0x151
1850 stop $SINGLEMDS || error "stop MDS failed"
1851 start $SINGLEMDS $mdsdev && error "start MDS should fail"
1852 do_facet $SINGLEMDS lctl set_param fail_loc=0
1853 start $SINGLEMDS $mdsdev || error "start MDS failed"
1855 run_test 111 "mdd setup fail should not cause umount oops"
1859 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1861 do_facet_random_file client $TMP/$tfile 100K ||
1862 error_noexit "Create random file $TMP/$tfile"
1864 pause_bulk "cp $TMP/$tfile $DIR/$tfile" $TIMEOUT ||
1865 error_noexit "Can't pause_bulk copy"
1868 # expect cmp to succeed, client resent bulk
1869 cmp $TMP/$tfile $DIR/$tfile ||
1870 error_noexit "Wrong data has been written"
1872 error_noexit "Can't remove file"
1875 run_test 112a "bulk resend while orignal request is in progress"
1877 # parameters: fail_loc CMD RC
1884 #define OBD_FAIL_LDLM_CP_CB_WAIT2 0x320
1885 #define OBD_FAIL_LDLM_CP_CB_WAIT3 0x321
1886 #define OBD_FAIL_LDLM_CP_CB_WAIT4 0x322
1887 #define OBD_FAIL_LDLM_CP_CB_WAIT5 0x323
1890 echo -n "** FLOCK REPLY vs. EVICTION race, lock $2"
1891 [ "$1" = "CLEANUP" ] &&
1892 fail=0x80000320 && echo ", $1 cp first"
1893 [ "$1" = "REPLY" ] &&
1894 fail=0x80000321 && echo ", $1 cp first"
1895 [ "$1" = "DEADLOCK CLEANUP" ] &&
1896 fail=0x80000322 && echo " DEADLOCK, CLEANUP cp first"
1897 [ "$1" = "DEADLOCK REPLY" ] &&
1898 fail=0x80000323 && echo " DEADLOCK, REPLY cp first"
1900 if [ x"$2" = x"get" ]; then
1901 #for TEST lock, take a conflict in advance
1902 # sleep longer than evictor to not confuse fail_loc: 2+2+4
1903 echo "** Taking conflict **"
1904 flocks_test 5 set read sleep 10 $DIR/$tfile &
1910 $LCTL set_param fail_loc=$fail
1912 flocks_test 5 $2 write $DIR/$tfile &
1916 echo "** Evicting and re-connecting client **"
1921 if [ x"$2" = x"get" ]; then
1928 # check if the return value is allowed
1929 [ $rc -eq $3 ] && rc=0
1931 $LCTL set_param fail_loc=0
1935 # a lock is taken, unlock vs. cleanup_resource() race for destroying
1936 # the ORIGINAL lock.
1941 flocks_test 5 set write sleep 4 $DIR/$tfile &
1945 # let unlock to sleep in CP CB
1946 $LCTL set_param fail_loc=$1
1949 # let cleanup to cleep in CP CB
1957 $LCTL set_param fail_loc=0
1962 flock_is_enabled || { skip "mount w/o flock enabled" && return; }
1965 test_120_reply "CLEANUP" set 5 || error "SET race failed"
1966 test_120_reply "CLEANUP" get 5 || error "GET race failed"
1967 test_120_reply "CLEANUP" unlock 5 || error "UNLOCK race failed"
1969 test_120_reply "REPLY" set 5 || error "SET race failed"
1970 test_120_reply "REPLY" get 5 || error "GET race failed"
1971 test_120_reply "REPLY" unlock 5 || error "UNLOCK race failed"
1974 test_120_reply "DEADLOCK CLEANUP" set 5 || error "DEADLOCK race failed"
1975 test_120_reply "DEADLOCK REPLY" set 35 || error "DEADLOCK race failed"
1977 test_120_destroy 0x320 || error "unlock-cleanup race failed"
1979 run_test 120 "flock race: completion vs. evict"
1982 local BEFORE=$(date +%s)
1985 # modify dir so that next revalidate would not obtain UPDATE lock
1988 # drop 1 reply with UPDATE lock,
1989 # resend should not create 2nd lock on server
1990 mcreate $DIR/$tfile || error "mcreate failed: $?"
1991 drop_ldlm_reply_once "stat $DIR/$tfile" || error "stat failed: $?"
1993 # 2 BL AST will be sent to client, both must find the same lock,
1994 # race them to not get EINVAL for 2nd BL AST
1995 #define OBD_FAIL_LDLM_PAUSE_CANCEL2 0x31f
1996 $LCTL set_param fail_loc=0x8000031f
1998 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=0 > /dev/null
1999 chmod 0777 $DIR/$tfile || error "chmod failed: $?"
2000 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=1 > /dev/null
2002 # let the client reconnect
2004 EVICT=$($LCTL get_param mdc.$FSNAME-MDT*.state |
2005 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
2007 [ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
2009 run_test 113 "ldlm enqueue dropped reply should not cause deadlocks"
2012 check_and_cleanup_lustre