5 PTLDEBUG=${PTLDEBUG:--1}
6 LUSTRE=${LUSTRE:-$(dirname $0)/..}
7 . $LUSTRE/tests/test-framework.sh
11 ALWAYS_EXCEPT="$RECOVERY_SMALL_EXCEPT "
14 always_except LU-17141 133
19 require_dsh_mds || exit 0
21 # Allow us to override the setup if we already have a mounted system by
22 # setting SETUP=" " and CLEANUP=" "
24 CLEANUP=${CLEANUP:-""}
26 check_and_setup_lustre
29 rm -rf $DIR/d[0-9]* $DIR/f.${TESTSUITE}*
32 local f1="$DIR/$tfile"
33 local f2="$DIR/$tfile.2"
35 drop_request "mcreate $f1" ||
36 error_noexit "create '$f1': drop req"
38 drop_reint_reply "mcreate $f2" ||
39 error_noexit "create '$f2': drop rep"
41 drop_request "tchmod 111 $f2" ||
42 error_noexit "chmod '$f2': drop req"
44 drop_reint_reply "tchmod 666 $f2" ||
45 error_noexit "chmod '$f2': drop rep"
47 drop_request "statone $f2" ||
48 error_noexit "stat '$f2': drop req"
50 drop_reply "statone $f2" ||
51 error_noexit "stat '$f2': drop rep"
53 run_test 1 "create, chmod, stat: drop req, drop rep"
57 do_facet_create_file client $t 10K ||
58 error_noexit "Create file $t"
60 drop_request "cat $t > /dev/null" ||
61 error_noexit "Open request for $t file"
63 drop_reply "cat $t > /dev/null" ||
64 error_noexit "Open replay for $t file"
66 run_test 4 "open: drop req, drop rep"
71 local RR="$T-renamed-again"
72 do_facet_create_file client $T 10K ||
73 error_noexit "Create file $T"
75 drop_request "mv $T $R" ||
76 error_noexit "Rename $T"
78 drop_reint_reply "mv $R $RR" ||
79 error_noexit "Failed rename replay on $R"
81 do_facet client "checkstat -v $RR" ||
82 error_noexit "checkstat error on $RR"
84 do_facet client "rm $RR" ||
85 error_noexit "Can't remove file $RR"
87 run_test 5 "rename: drop req, drop rep"
91 local LINK1=$DIR/$tfile.link1
92 local LINK2=$DIR/$tfile.link2
94 do_facet_create_file client $T 10K || error_noexit "Create file $T"
95 drop_request "link $T $LINK1" || error_noexit "link request for $T"
96 drop_reint_reply "link $T $LINK2" || error_noexit "link reply for $T"
97 drop_request "unlink $LINK1" || error_noexit "unlink request for $T"
98 drop_reint_reply "unlink $LINK2" || error_noexit "unlink reply for $T"
99 do_facet client "rm $T" || error_noexit "Can't remove file $T"
101 run_test 6 "link, unlink: drop req, drop rep"
105 drop_reint_reply "touch $DIR/$tfile" || return 1
107 run_test 8 "touch: drop rep (bug 1423)"
111 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
115 do_facet_random_file client $TMP/$tfile 1K ||
116 error_noexit "Create random file $TMP/$tfile"
117 # make this big, else test 9 doesn't wait for bulk -- bz 5595
118 do_facet_create_file client $TMP/$t1 4M ||
119 error_noexit "Create file $TMP/$t1"
120 do_facet client "cp $TMP/$t1 $DIR/$t1" ||
121 error_noexit "Can't copy to $DIR/$t1 file"
122 pause_bulk "cp $TMP/$tfile $DIR/$tfile" ||
123 error_noexit "Can't pause_bulk copy"
124 do_facet client "cp $TMP/$t1 $DIR/$t2" ||
125 error_noexit "Can't copy file"
126 do_facet client "sync"
127 do_facet client "rm $DIR/$tfile $DIR/$t2 $DIR/$t1" ||
128 error_noexit "Can't remove files"
129 do_facet client "rm $TMP/$t1 $TMP/$tfile"
131 run_test 9 "pause bulk on OST (bug 1420)"
135 local before=$(date +%s)
138 do_facet client "stat $DIR > /dev/null" ||
139 error "failed to stat $DIR: $?"
140 drop_bl_callback "chmod 0777 $DIR" ||
141 error "failed to chmod $DIR: $?"
143 # let the client reconnect
145 evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
146 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
147 [ ! -z "$evict" ] && [[ $evict -gt $before ]] ||
148 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
149 error "no eviction: $evict before:$before")
151 do_facet client checkstat -v -p 0777 $DIR ||
152 error "client checkstat failed: $?"
154 run_test 10a "finish request on server after client eviction (bug 1521)"
157 local before=$(date +%s)
160 [[ "$MDS1_VERSION" -lt $(version_code 2.6.53) ]] &&
161 skip "Need MDS version at least 2.6.53"
162 do_facet client "stat $DIR > /dev/null" ||
163 error "failed to stat $DIR: $?"
164 drop_bl_callback_once "chmod 0777 $DIR" ||
165 error "failed to chmod $DIR: $?"
167 # let the client reconnect
169 evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
170 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
172 [ -z "$evict" ] || [[ $evict -le $before ]] ||
173 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
174 error "eviction happened: $evict before:$before")
176 do_facet client checkstat -v -p 0777 $DIR ||
177 error "client checkstat failed: $?"
179 run_test 10b "re-send BL AST"
182 local before=$(date +%s)
191 workdir="${DIR}/${tdir}"
192 mkdir -p ${workdir} || error "can't create workdir $?"
193 stat ${workdir} > /dev/null ||
194 error "failed to stat ${workdir}: $?"
195 mdtidx=$($LFS getdirstripe -i ${workdir})
196 mdtname=$($LFS mdts ${workdir} | grep -e "^$mdtidx:" |
197 awk '{sub("_UUID", "", $2); print $2;}')
199 mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
200 conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid)
201 mdcpath="mdc.${mdccli}.import=connection=${conn_uuid}"
203 drop_bl_callback_once "chmod 0777 ${workdir}" &
208 # force client reconnect
209 $LCTL set_param "${mdcpath}"
211 # wait client reconnect
215 evict=$($LCTL get_param mdc.${mdccli}.state |
216 awk -F"[ [,]" '/EVICTED]$/ { if (t<$4) {t=$4;} } END { print t }')
218 [[ $evict -le $before ]] ||
219 ( $LCTL get_param mdc.$FSNAME-MDT*.state;
220 error "eviction happened: $EVICT before:$BEFORE" )
222 [ $rc -eq 0 ] || error "chmod must finished OK"
223 checkstat -v -p 0777 "${workdir}" ||
224 error "client checkstat failed: $?"
226 run_test 10c "re-send BL AST vs reconnect race (LU-5569)"
229 local before=$(date +%s)
232 [[ "$MDS1_VERSION" -lt $(version_code 2.6.90) ]] &&
233 skip "Need MDS version at least 2.6.90"
235 # sleep 1 is to make sure that BEFORE is not equal to EVICTED below
238 echo -n ", world" | dd of=$TMP/$tfile bs=1c seek=5
240 remount_client $MOUNT
244 $LFS setstripe -i 0 -c 1 $DIR1/$tfile
245 echo -n hello | dd of=$DIR1/$tfile bs=5
247 stat $DIR2/$tfile >& /dev/null
248 $LCTL set_param fail_err=71
249 drop_bl_callback "echo -n \\\", world\\\" >> $DIR2/$tfile"
254 cmp -l $DIR1/$tfile $DIR2/$tfile || error "file contents differ"
255 cmp -l $DIR1/$tfile $TMP/$tfile || error "wrong content found"
257 evict=$(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state | \
259 awk -F"[ [,]" '/EVICTED$/ { if (mx<$1) {mx=$1;} } END { print mx }')
261 [[ $evict -gt $before ]] ||
262 (do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state;
263 error "no eviction: $evict before:$before")
265 $LCTL set_param fail_err=0
267 umount_client $MOUNT2
269 run_test 10d "test failed blocking ast"
273 [[ "$OST1_VERSION" -le $(version_code 2.8.58) ]] &&
274 skip "Need OST version at least 2.8.59"
275 [ $CLIENTCOUNT -lt 2 ] && skip "need two clients"
276 [ $(facet_host client) == $(facet_host ost1) ] &&
277 skip "need ost1 and client on different nodes"
278 local -a clients=(${CLIENTS//,/ })
279 local client1=${clients[0]}
280 local client2=${clients[1]}
282 $LFS setstripe -c 1 -i 0 $DIR/$tfile-1 $DIR/$tfile-2
283 $MULTIOP $DIR/$tfile-1 Ow1048576c
285 #define OBD_FAIL_LDLM_BL_CALLBACK_NET 0x305
286 $LCTL set_param fail_loc=0x80000305
288 #define OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT 0x30e
289 do_facet ost1 "$LCTL set_param fail_loc=0x1000030e"
290 # hit OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT twice:
291 # 1. to return ENOTCONN from ldlm_handle_enqueue0
292 # 2. to pause reconnect handling between resend and setting
293 # import to LUSTRE_IMP_FULL state
294 do_facet ost1 "$LCTL set_param fail_val=3"
296 # client1 fails ro respond to bl ast
297 do_node $client2 "$MULTIOP $DIR/$tfile-1 Ow1048576c" &
300 # ost1 returns error on enqueue, which causes client1 to reconnect
301 do_node $client1 "$MULTIOP $DIR/$tfile-2 Ow1048576c" ||
302 error "multiop failed"
305 do_facet ost1 "$LCTL set_param fail_loc=0"
306 do_facet ost1 "$LCTL set_param fail_val=0"
308 run_test 10e "re-send BL AST vs reconnect race 2"
311 # wake up a thread waiting for completion after eviction
313 do_facet client $MULTIOP $DIR/$tfile Ow ||
314 { error "multiop write failed: $?"; return 1; }
315 do_facet client $MULTIOP $DIR/$tfile or ||
316 { error "multiop read failed: $?"; return 2; }
320 do_facet client $MULTIOP $DIR/$tfile or ||
321 { error "multiop read failed: $?"; return 3; }
322 drop_bl_callback_once $MULTIOP $DIR/$tfile Ow ||
323 echo "evicted as expected"
325 do_facet client unlink $DIR/$tfile ||
326 { error "unlink failed: $?"; return 4; }
327 # allow recovery to complete
328 client_up || client_up || sleep $TIMEOUT
330 run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
334 $LCTL mark $MULTIOP $DIR/$tfile OS_c
335 do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
336 clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
337 multiop_bg_pause $DIR/$tfile OS_c ||
338 { error "multiop failed: $?"; return 1; }
340 #define OBD_FAIL_MDS_CLOSE_NET 0x115
342 echo "waiting for multiop $PID"
343 wait $PID || { error "wait for multiop faile: $?"; return 2; }
344 do_facet client unlink $DIR/$tfile ||
345 { error "client unlink failed: $?"; return 3; }
346 # allow recovery to complete
347 client_up || client_up || sleep $TIMEOUT
349 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
351 # Bug 113, check that readdir lost recv timeout works.
353 mkdir_on_mdt0 $DIR/$tdir || { error "mkdir failed: $?"; return 1; }
354 touch $DIR/$tdir/newentry || { error "touch failed: $?"; return 2; }
355 # OBD_FAIL_MDS_READPAGE_NET|CFS_FAIL_ONCE
356 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104"
357 ls $DIR/$tdir || { error "ls failed: $?"; return 3; }
358 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
359 rm -rf $DIR/$tdir || { error "remove test dir failed: $?"; return 4; }
361 run_test 13 "mdc_readpage restart test (bug 1138)"
363 # Bug 113, check that readdir lost send timeout works.
365 mkdir_on_mdt0 $DIR/$tdir
366 touch $DIR/$tdir/newentry
367 # OBD_FAIL_MDS_SENDPAGE|CFS_FAIL_ONCE
368 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000106"
369 ls $DIR/$tdir || return 1
370 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
372 run_test 14 "mdc_readpage resend test (bug 1138)"
375 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
376 touch $DIR/$tfile && return 1
379 run_test 15 "failed open (-ENOMEM)"
381 READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1`
383 lctl set_param -n llite.*.max_read_ahead_mb 0
387 lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD
391 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
393 do_facet_random_file client $TMP/$tfile 100K ||
394 { error_noexit "Create random file $TMP/$T" ; return 0; }
395 do_facet client "cp $TMP/$tfile $DIR/$tfile" ||
396 { error_noexit "Copy to $DIR/$tfile file" ; return 0; }
400 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | CFS_FAIL_ONCE
401 do_facet ost1 "lctl set_param fail_loc=0x80000504"
403 # OST bulk will time out here, client resends
404 do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 1
405 do_facet ost1 lctl set_param fail_loc=0
406 # give recovery a chance to finish (shouldn't take long)
408 do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 2
412 run_test 16 "timeout bulk put, don't evict client (2732)"
417 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
419 local SAMPLE_FILE=$TMP/$tfile
420 do_facet_random_file client $SAMPLE_FILE 20K ||
421 { error_noexit "Create random file $SAMPLE_FILE" ; return 0; }
423 # With adaptive timeouts, bulk_get won't expire until
424 # adaptive_timeout_max
425 if at_is_enabled; then
426 at_max_saved=$(at_max_get ost1)
427 at_max_set $TIMEOUT ost1
430 # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | CFS_FAIL_ONCE
431 # OST bulk will time out here, client retries
432 do_facet ost1 lctl set_param fail_loc=0x80000503
433 # need to ensure we send an RPC
434 do_facet client cp $SAMPLE_FILE $DIR/$tfile
437 # with AT, client will wait adaptive_max*factor+net_latency before
438 # expiring the req, hopefully timeout*2 is enough
439 sleep $(($TIMEOUT*2))
441 do_facet ost1 lctl set_param fail_loc=0
442 do_facet client "df $DIR"
443 # expect cmp to succeed, client resent bulk
444 do_facet client "cmp $SAMPLE_FILE $DIR/$tfile" || return 3
445 do_facet client "rm $DIR/$tfile" || return 4
446 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved ost1
449 run_test 17a "timeout bulk get, don't evict client (2732)"
452 [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
454 # get one of the clients from client list
455 local rcli=$(echo $RCLIENTS | cut -d ' ' -f 1)
456 local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
457 local ldlm_enqueue_min=$(do_facet ost1 find /sys -name ldlm_enqueue_min)
458 [ -z "$ldlm_enqueue_min" ] &&
459 skip "missing /sys/.../ldlm_enqueue_min" && return 0
461 $LFS setstripe -i 0 -c 1 -S 1048576 $DIR/$tfile ||
462 error "setstripe failed"
463 $LFS setstripe -i 0 -c 1 -S 1048576 $DIR/${tfile}2 ||
464 error "setstripe 2 failed"
466 save_lustre_params ost1 "at_history" > $p
467 save_lustre_params ost1 "bulk_timeout" >> $p
468 local dev="${FSNAME}-OST0000"
469 save_lustre_params ost1 "obdfilter.$dev.brw_size" >> $p
470 local ldlm_enqueue_min_save=$(do_facet ost1 cat $ldlm_enqueue_min)
472 local new_at_history=15
474 do_facet ost1 "$LCTL set_param at_history=$new_at_history"
475 do_facet ost1 "$LCTL set_param bulk_timeout=30"
476 do_facet ost1 "echo 30 > /sys/module/ptlrpc/parameters/ldlm_enqueue_min"
478 # brw_size is required to be 4m so that bulk transfer timeout
479 # could be simulated with OBD_FAIL_PTLRPC_CLIENT_BULK_CB3
480 local brw_size=$($LCTL get_param -n \
481 osc.$FSNAME-OST0000-osc-[^M]*.import |
482 awk '/max_brw_size/{print $2}')
483 if [ $brw_size -ne 4194304 ]
485 save_lustre_params ost1 "obdfilter.$dev.brw_size" >> $p
486 do_facet ost1 "$LCTL set_param obdfilter.$dev.brw_size=4"
487 remount_client $MOUNT
490 # get service estimate expanded
491 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
492 do_facet ost1 "$LCTL set_param fail_loc=0x80000224 fail_val=35"
493 echo "delay rpc servicing by 35 seconds"
494 dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=fdatasync
497 estimate=$($LCTL get_param -n osc.$dev-osc-*.timeouts |
498 awk '/portal 6/ {print $5}')
499 echo "service estimates increased to $estimate"
501 # let current worst service estimate to get closer to obliteration
502 sleep $((new_at_history / 3))
504 # start i/o and simulate bulk transfer loss
505 #define OBD_FAIL_PTLRPC_CLIENT_BULK_CB3 0x520
506 do_facet ost1 "$LCTL set_param fail_loc=0xa0000520 fail_val=1"
507 dd if=/dev/zero of=$DIR/$tfile bs=2M count=1 conv=fdatasync,notrunc &
510 # start lock conflict handling
511 sleep $((new_at_history / 3))
512 do_node $rcli "dd if=$DIR/$tfile of=/dev/null bs=1M count=1" &
515 # obliterate the worst service estimate
516 sleep $((new_at_history / 3 + 1))
517 dd if=/dev/zero of=$DIR/${tfile}2 bs=1M count=1
519 estimate=$($LCTL get_param -n osc.$dev-osc-*.timeouts |
520 awk '/portal 6/ {print $5}')
521 echo "service estimate dropped to $estimate"
524 [[ $? == 0 ]] || error "write failed"
526 [[ $? == 0 ]] || error "read failed"
528 restore_lustre_params <$p
529 if [ $brw_size -ne 4194304 ]
531 remount_client $MOUNT || error "remount_client failed"
533 do_facet ost1 "echo $ldlm_enqueue_min_save > $ldlm_enqueue_min"
535 run_test 17b "timeout bulk get, dont evict client (3582)"
538 [ -z ${ost2_svc} ] && skip_env "needs 2 osts" && return 0
540 do_facet_create_file client $TMP/$tfile 20K ||
541 { error_noexit "Create file $TMP/$tfile" ; return 0; }
543 do_facet client mkdir -p $DIR/$tdir
547 pgcache_empty || return 1
550 $LFS setstripe -i 1 -c 1 $f
551 stripe_index=$($LFS getstripe -i $f)
552 if [ $stripe_index -ne 1 ]; then
554 error "$f: stripe_index $stripe_index != 1" && return
557 do_facet client cp $TMP/$tfile $f
559 local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | egrep -v 'MDT' | awk '{print $1}'`
560 $LCTL --device $osc2dev deactivate || return 3
561 # my understanding is that there should be nothing in the page
562 # cache after the client reconnects?
564 pgcache_empty || rc=2
565 $LCTL --device $osc2dev activate
569 run_test 18a "manual ost invalidate clears page cache immediately"
572 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
574 do_facet_create_file client $TMP/$tfile 20K ||
575 { error_noexit "Create file $TMP/$tfile" ; return 0; }
577 do_facet client mkdir -p $DIR/$tdir
581 pgcache_empty || return 1
583 $LFS setstripe -i 0 -c 1 $f
584 stripe_index=$($LFS getstripe -i $f)
585 if [ $stripe_index -ne 0 ]; then
587 error "$f: stripe_index $stripe_index != 0" && return
590 do_facet client cp $TMP/$tfile $f
593 # allow recovery to complete
594 sleep $((TIMEOUT + 2))
595 # my understanding is that there should be nothing in the page
596 # cache after the client reconnects?
598 pgcache_empty || rc=2
602 run_test 18b "eviction and reconnect clears page cache (2766)"
605 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
607 do_facet_create_file client $TMP/$tfile 20K ||
608 { error_noexit "Create file $TMP/$tfile" ; return 0; }
610 do_facet client mkdir -p $DIR/$tdir
614 pgcache_empty || return 1
616 $LFS setstripe -i 0 -c 1 $f
617 stripe_index=$($LFS getstripe -i $f)
618 if [ $stripe_index -ne 0 ]; then
620 error "$f: stripe_index $stripe_index != 0" && return
623 do_facet client cp $TMP/$tfile $f
627 # OBD_FAIL_OST_CONNECT_NET2
628 # lost reply to connect request
629 do_facet ost1 lctl set_param fail_loc=0x80000225
632 $LFS df $MOUNT > /dev/null 2>&1
634 # my understanding is that there should be nothing in the page
635 # cache after the client reconnects?
637 pgcache_empty || rc=2
641 run_test 18c "Dropped connect reply after eviction handing (14755)"
644 local BEFORE=`date +%s`
647 mount_client $DIR2 || error "failed to mount $DIR2"
649 # cancel cached locks from OST to avoid eviction from it
652 do_facet client "stat $DIR > /dev/null" ||
653 error "failed to stat $DIR: $?"
654 drop_ldlm_cancel "chmod 0777 $DIR2" ||
655 error "failed to chmod $DIR2"
659 # let the client reconnect
661 EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
662 awk -F"[ [,]" '/EVICTED ]$/ \
663 { if (mx<$5) {mx=$5;} } END { print mx }')
665 [ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
666 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
667 error "no eviction: $EVICT before:$BEFORE")
669 run_test 19a "test expired_lock_main on mds (2867)"
672 local BEFORE=`date +%s`
675 mount_client $DIR2 || error "failed to mount $DIR2: $?"
677 # cancel cached locks from MDT to avoid eviction from it
680 do_facet client $MULTIOP $DIR/$tfile Ow ||
681 error "failed to run multiop: $?"
682 drop_ldlm_cancel $MULTIOP $DIR2/$tfile Ow ||
683 error "failed to ldlm_cancel: $?"
685 umount_client $DIR2 || error "failed to unmount $DIR2: $?"
686 do_facet client unlink $DIR/$tfile ||
687 error "failed to unlink $DIR/$tfile: $?"
689 # let the client reconnect
691 EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state |
692 awk -F"[ [,]" '/EVICTED ]$/ \
693 { if (mx < $5) {mx = $5;} } END { print mx }')
695 [ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
696 (do_facet client $LCTL get_param osc.$FSNAME-OST*.state;
697 error "no eviction: $EVICT before:$BEFORE")
699 run_test 19b "test expired_lock_main on ost (2867)"
702 local BEFORE=`date +%s`
705 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=0
707 mkdir -p $DIR1/$tfile
710 #define OBD_FAIL_PTLRPC_CANCEL_RESEND 0x516
711 do_facet mds $LCTL set_param fail_loc=0x80000516
713 touch $DIR2/$tfile/file1 &
715 # let touch to get blocked on the server
719 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=1
722 # let the client reconnect
724 EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
725 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
727 [ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
729 run_test 19c "check reconnect and lock resend do not trigger expired_lock_main"
731 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
732 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
735 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
736 multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
739 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
740 do_facet ost1 lctl set_param fail_loc=0x80000308
741 kill -USR1 $MULTI_PID
744 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
746 run_test 20a "ldlm_handle_enqueue error (should return error)"
748 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
749 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
752 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
754 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
755 do_facet ost1 lctl set_param fail_loc=0x80000308
756 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
757 error "didn't fail open enqueue" || true
759 run_test 20b "ldlm_handle_enqueue error (should return error)"
762 mkdir_on_mdt0 $DIR/$tdir-1
763 mkdir_on_mdt0 $DIR/$tdir-2
764 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
767 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
768 $MULTIOP $DIR/$tdir-2/f Oc &
771 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
773 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
774 kill -USR1 $close_pid
776 wait $close_pid || return 1
777 wait $open_pid || return 2
778 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
780 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
781 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
785 run_test 21a "drop close request while close and open are both in flight"
788 mkdir_on_mdt0 $DIR/$tdir-1
789 mkdir_on_mdt0 $DIR/$tdir-2
790 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
793 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
794 mcreate $DIR/$tdir-2/f &
797 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
799 kill -USR1 $close_pid
801 wait $close_pid || return 1
802 wait $open_pid || return 3
804 $CHECKSTAT -t file $DIR/$tdir-1/f || return 4
805 $CHECKSTAT -t file $DIR/$tdir-2/f || return 5
808 run_test 21b "drop open request while close and open are both in flight"
811 mkdir_on_mdt0 $DIR/$tdir-1
812 mkdir_on_mdt0 $DIR/$tdir-2
813 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
816 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
817 mcreate $DIR/$tdir-2/f &
820 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
822 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
823 kill -USR1 $close_pid
825 wait $close_pid || return 1
826 wait $open_pid || return 2
828 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
830 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
831 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
834 run_test 21c "drop both request while close and open are both in flight"
837 mkdir_on_mdt0 $DIR/$tdir-1
838 mkdir_on_mdt0 $DIR/$tdir-2
839 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
842 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
843 $MULTIOP $DIR/$tdir-2/f Oc &
845 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
847 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
850 wait $pid || return 1
851 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
853 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
854 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
858 run_test 21d "drop close reply while close and open are both in flight"
861 mkdir_on_mdt0 $DIR/$tdir-1
862 mkdir_on_mdt0 $DIR/$tdir-2
863 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
866 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
867 touch $DIR/$tdir-2/f &
869 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
873 wait $pid || return 1
876 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
877 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
880 run_test 21e "drop open reply while close and open are both in flight"
883 mkdir_on_mdt0 $DIR/$tdir-1
884 mkdir_on_mdt0 $DIR/$tdir-2
885 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
888 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
889 touch $DIR/$tdir-2/f &
891 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
893 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
896 wait $pid || return 1
897 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
899 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
900 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
903 run_test 21f "drop both reply while close and open are both in flight"
906 mkdir_on_mdt0 $DIR/$tdir-1
907 mkdir_on_mdt0 $DIR/$tdir-2
908 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
911 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
912 touch $DIR/$tdir-2/f &
914 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
916 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
919 wait $pid || return 1
920 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
922 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
923 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
926 run_test 21g "drop open reply and close request while close and open are both in flight"
929 mkdir_on_mdt0 $DIR/$tdir-1
930 mkdir_on_mdt0 $DIR/$tdir-2
931 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
934 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
935 touch $DIR/$tdir-2/f &
938 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
940 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
943 wait $pid || return 1
944 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
946 wait $touch_pid || return 2
948 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
949 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
952 run_test 21h "drop open request and close reply while close and open are both in flight"
954 # bug 3462 - multiple MDC requests
959 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
964 $MULTIOP $f1 msu || return 1
967 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
969 wait $close_pid || return 2
970 rm -rf $f2 || return 4
972 run_test 22 "drop close request and do mknod"
975 multiop_bg_pause $DIR/$tfile O_c || return 1
977 # give a chance for open
981 drop_request "kill -USR1 $pid"
984 wait $pid || return 1
987 run_test 23 "client hang when close a file after mds crash"
989 test_24a() { # bug 11710 details correct fsync() behavior
990 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
993 $LFS setstripe -i 0 -c 1 $DIR/$tdir
995 multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
998 kill -USR1 $MULTI_PID
1001 lctl set_param fail_loc=0x0
1004 error_ignore bz5494 "multiop didn't fail fsync: rc $rc" || true
1006 run_test 24a "fsync error (should return error)"
1008 wait_client_evicted () {
1011 local varsvc=${facet}_svc
1013 wait_update $(facet_active_host $facet) \
1014 "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \
1019 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1021 dmesg -c > /dev/null
1023 lfs setstripe $DIR/$tdir -S 0 -i 0 -c 1 ||
1024 error "$LFS setstripe failed"
1025 cancel_lru_locks osc
1026 multiop_bg_pause $DIR/$tdir/$tfile-1 Ow8192_yc ||
1027 error "mulitop Ow8192_yc failed"
1030 multiop_bg_pause $DIR/$tdir/$tfile-2 Ow8192_c ||
1031 error "mulitop Ow8192_c failed"
1036 kill -USR1 $MULTI_PID1
1039 kill -USR1 $MULTI_PID2
1042 lctl set_param fail_loc=0x0
1044 [ $rc1 -eq 0 -o $rc2 -eq 0 ] &&
1045 error_ignore bz5494 "multiop didn't fail fsync: $rc1 or close: $rc2" ||
1048 dmesg | grep "dirty page discard:" ||
1049 error "no discarded dirty page found!"
1051 run_test 24b "test dirty page discard due to client eviction"
1053 test_26a() { # was test_26 bug 5921 - evict dead exports by pinger
1054 # this test can only run from a client on a separate node.
1055 remote_ost || { skip "local OST" && return 0; }
1056 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1057 remote_mds || { skip "local MDS" && return 0; }
1059 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
1060 skip "msg and ost1 are at the same node"
1064 check_timeout || return 1
1066 # make sure all imports are connected and not IDLE
1067 do_facet client lfs df > /dev/null
1068 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
1069 do_facet client lctl set_param fail_loc=0x505
1070 local before=$(date +%s)
1073 # evictor takes PING_EVICT_TIMEOUT to evict.
1074 # But if there's a race to start the evictor from various obds,
1075 # the loser might have to wait for the next ping.
1076 sleep $((TIMEOUT * 2))
1077 do_facet client lctl set_param fail_loc=0x0
1078 do_facet client lfs df > /dev/null
1080 local oscs=$(lctl dl | awk '/-osc-/ {print $4}')
1081 check_clients_evicted "$before ${oscs[@]}"
1082 check_clients_full 10 "${oscs[@]}"
1084 run_test 26a "evict dead exports"
1086 test_26b() { # bug 10140 - evict dead exports by pinger
1087 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1089 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
1090 skip "msg and ost1 are at the same node"
1094 check_timeout || return 1
1096 zconf_mount `hostname` $MOUNT2 ||
1097 { error "Failed to mount $MOUNT2"; return 2; }
1098 sleep 1 # wait connections being established
1100 local MDS_NEXP=$(do_facet $SINGLEMDS lctl get_param -n mdt.${mds1_svc}.num_exports | cut -d' ' -f2)
1101 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
1103 echo starting with $OST_NEXP OST and $MDS_NEXP MDS exports
1105 zconf_umount `hostname` $MOUNT2 -f
1107 # PING_INTERVAL max(obd_timeout / 4, 1U)
1108 # PING_EVICT_TIMEOUT (PING_INTERVAL * 6)
1110 # evictor takes PING_EVICT_TIMEOUT to evict.
1111 # But if there's a race to start the evictor from various obds,
1112 # the loser might have to wait for the next ping.
1113 # = 6 * PING_INTERVAL + PING_INTERVAL
1114 # = 7 PING_INTERVAL = 7 obd_timeout / 4 = (1+3/4)obd_timeout
1115 # let's wait $((TIMEOUT * 2)) # bug 19887
1116 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2)) ||
1117 error "Client was not evicted by ost"
1118 wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 2)) ||
1119 error "Client was not evicted by mds"
1121 run_test 26b "evict dead exports"
1124 mkdir_on_mdt0 $DIR/$tdir
1125 writemany -q -a $DIR/$tdir/$tfile 0 5 &
1128 local save_FAILURE_MODE=$FAILURE_MODE
1130 facet_failover $SINGLEMDS
1131 #define OBD_FAIL_OSC_SHUTDOWN 0x407
1132 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000407
1133 # need to wait for reconnect
1134 echo waiting for fail_loc
1135 wait_update_facet $SINGLEMDS "lctl get_param -n fail_loc" "-2147482617"
1136 facet_failover $SINGLEMDS
1137 #no crashes allowed!
1138 kill -USR1 $CLIENT_PID
1141 FAILURE_MODE=$save_FAILURE_MODE
1143 run_test 27 "fail LOV while using OSC's"
1145 test_28() { # bug 6086 - error adding new clients
1146 do_facet client mcreate $DIR/$tfile || return 1
1147 drop_bl_callback_once "chmod 0777 $DIR/$tfile" ||
1148 echo "evicted as expected"
1149 #define OBD_FAIL_MDS_CLIENT_ADD 0x12f
1150 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
1151 # fail once (evicted), reconnect fail (fail_loc), ok
1152 client_up || (sleep 10; client_up) || (sleep 10; client_up) || error "reconnect failed"
1154 fail $SINGLEMDS # verify MDS last_rcvd can be loaded
1156 run_test 28 "handle error adding new clients (bug 6086)"
1158 test_29a() { # bug 22273 - error adding new clients
1159 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
1160 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000711"
1161 # fail abort so client will be new again
1162 fail_abort $SINGLEMDS
1163 client_up || error "reconnect failed"
1164 wait_osc_import_state $SINGLEMDS ost FULL
1167 run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
1169 test_29b() { # bug 22273 - error adding new clients
1170 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
1171 do_facet ost1 "lctl set_param fail_loc=0x80000711"
1172 # fail abort so client will be new again
1174 client_up || error "reconnect failed"
1177 run_test 29b "error adding new clients doesn't cause LBUG (bug 22273)"
1181 # put a load of file creates/writes/deletes
1182 writemany -q $DIR/$tdir/$tfile 0 5 &
1184 echo writemany pid $CLIENT_PID
1188 # wait for client to reconnect to MDS
1193 # client process should see no problems even though MDS went down
1195 kill -USR1 $CLIENT_PID
1198 echo writemany returned $rc
1199 #these may fail because of eviction due to slow AST response.
1201 error_ignore bz13652 "writemany returned rc $rc" || true
1203 run_test 50 "failover MDS under load"
1206 #define OBD_FAIL_MDS_SYNC_CAPA_SL 0x1310
1207 do_facet ost1 lctl set_param fail_loc=0x00001310
1209 mkdir_on_mdt0 $DIR/$tdir
1210 # put a load of file creates/writes/deletes
1211 writemany -q $DIR/$tdir/$tfile 0 5 &
1215 facet_failover $SINGLEMDS
1216 # failover at various points during recovery
1217 SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
1218 echo will failover at $SEQ
1220 #echo failover in $i sec
1221 log "$TESTNAME: failover in $i sec"
1223 facet_failover $SINGLEMDS
1225 # client process should see no problems even though MDS went down
1226 # and recovery was interrupted
1228 kill -USR1 $CLIENT_PID
1231 echo writemany returned $rc
1233 error_ignore bz13652 "writemany returned rc $rc" || true
1235 run_test 51 "failover MDS during recovery"
1238 do_facet client "mkdir -p $DIR/$tdir"
1239 do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
1241 echo writemany pid $CLIENT_PID
1246 wait $CLIENT_PID || rc=$?
1247 # active client process should see an EIO for down OST
1248 [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
1249 # but timing or failover setup may allow success
1250 [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
1251 echo "writemany returned $rc"
1256 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1261 [ $rc -ne 0 ] && { return $rc; }
1262 # wait for client to reconnect to OST
1266 [ $rc -ne 0 ] && { return $rc; }
1273 run_test 52 "failover OST under load"
1275 # test of open reconstruct
1278 drop_mdt_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
1281 run_test 53a "touch: drop rep"
1286 drop_mdt_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||
1289 run_test 53b "touch: drop rep"
1294 drop_mdt_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||
1297 run_test 53c "touch: drop rep"
1300 zconf_mount $(hostname) $MOUNT2
1302 touch $DIR2/$tfile.1
1304 cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
1307 ERROR=$(dmesg | egrep "(test 54|went back in time)" | tail -n1 |
1308 grep "went back in time")
1309 [ x"$ERROR" == x ] || error "back in time occured"
1311 run_test 54 "back in time"
1313 # bug 11330 - liblustre application death during I/O locks up OST
1315 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1319 # This test assumes relatively small max_dirty_mb setting
1320 # which we want to walk away from, so just for it we will
1321 # temporarily lower the value
1322 local max_dirty_mb=$(lctl get_param -n osc.*.max_dirty_mb | head -n 1)
1323 lctl set_param -n osc.*.max_dirty_mb=32
1324 stack_trap "lctl set_param osc.*.max_dirty_mb=$max_dirty_mb" EXIT
1326 # Minimum pass speed is 2MBps
1328 # LU-2887/LU-3089 - set min pass speed to 500KBps
1329 [ "$ost1_FSTYPE" = zfs ] && ddtimeout=256
1331 # first dd should be finished quickly
1332 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
1333 dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
1336 echo "step1: testing ......"
1337 while kill -0 $DDPID 2> /dev/null; do
1339 if [ $count -gt $ddtimeout ]; then
1340 error "dd should be finished!"
1344 echo "(dd_pid=$DDPID, time=$count)successful"
1346 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2
1347 #define OBD_FAIL_OST_DROP_REQ 0x21d
1348 do_facet ost1 lctl set_param fail_loc=0x0000021d
1349 # second dd will be never finished
1350 dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 &
1353 echo "step2: testing ......"
1354 while [ $count -le $ddtimeout ]; do
1355 if ! kill -0 $DDPID 2> /dev/null; then
1357 error "dd shouldn't be finished! (time=$count)"
1362 echo "(dd_pid=$DDPID, time=$count)successful"
1364 #Recover fail_loc and dd will finish soon
1365 do_facet ost1 lctl set_param fail_loc=0
1367 echo "step3: testing ......"
1368 while kill -0 $DDPID 2> /dev/null; do
1370 if [ $count -gt $((ddtimeout + 440)) ]; then
1371 error "dd should be finished!"
1375 echo "(dd_pid=$DDPID, time=$count)successful"
1379 run_test 55 "ost_brw_read/write drops timed-out read/write request"
1381 test_56() { # b=11277
1382 #define OBD_FAIL_MDS_RESEND 0x136
1384 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
1385 stat $DIR/$tfile || error "stat failed"
1386 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1389 run_test 56 "do not fail on getattr resend"
1392 # no oscs means no client or mdt
1393 while lctl get_param osc.*.* > /dev/null 2>&1; do
1394 : # loop until proc file is removed
1398 test_57() { # bug 10866
1402 #define OBD_FAIL_LPROC_REMOVE 0xB00
1403 lctl set_param fail_loc=0x80000B00
1404 zconf_umount `hostname` $DIR
1405 lctl set_param fail_loc=0x80000B00
1406 fail_abort $SINGLEMDS
1408 lctl set_param fail_loc=0
1410 do_facet client "df $DIR"
1412 run_test 57 "read procfs entries causes kernel crash"
1414 test_58() { # bug 11546
1415 #define OBD_FAIL_MDC_ENQUEUE_PAUSE 0x801
1418 lctl set_param fail_loc=0x80000801
1419 cp $DIR/$tfile /dev/null &
1422 lctl set_param fail_loc=0
1423 drop_bl_callback_once rm -f $DIR/$tfile
1425 # the first 'df' could tigger the eviction caused by
1426 # 'drop_bl_callback_once', and it's normal case.
1427 # but the next 'df' should return successfully.
1428 do_facet client "df $DIR" || do_facet client "df $DIR"
1430 run_test 58 "Eviction in the middle of open RPC reply processing"
1432 test_59() { # bug 10589
1433 zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
1434 echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
1435 #define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311
1436 lctl set_param fail_loc=0x311
1437 writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
1438 [ $? = 0 ] || error "dd write failed"
1439 writes=$(echo $writes | awk -F '+' '/out/ {print $1}')
1440 lctl set_param fail_loc=0
1442 zconf_umount `hostname` $MOUNT2 -f
1443 reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
1444 [ $? = 0 ] || error "dd read failed"
1445 reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
1446 [ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
1448 run_test 59 "Read cancel race on client eviction"
1451 local num_files=${COUNT:-5000}
1452 test_mkdir $DIR/$tdir
1454 # Register (and start) changelog
1455 changelog_register || error "changelog_register failed"
1457 # Generate a large number of changelog entries
1458 createmany -o $DIR/$tdir/$tfile $num_files
1462 # Unlink files in the background
1463 unlinkmany $DIR/$tdir/$tfile $num_files &
1467 # Failover the MDS while unlinks are happening
1468 facet_failover $SINGLEMDS
1470 # Wait for unlinkmany to finish
1473 # Check if all the create/unlink events were recorded
1475 local cl_count=$(changelog_dump | grep -c UNLNK)
1476 echo "$cl_count unlinks in changelog"
1478 changelog_deregister || error "changelog_deregister failed"
1479 if ! changelog_users $SINGLEMDS | grep -q "^cl"; then
1480 [ $cl_count -eq $num_files ] ||
1481 error "Recorded $cl_count of $num_files unlinks"
1482 # Also make sure we can clear large changelogs
1483 cl_count=$(changelog_dump | wc -l)
1484 [ $cl_count -le 2 ] ||
1485 error "Changelog not empty: $cl_count entries"
1486 else # If other users, there may be other unlinks in the log
1487 [ $cl_count -ge $num_files ] ||
1488 error "Recorded $cl_count of $num_files unlinks"
1489 changelog_users $SINGLEMDS
1490 echo "other changelog users; can't verify clear"
1493 run_test 60 "Add Changelog entries during MDS failover"
1497 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0000)
1498 mdtosc=${mdtosc/-MDT*/-MDT\*}
1499 local cflags="osc.$mdtosc.connect_flags"
1500 do_facet $SINGLEMDS "lctl get_param -n $cflags" |grep -q skip_orphan
1501 [ $? -ne 0 ] && skip "don't have skip orphan feature" && return
1503 mkdir_on_mdt0 $DIR/$tdir || error "mkdir dir $DIR/$tdir failed"
1504 # Set the default stripe of $DIR/$tdir to put the files to ost1
1505 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1507 replay_barrier $SINGLEMDS
1508 createmany -o $DIR/$tdir/$tfile-%d 10
1509 local oid=$(do_facet ost1 "lctl get_param -n \
1510 obdfilter.${ost1_svc}.last_id" | sed -e 's/.*://')
1512 fail_abort $SINGLEMDS
1514 touch $DIR/$tdir/$tfile
1515 local id=$($LFS getstripe $DIR/$tdir/$tfile |
1516 awk '$1 == 0 { print $2 }')
1517 [ $id -le $oid ] && error "the orphan objid was reused, failed"
1522 run_test 61 "Verify to not reuse orphan objects - bug 17025"
1524 # test_62 as seen it b2_1 please do not reuse test_62
1527 # zconf_umount `hostname` $DIR
1528 # #define OBD_FAIL_PTLRPC_DELAY_IMP_FULL 0x516
1529 # lctl set_param fail_loc=0x516
1532 #run_test 62 "Verify connection flags race - bug LU-1716"
1537 #grant lock1, export2
1538 $LFS setstripe -i -0 $DIR2/$tfile || error "setstripe failed"
1539 $MULTIOP $DIR2/$tfile Ow || error "multiop failed"
1541 #define OBD_FAIL_LDLM_BL_EVICT 0x31e
1542 do_facet ost $LCTL set_param fail_loc=0x31e
1543 #get waiting lock2, export1
1544 $MULTIOP $DIR/$tfile Ow &
1546 # let enqueue to get asleep
1550 $MULTIOP $DIR2/$tfile Ow &
1558 do_facet ost $LCTL set_param fail_loc=0
1565 run_test 65 "lock enqueue for destroyed export"
1569 [[ "$MDS1_VERSION" -ge $(version_code 2.7.51) ]] ||
1570 skip "Need MDS version at least 2.7.51"
1572 local list=$(comma_list $(osts_nodes))
1574 # modify dir so that next revalidate would not obtain UPDATE lock
1577 # drop 1 reply with UPDATE lock
1578 mcreate $DIR/$tfile || error "mcreate failed: $?"
1579 drop_mdt_ldlm_reply_once "stat $DIR/$tfile" &
1582 # make the re-sent lock to sleep
1583 #define OBD_FAIL_MDS_RESEND 0x136
1584 do_nodes $list $LCTL set_param fail_loc=0x80000136
1586 #initiate the re-connect & re-send
1587 local mdtname="MDT0000"
1588 local mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
1589 local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid)
1590 $LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}"
1593 #initiate the client eviction while enqueue re-send is in progress
1599 run_test 66 "lock enqueue re-send vs client eviction"
1603 #define OBD_FAIL_PTLRPC_CONNECT_RACE 0x531
1604 $LCTL set_param fail_loc=0x80000531
1606 local mdtname="MDT0000"
1607 local mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
1608 local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid)
1609 $LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}" &
1618 run_test 67 "connect vs import invalidate race"
1620 check_cli_ir_state()
1622 local NODE=${1:-$HOSTNAME}
1624 st=$(do_node $NODE "lctl get_param mgc.*.ir_state |
1625 awk '/imperative_recovery:/ { print \\\$2}'")
1626 [ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
1627 error "Error state $st, must be ENABLED or DISABLED"
1631 check_target_ir_state()
1634 local name=${target}_svc
1635 local recovery_proc=obdfilter.${!name}.recovery_status
1639 st=$(do_facet $target "$LCTL get_param -n $recovery_proc |
1640 awk '/status:/{ print \\\$2}'")
1641 [ x$st = xRECOVERING ] || break
1643 st=$(do_facet $target "lctl get_param -n $recovery_proc |
1644 awk '/IR:/{ print \\\$2}'")
1645 [ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
1646 error "Error state $st, must be ENABLED or DISABLED"
1652 do_facet mgs lctl set_param -n mgs.MGS.live.$FSNAME="state=$1"
1657 local state=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1658 awk '/state:/{ print \\\$2 }'")
1662 nidtbl_version_mgs()
1664 local ver=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1665 awk '/nidtbl_version:/{ print \\\$2 }'")
1669 # nidtbl_version_client <mds1|client> [node]
1670 nidtbl_version_client()
1673 local node=${2:-$HOSTNAME}
1675 if [ X$cli = Xclient ]; then
1678 local obdtype=${cli/%[0-9]*/}
1679 [ $obdtype != mds ] && error "wrong parameters $cli"
1681 node=$(facet_active_host $cli)
1686 local vers=$(do_node $node "lctl get_param -n mgc.*.ir_state" |
1687 awk "/$cli/{print \$6}" |sort -u)
1689 # in case there are multiple mounts on the client node
1691 [ ${#arr[@]} -ne 1 ] && error "versions on client node mismatch"
1695 nidtbl_versions_match()
1697 [ $(nidtbl_version_mgs) -eq $(nidtbl_version_client ${1:-client}) ]
1700 target_instance_match()
1706 obdtype=${srv/%[0-9]*/}
1717 error "invalid target type" $srv
1722 local target=${srv}_svc
1723 local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance)
1724 local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import |
1725 awk '/instance/{ print $2 }' | head -n1)
1727 return $([ $si -eq $ci ])
1732 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1733 { skip "MGS without IR support"; return 0; }
1735 # MDT was just restarted in the previous test, make sure everything
1738 while [ $cnt -gt 0 ]; do
1739 nidtbl_versions_match && break
1745 set_ir_status disabled
1747 local prev_ver=$(nidtbl_version_client client)
1749 local saved_FAILURE_MODE=$FAILURE_MODE
1750 [ $(facet_host mgs) = $(facet_host ost1) ] && FAILURE_MODE="SOFT"
1754 [ $(nidtbl_version_client client) -eq $prev_ver ] ||
1755 error "version must not change due to IR disabled"
1756 target_instance_match ost1 || error "instance mismatch"
1760 FAILURE_MODE=$saved_FAILURE_MODE
1762 run_test 100 "IR: Make sure normal recovery still works w/o IR"
1766 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1767 skip "MGS without IR support"
1771 local ost1_imp=$(get_osc_import_name client ost1)
1773 # disable pinger recovery
1774 lctl set_param -n osc.$ost1_imp.pinger_recov=0
1775 stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT
1779 target_instance_match ost1 || error "instance mismatch"
1780 nidtbl_versions_match || error "version must match"
1782 run_test 101a "IR: Make sure IR works w/o normal recovery"
1786 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1787 skip "MGS without IR support"
1791 local ost1_imp=$(get_osc_import_name client ost1)
1793 #define OBD_FAIL_OST_PREPARE_DELAY 0x247
1794 do_facet ost1 "$LCTL set_param fail_loc=0x247"
1795 # disable pinger recovery
1796 $LCTL set_param -n osc.$ost1_imp.pinger_recov=0
1797 stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT
1799 #OST may return EAGAIN if it is not configured yet
1802 run_test 101b "IR: Make sure IR works w/o normal recovery and proceed EAGAIN"
1806 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1807 { skip "MGS without IR support"; return 0; }
1809 local clients=${CLIENTS:-$HOSTNAME}
1816 # let's have a new nidtbl version
1819 # sleep for a while so that clients can see the failure of ost
1820 # it must be MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC.
1821 # int mgc_request.c:
1822 # define MGC_TIMEOUT_MIN_SECONDS 5
1823 # define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 *
1824 local count=30 # 20 seconds at most
1825 while [ $count -gt 0 ]; do
1826 nidtbl_versions_match && break
1831 nidtbl_versions_match || error "nidtbl mismatch"
1834 old_version=$(nidtbl_version_client client)
1836 zconf_umount_clients $clients $MOUNT || error "Cannot umount client"
1839 combined_mgs_mds && mgsdev=mds1
1840 remount_facet $mgsdev
1843 zconf_mount_clients $clients $MOUNT || error "Cannot mount client"
1846 new_version=$(nidtbl_version_client client)
1847 [ $new_version -lt $old_version ] &&
1848 error "nidtbl version wrong after mgs restarts"
1851 run_test 102 "IR: New client gets updated nidtbl after MGS restart"
1855 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1856 { skip "MGS without IR support"; return 0; }
1858 combined_mgs_mds && skip "needs separate mgs and mds" && return 0
1860 # workaround solution to generate config log on the mds
1866 # We need this test because mds is like a client in IR context.
1867 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS ||
1868 error "MDS should start w/o mgs"
1870 # start mgs and remount mds w/ ir
1871 start mgs $(mgsdevname) $MGS_MOUNT_OPTS
1874 # remount client so that fsdb will be created on the MGS
1875 umount_client $MOUNT || error "umount failed"
1876 mount_client $MOUNT || error "mount failed"
1878 # sleep 30 seconds so the MDS has a chance to detect MGS restarting
1880 while [ $count -gt 0 ]; do
1881 [ $(nidtbl_version_client mds1) -ne 0 ] && break
1886 # after a while, mds should be able to reconnect to mgs and fetch
1887 # up-to-date nidtbl version
1888 nidtbl_versions_match mds1 || error "mds nidtbl mismatch"
1893 run_test 103 "IR: MDS can start w/o MGS and get updated nidtbl later"
1897 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1898 { skip "MGS without IR support"; return 0; }
1903 start ost1 $(ostdevname 1) "$OST_MOUNT_OPTS -onoir" ||
1904 error "OST1 cannot start"
1907 local ir_state=$(check_target_ir_state ost1)
1909 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1910 error "ir status on ost1 should be DISABLED"
1912 run_test 104 "IR: ost can disable IR voluntarily"
1916 [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
1917 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1918 { skip "MGS without IR support"; return 0; }
1922 # get one of the clients from client list
1923 local rcli=$(echo $RCLIENTS |cut -d' ' -f 1)
1925 local mount_opts=${MOUNT_OPTS:+$MOUNT_OPTS,}noir
1926 zconf_umount $rcli $MOUNT || error "umount failed"
1927 zconf_mount $rcli $MOUNT $mount_opts || error "mount failed"
1929 # make sure lustre mount at $rcli disabling IR
1930 local ir_state=$(check_cli_ir_state $rcli)
1931 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1932 error "IR state must be DISABLED at $rcli"
1934 # Since the client just mounted, its last_rcvd entry is not on disk.
1935 # Send an RPC so exp_need_sync forces last_rcvd to commit this export
1936 # so the client can reconnect during OST recovery (LU-924, LU-1582)
1937 $LFS setstripe -i 0 $DIR/$tfile
1938 dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=sync
1940 # make sure MGS's state is Partial
1941 [ $(get_ir_status) = "partial" ] || error "MGS IR state must be partial"
1944 # make sure IR on ost1 is DISABLED
1945 local ir_state=$(check_target_ir_state ost1)
1946 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1947 error "IR status on ost1 should be DISABLED"
1949 # remount with the default MOUNT_OPTS
1950 zconf_umount $rcli $MOUNT || error "umount failed"
1951 zconf_mount $rcli $MOUNT || error "mount failed"
1953 # make sure MGS's state is full
1954 [ $(get_ir_status) = "full" ] || error "MGS IR status must be full"
1957 # make sure IR on ost1 is ENABLED
1958 local ir_state=$(check_target_ir_state ost1)
1959 [ $ir_state = "ENABLED" -o $ir_state = "ON" ] ||
1960 error "IR status on ost1 should be ENABLED"
1964 run_test 105 "IR: NON IR clients support"
1972 test_106() { # LU-1789
1973 [[ "$MDS1_VERSION" -ge $(version_code 2.3.50) ]] ||
1974 skip "Need MDS version at least 2.3.50"
1976 #define OBD_FAIL_MDC_LIGHTWEIGHT 0x805
1977 $LCTL set_param fail_loc=0x805
1980 trap cleanup_106 EXIT
1982 # enable lightweight flag on mdc connection
1985 local MDS_NEXP=$(do_facet $SINGLEMDS \
1986 lctl get_param -n mdt.${mds1_svc}.num_exports |
1988 $LCTL set_param fail_loc=0
1990 touch $DIR2/$tfile || error "failed to create empty file"
1991 replay_barrier $SINGLEMDS
1993 $LCTL set_param debug=ha
1995 facet_failover $SINGLEMDS
1997 # lightweight goes through LUSTRE_IMP_RECOVER during failover
1998 touch -c $DIR2/$tfile || true
1999 $LCTL dk $TMP/lustre-log-$TESTNAME.log
2000 recovered=$(awk '/MDT0000-mdc-[0-9a-f]*. lwp recover/ { print }' \
2001 $TMP/lustre-log-$TESTNAME.log)
2002 [ -z "$recovered" ] && error "lightweight client was not recovered"
2004 # and all operations performed by lightweight client should be
2005 # synchronous, so the file created before mds restart should be there
2006 $CHECKSTAT -t file $DIR/$tfile || error "file not present"
2011 run_test 106 "lightweight connection support"
2017 mkdir_on_mdt0 $DIR/$tdir
2018 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2019 do_facet $SINGLEMDS lctl set_param fail_loc=0x119
2020 multiop $DIR/$tdir D_c &
2022 mkdir $DIR/$tdir/dir_106 &
2024 do_facet $SINGLEMDS lctl set_param fail_loc=0
2027 wait $CLIENT_PID || rc=$?
2028 checkstat -t dir $DIR/$tdir/dir_106 || return 1
2030 kill -USR1 $close_pid
2031 wait $close_pid || return 2
2035 run_test 107 "drop reint reply, then restart MDT"
2039 $LFS setstripe -c 1 -i 0 $DIR/$tdir
2041 dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=256 &
2049 client_up || error "reconnect failed"
2050 rm -f $DIR/$tdir/$tfile
2052 run_test 108 "client eviction don't crash"
2055 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2056 local remote_dir=$DIR/$tdir/remote_dir
2060 #prepare for 110 test, which need set striped dir on remote MDT.
2061 for num in $(seq $MDSCOUNT); do
2063 lctl set_param -n mdt.$FSNAME*.enable_remote_dir=1 \
2067 mkdir_on_mdt0 $DIR/$tdir
2068 drop_request "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
2069 error "lfs mkdir failed"
2070 local diridx=$($LFS getstripe -m $remote_dir)
2071 [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
2073 rm -rf $DIR/$tdir || error "rmdir failed"
2075 run_test 110a "create remote directory: drop client req"
2078 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2079 local remote_dir=$DIR/$tdir/remote_dir
2082 mkdir_on_mdt0 $DIR/$tdir
2083 drop_reint_reply "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
2084 error "lfs mkdir failed"
2086 diridx=$($LFS getstripe -m $remote_dir)
2087 [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
2089 rm -rf $DIR/$tdir || error "rmdir failed"
2091 run_test 110b "create remote directory: drop Master rep"
2094 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2095 local remote_dir=$DIR/$tdir/remote_dir
2098 mkdir_on_mdt0 $DIR/$tdir
2099 drop_update_reply $mdtidx "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
2100 error "lfs mkdir failed"
2102 diridx=$($LFS getstripe -m $remote_dir)
2103 [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
2105 rm -rf $DIR/$tdir || error "rmdir failed"
2107 run_test 110c "create remote directory: drop update rep on slave MDT"
2110 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2111 local remote_dir=$DIR/$tdir/remote_dir
2114 mkdir_on_mdt0 $DIR/$tdir
2115 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
2117 drop_request "rm -rf $remote_dir" || error "rm remote dir failed"
2119 rm -rf $DIR/$tdir || error "rmdir failed"
2123 run_test 110d "remove remote directory: drop client req"
2126 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2127 local remote_dir=$DIR/$tdir/remote_dir
2130 mkdir_on_mdt0 $DIR/$tdir
2131 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
2132 drop_reint_reply "rm -rf $remote_dir" || error "rm remote dir failed"
2134 rm -rf $DIR/$tdir || error "rmdir failed"
2138 run_test 110e "remove remote directory: drop master rep"
2141 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2142 local remote_dir=$DIR/$tdir/remote_dir
2145 mkdir_on_mdt0 $DIR/$tdir
2146 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
2147 drop_update_reply $MDTIDX "rm -rf $remote_dir" ||
2148 error "rm remote dir failed"
2150 rm -rf $DIR/$tdir || error "rmdir failed"
2152 run_test 110f "remove remote directory: drop slave rep"
2155 [[ "$MDS1_VERSION" -ge $(version_code 2.11.0) ]] ||
2156 skip "Need MDS version at least 2.11.0"
2158 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
2160 mkdir_on_mdt0 $DIR/$tdir
2161 touch $DIR/$tdir/$tfile
2163 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2164 do_facet mds1 $LCTL set_param fail_loc=0x119
2165 $LFS migrate -m 1 $DIR/$tdir &
2168 do_facet mds1 $LCTL set_param fail_loc=0
2172 mdt_index=$($LFS getstripe -m $DIR/$tdir)
2173 [ $mdt_index == 1 ] || error "$tdir is not on MDT1"
2174 mdt_index=$($LFS getstripe -m $DIR/$tdir/$tfile)
2175 [ $mdt_index == 1 ] || error "$tfile is not on MDT1"
2177 rm -rf $DIR/$tdir || error "rmdir failed"
2179 run_test 110g "drop reply during migration"
2182 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
2183 [[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
2184 skip "Need MDS version at least 2.7.56"
2186 local src_dir=$DIR/$tdir/source_dir
2187 local tgt_dir=$DIR/$tdir/target_dir
2191 $LFS mkdir -i $MDTIDX $tgt_dir
2193 dd if=/etc/hosts of=$src_dir/src_file
2194 touch $tgt_dir/tgt_file
2195 drop_update_reply $MDTIDX \
2196 "mrename $src_dir/src_file $tgt_dir/tgt_file" ||
2197 error "mrename failed"
2199 $CHECKSTAT -t file $src_dir/src_file &&
2200 error "src_file present after rename"
2202 diff /etc/hosts $tgt_dir/tgt_file ||
2203 error "file changed after rename"
2206 run_test 110h "drop update reply during cross-MDT file rename"
2209 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
2210 [[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
2211 skip "Need MDS version at least 2.7.56"
2213 local src_dir=$DIR/$tdir/source_dir
2214 local tgt_dir=$DIR/$tdir/target_dir
2218 $LFS mkdir -i $MDTIDX $tgt_dir
2220 mkdir $src_dir/src_dir
2221 touch $src_dir/src_dir/a
2222 mkdir $tgt_dir/tgt_dir
2223 drop_update_reply $MDTIDX \
2224 "mrename $src_dir/src_dir $tgt_dir/tgt_dir" ||
2225 error "mrename failed"
2227 $CHECKSTAT -t dir $src_dir/src_dir &&
2228 error "src_dir present after rename"
2230 $CHECKSTAT -t dir $tgt_dir/tgt_dir ||
2231 error "tgt_dir not present after rename"
2233 $CHECKSTAT -t file $tgt_dir/tgt_dir/a ||
2234 error "a not present after rename"
2236 run_test 110i "drop update reply during cross-MDT dir rename"
2239 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
2240 [[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
2241 skip "Need MDS version at least 2.7.56"
2243 local remote_dir=$DIR/$tdir/remote_dir
2244 local local_dir=$DIR/$tdir/local_dir
2247 mkdir_on_mdt0 $DIR/$tdir
2248 mkdir $DIR/$tdir/local_dir
2249 $LFS mkdir -i $MDTIDX $remote_dir
2251 touch $local_dir/local_file
2252 drop_update_reply $MDTIDX \
2253 "ln $local_dir/local_file $remote_dir/remote_file" ||
2256 $CHECKSTAT -t file $remote_dir/remote_file ||
2257 error "remote not present after ln"
2259 run_test 110j "drop update reply during cross-MDT ln"
2262 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTS"
2263 [[ "$MDS1_VERSION" -ge $(version_code 2.12.55) ]] ||
2264 skip "Need MDS version at least 2.12.55"
2267 stop mds2 || error "stop mds2 failed"
2269 #define OBD_FAIL_FLD_QUERY_REQ 0x1103
2270 do_facet mds2 lctl set_param fail_loc=0x80001103
2271 local OPTS="$MDS_MOUNT_OPTS -o abort_recovery"
2272 start mds2 $(mdsdevname 2) $OPTS ||
2273 error "start MDS with abort_recovery should succeed"
2274 do_facet mds2 lctl set_param fail_loc=0
2277 stop mds2 || error "cleanup: stop mds2 failed"
2278 start mds2 $(mdsdevname 2) $MDS_MOUNT_OPTS ||
2279 error "cleanup: start mds2 failed"
2280 zconf_mount $(hostname) $MOUNT || error "cleanup: mount failed"
2281 client_up || error "post-failover df failed"
2284 run_test 110k "FID_QUERY failed during recovery"
2287 (( $(lustre_version_code $SINGLEMDS) >= $(version_code 2.14.52) )) ||
2288 skip "Need MDS version at least 2.14.52"
2289 (( $MDSCOUNT >= 2 )) || skip "needs at least 2 MDTs"
2290 local remote_dir=$DIR/$tdir/remote_dir
2295 mkdir_on_mdt0 $DIR/$tdir
2297 #define OBD_FAIL_PTLRPC_RESEND_RACE 0x0525
2298 do_facet mds1 $LCTL set_param fail_loc=0x80000525
2299 $LFS mkdir -i 1 -c2 $remote_dir &
2302 # initiate the re-connect & re-send
2303 mdccli=$(do_facet mds2 $LCTL dl |
2304 awk '/MDT0000-osp-MDT0001/ {print $4;}')
2305 uuid=$(do_facet mds2 $LCTL get_param -n osp.$mdccli.mdt_conn_uuid)
2306 echo "conn_uuid=$uuid"
2307 do_facet mds2 $LCTL set_param "osp.$mdccli.import=connection=$uuid"
2310 (( $? == 0 )) || error "mkdir failed"
2312 diridx=$($LFS getstripe -m $remote_dir)
2313 (( $diridx == 1 )) || error "$diridx != 1"
2314 rm -rf $DIR/$tdir || error "rmdir failed"
2316 run_test 110m "update resent vs original RPC race"
2318 # LU-2844 mdt prepare fail should not cause umount oops
2321 [[ "$MDS1_VERSION" -ge $(version_code 2.3.62) ]] ||
2322 skip "Need MDS version at least 2.3.62"
2324 #define OBD_FAIL_MDS_CHANGELOG_INIT 0x151
2325 do_facet $SINGLEMDS lctl set_param fail_loc=0x151
2326 stop $SINGLEMDS || error "stop MDS failed"
2327 start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS &&
2328 error "start MDS should fail"
2329 do_facet $SINGLEMDS lctl set_param fail_loc=0
2330 start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS ||
2331 error "start MDS failed"
2333 run_test 111 "mdd setup fail should not cause umount oops"
2337 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
2339 do_facet_random_file client $TMP/$tfile 100K ||
2340 error_noexit "Create random file $TMP/$tfile"
2342 pause_bulk "cp $TMP/$tfile $DIR/$tfile" $TIMEOUT ||
2343 error_noexit "Can't pause_bulk copy"
2346 # expect cmp to succeed, client resent bulk
2347 cmp $TMP/$tfile $DIR/$tfile ||
2348 error_noexit "Wrong data has been written"
2350 error_noexit "Can't remove file"
2353 run_test 112a "bulk resend while orignal request is in progress"
2360 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1
2361 cancel_lru_locks osc
2364 $LCTL set_param fail_loc=$fail1 fail_val=3
2365 dd of=/dev/null if=$DIR/$tfile bs=4096 count=1 &
2369 set_nodes_failloc "$(osts_nodes)" $fail2
2371 wait $pid || error "dd failed"
2379 local fail_val2=${4:-0}
2385 $LCTL set_param fail_loc=$fail1 fail_val=4
2386 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 oflag=dsync &
2391 set_nodes_failloc "$(osts_nodes)" $fail2 $fail_val2
2395 [ $error -eq 0 ] && [ $rc -ne 0 ] && error "dd error ($rc)"
2396 [ $error -ne 0 ] && [ $rc -eq 0 ] && error "dd success"
2401 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2402 skip "need at least 2.8.50 on OST"
2404 #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
2405 #define OBD_FAIL_PTLRPC_DROP_BULK 0x51a
2406 test_115_read 0x8000051b 0x8000051a
2408 run_test 115a "read: late REQ MDunlink and no bulk"
2411 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2412 skip "need at least 2.8.50 on OST"
2414 #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
2415 #define OBD_FAIL_OST_ENOSPC 0x215
2417 # pass $OSTCOUNT for the fail_loc to be caught
2418 # appropriately by the IO thread
2419 test_115_write 0x8000051b 0x80000215 1 $OSTCOUNT
2421 run_test 115b "write: late REQ MDunlink and no bulk"
2424 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2425 skip "need at least 2.8.50 on OST"
2427 #define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
2428 #define OBD_FAIL_PTLRPC_DROP_BULK 0x51a
2429 test_115_read 0x8000050f 0x8000051a
2431 run_test 115c "read: late Reply MDunlink and no bulk"
2434 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2435 skip "need at least 2.8.50 on OST"
2437 #define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
2438 #define OBD_FAIL_OST_ENOSPC 0x215
2439 test_115_write 0x8000050f 0x80000215 0
2441 run_test 115d "write: late Reply MDunlink and no bulk"
2444 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2445 skip "need at least 2.8.50 on OST"
2447 #define OBD_FAIL_PTLRPC_LONG_BULK_UNLINK 0x510
2448 #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
2449 test_115_read 0x80000510 0x80000211
2451 run_test 115e "read: late Bulk MDunlink and no reply"
2454 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2455 skip "need at least 2.8.50 on OST"
2457 #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
2458 #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
2459 test_115_read 0x8000051b 0x80000211
2461 run_test 115f "read: late REQ MDunlink and no reply"
2464 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2465 skip "need at least 2.8.50 on OST"
2467 #define OBD_FAIL_PTLRPC_LONG_BOTH_UNLINK 0x51c
2468 test_115_read 0x8000051c 0
2470 run_test 115g "read: late REQ MDunlink and Reply MDunlink"
2472 # parameters: fail_loc CMD RC
2479 #define OBD_FAIL_LDLM_CP_CB_WAIT2 0x320
2480 #define OBD_FAIL_LDLM_CP_CB_WAIT3 0x321
2481 #define OBD_FAIL_LDLM_CP_CB_WAIT4 0x322
2482 #define OBD_FAIL_LDLM_CP_CB_WAIT5 0x323
2485 echo -n "** FLOCK REPLY vs. EVICTION race, lock $2"
2486 [ "$1" = "CLEANUP" ] &&
2487 fail=0x80000320 && echo ", $1 cp first"
2488 [ "$1" = "REPLY" ] &&
2489 fail=0x80000321 && echo ", $1 cp first"
2490 [ "$1" = "DEADLOCK CLEANUP" ] &&
2491 fail=0x80000322 && echo " DEADLOCK, CLEANUP cp first"
2492 [ "$1" = "DEADLOCK REPLY" ] &&
2493 fail=0x80000323 && echo " DEADLOCK, REPLY cp first"
2495 if [ x"$2" = x"get" ]; then
2496 #for TEST lock, take a conflict in advance
2497 # sleep longer than evictor to not confuse fail_loc: 2+2+4
2498 echo "** Taking conflict **"
2499 flocks_test 5 set read sleep 10 $DIR/$tfile &
2505 $LCTL set_param fail_loc=$fail
2507 flocks_test 5 $2 write $DIR/$tfile &
2511 echo "** Evicting and re-connecting client **"
2516 if [ x"$2" = x"get" ]; then
2523 # check if the return value is allowed
2524 [ $rc -eq $3 ] && rc=0
2526 $LCTL set_param fail_loc=0
2530 # a lock is taken, unlock vs. cleanup_resource() race for destroying
2531 # the ORIGINAL lock.
2536 flocks_test 5 set write sleep 4 $DIR/$tfile &
2540 # let unlock to sleep in CP CB
2541 $LCTL set_param fail_loc=$1
2544 # let cleanup to cleep in CP CB
2552 $LCTL set_param fail_loc=0
2557 flock_is_enabled || { skip "mount w/o flock enabled" && return; }
2560 test_120_reply "CLEANUP" set 5 || error "SET race failed"
2561 test_120_reply "CLEANUP" get 5 || error "GET race failed"
2562 test_120_reply "CLEANUP" unlock 5 || error "UNLOCK race failed"
2564 test_120_reply "REPLY" set 5 || error "SET race failed"
2565 test_120_reply "REPLY" get 5 || error "GET race failed"
2566 test_120_reply "REPLY" unlock 5 || error "UNLOCK race failed"
2569 test_120_reply "DEADLOCK CLEANUP" set 5 || error "DEADLOCK race failed"
2570 test_120_reply "DEADLOCK REPLY" set 35 || error "DEADLOCK race failed"
2572 test_120_destroy 0x320 || error "unlock-cleanup race failed"
2574 run_test 120 "flock race: completion vs. evict"
2577 local BEFORE=$(date +%s)
2580 # modify dir so that next revalidate would not obtain UPDATE lock
2583 # drop 1 reply with UPDATE lock,
2584 # resend should not create 2nd lock on server
2585 mcreate $DIR/$tfile || error "mcreate failed: $?"
2586 drop_mdt_ldlm_reply_once "stat $DIR/$tfile" || error "stat failed: $?"
2588 # 2 BL AST will be sent to client, both must find the same lock,
2589 # race them to not get EINVAL for 2nd BL AST
2590 #define OBD_FAIL_LDLM_PAUSE_CANCEL2 0x31f
2591 $LCTL set_param fail_loc=0x8000031f
2593 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=0 > /dev/null
2594 chmod 0777 $DIR/$tfile || error "chmod failed: $?"
2595 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=1 > /dev/null
2597 # let the client reconnect
2599 EVICT=$($LCTL get_param mdc.$FSNAME-MDT*.state |
2600 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
2602 [ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
2604 run_test 113 "ldlm enqueue dropped reply should not cause deadlocks"
2608 test_mkdir -p -c1 $DIR/$tdir
2610 # Prevent interference from layout intent RPCs due to
2611 # asynchronous writeback. These will be tested in 130c below.
2612 do_nodes ${CLIENTS:-$HOSTNAME} sync
2614 # get only LOOKUP lock on $tdir
2615 cancel_lru_locks mdc
2616 ls $DIR/$tdir/$tfile 2>/dev/null
2618 # get getattr by fid on $tdir
2620 # we need to race with unlink, unlink must complete before we will
2621 # take a DLM lock, otherwise unlink will wait until getattr will
2622 # complete; but later than getattr starts so that getattr found
2624 #define OBD_FAIL_MDS_INTENT_DELAY 0x160
2625 set_nodes_failloc "$(mdts_nodes)" 0x80000160
2632 # drop the reply so that resend happens on an unlinked file.
2633 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2634 set_nodes_failloc "$(mdts_nodes)" 0x80000157
2638 remote_mds_nodsh && skip "remote MDS with nodsh"
2639 [[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
2640 skip "Need server version newer than 2.7.1"
2644 wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
2647 run_test 130a "enqueue resend on not existing file"
2650 remote_mds_nodsh && skip "remote MDS with nodsh"
2651 [[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
2652 skip "Need server version newer than 2.7.1"
2655 # let the reply to be dropped
2658 #define OBD_FAIL_SRV_ENOENT 0x217
2659 set_nodes_failloc "$(mdts_nodes)" 0x80000217
2661 wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
2664 run_test 130b "enqueue resend on a stale inode"
2667 remote_mds_nodsh && skip "remote MDS with nodsh" && return
2669 do_nodes ${CLIENTS:-$HOSTNAME} sync
2670 echo XXX > $DIR/$tfile
2672 cancel_lru_locks mdc
2674 # Trigger writeback on $tfile.
2676 # we need to race with unlink, unlink must complete before we will
2677 # take a DLM lock, otherwise unlink will wait until intent will
2678 # complete; but later than intent starts so that intent found
2680 #define OBD_FAIL_MDS_INTENT_DELAY 0x160
2681 set_nodes_failloc "$(mdts_nodes)" 0x80000160
2688 # drop the reply so that resend happens on an unlinked file.
2689 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2690 set_nodes_failloc "$(mdts_nodes)" 0x80000157
2692 # let the reply to be dropped
2695 #define OBD_FAIL_SRV_ENOENT 0x217
2696 set_nodes_failloc "$(mdts_nodes)" 0x80000217
2702 run_test 130c "layout intent resend on a stale inode"
2705 local before=$(date +%s)
2708 mount_client $MOUNT2 || error "mount filed"
2711 # get a lock on client so that export would reach the stale list
2712 $LFS setstripe -i 0 $DIR/$tfile || error "setstripe failed"
2713 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=fsync ||
2716 #define OBD_FAIL_OST_PAUSE_PUNCH 0x236
2717 do_facet ost1 $LCTL set_param fail_val=120 fail_loc=0x80000236
2719 $TRUNCATE $DIR/$tfile 100 &
2722 dd if=/dev/zero of=$DIR2/$tfile bs=4096 count=1 conv=notrunc ||
2726 umount_client $MOUNT2
2728 evict=$(do_facet client $LCTL get_param \
2729 osc.$FSNAME-OST0000-osc-*/state |
2730 awk -F"[ [,]" '/EVICTED ]$/ { if (t<$5) {t=$5;} } END { print t }')
2732 [ -z "$evict" ] || [[ $evict -le $before ]] ||
2733 (do_facet client $LCTL get_param \
2734 osc.$FSNAME-OST0000-osc-*/state;
2735 error "eviction happened: $evict before:$before")
2737 run_test 132 "long punch"
2740 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
2743 # get a lock on client so that export would reach the stale list
2744 $LFS setstripe -i 0 $DIR/$tfile || error "setstripe failed"
2745 dd if=/dev/zero of=$DIR/$tfile count=1 || error "dd failed"
2747 # another IO under the same lock
2748 #define OBD_FAIL_OSC_DELAY_IO 0x414
2749 $LCTL set_param fail_loc=0x80000414
2750 $LCTL set_param fail_val=4 fail_loc=0x80000414
2751 dd if=/dev/zero of=$DIR/$tfile count=1 conv=notrunc oflag=dsync &
2755 #define OBD_FAIL_LDLM_BL_EVICT 0x31e
2756 set_nodes_failloc "$(osts_nodes)" 0x8000031e
2760 wait $pid && error "dd succeeded"
2763 run_test 131 "IO vs evict results to IO under staled lock"
2766 local list=$(comma_list $(mdts_nodes))
2768 local t=$((TIMEOUT * 2))
2771 flock $DIR/$tfile -c "echo bl lock;sleep $t;echo bl flock unlocked" &
2773 multiop_bg_pause $DIR/$tfile O_jc || return 1
2776 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2777 do_nodes $list $LCTL set_param fail_loc=0x80000157
2779 echo "waiting for multiop $PID"
2780 wait $PID || return 2
2786 run_test 133 "don't fail on flock resend"
2789 [ $CLIENTCOUNT -lt 2 ] &&
2790 { skip "Need 2+ clients, have $CLIENTCOUNT" && return; }
2792 mkdir -p $MOUNT/$tdir/1 $MOUNT/$tdir/2 || error "mkdir failed"
2793 touch $MOUNT/$tdir/1/$tfile $MOUNT/$tdir/2/$tfile ||
2794 error "touch failed"
2795 zconf_umount_clients $CLIENTS $MOUNT
2796 zconf_mount_clients $CLIENTS $MOUNT
2798 #define OBD_FAIL_TGT_REPLY_DATA_RACE 0x722
2799 # assume commit interval is 5
2800 do_facet mds1 "$LCTL set_param fail_loc=0x722 fail_val=5"
2802 local -a clients=(${CLIENTS//,/ })
2803 local client1=${clients[0]}
2804 local client2=${clients[1]}
2806 do_node $client1 rm $MOUNT/$tdir/1/$tfile &
2808 do_node $client2 mv $MOUNT/$tdir/2/$tfile $MOUNT/$tdir/2/${tfile}_2 &
2811 wait $rmpid || error "rm failed"
2812 wait $mvpid || error "mv failed"
2815 run_test 134 "race between failover and search for reply data free slot"
2818 [ "$MDS1_VERSION" -lt $(version_code 2.12.51) ] &&
2819 skip "Need MDS version at least 2.12.51"
2822 $LFS setstripe -E 1M -L mdt $DIR/$tdir
2823 # to have parent dir write lock before open/resend
2824 touch $DIR/$tdir/$tfile
2825 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2826 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x80000157
2827 openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tdir/$tfile ||
2828 error "Failed to open DOM file"
2830 run_test 135 "DOM: open/create resend to return size"
2833 remote_mds_nodsh && skip "remote MDS with nodsh"
2834 [[ "$MDS1_VERSION" -ge $(version_code 2.12.52) ]] ||
2835 skip "Need MDS version at least 2.12.52"
2837 local mdts=$(comma_list $(mdts_nodes))
2838 local MDT0=$(facet_svc $SINGLEMDS)
2840 local clog=$(do_facet mds1 $LCTL --device $MDT0 changelog_register -n)
2841 [ -n "$clog" ] || error "changelog_register failed"
2842 cl_mask=$(do_facet mds1 $LCTL get_param \
2843 mdd.$MDT0.changelog_mask -n)
2844 changelog_chmask "ALL"
2846 # generate some changelog records to accumulate
2847 test_mkdir -i 0 -c 0 $DIR/$tdir || error "mkdir $tdir failed"
2848 createmany -m $DIR/$tdir/$tfile 10000 ||
2849 error "create $DIR/$tdir/$tfile failed"
2851 local size1=$(do_facet $SINGLEMDS \
2852 $LCTL get_param -n mdd.$MDT0.changelog_size)
2853 echo "Changelog size $size1"
2855 #define OBD_FAIL_LLOG_PURGE_DELAY 0x1318
2856 do_nodes $mdts $LCTL set_param fail_loc=0x1318 fail_val=30
2858 # launch changelog_deregister in background on MDS
2859 do_facet mds1 "nohup $LCTL --device $MDT0 changelog_deregister $clog \
2860 > foo.out 2> foo.err < /dev/null &"
2861 # give time to reach fail_loc
2864 # fail_loc will make MDS sleep in the middle of changelog_deregister
2865 # take this opportunity to abruptly kill MDS
2866 FAILURE_MODE_save=$FAILURE_MODE
2869 FAILURE_MODE=$FAILURE_MODE_save
2871 do_nodes $mdts $LCTL set_param fail_loc=0x0 fail_val=0
2873 local size2=$(do_facet $SINGLEMDS \
2874 $LCTL get_param -n mdd.$MDT0.changelog_size)
2875 echo "Changelog size $size2"
2876 local clog2=$(do_facet $SINGLEMDS "$LCTL get_param -n \
2877 mdd.$MDT0.changelog_users | grep $clog")
2878 echo "After crash, changelog user $clog2"
2880 [ -n "$clog2" -o $size2 -lt $size1 ] ||
2881 error "changelog record count unchanged"
2883 do_facet mds1 $LCTL set_param mdd.$MDT0.changelog_mask=\'$cl_mask\' -n
2885 run_test 136 "changelog_deregister leaving pending records"
2889 mkdir_on_mdt0 $DIR/$tdir
2892 dd if=/dev/zero of=$DIR/$tdir/d1/$tfile bs=4096 count=1
2893 dd if=/dev/zero of=$DIR/$tdir/d2/$tfile bs=4096 count=1
2894 cancel_lru_locks osc
2896 #define OBD_FAIL_PTLRPC_RESEND_RACE 0x525
2897 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000525"
2899 # RPC1: any reply is to be delayed to disable last_xid logic
2900 ln $DIR/$tdir/d1/$tfile $DIR/$tdir/d1/f2 &
2903 # RPC2: setattr1 reply is delayed & resent
2904 # original reply comes to client; the resend get asleep
2905 chmod 666 $DIR/$tdir/d2/$tfile
2907 # RPC3: setattr2 on the same file; run ahead of RPC2 resend
2908 chmod 777 $DIR/$tdir/d2/$tfile
2910 # RPC2 resend wakes up
2912 [ $(stat -c "%a" $DIR/$tdir/d2/$tfile) == 777 ] ||
2913 error "resend got applied"
2915 run_test 137 "late resend must be skipped if already applied"
2918 remote_mds_nodsh && skip "remote MDS with nodsh"
2919 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2920 [[ "$MDS1_VERSION" -ge $(version_code 2.12.59) ]] ||
2921 skip "Need server version newer than 2.12.59"
2923 zconf_umount_clients $CLIENTS $MOUNT
2925 #define OBD_FAIL_TGT_RECOVERY_CONNECT 0x724
2926 #delay a first step of recovey when MDS waiting clients
2927 #and failing to get osp logs
2928 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x724 fail_val=5
2930 facet_failover $SINGLEMDS
2932 #waiting failover and recovery timer
2933 #the valuse is based on target_recovery_overseer() wait_event timeout
2935 stop $SINGLEMDS || error "stop MDS failed"
2936 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
2937 start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS ||
2938 error "start MDS failed"
2939 zconf_mount_clients $CLIENTS $MOUNT
2941 run_test 138 "Umount MDT during recovery"
2944 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2945 [ $MDS1_VERSION -lt $(version_code 2.13.50) ] &&
2946 skip "Need MDS version at least 2.13.50"
2948 stop $SINGLEMDS || error "stop $SINGLEMDS failed"
2950 #define OBD_FAIL_OSP_INVALID_LOGID 0x2106
2951 do_facet $SINGLEMDS $LCTL set_param fail_val=0x68 fail_loc=0x80002106
2952 start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS ||
2953 error "Fail to start $SINGLEMDS"
2955 run_test 139 "corrupted catid won't cause crash"
2958 [ $MDS1_VERSION -lt $(version_code 2.12.58) ] &&
2959 skip "Need MDS version at least 2.13.50"
2961 slr=$(do_facet mds1 \
2962 $LCTL get_param -n mdt.$FSNAME-MDT0000.local_recovery)
2963 stack_trap "do_facet mds1 $LCTL set_param \
2964 mdt.*.local_recovery=$slr" EXIT
2966 # disable recovery for local clients
2967 # so local clients should be marked with no_recovery flag
2968 do_facet mds1 $LCTL set_param mdt.*.local_recovery=0
2972 cnt=$(do_facet mds1 $LCTL get_param "mdt.*MDT0000.exports.*.export" |
2973 grep export_flags.*no_recovery | wc -l)
2974 echo "$cnt clients with recovery disabled"
2976 [ $cnt -eq 0 ] && error "no clients with recovery disabled"
2978 # enable recovery for local clients
2979 # so no local clients should be marked with no_recovery flag
2980 do_facet mds1 $LCTL set_param mdt.*.local_recovery=1
2983 cnt=$(do_facet mds1 $LCTL get_param "mdt.*MDT0000.exports.*.export" |
2984 grep export_flags.*no_recovery | wc -l)
2985 echo "$cnt clients with recovery disabled"
2987 [ $cnt -eq 0 ] || error "$cnt clients with recovery disabled"
2989 run_test 140a "local mount is flagged properly"
2992 [ $MDS1_VERSION -lt $(version_code 2.12.58) ] &&
2993 skip "Need MDS version at least 2.13.50"
2995 slr=$(do_facet mds1 \
2996 $LCTL get_param -n mdt.$FSNAME-MDT0000.local_recovery)
2997 stack_trap "do_facet mds1 $LCTL set_param \
2998 mdt.*.local_recovery=$slr" EXIT
3000 # disable recovery for local clients
3001 do_facet mds1 $LCTL set_param mdt.*.local_recovery=0
3007 # Lustre: tfs-MDT0000: Recovery over after 0:03, of 2 clients 2 rec...
3008 local recovery=$(do_facet mds1 dmesg |
3009 awk '/Recovery over after/ { print $6 }' | tail -1 |
3010 awk -F: '{ print $1 * 60 + $2 }')
3011 (( recovery < TIMEOUT * 2 + 5 )) ||
3012 error "recovery took too long $recovery > $((TIMEOUT * 2 + 5))"
3014 run_test 140b "local mount is excluded from recovery"
3020 [ "$PARALLEL" == "yes" ] && skip "skip parallel run"
3021 combined_mgs_mds || skip "needs combined MGS/MDT"
3022 ( local_mode || from_build_tree ) &&
3023 skip "cannot run in local mode or from build tree"
3025 # some get_param have a bug to handle dot in param name
3026 do_rpc_nodes $(facet_active_host $SINGLEMDS) cancel_lru_locks MGC
3027 oldc=$(do_facet $SINGLEMDS $LCTL get_param -n \
3028 'ldlm.namespaces.MGC*.lock_count')
3030 do_rpc_nodes $(facet_active_host $SINGLEMDS) cancel_lru_locks MGC
3031 newc=$(do_facet $SINGLEMDS $LCTL get_param -n \
3032 'ldlm.namespaces.MGC*.lock_count')
3034 [ $oldc -eq $newc ] || error "mgc lost locks ($oldc != $newc)"
3037 run_test 141 "do not lose locks on MGS restart"
3040 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
3041 skip "Need MDS version at least 2.11.56"
3043 #define OBD_FAIL_MDS_ORPHAN_DELETE 0x165
3044 do_facet mds1 $LCTL set_param fail_loc=0x165
3045 $MULTIOP $DIR/$tfile Ouc || error "multiop failed"
3048 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS
3050 wait_update_facet mds1 "pgrep orph_.*-MDD | wc -l" "0" ||
3051 error "MDD orphan cleanup thread not quit"
3053 run_test 142 "orphan name stub can be cleaned up in startup"
3056 (( $MDS1_VERSION >= $(version_code 2.13.0) )) ||
3057 skip "Need MDS version at least 2.13.0"
3058 [ "$PARALLEL" == "yes" ] && skip "skip parallel run"
3060 local mntpt=$(facet_mntpt $SINGLEMDS)
3062 mount_fstype $SINGLEMDS || error "mount as fstype $SINGLEMDS failed"
3063 do_facet $SINGLEMDS touch $mntpt/PENDING/$tfile
3064 unmount_fstype $SINGLEMDS
3065 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "mds1 start fail"
3067 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
3068 wait_update_facet mds1 "pgrep orph_.*-MDD | wc -l" "0" ||
3069 error "MDD orphan cleanup thread not quit"
3071 run_test 143 "orphan cleanup thread shouldn't be blocked even delete failed"
3074 [[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
3075 skip "server does not support overstriping"
3076 (( MDS1_VERSION >= $(version_code 2.15.50.49) )) ||
3077 skip "Need MDS version at least 2.15.50.49"
3086 large_xattr_enabled || skip_env "ea_inode feature disabled"
3087 test_mkdir -i 0 -c 1 -p $DIR/$tdir
3088 stack_trap "rm -rf $DIR/$tdir" EXIT
3090 mds_timeout=$(do_facet mds1 $LCTL get_param -n timeout)
3091 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param timeout=300
3092 stack_trap "do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param timeout=$mds_timeout" EXIT
3094 $LFS setstripe -i 0 -C $setcount $DIR/$tdir || error "setstripe failed"
3096 for (( i = 0; i < 50; i++)); do
3097 touch $DIR/$tdir/$tfile_$i &
3104 for pid in $pids; do
3105 kill -9 $pid >/dev/null 2>&1
3111 # here we measure MDT stop + MDT start time. For error case MDT stop takes
3112 # about obd_timeout-60 (240) seconds. Without error - less than 30s.
3113 # MDT start takes different time depends on a configuration, let's check
3115 diff=$((after - before))
3116 (( $diff < 240 )) || error "MDT failover took $diff seconds"
3118 # failover mds1 back to the primary server
3121 run_test 144a "MDT failover should stop precreation threads"
3124 [[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
3125 skip "server does not support overstriping"
3126 (( MDS1_VERSION >= $(version_code 2.15.51.50) )) ||
3127 skip "Need MDS version at least 2.15.51.50"
3134 large_xattr_enabled || skip_env "ea_inode feature disabled"
3135 test_mkdir -i 0 -c 1 -p $DIR/$tdir
3136 stack_trap "rm -rf $DIR/$tdir" EXIT
3138 mds_timeout=$(do_facet mds1 $LCTL get_param -n timeout)
3139 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param timeout=300
3140 stack_trap "do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param timeout=$mds_timeout" EXIT
3142 $LFS setstripe -i 0 -C $setcount $DIR/$tdir || error "setstripe failed"
3144 for (( i = 0; i < 50; i++)); do
3145 touch $DIR/$tdir/$tfile_$i &
3151 # 60s is not enought with ZFS which is still significatnly
3152 # slower on some operations like record writing
3154 [[ "$mds1_FSTYPE" != "zfs" ]] || stime=120
3157 for pid in $pids; do
3158 ps -p $pid > /dev/null
3159 (( $? == 0 )) && rc=1
3160 kill -9 $pid >/dev/null 2>&1
3163 (( $rc == 0 )) || { fail mds1; error "Create thread still running"; }
3165 run_test 144b "orphan cleanup shouldn't be blocked for no objects+failover situation"
3168 [ "$PARALLEL" == "yes" ] && skip "skip parallel run"
3169 remote_mds_nodsh && skip "remote MDS with nodsh"
3170 remote_ost_nodsh && skip "remote OST with nodsh"
3171 (( OST1_VERSION >= $(version_code 2.15.59.53) )) ||
3172 skip "need OSS >= v2_15_59.53 for reconnect fix"
3175 #increase a precreation window
3176 mkdir_on_mdt0 $DIR/$tdir
3177 $LFS setstripe -c 1 -i 0 $DIR/$tdir
3178 createmany -o $DIR/$tdir/$tfile 9000
3181 #define OBD_FAIL_OST_DELORPHAN_DELAY 0x254
3182 #delay delorphan request to reconnection 5seconds
3183 do_facet ost1 $LCTL set_param fail_loc=0x0000254 fail_val=5
3185 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "mds1 start fail"
3187 wait_recovery_complete mds1 || error "MDS recovery not done"
3191 do_facet mds1 $LCTL --device $FSNAME-OST0000-osc-MDT0000 recover
3193 do_facet ost1 $LCTL set_param fail_loc=0 fail_val=0
3195 #first and second orphan request delayed for 5seconds
3198 local testid=${TESTNAME//_/ }
3200 do_facet ost1 "dmesg | tac | sed '/$testid/,$ d'" |
3201 grep "trust the OST"
3203 if (( rc == 0 )); then
3205 error "LAST_ID synchronization failed"
3210 run_test 144c "reconnection during orphan cleanup shouldn't lose LAST_ID synchronization"
3213 [ $MDSCOUNT -lt 3 ] && skip "needs >= 3 MDTs"
3214 [ $(facet_active_host mds2) = $(facet_active_host mds3) ] &&
3215 skip "needs mds2 and mds3 on separate nodes"
3221 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
3222 echo block mds_connect from mds2
3223 do_facet mds2 "$LCTL set_param fail_loc=0x507"
3225 #define OBD_FAIL_OUT_UPDATE_DROP 0x1707
3226 echo block recovery updates from mds3
3227 do_facet mds3 "$LCTL set_param fail_loc=0x1707"
3229 local hard_timeout=\
3230 $(do_facet mds1 $LCTL get_param -n mdt.$FSNAME-MDT0000.recovery_time_hard)
3234 local get_soft_timeout_cmd=\
3235 "$LCTL get_param -n mdt.$FSNAME-MDT0000.recovery_time_soft 2>/dev/null"
3237 echo wait until mds1 recovery_time_soft is $hard_timeout
3238 wait_update $(facet_host mds1) "$get_soft_timeout_cmd" \
3239 "$hard_timeout" $hard_timeout
3241 echo unblock mds_connect from mds2
3242 do_facet mds2 "$LCTL set_param fail_loc=0"
3244 echo upblock recovery updates from mds3
3245 do_facet mds3 "$LCTL set_param fail_loc=0"
3248 [ -f $DIR/$tfile ] || error "$DIR/$tfile does not exist"
3250 run_test 145 "connect mdtlovs and process update logs after recovery expire"
3253 (( $MDS1_VERSION >= $(version_code 2.15.54.6) )) ||
3254 skip "Need MDS >= v2_15_54-6-g3c69d46e17 for eviction_count"
3256 local prev_count=$(do_facet $SINGLEMDS \
3257 $LCTL get_param -n "mdt.${mds1_svc}.eviction_count")
3263 local next_count=$(do_facet $SINGLEMDS \
3264 $LCTL get_param -n "mdt.${mds1_svc}.eviction_count")
3266 [ "$prev_count" -lt "$next_count" ] ||
3267 error "wrong eviction count ($prev_count >= $next_count)"
3269 run_test 146 "test eviction is counted properly"
3272 local obd_timeout=200
3273 local old=$($LCTL get_param -n timeout)
3275 local connection_count
3277 $LFS setstripe -i 0 -c 1 $f
3278 stripe_index=$($LFS getstripe -i $f)
3279 if [ $stripe_index -ne 0 ]; then
3281 error "$f: stripe_index $stripe_index != 0" && return
3284 $LCTL set_param timeout=$obd_timeout
3285 stack_trap "$LCTL set_param timeout=$old && client_reconnect" EXIT
3287 # OBD_FAIL_OST_CONNECT_NET2
3288 # lost reply to connect request
3289 do_facet ost1 lctl set_param fail_loc=0x00000225 timeout=$obd_timeout
3290 stack_trap "do_facet ost1 $LCTL set_param fail_loc=0 timeout=$old" EXIT
3295 $LFS df $MOUNT > /dev/null 2>&1 &
3296 sleep $((obd_timeout * 3 / 4))
3298 $LCTL get_param osc.$FSNAME-OST0000-osc-*.state
3299 connection_count=$($LCTL get_param osc.$FSNAME-OST0000-osc-*.state |
3300 tac | sed "/FULL/,$ d" | grep CONNECTING | wc -l)
3302 echo $connection_count
3303 (($connection_count >= 6)) || error "Client reconnected too slow"
3305 run_test 147 "Check client reconnect"
3308 local wce_param="obdfilter.$FSNAME-OST0000.writethrough_cache_enable"
3309 local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
3310 local amc=$(at_max_get client)
3311 local amo=$(at_max_get ost1)
3316 timeout=$(request_timeout client)
3318 [ "$(facet_fstype ost1)" = "ldiskfs" ] && {
3319 # save old r/o cache settings
3320 save_lustre_params ost1 $wce_param > $p
3323 do_facet ost1 "$LCTL set_param -n $wce_param=0"
3326 $LFS setstripe -i 0 -c 1 $DIR/$tfile
3327 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 oflag=direct
3328 cp $DIR/$tfile $TMP/$tfile
3329 #define OBD_FAIL_OST_BRW_PAUSE_BULK2 0x227
3330 do_facet ost1 $LCTL set_param fail_loc=0x80000227
3331 do_facet ost1 $LCTL set_param fail_val=$((timeout+2))
3332 dd if=/dev/urandom of=$DIR/$tfile bs=4096 count=1 conv=notrunc,fdatasync
3333 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=notrunc,fdatasync
3335 cancel_lru_locks osc
3336 cmp -b $DIR/$tfile $TMP/$tfile || error "wrong data"
3338 rm -f $DIR/$tfile $TMP/$tfile
3340 at_max_set $amc client
3341 at_max_set $amo ost1
3343 [ "$(facet_fstype ost1)" = "ldiskfs" ] && {
3344 # restore initial r/o cache settings
3345 restore_lustre_params < $p
3350 run_test 148 "data corruption through resend"
3353 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
3355 test_mkdir -i 0 -c $MDSCOUNT $DIR/$tdir || error "mkdir $tdir failed"
3357 # make an orphan striped dir
3358 $MULTIOP $DIR/$tdir D_c &
3361 rmdir $DIR/$tdir || error "can't rmdir"
3363 # stop a slave MDT where one ons stripe is located
3366 # stopping should not cause orphan as another MDT can
3370 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "mds1 start fail"
3371 start mds2 $(mdsdevname 2) $MDS_MOUNT_OPTS || error "mds1 start fail"
3379 run_test 149 "skip orphan removal at umount"
3382 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
3386 lazystatfs=$($LCTL get_param -n llite.$FSNAME-*.lazystatfs | head -1)
3387 max=$($LCTL get_param -n llite.$FSNAME-*.statahead_max | head -1)
3389 $LCTL set_param llite.$FSNAME-*.lazystatfs=1
3390 $LCTL set_param llite.$FSNAME-*.statahead_max=0
3391 stack_trap "$LCTL set_param llite.$FSNAME-*.lazystatfs=$lazystatfs" EXIT
3392 stack_trap "$LCTL set_param llite.$FSNAME-*.statahead_max=$max" EXIT
3393 # stop a slave MDT where one ons stripe is located
3396 stack_trap "start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS && \
3397 wait_recovery_complete mds1 && clients_up && true" EXIT
3399 df $MOUNT || error "statfs failed"
3402 run_test 150 "statfs when MDT0 offline with lazystatfs option"
3405 [[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
3406 skip "server does not support overstriping"
3417 large_xattr_enabled || skip_env "ea_inode feature disabled"
3418 version=$(do_facet mds1 \
3419 uname -r | sed -e "s/\([0-9]*\.[0-9]*\.[0-9]*\).*/\1/")
3420 version=$(version_code ${version//\./ })
3421 if (( $version < $(version_code 4.6.0) )); then
3422 skip "MDS Linux kernel does not support killable semaphore"
3425 test_mkdir -i 0 -c 1 -p $DIR/$tdir
3426 test_mkdir -i 0 -c 1 -p $DIR/$tdir/rr
3427 test_mkdir -i 0 -c 1 -p $DIR/$tdir/qos
3428 stack_trap "rm -rf $DIR/$tdir" EXIT
3430 $LFS setstripe -C $setcount $DIR/$tdir/rr/ || error "setstripe failed"
3433 #define OBD_FAIL_MDS_LOD_CREATE_PAUSE 0x173
3434 #Simulate OST failover and sleep RR allocation under lq_rw_sem
3435 do_facet mds1 $LCTL set_param fail_loc=0x80000173 fail_val=20
3437 for (( i = 0; i < 2; i++)); do
3438 touch $DIR/$tdir/rr/$tfile_$i &
3439 pids_rr="$pids_rr $!"
3443 saved=$(do_facet mds1 $LCTL get_param -n lov.*0000*.qos_threshold_rr)
3444 do_facet mds1 $LCTL set_param lov.*.qos_threshold_rr=0
3445 stack_trap "do_facet mds1 $LCTL set_param lov.*.qos_threshold_rr=$saved" EXIT
3447 #create files with QoS algo, killable semaphore sleeps for 2seconds
3448 for (( i = 0; i < 3; i++)); do
3449 touch $DIR/$tdir/qos/$tfile_$i &
3450 pids_qos="$pids_qos $!"
3453 for pid in $pids_qos; do
3458 diff=$((after - before))
3459 echo "QoS allocation took $diff seconds"
3460 for pid in $pids_rr; do
3465 error "QoS allocation slower than RR, killable semaphore doesn't work"
3467 run_test 152 "QoS object allocation could be awakened in case of OST failover"
3470 #define OBD_FAIL_MDS_CONNECT_VS_EVICT 0x174
3471 do_facet mds1 "$LCTL set_param fail_loc=0x174"
3472 # first drop ping reply from MDS and then
3473 # evict on the subsequent reconnect
3474 # (see target_handle_connect)
3475 sleep $((TIMEOUT + 3))
3477 do_facet mds1 "$LCTL set_param fail_loc=0"
3478 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS ||
3479 error "Fail to start $SINGLEMDS"
3480 wait_recovery_complete mds1 || error "MDS recovery not done"
3482 run_test 153 "evict vs reconnect race"
3485 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
3489 local mnt=$(facet_mntpt mds2)
3492 local ula=($(do_facet mds2 od -t u8 $mnt/update_log | \
3493 awk '{ if($2 > 0) printf "0x%x:0x%x:0x0 ", $3, $2 }'))
3494 [ ${#ula[@]} -lt 2 ] && error "update_log file corruption"
3496 local catfile=$mnt/update_log_dir/\[${ula[0]}\]
3498 echo replace file $catfile
3499 do_facet mds2 dd if=/dev/urandom of=$catfile count=100 ||
3500 error "Write file $catfile error"
3503 start mds2 $(mdsdevname 2) $MDS_MOUNT_OPTS || error "mds2 start fail"
3506 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "mds1 start fail"
3508 echo waiting mds1 recovery....
3509 wait_recovery_complete mds1 20
3511 do_facet mds1 $LCTL get_param mdt.$FSNAME-MDT0000.recovery_status | \
3512 grep -w COMPLETE || error "Recovery was blocked"
3514 run_test 154a "corruption update llog can be skipped"
3517 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
3521 stack_trap "do_facet mds1 $LCTL set_param fail_loc=0"
3522 #define OBD_FAIL_TGT_RECOVERY_CONNECT 0x724
3523 do_facet mds1 $LCTL set_param fail_loc=0x0724 fail_val=5
3524 local OPTS="$MDS_MOUNT_OPTS -o abort_recov"
3525 start mds1 $(mdsdevname 1) $OPTS ||
3526 error "start MDS with abort_recovery failed"
3528 echo waiting mds1 recovery....
3529 wait_recovery_complete mds1 30
3530 do_facet mds1 $LCTL get_param mdt.$FSNAME-MDT0000.recovery_status | \
3531 grep -w COMPLETE || error "Recovery was blocked"
3533 mount_client $MOUNT2
3534 umount_client $MOUNT2
3535 remount_client $MOUNT
3537 run_test 154b "restore update llog after failed recovery"
3544 lsoutput1=$(ls -l $DIR)
3546 zconf_umount $HOSTNAME $MOUNT || error "umount failed"
3547 # make sure that last_rcvd update is committed
3549 zconf_mount $HOSTNAME $MOUNT || error "mount failed"
3551 replay_barrier_nosync mds1
3555 lsoutput2=$(ls -l $DIR) || error "ls failed"
3556 [[ $lsoutput1 == $lsoutput2 ]] || error "$lsoutput1 != $lsoutput2"
3558 run_test 155 "failover after client remount"
3562 (( OST1_VERSION >= $(version_code v2_15_60-90-g9df01eee75) )) ||
3563 skip "Need OST >= 2.15.60.90 for tot_granted miscount fix"
3565 # on failover recovery time hard will be 9 * 5
3566 local saved_timeout=$(do_facet ost1 $LCTL get_param -n timeout)
3568 do_facet mgs $LCTL set_param -P timeout=5 ||
3569 error "failed to set obd_timeout"
3570 stack_trap "do_facet mgs $LCTL set_param -P timeout=$saved_timeout" \
3573 $LFS setstripe -c 1 -i 0 $DIR/$tfile || error "setstripe failed"
3575 # this is to sync last_rcvd, so that the client will have to
3576 # send replay on recovery
3582 $MULTIOP $DIR/$tfile oO_RDWR:O_SYNC:w1048576c || error "multiop failed"
3584 # delay write replay for 45 sec (OBD_RECOVERY_TIME_HARD) to
3585 # get the client evicted as not sending replays
3587 #define OBD_FAIL_PTLRPC_REPLAY_PAUSE 0x536
3588 $LCTL set_param fail_loc=0x80000536 fail_val=45
3592 # check that ost1 evicted the client in recovery
3594 clients=($(do_facet ost1 \
3595 $LCTL get_param -n obdfilter.$FSNAME-OST0000.recovery_status |
3596 awk '/completed_clients/ { print $2 }' | tr '/' '\n'))
3597 [[ $((${clients[0]} + 1)) == ${clients[1]} ]] ||
3598 error "client not evicted by ost1"
3600 local testid=$(echo $TESTNAME | tr '_' ' ')
3601 do_facet ost1 dmesg | tac | sed "/$testid/,$ d" |
3602 grep "ofd_obd_disconnect: tot_granted" &&
3603 error "grant miscount" || true
3605 run_test 156 "tot_granted miscount after client eviction"
3609 $LFS setstripe -i 0 -c 1 $DIR/$tfile || error "setstripe failed"
3610 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1
3611 cancel_lru_locks osc
3613 #define OBD_FAIL_LLITE_FAULT_PAUSE 0x1432
3614 $LCTL set_param fail_loc=0x80001432 fail_val=3
3615 $MULTIOP $DIR/$tfile soO_RDWR:MRUc &
3620 [[ $? == 135 ]] || error "multiop failed with not SIGBUS"
3622 run_test 157 "eviction during mmaped i/o"
3624 complete_test $SECONDS
3625 check_and_cleanup_lustre