5 PTLDEBUG=${PTLDEBUG:--1}
6 LUSTRE=${LUSTRE:-$(dirname $0)/..}
7 . $LUSTRE/tests/test-framework.sh
11 ALWAYS_EXCEPT="$RECOVERY_SMALL_EXCEPT "
15 require_dsh_mds || exit 0
17 # Allow us to override the setup if we already have a mounted system by
18 # setting SETUP=" " and CLEANUP=" "
20 CLEANUP=${CLEANUP:-""}
22 check_and_setup_lustre
25 rm -rf $DIR/d[0-9]* $DIR/f.${TESTSUITE}*
30 local f1="$DIR/$tfile"
31 local f2="$DIR/$tfile.2"
33 drop_request "mcreate $f1" ||
34 error_noexit "create '$f1': drop req"
36 drop_reint_reply "mcreate $f2" ||
37 error_noexit "create '$f2': drop rep"
39 drop_request "tchmod 111 $f2" ||
40 error_noexit "chmod '$f2': drop req"
42 drop_reint_reply "tchmod 666 $f2" ||
43 error_noexit "chmod '$f2': drop rep"
45 drop_request "statone $f2" ||
46 error_noexit "stat '$f2': drop req"
48 drop_reply "statone $f2" ||
49 error_noexit "stat '$f2': drop rep"
51 run_test 1 "create, chmod, stat: drop req, drop rep"
55 do_facet_create_file client $t 10K ||
56 error_noexit "Create file $t"
58 drop_request "cat $t > /dev/null" ||
59 error_noexit "Open request for $t file"
61 drop_reply "cat $t > /dev/null" ||
62 error_noexit "Open replay for $t file"
64 run_test 4 "open: drop req, drop rep"
69 local RR="$T-renamed-again"
70 do_facet_create_file client $T 10K ||
71 error_noexit "Create file $T"
73 drop_request "mv $T $R" ||
74 error_noexit "Rename $T"
76 drop_reint_reply "mv $R $RR" ||
77 error_noexit "Failed rename replay on $R"
79 do_facet client "checkstat -v $RR" ||
80 error_noexit "checkstat error on $RR"
82 do_facet client "rm $RR" ||
83 error_noexit "Can't remove file $RR"
85 run_test 5 "rename: drop req, drop rep"
89 local LINK1=$DIR/$tfile.link1
90 local LINK2=$DIR/$tfile.link2
92 do_facet_create_file client $T 10K || error_noexit "Create file $T"
93 drop_request "link $T $LINK1" || error_noexit "link request for $T"
94 drop_reint_reply "link $T $LINK2" || error_noexit "link reply for $T"
95 drop_request "unlink $LINK1" || error_noexit "unlink request for $T"
96 drop_reint_reply "unlink $LINK2" || error_noexit "unlink reply for $T"
97 do_facet client "rm $T" || error_noexit "Can't remove file $T"
99 run_test 6 "link, unlink: drop req, drop rep"
103 drop_reint_reply "touch $DIR/$tfile" || return 1
105 run_test 8 "touch: drop rep (bug 1423)"
109 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
113 do_facet_random_file client $TMP/$tfile 1K ||
114 error_noexit "Create random file $TMP/$tfile"
115 # make this big, else test 9 doesn't wait for bulk -- bz 5595
116 do_facet_create_file client $TMP/$t1 4M ||
117 error_noexit "Create file $TMP/$t1"
118 do_facet client "cp $TMP/$t1 $DIR/$t1" ||
119 error_noexit "Can't copy to $DIR/$t1 file"
120 pause_bulk "cp $TMP/$tfile $DIR/$tfile" ||
121 error_noexit "Can't pause_bulk copy"
122 do_facet client "cp $TMP/$t1 $DIR/$t2" ||
123 error_noexit "Can't copy file"
124 do_facet client "sync"
125 do_facet client "rm $DIR/$tfile $DIR/$t2 $DIR/$t1" ||
126 error_noexit "Can't remove files"
127 do_facet client "rm $TMP/$t1 $TMP/$tfile"
129 run_test 9 "pause bulk on OST (bug 1420)"
133 local before=$(date +%s)
136 do_facet client "stat $DIR > /dev/null" ||
137 error "failed to stat $DIR: $?"
138 drop_bl_callback "chmod 0777 $DIR" ||
139 error "failed to chmod $DIR: $?"
141 # let the client reconnect
143 evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
144 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
145 [ ! -z "$evict" ] && [[ $evict -gt $before ]] ||
146 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
147 error "no eviction: $evict before:$before")
149 do_facet client checkstat -v -p 0777 $DIR ||
150 error "client checkstat failed: $?"
152 run_test 10a "finish request on server after client eviction (bug 1521)"
155 local before=$(date +%s)
158 [[ "$MDS1_VERSION" -lt $(version_code 2.6.53) ]] &&
159 skip "Need MDS version at least 2.6.53"
160 do_facet client "stat $DIR > /dev/null" ||
161 error "failed to stat $DIR: $?"
162 drop_bl_callback_once "chmod 0777 $DIR" ||
163 error "failed to chmod $DIR: $?"
165 # let the client reconnect
167 evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
168 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
170 [ -z "$evict" ] || [[ $evict -le $before ]] ||
171 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
172 error "eviction happened: $evict before:$before")
174 do_facet client checkstat -v -p 0777 $DIR ||
175 error "client checkstat failed: $?"
177 run_test 10b "re-send BL AST"
180 local before=$(date +%s)
189 workdir="${DIR}/${tdir}"
190 mkdir -p ${workdir} || error "can't create workdir $?"
191 stat ${workdir} > /dev/null ||
192 error "failed to stat ${workdir}: $?"
193 mdtidx=$($LFS getdirstripe -i ${workdir})
194 mdtname=$($LFS mdts ${workdir} | grep -e "^$mdtidx:" |
195 awk '{sub("_UUID", "", $2); print $2;}')
197 mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
198 conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid)
199 mdcpath="mdc.${mdccli}.import=connection=${conn_uuid}"
201 drop_bl_callback_once "chmod 0777 ${workdir}" &
206 # force client reconnect
207 $LCTL set_param "${mdcpath}"
209 # wait client reconnect
213 evict=$($LCTL get_param mdc.${mdccli}.state |
214 awk -F"[ [,]" '/EVICTED]$/ { if (t<$4) {t=$4;} } END { print t }')
216 [[ $evict -le $before ]] ||
217 ( $LCTL get_param mdc.$FSNAME-MDT*.state;
218 error "eviction happened: $EVICT before:$BEFORE" )
220 [ $rc -eq 0 ] || error "chmod must finished OK"
221 checkstat -v -p 0777 "${workdir}" ||
222 error "client checkstat failed: $?"
224 run_test 10c "re-send BL AST vs reconnect race (LU-5569)"
227 local before=$(date +%s)
230 [[ "$MDS1_VERSION" -lt $(version_code 2.6.90) ]] &&
231 skip "Need MDS version at least 2.6.90"
233 # sleep 1 is to make sure that BEFORE is not equal to EVICTED below
236 echo -n ", world" | dd of=$TMP/$tfile bs=1c seek=5
238 remount_client $MOUNT
242 $LFS setstripe -i 0 -c 1 $DIR1/$tfile
243 echo -n hello | dd of=$DIR1/$tfile bs=5
245 stat $DIR2/$tfile >& /dev/null
246 $LCTL set_param fail_err=71
247 drop_bl_callback "echo -n \\\", world\\\" >> $DIR2/$tfile"
252 cmp -l $DIR1/$tfile $DIR2/$tfile || error "file contents differ"
253 cmp -l $DIR1/$tfile $TMP/$tfile || error "wrong content found"
255 evict=$(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state | \
257 awk -F"[ [,]" '/EVICTED$/ { if (mx<$1) {mx=$1;} } END { print mx }')
259 [[ $evict -gt $before ]] ||
260 (do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state;
261 error "no eviction: $evict before:$before")
263 $LCTL set_param fail_err=0
265 umount_client $MOUNT2
267 run_test 10d "test failed blocking ast"
271 [[ "$OST1_VERSION" -le $(version_code 2.8.58) ]] &&
272 skip "Need OST version at least 2.8.59"
273 [ $CLIENTCOUNT -lt 2 ] && skip "need two clients"
274 [ $(facet_host client) == $(facet_host ost1) ] &&
275 skip "need ost1 and client on different nodes"
276 local -a clients=(${CLIENTS//,/ })
277 local client1=${clients[0]}
278 local client2=${clients[1]}
280 $LFS setstripe -c 1 -i 0 $DIR/$tfile-1 $DIR/$tfile-2
281 $MULTIOP $DIR/$tfile-1 Ow1048576c
283 #define OBD_FAIL_LDLM_BL_CALLBACK_NET 0x305
284 $LCTL set_param fail_loc=0x80000305
286 #define OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT 0x30e
287 do_facet ost1 "$LCTL set_param fail_loc=0x1000030e"
288 # hit OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT twice:
289 # 1. to return ENOTCONN from ldlm_handle_enqueue0
290 # 2. to pause reconnect handling between resend and setting
291 # import to LUSTRE_IMP_FULL state
292 do_facet ost1 "$LCTL set_param fail_val=3"
294 # client1 fails ro respond to bl ast
295 do_node $client2 "$MULTIOP $DIR/$tfile-1 Ow1048576c" &
298 # ost1 returns error on enqueue, which causes client1 to reconnect
299 do_node $client1 "$MULTIOP $DIR/$tfile-2 Ow1048576c" ||
300 error "multiop failed"
303 do_facet ost1 "$LCTL set_param fail_loc=0"
304 do_facet ost1 "$LCTL set_param fail_val=0"
306 run_test 10e "re-send BL AST vs reconnect race 2"
309 # wake up a thread waiting for completion after eviction
311 do_facet client $MULTIOP $DIR/$tfile Ow ||
312 { error "multiop write failed: $?"; return 1; }
313 do_facet client $MULTIOP $DIR/$tfile or ||
314 { error "multiop read failed: $?"; return 2; }
318 do_facet client $MULTIOP $DIR/$tfile or ||
319 { error "multiop read failed: $?"; return 3; }
320 drop_bl_callback_once $MULTIOP $DIR/$tfile Ow ||
321 echo "evicted as expected"
323 do_facet client unlink $DIR/$tfile ||
324 { error "unlink failed: $?"; return 4; }
325 # allow recovery to complete
326 client_up || client_up || sleep $TIMEOUT
328 run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
332 $LCTL mark $MULTIOP $DIR/$tfile OS_c
333 do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
334 clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
335 multiop_bg_pause $DIR/$tfile OS_c ||
336 { error "multiop failed: $?"; return 1; }
338 #define OBD_FAIL_MDS_CLOSE_NET 0x115
340 echo "waiting for multiop $PID"
341 wait $PID || { error "wait for multiop faile: $?"; return 2; }
342 do_facet client unlink $DIR/$tfile ||
343 { error "client unlink failed: $?"; return 3; }
344 # allow recovery to complete
345 client_up || client_up || sleep $TIMEOUT
347 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
349 # Bug 113, check that readdir lost recv timeout works.
351 mkdir_on_mdt0 $DIR/$tdir || { error "mkdir failed: $?"; return 1; }
352 touch $DIR/$tdir/newentry || { error "touch failed: $?"; return 2; }
353 # OBD_FAIL_MDS_READPAGE_NET|CFS_FAIL_ONCE
354 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104"
355 ls $DIR/$tdir || { error "ls failed: $?"; return 3; }
356 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
357 rm -rf $DIR/$tdir || { error "remove test dir failed: $?"; return 4; }
359 run_test 13 "mdc_readpage restart test (bug 1138)"
361 # Bug 113, check that readdir lost send timeout works.
363 mkdir_on_mdt0 $DIR/$tdir
364 touch $DIR/$tdir/newentry
365 # OBD_FAIL_MDS_SENDPAGE|CFS_FAIL_ONCE
366 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000106"
367 ls $DIR/$tdir || return 1
368 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
370 run_test 14 "mdc_readpage resend test (bug 1138)"
373 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
374 touch $DIR/$tfile && return 1
377 run_test 15 "failed open (-ENOMEM)"
379 READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1`
381 lctl set_param -n llite.*.max_read_ahead_mb 0
385 lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD
389 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
391 do_facet_random_file client $TMP/$tfile 100K ||
392 { error_noexit "Create random file $TMP/$T" ; return 0; }
393 do_facet client "cp $TMP/$tfile $DIR/$tfile" ||
394 { error_noexit "Copy to $DIR/$tfile file" ; return 0; }
398 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | CFS_FAIL_ONCE
399 do_facet ost1 "lctl set_param fail_loc=0x80000504"
401 # OST bulk will time out here, client resends
402 do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 1
403 do_facet ost1 lctl set_param fail_loc=0
404 # give recovery a chance to finish (shouldn't take long)
406 do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 2
410 run_test 16 "timeout bulk put, don't evict client (2732)"
415 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
417 local SAMPLE_FILE=$TMP/$tfile
418 do_facet_random_file client $SAMPLE_FILE 20K ||
419 { error_noexit "Create random file $SAMPLE_FILE" ; return 0; }
421 # With adaptive timeouts, bulk_get won't expire until
422 # adaptive_timeout_max
423 if at_is_enabled; then
424 at_max_saved=$(at_max_get ost1)
425 at_max_set $TIMEOUT ost1
428 # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | CFS_FAIL_ONCE
429 # OST bulk will time out here, client retries
430 do_facet ost1 lctl set_param fail_loc=0x80000503
431 # need to ensure we send an RPC
432 do_facet client cp $SAMPLE_FILE $DIR/$tfile
435 # with AT, client will wait adaptive_max*factor+net_latency before
436 # expiring the req, hopefully timeout*2 is enough
437 sleep $(($TIMEOUT*2))
439 do_facet ost1 lctl set_param fail_loc=0
440 do_facet client "df $DIR"
441 # expect cmp to succeed, client resent bulk
442 do_facet client "cmp $SAMPLE_FILE $DIR/$tfile" || return 3
443 do_facet client "rm $DIR/$tfile" || return 4
444 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved ost1
447 run_test 17a "timeout bulk get, don't evict client (2732)"
450 [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
452 # get one of the clients from client list
453 local rcli=$(echo $RCLIENTS | cut -d ' ' -f 1)
454 local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
455 local ldlm_enqueue_min=$(do_facet ost1 find /sys -name ldlm_enqueue_min)
456 [ -z "$ldlm_enqueue_min" ] &&
457 skip "missing /sys/.../ldlm_enqueue_min" && return 0
459 $LFS setstripe -i 0 -c 1 -S 1048576 $DIR/$tfile ||
460 error "setstripe failed"
461 $LFS setstripe -i 0 -c 1 -S 1048576 $DIR/${tfile}2 ||
462 error "setstripe 2 failed"
464 save_lustre_params ost1 "at_history" > $p
465 save_lustre_params ost1 "bulk_timeout" >> $p
466 local dev="${FSNAME}-OST0000"
467 save_lustre_params ost1 "obdfilter.$dev.brw_size" >> $p
468 local ldlm_enqueue_min_save=$(do_facet ost1 cat $ldlm_enqueue_min)
470 local new_at_history=15
472 do_facet ost1 "$LCTL set_param at_history=$new_at_history"
473 do_facet ost1 "$LCTL set_param bulk_timeout=30"
474 do_facet ost1 "echo 30 > /sys/module/ptlrpc/parameters/ldlm_enqueue_min"
476 # brw_size is required to be 4m so that bulk transfer timeout
477 # could be simulated with OBD_FAIL_PTLRPC_CLIENT_BULK_CB3
478 local brw_size=$($LCTL get_param -n \
479 osc.$FSNAME-OST0000-osc-[^M]*.import |
480 awk '/max_brw_size/{print $2}')
481 if [ $brw_size -ne 4194304 ]
483 save_lustre_params ost1 "obdfilter.$dev.brw_size" >> $p
484 do_facet ost1 "$LCTL set_param obdfilter.$dev.brw_size=4"
485 remount_client $MOUNT
488 # get service estimate expanded
489 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
490 do_facet ost1 "$LCTL set_param fail_loc=0x80000224 fail_val=35"
491 echo "delay rpc servicing by 35 seconds"
492 dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=fdatasync
495 estimate=$($LCTL get_param -n osc.$dev-osc-*.timeouts |
496 awk '/portal 6/ {print $5}')
497 echo "service estimates increased to $estimate"
499 # let current worst service estimate to get closer to obliteration
500 sleep $((new_at_history / 3))
502 # start i/o and simulate bulk transfer loss
503 #define OBD_FAIL_PTLRPC_CLIENT_BULK_CB3 0x520
504 do_facet ost1 "$LCTL set_param fail_loc=0xa0000520 fail_val=1"
505 dd if=/dev/zero of=$DIR/$tfile bs=2M count=1 conv=fdatasync,notrunc &
508 # start lock conflict handling
509 sleep $((new_at_history / 3))
510 do_node $rcli "dd if=$DIR/$tfile of=/dev/null bs=1M count=1" &
513 # obliterate the worst service estimate
514 sleep $((new_at_history / 3 + 1))
515 dd if=/dev/zero of=$DIR/${tfile}2 bs=1M count=1
517 estimate=$($LCTL get_param -n osc.$dev-osc-*.timeouts |
518 awk '/portal 6/ {print $5}')
519 echo "service estimate dropped to $estimate"
522 [[ $? == 0 ]] || error "write failed"
524 [[ $? == 0 ]] || error "read failed"
526 restore_lustre_params <$p
527 if [ $brw_size -ne 4194304 ]
529 remount_client $MOUNT || error "remount_client failed"
531 do_facet ost1 "echo $ldlm_enqueue_min_save > $ldlm_enqueue_min"
533 run_test 17b "timeout bulk get, dont evict client (3582)"
536 [ -z ${ost2_svc} ] && skip_env "needs 2 osts" && return 0
538 do_facet_create_file client $TMP/$tfile 20K ||
539 { error_noexit "Create file $TMP/$tfile" ; return 0; }
541 do_facet client mkdir -p $DIR/$tdir
545 pgcache_empty || return 1
548 $LFS setstripe -i 1 -c 1 $f
549 stripe_index=$($LFS getstripe -i $f)
550 if [ $stripe_index -ne 1 ]; then
552 error "$f: stripe_index $stripe_index != 1" && return
555 do_facet client cp $TMP/$tfile $f
557 local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | egrep -v 'MDT' | awk '{print $1}'`
558 $LCTL --device $osc2dev deactivate || return 3
559 # my understanding is that there should be nothing in the page
560 # cache after the client reconnects?
562 pgcache_empty || rc=2
563 $LCTL --device $osc2dev activate
567 run_test 18a "manual ost invalidate clears page cache immediately"
570 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
572 do_facet_create_file client $TMP/$tfile 20K ||
573 { error_noexit "Create file $TMP/$tfile" ; return 0; }
575 do_facet client mkdir -p $DIR/$tdir
579 pgcache_empty || return 1
581 $LFS setstripe -i 0 -c 1 $f
582 stripe_index=$($LFS getstripe -i $f)
583 if [ $stripe_index -ne 0 ]; then
585 error "$f: stripe_index $stripe_index != 0" && return
588 do_facet client cp $TMP/$tfile $f
591 # allow recovery to complete
592 sleep $((TIMEOUT + 2))
593 # my understanding is that there should be nothing in the page
594 # cache after the client reconnects?
596 pgcache_empty || rc=2
600 run_test 18b "eviction and reconnect clears page cache (2766)"
603 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
605 do_facet_create_file client $TMP/$tfile 20K ||
606 { error_noexit "Create file $TMP/$tfile" ; return 0; }
608 do_facet client mkdir -p $DIR/$tdir
612 pgcache_empty || return 1
614 $LFS setstripe -i 0 -c 1 $f
615 stripe_index=$($LFS getstripe -i $f)
616 if [ $stripe_index -ne 0 ]; then
618 error "$f: stripe_index $stripe_index != 0" && return
621 do_facet client cp $TMP/$tfile $f
625 # OBD_FAIL_OST_CONNECT_NET2
626 # lost reply to connect request
627 do_facet ost1 lctl set_param fail_loc=0x80000225
630 $LFS df $MOUNT > /dev/null 2>&1
632 # my understanding is that there should be nothing in the page
633 # cache after the client reconnects?
635 pgcache_empty || rc=2
639 run_test 18c "Dropped connect reply after eviction handing (14755)"
642 local BEFORE=`date +%s`
645 mount_client $DIR2 || error "failed to mount $DIR2"
647 # cancel cached locks from OST to avoid eviction from it
650 do_facet client "stat $DIR > /dev/null" ||
651 error "failed to stat $DIR: $?"
652 drop_ldlm_cancel "chmod 0777 $DIR2" ||
653 error "failed to chmod $DIR2"
657 # let the client reconnect
659 EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
660 awk -F"[ [,]" '/EVICTED ]$/ \
661 { if (mx<$5) {mx=$5;} } END { print mx }')
663 [ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
664 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
665 error "no eviction: $EVICT before:$BEFORE")
667 run_test 19a "test expired_lock_main on mds (2867)"
670 local BEFORE=`date +%s`
673 mount_client $DIR2 || error "failed to mount $DIR2: $?"
675 # cancel cached locks from MDT to avoid eviction from it
678 do_facet client $MULTIOP $DIR/$tfile Ow ||
679 error "failed to run multiop: $?"
680 drop_ldlm_cancel $MULTIOP $DIR2/$tfile Ow ||
681 error "failed to ldlm_cancel: $?"
683 umount_client $DIR2 || error "failed to unmount $DIR2: $?"
684 do_facet client unlink $DIR/$tfile ||
685 error "failed to unlink $DIR/$tfile: $?"
687 # let the client reconnect
689 EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state |
690 awk -F"[ [,]" '/EVICTED ]$/ \
691 { if (mx < $5) {mx = $5;} } END { print mx }')
693 [ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
694 (do_facet client $LCTL get_param osc.$FSNAME-OST*.state;
695 error "no eviction: $EVICT before:$BEFORE")
697 run_test 19b "test expired_lock_main on ost (2867)"
700 local BEFORE=`date +%s`
703 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=0
705 mkdir -p $DIR1/$tfile
708 #define OBD_FAIL_PTLRPC_CANCEL_RESEND 0x516
709 do_facet mds $LCTL set_param fail_loc=0x80000516
711 touch $DIR2/$tfile/file1 &
713 # let touch to get blocked on the server
717 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=1
720 # let the client reconnect
722 EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
723 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
725 [ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
727 run_test 19c "check reconnect and lock resend do not trigger expired_lock_main"
729 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
730 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
733 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
734 multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
737 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
738 do_facet ost1 lctl set_param fail_loc=0x80000308
739 kill -USR1 $MULTI_PID
742 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
744 run_test 20a "ldlm_handle_enqueue error (should return error)"
746 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
747 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
750 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
752 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
753 do_facet ost1 lctl set_param fail_loc=0x80000308
754 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
755 error "didn't fail open enqueue" || true
757 run_test 20b "ldlm_handle_enqueue error (should return error)"
760 mkdir_on_mdt0 $DIR/$tdir-1
761 mkdir_on_mdt0 $DIR/$tdir-2
762 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
765 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
766 $MULTIOP $DIR/$tdir-2/f Oc &
769 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
771 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
772 kill -USR1 $close_pid
774 wait $close_pid || return 1
775 wait $open_pid || return 2
776 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
778 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
779 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
783 run_test 21a "drop close request while close and open are both in flight"
786 mkdir_on_mdt0 $DIR/$tdir-1
787 mkdir_on_mdt0 $DIR/$tdir-2
788 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
791 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
792 mcreate $DIR/$tdir-2/f &
795 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
797 kill -USR1 $close_pid
799 wait $close_pid || return 1
800 wait $open_pid || return 3
802 $CHECKSTAT -t file $DIR/$tdir-1/f || return 4
803 $CHECKSTAT -t file $DIR/$tdir-2/f || return 5
806 run_test 21b "drop open request while close and open are both in flight"
809 mkdir_on_mdt0 $DIR/$tdir-1
810 mkdir_on_mdt0 $DIR/$tdir-2
811 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
814 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
815 mcreate $DIR/$tdir-2/f &
818 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
820 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
821 kill -USR1 $close_pid
823 wait $close_pid || return 1
824 wait $open_pid || return 2
826 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
828 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
829 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
832 run_test 21c "drop both request while close and open are both in flight"
835 mkdir_on_mdt0 $DIR/$tdir-1
836 mkdir_on_mdt0 $DIR/$tdir-2
837 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
840 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
841 $MULTIOP $DIR/$tdir-2/f Oc &
843 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
845 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
848 wait $pid || return 1
849 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
851 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
852 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
856 run_test 21d "drop close reply while close and open are both in flight"
859 mkdir_on_mdt0 $DIR/$tdir-1
860 mkdir_on_mdt0 $DIR/$tdir-2
861 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
864 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
865 touch $DIR/$tdir-2/f &
867 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
871 wait $pid || return 1
874 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
875 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
878 run_test 21e "drop open reply while close and open are both in flight"
881 mkdir_on_mdt0 $DIR/$tdir-1
882 mkdir_on_mdt0 $DIR/$tdir-2
883 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
886 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
887 touch $DIR/$tdir-2/f &
889 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
891 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
894 wait $pid || return 1
895 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
897 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
898 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
901 run_test 21f "drop both reply while close and open are both in flight"
904 mkdir_on_mdt0 $DIR/$tdir-1
905 mkdir_on_mdt0 $DIR/$tdir-2
906 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
909 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
910 touch $DIR/$tdir-2/f &
912 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
914 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
917 wait $pid || return 1
918 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
920 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
921 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
924 run_test 21g "drop open reply and close request while close and open are both in flight"
927 mkdir_on_mdt0 $DIR/$tdir-1
928 mkdir_on_mdt0 $DIR/$tdir-2
929 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
932 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
933 touch $DIR/$tdir-2/f &
936 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
938 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
941 wait $pid || return 1
942 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
944 wait $touch_pid || return 2
946 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
947 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
950 run_test 21h "drop open request and close reply while close and open are both in flight"
952 # bug 3462 - multiple MDC requests
957 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
962 $MULTIOP $f1 msu || return 1
965 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
967 wait $close_pid || return 2
968 rm -rf $f2 || return 4
970 run_test 22 "drop close request and do mknod"
973 multiop_bg_pause $DIR/$tfile O_c || return 1
975 # give a chance for open
979 drop_request "kill -USR1 $pid"
982 wait $pid || return 1
985 run_test 23 "client hang when close a file after mds crash"
987 test_24a() { # bug 11710 details correct fsync() behavior
988 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
991 $LFS setstripe -i 0 -c 1 $DIR/$tdir
993 multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
996 kill -USR1 $MULTI_PID
999 lctl set_param fail_loc=0x0
1002 error_ignore bz5494 "multiop didn't fail fsync: rc $rc" || true
1004 run_test 24a "fsync error (should return error)"
1006 wait_client_evicted () {
1009 local varsvc=${facet}_svc
1011 wait_update $(facet_active_host $facet) \
1012 "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \
1017 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1019 dmesg -c > /dev/null
1021 lfs setstripe $DIR/$tdir -S 0 -i 0 -c 1 ||
1022 error "$LFS setstripe failed"
1023 cancel_lru_locks osc
1024 multiop_bg_pause $DIR/$tdir/$tfile-1 Ow8192_yc ||
1025 error "mulitop Ow8192_yc failed"
1028 multiop_bg_pause $DIR/$tdir/$tfile-2 Ow8192_c ||
1029 error "mulitop Ow8192_c failed"
1034 kill -USR1 $MULTI_PID1
1037 kill -USR1 $MULTI_PID2
1040 lctl set_param fail_loc=0x0
1042 [ $rc1 -eq 0 -o $rc2 -eq 0 ] &&
1043 error_ignore bz5494 "multiop didn't fail fsync: $rc1 or close: $rc2" ||
1046 dmesg | grep "dirty page discard:" ||
1047 error "no discarded dirty page found!"
1049 run_test 24b "test dirty page discard due to client eviction"
1051 test_26a() { # was test_26 bug 5921 - evict dead exports by pinger
1052 # this test can only run from a client on a separate node.
1053 remote_ost || { skip "local OST" && return 0; }
1054 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1055 remote_mds || { skip "local MDS" && return 0; }
1057 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
1058 skip "msg and ost1 are at the same node"
1062 check_timeout || return 1
1064 # make sure all imports are connected and not IDLE
1065 do_facet client lfs df > /dev/null
1066 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
1067 do_facet client lctl set_param fail_loc=0x505
1068 local before=$(date +%s)
1071 # evictor takes PING_EVICT_TIMEOUT to evict.
1072 # But if there's a race to start the evictor from various obds,
1073 # the loser might have to wait for the next ping.
1074 sleep $((TIMEOUT * 2))
1075 do_facet client lctl set_param fail_loc=0x0
1076 do_facet client lfs df > /dev/null
1078 local oscs=$(lctl dl | awk '/-osc-/ {print $4}')
1079 check_clients_evicted "$before ${oscs[@]}"
1080 check_clients_full 10 "${oscs[@]}"
1082 run_test 26a "evict dead exports"
1084 test_26b() { # bug 10140 - evict dead exports by pinger
1085 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1087 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
1088 skip "msg and ost1 are at the same node"
1092 check_timeout || return 1
1094 zconf_mount `hostname` $MOUNT2 ||
1095 { error "Failed to mount $MOUNT2"; return 2; }
1096 sleep 1 # wait connections being established
1098 local MDS_NEXP=$(do_facet $SINGLEMDS lctl get_param -n mdt.${mds1_svc}.num_exports | cut -d' ' -f2)
1099 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
1101 echo starting with $OST_NEXP OST and $MDS_NEXP MDS exports
1103 zconf_umount `hostname` $MOUNT2 -f
1105 # PING_INTERVAL max(obd_timeout / 4, 1U)
1106 # PING_EVICT_TIMEOUT (PING_INTERVAL * 6)
1108 # evictor takes PING_EVICT_TIMEOUT to evict.
1109 # But if there's a race to start the evictor from various obds,
1110 # the loser might have to wait for the next ping.
1111 # = 6 * PING_INTERVAL + PING_INTERVAL
1112 # = 7 PING_INTERVAL = 7 obd_timeout / 4 = (1+3/4)obd_timeout
1113 # let's wait $((TIMEOUT * 2)) # bug 19887
1114 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 2)) ||
1115 error "Client was not evicted by ost"
1116 wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 2)) ||
1117 error "Client was not evicted by mds"
1119 run_test 26b "evict dead exports"
1122 mkdir_on_mdt0 $DIR/$tdir
1123 writemany -q -a $DIR/$tdir/$tfile 0 5 &
1126 local save_FAILURE_MODE=$FAILURE_MODE
1128 facet_failover $SINGLEMDS
1129 #define OBD_FAIL_OSC_SHUTDOWN 0x407
1130 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000407
1131 # need to wait for reconnect
1132 echo waiting for fail_loc
1133 wait_update_facet $SINGLEMDS "lctl get_param -n fail_loc" "-2147482617"
1134 facet_failover $SINGLEMDS
1135 #no crashes allowed!
1136 kill -USR1 $CLIENT_PID
1139 FAILURE_MODE=$save_FAILURE_MODE
1141 run_test 27 "fail LOV while using OSC's"
1143 test_28() { # bug 6086 - error adding new clients
1144 do_facet client mcreate $DIR/$tfile || return 1
1145 drop_bl_callback_once "chmod 0777 $DIR/$tfile" ||
1146 echo "evicted as expected"
1147 #define OBD_FAIL_MDS_CLIENT_ADD 0x12f
1148 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
1149 # fail once (evicted), reconnect fail (fail_loc), ok
1150 client_up || (sleep 10; client_up) || (sleep 10; client_up) || error "reconnect failed"
1152 fail $SINGLEMDS # verify MDS last_rcvd can be loaded
1154 run_test 28 "handle error adding new clients (bug 6086)"
1156 test_29a() { # bug 22273 - error adding new clients
1157 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
1158 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000711"
1159 # fail abort so client will be new again
1160 fail_abort $SINGLEMDS
1161 client_up || error "reconnect failed"
1162 wait_osc_import_state $SINGLEMDS ost FULL
1165 run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
1167 test_29b() { # bug 22273 - error adding new clients
1168 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
1169 do_facet ost1 "lctl set_param fail_loc=0x80000711"
1170 # fail abort so client will be new again
1172 client_up || error "reconnect failed"
1175 run_test 29b "error adding new clients doesn't cause LBUG (bug 22273)"
1179 # put a load of file creates/writes/deletes
1180 writemany -q $DIR/$tdir/$tfile 0 5 &
1182 echo writemany pid $CLIENT_PID
1186 # wait for client to reconnect to MDS
1191 # client process should see no problems even though MDS went down
1193 kill -USR1 $CLIENT_PID
1196 echo writemany returned $rc
1197 #these may fail because of eviction due to slow AST response.
1199 error_ignore bz13652 "writemany returned rc $rc" || true
1201 run_test 50 "failover MDS under load"
1204 #define OBD_FAIL_MDS_SYNC_CAPA_SL 0x1310
1205 do_facet ost1 lctl set_param fail_loc=0x00001310
1207 mkdir_on_mdt0 $DIR/$tdir
1208 # put a load of file creates/writes/deletes
1209 writemany -q $DIR/$tdir/$tfile 0 5 &
1213 facet_failover $SINGLEMDS
1214 # failover at various points during recovery
1215 SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
1216 echo will failover at $SEQ
1218 #echo failover in $i sec
1219 log "$TESTNAME: failover in $i sec"
1221 facet_failover $SINGLEMDS
1223 # client process should see no problems even though MDS went down
1224 # and recovery was interrupted
1226 kill -USR1 $CLIENT_PID
1229 echo writemany returned $rc
1231 error_ignore bz13652 "writemany returned rc $rc" || true
1233 run_test 51 "failover MDS during recovery"
1236 do_facet client "mkdir -p $DIR/$tdir"
1237 do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
1239 echo writemany pid $CLIENT_PID
1244 wait $CLIENT_PID || rc=$?
1245 # active client process should see an EIO for down OST
1246 [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
1247 # but timing or failover setup may allow success
1248 [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
1249 echo "writemany returned $rc"
1254 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1259 [ $rc -ne 0 ] && { return $rc; }
1260 # wait for client to reconnect to OST
1264 [ $rc -ne 0 ] && { return $rc; }
1271 run_test 52 "failover OST under load"
1273 # test of open reconstruct
1276 drop_mdt_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
1279 run_test 53a "touch: drop rep"
1284 drop_mdt_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||
1287 run_test 53b "touch: drop rep"
1292 drop_mdt_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||
1295 run_test 53c "touch: drop rep"
1298 zconf_mount $(hostname) $MOUNT2
1300 touch $DIR2/$tfile.1
1302 cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
1305 ERROR=$(dmesg | egrep "(test 54|went back in time)" | tail -n1 |
1306 grep "went back in time")
1307 [ x"$ERROR" == x ] || error "back in time occured"
1309 run_test 54 "back in time"
1311 # bug 11330 - liblustre application death during I/O locks up OST
1313 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1317 # This test assumes relatively small max_dirty_mb setting
1318 # which we want to walk away from, so just for it we will
1319 # temporarily lower the value
1320 local max_dirty_mb=$(lctl get_param -n osc.*.max_dirty_mb | head -n 1)
1321 lctl set_param -n osc.*.max_dirty_mb=32
1322 stack_trap "lctl set_param osc.*.max_dirty_mb=$max_dirty_mb" EXIT
1324 # Minimum pass speed is 2MBps
1326 # LU-2887/LU-3089 - set min pass speed to 500KBps
1327 [ "$ost1_FSTYPE" = zfs ] && ddtimeout=256
1329 # first dd should be finished quickly
1330 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
1331 dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
1334 echo "step1: testing ......"
1335 while kill -0 $DDPID 2> /dev/null; do
1337 if [ $count -gt $ddtimeout ]; then
1338 error "dd should be finished!"
1342 echo "(dd_pid=$DDPID, time=$count)successful"
1344 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2
1345 #define OBD_FAIL_OST_DROP_REQ 0x21d
1346 do_facet ost1 lctl set_param fail_loc=0x0000021d
1347 # second dd will be never finished
1348 dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 &
1351 echo "step2: testing ......"
1352 while [ $count -le $ddtimeout ]; do
1353 if ! kill -0 $DDPID 2> /dev/null; then
1355 error "dd shouldn't be finished! (time=$count)"
1360 echo "(dd_pid=$DDPID, time=$count)successful"
1362 #Recover fail_loc and dd will finish soon
1363 do_facet ost1 lctl set_param fail_loc=0
1365 echo "step3: testing ......"
1366 while kill -0 $DDPID 2> /dev/null; do
1368 if [ $count -gt $((ddtimeout + 440)) ]; then
1369 error "dd should be finished!"
1373 echo "(dd_pid=$DDPID, time=$count)successful"
1377 run_test 55 "ost_brw_read/write drops timed-out read/write request"
1379 test_56() { # b=11277
1380 #define OBD_FAIL_MDS_RESEND 0x136
1382 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
1383 stat $DIR/$tfile || error "stat failed"
1384 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1387 run_test 56 "do not fail on getattr resend"
1390 # no oscs means no client or mdt
1391 while lctl get_param osc.*.* > /dev/null 2>&1; do
1392 : # loop until proc file is removed
1396 test_57() { # bug 10866
1400 #define OBD_FAIL_LPROC_REMOVE 0xB00
1401 lctl set_param fail_loc=0x80000B00
1402 zconf_umount `hostname` $DIR
1403 lctl set_param fail_loc=0x80000B00
1404 fail_abort $SINGLEMDS
1406 lctl set_param fail_loc=0
1408 do_facet client "df $DIR"
1410 run_test 57 "read procfs entries causes kernel crash"
1412 test_58() { # bug 11546
1413 #define OBD_FAIL_MDC_ENQUEUE_PAUSE 0x801
1416 lctl set_param fail_loc=0x80000801
1417 cp $DIR/$tfile /dev/null &
1420 lctl set_param fail_loc=0
1421 drop_bl_callback_once rm -f $DIR/$tfile
1423 # the first 'df' could tigger the eviction caused by
1424 # 'drop_bl_callback_once', and it's normal case.
1425 # but the next 'df' should return successfully.
1426 do_facet client "df $DIR" || do_facet client "df $DIR"
1428 run_test 58 "Eviction in the middle of open RPC reply processing"
1430 test_59() { # bug 10589
1431 zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
1432 echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
1433 #define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311
1434 lctl set_param fail_loc=0x311
1435 writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
1436 [ $? = 0 ] || error "dd write failed"
1437 writes=$(echo $writes | awk -F '+' '/out/ {print $1}')
1438 lctl set_param fail_loc=0
1440 zconf_umount `hostname` $MOUNT2 -f
1441 reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
1442 [ $? = 0 ] || error "dd read failed"
1443 reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
1444 [ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
1446 run_test 59 "Read cancel race on client eviction"
1449 local num_files=${COUNT:-5000}
1450 test_mkdir $DIR/$tdir
1452 # Register (and start) changelog
1453 changelog_register || error "changelog_register failed"
1455 # Generate a large number of changelog entries
1456 createmany -o $DIR/$tdir/$tfile $num_files
1460 # Unlink files in the background
1461 unlinkmany $DIR/$tdir/$tfile $num_files &
1465 # Failover the MDS while unlinks are happening
1466 facet_failover $SINGLEMDS
1468 # Wait for unlinkmany to finish
1471 # Check if all the create/unlink events were recorded
1473 local cl_count=$(changelog_dump | grep -c UNLNK)
1474 echo "$cl_count unlinks in changelog"
1476 changelog_deregister || error "changelog_deregister failed"
1477 if ! changelog_users $SINGLEMDS | grep -q "^cl"; then
1478 [ $cl_count -eq $num_files ] ||
1479 error "Recorded $cl_count of $num_files unlinks"
1480 # Also make sure we can clear large changelogs
1481 cl_count=$(changelog_dump | wc -l)
1482 [ $cl_count -le 2 ] ||
1483 error "Changelog not empty: $cl_count entries"
1484 else # If other users, there may be other unlinks in the log
1485 [ $cl_count -ge $num_files ] ||
1486 error "Recorded $cl_count of $num_files unlinks"
1487 changelog_users $SINGLEMDS
1488 echo "other changelog users; can't verify clear"
1491 run_test 60 "Add Changelog entries during MDS failover"
1495 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0000)
1496 mdtosc=${mdtosc/-MDT*/-MDT\*}
1497 local cflags="osc.$mdtosc.connect_flags"
1498 do_facet $SINGLEMDS "lctl get_param -n $cflags" |grep -q skip_orphan
1499 [ $? -ne 0 ] && skip "don't have skip orphan feature" && return
1501 mkdir_on_mdt0 $DIR/$tdir || error "mkdir dir $DIR/$tdir failed"
1502 # Set the default stripe of $DIR/$tdir to put the files to ost1
1503 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1505 replay_barrier $SINGLEMDS
1506 createmany -o $DIR/$tdir/$tfile-%d 10
1507 local oid=$(do_facet ost1 "lctl get_param -n \
1508 obdfilter.${ost1_svc}.last_id" | sed -e 's/.*://')
1510 fail_abort $SINGLEMDS
1512 touch $DIR/$tdir/$tfile
1513 local id=$($LFS getstripe $DIR/$tdir/$tfile |
1514 awk '$1 == 0 { print $2 }')
1515 [ $id -le $oid ] && error "the orphan objid was reused, failed"
1520 run_test 61 "Verify to not reuse orphan objects - bug 17025"
1522 # test_62 as seen it b2_1 please do not reuse test_62
1525 # zconf_umount `hostname` $DIR
1526 # #define OBD_FAIL_PTLRPC_DELAY_IMP_FULL 0x516
1527 # lctl set_param fail_loc=0x516
1530 #run_test 62 "Verify connection flags race - bug LU-1716"
1535 #grant lock1, export2
1536 $LFS setstripe -i -0 $DIR2/$tfile || error "setstripe failed"
1537 $MULTIOP $DIR2/$tfile Ow || error "multiop failed"
1539 #define OBD_FAIL_LDLM_BL_EVICT 0x31e
1540 do_facet ost $LCTL set_param fail_loc=0x31e
1541 #get waiting lock2, export1
1542 $MULTIOP $DIR/$tfile Ow &
1544 # let enqueue to get asleep
1548 $MULTIOP $DIR2/$tfile Ow &
1556 do_facet ost $LCTL set_param fail_loc=0
1563 run_test 65 "lock enqueue for destroyed export"
1567 [[ "$MDS1_VERSION" -ge $(version_code 2.7.51) ]] ||
1568 skip "Need MDS version at least 2.7.51"
1570 local list=$(comma_list $(osts_nodes))
1572 # modify dir so that next revalidate would not obtain UPDATE lock
1575 # drop 1 reply with UPDATE lock
1576 mcreate $DIR/$tfile || error "mcreate failed: $?"
1577 drop_mdt_ldlm_reply_once "stat $DIR/$tfile" &
1580 # make the re-sent lock to sleep
1581 #define OBD_FAIL_MDS_RESEND 0x136
1582 do_nodes $list $LCTL set_param fail_loc=0x80000136
1584 #initiate the re-connect & re-send
1585 local mdtname="MDT0000"
1586 local mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
1587 local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid)
1588 $LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}"
1591 #initiate the client eviction while enqueue re-send is in progress
1597 run_test 66 "lock enqueue re-send vs client eviction"
1601 #define OBD_FAIL_PTLRPC_CONNECT_RACE 0x531
1602 $LCTL set_param fail_loc=0x80000531
1604 local mdtname="MDT0000"
1605 local mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
1606 local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid)
1607 $LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}" &
1616 run_test 67 "connect vs import invalidate race"
1618 check_cli_ir_state()
1620 local NODE=${1:-$HOSTNAME}
1622 st=$(do_node $NODE "lctl get_param mgc.*.ir_state |
1623 awk '/imperative_recovery:/ { print \\\$2}'")
1624 [ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
1625 error "Error state $st, must be ENABLED or DISABLED"
1629 check_target_ir_state()
1632 local name=${target}_svc
1633 local recovery_proc=obdfilter.${!name}.recovery_status
1637 st=$(do_facet $target "$LCTL get_param -n $recovery_proc |
1638 awk '/status:/{ print \\\$2}'")
1639 [ x$st = xRECOVERING ] || break
1641 st=$(do_facet $target "lctl get_param -n $recovery_proc |
1642 awk '/IR:/{ print \\\$2}'")
1643 [ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
1644 error "Error state $st, must be ENABLED or DISABLED"
1650 do_facet mgs lctl set_param -n mgs.MGS.live.$FSNAME="state=$1"
1655 local state=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1656 awk '/state:/{ print \\\$2 }'")
1660 nidtbl_version_mgs()
1662 local ver=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1663 awk '/nidtbl_version:/{ print \\\$2 }'")
1667 # nidtbl_version_client <mds1|client> [node]
1668 nidtbl_version_client()
1671 local node=${2:-$HOSTNAME}
1673 if [ X$cli = Xclient ]; then
1676 local obdtype=${cli/%[0-9]*/}
1677 [ $obdtype != mds ] && error "wrong parameters $cli"
1679 node=$(facet_active_host $cli)
1684 local vers=$(do_node $node "lctl get_param -n mgc.*.ir_state" |
1685 awk "/$cli/{print \$6}" |sort -u)
1687 # in case there are multiple mounts on the client node
1689 [ ${#arr[@]} -ne 1 ] && error "versions on client node mismatch"
1693 nidtbl_versions_match()
1695 [ $(nidtbl_version_mgs) -eq $(nidtbl_version_client ${1:-client}) ]
1698 target_instance_match()
1704 obdtype=${srv/%[0-9]*/}
1715 error "invalid target type" $srv
1720 local target=${srv}_svc
1721 local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance)
1722 local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import |
1723 awk '/instance/{ print $2 }' | head -n1)
1725 return $([ $si -eq $ci ])
1730 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1731 { skip "MGS without IR support"; return 0; }
1733 # MDT was just restarted in the previous test, make sure everything
1736 while [ $cnt -gt 0 ]; do
1737 nidtbl_versions_match && break
1743 set_ir_status disabled
1745 local prev_ver=$(nidtbl_version_client client)
1747 local saved_FAILURE_MODE=$FAILURE_MODE
1748 [ $(facet_host mgs) = $(facet_host ost1) ] && FAILURE_MODE="SOFT"
1752 [ $(nidtbl_version_client client) -eq $prev_ver ] ||
1753 error "version must not change due to IR disabled"
1754 target_instance_match ost1 || error "instance mismatch"
1758 FAILURE_MODE=$saved_FAILURE_MODE
1760 run_test 100 "IR: Make sure normal recovery still works w/o IR"
1764 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1765 skip "MGS without IR support"
1769 local ost1_imp=$(get_osc_import_name client ost1)
1771 # disable pinger recovery
1772 lctl set_param -n osc.$ost1_imp.pinger_recov=0
1773 stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT
1777 target_instance_match ost1 || error "instance mismatch"
1778 nidtbl_versions_match || error "version must match"
1780 run_test 101a "IR: Make sure IR works w/o normal recovery"
1784 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1785 skip "MGS without IR support"
1789 local ost1_imp=$(get_osc_import_name client ost1)
1791 #define OBD_FAIL_OST_PREPARE_DELAY 0x247
1792 do_facet ost1 "$LCTL set_param fail_loc=0x247"
1793 # disable pinger recovery
1794 $LCTL set_param -n osc.$ost1_imp.pinger_recov=0
1795 stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT
1797 #OST may return EAGAIN if it is not configured yet
1800 run_test 101b "IR: Make sure IR works w/o normal recovery and proceed EAGAIN"
1804 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1805 { skip "MGS without IR support"; return 0; }
1807 local clients=${CLIENTS:-$HOSTNAME}
1814 # let's have a new nidtbl version
1817 # sleep for a while so that clients can see the failure of ost
1818 # it must be MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC.
1819 # int mgc_request.c:
1820 # define MGC_TIMEOUT_MIN_SECONDS 5
1821 # define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 *
1822 local count=30 # 20 seconds at most
1823 while [ $count -gt 0 ]; do
1824 nidtbl_versions_match && break
1829 nidtbl_versions_match || error "nidtbl mismatch"
1832 old_version=$(nidtbl_version_client client)
1834 zconf_umount_clients $clients $MOUNT || error "Cannot umount client"
1837 combined_mgs_mds && mgsdev=mds1
1838 remount_facet $mgsdev
1841 zconf_mount_clients $clients $MOUNT || error "Cannot mount client"
1844 new_version=$(nidtbl_version_client client)
1845 [ $new_version -lt $old_version ] &&
1846 error "nidtbl version wrong after mgs restarts"
1849 run_test 102 "IR: New client gets updated nidtbl after MGS restart"
1853 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1854 { skip "MGS without IR support"; return 0; }
1856 combined_mgs_mds && skip "needs separate mgs and mds" && return 0
1858 # workaround solution to generate config log on the mds
1864 # We need this test because mds is like a client in IR context.
1865 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS ||
1866 error "MDS should start w/o mgs"
1868 # start mgs and remount mds w/ ir
1869 start mgs $(mgsdevname) $MGS_MOUNT_OPTS
1872 # remount client so that fsdb will be created on the MGS
1873 umount_client $MOUNT || error "umount failed"
1874 mount_client $MOUNT || error "mount failed"
1876 # sleep 30 seconds so the MDS has a chance to detect MGS restarting
1878 while [ $count -gt 0 ]; do
1879 [ $(nidtbl_version_client mds1) -ne 0 ] && break
1884 # after a while, mds should be able to reconnect to mgs and fetch
1885 # up-to-date nidtbl version
1886 nidtbl_versions_match mds1 || error "mds nidtbl mismatch"
1891 run_test 103 "IR: MDS can start w/o MGS and get updated nidtbl later"
1895 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1896 { skip "MGS without IR support"; return 0; }
1901 start ost1 $(ostdevname 1) "$OST_MOUNT_OPTS -onoir" ||
1902 error "OST1 cannot start"
1905 local ir_state=$(check_target_ir_state ost1)
1907 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1908 error "ir status on ost1 should be DISABLED"
1910 run_test 104 "IR: ost can disable IR voluntarily"
1914 [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
1915 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1916 { skip "MGS without IR support"; return 0; }
1920 # get one of the clients from client list
1921 local rcli=$(echo $RCLIENTS |cut -d' ' -f 1)
1923 local mount_opts=${MOUNT_OPTS:+$MOUNT_OPTS,}noir
1924 zconf_umount $rcli $MOUNT || error "umount failed"
1925 zconf_mount $rcli $MOUNT $mount_opts || error "mount failed"
1927 # make sure lustre mount at $rcli disabling IR
1928 local ir_state=$(check_cli_ir_state $rcli)
1929 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1930 error "IR state must be DISABLED at $rcli"
1932 # Since the client just mounted, its last_rcvd entry is not on disk.
1933 # Send an RPC so exp_need_sync forces last_rcvd to commit this export
1934 # so the client can reconnect during OST recovery (LU-924, LU-1582)
1935 $LFS setstripe -i 0 $DIR/$tfile
1936 dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=sync
1938 # make sure MGS's state is Partial
1939 [ $(get_ir_status) = "partial" ] || error "MGS IR state must be partial"
1942 # make sure IR on ost1 is DISABLED
1943 local ir_state=$(check_target_ir_state ost1)
1944 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1945 error "IR status on ost1 should be DISABLED"
1947 # remount with the default MOUNT_OPTS
1948 zconf_umount $rcli $MOUNT || error "umount failed"
1949 zconf_mount $rcli $MOUNT || error "mount failed"
1951 # make sure MGS's state is full
1952 [ $(get_ir_status) = "full" ] || error "MGS IR status must be full"
1955 # make sure IR on ost1 is ENABLED
1956 local ir_state=$(check_target_ir_state ost1)
1957 [ $ir_state = "ENABLED" -o $ir_state = "ON" ] ||
1958 error "IR status on ost1 should be ENABLED"
1962 run_test 105 "IR: NON IR clients support"
1970 test_106() { # LU-1789
1971 [[ "$MDS1_VERSION" -ge $(version_code 2.3.50) ]] ||
1972 skip "Need MDS version at least 2.3.50"
1974 #define OBD_FAIL_MDC_LIGHTWEIGHT 0x805
1975 $LCTL set_param fail_loc=0x805
1978 trap cleanup_106 EXIT
1980 # enable lightweight flag on mdc connection
1983 local MDS_NEXP=$(do_facet $SINGLEMDS \
1984 lctl get_param -n mdt.${mds1_svc}.num_exports |
1986 $LCTL set_param fail_loc=0
1988 touch $DIR2/$tfile || error "failed to create empty file"
1989 replay_barrier $SINGLEMDS
1991 $LCTL set_param debug=ha
1993 facet_failover $SINGLEMDS
1995 # lightweight goes through LUSTRE_IMP_RECOVER during failover
1996 touch -c $DIR2/$tfile || true
1997 $LCTL dk $TMP/lustre-log-$TESTNAME.log
1998 recovered=$(awk '/MDT0000-mdc-[0-9a-f]*. lwp recover/ { print }' \
1999 $TMP/lustre-log-$TESTNAME.log)
2000 [ -z "$recovered" ] && error "lightweight client was not recovered"
2002 # and all operations performed by lightweight client should be
2003 # synchronous, so the file created before mds restart should be there
2004 $CHECKSTAT -t file $DIR/$tfile || error "file not present"
2009 run_test 106 "lightweight connection support"
2015 mkdir_on_mdt0 $DIR/$tdir
2016 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2017 do_facet $SINGLEMDS lctl set_param fail_loc=0x119
2018 multiop $DIR/$tdir D_c &
2020 mkdir $DIR/$tdir/dir_106 &
2022 do_facet $SINGLEMDS lctl set_param fail_loc=0
2025 wait $CLIENT_PID || rc=$?
2026 checkstat -t dir $DIR/$tdir/dir_106 || return 1
2028 kill -USR1 $close_pid
2029 wait $close_pid || return 2
2033 run_test 107 "drop reint reply, then restart MDT"
2037 $LFS setstripe -c 1 -i 0 $DIR/$tdir
2039 dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=256 &
2047 client_up || error "reconnect failed"
2048 rm -f $DIR/$tdir/$tfile
2050 run_test 108 "client eviction don't crash"
2053 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2054 local remote_dir=$DIR/$tdir/remote_dir
2058 #prepare for 110 test, which need set striped dir on remote MDT.
2059 for num in $(seq $MDSCOUNT); do
2061 lctl set_param -n mdt.$FSNAME*.enable_remote_dir=1 \
2065 mkdir_on_mdt0 $DIR/$tdir
2066 drop_request "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
2067 error "lfs mkdir failed"
2068 local diridx=$($LFS getstripe -m $remote_dir)
2069 [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
2071 rm -rf $DIR/$tdir || error "rmdir failed"
2073 run_test 110a "create remote directory: drop client req"
2076 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2077 local remote_dir=$DIR/$tdir/remote_dir
2080 mkdir_on_mdt0 $DIR/$tdir
2081 drop_reint_reply "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
2082 error "lfs mkdir failed"
2084 diridx=$($LFS getstripe -m $remote_dir)
2085 [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
2087 rm -rf $DIR/$tdir || error "rmdir failed"
2089 run_test 110b "create remote directory: drop Master rep"
2092 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2093 local remote_dir=$DIR/$tdir/remote_dir
2096 mkdir_on_mdt0 $DIR/$tdir
2097 drop_update_reply $mdtidx "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
2098 error "lfs mkdir failed"
2100 diridx=$($LFS getstripe -m $remote_dir)
2101 [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
2103 rm -rf $DIR/$tdir || error "rmdir failed"
2105 run_test 110c "create remote directory: drop update rep on slave MDT"
2108 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2109 local remote_dir=$DIR/$tdir/remote_dir
2112 mkdir_on_mdt0 $DIR/$tdir
2113 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
2115 drop_request "rm -rf $remote_dir" || error "rm remote dir failed"
2117 rm -rf $DIR/$tdir || error "rmdir failed"
2121 run_test 110d "remove remote directory: drop client req"
2124 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2125 local remote_dir=$DIR/$tdir/remote_dir
2128 mkdir_on_mdt0 $DIR/$tdir
2129 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
2130 drop_reint_reply "rm -rf $remote_dir" || error "rm remote dir failed"
2132 rm -rf $DIR/$tdir || error "rmdir failed"
2136 run_test 110e "remove remote directory: drop master rep"
2139 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2140 local remote_dir=$DIR/$tdir/remote_dir
2143 mkdir_on_mdt0 $DIR/$tdir
2144 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
2145 drop_update_reply $MDTIDX "rm -rf $remote_dir" ||
2146 error "rm remote dir failed"
2148 rm -rf $DIR/$tdir || error "rmdir failed"
2150 run_test 110f "remove remote directory: drop slave rep"
2153 [[ "$MDS1_VERSION" -ge $(version_code 2.11.0) ]] ||
2154 skip "Need MDS version at least 2.11.0"
2156 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
2158 mkdir_on_mdt0 $DIR/$tdir
2159 touch $DIR/$tdir/$tfile
2161 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2162 do_facet mds1 $LCTL set_param fail_loc=0x119
2163 $LFS migrate -m 1 $DIR/$tdir &
2166 do_facet mds1 $LCTL set_param fail_loc=0
2170 mdt_index=$($LFS getstripe -m $DIR/$tdir)
2171 [ $mdt_index == 1 ] || error "$tdir is not on MDT1"
2172 mdt_index=$($LFS getstripe -m $DIR/$tdir/$tfile)
2173 [ $mdt_index == 1 ] || error "$tfile is not on MDT1"
2175 rm -rf $DIR/$tdir || error "rmdir failed"
2177 run_test 110g "drop reply during migration"
2180 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
2181 [[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
2182 skip "Need MDS version at least 2.7.56"
2184 local src_dir=$DIR/$tdir/source_dir
2185 local tgt_dir=$DIR/$tdir/target_dir
2189 $LFS mkdir -i $MDTIDX $tgt_dir
2191 dd if=/etc/hosts of=$src_dir/src_file
2192 touch $tgt_dir/tgt_file
2193 drop_update_reply $MDTIDX \
2194 "mrename $src_dir/src_file $tgt_dir/tgt_file" ||
2195 error "mrename failed"
2197 $CHECKSTAT -t file $src_dir/src_file &&
2198 error "src_file present after rename"
2200 diff /etc/hosts $tgt_dir/tgt_file ||
2201 error "file changed after rename"
2204 run_test 110h "drop update reply during cross-MDT file rename"
2207 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
2208 [[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
2209 skip "Need MDS version at least 2.7.56"
2211 local src_dir=$DIR/$tdir/source_dir
2212 local tgt_dir=$DIR/$tdir/target_dir
2216 $LFS mkdir -i $MDTIDX $tgt_dir
2218 mkdir $src_dir/src_dir
2219 touch $src_dir/src_dir/a
2220 mkdir $tgt_dir/tgt_dir
2221 drop_update_reply $MDTIDX \
2222 "mrename $src_dir/src_dir $tgt_dir/tgt_dir" ||
2223 error "mrename failed"
2225 $CHECKSTAT -t dir $src_dir/src_dir &&
2226 error "src_dir present after rename"
2228 $CHECKSTAT -t dir $tgt_dir/tgt_dir ||
2229 error "tgt_dir not present after rename"
2231 $CHECKSTAT -t file $tgt_dir/tgt_dir/a ||
2232 error "a not present after rename"
2234 run_test 110i "drop update reply during cross-MDT dir rename"
2237 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
2238 [[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
2239 skip "Need MDS version at least 2.7.56"
2241 local remote_dir=$DIR/$tdir/remote_dir
2242 local local_dir=$DIR/$tdir/local_dir
2245 mkdir_on_mdt0 $DIR/$tdir
2246 mkdir $DIR/$tdir/local_dir
2247 $LFS mkdir -i $MDTIDX $remote_dir
2249 touch $local_dir/local_file
2250 drop_update_reply $MDTIDX \
2251 "ln $local_dir/local_file $remote_dir/remote_file" ||
2254 $CHECKSTAT -t file $remote_dir/remote_file ||
2255 error "remote not present after ln"
2257 run_test 110j "drop update reply during cross-MDT ln"
2260 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTS"
2261 [[ "$MDS1_VERSION" -ge $(version_code 2.12.55) ]] ||
2262 skip "Need MDS version at least 2.12.55"
2265 stop mds2 || error "stop mds2 failed"
2267 #define OBD_FAIL_FLD_QUERY_REQ 0x1103
2268 do_facet mds2 lctl set_param fail_loc=0x80001103
2269 local OPTS="$MDS_MOUNT_OPTS -o abort_recovery"
2270 start mds2 $(mdsdevname 2) $OPTS ||
2271 error "start MDS with abort_recovery should succeed"
2272 do_facet mds2 lctl set_param fail_loc=0
2275 stop mds2 || error "cleanup: stop mds2 failed"
2276 start mds2 $(mdsdevname 2) $MDS_MOUNT_OPTS ||
2277 error "cleanup: start mds2 failed"
2278 zconf_mount $(hostname) $MOUNT || error "cleanup: mount failed"
2279 client_up || error "post-failover df failed"
2282 run_test 110k "FID_QUERY failed during recovery"
2285 (( $(lustre_version_code $SINGLEMDS) >= $(version_code 2.14.52) )) ||
2286 skip "Need MDS version at least 2.14.52"
2287 (( $MDSCOUNT >= 2 )) || skip "needs at least 2 MDTs"
2288 local remote_dir=$DIR/$tdir/remote_dir
2293 mkdir_on_mdt0 $DIR/$tdir
2295 #define OBD_FAIL_PTLRPC_RESEND_RACE 0x0525
2296 do_facet mds1 $LCTL set_param fail_loc=0x80000525
2297 $LFS mkdir -i 1 -c2 $remote_dir &
2300 # initiate the re-connect & re-send
2301 mdccli=$(do_facet mds2 $LCTL dl |
2302 awk '/MDT0000-osp-MDT0001/ {print $4;}')
2303 uuid=$(do_facet mds2 $LCTL get_param -n osp.$mdccli.mdt_conn_uuid)
2304 echo "conn_uuid=$uuid"
2305 do_facet mds2 $LCTL set_param "osp.$mdccli.import=connection=$uuid"
2308 (( $? == 0 )) || error "mkdir failed"
2310 diridx=$($LFS getstripe -m $remote_dir)
2311 (( $diridx == 1 )) || error "$diridx != 1"
2312 rm -rf $DIR/$tdir || error "rmdir failed"
2314 run_test 110m "update resent vs original RPC race"
2316 # LU-2844 mdt prepare fail should not cause umount oops
2319 [[ "$MDS1_VERSION" -ge $(version_code 2.3.62) ]] ||
2320 skip "Need MDS version at least 2.3.62"
2322 #define OBD_FAIL_MDS_CHANGELOG_INIT 0x151
2323 do_facet $SINGLEMDS lctl set_param fail_loc=0x151
2324 stop $SINGLEMDS || error "stop MDS failed"
2325 start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS &&
2326 error "start MDS should fail"
2327 do_facet $SINGLEMDS lctl set_param fail_loc=0
2328 start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS ||
2329 error "start MDS failed"
2331 run_test 111 "mdd setup fail should not cause umount oops"
2335 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
2337 do_facet_random_file client $TMP/$tfile 100K ||
2338 error_noexit "Create random file $TMP/$tfile"
2340 pause_bulk "cp $TMP/$tfile $DIR/$tfile" $TIMEOUT ||
2341 error_noexit "Can't pause_bulk copy"
2344 # expect cmp to succeed, client resent bulk
2345 cmp $TMP/$tfile $DIR/$tfile ||
2346 error_noexit "Wrong data has been written"
2348 error_noexit "Can't remove file"
2351 run_test 112a "bulk resend while orignal request is in progress"
2358 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1
2359 cancel_lru_locks osc
2362 $LCTL set_param fail_loc=$fail1 fail_val=3
2363 dd of=/dev/null if=$DIR/$tfile bs=4096 count=1 &
2367 set_nodes_failloc "$(osts_nodes)" $fail2
2369 wait $pid || error "dd failed"
2377 local fail_val2=${4:-0}
2383 $LCTL set_param fail_loc=$fail1 fail_val=4
2384 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 oflag=dsync &
2389 set_nodes_failloc "$(osts_nodes)" $fail2 $fail_val2
2393 [ $error -eq 0 ] && [ $rc -ne 0 ] && error "dd error ($rc)"
2394 [ $error -ne 0 ] && [ $rc -eq 0 ] && error "dd success"
2399 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2400 skip "need at least 2.8.50 on OST"
2402 #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
2403 #define OBD_FAIL_PTLRPC_DROP_BULK 0x51a
2404 test_115_read 0x8000051b 0x8000051a
2406 run_test 115a "read: late REQ MDunlink and no bulk"
2409 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2410 skip "need at least 2.8.50 on OST"
2412 #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
2413 #define OBD_FAIL_OST_ENOSPC 0x215
2415 # pass $OSTCOUNT for the fail_loc to be caught
2416 # appropriately by the IO thread
2417 test_115_write 0x8000051b 0x80000215 1 $OSTCOUNT
2419 run_test 115b "write: late REQ MDunlink and no bulk"
2422 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2423 skip "need at least 2.8.50 on OST"
2425 #define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
2426 #define OBD_FAIL_PTLRPC_DROP_BULK 0x51a
2427 test_115_read 0x8000050f 0x8000051a
2429 run_test 115c "read: late Reply MDunlink and no bulk"
2432 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2433 skip "need at least 2.8.50 on OST"
2435 #define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
2436 #define OBD_FAIL_OST_ENOSPC 0x215
2437 test_115_write 0x8000050f 0x80000215 0
2439 run_test 115d "write: late Reply MDunlink and no bulk"
2442 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2443 skip "need at least 2.8.50 on OST"
2445 #define OBD_FAIL_PTLRPC_LONG_BULK_UNLINK 0x510
2446 #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
2447 test_115_read 0x80000510 0x80000211
2449 run_test 115e "read: late Bulk MDunlink and no reply"
2452 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2453 skip "need at least 2.8.50 on OST"
2455 #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
2456 #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
2457 test_115_read 0x8000051b 0x80000211
2459 run_test 115f "read: late REQ MDunlink and no reply"
2462 [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
2463 skip "need at least 2.8.50 on OST"
2465 #define OBD_FAIL_PTLRPC_LONG_BOTH_UNLINK 0x51c
2466 test_115_read 0x8000051c 0
2468 run_test 115g "read: late REQ MDunlink and Reply MDunlink"
2470 # parameters: fail_loc CMD RC
2477 #define OBD_FAIL_LDLM_CP_CB_WAIT2 0x320
2478 #define OBD_FAIL_LDLM_CP_CB_WAIT3 0x321
2479 #define OBD_FAIL_LDLM_CP_CB_WAIT4 0x322
2480 #define OBD_FAIL_LDLM_CP_CB_WAIT5 0x323
2483 echo -n "** FLOCK REPLY vs. EVICTION race, lock $2"
2484 [ "$1" = "CLEANUP" ] &&
2485 fail=0x80000320 && echo ", $1 cp first"
2486 [ "$1" = "REPLY" ] &&
2487 fail=0x80000321 && echo ", $1 cp first"
2488 [ "$1" = "DEADLOCK CLEANUP" ] &&
2489 fail=0x80000322 && echo " DEADLOCK, CLEANUP cp first"
2490 [ "$1" = "DEADLOCK REPLY" ] &&
2491 fail=0x80000323 && echo " DEADLOCK, REPLY cp first"
2493 if [ x"$2" = x"get" ]; then
2494 #for TEST lock, take a conflict in advance
2495 # sleep longer than evictor to not confuse fail_loc: 2+2+4
2496 echo "** Taking conflict **"
2497 flocks_test 5 set read sleep 10 $DIR/$tfile &
2503 $LCTL set_param fail_loc=$fail
2505 flocks_test 5 $2 write $DIR/$tfile &
2509 echo "** Evicting and re-connecting client **"
2514 if [ x"$2" = x"get" ]; then
2521 # check if the return value is allowed
2522 [ $rc -eq $3 ] && rc=0
2524 $LCTL set_param fail_loc=0
2528 # a lock is taken, unlock vs. cleanup_resource() race for destroying
2529 # the ORIGINAL lock.
2534 flocks_test 5 set write sleep 4 $DIR/$tfile &
2538 # let unlock to sleep in CP CB
2539 $LCTL set_param fail_loc=$1
2542 # let cleanup to cleep in CP CB
2550 $LCTL set_param fail_loc=0
2555 flock_is_enabled || { skip "mount w/o flock enabled" && return; }
2558 test_120_reply "CLEANUP" set 5 || error "SET race failed"
2559 test_120_reply "CLEANUP" get 5 || error "GET race failed"
2560 test_120_reply "CLEANUP" unlock 5 || error "UNLOCK race failed"
2562 test_120_reply "REPLY" set 5 || error "SET race failed"
2563 test_120_reply "REPLY" get 5 || error "GET race failed"
2564 test_120_reply "REPLY" unlock 5 || error "UNLOCK race failed"
2567 test_120_reply "DEADLOCK CLEANUP" set 5 || error "DEADLOCK race failed"
2568 test_120_reply "DEADLOCK REPLY" set 35 || error "DEADLOCK race failed"
2570 test_120_destroy 0x320 || error "unlock-cleanup race failed"
2572 run_test 120 "flock race: completion vs. evict"
2575 local BEFORE=$(date +%s)
2578 # modify dir so that next revalidate would not obtain UPDATE lock
2581 # drop 1 reply with UPDATE lock,
2582 # resend should not create 2nd lock on server
2583 mcreate $DIR/$tfile || error "mcreate failed: $?"
2584 drop_mdt_ldlm_reply_once "stat $DIR/$tfile" || error "stat failed: $?"
2586 # 2 BL AST will be sent to client, both must find the same lock,
2587 # race them to not get EINVAL for 2nd BL AST
2588 #define OBD_FAIL_LDLM_PAUSE_CANCEL2 0x31f
2589 $LCTL set_param fail_loc=0x8000031f
2591 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=0 > /dev/null
2592 chmod 0777 $DIR/$tfile || error "chmod failed: $?"
2593 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=1 > /dev/null
2595 # let the client reconnect
2597 EVICT=$($LCTL get_param mdc.$FSNAME-MDT*.state |
2598 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
2600 [ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
2602 run_test 113 "ldlm enqueue dropped reply should not cause deadlocks"
2606 test_mkdir -p $DIR/$tdir
2608 # Prevent interference from layout intent RPCs due to
2609 # asynchronous writeback. These will be tested in 130c below.
2610 do_nodes ${CLIENTS:-$HOSTNAME} sync
2612 # get only LOOKUP lock on $tdir
2613 cancel_lru_locks mdc
2614 ls $DIR/$tdir/$tfile 2>/dev/null
2616 # get getattr by fid on $tdir
2618 # we need to race with unlink, unlink must complete before we will
2619 # take a DLM lock, otherwise unlink will wait until getattr will
2620 # complete; but later than getattr starts so that getattr found
2622 #define OBD_FAIL_MDS_INTENT_DELAY 0x160
2623 set_nodes_failloc "$(mdts_nodes)" 0x80000160
2630 # drop the reply so that resend happens on an unlinked file.
2631 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2632 set_nodes_failloc "$(mdts_nodes)" 0x80000157
2636 remote_mds_nodsh && skip "remote MDS with nodsh"
2637 [[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
2638 skip "Need server version newer than 2.7.1"
2642 wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
2645 run_test 130a "enqueue resend on not existing file"
2648 remote_mds_nodsh && skip "remote MDS with nodsh"
2649 [[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
2650 skip "Need server version newer than 2.7.1"
2653 # let the reply to be dropped
2656 #define OBD_FAIL_SRV_ENOENT 0x217
2657 set_nodes_failloc "$(mdts_nodes)" 0x80000217
2659 wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
2662 run_test 130b "enqueue resend on a stale inode"
2665 remote_mds_nodsh && skip "remote MDS with nodsh" && return
2667 do_nodes ${CLIENTS:-$HOSTNAME} sync
2668 echo XXX > $DIR/$tfile
2670 cancel_lru_locks mdc
2672 # Trigger writeback on $tfile.
2674 # we need to race with unlink, unlink must complete before we will
2675 # take a DLM lock, otherwise unlink will wait until intent will
2676 # complete; but later than intent starts so that intent found
2678 #define OBD_FAIL_MDS_INTENT_DELAY 0x160
2679 set_nodes_failloc "$(mdts_nodes)" 0x80000160
2686 # drop the reply so that resend happens on an unlinked file.
2687 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2688 set_nodes_failloc "$(mdts_nodes)" 0x80000157
2690 # let the reply to be dropped
2693 #define OBD_FAIL_SRV_ENOENT 0x217
2694 set_nodes_failloc "$(mdts_nodes)" 0x80000217
2700 run_test 130c "layout intent resend on a stale inode"
2703 local before=$(date +%s)
2706 mount_client $MOUNT2 || error "mount filed"
2709 # get a lock on client so that export would reach the stale list
2710 $LFS setstripe -i 0 $DIR/$tfile || error "setstripe failed"
2711 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=fsync ||
2714 #define OBD_FAIL_OST_PAUSE_PUNCH 0x236
2715 do_facet ost1 $LCTL set_param fail_val=120 fail_loc=0x80000236
2717 $TRUNCATE $DIR/$tfile 100 &
2720 dd if=/dev/zero of=$DIR2/$tfile bs=4096 count=1 conv=notrunc ||
2724 umount_client $MOUNT2
2726 evict=$(do_facet client $LCTL get_param \
2727 osc.$FSNAME-OST0000-osc-*/state |
2728 awk -F"[ [,]" '/EVICTED ]$/ { if (t<$5) {t=$5;} } END { print t }')
2730 [ -z "$evict" ] || [[ $evict -le $before ]] ||
2731 (do_facet client $LCTL get_param \
2732 osc.$FSNAME-OST0000-osc-*/state;
2733 error "eviction happened: $evict before:$before")
2735 run_test 132 "long punch"
2738 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
2741 # get a lock on client so that export would reach the stale list
2742 $LFS setstripe -i 0 $DIR/$tfile || error "setstripe failed"
2743 dd if=/dev/zero of=$DIR/$tfile count=1 || error "dd failed"
2745 # another IO under the same lock
2746 #define OBD_FAIL_OSC_DELAY_IO 0x414
2747 $LCTL set_param fail_loc=0x80000414
2748 $LCTL set_param fail_val=4 fail_loc=0x80000414
2749 dd if=/dev/zero of=$DIR/$tfile count=1 conv=notrunc oflag=dsync &
2753 #define OBD_FAIL_LDLM_BL_EVICT 0x31e
2754 set_nodes_failloc "$(osts_nodes)" 0x8000031e
2758 wait $pid && error "dd succeeded"
2761 run_test 131 "IO vs evict results to IO under staled lock"
2764 local list=$(comma_list $(mdts_nodes))
2766 local t=$((TIMEOUT * 2))
2769 flock $DIR/$tfile -c "echo bl lock;sleep $t;echo bl flock unlocked" &
2771 multiop_bg_pause $DIR/$tfile O_jc || return 1
2774 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2775 do_nodes $list $LCTL set_param fail_loc=0x80000157
2777 echo "waiting for multiop $PID"
2778 wait $PID || return 2
2784 run_test 133 "don't fail on flock resend"
2787 [ $CLIENTCOUNT -lt 2 ] &&
2788 { skip "Need 2+ clients, have $CLIENTCOUNT" && return; }
2790 mkdir -p $MOUNT/$tdir/1 $MOUNT/$tdir/2 || error "mkdir failed"
2791 touch $MOUNT/$tdir/1/$tfile $MOUNT/$tdir/2/$tfile ||
2792 error "touch failed"
2793 zconf_umount_clients $CLIENTS $MOUNT
2794 zconf_mount_clients $CLIENTS $MOUNT
2796 #define OBD_FAIL_TGT_REPLY_DATA_RACE 0x722
2797 # assume commit interval is 5
2798 do_facet mds1 "$LCTL set_param fail_loc=0x722 fail_val=5"
2800 local -a clients=(${CLIENTS//,/ })
2801 local client1=${clients[0]}
2802 local client2=${clients[1]}
2804 do_node $client1 rm $MOUNT/$tdir/1/$tfile &
2806 do_node $client2 mv $MOUNT/$tdir/2/$tfile $MOUNT/$tdir/2/${tfile}_2 &
2809 wait $rmpid || error "rm failed"
2810 wait $mvpid || error "mv failed"
2813 run_test 134 "race between failover and search for reply data free slot"
2816 [ "$MDS1_VERSION" -lt $(version_code 2.12.51) ] &&
2817 skip "Need MDS version at least 2.12.51"
2820 $LFS setstripe -E 1M -L mdt $DIR/$tdir
2821 # to have parent dir write lock before open/resend
2822 touch $DIR/$tdir/$tfile
2823 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2824 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x80000157
2825 openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tdir/$tfile ||
2826 error "Failed to open DOM file"
2828 run_test 135 "DOM: open/create resend to return size"
2831 remote_mds_nodsh && skip "remote MDS with nodsh"
2832 [[ "$MDS1_VERSION" -ge $(version_code 2.12.52) ]] ||
2833 skip "Need MDS version at least 2.12.52"
2835 local mdts=$(comma_list $(mdts_nodes))
2836 local MDT0=$(facet_svc $SINGLEMDS)
2838 local clog=$(do_facet mds1 $LCTL --device $MDT0 changelog_register -n)
2839 [ -n "$clog" ] || error "changelog_register failed"
2840 cl_mask=$(do_facet mds1 $LCTL get_param \
2841 mdd.$MDT0.changelog_mask -n)
2842 changelog_chmask "ALL"
2844 # generate some changelog records to accumulate
2845 test_mkdir -i 0 -c 0 $DIR/$tdir || error "mkdir $tdir failed"
2846 createmany -m $DIR/$tdir/$tfile 10000 ||
2847 error "create $DIR/$tdir/$tfile failed"
2849 local size1=$(do_facet $SINGLEMDS \
2850 $LCTL get_param -n mdd.$MDT0.changelog_size)
2851 echo "Changelog size $size1"
2853 #define OBD_FAIL_LLOG_PURGE_DELAY 0x1318
2854 do_nodes $mdts $LCTL set_param fail_loc=0x1318 fail_val=30
2856 # launch changelog_deregister in background on MDS
2857 do_facet mds1 "nohup $LCTL --device $MDT0 changelog_deregister $clog \
2858 > foo.out 2> foo.err < /dev/null &"
2859 # give time to reach fail_loc
2862 # fail_loc will make MDS sleep in the middle of changelog_deregister
2863 # take this opportunity to abruptly kill MDS
2864 FAILURE_MODE_save=$FAILURE_MODE
2867 FAILURE_MODE=$FAILURE_MODE_save
2869 do_nodes $mdts $LCTL set_param fail_loc=0x0 fail_val=0
2871 local size2=$(do_facet $SINGLEMDS \
2872 $LCTL get_param -n mdd.$MDT0.changelog_size)
2873 echo "Changelog size $size2"
2874 local clog2=$(do_facet $SINGLEMDS "$LCTL get_param -n \
2875 mdd.$MDT0.changelog_users | grep $clog")
2876 echo "After crash, changelog user $clog2"
2878 [ -n "$clog2" -o $size2 -lt $size1 ] ||
2879 error "changelog record count unchanged"
2881 do_facet mds1 $LCTL set_param mdd.$MDT0.changelog_mask=\'$cl_mask\' -n
2883 run_test 136 "changelog_deregister leaving pending records"
2887 mkdir_on_mdt0 $DIR/$tdir
2890 dd if=/dev/zero of=$DIR/$tdir/d1/$tfile bs=4096 count=1
2891 dd if=/dev/zero of=$DIR/$tdir/d2/$tfile bs=4096 count=1
2892 cancel_lru_locks osc
2894 #define OBD_FAIL_PTLRPC_RESEND_RACE 0x525
2895 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000525"
2897 # RPC1: any reply is to be delayed to disable last_xid logic
2898 ln $DIR/$tdir/d1/$tfile $DIR/$tdir/d1/f2 &
2901 # RPC2: setattr1 reply is delayed & resent
2902 # original reply comes to client; the resend get asleep
2903 chmod 666 $DIR/$tdir/d2/$tfile
2905 # RPC3: setattr2 on the same file; run ahead of RPC2 resend
2906 chmod 777 $DIR/$tdir/d2/$tfile
2908 # RPC2 resend wakes up
2910 [ $(stat -c "%a" $DIR/$tdir/d2/$tfile) == 777 ] ||
2911 error "resend got applied"
2913 run_test 137 "late resend must be skipped if already applied"
2916 remote_mds_nodsh && skip "remote MDS with nodsh"
2917 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2918 [[ "$MDS1_VERSION" -ge $(version_code 2.12.59) ]] ||
2919 skip "Need server version newer than 2.12.59"
2921 zconf_umount_clients $CLIENTS $MOUNT
2923 #define OBD_FAIL_TGT_RECOVERY_CONNECT 0x724
2924 #delay a first step of recovey when MDS waiting clients
2925 #and failing to get osp logs
2926 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x724 fail_val=5
2928 facet_failover $SINGLEMDS
2930 #waiting failover and recovery timer
2931 #the valuse is based on target_recovery_overseer() wait_event timeout
2933 stop $SINGLEMDS || error "stop MDS failed"
2934 do_facet $SINGLEMDS $LCTL set_param fail_loc=0
2935 start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS ||
2936 error "start MDS failed"
2937 zconf_mount_clients $CLIENTS $MOUNT
2939 run_test 138 "Umount MDT during recovery"
2942 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2943 [ $MDS1_VERSION -lt $(version_code 2.13.50) ] &&
2944 skip "Need MDS version at least 2.13.50"
2946 stop $SINGLEMDS || error "stop $SINGLEMDS failed"
2948 #define OBD_FAIL_OSP_INVALID_LOGID 0x2106
2949 do_facet $SINGLEMDS $LCTL set_param fail_val=0x68 fail_loc=0x80002106
2950 start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) $MDS_MOUNT_OPTS ||
2951 error "Fail to start $SINGLEMDS"
2953 run_test 139 "corrupted catid won't cause crash"
2956 [ $MDS1_VERSION -lt $(version_code 2.12.58) ] &&
2957 skip "Need MDS version at least 2.13.50"
2959 slr=$(do_facet mds1 \
2960 $LCTL get_param -n mdt.$FSNAME-MDT0000.local_recovery)
2961 stack_trap "do_facet mds1 $LCTL set_param \
2962 mdt.*.local_recovery=$slr" EXIT
2964 # disable recovery for local clients
2965 # so local clients should be marked with no_recovery flag
2966 do_facet mds1 $LCTL set_param mdt.*.local_recovery=0
2970 cnt=$(do_facet mds1 $LCTL get_param "mdt.*MDT0000.exports.*.export" |
2971 grep export_flags.*no_recovery | wc -l)
2972 echo "$cnt clients with recovery disabled"
2974 [ $cnt -eq 0 ] && error "no clients with recovery disabled"
2976 # enable recovery for local clients
2977 # so no local clients should be marked with no_recovery flag
2978 do_facet mds1 $LCTL set_param mdt.*.local_recovery=1
2981 cnt=$(do_facet mds1 $LCTL get_param "mdt.*MDT0000.exports.*.export" |
2982 grep export_flags.*no_recovery | wc -l)
2983 echo "$cnt clients with recovery disabled"
2985 [ $cnt -eq 0 ] || error "$cnt clients with recovery disabled"
2987 run_test 140a "local mount is flagged properly"
2990 [ $MDS1_VERSION -lt $(version_code 2.12.58) ] &&
2991 skip "Need MDS version at least 2.13.50"
2993 slr=$(do_facet mds1 \
2994 $LCTL get_param -n mdt.$FSNAME-MDT0000.local_recovery)
2995 stack_trap "do_facet mds1 $LCTL set_param \
2996 mdt.*.local_recovery=$slr" EXIT
2998 # disable recovery for local clients
2999 do_facet mds1 $LCTL set_param mdt.*.local_recovery=0
3005 # Lustre: tfs-MDT0000: Recovery over after 0:03, of 2 clients 2 rec...
3006 local recovery=$(do_facet mds1 dmesg |
3007 awk '/Recovery over after/ { print $6 }' | tail -1 |
3008 awk -F: '{ print $1 * 60 + $2 }')
3009 (( recovery < TIMEOUT * 2 + 5 )) ||
3010 error "recovery took too long $recovery > $((TIMEOUT * 2 + 5))"
3012 run_test 140b "local mount is excluded from recovery"
3018 [ "$PARALLEL" == "yes" ] && skip "skip parallel run"
3019 combined_mgs_mds || skip "needs combined MGS/MDT"
3020 ( local_mode || from_build_tree ) &&
3021 skip "cannot run in local mode or from build tree"
3023 # some get_param have a bug to handle dot in param name
3024 do_rpc_nodes $(facet_active_host $SINGLEMDS) cancel_lru_locks MGC
3025 oldc=$(do_facet $SINGLEMDS $LCTL get_param -n \
3026 'ldlm.namespaces.MGC*.lock_count')
3028 do_rpc_nodes $(facet_active_host $SINGLEMDS) cancel_lru_locks MGC
3029 newc=$(do_facet $SINGLEMDS $LCTL get_param -n \
3030 'ldlm.namespaces.MGC*.lock_count')
3032 [ $oldc -eq $newc ] || error "mgc lost locks ($oldc != $newc)"
3035 run_test 141 "do not lose locks on MGS restart"
3038 [ $MDS1_VERSION -lt $(version_code 2.11.56) ] &&
3039 skip "Need MDS version at least 2.11.56"
3041 #define OBD_FAIL_MDS_ORPHAN_DELETE 0x165
3042 do_facet mds1 $LCTL set_param fail_loc=0x165
3043 $MULTIOP $DIR/$tfile Ouc || error "multiop failed"
3046 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS
3048 wait_update_facet mds1 "pgrep orph_.*-MDD | wc -l" "0" ||
3049 error "MDD orphan cleanup thread not quit"
3051 run_test 142 "orphan name stub can be cleaned up in startup"
3054 (( $MDS1_VERSION >= $(version_code 2.13.0) )) ||
3055 skip "Need MDS version at least 2.13.0"
3056 [ "$PARALLEL" == "yes" ] && skip "skip parallel run"
3058 local mntpt=$(facet_mntpt $SINGLEMDS)
3060 mount_fstype $SINGLEMDS || error "mount as fstype $SINGLEMDS failed"
3061 do_facet $SINGLEMDS touch $mntpt/PENDING/$tfile
3062 unmount_fstype $SINGLEMDS
3063 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "mds1 start fail"
3065 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
3066 wait_update_facet mds1 "pgrep orph_.*-MDD | wc -l" "0" ||
3067 error "MDD orphan cleanup thread not quit"
3069 run_test 143 "orphan cleanup thread shouldn't be blocked even delete failed"
3072 [[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
3073 skip "server does not support overstriping"
3074 (( MDS1_VERSION >= $(version_code 2.15.50.49) )) ||
3075 skip "Need MDS version at least 2.15.50.49"
3084 large_xattr_enabled || skip_env "ea_inode feature disabled"
3085 test_mkdir -i 0 -c 1 -p $DIR/$tdir
3086 stack_trap "rm -rf $DIR/$tdir" EXIT
3088 mds_timeout=$(do_facet mds1 $LCTL get_param -n timeout)
3089 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param timeout=300
3090 stack_trap "do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param timeout=$mds_timeout" EXIT
3092 $LFS setstripe -i 0 -C $setcount $DIR/$tdir || error "setstripe failed"
3094 for (( i = 0; i < 50; i++)); do
3095 touch $DIR/$tdir/$tfile_$i &
3102 for pid in $pids; do
3103 kill -9 $pid >/dev/null 2>&1
3109 # here we measure MDT stop + MDT start time. For error case MDT stop takes
3110 # about obd_timeout-60 (240) seconds. Without error - less than 30s.
3111 # MDT start takes different time depends on a configuration, let's check
3113 diff=$((after - before))
3114 (( $diff < 240 )) || error "MDT failover took $diff seconds"
3116 # failover mds1 back to the primary server
3119 run_test 144a "MDT failover should stop precreation threads"
3122 [[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
3123 skip "server does not support overstriping"
3124 (( MDS1_VERSION >= $(version_code 2.15.51.50) )) ||
3125 skip "Need MDS version at least 2.15.51.50"
3132 large_xattr_enabled || skip_env "ea_inode feature disabled"
3133 test_mkdir -i 0 -c 1 -p $DIR/$tdir
3134 stack_trap "rm -rf $DIR/$tdir" EXIT
3136 mds_timeout=$(do_facet mds1 $LCTL get_param -n timeout)
3137 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param timeout=300
3138 stack_trap "do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param timeout=$mds_timeout" EXIT
3140 $LFS setstripe -i 0 -C $setcount $DIR/$tdir || error "setstripe failed"
3142 for (( i = 0; i < 50; i++)); do
3143 touch $DIR/$tdir/$tfile_$i &
3149 # 60s is not enought with ZFS which is still significatnly
3150 # slower on some operations like record writing
3152 [[ "$mds1_FSTYPE" != "zfs" ]] || stime=120
3155 for pid in $pids; do
3156 ps -p $pid > /dev/null
3157 (( $? == 0 )) && rc=1
3158 kill -9 $pid >/dev/null 2>&1
3161 (( $rc == 0 )) || { fail mds1; error "Create thread still running"; }
3163 run_test 144b "orphan cleanup shouldn't be blocked for no objects+failover situation"
3166 [ $MDSCOUNT -lt 3 ] && skip "needs >= 3 MDTs"
3167 [ $(facet_active_host mds2) = $(facet_active_host mds3) ] &&
3168 skip "needs mds2 and mds3 on separate nodes"
3174 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
3175 echo block mds_connect from mds2
3176 do_facet mds2 "$LCTL set_param fail_loc=0x507"
3178 #define OBD_FAIL_OUT_UPDATE_DROP 0x1707
3179 echo block recovery updates from mds3
3180 do_facet mds3 "$LCTL set_param fail_loc=0x1707"
3182 local hard_timeout=\
3183 $(do_facet mds1 $LCTL get_param -n mdt.$FSNAME-MDT0000.recovery_time_hard)
3187 local get_soft_timeout_cmd=\
3188 "$LCTL get_param -n mdt.$FSNAME-MDT0000.recovery_time_soft 2>/dev/null"
3190 echo wait until mds1 recovery_time_soft is $hard_timeout
3191 wait_update $(facet_host mds1) "$get_soft_timeout_cmd" \
3192 "$hard_timeout" $hard_timeout
3194 echo unblock mds_connect from mds2
3195 do_facet mds2 "$LCTL set_param fail_loc=0"
3197 echo upblock recovery updates from mds3
3198 do_facet mds3 "$LCTL set_param fail_loc=0"
3201 [ -f $DIR/$tfile ] || error "$DIR/$tfile does not exist"
3203 run_test 145 "connect mdtlovs and process update logs after recovery expire"
3206 local prev_count=$(do_facet $SINGLEMDS \
3207 $LCTL get_param -n "mdt.${mds1_svc}.eviction_count")
3213 local next_count=$(do_facet $SINGLEMDS \
3214 $LCTL get_param -n "mdt.${mds1_svc}.eviction_count")
3216 [ "$prev_count" -lt "$next_count" ] ||
3217 error "wrong eviction count ($prev_count >= $next_count)"
3219 run_test 146 "test eviction is counted properly"
3222 local obd_timeout=200
3223 local old=$($LCTL get_param -n timeout)
3225 local connection_count
3227 $LFS setstripe -i 0 -c 1 $f
3228 stripe_index=$($LFS getstripe -i $f)
3229 if [ $stripe_index -ne 0 ]; then
3231 error "$f: stripe_index $stripe_index != 0" && return
3234 $LCTL set_param timeout=$obd_timeout
3235 stack_trap "$LCTL set_param timeout=$old && client_reconnect" EXIT
3237 # OBD_FAIL_OST_CONNECT_NET2
3238 # lost reply to connect request
3239 do_facet ost1 lctl set_param fail_loc=0x00000225 timeout=$obd_timeout
3240 stack_trap "do_facet ost1 $LCTL set_param fail_loc=0 timeout=$old" EXIT
3245 $LFS df $MOUNT > /dev/null 2>&1 &
3246 sleep $((obd_timeout * 3 / 4))
3248 $LCTL get_param osc.$FSNAME-OST0000-osc-*.state
3249 connection_count=$($LCTL get_param osc.$FSNAME-OST0000-osc-*.state |
3250 tac | sed "/FULL/,$ d" | grep CONNECTING | wc -l)
3252 echo $connection_count
3253 (($connection_count >= 6)) || error "Client reconnected too slow"
3255 run_test 147 "Check client reconnect"
3258 local wce_param="obdfilter.$FSNAME-OST0000.writethrough_cache_enable"
3259 local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
3260 local amc=$(at_max_get client)
3261 local amo=$(at_max_get ost1)
3266 timeout=$(request_timeout client)
3268 [ "$(facet_fstype ost1)" = "ldiskfs" ] && {
3269 # save old r/o cache settings
3270 save_lustre_params ost1 $wce_param > $p
3273 do_facet ost1 "$LCTL set_param -n $wce_param=0"
3276 $LFS setstripe -i 0 -c 1 $DIR/$tfile
3277 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 oflag=direct
3278 cp $DIR/$tfile $TMP/$tfile
3279 #define OBD_FAIL_OST_BRW_PAUSE_BULK2 0x227
3280 do_facet ost1 $LCTL set_param fail_loc=0x80000227
3281 do_facet ost1 $LCTL set_param fail_val=$((timeout+2))
3282 dd if=/dev/urandom of=$DIR/$tfile bs=4096 count=1 conv=notrunc,fdatasync
3283 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=notrunc,fdatasync
3285 cancel_lru_locks osc
3286 cmp -b $DIR/$tfile $TMP/$tfile || error "wrong data"
3288 rm -f $DIR/$tfile $TMP/$tfile
3290 at_max_set $amc client
3291 at_max_set $amo ost1
3293 [ "$(facet_fstype ost1)" = "ldiskfs" ] && {
3294 # restore initial r/o cache settings
3295 restore_lustre_params < $p
3300 run_test 148 "data corruption through resend"
3303 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
3305 test_mkdir -i 0 -c $MDSCOUNT $DIR/$tdir || error "mkdir $tdir failed"
3307 # make an orphan striped dir
3308 $MULTIOP $DIR/$tdir D_c &
3311 rmdir $DIR/$tdir || error "can't rmdir"
3313 # stop a slave MDT where one ons stripe is located
3316 # stopping should not cause orphan as another MDT can
3320 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "mds1 start fail"
3321 start mds2 $(mdsdevname 2) $MDS_MOUNT_OPTS || error "mds1 start fail"
3329 run_test 149 "skip orphan removal at umount"
3332 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
3336 lazystatfs=$($LCTL get_param -n llite.$FSNAME-*.lazystatfs | head -1)
3337 max=$($LCTL get_param -n llite.$FSNAME-*.statahead_max | head -1)
3339 $LCTL set_param llite.$FSNAME-*.lazystatfs=1
3340 $LCTL set_param llite.$FSNAME-*.statahead_max=0
3341 stack_trap "$LCTL set_param llite.$FSNAME-*.lazystatfs=$lazystatfs" EXIT
3342 stack_trap "$LCTL set_param llite.$FSNAME-*.statahead_max=$max" EXIT
3343 # stop a slave MDT where one ons stripe is located
3346 stack_trap "start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS && \
3347 wait_recovery_complete mds1 && clients_up && true" EXIT
3349 df $MOUNT || error "statfs failed"
3352 run_test 150 "statfs when MDT0 offline with lazystatfs option"
3355 [[ $($LCTL get_param mdc.*.import) =~ connect_flags.*overstriping ]] ||
3356 skip "server does not support overstriping"
3367 large_xattr_enabled || skip_env "ea_inode feature disabled"
3368 version=$(do_facet mds1 \
3369 uname -r | sed -e "s/\([0-9]*\.[0-9]*\.[0-9]*\).*/\1/")
3370 version=$(version_code ${version//\./ })
3371 if (( $version < $(version_code 4.6.0) )); then
3372 skip "MDS Linux kernel does not support killable semaphore"
3375 test_mkdir -i 0 -c 1 -p $DIR/$tdir
3376 test_mkdir -i 0 -c 1 -p $DIR/$tdir/rr
3377 test_mkdir -i 0 -c 1 -p $DIR/$tdir/qos
3378 stack_trap "rm -rf $DIR/$tdir" EXIT
3380 $LFS setstripe -C $setcount $DIR/$tdir/rr/ || error "setstripe failed"
3383 #define OBD_FAIL_MDS_LOD_CREATE_PAUSE 0x173
3384 #Simulate OST failover and sleep RR allocation under lq_rw_sem
3385 do_facet mds1 $LCTL set_param fail_loc=0x80000173 fail_val=20
3387 for (( i = 0; i < 2; i++)); do
3388 touch $DIR/$tdir/rr/$tfile_$i &
3389 pids_rr="$pids_rr $!"
3393 saved=$(do_facet mds1 $LCTL get_param -n lov.*0000*.qos_threshold_rr)
3394 do_facet mds1 $LCTL set_param lov.*.qos_threshold_rr=0
3395 stack_trap "do_facet mds1 $LCTL set_param lov.*.qos_threshold_rr=$saved" EXIT
3397 #create files with QoS algo, killable semaphore sleeps for 2seconds
3398 for (( i = 0; i < 3; i++)); do
3399 touch $DIR/$tdir/qos/$tfile_$i &
3400 pids_qos="$pids_qos $!"
3403 for pid in $pids_qos; do
3408 diff=$((after - before))
3409 echo "QoS allocation took $diff seconds"
3410 for pid in $pids_rr; do
3415 error "QoS allocation slower than RR, killable semaphore doesn't work"
3417 run_test 152 "QoS object allocation could be awakened in case of OST failover"
3420 #define OBD_FAIL_MDS_CONNECT_VS_EVICT 0x174
3421 do_facet mds1 "$LCTL set_param fail_loc=0x174"
3422 # first drop ping reply from MDS and then
3423 # evict on the subsequent reconnect
3424 # (see target_handle_connect)
3425 sleep $((TIMEOUT + 3))
3427 do_facet mds1 "$LCTL set_param fail_loc=0"
3428 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS ||
3429 error "Fail to start $SINGLEMDS"
3430 wait_recovery_complete mds1 || error "MDS recovery not done"
3432 run_test 153 "evict vs reconnect race"
3435 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
3439 local mnt=$(facet_mntpt mds2)
3442 local ula=($(do_facet mds2 od -t u8 $mnt/update_log | \
3443 awk '{ if($2 > 0) printf "0x%x:0x%x:0x0 ", $3, $2 }'))
3444 [ ${#ula[@]} -lt 2 ] && error "update_log file corruption"
3446 local catfile=$mnt/update_log_dir/\[${ula[0]}\]
3448 echo replace file $catfile
3449 do_facet mds2 dd if=/dev/urandom of=$catfile count=100 ||
3450 error "Write file $catfile error"
3453 start mds2 $(mdsdevname 2) $MDS_MOUNT_OPTS || error "mds2 start fail"
3456 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS || error "mds1 start fail"
3458 echo waiting mds1 recovery....
3459 wait_recovery_complete mds1 20
3461 do_facet mds1 $LCTL get_param mdt.$FSNAME-MDT0000.recovery_status | \
3462 grep -w COMPLETE || error "Recovery was blocked"
3464 run_test 154 "corruption update llog can be skipped"
3466 complete_test $SECONDS
3467 check_and_cleanup_lustre