5 # bug number for skipped test:
6 ALWAYS_EXCEPT="$RECOVERY_SMALL_EXCEPT"
7 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
9 export MULTIOP=${MULTIOP:-multiop}
10 PTLDEBUG=${PTLDEBUG:--1}
11 LUSTRE=${LUSTRE:-`dirname $0`/..}
12 . $LUSTRE/tests/test-framework.sh
14 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
17 require_dsh_mds || exit 0
19 # also long tests: 19, 21a, 21e, 21f, 23, 27
21 [ "$SLOW" = "no" ] && EXCEPT_SLOW=""
25 # Allow us to override the setup if we already have a mounted system by
26 # setting SETUP=" " and CLEANUP=" "
28 CLEANUP=${CLEANUP:-""}
30 check_and_setup_lustre
33 rm -rf $DIR/d[0-9]* $DIR/f.${TESTSUITE}*
36 local f1="$DIR/$tfile"
37 local f2="$DIR/$tfile.2"
39 drop_request "mcreate $f1" ||
40 error_noexit "create '$f1': drop req"
42 drop_reint_reply "mcreate $f2" ||
43 error_noexit "create '$f2': drop rep"
45 drop_request "tchmod 111 $f2" ||
46 error_noexit "chmod '$f2': drop req"
48 drop_reint_reply "tchmod 666 $f2" ||
49 error_noexit "chmod '$f2': drop rep"
51 drop_request "statone $f2" ||
52 error_noexit "stat '$f2': drop req"
54 drop_reply "statone $f2" ||
55 error_noexit "stat '$f2': drop rep"
57 run_test 1 "create, chmod, stat: drop req, drop rep"
61 do_facet_create_file client $t 10K ||
62 error_noexit "Create file $t"
64 drop_request "cat $t > /dev/null" ||
65 error_noexit "Open request for $t file"
67 drop_reply "cat $t > /dev/null" ||
68 error_noexit "Open replay for $t file"
70 run_test 4 "open: drop req, drop rep"
75 local RR="$T-renamed-again"
76 do_facet_create_file client $T 10K ||
77 error_noexit "Create file $T"
79 drop_request "mv $T $R" ||
80 error_noexit "Rename $T"
82 drop_reint_reply "mv $R $RR" ||
83 error_noexit "Failed rename replay on $R"
85 do_facet client "checkstat -v $RR" ||
86 error_noexit "checkstat error on $RR"
88 do_facet client "rm $RR" ||
89 error_noexit "Can't remove file $RR"
91 run_test 5 "rename: drop req, drop rep"
95 local LINK1=$DIR/$tfile.link1
96 local LINK2=$DIR/$tfile.link2
98 do_facet_create_file client $T 10K ||
99 error_noexit "Create file $T"
101 drop_request "mlink $T $LINK1" ||
102 error_noexit "mlink request for $T"
104 drop_reint_reply "mlink $T $LINK2" ||
105 error_noexit "mlink reply for $T"
107 drop_request "munlink $LINK1" ||
108 error_noexit "munlink request for $T"
110 drop_reint_reply "munlink $LINK2" ||
111 error_noexit "munlink reply for $T"
113 do_facet client "rm $T" ||
114 error_noexit "Can't remove file $T"
116 run_test 6 "link, unlink: drop req, drop rep"
120 drop_reint_reply "touch $DIR/$tfile" || return 1
122 run_test 8 "touch: drop rep (bug 1423)"
126 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
130 do_facet_random_file client $TMP/$tfile 1K ||
131 error_noexit "Create random file $TMP/$tfile"
132 # make this big, else test 9 doesn't wait for bulk -- bz 5595
133 do_facet_create_file client $TMP/$t1 4M ||
134 error_noexit "Create file $TMP/$t1"
135 do_facet client "cp $TMP/$t1 $DIR/$t1" ||
136 error_noexit "Can't copy to $DIR/$t1 file"
137 pause_bulk "cp $TMP/$tfile $DIR/$tfile" ||
138 error_noexit "Can't pause_bulk copy"
139 do_facet client "cp $TMP/$t1 $DIR/$t2" ||
140 error_noexit "Can't copy file"
141 do_facet client "sync"
142 do_facet client "rm $DIR/$tfile $DIR/$t2 $DIR/$t1" ||
143 error_noexit "Can't remove files"
144 do_facet client "rm $TMP/$t1 $TMP/$tfile"
146 run_test 9 "pause bulk on OST (bug 1420)"
150 local before=$(date +%s)
153 do_facet client "stat $DIR > /dev/null" ||
154 error "failed to stat $DIR: $?"
155 drop_bl_callback "chmod 0777 $DIR" ||
156 error "failed to chmod $DIR: $?"
158 # let the client reconnect
160 evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
161 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
162 [ ! -z "$evict" ] && [[ $evict -gt $before ]] ||
163 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
164 error "no eviction: $evict before:$before")
166 do_facet client checkstat -v -p 0777 $DIR ||
167 error "client checkstat failed: $?"
169 run_test 10a "finish request on server after client eviction (bug 1521)"
172 local before=$(date +%s)
175 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.53) ]] &&
176 skip "Need MDS version at least 2.6.53" && return
177 do_facet client "stat $DIR > /dev/null" ||
178 error "failed to stat $DIR: $?"
179 drop_bl_callback_once "chmod 0777 $DIR" ||
180 error "failed to chmod $DIR: $?"
182 # let the client reconnect
184 evict=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
185 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
187 [ -z "$evict" ] || [[ $evict -le $before ]] ||
188 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
189 error "eviction happened: $evict before:$before")
191 do_facet client checkstat -v -p 0777 $DIR ||
192 error "client checkstat failed: $?"
194 run_test 10b "re-send BL AST"
197 local before=$(date +%s)
206 workdir="${DIR}/${tdir}"
207 mkdir -p ${workdir} || error "can't create workdir $?"
208 stat ${workdir} > /dev/null ||
209 error "failed to stat ${workdir}: $?"
210 mdtidx=$($LFS getdirstripe -i ${workdir})
211 mdtname=$($LFS mdts ${workdir} | grep -e "^$mdtidx:" |
212 awk '{sub("_UUID", "", $2); print $2;}')
214 mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
215 conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid)
216 mdcpath="mdc.${mdccli}.import=connection=${conn_uuid}"
218 drop_bl_callback_once "chmod 0777 ${workdir}" &
223 # force client reconnect
224 $LCTL set_param "${mdcpath}"
226 # wait client reconnect
230 evict=$($LCTL get_param mdc.${mdccli}.state |
231 awk -F"[ [,]" '/EVICTED]$/ { if (t<$4) {t=$4;} } END { print t }')
233 [[ $evict -le $before ]] ||
234 ( $LCTL get_param mdc.$FSNAME-MDT*.state;
235 error "eviction happened: $EVICT before:$BEFORE" )
237 [ $rc -eq 0 ] || error "chmod must finished OK"
238 checkstat -v -p 0777 "${workdir}" ||
239 error "client checkstat failed: $?"
241 run_test 10c "re-send BL AST vs reconnect race (LU-5569)"
244 local before=$(date +%s)
247 [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ]] &&
248 skip "Need MDS version at least 2.6.90" && return
250 # sleep 1 is to make sure that BEFORE is not equal to EVICTED below
253 echo -n ", world" | dd of=$TMP/$tfile bs=1c seek=5
255 remount_client $MOUNT
259 $LFS setstripe -i 0 -c 1 $DIR1/$tfile
260 echo -n hello | dd of=$DIR1/$tfile bs=5
262 stat $DIR2/$tfile >& /dev/null
263 $LCTL set_param fail_err=71
264 drop_bl_callback "echo -n \\\", world\\\" >> $DIR2/$tfile"
269 cmp -l $DIR1/$tfile $DIR2/$tfile || error "file contents differ"
270 cmp -l $DIR1/$tfile $TMP/$tfile || error "wrong content found"
272 evict=$(do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state | \
274 awk -F"[ [,]" '/EVICTED$/ { if (mx<$1) {mx=$1;} } END { print mx }')
276 [[ $evict -gt $before ]] ||
277 (do_facet client $LCTL get_param osc.$FSNAME-OST0000*.state;
278 error "no eviction: $evict before:$before")
280 $LCTL set_param fail_err=0
282 umount_client $MOUNT2
284 run_test 10d "test failed blocking ast"
288 [[ $(lustre_version_code ost1) -le $(version_code 2.8.58) ]] &&
289 skip "Need OST version at least 2.8.59" && return 0
290 [ $CLIENTCOUNT -lt 2 ] && skip "need two clients" && return 0
291 [ $(facet_host client) == $(facet_host ost1) ] &&
292 skip "need ost1 and client on different nodes" && return 0
293 local -a clients=(${CLIENTS//,/ })
294 local client1=${clients[0]}
295 local client2=${clients[1]}
297 $LFS setstripe -c 1 -i 0 $DIR/$tfile-1 $DIR/$tfile-2
298 $MULTIOP $DIR/$tfile-1 Ow1048576c
300 #define OBD_FAIL_LDLM_BL_CALLBACK_NET 0x305
301 $LCTL set_param fail_loc=0x80000305
303 #define OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT 0x30e
304 do_facet ost1 "$LCTL set_param fail_loc=0x1000030e"
305 # hit OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT twice:
306 # 1. to return ENOTCONN from ldlm_handle_enqueue0
307 # 2. to pause reconnect handling between resend and setting
308 # import to LUSTRE_IMP_FULL state
309 do_facet ost1 "$LCTL set_param fail_val=3"
311 # client1 fails ro respond to bl ast
312 do_node $client2 "$MULTIOP $DIR/$tfile-1 Ow1048576c" &
315 # ost1 returns error on enqueue, which causes client1 to reconnect
316 do_node $client1 "$MULTIOP $DIR/$tfile-2 Ow1048576c" ||
317 error "multiop failed"
320 do_facet ost1 "$LCTL set_param fail_loc=0"
321 do_facet ost1 "$LCTL set_param fail_val=0"
323 run_test 10e "re-send BL AST vs reconnect race 2"
326 # wake up a thread waiting for completion after eviction
328 do_facet client $MULTIOP $DIR/$tfile Ow ||
329 { error "multiop write failed: $?"; return 1; }
330 do_facet client $MULTIOP $DIR/$tfile or ||
331 { error "multiop read failed: $?"; return 2; }
335 do_facet client $MULTIOP $DIR/$tfile or ||
336 { error "multiop read failed: $?"; return 3; }
337 drop_bl_callback_once $MULTIOP $DIR/$tfile Ow ||
338 echo "evicted as expected"
340 do_facet client munlink $DIR/$tfile ||
341 { error "munlink failed: $?"; return 4; }
342 # allow recovery to complete
343 client_up || client_up || sleep $TIMEOUT
345 run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
349 $LCTL mark $MULTIOP $DIR/$tfile OS_c
350 do_facet $SINGLEMDS "lctl set_param fail_loc=0x115"
351 clear_failloc $SINGLEMDS $((TIMEOUT * 2)) &
352 multiop_bg_pause $DIR/$tfile OS_c ||
353 { error "multiop failed: $?"; return 1; }
355 #define OBD_FAIL_MDS_CLOSE_NET 0x115
357 echo "waiting for multiop $PID"
358 wait $PID || { error "wait for multiop faile: $?"; return 2; }
359 do_facet client munlink $DIR/$tfile ||
360 { error "client munlink failed: $?"; return 3; }
361 # allow recovery to complete
362 client_up || client_up || sleep $TIMEOUT
364 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
366 # Bug 113, check that readdir lost recv timeout works.
368 mkdir -p $DIR/$tdir || { error "mkdir failed: $?"; return 1; }
369 touch $DIR/$tdir/newentry || { error "touch failed: $?"; return 2; }
370 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
371 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000104"
372 ls $DIR/$tdir || { error "ls failed: $?"; return 3; }
373 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
374 rm -rf $DIR/$tdir || { error "remove test dir failed: $?"; return 4; }
376 run_test 13 "mdc_readpage restart test (bug 1138)"
378 # Bug 113, check that readdir lost send timeout works.
381 touch $DIR/$tdir/newentry
382 # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
383 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000106"
384 ls $DIR/$tdir || return 1
385 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
387 run_test 14 "mdc_readpage resend test (bug 1138)"
390 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
391 touch $DIR/$tfile && return 1
394 run_test 15 "failed open (-ENOMEM)"
396 READ_AHEAD=`lctl get_param -n llite.*.max_read_ahead_mb | head -n 1`
398 lctl set_param -n llite.*.max_read_ahead_mb 0
402 lctl set_param -n llite.*.max_read_ahead_mb $READ_AHEAD
406 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
408 do_facet_random_file client $TMP/$tfile 100K ||
409 { error_noexit "Create random file $TMP/$T" ; return 0; }
410 do_facet client "cp $TMP/$tfile $DIR/$tfile" ||
411 { error_noexit "Copy to $DIR/$tfile file" ; return 0; }
415 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
416 do_facet ost1 "lctl set_param fail_loc=0x80000504"
418 # OST bulk will time out here, client resends
419 do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 1
420 do_facet ost1 lctl set_param fail_loc=0
421 # give recovery a chance to finish (shouldn't take long)
423 do_facet client "cmp $TMP/$tfile $DIR/$tfile" || return 2
427 run_test 16 "timeout bulk put, don't evict client (2732)"
432 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
434 local SAMPLE_FILE=$TMP/$tfile
435 do_facet_random_file client $SAMPLE_FILE 20K ||
436 { error_noexit "Create random file $SAMPLE_FILE" ; return 0; }
438 # With adaptive timeouts, bulk_get won't expire until adaptive_timeout_max
439 if at_is_enabled; then
440 at_max_saved=$(at_max_get ost1)
441 at_max_set $TIMEOUT ost1
444 # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
445 # OST bulk will time out here, client retries
446 do_facet ost1 lctl set_param fail_loc=0x80000503
447 # need to ensure we send an RPC
448 do_facet client cp $SAMPLE_FILE $DIR/$tfile
451 # with AT, client will wait adaptive_max*factor+net_latency before
452 # expiring the req, hopefully timeout*2 is enough
453 sleep $(($TIMEOUT*2))
455 do_facet ost1 lctl set_param fail_loc=0
456 do_facet client "df $DIR"
457 # expect cmp to succeed, client resent bulk
458 do_facet client "cmp $SAMPLE_FILE $DIR/$tfile" || return 3
459 do_facet client "rm $DIR/$tfile" || return 4
460 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved ost1
463 run_test 17a "timeout bulk get, don't evict client (2732)"
466 [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
468 # get one of the clients from client list
469 local rcli=$(echo $RCLIENTS | cut -d ' ' -f 1)
470 local p="$TMP/$TESTSUITE-$TESTNAME.parameters"
471 local ldlm_enqueue_min=$(do_facet ost1 find /sys -name ldlm_enqueue_min)
472 [ -z "$ldlm_enqueue_min" ] &&
473 skip "missing /sys/.../ldlm_enqueue_min" && return 0
475 $LFS setstripe -i 0 -c 1 -S 1048576 $DIR/$tfile ||
476 error "setstripe failed"
477 $LFS setstripe -i 0 -c 1 -S 1048576 $DIR/${tfile}2 ||
478 error "setstripe 2 failed"
480 save_lustre_params ost1 "at_history" > $p
481 save_lustre_params ost1 "bulk_timeout" >> $p
482 local dev="${FSNAME}-OST0000"
483 save_lustre_params ost1 "obdfilter.$dev.brw_size" >> $p
484 local ldlm_enqueue_min_save=$(do_facet ost1 cat $ldlm_enqueue_min)
486 local new_at_history=15
488 do_facet ost1 "$LCTL set_param at_history=$new_at_history"
489 do_facet ost1 "$LCTL set_param bulk_timeout=30"
490 do_facet ost1 "echo 30 > /sys/module/ptlrpc/parameters/ldlm_enqueue_min"
492 # brw_size is required to be 4m so that bulk transfer timeout
493 # could be simulated with OBD_FAIL_PTLRPC_CLIENT_BULK_CB3
494 local brw_size=$($LCTL get_param -n \
495 osc.$FSNAME-OST0000-osc-[^M]*.import |
496 awk '/max_brw_size/{print $2}')
497 if [ $brw_size -ne 4194304 ]
499 save_lustre_params ost1 "obdfilter.$dev.brw_size" >> $p
500 do_facet ost1 "$LCTL set_param obdfilter.$dev.brw_size=4"
501 remount_client $MOUNT
504 # get service estimate expanded
505 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
506 do_facet ost1 "$LCTL set_param fail_loc=0x80000224 fail_val=35"
507 echo "delay rpc servicing by 35 seconds"
508 dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=fdatasync
511 estimate=$($LCTL get_param -n osc.$dev-osc-*.timeouts |
512 awk '/portal 6/ {print $5}')
513 echo "service estimates increased to $estimate"
515 # let current worst service estimate to get closer to obliteration
516 sleep $((new_at_history / 3))
518 # start i/o and simulate bulk transfer loss
519 #define OBD_FAIL_PTLRPC_CLIENT_BULK_CB3 0x520
520 do_facet ost1 "$LCTL set_param fail_loc=0xa0000520 fail_val=1"
521 dd if=/dev/zero of=$DIR/$tfile bs=2M count=1 conv=fdatasync,notrunc &
524 # start lock conflict handling
525 sleep $((new_at_history / 3))
526 do_node $rcli "dd if=$DIR/$tfile of=/dev/null bs=1M count=1" &
529 # obliterate the worst service estimate
530 sleep $((new_at_history / 3 + 1))
531 dd if=/dev/zero of=$DIR/${tfile}2 bs=1M count=1
533 estimate=$($LCTL get_param -n osc.$dev-osc-*.timeouts |
534 awk '/portal 6/ {print $5}')
535 echo "service estimate dropped to $estimate"
538 [[ $? == 0 ]] || error "write failed"
540 [[ $? == 0 ]] || error "read failed"
542 restore_lustre_params <$p
543 if [ $brw_size -ne 4194304 ]
545 remount_client $MOUNT || error "remount_client failed"
547 do_facet ost1 "echo $ldlm_enqueue_min_save > $ldlm_enqueue_min"
549 run_test 17b "timeout bulk get, dont evict client (3582)"
552 [ -z ${ost2_svc} ] && skip_env "needs 2 osts" && return 0
554 do_facet_create_file client $TMP/$tfile 20K ||
555 { error_noexit "Create file $TMP/$tfile" ; return 0; }
557 do_facet client mkdir -p $DIR/$tdir
561 pgcache_empty || return 1
564 $LFS setstripe -i 1 -c 1 $f
565 stripe_index=$($LFS getstripe -i $f)
566 if [ $stripe_index -ne 1 ]; then
568 error "$f: stripe_index $stripe_index != 1" && return
571 do_facet client cp $TMP/$tfile $f
573 local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | egrep -v 'MDT' | awk '{print $1}'`
574 $LCTL --device $osc2dev deactivate || return 3
575 # my understanding is that there should be nothing in the page
576 # cache after the client reconnects?
578 pgcache_empty || rc=2
579 $LCTL --device $osc2dev activate
583 run_test 18a "manual ost invalidate clears page cache immediately"
586 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
588 do_facet_create_file client $TMP/$tfile 20K ||
589 { error_noexit "Create file $TMP/$tfile" ; return 0; }
591 do_facet client mkdir -p $DIR/$tdir
595 pgcache_empty || return 1
597 $LFS setstripe -i 0 -c 1 $f
598 stripe_index=$($LFS getstripe -i $f)
599 if [ $stripe_index -ne 0 ]; then
601 error "$f: stripe_index $stripe_index != 0" && return
604 do_facet client cp $TMP/$tfile $f
607 # allow recovery to complete
608 sleep $((TIMEOUT + 2))
609 # my understanding is that there should be nothing in the page
610 # cache after the client reconnects?
612 pgcache_empty || rc=2
616 run_test 18b "eviction and reconnect clears page cache (2766)"
619 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
621 do_facet_create_file client $TMP/$tfile 20K ||
622 { error_noexit "Create file $TMP/$tfile" ; return 0; }
624 do_facet client mkdir -p $DIR/$tdir
628 pgcache_empty || return 1
630 $LFS setstripe -i 0 -c 1 $f
631 stripe_index=$($LFS getstripe -i $f)
632 if [ $stripe_index -ne 0 ]; then
634 error "$f: stripe_index $stripe_index != 0" && return
637 do_facet client cp $TMP/$tfile $f
641 # OBD_FAIL_OST_CONNECT_NET2
642 # lost reply to connect request
643 do_facet ost1 lctl set_param fail_loc=0x80000225
646 $LFS df $MOUNT > /dev/null 2>&1
648 # my understanding is that there should be nothing in the page
649 # cache after the client reconnects?
651 pgcache_empty || rc=2
655 run_test 18c "Dropped connect reply after eviction handing (14755)"
658 local BEFORE=`date +%s`
661 mount_client $DIR2 || error "failed to mount $DIR2"
663 # cancel cached locks from OST to avoid eviction from it
666 do_facet client "stat $DIR > /dev/null" ||
667 error "failed to stat $DIR: $?"
668 drop_ldlm_cancel "chmod 0777 $DIR2" ||
669 error "failed to chmod $DIR2"
673 # let the client reconnect
675 EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
676 awk -F"[ [,]" '/EVICTED ]$/ \
677 { if (mx<$5) {mx=$5;} } END { print mx }')
679 [ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
680 (do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state;
681 error "no eviction: $EVICT before:$BEFORE")
683 run_test 19a "test expired_lock_main on mds (2867)"
686 local BEFORE=`date +%s`
689 mount_client $DIR2 || error "failed to mount $DIR2: $?"
691 # cancel cached locks from MDT to avoid eviction from it
694 do_facet client $MULTIOP $DIR/$tfile Ow ||
695 error "failed to run multiop: $?"
696 drop_ldlm_cancel $MULTIOP $DIR2/$tfile Ow ||
697 error "failed to ldlm_cancel: $?"
699 umount_client $DIR2 || error "failed to unmount $DIR2: $?"
700 do_facet client munlink $DIR/$tfile ||
701 error "failed to unlink $DIR/$tfile: $?"
703 # let the client reconnect
705 EVICT=$(do_facet client $LCTL get_param osc.$FSNAME-OST*.state |
706 awk -F"[ [,]" '/EVICTED ]$/ \
707 { if (mx < $5) {mx = $5;} } END { print mx }')
709 [ ! -z "$EVICT" ] && [[ $EVICT -gt $BEFORE ]] ||
710 (do_facet client $LCTL get_param osc.$FSNAME-OST*.state;
711 error "no eviction: $EVICT before:$BEFORE")
713 run_test 19b "test expired_lock_main on ost (2867)"
716 local BEFORE=`date +%s`
719 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=0
721 mkdir -p $DIR1/$tfile
724 #define OBD_FAIL_PTLRPC_CANCEL_RESEND 0x516
725 do_facet mds $LCTL set_param fail_loc=0x80000516
727 touch $DIR2/$tfile/file1 &
729 # let touch to get blocked on the server
733 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=1
736 # let the client reconnect
738 EVICT=$(do_facet client $LCTL get_param mdc.$FSNAME-MDT*.state |
739 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
741 [ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
743 run_test 19c "check reconnect and lock resend do not trigger expired_lock_main"
745 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
746 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
749 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
750 multiop_bg_pause $DIR/$tdir/${tfile} O_wc || return 1
753 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
754 do_facet ost1 lctl set_param fail_loc=0x80000308
755 kill -USR1 $MULTI_PID
758 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
760 run_test 20a "ldlm_handle_enqueue error (should return error)"
762 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
763 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
766 $LFS setstripe -i 0 -c 1 $DIR/$tdir/${tfile}
768 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
769 do_facet ost1 lctl set_param fail_loc=0x80000308
770 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
771 error "didn't fail open enqueue" || true
773 run_test 20b "ldlm_handle_enqueue error (should return error)"
776 mkdir -p $DIR/$tdir-1
777 mkdir -p $DIR/$tdir-2
778 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
781 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
782 $MULTIOP $DIR/$tdir-2/f Oc &
785 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
787 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
788 kill -USR1 $close_pid
790 wait $close_pid || return 1
791 wait $open_pid || return 2
792 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
794 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
795 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
799 run_test 21a "drop close request while close and open are both in flight"
802 mkdir -p $DIR/$tdir-1
803 mkdir -p $DIR/$tdir-2
804 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
807 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
808 mcreate $DIR/$tdir-2/f &
811 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
813 kill -USR1 $close_pid
815 wait $close_pid || return 1
816 wait $open_pid || return 3
818 $CHECKSTAT -t file $DIR/$tdir-1/f || return 4
819 $CHECKSTAT -t file $DIR/$tdir-2/f || return 5
822 run_test 21b "drop open request while close and open are both in flight"
825 mkdir -p $DIR/$tdir-1
826 mkdir -p $DIR/$tdir-2
827 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
830 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
831 mcreate $DIR/$tdir-2/f &
834 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
836 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
837 kill -USR1 $close_pid
839 wait $close_pid || return 1
840 wait $open_pid || return 2
842 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
844 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
845 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
848 run_test 21c "drop both request while close and open are both in flight"
851 mkdir -p $DIR/$tdir-1
852 mkdir -p $DIR/$tdir-2
853 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
856 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000129"
857 $MULTIOP $DIR/$tdir-2/f Oc &
859 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
861 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
864 wait $pid || return 1
865 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
867 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
868 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
872 run_test 21d "drop close reply while close and open are both in flight"
875 mkdir -p $DIR/$tdir-1
876 mkdir -p $DIR/$tdir-2
877 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
880 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
881 touch $DIR/$tdir-2/f &
883 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
887 wait $pid || return 1
890 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
891 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
894 run_test 21e "drop open reply while close and open are both in flight"
897 mkdir -p $DIR/$tdir-1
898 mkdir -p $DIR/$tdir-2
899 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
902 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
903 touch $DIR/$tdir-2/f &
905 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
907 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
910 wait $pid || return 1
911 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
913 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
914 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
917 run_test 21f "drop both reply while close and open are both in flight"
920 mkdir -p $DIR/$tdir-1
921 mkdir -p $DIR/$tdir-2
922 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
925 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
926 touch $DIR/$tdir-2/f &
928 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
930 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
933 wait $pid || return 1
934 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
936 $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
937 $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
940 run_test 21g "drop open reply and close request while close and open are both in flight"
943 mkdir -p $DIR/$tdir-1
944 mkdir -p $DIR/$tdir-2
945 multiop_bg_pause $DIR/$tdir-1/f O_c || return 1
948 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
949 touch $DIR/$tdir-2/f &
952 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
954 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000122"
957 wait $pid || return 1
958 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
960 wait $touch_pid || return 2
962 $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
963 $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
966 run_test 21h "drop open request and close reply while close and open are both in flight"
968 # bug 3462 - multiple MDC requests
973 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
978 $MULTIOP $f1 msu || return 1
981 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
983 wait $close_pid || return 2
984 rm -rf $f2 || return 4
986 run_test 22 "drop close request and do mknod"
989 multiop_bg_pause $DIR/$tfile O_c || return 1
991 # give a chance for open
995 drop_request "kill -USR1 $pid"
998 wait $pid || return 1
1001 run_test 23 "client hang when close a file after mds crash"
1003 test_24a() { # bug 11710 details correct fsync() behavior
1004 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1007 $LFS setstripe -i 0 -c 1 $DIR/$tdir
1008 cancel_lru_locks osc
1009 multiop_bg_pause $DIR/$tdir/$tfile Owy_wyc || return 1
1012 kill -USR1 $MULTI_PID
1015 lctl set_param fail_loc=0x0
1018 error_ignore bz5494 "multiop didn't fail fsync: rc $rc" || true
1020 run_test 24a "fsync error (should return error)"
1022 wait_client_evicted () {
1025 local varsvc=${facet}_svc
1027 wait_update $(facet_active_host $facet) \
1028 "lctl get_param -n *.${!varsvc}.num_exports | cut -d' ' -f2" \
1033 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1035 dmesg -c > /dev/null
1037 lfs setstripe $DIR/$tdir -S 0 -i 0 -c 1 ||
1038 error "$LFS setstripe failed"
1039 cancel_lru_locks osc
1040 multiop_bg_pause $DIR/$tdir/$tfile-1 Ow8192_yc ||
1041 error "mulitop Ow8192_yc failed"
1044 multiop_bg_pause $DIR/$tdir/$tfile-2 Ow8192_c ||
1045 error "mulitop Ow8192_c failed"
1050 kill -USR1 $MULTI_PID1
1053 kill -USR1 $MULTI_PID2
1056 lctl set_param fail_loc=0x0
1058 [ $rc1 -eq 0 -o $rc2 -eq 0 ] &&
1059 error_ignore bz5494 "multiop didn't fail fsync: $rc1 or close: $rc2" ||
1062 dmesg | grep "dirty page discard:" ||
1063 error "no discarded dirty page found!"
1065 run_test 24b "test dirty page discard due to client eviction"
1067 test_26a() { # was test_26 bug 5921 - evict dead exports by pinger
1068 # this test can only run from a client on a separate node.
1069 remote_ost || { skip "local OST" && return 0; }
1070 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1071 remote_mds || { skip "local MDS" && return 0; }
1073 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
1074 skip "msg and ost1 are at the same node"
1078 check_timeout || return 1
1080 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
1081 do_facet client lctl set_param fail_loc=0x505
1082 local before=$(date +%s)
1085 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
1086 # But if there's a race to start the evictor from various obds,
1087 # the loser might have to wait for the next ping.
1088 sleep $((TIMEOUT * 2 + TIMEOUT * 3 / 4))
1089 do_facet client lctl set_param fail_loc=0x0
1090 do_facet client df > /dev/null
1092 local oscs=$(lctl dl | awk '/-osc-/ {print $4}')
1093 check_clients_evicted $before ${oscs[@]}
1094 check_clients_full 10 ${oscs[@]}
1096 run_test 26a "evict dead exports"
1098 test_26b() { # bug 10140 - evict dead exports by pinger
1099 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1101 if [ $(facet_host mgs) = $(facet_host ost1) ]; then
1102 skip "msg and ost1 are at the same node"
1106 check_timeout || return 1
1108 zconf_mount `hostname` $MOUNT2 ||
1109 { error "Failed to mount $MOUNT2"; return 2; }
1110 sleep 1 # wait connections being established
1112 local MDS_NEXP=$(do_facet $SINGLEMDS lctl get_param -n mdt.${mds1_svc}.num_exports | cut -d' ' -f2)
1113 local OST_NEXP=$(do_facet ost1 lctl get_param -n obdfilter.${ost1_svc}.num_exports | cut -d' ' -f2)
1115 echo starting with $OST_NEXP OST and $MDS_NEXP MDS exports
1117 zconf_umount `hostname` $MOUNT2 -f
1119 # PING_INTERVAL max(obd_timeout / 4, 1U)
1120 # PING_EVICT_TIMEOUT (PING_INTERVAL * 6)
1122 # evictor takes PING_EVICT_TIMEOUT + 3 * PING_INTERVAL to evict.
1123 # But if there's a race to start the evictor from various obds,
1124 # the loser might have to wait for the next ping.
1125 # = 9 * PING_INTERVAL + PING_INTERVAL
1126 # = 10 PING_INTERVAL = 10 obd_timeout / 4 = 2.5 obd_timeout
1127 # let's wait $((TIMEOUT * 3)) # bug 19887
1129 wait_client_evicted ost1 $OST_NEXP $((TIMEOUT * 3)) || \
1130 error "Client was not evicted by ost" rc=1
1131 wait_client_evicted $SINGLEMDS $MDS_NEXP $((TIMEOUT * 3)) || \
1132 error "Client was not evicted by mds"
1134 run_test 26b "evict dead exports"
1138 writemany -q -a $DIR/$tdir/$tfile 0 5 &
1141 local save_FAILURE_MODE=$FAILURE_MODE
1143 facet_failover $SINGLEMDS
1144 #define OBD_FAIL_OSC_SHUTDOWN 0x407
1145 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000407
1146 # need to wait for reconnect
1147 echo waiting for fail_loc
1148 wait_update_facet $SINGLEMDS "lctl get_param -n fail_loc" "-2147482617"
1149 facet_failover $SINGLEMDS
1150 #no crashes allowed!
1151 kill -USR1 $CLIENT_PID
1154 FAILURE_MODE=$save_FAILURE_MODE
1156 run_test 27 "fail LOV while using OSC's"
1158 test_28() { # bug 6086 - error adding new clients
1159 do_facet client mcreate $DIR/$tfile || return 1
1160 drop_bl_callback_once "chmod 0777 $DIR/$tfile" ||
1161 echo "evicted as expected"
1162 #define OBD_FAIL_MDS_CLIENT_ADD 0x12f
1163 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012f"
1164 # fail once (evicted), reconnect fail (fail_loc), ok
1165 client_up || (sleep 10; client_up) || (sleep 10; client_up) || error "reconnect failed"
1167 fail $SINGLEMDS # verify MDS last_rcvd can be loaded
1169 run_test 28 "handle error adding new clients (bug 6086)"
1171 test_29a() { # bug 22273 - error adding new clients
1172 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
1173 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000711"
1174 # fail abort so client will be new again
1175 fail_abort $SINGLEMDS
1176 client_up || error "reconnect failed"
1177 wait_osc_import_state $SINGLEMDS ost FULL
1180 run_test 29a "error adding new clients doesn't cause LBUG (bug 22273)"
1182 test_29b() { # bug 22273 - error adding new clients
1183 #define OBD_FAIL_TGT_CLIENT_ADD 0x711
1184 do_facet ost1 "lctl set_param fail_loc=0x80000711"
1185 # fail abort so client will be new again
1187 client_up || error "reconnect failed"
1190 run_test 29b "error adding new clients doesn't cause LBUG (bug 22273)"
1194 # put a load of file creates/writes/deletes
1195 writemany -q $DIR/$tdir/$tfile 0 5 &
1197 echo writemany pid $CLIENT_PID
1201 # wait for client to reconnect to MDS
1206 # client process should see no problems even though MDS went down
1208 kill -USR1 $CLIENT_PID
1211 echo writemany returned $rc
1212 #these may fail because of eviction due to slow AST response.
1214 error_ignore bz13652 "writemany returned rc $rc" || true
1216 run_test 50 "failover MDS under load"
1219 #define OBD_FAIL_MDS_SYNC_CAPA_SL 0x1310
1220 do_facet ost1 lctl set_param fail_loc=0x00001310
1223 # put a load of file creates/writes/deletes
1224 writemany -q $DIR/$tdir/$tfile 0 5 &
1228 facet_failover $SINGLEMDS
1229 # failover at various points during recovery
1230 SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
1231 echo will failover at $SEQ
1234 #echo failover in $i sec
1235 log "test_$testnum: failover in $i sec"
1237 facet_failover $SINGLEMDS
1239 # client process should see no problems even though MDS went down
1240 # and recovery was interrupted
1242 kill -USR1 $CLIENT_PID
1245 echo writemany returned $rc
1247 error_ignore bz13652 "writemany returned rc $rc" || true
1249 run_test 51 "failover MDS during recovery"
1252 do_facet client "mkdir -p $DIR/$tdir"
1253 do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
1255 echo writemany pid $CLIENT_PID
1260 wait $CLIENT_PID || rc=$?
1261 # active client process should see an EIO for down OST
1262 [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
1263 # but timing or failover setup may allow success
1264 [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
1265 echo "writemany returned $rc"
1270 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1275 [ $rc -ne 0 ] && { return $rc; }
1276 # wait for client to reconnect to OST
1280 [ $rc -ne 0 ] && { return $rc; }
1287 run_test 52 "failover OST under load"
1289 # test of open reconstruct
1292 drop_mdt_ldlm_reply "openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tfile" ||\
1295 run_test 53 "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 | grep "went back in time"`
1306 [ x"$ERROR" == x ] || error "back in time occured"
1308 run_test 54 "back in time"
1310 # bug 11330 - liblustre application death during I/O locks up OST
1312 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1316 # This test assumes relatively small max_dirty_mb setting
1317 # which we want to walk away from, so just for it we will
1318 # temporarily lower the value
1319 local max_dirty_mb=$(lctl get_param -n osc.*.max_dirty_mb | head -n 1)
1320 lctl set_param -n osc.*.max_dirty_mb=32
1321 stack_trap "lctl set_param osc.*.max_dirty_mb=$max_dirty_mb" EXIT
1323 # Minimum pass speed is 2MBps
1325 # LU-2887/LU-3089 - set min pass speed to 500KBps
1326 [ "$(facet_fstype ost1)" = "zfs" ] && ddtimeout=256
1328 # first dd should be finished quickly
1329 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
1330 dd if=/dev/zero of=$DIR/$tdir/$tfile-1 bs=32M count=4 &
1333 echo "step1: testing ......"
1334 while kill -0 $DDPID 2> /dev/null; do
1336 if [ $count -gt $ddtimeout ]; then
1337 error "dd should be finished!"
1341 echo "(dd_pid=$DDPID, time=$count)successful"
1343 $LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-2
1344 #define OBD_FAIL_OST_DROP_REQ 0x21d
1345 do_facet ost1 lctl set_param fail_loc=0x0000021d
1346 # second dd will be never finished
1347 dd if=/dev/zero of=$DIR/$tdir/$tfile-2 bs=32M count=4 &
1350 echo "step2: testing ......"
1351 while [ $count -le $ddtimeout ]; do
1352 if ! kill -0 $DDPID 2> /dev/null; then
1354 error "dd shouldn't be finished! (time=$count)"
1359 echo "(dd_pid=$DDPID, time=$count)successful"
1361 #Recover fail_loc and dd will finish soon
1362 do_facet ost1 lctl set_param fail_loc=0
1364 echo "step3: testing ......"
1365 while kill -0 $DDPID 2> /dev/null; do
1367 if [ $count -gt $((ddtimeout + 440)) ]; then
1368 error "dd should be finished!"
1372 echo "(dd_pid=$DDPID, time=$count)successful"
1376 run_test 55 "ost_brw_read/write drops timed-out read/write request"
1378 test_56() { # b=11277
1379 #define OBD_FAIL_MDS_RESEND 0x136
1381 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000136"
1382 stat $DIR/$tfile || error "stat failed"
1383 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1386 run_test 56 "do not fail on getattr resend"
1389 # no oscs means no client or mdt
1390 while lctl get_param osc.*.* > /dev/null 2>&1; do
1391 : # loop until proc file is removed
1395 test_57() { # bug 10866
1399 #define OBD_FAIL_LPROC_REMOVE 0xB00
1400 lctl set_param fail_loc=0x80000B00
1401 zconf_umount `hostname` $DIR
1402 lctl set_param fail_loc=0x80000B00
1403 fail_abort $SINGLEMDS
1405 lctl set_param fail_loc=0
1407 do_facet client "df $DIR"
1409 run_test 57 "read procfs entries causes kernel crash"
1411 test_58() { # bug 11546
1412 #define OBD_FAIL_MDC_ENQUEUE_PAUSE 0x801
1415 lctl set_param fail_loc=0x80000801
1416 cp $DIR/$tfile /dev/null &
1419 lctl set_param fail_loc=0
1420 drop_bl_callback_once rm -f $DIR/$tfile
1422 # the first 'df' could tigger the eviction caused by
1423 # 'drop_bl_callback_once', and it's normal case.
1424 # but the next 'df' should return successfully.
1425 do_facet client "df $DIR" || do_facet client "df $DIR"
1427 run_test 58 "Eviction in the middle of open RPC reply processing"
1429 test_59() { # bug 10589
1430 zconf_mount `hostname` $MOUNT2 || error "Failed to mount $MOUNT2"
1431 echo $DIR2 | grep -q $MOUNT2 || error "DIR2 is not set properly: $DIR2"
1432 #define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311
1433 lctl set_param fail_loc=0x311
1434 writes=$(LANG=C dd if=/dev/zero of=$DIR2/$tfile count=1 2>&1)
1435 [ $? = 0 ] || error "dd write failed"
1436 writes=$(echo $writes | awk -F '+' '/out/ {print $1}')
1437 lctl set_param fail_loc=0
1439 zconf_umount `hostname` $MOUNT2 -f
1440 reads=$(LANG=C dd if=$DIR/$tfile of=/dev/null 2>&1)
1441 [ $? = 0 ] || error "dd read failed"
1442 reads=$(echo $reads | awk -F '+' '/in/ {print $1}')
1443 [ "$reads" -eq "$writes" ] || error "read" $reads "blocks, must be" $writes
1445 run_test 59 "Read cancel race on client eviction"
1448 local num_files=${COUNT:-5000}
1449 test_mkdir $DIR/$tdir
1451 # Register (and start) changelog
1452 changelog_register || error "changelog_register failed"
1454 # Generate a large number of changelog entries
1455 createmany -o $DIR/$tdir/$tfile $num_files
1459 # Unlink files in the background
1460 unlinkmany $DIR/$tdir/$tfile $num_files &
1464 # Failover the MDS while unlinks are happening
1465 facet_failover $SINGLEMDS
1467 # Wait for unlinkmany to finish
1470 # Check if all the create/unlink events were recorded
1472 local cl_count=$(changelog_dump | grep -c UNLNK)
1473 echo "$cl_count unlinks in changelog"
1475 changelog_deregister || error "changelog_deregister failed"
1476 if ! changelog_users $SINGLEMDS | grep -q "^cl"; then
1477 [ $cl_count -eq $num_files ] ||
1478 error "Recorded $cl_count of $num_files unlinks"
1479 # Also make sure we can clear large changelogs
1480 cl_count=$(changelog_dump | wc -l)
1481 [ $cl_count -le 2 ] ||
1482 error "Changelog not empty: $cl_count entries"
1483 else # If other users, there may be other unlinks in the log
1484 [ $cl_count -ge $num_files ] ||
1485 error "Recorded $cl_count of $num_files unlinks"
1486 changelog_users $SINGLEMDS
1487 echo "other changelog users; can't verify clear"
1490 run_test 60 "Add Changelog entries during MDS failover"
1494 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $FSNAME-OST0000)
1495 mdtosc=${mdtosc/-MDT*/-MDT\*}
1496 local cflags="osc.$mdtosc.connect_flags"
1497 do_facet $SINGLEMDS "lctl get_param -n $cflags" |grep -q skip_orphan
1498 [ $? -ne 0 ] && skip "don't have skip orphan feature" && return
1500 mkdir -p $DIR/$tdir || error "mkdir dir $DIR/$tdir failed"
1501 # Set the default stripe of $DIR/$tdir to put the files to ost1
1502 $LFS setstripe -c 1 -i 0 $DIR/$tdir
1504 replay_barrier $SINGLEMDS
1505 createmany -o $DIR/$tdir/$tfile-%d 10
1506 local oid=$(do_facet ost1 "lctl get_param -n \
1507 obdfilter.${ost1_svc}.last_id" | sed -e 's/.*://')
1509 fail_abort $SINGLEMDS
1511 touch $DIR/$tdir/$tfile
1512 local id=$($LFS getstripe $DIR/$tdir/$tfile |
1513 awk '$1 == 0 { print $2 }')
1514 [ $id -le $oid ] && error "the orphan objid was reused, failed"
1519 run_test 61 "Verify to not reuse orphan objects - bug 17025"
1521 # test_62 as seen it b2_1 please do not reuse test_62
1524 # zconf_umount `hostname` $DIR
1525 # #define OBD_FAIL_PTLRPC_DELAY_IMP_FULL 0x516
1526 # lctl set_param fail_loc=0x516
1529 #run_test 62 "Verify connection flags race - bug LU-1716"
1534 #grant lock1, export2
1535 $SETSTRIPE -i -0 $DIR2/$tfile || return 1
1536 $MULTIOP $DIR2/$tfile Ow || return 2
1538 #define OBD_FAIL_LDLM_BL_EVICT 0x31e
1539 do_facet ost $LCTL set_param fail_loc=0x31e
1540 #get waiting lock2, export1
1541 $MULTIOP $DIR/$tfile Ow &
1543 # let enqueue to get asleep
1547 $MULTIOP $DIR2/$tfile Ow &
1555 do_facet ost $LCTL set_param fail_loc=0
1562 run_test 65 "lock enqueue for destroyed export"
1566 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.51) ]] ||
1567 { skip "Need MDS version at least 2.7.51"; return 0; }
1569 local list=$(comma_list $(osts_nodes))
1571 # modify dir so that next revalidate would not obtain UPDATE lock
1574 # drop 1 reply with UPDATE lock
1575 mcreate $DIR/$tfile || error "mcreate failed: $?"
1576 drop_mdt_ldlm_reply_once "stat $DIR/$tfile" &
1579 # make the re-sent lock to sleep
1580 #define OBD_FAIL_MDS_RESEND 0x136
1581 do_nodes $list $LCTL set_param fail_loc=0x80000136
1583 #initiate the re-connect & re-send
1584 local mdccli=$($LCTL dl | awk '/-MDT0000-mdc-/ {print $4;}')
1585 local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid)
1586 $LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}"
1589 #initiate the client eviction while enqueue re-send is in progress
1595 run_test 66 "lock enqueue re-send vs client eviction"
1599 #define OBD_FAIL_PTLRPC_CONNECT_RACE 0x531
1600 $LCTL set_param fail_loc=0x80000531
1602 local mdtname="MDT0000"
1603 local mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
1604 local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.mds_conn_uuid)
1605 $LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}" &
1614 run_test 67 "connect vs import invalidate race"
1616 check_cli_ir_state()
1618 local NODE=${1:-$HOSTNAME}
1620 st=$(do_node $NODE "lctl get_param mgc.*.ir_state |
1621 awk '/imperative_recovery:/ { print \\\$2}'")
1622 [ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
1623 error "Error state $st, must be ENABLED or DISABLED"
1627 check_target_ir_state()
1630 local name=${target}_svc
1631 local recovery_proc=obdfilter.${!name}.recovery_status
1635 st=$(do_facet $target "$LCTL get_param -n $recovery_proc |
1636 awk '/status:/{ print \\\$2}'")
1637 [ x$st = xRECOVERING ] || break
1639 st=$(do_facet $target "lctl get_param -n $recovery_proc |
1640 awk '/IR:/{ print \\\$2}'")
1641 [ $st != ON -o $st != OFF -o $st != ENABLED -o $st != DISABLED ] ||
1642 error "Error state $st, must be ENABLED or DISABLED"
1648 do_facet mgs lctl set_param -n mgs.MGS.live.$FSNAME="state=$1"
1653 local state=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1654 awk '/state:/{ print \\\$2 }'")
1658 nidtbl_version_mgs()
1660 local ver=$(do_facet mgs "lctl get_param -n mgs.MGS.live.$FSNAME |
1661 awk '/nidtbl_version:/{ print \\\$2 }'")
1665 # nidtbl_version_client <mds1|client> [node]
1666 nidtbl_version_client()
1669 local node=${2:-$HOSTNAME}
1671 if [ X$cli = Xclient ]; then
1674 local obdtype=${cli/%[0-9]*/}
1675 [ $obdtype != mds ] && error "wrong parameters $cli"
1677 node=$(facet_active_host $cli)
1682 local vers=$(do_node $node "lctl get_param -n mgc.*.ir_state" |
1683 awk "/$cli/{print \$6}" |sort -u)
1685 # in case there are multiple mounts on the client node
1687 [ ${#arr[@]} -ne 1 ] && error "versions on client node mismatch"
1691 nidtbl_versions_match()
1693 [ $(nidtbl_version_mgs) -eq $(nidtbl_version_client ${1:-client}) ]
1696 target_instance_match()
1702 obdtype=${srv/%[0-9]*/}
1713 error "invalid target type" $srv
1718 local target=${srv}_svc
1719 local si=$(do_facet $srv lctl get_param -n $obdname.${!target}.instance)
1720 local ci=$(lctl get_param -n $cliname.${!target}-${cliname}-*.import |
1721 awk '/instance/{ print $2 }' | head -n1)
1723 return $([ $si -eq $ci ])
1728 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1729 { skip "MGS without IR support"; return 0; }
1731 # MDT was just restarted in the previous test, make sure everything
1734 while [ $cnt -gt 0 ]; do
1735 nidtbl_versions_match && break
1741 set_ir_status disabled
1743 local prev_ver=$(nidtbl_version_client client)
1745 local saved_FAILURE_MODE=$FAILURE_MODE
1746 [ $(facet_host mgs) = $(facet_host ost1) ] && FAILURE_MODE="SOFT"
1750 [ $(nidtbl_version_client client) -eq $prev_ver ] ||
1751 error "version must not change due to IR disabled"
1752 target_instance_match ost1 || error "instance mismatch"
1756 FAILURE_MODE=$saved_FAILURE_MODE
1758 run_test 100 "IR: Make sure normal recovery still works w/o IR"
1762 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1763 skip "MGS without IR support"
1767 local ost1_imp=$(get_osc_import_name client ost1)
1769 # disable pinger recovery
1770 lctl set_param -n osc.$ost1_imp.pinger_recov=0
1771 stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT
1775 target_instance_match ost1 || error "instance mismatch"
1776 nidtbl_versions_match || error "version must match"
1778 run_test 101a "IR: Make sure IR works w/o normal recovery"
1782 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1783 skip "MGS without IR support"
1787 local ost1_imp=$(get_osc_import_name client ost1)
1789 #define OBD_FAIL_OST_PREPARE_DELAY 0x247
1790 do_facet ost1 "$LCTL set_param fail_loc=0x247"
1791 # disable pinger recovery
1792 $LCTL set_param -n osc.$ost1_imp.pinger_recov=0
1793 stack_trap "$LCTL set_param -n osc.$ost1_imp.pinger_recov=1" EXIT
1795 #OST may return EAGAIN if it is not configured yet
1798 run_test 101b "IR: Make sure IR works w/o normal recovery and proceed EAGAIN"
1802 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1803 { skip "MGS without IR support"; return 0; }
1805 local clients=${CLIENTS:-$HOSTNAME}
1812 # let's have a new nidtbl version
1815 # sleep for a while so that clients can see the failure of ost
1816 # it must be MGC_TIMEOUT_MIN_SECONDS + MGC_TIMEOUT_RAND_CENTISEC.
1817 # int mgc_request.c:
1818 # define MGC_TIMEOUT_MIN_SECONDS 5
1819 # define MGC_TIMEOUT_RAND_CENTISEC 0x1ff /* ~500 *
1820 local count=30 # 20 seconds at most
1821 while [ $count -gt 0 ]; do
1822 nidtbl_versions_match && break
1827 nidtbl_versions_match || error "nidtbl mismatch"
1830 old_version=$(nidtbl_version_client client)
1832 zconf_umount_clients $clients $MOUNT || error "Cannot umount client"
1835 combined_mgs_mds && mgsdev=mds1
1836 remount_facet $mgsdev
1839 zconf_mount_clients $clients $MOUNT || error "Cannot mount client"
1842 new_version=$(nidtbl_version_client client)
1843 [ $new_version -lt $old_version ] &&
1844 error "nidtbl version wrong after mgs restarts"
1847 run_test 102 "IR: New client gets updated nidtbl after MGS restart"
1851 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1852 { skip "MGS without IR support"; return 0; }
1854 combined_mgs_mds && skip "needs separate mgs and mds" && return 0
1856 # workaround solution to generate config log on the mds
1862 # We need this test because mds is like a client in IR context.
1863 start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS ||
1864 error "MDS should start w/o mgs"
1866 # start mgs and remount mds w/ ir
1867 start mgs $(mgsdevname) $MGS_MOUNT_OPTS
1870 # remount client so that fsdb will be created on the MGS
1871 umount_client $MOUNT || error "umount failed"
1872 mount_client $MOUNT || error "mount failed"
1874 # sleep 30 seconds so the MDS has a chance to detect MGS restarting
1876 while [ $count -gt 0 ]; do
1877 [ $(nidtbl_version_client mds1) -ne 0 ] && break
1882 # after a while, mds should be able to reconnect to mgs and fetch
1883 # up-to-date nidtbl version
1884 nidtbl_versions_match mds1 || error "mds nidtbl mismatch"
1889 run_test 103 "IR: MDS can start w/o MGS and get updated nidtbl later"
1893 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1894 { skip "MGS without IR support"; return 0; }
1899 start ost1 $(ostdevname 1) "$OST_MOUNT_OPTS -onoir" ||
1900 error "OST1 cannot start"
1903 local ir_state=$(check_target_ir_state ost1)
1904 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1905 error "ir status on ost1 should be DISABLED"
1907 run_test 104 "IR: ost can disable IR voluntarily"
1911 [ -z "$RCLIENTS" ] && skip "Needs multiple clients" && return 0
1912 do_facet mgs $LCTL list_param mgs.*.ir_timeout ||
1913 { skip "MGS without IR support"; return 0; }
1917 # get one of the clients from client list
1918 local rcli=$(echo $RCLIENTS |cut -d' ' -f 1)
1920 local mount_opts=${MOUNT_OPTS:+$MOUNT_OPTS,}noir
1921 zconf_umount $rcli $MOUNT || error "umount failed"
1922 zconf_mount $rcli $MOUNT $mount_opts || error "mount failed"
1924 # make sure lustre mount at $rcli disabling IR
1925 local ir_state=$(check_cli_ir_state $rcli)
1926 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1927 error "IR state must be DISABLED at $rcli"
1929 # Since the client just mounted, its last_rcvd entry is not on disk.
1930 # Send an RPC so exp_need_sync forces last_rcvd to commit this export
1931 # so the client can reconnect during OST recovery (LU-924, LU-1582)
1932 $SETSTRIPE -i 0 $DIR/$tfile
1933 dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=sync
1935 # make sure MGS's state is Partial
1936 [ $(get_ir_status) = "partial" ] || error "MGS IR state must be partial"
1939 # make sure IR on ost1 is DISABLED
1940 local ir_state=$(check_target_ir_state ost1)
1941 [ $ir_state = "DISABLED" -o $ir_state = "OFF" ] ||
1942 error "IR status on ost1 should be DISABLED"
1944 # remount with the default MOUNT_OPTS
1945 zconf_umount $rcli $MOUNT || error "umount failed"
1946 zconf_mount $rcli $MOUNT || error "mount failed"
1948 # make sure MGS's state is full
1949 [ $(get_ir_status) = "full" ] || error "MGS IR status must be full"
1952 # make sure IR on ost1 is ENABLED
1953 local ir_state=$(check_target_ir_state ost1)
1954 [ $ir_state = "ENABLED" -o $ir_state = "ON" ] ||
1955 error "IR status on ost1 should be ENABLED"
1959 run_test 105 "IR: NON IR clients support"
1967 test_106() { # LU-1789
1968 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.50) ]] ||
1969 { skip "Need MDS version at least 2.3.50"; return 0; }
1971 #define OBD_FAIL_MDC_LIGHTWEIGHT 0x805
1972 $LCTL set_param fail_loc=0x805
1975 trap cleanup_106 EXIT
1977 # enable lightweight flag on mdc connection
1980 local MDS_NEXP=$(do_facet $SINGLEMDS \
1981 lctl get_param -n mdt.${mds1_svc}.num_exports |
1983 $LCTL set_param fail_loc=0
1985 touch $DIR2/$tfile || error "failed to create empty file"
1986 replay_barrier $SINGLEMDS
1988 $LCTL set_param debug=ha
1990 facet_failover $SINGLEMDS
1992 # lightweight goes through LUSTRE_IMP_RECOVER during failover
1993 touch -c $DIR2/$tfile || true
1994 $LCTL dk $TMP/lustre-log-$TESTNAME.log
1995 recovered=`awk '/MDT0000-mdc-[0-9a-f]*: lwp recover/ {
1997 }' $TMP/lustre-log-$TESTNAME.log`
1998 [ -z "$recovered" ] && error "lightweight client was not recovered"
2000 # and all operations performed by lightweight client should be
2001 # synchronous, so the file created before mds restart should be there
2002 $CHECKSTAT -t file $DIR/$tfile || error "file not present"
2007 run_test 106 "lightweight connection support"
2014 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2015 do_facet $SINGLEMDS lctl set_param fail_loc=0x119
2016 multiop $DIR/$tdir D_c &
2018 mkdir $DIR/$tdir/dir_106 &
2020 do_facet $SINGLEMDS lctl set_param fail_loc=0
2023 wait $CLIENT_PID || rc=$?
2024 checkstat -t dir $DIR/$tdir/dir_106 || return 1
2026 kill -USR1 $close_pid
2027 wait $close_pid || return 2
2031 run_test 107 "drop reint reply, then restart MDT"
2035 $SETSTRIPE -c 1 -i 0 $DIR/$tdir
2037 dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=256 &
2045 client_up || error "reconnect failed"
2046 rm -f $DIR/$tdir/$tfile
2048 run_test 108 "client eviction don't crash"
2051 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2052 local remote_dir=$DIR/$tdir/remote_dir
2056 #prepare for 110 test, which need set striped dir on remote MDT.
2057 for num in $(seq $MDSCOUNT); do
2059 lctl set_param -n mdt.$FSNAME*.enable_remote_dir=1 \
2064 drop_request "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
2065 error "lfs mkdir failed"
2066 local diridx=$($LFS getstripe -m $remote_dir)
2067 [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
2069 rm -rf $DIR/$tdir || error "rmdir failed"
2071 run_test 110a "create remote directory: drop client req"
2074 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2075 local remote_dir=$DIR/$tdir/remote_dir
2079 drop_reint_reply "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
2080 error "lfs mkdir failed"
2082 diridx=$($LFS getstripe -m $remote_dir)
2083 [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
2085 rm -rf $DIR/$tdir || error "rmdir failed"
2087 run_test 110b "create remote directory: drop Master rep"
2090 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2091 local remote_dir=$DIR/$tdir/remote_dir
2095 drop_update_reply $mdtidx "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
2096 error "lfs mkdir failed"
2098 diridx=$($GETSTRIPE -m $remote_dir)
2099 [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
2101 rm -rf $DIR/$tdir || error "rmdir failed"
2103 run_test 110c "create remote directory: drop update rep on slave MDT"
2106 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2107 local remote_dir=$DIR/$tdir/remote_dir
2111 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
2113 drop_request "rm -rf $remote_dir" || error "rm remote dir failed"
2115 rm -rf $DIR/$tdir || error "rmdir failed"
2119 run_test 110d "remove remote directory: drop client req"
2122 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2123 local remote_dir=$DIR/$tdir/remote_dir
2127 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
2128 drop_reint_reply "rm -rf $remote_dir" || error "rm remote dir failed"
2130 rm -rf $DIR/$tdir || error "rmdir failed"
2134 run_test 110e "remove remote directory: drop master rep"
2137 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2138 local remote_dir=$DIR/$tdir/remote_dir
2142 $LFS mkdir -i $MDTIDX -c2 $remote_dir || error "lfs mkdir failed"
2143 drop_update_reply $MDTIDX "rm -rf $remote_dir" ||
2144 error "rm remote dir failed"
2146 rm -rf $DIR/$tdir || error "rmdir failed"
2148 run_test 110f "remove remote directory: drop slave rep"
2151 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.11.0) ]] ||
2152 { skip "Need MDS version at least 2.11.0"; return 0; }
2154 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2157 touch $DIR/$tdir/$tfile
2159 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2160 do_facet mds1 $LCTL set_param fail_loc=0x119
2161 $LFS migrate -m 1 $DIR/$tdir &
2164 do_facet mds1 $LCTL set_param fail_loc=0
2168 mdt_index=$($LFS getstripe -m $DIR/$tdir)
2169 [ $mdt_index == 1 ] || error "$tdir is not on MDT1"
2170 mdt_index=$($LFS getstripe -m $DIR/$tdir/$tfile)
2171 [ $mdt_index == 1 ] || error "$tfile is not on MDT1"
2173 rm -rf $DIR/$tdir || error "rmdir failed"
2175 run_test 110g "drop reply during migration"
2178 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2179 local server_version=$(lustre_version_code mds1)
2180 [[ $server_version -ge $(version_code 2.7.56) ]] ||
2181 { skip "Need MDS version at least 2.7.56"; return 0; }
2183 local src_dir=$DIR/$tdir/source_dir
2184 local tgt_dir=$DIR/$tdir/target_dir
2188 $LFS mkdir -i $MDTIDX $tgt_dir
2190 dd if=/etc/hosts of=$src_dir/src_file
2191 touch $tgt_dir/tgt_file
2192 drop_update_reply $MDTIDX \
2193 "mrename $src_dir/src_file $tgt_dir/tgt_file" ||
2194 error "mrename failed"
2196 $CHECKSTAT -t file $src_dir/src_file &&
2197 error "src_file present after rename"
2199 diff /etc/hosts $tgt_dir/tgt_file ||
2200 error "file changed after rename"
2203 run_test 110h "drop update reply during cross-MDT file rename"
2206 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2207 local server_version=$(lustre_version_code mds1)
2208 [[ $server_version -ge $(version_code 2.7.56) ]] ||
2209 { skip "Need MDS version at least 2.7.56"; return 0; }
2211 local src_dir=$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 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2238 local server_version=$(lustre_version_code mds1)
2239 [[ $server_version -ge $(version_code 2.7.56) ]] ||
2240 { skip "Need MDS version at least 2.7.56"; return 0; }
2242 local remote_dir=$DIR/$tdir/remote_dir
2243 local local_dir=$DIR/$tdir/local_dir
2247 mkdir $DIR/$tdir/local_dir
2248 $LFS mkdir -i $MDTIDX $remote_dir
2250 touch $local_dir/local_file
2251 drop_update_reply $MDTIDX \
2252 "ln $local_dir/local_file $remote_dir/remote_file" ||
2255 $CHECKSTAT -t file $remote_dir/remote_file ||
2256 error "remote not present after ln"
2258 run_test 110j "drop update reply during cross-MDT ln"
2260 # LU-2844 mdt prepare fail should not cause umount oops
2263 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.62) ]] ||
2264 { skip "Need MDS version at least 2.3.62"; return 0; }
2266 #define OBD_FAIL_MDS_CHANGELOG_INIT 0x151
2267 do_facet $SINGLEMDS lctl set_param fail_loc=0x151
2268 stop $SINGLEMDS || error "stop MDS failed"
2269 start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) &&
2270 error "start MDS should fail"
2271 do_facet $SINGLEMDS lctl set_param fail_loc=0
2272 start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) ||
2273 error "start MDS failed"
2275 run_test 111 "mdd setup fail should not cause umount oops"
2279 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
2281 do_facet_random_file client $TMP/$tfile 100K ||
2282 error_noexit "Create random file $TMP/$tfile"
2284 pause_bulk "cp $TMP/$tfile $DIR/$tfile" $TIMEOUT ||
2285 error_noexit "Can't pause_bulk copy"
2288 # expect cmp to succeed, client resent bulk
2289 cmp $TMP/$tfile $DIR/$tfile ||
2290 error_noexit "Wrong data has been written"
2292 error_noexit "Can't remove file"
2295 run_test 112a "bulk resend while orignal request is in progress"
2302 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1
2303 cancel_lru_locks osc
2306 $LCTL set_param fail_loc=$fail1 fail_val=3
2307 dd of=/dev/null if=$DIR/$tfile bs=4096 count=1 &
2311 set_nodes_failloc "$(osts_nodes)" $fail2
2313 wait $pid || error "dd failed"
2321 local fail_val2=${4:-0}
2327 $LCTL set_param fail_loc=$fail1 fail_val=4
2328 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 oflag=dsync &
2333 set_nodes_failloc "$(osts_nodes)" $fail2 $fail_val2
2337 [ $error -eq 0 ] && [ $rc -ne 0 ] && error "dd error ($rc)"
2338 [ $error -ne 0 ] && [ $rc -eq 0 ] && error "dd success"
2343 [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
2344 skip "need at least 2.8.50 on OST" && return 0
2346 #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
2347 #define OBD_FAIL_PTLRPC_DROP_BULK 0x51a
2348 test_115_read 0x8000051b 0x8000051a
2350 run_test 115a "read: late REQ MDunlink and no bulk"
2353 [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
2354 skip "need at least 2.8.50 on OST" && return 0
2356 #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
2357 #define OBD_FAIL_OST_ENOSPC 0x215
2359 # pass $OSTCOUNT for the fail_loc to be caught
2360 # appropriately by the IO thread
2361 test_115_write 0x8000051b 0x80000215 1 $OSTCOUNT
2363 run_test 115b "write: late REQ MDunlink and no bulk"
2366 [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
2367 skip "need at least 2.8.50 on OST" && return 0
2369 #define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
2370 #define OBD_FAIL_PTLRPC_DROP_BULK 0x51a
2371 test_115_read 0x8000050f 0x8000051a
2373 run_test 115c "read: late Reply MDunlink and no bulk"
2376 [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
2377 skip "need at least 2.8.50 on OST" && return 0
2379 #define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
2380 #define OBD_FAIL_OST_ENOSPC 0x215
2381 test_115_write 0x8000050f 0x80000215 0
2383 run_test 115d "write: late Reply MDunlink and no bulk"
2386 [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
2387 skip "need at least 2.8.50 on OST" && return 0
2389 #define OBD_FAIL_PTLRPC_LONG_BULK_UNLINK 0x510
2390 #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
2391 test_115_read 0x80000510 0x80000211
2393 run_test 115e "read: late Bulk MDunlink and no reply"
2396 [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
2397 skip "need at least 2.8.50 on OST" && return 0
2399 #define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
2400 #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
2401 test_115_read 0x8000051b 0x80000211
2403 run_test 115f "read: late REQ MDunlink and no reply"
2406 [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
2407 skip "need at least 2.8.50 on OST" && return 0
2409 #define OBD_FAIL_PTLRPC_LONG_BOTH_UNLINK 0x51c
2410 test_115_read 0x8000051c 0
2412 run_test 115g "read: late REQ MDunlink and Reply MDunlink"
2414 # parameters: fail_loc CMD RC
2421 #define OBD_FAIL_LDLM_CP_CB_WAIT2 0x320
2422 #define OBD_FAIL_LDLM_CP_CB_WAIT3 0x321
2423 #define OBD_FAIL_LDLM_CP_CB_WAIT4 0x322
2424 #define OBD_FAIL_LDLM_CP_CB_WAIT5 0x323
2427 echo -n "** FLOCK REPLY vs. EVICTION race, lock $2"
2428 [ "$1" = "CLEANUP" ] &&
2429 fail=0x80000320 && echo ", $1 cp first"
2430 [ "$1" = "REPLY" ] &&
2431 fail=0x80000321 && echo ", $1 cp first"
2432 [ "$1" = "DEADLOCK CLEANUP" ] &&
2433 fail=0x80000322 && echo " DEADLOCK, CLEANUP cp first"
2434 [ "$1" = "DEADLOCK REPLY" ] &&
2435 fail=0x80000323 && echo " DEADLOCK, REPLY cp first"
2437 if [ x"$2" = x"get" ]; then
2438 #for TEST lock, take a conflict in advance
2439 # sleep longer than evictor to not confuse fail_loc: 2+2+4
2440 echo "** Taking conflict **"
2441 flocks_test 5 set read sleep 10 $DIR/$tfile &
2447 $LCTL set_param fail_loc=$fail
2449 flocks_test 5 $2 write $DIR/$tfile &
2453 echo "** Evicting and re-connecting client **"
2458 if [ x"$2" = x"get" ]; then
2465 # check if the return value is allowed
2466 [ $rc -eq $3 ] && rc=0
2468 $LCTL set_param fail_loc=0
2472 # a lock is taken, unlock vs. cleanup_resource() race for destroying
2473 # the ORIGINAL lock.
2478 flocks_test 5 set write sleep 4 $DIR/$tfile &
2482 # let unlock to sleep in CP CB
2483 $LCTL set_param fail_loc=$1
2486 # let cleanup to cleep in CP CB
2494 $LCTL set_param fail_loc=0
2499 flock_is_enabled || { skip "mount w/o flock enabled" && return; }
2502 test_120_reply "CLEANUP" set 5 || error "SET race failed"
2503 test_120_reply "CLEANUP" get 5 || error "GET race failed"
2504 test_120_reply "CLEANUP" unlock 5 || error "UNLOCK race failed"
2506 test_120_reply "REPLY" set 5 || error "SET race failed"
2507 test_120_reply "REPLY" get 5 || error "GET race failed"
2508 test_120_reply "REPLY" unlock 5 || error "UNLOCK race failed"
2511 test_120_reply "DEADLOCK CLEANUP" set 5 || error "DEADLOCK race failed"
2512 test_120_reply "DEADLOCK REPLY" set 35 || error "DEADLOCK race failed"
2514 test_120_destroy 0x320 || error "unlock-cleanup race failed"
2516 run_test 120 "flock race: completion vs. evict"
2519 local BEFORE=$(date +%s)
2522 # modify dir so that next revalidate would not obtain UPDATE lock
2525 # drop 1 reply with UPDATE lock,
2526 # resend should not create 2nd lock on server
2527 mcreate $DIR/$tfile || error "mcreate failed: $?"
2528 drop_mdt_ldlm_reply_once "stat $DIR/$tfile" || error "stat failed: $?"
2530 # 2 BL AST will be sent to client, both must find the same lock,
2531 # race them to not get EINVAL for 2nd BL AST
2532 #define OBD_FAIL_LDLM_PAUSE_CANCEL2 0x31f
2533 $LCTL set_param fail_loc=0x8000031f
2535 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=0 > /dev/null
2536 chmod 0777 $DIR/$tfile || error "chmod failed: $?"
2537 $LCTL set_param ldlm.namespaces.*.early_lock_cancel=1 > /dev/null
2539 # let the client reconnect
2541 EVICT=$($LCTL get_param mdc.$FSNAME-MDT*.state |
2542 awk -F"[ [,]" '/EVICTED ]$/ { if (mx<$5) {mx=$5;} } END { print mx }')
2544 [ -z "$EVICT" ] || [[ $EVICT -le $BEFORE ]] || error "eviction happened"
2546 run_test 113 "ldlm enqueue dropped reply should not cause deadlocks"
2550 test_mkdir -p $DIR/$tdir
2552 # Prevent interference from layout intent RPCs due to
2553 # asynchronous writeback. These will be tested in 130c below.
2554 do_nodes ${CLIENTS:-$HOSTNAME} sync
2556 # get only LOOKUP lock on $tdir
2557 cancel_lru_locks mdc
2558 ls $DIR/$tdir/$tfile 2>/dev/null
2560 # get getattr by fid on $tdir
2562 # we need to race with unlink, unlink must complete before we will
2563 # take a DLM lock, otherwise unlink will wait until getattr will
2564 # complete; but later than getattr starts so that getattr found
2566 #define OBD_FAIL_MDS_INTENT_DELAY 0x160
2567 set_nodes_failloc "$(mdts_nodes)" 0x80000160
2574 # drop the reply so that resend happens on an unlinked file.
2575 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2576 set_nodes_failloc "$(mdts_nodes)" 0x80000157
2580 remote_mds_nodsh && skip "remote MDS with nodsh" && return
2581 local server_version=$(lustre_version_code $SINGLEMDS)
2582 [[ $server_version -ge $(version_code 2.7.2) ]] ||
2583 { skip "Need server version newer than 2.7.1"; return 0; }
2587 wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
2590 run_test 130a "enqueue resend on not existing file"
2593 remote_mds_nodsh && skip "remote MDS with nodsh" && return
2594 local server_version=$(lustre_version_code $SINGLEMDS)
2595 [[ $server_version -ge $(version_code 2.7.2) ]] ||
2596 { skip "Need server version newer than 2.7.1"; return 0; }
2599 # let the reply to be dropped
2602 #define OBD_FAIL_SRV_ENOENT 0x217
2603 set_nodes_failloc "$(mdts_nodes)" 0x80000217
2605 wait $T130_PID || [ $? -eq 0 ] && error "stat should fail"
2608 run_test 130b "enqueue resend on a stale inode"
2611 remote_mds_nodsh && skip "remote MDS with nodsh" && return
2613 do_nodes ${CLIENTS:-$HOSTNAME} sync
2614 echo XXX > $DIR/$tfile
2616 cancel_lru_locks mdc
2618 # Trigger writeback on $tfile.
2620 # we need to race with unlink, unlink must complete before we will
2621 # take a DLM lock, otherwise unlink will wait until intent will
2622 # complete; but later than intent starts so that intent found
2624 #define OBD_FAIL_MDS_INTENT_DELAY 0x160
2625 set_nodes_failloc "$(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 "$(mdts_nodes)" 0x80000157
2636 # let the reply to be dropped
2639 #define OBD_FAIL_SRV_ENOENT 0x217
2640 set_nodes_failloc "$(mdts_nodes)" 0x80000217
2646 run_test 130c "layout intent resend on a stale inode"
2649 local before=$(date +%s)
2652 mount_client $MOUNT2 || error "mount filed"
2655 # get a lock on client so that export would reach the stale list
2656 $SETSTRIPE -i 0 $DIR/$tfile || error "setstripe failed"
2657 dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=fsync ||
2660 #define OBD_FAIL_OST_PAUSE_PUNCH 0x236
2661 do_facet ost1 $LCTL set_param fail_val=120 fail_loc=0x80000236
2663 $TRUNCATE $DIR/$tfile 100 &
2666 dd if=/dev/zero of=$DIR2/$tfile bs=4096 count=1 conv=notrunc ||
2670 umount_client $MOUNT2
2672 evict=$(do_facet client $LCTL get_param \
2673 osc.$FSNAME-OST0000-osc-*/state |
2674 awk -F"[ [,]" '/EVICTED ]$/ { if (t<$5) {t=$5;} } END { print t }')
2676 [ -z "$evict" ] || [[ $evict -le $before ]] ||
2677 (do_facet client $LCTL get_param \
2678 osc.$FSNAME-OST0000-osc-*/state;
2679 error "eviction happened: $evict before:$before")
2681 run_test 132 "long punch"
2684 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
2687 # get a lock on client so that export would reach the stale list
2688 $SETSTRIPE -i 0 $DIR/$tfile || error "setstripe failed"
2689 dd if=/dev/zero of=$DIR/$tfile count=1 || error "dd failed"
2691 # another IO under the same lock
2692 #define OBD_FAIL_OSC_DELAY_IO 0x414
2693 $LCTL set_param fail_loc=0x80000414
2694 $LCTL set_param fail_val=4 fail_loc=0x80000414
2695 dd if=/dev/zero of=$DIR/$tfile count=1 conv=notrunc oflag=dsync &
2699 #define OBD_FAIL_LDLM_BL_EVICT 0x31e
2700 set_nodes_failloc "$(osts_nodes)" 0x8000031e
2704 wait $pid && error "dd succeeded"
2707 run_test 131 "IO vs evict results to IO under staled lock"
2710 local list=$(comma_list $(mdts_nodes))
2712 local t=$((TIMEOUT * 2))
2715 flock $DIR/$tfile -c "echo bl lock;sleep $t;echo bl flock unlocked" &
2717 multiop_bg_pause $DIR/$tfile O_jc || return 1
2720 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2721 do_nodes $list $LCTL set_param fail_loc=0x80000157
2723 echo "waiting for multiop $PID"
2724 wait $PID || return 2
2730 run_test 133 "don't fail on flock resend"
2733 [ -z "$CLIENTS" ] && skip "Need two or more clients" && return
2734 [ $CLIENTCOUNT -lt 2 ] &&
2735 { skip "Need 2+ clients, have $CLIENTCOUNT" && return; }
2737 mkdir -p $MOUNT/$tdir/1 $MOUNT/$tdir/2 || error "mkdir failed"
2738 touch $MOUNT/$tdir/1/$tfile $MOUNT/$tdir/2/$tfile ||
2739 error "touch failed"
2740 zconf_umount_clients $CLIENTS $MOUNT
2741 zconf_mount_clients $CLIENTS $MOUNT
2743 #define OBD_FAIL_TGT_REPLY_DATA_RACE 0x722
2744 # assume commit interval is 5
2745 do_facet mds1 "$LCTL set_param fail_loc=0x722 fail_val=5"
2747 local -a clients=(${CLIENTS//,/ })
2748 local client1=${clients[0]}
2749 local client2=${clients[1]}
2751 do_node $client1 rm $MOUNT/$tdir/1/$tfile &
2753 do_node $client2 mv $MOUNT/$tdir/2/$tfile $MOUNT/$tdir/2/${tfile}_2 &
2756 wait $rmpid || error "rm failed"
2757 wait $mvpid || error "mv failed"
2760 run_test 134 "race between failover and search for reply data free slot"
2763 [ $MDS1_VERSION -lt $(version_code 2.12.51) ] &&
2764 skip "Need MDS version at least 2.12.51"
2767 $LFS setstripe -E 1M -L mdt $DIR/$tdir
2768 # to have parent dir write lock before open/resend
2769 touch $DIR/$tdir/$tfile
2770 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
2771 do_nodes $(comma_list $(mdts_nodes)) $LCTL set_param fail_loc=0x80000157
2772 openfile -f O_RDWR:O_CREAT -m 0755 $DIR/$tdir/$tfile ||
2773 error "Failed to open DOM file"
2775 run_test 135 "DOM: open/create resend to return size"
2778 remote_mds_nodsh && skip "remote MDS with nodsh" && return
2779 [[ $MDS1_VERSION -ge $(version_code 2.12.52) ]] ||
2780 skip "Need MDS version at least 2.12.52"
2782 local mdts=$(comma_list $(mdts_nodes))
2783 local MDT0=$(facet_svc $SINGLEMDS)
2785 local clog=$(do_facet mds1 $LCTL --device $MDT0 changelog_register -n)
2786 [ -n "$clog" ] || error "changelog_register failed"
2787 cl_mask=$(do_facet mds1 $LCTL get_param \
2788 mdd.$MDT0.changelog_mask -n)
2789 changelog_chmask "ALL"
2791 # generate some changelog records to accumulate
2792 test_mkdir -i 0 -c 0 $DIR/$tdir || error "mkdir $tdir failed"
2793 createmany -m $DIR/$tdir/$tfile 10000 ||
2794 error "create $DIR/$tdir/$tfile failed"
2796 local size1=$(do_facet $SINGLEMDS \
2797 $LCTL get_param -n mdd.$MDT0.changelog_size)
2798 echo "Changelog size $size1"
2800 #define OBD_FAIL_LLOG_PURGE_DELAY 0x1318
2801 do_nodes $mdts $LCTL set_param fail_loc=0x1318 fail_val=30
2803 # launch changelog_deregister in background on MDS
2804 do_facet mds1 "nohup $LCTL --device $MDT0 changelog_deregister $clog \
2805 > foo.out 2> foo.err < /dev/null &"
2806 # give time to reach fail_loc
2809 # fail_loc will make MDS sleep in the middle of changelog_deregister
2810 # take this opportunity to abruptly kill MDS
2811 FAILURE_MODE_save=$FAILURE_MODE
2814 FAILURE_MODE=$FAILURE_MODE_save
2816 do_nodes $mdts $LCTL set_param fail_loc=0x0 fail_val=0
2818 local size2=$(do_facet $SINGLEMDS \
2819 $LCTL get_param -n mdd.$MDT0.changelog_size)
2820 echo "Changelog size $size2"
2821 local clog2=$(do_facet $SINGLEMDS "$LCTL get_param -n \
2822 mdd.$MDT0.changelog_users | grep $clog")
2823 echo "After crash, changelog user $clog2"
2825 [ -n "$clog2" -o $size2 -lt $size1 ] ||
2826 error "changelog record count unchanged"
2828 do_facet mds1 $LCTL set_param mdd.$MDT0.changelog_mask=\'$cl_mask\' -n
2830 run_test 136 "changelog_deregister leaving pending records"
2833 check_and_cleanup_lustre