7 # This test needs to be run on the client
10 export MULTIOP=${MULTIOP:-multiop}
11 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
14 . $LUSTRE/tests/test-framework.sh
16 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
18 CHECK_GRANT=${CHECK_GRANT:-"yes"}
19 GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
21 require_dsh_mds || exit 0
24 # bug number for skipped tests:
25 ALWAYS_EXCEPT="$REPLAY_SINGLE_EXCEPT "
26 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
29 [ "$SLOW" = "no" ] && EXCEPT_SLOW="44b"
31 [ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
32 # bug number for skipped test:
33 ALWAYS_EXCEPT="$ALWAYS_EXCEPT "
37 check_and_setup_lustre
42 rm -rf $DIR/[df][0-9]* $DIR/f.$TESTSUITE.*
44 # LU-482 Avert LVM and VM inability to flush caches in pre .33 kernels
45 if [ $LINUX_VERSION_CODE -lt $(version_code 2.6.33) ]; then
47 do_facet $SINGLEMDS sync
50 test_0a() { # was test_0
51 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
52 replay_barrier $SINGLEMDS
56 run_test 0a "empty replay"
59 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
61 # this test attempts to trigger a race in the precreation code,
62 # and must run before any other objects are created on the filesystem
64 createmany -o $DIR/$tfile 20 || error "createmany -o $DIR/$tfile failed"
65 unlinkmany $DIR/$tfile 20 || error "unlinkmany $DIR/$tfile failed"
67 run_test 0b "ensure object created after recover exists. (3284)"
70 replay_barrier $SINGLEMDS
73 facet_failover $SINGLEMDS
74 zconf_mount $(hostname) $MOUNT || error "mount fails"
75 client_up || error "post-failover df failed"
76 # file shouldn't exist if replay-barrier works as expected
77 rm $DIR/$tfile && error "File exists and it shouldn't"
80 run_test 0c "check replay-barrier"
83 replay_barrier $SINGLEMDS
85 facet_failover $SINGLEMDS
86 zconf_mount $(hostname) $MOUNT || error "mount fails"
87 client_up || error "post-failover df failed"
89 run_test 0d "expired recovery with no clients"
92 replay_barrier $SINGLEMDS
95 $CHECKSTAT -t file $DIR/$tfile ||
96 error "$CHECKSTAT $DIR/$tfile attribute check failed"
99 run_test 1 "simple create"
102 replay_barrier $SINGLEMDS
105 $CHECKSTAT -t file $DIR/$tfile ||
106 error "$CHECKSTAT $DIR/$tfile attribute check failed"
112 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
113 replay_barrier $SINGLEMDS
116 $CHECKSTAT -t file $DIR/$tfile ||
117 error "$CHECKSTAT $DIR/$tfile attribute check failed"
123 replay_barrier $SINGLEMDS
124 $LFS setstripe -c $OSTCOUNT $DIR/$tfile
126 $CHECKSTAT -t file $DIR/$tfile ||
127 error "$CHECKSTAT $DIR/$tfile check failed"
129 run_test 2c "setstripe replay"
132 replay_barrier $SINGLEMDS
133 $LFS setdirstripe -i 0 -c $MDSCOUNT $DIR/$tdir
135 $CHECKSTAT -t dir $DIR/$tdir ||
136 error "$CHECKSTAT $DIR/$tdir check failed"
138 run_test 2d "setdirstripe replay"
141 local file=$DIR/$tfile
142 replay_barrier $SINGLEMDS
144 openfile -f O_DIRECTORY $file
146 $CHECKSTAT -t file $file ||
147 error "$CHECKSTAT $file attribute check failed"
150 run_test 3a "replay failed open(O_DIRECTORY)"
153 replay_barrier $SINGLEMDS
154 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
155 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000114"
157 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
159 $CHECKSTAT -t file $DIR/$tfile &&
160 error "$CHECKSTAT $DIR/$tfile attribute check should fail"
163 run_test 3b "replay failed open -ENOMEM"
166 replay_barrier $SINGLEMDS
167 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
168 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
170 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
173 $CHECKSTAT -t file $DIR/$tfile &&
174 error "$CHECKSTAT $DIR/$tfile attribute check should fail"
177 run_test 3c "replay failed open -ENOMEM"
179 test_4a() { # was test_4
180 replay_barrier $SINGLEMDS
181 for i in $(seq 10); do
182 echo "tag-$i" > $DIR/$tfile-$i
185 for i in $(seq 10); do
186 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
189 run_test 4a "|x| 10 open(O_CREAT)s"
192 for i in $(seq 10); do
193 echo "tag-$i" > $DIR/$tfile-$i
195 replay_barrier $SINGLEMDS
198 $CHECKSTAT -t file $DIR/$tfile-* &&
199 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
202 run_test 4b "|x| rm 10 files"
204 # The idea is to get past the first block of precreated files on both
205 # osts, and then replay.
207 replay_barrier $SINGLEMDS
208 for i in $(seq 220); do
209 echo "tag-$i" > $DIR/$tfile-$i
212 for i in $(seq 220); do
213 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
217 # waiting for commitment of removal
219 run_test 5 "|x| 220 open(O_CREAT)"
221 test_6a() { # was test_6
222 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
223 replay_barrier $SINGLEMDS
224 mcreate $DIR/$tdir/$tfile
226 $CHECKSTAT -t dir $DIR/$tdir ||
227 error "$CHECKSTAT $DIR/$tdir attribute check failed"
228 $CHECKSTAT -t file $DIR/$tdir/$tfile ||
229 error "$CHECKSTAT $DIR/$tdir/$tfile attribute check failed"
231 # waiting for log process thread
233 run_test 6a "mkdir + contained create"
236 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
237 replay_barrier $SINGLEMDS
240 $CHECKSTAT -t dir $DIR/$tdir &&
241 error "$CHECKSTAT $DIR/$tdir attribute check should fail" ||
244 run_test 6b "|X| rmdir"
247 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
248 replay_barrier $SINGLEMDS
249 mcreate $DIR/$tdir/$tfile
251 $CHECKSTAT -t dir $DIR/$tdir ||
252 error "$CHECKSTAT $DIR/$tdir attribute check failed"
253 $CHECKSTAT -t file $DIR/$tdir/$tfile ||
254 error "$CHECKSTAT $DIR/$tdir/$tfile attribute check failed"
257 run_test 7 "mkdir |X| contained create"
260 replay_barrier $SINGLEMDS
261 multiop_bg_pause $DIR/$tfile mo_c ||
262 error "multiop mknod $DIR/$tfile failed"
266 $CHECKSTAT -t file $DIR/$tfile ||
267 error "$CHECKSTAT $DIR/$tfile attribute check failed"
268 kill -USR1 $MULTIPID || error "multiop mknod $MULTIPID not running"
269 wait $MULTIPID || error "multiop mknod $MULTIPID failed"
272 run_test 8 "creat open |X| close"
275 replay_barrier $SINGLEMDS
277 local old_inum=$(ls -i $DIR/$tfile | awk '{print $1}')
279 local new_inum=$(ls -i $DIR/$tfile | awk '{print $1}')
281 echo " old_inum == $old_inum, new_inum == $new_inum"
282 if [ $old_inum -eq $new_inum ] ;
284 echo "old_inum and new_inum match"
286 echo " old_inum and new_inum do not match"
287 error "old index($old_inum) does not match new index($new_inum)"
291 run_test 9 "|X| create (same inum/gen)"
294 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
295 replay_barrier $SINGLEMDS
296 mv $DIR/$tfile $DIR/$tfile-2
299 $CHECKSTAT $DIR/$tfile &&
300 error "$CHECKSTAT $DIR/$tfile attribute check should fail"
301 $CHECKSTAT $DIR/$tfile-2 ||
302 error "$CHECKSTAT $DIR/$tfile-2 attribute check failed"
306 run_test 10 "create |X| rename unlink"
309 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
310 echo "old" > $DIR/$tfile
311 mv $DIR/$tfile $DIR/$tfile-2
312 replay_barrier $SINGLEMDS
313 echo "new" > $DIR/$tfile
315 grep old $DIR/$tfile-2
317 grep new $DIR/$tfile || error "grep $DIR/$tfile failed"
318 grep old $DIR/$tfile-2 || error "grep $DIR/$tfile-2 failed"
320 run_test 11 "create open write rename |X| create-old-name read"
323 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
324 multiop_bg_pause $DIR/$tfile o_tSc ||
325 error "multiop_bg_pause $DIR/$tfile failed"
328 replay_barrier $SINGLEMDS
329 kill -USR1 $pid || error "multiop $pid not running"
330 wait $pid || error "multiop $pid failed"
333 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
336 run_test 12 "open, unlink |X| close"
338 # 1777 - replay open after committed chmod that would make
339 # a regular open a failure
341 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
342 multiop_bg_pause $DIR/$tfile O_wc ||
343 error "multiop_bg_pause $DIR/$tfile failed"
346 $CHECKSTAT -p 0 $DIR/$tfile ||
347 error "$CHECKSTAT $DIR/$tfile attribute check failed"
348 replay_barrier $SINGLEMDS
350 kill -USR1 $pid || error "multiop $pid not running"
351 wait $pid || error "multiop $pid failed"
353 $CHECKSTAT -s 1 -p 0 $DIR/$tfile ||
354 error "second $CHECKSTAT $DIR/$tfile attribute check failed"
355 rm $DIR/$tfile || error "rm $DIR/$tfile failed"
358 run_test 13 "open chmod 0 |x| write close"
361 multiop_bg_pause $DIR/$tfile O_tSc ||
362 error "multiop_bg_pause $DIR/$tfile failed"
365 replay_barrier $SINGLEMDS
366 kill -USR1 $pid || error "multiop $pid not running"
367 wait $pid || error "multiop $pid failed"
370 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
373 run_test 14 "open(O_CREAT), unlink |X| close"
376 multiop_bg_pause $DIR/$tfile O_tSc ||
377 error "multiop_bg_pause $DIR/$tfile failed"
380 replay_barrier $SINGLEMDS
381 touch $DIR/$tfile-1 || error "touch $DIR/$tfile-1 failed"
382 kill -USR1 $pid || error "multiop $pid not running"
383 wait $pid || error "multiop $pid failed"
386 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
387 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
390 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
393 replay_barrier $SINGLEMDS
396 mcreate $DIR/$tfile-2
398 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
399 [ -e $DIR/$tfile-2 ] || error "file $DIR/$tfile-2 does not exist"
400 munlink $DIR/$tfile-2 || error "munlink $DIR/$tfile-2 failed"
402 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
405 replay_barrier $SINGLEMDS
406 multiop_bg_pause $DIR/$tfile O_c ||
407 error "multiop_bg_pause $DIR/$tfile failed"
410 kill -USR1 $pid || error "multiop $pid not running"
411 wait $pid || error "multiop $pid failed"
412 $CHECKSTAT -t file $DIR/$tfile ||
413 error "$CHECKSTAT $DIR/$tfile attribute check failed"
416 run_test 17 "|X| open(O_CREAT), |replay| close"
419 replay_barrier $SINGLEMDS
420 multiop_bg_pause $DIR/$tfile O_tSc ||
421 error "multiop_bg_pause $DIR/$tfile failed"
424 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
425 echo "pid: $pid will close"
426 kill -USR1 $pid || error "multiop $pid not running"
427 wait $pid || error "multiop $pid failed"
430 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
431 [ -e $DIR/$tfile-2 ] || error "file $DIR/$tfile-2 does not exist"
432 # this touch frequently fails
433 touch $DIR/$tfile-3 || error "touch $DIR/$tfile-3 failed"
434 munlink $DIR/$tfile-2 || error "munlink $DIR/$tfile-2 failed"
435 munlink $DIR/$tfile-3 || error "munlink $DIR/$tfile-3 failed"
438 run_test 18 "open(O_CREAT), unlink, touch new, close, touch, unlink"
440 # bug 1855 (a simpler form of test_11 above)
442 replay_barrier $SINGLEMDS
444 echo "old" > $DIR/$tfile
445 mv $DIR/$tfile $DIR/$tfile-2
446 grep old $DIR/$tfile-2
448 grep old $DIR/$tfile-2 || error "grep $DIR/$tfile-2 failed"
450 run_test 19 "mcreate, open, write, rename "
452 test_20a() { # was test_20
453 replay_barrier $SINGLEMDS
454 multiop_bg_pause $DIR/$tfile O_tSc ||
455 error "multiop_bg_pause $DIR/$tfile failed"
460 kill -USR1 $pid || error "multiop $pid not running"
461 wait $pid || error "multiop $pid failed"
462 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
465 run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
467 test_20b() { # bug 10480
468 local wait_timeout=$((TIMEOUT * 4))
469 local extra=$(fs_log_size)
473 $LFS setstripe -i 0 -c 1 $DIR
475 local beforeused=$(df -P $DIR | tail -1 | awk '{ print $3 }')
477 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
478 while [ ! -e $DIR/$tfile ] ; do
479 usleep 60 # give dd a chance to start
482 $LFS getstripe $DIR/$tfile || error "$LFS getstripe $DIR/$tfile failed"
484 rm -f $DIR/$tfile || error "rm -f $DIR/$tfile failed"
486 client_up || client_up || true # reconnect
488 do_facet $SINGLEMDS "lctl set_param -n osd*.*MDT*.force_sync=1"
490 fail $SINGLEMDS # start orphan recovery
491 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
492 wait_delete_completed $wait_timeout || error "delete did not finish"
496 local afterused=$(df -P $DIR | tail -1 | awk '{ print $3 }')
497 log "before $beforeused, after $afterused"
499 (( $beforeused + $extra >= $afterused )) && break
500 n_attempts=$((n_attempts + 1))
501 [ $n_attempts -gt 3 ] &&
502 error "after $afterused > before $beforeused + $extra"
504 wait_zfs_commit $SINGLEMDS 5
509 run_test 20b "write, unlink, eviction, replay (test mds_cleanup_orphans)"
511 test_20c() { # bug 10480
512 multiop_bg_pause $DIR/$tfile Ow_c ||
513 error "multiop_bg_pause $DIR/$tfile failed"
519 client_up || client_up || true # reconnect
521 kill -USR1 $pid || error "multiop $pid not running"
522 wait $pid || error "multiop $pid failed"
523 [ -s $DIR/$tfile ] || error "File was truncated"
527 run_test 20c "check that client eviction does not affect file content"
530 replay_barrier $SINGLEMDS
531 multiop_bg_pause $DIR/$tfile O_tSc ||
532 error "multiop_bg_pause $DIR/$tfile failed"
535 touch $DIR/$tfile-1 || error "touch $DIR/$tfile-1 failed"
538 kill -USR1 $pid || error "multiop $pid not running"
539 wait $pid || error "multiop $pid failed"
540 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
541 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
544 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
547 multiop_bg_pause $DIR/$tfile O_tSc ||
548 error "multiop_bg_pause $DIR/$tfile failed"
551 replay_barrier $SINGLEMDS
555 kill -USR1 $pid || error "multiop $pid not running"
556 wait $pid || error "multiop $pid failed"
557 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
560 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
563 multiop_bg_pause $DIR/$tfile O_tSc ||
564 error "multiop_bg_pause $DIR/$tfile failed"
567 replay_barrier $SINGLEMDS
569 touch $DIR/$tfile-1 || error "touch $DIR/$tfile-1 failed"
572 kill -USR1 $pid || error "multiop $pid not running"
573 wait $pid || error "multiop $pid failed"
574 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
575 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
578 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
581 multiop_bg_pause $DIR/$tfile O_tSc ||
582 error "multiop_bg_pause $DIR/$tfile failed"
585 replay_barrier $SINGLEMDS
588 kill -USR1 $pid || error "multiop $pid not running"
589 wait $pid || error "multiop $pid failed"
590 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
593 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
596 multiop_bg_pause $DIR/$tfile O_tSc ||
597 error "multiop_bg_pause $DIR/$tfile failed"
601 replay_barrier $SINGLEMDS
603 kill -USR1 $pid || error "multiop $pid not running"
604 wait $pid || error "multiop $pid failed"
605 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
608 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
611 replay_barrier $SINGLEMDS
612 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
613 error "multiop_bg_pause $DIR/$tfile-1 failed"
615 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
616 error "multiop_bg_pause $DIR/$tfile-2 failed"
620 kill -USR1 $pid2 || error "second multiop $pid2 not running"
621 wait $pid2 || error "second multiop $pid2 failed"
624 kill -USR1 $pid1 || error "multiop $pid1 not running"
625 wait $pid1 || error "multiop $pid1 failed"
626 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
627 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
630 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
633 replay_barrier $SINGLEMDS
634 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
635 error "multiop_bg_pause $DIR/$tfile-1 failed"
637 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
638 error "multiop_bg_pause $DIR/$tfile-2 failed"
644 kill -USR1 $pid1 || error "multiop $pid1 not running"
645 wait $pid1 || error "multiop $pid1 failed"
646 kill -USR1 $pid2 || error "second multiop $pid2 not running"
647 wait $pid2 || error "second multiop $pid2 failed"
648 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
649 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
652 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
655 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
656 error "multiop_bg_pause $DIR/$tfile-1 failed"
658 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
659 error "multiop_bg_pause $DIR/$tfile-2 failed"
661 replay_barrier $SINGLEMDS
664 kill -USR1 $pid2 || error "second multiop $pid2 not running"
665 wait $pid2 || error "second multiop $pid2 failed"
668 kill -USR1 $pid1 || error "multiop $pid1 not running"
669 wait $pid1 || error "multiop $pid1 failed"
670 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
671 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
674 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
677 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
678 error "multiop_bg_pause $DIR/$tfile-1 failed"
680 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
681 error "multiop_bg_pause $DIR/$tfile-2 failed"
683 replay_barrier $SINGLEMDS
688 kill -USR1 $pid1 || error "multiop $pid1 not running"
689 wait $pid1 || error "multiop $pid1 failed"
690 kill -USR1 $pid2 || error "second multiop $pid2 not running"
691 wait $pid2 || error "second multiop $pid2 failed"
692 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
693 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
696 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
699 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
700 error "multiop_bg_pause $DIR/$tfile-1 failed"
702 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
703 error "multiop_bg_pause $DIR/$tfile-2 failed"
708 replay_barrier $SINGLEMDS
710 kill -USR1 $pid1 || error "multiop $pid1 not running"
711 wait $pid1 || error "multiop $pid1 failed"
712 kill -USR1 $pid2 || error "second multiop $pid2 not running"
713 wait $pid2 || error "second multiop $pid2 failed"
714 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
715 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
718 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
721 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
722 error "multiop_bg_pause $DIR/$tfile-1 failed"
724 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
725 error "multiop_bg_pause $DIR/$tfile-2 failed"
729 replay_barrier $SINGLEMDS
732 kill -USR1 $pid1 || error "multiop $pid1 not running"
733 wait $pid1 || error "multiop $pid1 failed"
734 kill -USR1 $pid2 || error "second multiop $pid2 not running"
735 wait $pid2 || error "second multiop $pid2 failed"
736 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
737 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
740 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
742 # tests for bug 2104; completion without crashing is success. The close is
743 # stale, but we always return 0 for close, so the app never sees it.
745 multiop_bg_pause $DIR/$tfile O_c ||
746 error "multiop_bg_pause $DIR/$tfile failed"
748 multiop_bg_pause $DIR/$tfile O_c ||
749 error "second multiop_bg_pause $DIR/$tfile failed"
752 client_up || client_up || error "client_up failed"
753 kill -USR1 $pid1 || error "multiop $pid1 not running"
754 kill -USR1 $pid2 || error "second multiop $pid2 not running"
755 wait $pid1 || error "multiop $pid1 failed"
756 wait $pid2 || error "second multiop $pid2 failed"
759 run_test 32 "close() notices client eviction; close() after client eviction"
762 createmany -o $DIR/$tfile-%d 10 ||
763 error "createmany create $DIR/$tfile failed"
764 replay_barrier_nosync $SINGLEMDS
765 fail_abort $SINGLEMDS
766 # recreate shouldn't fail
767 createmany -o $DIR/$tfile--%d 10 ||
768 error "createmany recreate $DIR/$tfile failed"
772 run_test 33a "fid seq shouldn't be reused after abort recovery"
775 #define OBD_FAIL_SEQ_ALLOC 0x1311
776 do_facet $SINGLEMDS "lctl set_param fail_loc=0x1311"
778 createmany -o $DIR/$tfile-%d 10
779 replay_barrier_nosync $SINGLEMDS
780 fail_abort $SINGLEMDS
781 # recreate shouldn't fail
782 createmany -o $DIR/$tfile--%d 10 ||
783 error "createmany recreate $DIR/$tfile failed"
787 run_test 33b "test fid seq allocation"
790 multiop_bg_pause $DIR/$tfile O_c ||
791 error "multiop_bg_pause $DIR/$tfile failed"
795 replay_barrier $SINGLEMDS
796 fail_abort $SINGLEMDS
797 kill -USR1 $pid || error "multiop $pid not running"
798 wait $pid || error "multiop $pid failed"
799 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
803 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
805 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
807 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
809 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
810 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
815 # give a chance to remove from MDS
816 fail_abort $SINGLEMDS
817 $CHECKSTAT -t file $DIR/$tfile &&
818 error "$CHECKSTAT $DIR/$tfile attribute check should fail" ||
821 run_test 35 "test recovery from llog for unlink op"
823 # b=2432 resent cancel after replay uses wrong cookie,
824 # so don't resend cancels
826 replay_barrier $SINGLEMDS
828 checkstat $DIR/$tfile
829 facet_failover $SINGLEMDS
831 if dmesg | grep "unknown lock cookie"; then
832 error "cancel after replay failed"
835 run_test 36 "don't resend cancel"
838 # directory orphans can't be unlinked from PENDING directory
840 rmdir $DIR/$tfile 2>/dev/null
841 multiop_bg_pause $DIR/$tfile dD_c ||
842 error "multiop_bg_pause $DIR/$tfile failed"
846 replay_barrier $SINGLEMDS
847 # clear the dmesg buffer so we only see errors from this recovery
848 do_facet $SINGLEMDS dmesg -c >/dev/null
849 fail_abort $SINGLEMDS
850 kill -USR1 $pid || error "multiop $pid not running"
851 do_facet $SINGLEMDS dmesg | grep "error .* unlinking .* from PENDING" &&
852 error "error unlinking files"
853 wait $pid || error "multiop $pid failed"
857 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
860 createmany -o $DIR/$tfile-%d 800 ||
861 error "createmany -o $DIR/$tfile failed"
862 unlinkmany $DIR/$tfile-%d 0 400 || error "unlinkmany $DIR/$tfile failed"
863 replay_barrier $SINGLEMDS
865 unlinkmany $DIR/$tfile-%d 400 400 ||
866 error "unlinkmany $DIR/$tfile 400 failed"
868 $CHECKSTAT -t file $DIR/$tfile-* &&
869 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
872 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
874 test_39() { # bug 4176
875 createmany -o $DIR/$tfile-%d 800 ||
876 error "createmany -o $DIR/$tfile failed"
877 replay_barrier $SINGLEMDS
878 unlinkmany $DIR/$tfile-%d 0 400
880 unlinkmany $DIR/$tfile-%d 400 400 ||
881 error "unlinkmany $DIR/$tfile 400 failed"
883 $CHECKSTAT -t file $DIR/$tfile-* &&
884 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
887 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
890 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
895 # always need connection to MDS to verify layout during IO. LU-2628.
896 lctl get_param mdc.*.connect_flags | grep -q layout_lock &&
897 skip "layout_lock needs MDS connection for IO" && return 0
899 $LCTL mark multiop $MOUNT/$tfile OS_c
900 multiop $MOUNT/$tfile OS_c &
902 writeme -s $MOUNT/${tfile}-2 &
905 facet_failover $SINGLEMDS
906 #define OBD_FAIL_MDS_CONNECT_NET 0x117
907 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
909 stat1=$(count_ost_writes)
911 stat2=$(count_ost_writes)
912 echo "$stat1, $stat2"
913 if [ $stat1 -lt $stat2 ]; then
914 echo "writes continuing during recovery"
917 echo "writes not continuing during recovery, bug 2477"
920 echo "waiting for writeme $WRITE_PID"
924 echo "waiting for multiop $PID"
925 wait $PID || error "multiop $PID failed"
926 do_facet client munlink $MOUNT/$tfile ||
927 error "munlink $MOUNT/$tfile failed"
928 do_facet client munlink $MOUNT/${tfile}-2 ||
929 error "munlink $MOUNT/$tfile-2 failed"
932 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
935 # make sure that a read to one osc doesn't try to double-unlock its page just
936 # because another osc is invalid. trigger_group_io used to mistakenly return
937 # an error if any oscs were invalid even after having successfully put rpcs
938 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
939 # the page, guarnateeing that the unlock from the RPC completion would
940 # assert on trying to unlock the unlocked page.
942 [ $OSTCOUNT -lt 2 ] && skip_env "needs >= 2 OSTs" && return
944 local f=$MOUNT/$tfile
945 # make sure the start of the file is ost1
946 $SETSTRIPE -S $((128 * 1024)) -i 0 $f
947 do_facet client dd if=/dev/zero of=$f bs=4k count=1 ||
948 error "dd on client failed"
950 # fail ost2 and read from ost1
951 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $ost2_svc)
952 local osc2dev=$(do_facet $SINGLEMDS "lctl get_param -n devices" |
953 grep $mdtosc | awk '{print $1}')
954 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" &&
955 lctl get_param -n devices &&
956 error "OST 2 $osc2dev does not exist"
957 do_facet $SINGLEMDS $LCTL --device $osc2dev deactivate ||
958 error "deactive device on $SINGLEMDS failed"
959 do_facet client dd if=$f of=/dev/null bs=4k count=1 ||
960 error "second dd on client failed"
961 do_facet $SINGLEMDS $LCTL --device $osc2dev activate ||
962 error "active device on $SINGLEMDS failed"
965 run_test 41 "read from a valid osc while other oscs are invalid"
967 # test MDS recovery after ost failure
969 blocks=$(df -P $MOUNT | tail -n 1 | awk '{ print $2 }')
970 createmany -o $DIR/$tfile-%d 800 ||
971 error "createmany -o $DIR/$tfile failed"
973 unlinkmany $DIR/$tfile-%d 0 400
975 lctl set_param debug=-1
978 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
979 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
980 #[ $blocks_after -lt $blocks ] || return 1
981 echo "wait for MDS to timeout and recover"
982 sleep $((TIMEOUT * 2))
984 unlinkmany $DIR/$tfile-%d 400 400 ||
985 error "unlinkmany $DIR/$tfile 400 failed"
986 $CHECKSTAT -t file $DIR/$tfile-* &&
987 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
990 run_test 42 "recovery after ost failure"
992 # timeout in MDS/OST recovery RPC will LBUG MDS
993 test_43() { # bug 2530
994 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
996 replay_barrier $SINGLEMDS
998 # OBD_FAIL_OST_CREATE_NET 0x204
999 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1005 run_test 43 "mds osc import failure during recovery; don't LBUG"
1007 test_44a() { # was test_44
1008 local at_max_saved=0
1010 local mdcdev=$($LCTL dl |
1011 awk "/${FSNAME}-MDT0000-mdc-/ {if (\$2 == \"UP\") {print \$1}}")
1012 [ "$mdcdev" ] || error "${FSNAME}-MDT0000-mdc- not UP"
1013 [ $(echo $mdcdev | wc -w) -eq 1 ] ||
1014 { $LCTL dl; error "looking for mdcdev=$mdcdev"; }
1016 # adaptive timeouts slow this way down
1017 if at_is_enabled; then
1018 at_max_saved=$(at_max_get mds)
1022 for i in $(seq 1 10); do
1023 echo "$i of 10 ($(date +%s))"
1024 do_facet $SINGLEMDS \
1025 "lctl get_param -n md[ts].*.mdt.timeouts | grep service"
1026 #define OBD_FAIL_TGT_CONN_RACE 0x701
1027 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
1028 # lctl below may fail, it is valid case
1029 $LCTL --device $mdcdev recover
1032 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1033 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
1036 run_test 44a "race in target handle connect"
1039 local mdcdev=$($LCTL dl |
1040 awk "/${FSNAME}-MDT0000-mdc-/ {if (\$2 == \"UP\") {print \$1}}")
1041 [ "$mdcdev" ] || error "${FSNAME}-MDT0000-mdc not up"
1042 [ $(echo $mdcdev | wc -w) -eq 1 ] ||
1043 { echo mdcdev=$mdcdev; $LCTL dl;
1044 error "more than one ${FSNAME}-MDT0000-mdc"; }
1046 for i in $(seq 1 10); do
1047 echo "$i of 10 ($(date +%s))"
1048 do_facet $SINGLEMDS \
1049 "lctl get_param -n md[ts].*.mdt.timeouts | grep service"
1050 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
1051 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
1052 # lctl below may fail, it is valid case
1053 $LCTL --device $mdcdev recover
1058 run_test 44b "race in target handle connect"
1061 replay_barrier $SINGLEMDS
1062 createmany -m $DIR/$tfile-%d 100 || error "failed to create directories"
1063 #define OBD_FAIL_TGT_RCVG_FLAG 0x712
1064 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000712"
1065 fail_abort $SINGLEMDS
1066 unlinkmany $DIR/$tfile-%d 100 && error "unliked after fail abort"
1068 unlinkmany $DIR/$tfile-%d 100 && error "unliked after fail"
1071 run_test 44c "race in target handle connect"
1073 # Handle failed close
1075 local mdcdev=$($LCTL get_param -n devices |
1076 awk "/ ${FSNAME}-MDT0000-mdc-/ {print \$1}")
1077 [ "$mdcdev" ] || error "${FSNAME}-MDT0000-mdc not up"
1078 [ $(echo $mdcdev | wc -w) -eq 1 ] ||
1079 { echo mdcdev=$mdcdev; $LCTL dl;
1080 error "more than one ${FSNAME}-MDT0000-mdc"; }
1082 $LCTL --device $mdcdev recover ||
1083 error "$LCTL --device $mdcdev recover failed"
1085 multiop_bg_pause $DIR/$tfile O_c ||
1086 error "multiop_bg_pause $DIR/$tfile failed"
1089 # This will cause the CLOSE to fail before even
1090 # allocating a reply buffer
1091 $LCTL --device $mdcdev deactivate ||
1092 error "$LCTL --device $mdcdev deactivate failed"
1095 kill -USR1 $pid || error "multiop $pid not running"
1096 wait $pid || error "multiop $pid failed"
1098 $LCTL --device $mdcdev activate ||
1099 error "$LCTL --device $mdcdev activate failed"
1102 $CHECKSTAT -t file $DIR/$tfile ||
1103 error "$CHECKSTAT $DIR/$tfile attribute check failed"
1106 run_test 45 "Handle failed close"
1110 drop_reply "touch $DIR/$tfile"
1112 # ironically, the previous test, 45, will cause a real forced close,
1113 # so just look for one for this test
1114 dmesg | grep -i "force closing client file handle for $tfile" &&
1115 error "found force closing in dmesg"
1118 run_test 46 "Don't leak file handle after open resend (3325)"
1120 test_47() { # bug 2824
1121 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1123 # create some files to make sure precreate has been done on all
1124 # OSTs. (just in case this test is run independently)
1125 createmany -o $DIR/$tfile 20 ||
1126 error "createmany create $DIR/$tfile failed"
1128 # OBD_FAIL_OST_CREATE_NET 0x204
1130 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1131 client_up || error "client_up failed"
1133 # let the MDS discover the OST failure, attempt to recover, fail
1134 # and recover again.
1135 sleep $((3 * TIMEOUT))
1137 # Without 2824, this createmany would hang
1138 createmany -o $DIR/$tfile 20 ||
1139 error "createmany recraete $DIR/$tfile failed"
1140 unlinkmany $DIR/$tfile 20 || error "unlinkmany $DIR/$tfile failed"
1144 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1147 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1148 [ "$OSTCOUNT" -lt "2" ] && skip_env "needs >= 2 OSTs" && return
1150 replay_barrier $SINGLEMDS
1151 createmany -o $DIR/$tfile 20 ||
1152 error "createmany -o $DIR/$tfile failed"
1153 # OBD_FAIL_OST_EROFS 0x216
1154 facet_failover $SINGLEMDS
1155 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1156 client_up || error "client_up failed"
1158 # let the MDS discover the OST failure, attempt to recover, fail
1159 # and recover again.
1160 sleep $((3 * TIMEOUT))
1162 createmany -o $DIR/$tfile 20 20 ||
1163 error "createmany recraete $DIR/$tfile failed"
1164 unlinkmany $DIR/$tfile 40 || error "unlinkmany $DIR/$tfile failed"
1167 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1170 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $ost1_svc)
1171 local oscdev=$(do_facet $SINGLEMDS "lctl get_param -n devices" |
1172 grep $mdtosc | awk '{print $1}')
1173 [ "$oscdev" ] || error "could not find OSC device on MDS"
1174 do_facet $SINGLEMDS $LCTL --device $oscdev recover ||
1175 error "OSC device $oscdev recovery failed"
1176 do_facet $SINGLEMDS $LCTL --device $oscdev recover ||
1177 error "second OSC device $oscdev recovery failed"
1178 # give the mds_lov_sync threads a chance to run
1181 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1183 # b3764 timed out lock replay
1185 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ] &&
1186 skip "MDS prior to 2.6.90 handle LDLM_REPLY_NET incorrectly" &&
1189 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1190 cancel_lru_locks mdc
1192 multiop_bg_pause $DIR/$tfile s_s || error "multiop $DIR/$tfile failed"
1195 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
1196 lctl set_param -n ldlm.cancel_unused_locks_before_replay "0"
1197 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000157"
1199 fail $SINGLEMDS || error "fail $SINGLEMDS failed"
1201 wait $mpid || error "multiop_bg_pause pid failed"
1203 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1204 lctl set_param fail_loc=0x0
1205 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
1208 run_test 52 "time out lock replay (3764)"
1210 # bug 3462 - simultaneous MDC requests
1212 [[ $(lctl get_param mdc.*.import |
1213 grep "connect_flags:.*multi_mod_rpc") ]] ||
1214 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
1216 cancel_lru_locks mdc # cleanup locks from former test cases
1217 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1218 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1219 multiop $DIR/${tdir}-1/f O_c &
1221 # give multiop a change to open
1224 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1225 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1226 kill -USR1 $close_pid
1227 cancel_lru_locks mdc # force the close
1228 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1230 mcreate $DIR/${tdir}-2/f || error "mcreate $DIR/${tdir}-2/f failed"
1232 # close should still be here
1233 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1235 replay_barrier_nodf $SINGLEMDS
1237 wait $close_pid || error "close_pid $close_pid failed"
1239 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1240 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1241 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1242 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1243 rm -rf $DIR/${tdir}-*
1245 run_test 53a "|X| close request while two MDC requests in flight"
1248 cancel_lru_locks mdc # cleanup locks from former test cases
1250 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1251 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1252 multiop_bg_pause $DIR/${tdir}-1/f O_c ||
1253 error "multiop_bg_pause $DIR/${tdir}-1/f failed"
1256 #define OBD_FAIL_MDS_REINT_NET 0x107
1257 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1258 mcreate $DIR/${tdir}-2/f &
1262 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1263 kill -USR1 $close_pid
1264 cancel_lru_locks mdc # force the close
1265 wait $close_pid || error "close_pid $close_pid failed"
1266 # open should still be here
1267 [ -d /proc/$open_pid ] || error "open_pid doesn't exist"
1269 replay_barrier_nodf $SINGLEMDS
1271 wait $open_pid || error "open_pid failed"
1273 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1274 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1275 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1276 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1277 rm -rf $DIR/${tdir}-*
1279 run_test 53b "|X| open request while two MDC requests in flight"
1282 cancel_lru_locks mdc # cleanup locks from former test cases
1284 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1285 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1286 multiop $DIR/${tdir}-1/f O_c &
1289 #define OBD_FAIL_MDS_REINT_NET 0x107
1290 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1291 mcreate $DIR/${tdir}-2/f &
1295 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1296 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1297 kill -USR1 $close_pid
1298 cancel_lru_locks mdc # force the close
1300 #bz20647: make sure all pids exist before failover
1301 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1302 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1303 replay_barrier_nodf $SINGLEMDS
1304 fail_nodf $SINGLEMDS
1305 wait $open_pid || error "open_pid failed"
1307 # close should be gone
1308 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1309 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1311 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1312 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1313 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1314 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1315 rm -rf $DIR/${tdir}-*
1317 run_test 53c "|X| open request and close request while two MDC requests in flight"
1320 [[ $(lctl get_param mdc.*.import |
1321 grep "connect_flags:.*multi_mod_rpc") ]] ||
1322 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
1324 cancel_lru_locks mdc # cleanup locks from former test cases
1326 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1327 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1328 multiop $DIR/${tdir}-1/f O_c &
1330 # give multiop a chance to open
1333 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1334 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1335 kill -USR1 $close_pid
1336 cancel_lru_locks mdc # force the close
1337 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1338 mcreate $DIR/${tdir}-2/f || error "mcreate $DIR/${tdir}-2/f failed"
1340 # close should still be here
1341 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1343 wait $close_pid || error "close_pid failed"
1345 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1346 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1347 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1348 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1349 rm -rf $DIR/${tdir}-*
1351 run_test 53d "close reply while two MDC requests in flight"
1354 cancel_lru_locks mdc # cleanup locks from former test cases
1356 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1357 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1358 multiop $DIR/${tdir}-1/f O_c &
1361 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1362 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1363 mcreate $DIR/${tdir}-2/f &
1367 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1368 kill -USR1 $close_pid
1369 cancel_lru_locks mdc # force the close
1370 wait $close_pid || error "close_pid failed"
1371 # open should still be here
1372 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1374 replay_barrier_nodf $SINGLEMDS
1376 wait $open_pid || error "open_pid failed"
1378 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1379 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1380 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1381 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1382 rm -rf $DIR/${tdir}-*
1384 run_test 53e "|X| open reply while two MDC requests in flight"
1387 cancel_lru_locks mdc # cleanup locks from former test cases
1389 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1390 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1391 multiop $DIR/${tdir}-1/f O_c &
1394 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1395 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1396 mcreate $DIR/${tdir}-2/f &
1400 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1401 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1402 kill -USR1 $close_pid
1403 cancel_lru_locks mdc # force the close
1405 #bz20647: make sure all pids are exists before failover
1406 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1407 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1408 replay_barrier_nodf $SINGLEMDS
1409 fail_nodf $SINGLEMDS
1410 wait $open_pid || error "open_pid failed"
1412 # close should be gone
1413 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1414 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1416 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1417 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1418 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1419 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1420 rm -rf $DIR/${tdir}-*
1422 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1425 cancel_lru_locks mdc # cleanup locks from former test cases
1427 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1428 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1429 multiop $DIR/${tdir}-1/f O_c &
1432 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1433 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1434 mcreate $DIR/${tdir}-2/f &
1438 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1439 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1440 kill -USR1 $close_pid
1441 cancel_lru_locks mdc # force the close
1442 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1444 #bz20647: make sure all pids are exists before failover
1445 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1446 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1447 replay_barrier_nodf $SINGLEMDS
1448 fail_nodf $SINGLEMDS
1449 wait $open_pid || error "open_pid failed"
1451 # close should be gone
1452 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1454 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1455 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1456 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1457 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1458 rm -rf $DIR/${tdir}-*
1460 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1463 cancel_lru_locks mdc # cleanup locks from former test cases
1465 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1466 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1467 multiop $DIR/${tdir}-1/f O_c &
1470 #define OBD_FAIL_MDS_REINT_NET 0x107
1471 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1472 mcreate $DIR/${tdir}-2/f &
1476 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1477 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1478 kill -USR1 $close_pid
1479 cancel_lru_locks mdc # force the close
1482 #bz20647: make sure all pids are exists before failover
1483 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1484 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1485 replay_barrier_nodf $SINGLEMDS
1486 fail_nodf $SINGLEMDS
1487 wait $open_pid || error "open_pid failed"
1489 # close should be gone
1490 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1491 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1493 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1494 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1495 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1496 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1497 rm -rf $DIR/${tdir}-*
1499 run_test 53h "open request and close reply while two MDC requests in flight"
1501 #b3761 ASSERTION(hash != 0) failed
1503 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1504 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1506 # give touch a chance to run
1508 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1512 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1514 #b3440 ASSERTION(rec->ur_fid2->id) failed
1516 ln -s foo $DIR/$tfile
1517 replay_barrier $SINGLEMDS
1518 #drop_reply "cat $DIR/$tfile"
1522 run_test 56 "don't replay a symlink open request (3440)"
1524 #recovery one mds-ost setattr from llog
1526 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1527 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1528 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1529 replay_barrier $SINGLEMDS
1531 wait_recovery_complete $SINGLEMDS || error "MDS recovery is not done"
1532 wait_mds_ost_sync || error "wait_mds_ost_sync failed"
1533 $CHECKSTAT -t file $DIR/$tfile ||
1534 error "$CHECKSTAT $DIR/$tfile attribute check failed"
1535 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1538 run_test 57 "test recovery from llog for setattr op"
1541 zconf_umount $(hostname) $MOUNT2
1545 #recovery many mds-ost setattr from llog
1547 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1548 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1549 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1550 createmany -o $DIR/$tdir/$tfile-%d 2500
1551 replay_barrier $SINGLEMDS
1554 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null ||
1555 error "$CHECKSTAT $DIR/$tfile-* attribute check failed"
1556 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1557 unlinkmany $DIR/$tdir/$tfile-%d 2500 ||
1558 error "unlinkmany $DIR/$tfile failed"
1561 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
1567 trap cleanup_58 EXIT
1569 large_xattr_enabled &&
1570 orig="$(generate_string $(max_xattr_size))" || orig="bar"
1571 # Original extended attribute can be long. Print a small version of
1572 # attribute if an error occurs
1573 local sm_msg=$(printf "%.9s" $orig)
1575 mount_client $MOUNT2 || error "mount_client on $MOUNT2 failed"
1576 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1577 touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
1578 replay_barrier $SINGLEMDS
1579 setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile
1581 new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
1582 [[ "$new" = "$orig" ]] ||
1583 error "xattr set ($sm_msg...) differs from xattr get ($new)"
1584 rm -f $DIR/$tdir/$tfile
1587 wait_clients_import_state ${CLIENTS:-$HOSTNAME} "mgs" FULL
1589 run_test 58b "test replay of setxattr op"
1591 test_58c() { # bug 16570
1596 trap cleanup_58 EXIT
1598 if large_xattr_enabled; then
1599 local xattr_size=$(max_xattr_size)
1600 orig="$(generate_string $((xattr_size / 2)))"
1601 orig1="$(generate_string $xattr_size)"
1607 # PING_INTERVAL max(obd_timeout / 4, 1U)
1608 sleep $((TIMEOUT / 4))
1610 # Original extended attribute can be long. Print a small version of
1611 # attribute if an error occurs
1612 local sm_msg=$(printf "%.9s" $orig)
1613 local sm_msg1=$(printf "%.9s" $orig1)
1615 mount_client $MOUNT2 || error "mount_client on $MOUNT2 failed"
1616 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1617 touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
1618 drop_request "setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile" ||
1619 error "drop_request for setfattr failed"
1620 new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
1621 [[ "$new" = "$orig" ]] ||
1622 error "xattr set ($sm_msg...) differs from xattr get ($new)"
1623 drop_reint_reply "setfattr -n trusted.foo1 \
1624 -v $orig1 $DIR/$tdir/$tfile" ||
1625 error "drop_reint_reply for setfattr failed"
1626 new=$(get_xattr_value trusted.foo1 $MOUNT2/$tdir/$tfile)
1627 [[ "$new" = "$orig1" ]] ||
1628 error "second xattr set ($sm_msg1...) differs xattr get ($new)"
1629 rm -f $DIR/$tdir/$tfile
1633 run_test 58c "resend/reconstruct setxattr op"
1635 # log_commit_thread vs filter_destroy race used to lead to import use after free
1638 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1640 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1641 createmany -o $DIR/$tdir/$tfile-%d 200 ||
1642 error "createmany create files failed"
1644 unlinkmany $DIR/$tdir/$tfile-%d 200 ||
1645 error "unlinkmany $DIR/$tdir/$tfile failed"
1646 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1647 do_facet ost1 "lctl set_param fail_loc=0x507"
1650 do_facet ost1 "lctl set_param fail_loc=0x0"
1654 run_test 59 "test log_commit_thread vs filter_destroy race"
1656 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1657 # bug 12086: should no oops and No ctxt error for this test
1659 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1660 createmany -o $DIR/$tdir/$tfile-%d 200 ||
1661 error "createmany create files failed"
1662 replay_barrier $SINGLEMDS
1663 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1665 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1666 local no_ctxt=$(dmesg | grep "No ctxt")
1667 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1669 run_test 60 "test llog post recovery init vs llog unlink"
1671 #test race llog recovery thread vs llog cleanup
1672 test_61a() { # was test_61
1673 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1675 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1676 createmany -o $DIR/$tdir/$tfile-%d 800 ||
1677 error "createmany create files failed"
1679 unlinkmany $DIR/$tdir/$tfile-%d 800
1680 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1681 set_nodes_failloc "$(osts_nodes)" 0x80000221
1686 set_nodes_failloc "$(osts_nodes)" 0x0
1688 $CHECKSTAT -t file $DIR/$tdir/$tfile-* &&
1689 error "$CHECKSTAT $DIR/$tdir/$tfile attribute check should fail"
1692 run_test 61a "test race llog recovery vs llog cleanup"
1694 #test race mds llog sync vs llog cleanup
1696 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1697 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
1698 facet_failover $SINGLEMDS
1701 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 ||
1704 run_test 61b "test race mds llog sync vs llog cleanup"
1706 #test race cancel cookie cb vs llog cleanup
1708 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1710 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1711 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1712 set_nodes_failloc "$(osts_nodes)" 0x80000222
1716 set_nodes_failloc "$(osts_nodes)" 0x0
1718 run_test 61c "test race mds llog sync vs llog cleanup"
1720 test_61d() { # bug 16002 # bug 17466 # bug 22137
1721 # OBD_FAIL_OBD_LLOG_SETUP 0x605
1723 do_facet mgs "lctl set_param fail_loc=0x80000605"
1724 start mgs $(mgsdevname) $MGS_MOUNT_OPTS &&
1725 error "mgs start should have failed"
1726 do_facet mgs "lctl set_param fail_loc=0"
1727 start mgs $(mgsdevname) $MGS_MOUNT_OPTS || error "cannot restart mgs"
1729 run_test 61d "error in llog_setup should cleanup the llog context correctly"
1731 test_62() { # Bug 15756 - don't mis-drop resent replay
1732 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1733 replay_barrier $SINGLEMDS
1734 createmany -o $DIR/$tdir/$tfile- 25 ||
1735 error "createmany create files failed"
1736 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1737 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1739 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1740 unlinkmany $DIR/$tdir/$tfile- 25 ||
1741 error "unlinkmany $DIR/$tdir/$tfile failed"
1744 run_test 62 "don't mis-drop resent replay"
1746 #Adaptive Timeouts (bug 3055)
1754 echo "Cleaning up AT ..."
1755 if [ -n "$ATOLDBASE" ]; then
1756 local at_history=$($LCTL get_param -n at_history)
1757 do_facet $SINGLEMDS "lctl set_param at_history=$at_history" || true
1758 do_facet ost1 "lctl set_param at_history=$at_history" || true
1761 if [ $AT_MAX_SET -ne 0 ]; then
1762 for facet in mds client ost; do
1763 var=AT_MAX_SAVE_${facet}
1764 echo restore AT on $facet to saved value ${!var}
1765 at_max_set ${!var} $facet
1766 at_new=$(at_max_get $facet)
1767 echo Restored AT value on $facet $at_new
1768 [ $at_new -eq ${!var} ] ||
1769 error "AT value not restored SAVED ${!var} NEW $at_new"
1776 local at_max_new=600
1778 # Save at_max original values
1780 if [ $AT_MAX_SET -eq 0 ]; then
1781 # Suppose that all osts have the same at_max
1782 for facet in mds client ost; do
1783 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1787 for facet in mds client ost; do
1788 at_max=$(at_max_get $facet)
1789 if [ $at_max -ne $at_max_new ]; then
1790 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1791 at_max_set $at_max_new $facet
1796 if [ -z "$ATOLDBASE" ]; then
1797 ATOLDBASE=$(do_facet $SINGLEMDS "lctl get_param -n at_history")
1798 # speed up the timebase so we can check decreasing AT
1799 do_facet $SINGLEMDS "lctl set_param at_history=8" || true
1800 do_facet ost1 "lctl set_param at_history=8" || true
1802 # sleep for a while to cool down, should be > 8s and also allow
1803 # at least one ping to be sent. simply use TIMEOUT to be safe.
1808 test_65a() #bug 3055
1810 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1812 at_start || return 0
1813 $LCTL dk > /dev/null
1815 $LCTL set_param debug="other"
1816 # Slow down a request to the current service time, this is critical
1817 # because previous tests may have caused this value to increase.
1818 REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
1819 awk '/portal 12/ {print $5}'`
1820 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1822 do_facet $SINGLEMDS lctl set_param fail_val=$((${REQ_DELAY} * 1000))
1823 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1824 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x8000050a
1825 createmany -o $DIR/$tfile 10 > /dev/null
1826 unlinkmany $DIR/$tfile 10 > /dev/null
1827 # check for log message
1828 $LCTL dk | grep "Early reply #" || error "No early reply"
1830 # client should show REQ_DELAY estimates
1831 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1833 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1835 run_test 65a "AT: verify early replies"
1837 test_65b() #bug 3055
1839 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1841 at_start || return 0
1844 $LCTL set_param debug="other trace"
1845 $LCTL dk > /dev/null
1846 # Slow down a request to the current service time, this is critical
1847 # because previous tests may have caused this value to increase.
1848 $SETSTRIPE --stripe-index=0 --stripe-count=1 $DIR/$tfile ||
1849 error "$SETSTRIPE failed for $DIR/$tfile"
1851 multiop $DIR/$tfile Ow1yc
1852 REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
1853 awk '/portal 6/ {print $5}'`
1854 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1856 do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
1857 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1858 do_facet ost1 $LCTL set_param fail_loc=0x224
1861 $SETSTRIPE --stripe-index=0 --stripe-count=1 $DIR/$tfile ||
1862 error "$SETSTRIPE failed"
1863 # force some real bulk transfer
1864 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1866 do_facet ost1 $LCTL set_param fail_loc=0
1867 # check for log message
1868 $LCTL dk | grep "Early reply #" || error "No early reply"
1870 # client should show REQ_DELAY estimates
1871 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1873 run_test 65b "AT: verify early replies on packed reply / bulk"
1875 test_66a() #bug 3055
1877 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1879 at_start || return 0
1880 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1881 # adjust 5s at a time so no early reply is sent (within deadline)
1882 do_facet $SINGLEMDS "$LCTL set_param fail_val=5000"
1883 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1884 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x8000050a"
1885 createmany -o $DIR/$tfile 20 > /dev/null
1886 unlinkmany $DIR/$tfile 20 > /dev/null
1887 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1888 do_facet $SINGLEMDS "$LCTL set_param fail_val=10000"
1889 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x8000050a"
1890 createmany -o $DIR/$tfile 20 > /dev/null
1891 unlinkmany $DIR/$tfile 20 > /dev/null
1892 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1893 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0"
1895 createmany -o $DIR/$tfile 20 > /dev/null
1896 unlinkmany $DIR/$tfile 20 > /dev/null
1897 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1898 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1899 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1900 echo "Current MDT timeout $CUR, worst $WORST"
1901 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1903 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1905 test_66b() #bug 3055
1907 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1909 at_start || return 0
1910 ORIG=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1911 awk '/network/ {print $4}')
1912 $LCTL set_param fail_val=$(($ORIG + 5))
1913 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1914 $LCTL set_param fail_loc=0x50c
1915 touch $DIR/$tfile > /dev/null 2>&1
1916 $LCTL set_param fail_loc=0
1917 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1918 awk '/network/ {print $4}')
1919 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1920 awk '/network/ {print $6}')
1921 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1922 [ $WORST -gt $ORIG ] ||
1923 error "Worst $WORST should be worse than orig $ORIG"
1925 run_test 66b "AT: verify net latency adjusts"
1927 test_67a() #bug 3055
1929 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1931 at_start || return 0
1932 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1933 # sleeping threads may drive values above this
1934 do_facet ost1 "$LCTL set_param fail_val=400"
1935 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1936 do_facet ost1 "$LCTL set_param fail_loc=0x50a"
1937 createmany -o $DIR/$tfile 20 > /dev/null
1938 unlinkmany $DIR/$tfile 20 > /dev/null
1939 do_facet ost1 "$LCTL set_param fail_loc=0"
1940 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1941 ATTEMPTS=$(($CONN2 - $CONN1))
1942 echo "$ATTEMPTS osc reconnect attempts on gradual slow"
1943 [ $ATTEMPTS -gt 0 ] &&
1944 error_ignore bz13721 "AT should have prevented reconnect"
1947 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1949 test_67b() #bug 3055
1951 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1953 at_start || return 0
1954 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1956 # exhaust precreations on ost1
1957 local OST=$(ostname_from_index 0)
1958 local mdtosc=$(get_mdtosc_proc_path mds $OST)
1959 local last_id=$(do_facet $SINGLEMDS lctl get_param -n \
1960 osc.$mdtosc.prealloc_last_id)
1961 local next_id=$(do_facet $SINGLEMDS lctl get_param -n \
1962 osc.$mdtosc.prealloc_next_id)
1964 mkdir -p $DIR/$tdir/${OST} || error "mkdir $DIR/$tdir/${OST} failed"
1965 $SETSTRIPE -i 0 -c 1 $DIR/$tdir/${OST} || error "$SETSTRIPE failed"
1966 echo "Creating to objid $last_id on ost $OST..."
1967 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1968 do_facet ost1 "$LCTL set_param fail_val=20000"
1969 do_facet ost1 "$LCTL set_param fail_loc=0x80000223"
1970 createmany -o $DIR/$tdir/${OST}/f $next_id $((last_id - next_id + 2))
1973 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1975 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1976 ATTEMPTS=$(($CONN2 - $CONN1))
1977 echo "$ATTEMPTS osc reconnect attempts on instant slow"
1978 # do it again; should not timeout
1979 do_facet ost1 "$LCTL set_param fail_loc=0x80000223"
1980 cp /etc/profile $DIR/$tfile || error "cp failed"
1981 do_facet ost1 "$LCTL set_param fail_loc=0"
1983 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1984 CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1985 ATTEMPTS=$(($CONN3 - $CONN2))
1986 echo "$ATTEMPTS osc reconnect attempts on 2nd slow"
1987 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1990 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1992 test_68 () #bug 13813
1994 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1996 at_start || return 0
1997 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1998 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1999 local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
2000 [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
2001 local ENQ_MIN=$(cat $ldlm_enqueue_min)
2002 local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
2003 echo $TIMEOUT >> $ldlm_enqueue_min
2004 do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
2006 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2007 $SETSTRIPE --stripe-index=0 -c 1 $DIR/$tdir ||
2008 error "$SETSTRIPE failed for $DIR/$tdir"
2009 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
2010 $LCTL set_param fail_val=$(($TIMEOUT - 1))
2011 $LCTL set_param fail_loc=0x80000312
2012 cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
2013 $LCTL set_param fail_val=$((TIMEOUT * 5 / 4))
2014 $LCTL set_param fail_loc=0x80000312
2015 cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
2016 $LCTL set_param fail_loc=0
2018 echo $ENQ_MIN >> $ldlm_enqueue_min
2019 do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
2023 run_test 68 "AT: verify slowing locks"
2026 # end of AT tests includes above lines
2028 # start multi-client tests
2030 [ -z "$CLIENTS" ] &&
2031 { skip "Need two or more clients." && return; }
2032 [ $CLIENTCOUNT -lt 2 ] &&
2033 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
2035 echo "mount clients $CLIENTS ..."
2036 zconf_mount_clients $CLIENTS $MOUNT
2038 local clients=${CLIENTS//,/ }
2039 echo "Write/read files on $DIR ; clients $CLIENTS ... "
2040 for CLIENT in $clients; do
2041 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
2042 of=$DIR/${tfile}_${CLIENT} 2>/dev/null ||
2043 error "dd failed on $CLIENT"
2046 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
2047 for C in ${CLIENTS//,/ }; do
2048 do_node $prev_client dd if=$DIR/${tfile}_${C} \
2049 of=/dev/null 2>/dev/null ||
2050 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
2056 run_test 70a "check multi client t-f"
2058 check_for_process () {
2063 killall_process $clients "$prog" -0
2067 local clients=${CLIENTS:-$HOSTNAME}
2069 zconf_mount_clients $clients $MOUNT
2072 [ "$SLOW" = "no" ] && duration=120
2073 # set duration to 900 because it takes some time to boot node
2074 [ "$FAILURE_MODE" = HARD ] && duration=900
2077 local start_ts=$(date +%s)
2078 local cmd="rundbench 1 -t $duration"
2080 if [ $MDSCOUNT -ge 2 ]; then
2081 test_mkdir -p -c$MDSCOUNT $DIR/$tdir
2082 $LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir
2084 do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
2085 PATH=\$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
2086 DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
2087 MOUNT=$MOUNT DIR=$DIR/$tdir/\\\$(hostname) LCTL=$LCTL $cmd" &
2090 #LU-1897 wait for all dbench copies to start
2091 while ! check_for_process $clients dbench; do
2092 elapsed=$(($(date +%s) - start_ts))
2093 if [ $elapsed -gt $duration ]; then
2094 killall_process $clients dbench
2095 error "dbench failed to start on $clients!"
2100 log "Started rundbench load pid=$pid ..."
2102 elapsed=$(($(date +%s) - start_ts))
2103 local num_failovers=0
2105 while [ $elapsed -lt $duration ]; do
2106 if ! check_for_process $clients dbench; then
2107 error_noexit "dbench stopped on some of $clients!"
2108 killall_process $clients dbench
2112 replay_barrier mds$fail_index
2113 sleep 1 # give clients a time to do operations
2114 # Increment the number of failovers
2115 num_failovers=$((num_failovers+1))
2116 log "$TESTNAME fail mds$fail_index $num_failovers times"
2118 elapsed=$(($(date +%s) - start_ts))
2119 if [ $fail_index -ge $MDSCOUNT ]; then
2122 fail_index=$((fail_index+1))
2126 wait $pid || error "rundbench load on $clients failed!"
2128 run_test 70b "dbench ${MDSCOUNT}mdts recovery; $CLIENTCOUNT clients"
2129 # end multi-client tests
2134 local monitor_pid=$3
2136 local start_ts=$(date +%s)
2137 local num_failovers=0
2140 elapsed=$(($(date +%s) - start_ts))
2141 while [ $elapsed -lt $duration ]; do
2142 fail_index=$((RANDOM%max_index+1))
2143 kill -0 $monitor_pid ||
2144 error "$monitor_pid stopped"
2146 replay_barrier mds$fail_index
2148 # Increment the number of failovers
2149 num_failovers=$((num_failovers+1))
2150 log "$TESTNAME fail mds$fail_index $num_failovers times"
2152 elapsed=$(($(date +%s) - start_ts))
2158 rm -f $DIR/replay-single.70c.lck
2163 local clients=${CLIENTS:-$HOSTNAME}
2166 zconf_mount_clients $clients $MOUNT
2169 [ "$SLOW" = "no" ] && duration=180
2170 # set duration to 900 because it takes some time to boot node
2171 [ "$FAILURE_MODE" = HARD ] && duration=600
2174 local start_ts=$(date +%s)
2176 trap cleanup_70c EXIT
2178 while [ ! -e $DIR/replay-single.70c.lck ]; do
2179 test_mkdir -p -c$MDSCOUNT $DIR/$tdir || break
2180 if [ $MDSCOUNT -ge 2 ]; then
2181 $LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir ||
2182 error "set default dirstripe failed"
2184 cd $DIR/$tdir || break
2185 tar cf - /etc | tar xf - || error "tar failed in loop"
2189 echo "Started tar $tar_70c_pid"
2191 random_fail_mdt $MDSCOUNT $duration $tar_70c_pid
2192 kill -0 $tar_70c_pid || error "tar $tar_70c_pid stopped"
2194 touch $DIR/replay-single.70c.lck
2195 wait $tar_70c_pid || error "$?: tar failed"
2200 run_test 70c "tar ${MDSCOUNT}mdts recovery"
2204 kill -9 $mkdir_70d_pid
2208 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2209 local clients=${CLIENTS:-$HOSTNAME}
2212 zconf_mount_clients $clients $MOUNT
2215 [ "$SLOW" = "no" ] && duration=180
2216 # set duration to 900 because it takes some time to boot node
2217 [ "$FAILURE_MODE" = HARD ] && duration=900
2222 local start_ts=$(date +%s)
2224 trap cleanup_70d EXIT
2227 $LFS mkdir -i0 -c2 $DIR/$tdir/test || {
2231 $LFS mkdir -i1 -c2 $DIR/$tdir/test1 || {
2236 touch $DIR/$tdir/test/a || {
2240 mkdir $DIR/$tdir/test/b || {
2244 rm -rf $DIR/$tdir/test || {
2249 touch $DIR/$tdir/test1/a || {
2253 mkdir $DIR/$tdir/test1/b || {
2258 rm -rf $DIR/$tdir/test1 || {
2265 echo "Started $mkdir_70d_pid"
2267 random_fail_mdt $MDSCOUNT $duration $mkdir_70d_pid
2268 kill -0 $mkdir_70d_pid || error "mkdir/rmdir $mkdir_70d_pid stopped"
2273 run_test 70d "mkdir/rmdir striped dir ${MDSCOUNT}mdts recovery"
2276 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2277 local clients=${CLIENTS:-$HOSTNAME}
2280 lctl set_param debug=+ha
2281 zconf_mount_clients $clients $MOUNT
2284 [ "$SLOW" = "no" ] && duration=180
2285 # set duration to 900 because it takes some time to boot node
2286 [ "$FAILURE_MODE" = HARD ] && duration=900
2289 $LFS mkdir -i0 $DIR/$tdir/test_0
2290 $LFS mkdir -i0 $DIR/$tdir/test_1
2291 touch $DIR/$tdir/test_0/a
2292 touch $DIR/$tdir/test_1/b
2295 mrename $DIR/$tdir/test_0/a $DIR/$tdir/test_1/b > /dev/null || {
2300 checkstat $DIR/$tdir/test_0/a && {
2301 echo "a still exists"
2305 checkstat $DIR/$tdir/test_1/b || {
2306 echo "b still exists"
2310 touch $DIR/$tdir/test_0/a || {
2311 echo "touch a fails"
2315 mrename $DIR/$tdir/test_1/b $DIR/$tdir/test_0/a > /dev/null || {
2322 stack_trap "kill -9 $rename_70e_pid" EXIT
2323 echo "Started PID=$rename_70e_pid"
2325 random_fail_mdt 2 $duration $rename_70e_pid
2326 kill -0 $rename_70e_pid || error "rename $rename_70e_pid stopped"
2328 run_test 70e "rename cross-MDT with random fails"
2330 test_70f_write_and_read(){
2335 echo "Write/read files in: '$DIR/$tdir', clients: '$CLIENTS' ..."
2336 for client in ${CLIENTS//,/ }; do
2337 [ -f $stopflag ] || return
2339 local tgtfile=$DIR/$tdir/$tfile.$client
2340 do_node $client dd $DD_OPTS bs=1M count=10 if=$srcfile \
2341 of=$tgtfile 2>/dev/null ||
2342 error "dd $DD_OPTS bs=1M count=10 if=$srcfile " \
2343 "of=$tgtfile failed on $client, rc=$?"
2346 local prev_client=$(echo ${CLIENTS//,/ } | awk '{ print $NF }')
2349 for client in ${CLIENTS//,/ }; do
2350 [ -f $stopflag ] || return
2352 # flush client cache in case test is running on only one client
2353 # do_node $client cancel_lru_locks osc
2354 do_node $client $LCTL set_param ldlm.namespaces.*.lru_size=clear
2356 tgtfile=$DIR/$tdir/$tfile.$client
2357 local md5=$(do_node $prev_client "md5sum $tgtfile")
2358 [ ${checksum[$index]// */} = ${md5// */} ] ||
2359 error "$tgtfile: checksum doesn't match on $prev_client"
2360 index=$((index + 1))
2370 mkdir -p $DIR/$tdir || error "cannot create $DIR/$tdir directory"
2371 $SETSTRIPE -c -1 $DIR/$tdir || error "cannot $SETSTRIPE $DIR/$tdir"
2374 while [ -f $stopflag ]; do
2375 test_70f_write_and_read $srcfile $stopflag
2376 # use direct IO and buffer cache in turns if loop
2377 [ -n "$DD_OPTS" ] && DD_OPTS="" || DD_OPTS="oflag=direct"
2381 test_70f_cleanup() {
2383 rm -f $TMP/$tfile.stop
2384 do_nodes $CLIENTS rm -f $TMP/$tfile
2385 rm -f $DIR/$tdir/$tfile.*
2389 # [ x$ost1failover_HOST = x$ost_HOST ] &&
2390 # { skip "Failover host not defined" && return; }
2391 # [ -z "$CLIENTS" ] &&
2392 # { skip "CLIENTS are not specified." && return; }
2393 # [ $CLIENTCOUNT -lt 2 ] &&
2394 # { skip "Need 2 or more clients, have $CLIENTCOUNT" && return; }
2396 [[ $(lustre_version_code ost1) -lt $(version_code 2.9.53) ]] &&
2397 skip "Need server version at least 2.9.53" && return
2399 echo "mount clients $CLIENTS ..."
2400 zconf_mount_clients $CLIENTS $MOUNT
2402 local srcfile=$TMP/$tfile
2406 trap test_70f_cleanup EXIT
2407 # create a different source file local to each client node so we can
2408 # detect if the file wasn't written out properly after failover
2409 do_nodes $CLIENTS dd bs=1M count=10 if=/dev/urandom of=$srcfile \
2410 2>/dev/null || error "can't create $srcfile on $CLIENTS"
2411 for client in ${CLIENTS//,/ }; do
2412 checksum[$index]=$(do_node $client "md5sum $srcfile")
2413 index=$((index + 1))
2417 [ "$SLOW" = "no" ] && duration=60
2418 # set duration to 900 because it takes some time to boot node
2419 [ "$FAILURE_MODE" = HARD ] && duration=900
2421 local stopflag=$TMP/$tfile.stop
2422 test_70f_loop $srcfile $stopflag &
2426 local num_failovers=0
2427 local start_ts=$SECONDS
2428 while [ $elapsed -lt $duration ]; do
2432 num_failovers=$((num_failovers + 1))
2433 log "$TESTNAME failing OST $num_failovers times"
2436 elapsed=$((SECONDS - start_ts))
2443 run_test 70f "OSS O_DIRECT recovery with $CLIENTCOUNT clients"
2447 kill -9 $mkdir_71a_pid
2450 random_double_fail_mdt() {
2453 local monitor_pid=$3
2455 local start_ts=$(date +%s)
2456 local num_failovers=0
2460 elapsed=$(($(date +%s) - start_ts))
2461 while [ $elapsed -lt $duration ]; do
2462 fail_index=$((RANDOM%max_index + 1))
2463 if [ $fail_index -eq $max_index ]; then
2466 second_index=$((fail_index + 1))
2468 kill -0 $monitor_pid ||
2469 error "$monitor_pid stopped"
2471 replay_barrier mds$fail_index
2472 replay_barrier mds$second_index
2474 # Increment the number of failovers
2475 num_failovers=$((num_failovers+1))
2476 log "fail mds$fail_index mds$second_index $num_failovers times"
2477 fail mds${fail_index},mds${second_index}
2478 elapsed=$(($(date +%s) - start_ts))
2483 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2484 local clients=${CLIENTS:-$HOSTNAME}
2487 zconf_mount_clients $clients $MOUNT
2490 [ "$SLOW" = "no" ] && duration=180
2491 # set duration to 900 because it takes some time to boot node
2492 [ "$FAILURE_MODE" = HARD ] && duration=900
2497 local start_ts=$(date +%s)
2499 trap cleanup_71a EXIT
2502 $LFS mkdir -i0 -c2 $DIR/$tdir/test
2503 rmdir $DIR/$tdir/test
2507 echo "Started $mkdir_71a_pid"
2509 random_double_fail_mdt 2 $duration $mkdir_71a_pid
2510 kill -0 $mkdir_71a_pid || error "mkdir/rmdir $mkdir_71a_pid stopped"
2515 run_test 71a "mkdir/rmdir striped dir with 2 mdts recovery"
2518 multiop_bg_pause $DIR/$tfile O_tSc ||
2519 error "multiop_bg_pause $DIR/$tfile failed"
2523 replay_barrier $SINGLEMDS
2524 #define OBD_FAIL_LDLM_ENQUEUE_NET 0x302
2525 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
2528 wait $pid || error "multiop pid failed"
2529 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
2532 run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay, close"
2535 multiop_bg_pause $DIR/$tfile O_tSc ||
2536 error "multiop_bg_pause $DIR/$tfile failed"
2540 replay_barrier $SINGLEMDS
2541 #define OBD_FAIL_LDLM_REPLY 0x30c
2542 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
2545 wait $pid || error "multiop pid failed"
2546 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
2549 run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
2553 local clients=${CLIENTS:-$HOSTNAME}
2555 zconf_umount_clients $clients $MOUNT
2557 facet_failover $SINGLEMDS
2558 zconf_mount_clients $clients $MOUNT
2560 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
2561 rm $DIR/$tfile || error "rm $DIR/$tfile failed"
2562 clients_up || error "client evicted: $?"
2565 run_test 74 "Ensure applications don't fail waiting for OST recovery"
2567 remote_dir_check_80() {
2572 diridx=$($LFS getstripe -m $remote_dir) ||
2573 error "$LFS getstripe -m $remote_dir failed"
2574 [ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
2576 createmany -o $remote_dir/f-%d 20 || error "creation failed"
2577 fileidx=$($LFS getstripe -m $remote_dir/f-1) ||
2578 error "$LFS getstripe -m $remote_dir/f-1 failed"
2579 [ $fileidx -eq $mdtidx ] || error "$fileidx != $mdtidx"
2585 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2586 ([ $FAILURE_MODE == "HARD" ] &&
2587 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2588 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2592 local remote_dir=$DIR/$tdir/remote_dir
2594 mkdir -p $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2595 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2596 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2597 $LFS mkdir -i $MDTIDX $remote_dir &
2603 wait $CLIENT_PID || error "remote creation failed"
2605 remote_dir_check_80 || error "remote dir check failed"
2606 rm -rf $DIR/$tdir || error "rmdir failed"
2610 run_test 80a "DNE: create remote dir, drop update rep from MDT0, fail MDT0"
2613 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2614 ([ $FAILURE_MODE == "HARD" ] &&
2615 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2616 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2620 local remote_dir=$DIR/$tdir/remote_dir
2622 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2623 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2624 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2625 $LFS mkdir -i $MDTIDX $remote_dir &
2630 fail mds$((MDTIDX + 1))
2632 wait $CLIENT_PID || error "remote creation failed"
2634 remote_dir_check_80 || error "remote dir check failed"
2635 rm -rf $DIR/$tdir || error "rmdir failed"
2639 run_test 80b "DNE: create remote dir, drop update rep from MDT0, fail MDT1"
2642 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2643 ([ $FAILURE_MODE == "HARD" ] &&
2644 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2645 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2649 local remote_dir=$DIR/$tdir/remote_dir
2651 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2652 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2653 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2654 $LFS mkdir -i $MDTIDX $remote_dir &
2660 fail mds$((MDTIDX + 1))
2662 wait $CLIENT_PID || error "remote creation failed"
2664 remote_dir_check_80 || error "remote dir check failed"
2665 rm -rf $DIR/$tdir || error "rmdir failed"
2669 run_test 80c "DNE: create remote dir, drop update rep from MDT1, fail MDT[0,1]"
2672 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2674 local remote_dir=$DIR/$tdir/remote_dir
2676 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2677 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2678 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2679 $LFS mkdir -i $MDTIDX $remote_dir &
2682 # sleep 3 seconds to make sure MDTs are failed after
2683 # lfs mkdir -i has finished on all of MDTs.
2688 fail mds${MDTIDX},mds$((MDTIDX + 1))
2690 wait $CLIENT_PID || error "remote creation failed"
2692 remote_dir_check_80 || error "remote dir check failed"
2693 rm -rf $DIR/$tdir || error "rmdir failed"
2697 run_test 80d "DNE: create remote dir, drop update rep from MDT1, fail 2 MDTs"
2700 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2701 ([ $FAILURE_MODE == "HARD" ] &&
2702 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2703 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2707 local remote_dir=$DIR/$tdir/remote_dir
2709 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2710 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2711 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2712 $LFS mkdir -i $MDTIDX $remote_dir &
2715 # sleep 3 seconds to make sure MDTs are failed after
2716 # lfs mkdir -i has finished on all of MDTs.
2722 wait $CLIENT_PID || error "remote creation failed"
2724 remote_dir_check_80 || error "remote dir check failed"
2725 rm -rf $DIR/$tdir || error "rmdir failed"
2729 run_test 80e "DNE: create remote dir, drop MDT1 rep, fail MDT0"
2732 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2733 ([ $FAILURE_MODE == "HARD" ] &&
2734 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2735 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2738 local remote_dir=$DIR/$tdir/remote_dir
2740 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2741 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2742 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2743 $LFS mkdir -i $MDTIDX $remote_dir &
2747 fail mds$((MDTIDX + 1))
2749 wait $CLIENT_PID || error "remote creation failed"
2751 remote_dir_check_80 || error "remote dir check failed"
2752 rm -rf $DIR/$tdir || error "rmdir failed"
2756 run_test 80f "DNE: create remote dir, drop MDT1 rep, fail MDT1"
2759 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2760 ([ $FAILURE_MODE == "HARD" ] &&
2761 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2762 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2766 local remote_dir=$DIR/$tdir/remote_dir
2768 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2769 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2770 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2771 $LFS mkdir -i $MDTIDX $remote_dir &
2774 # sleep 3 seconds to make sure MDTs are failed after
2775 # lfs mkdir -i has finished on all of MDTs.
2781 fail mds$((MDTIDX + 1))
2783 wait $CLIENT_PID || error "remote creation failed"
2785 remote_dir_check_80 || error "remote dir check failed"
2786 rm -rf $DIR/$tdir || error "rmdir failed"
2790 run_test 80g "DNE: create remote dir, drop MDT1 rep, fail MDT0, then MDT1"
2793 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2795 local remote_dir=$DIR/$tdir/remote_dir
2797 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2798 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2799 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2800 $LFS mkdir -i $MDTIDX $remote_dir &
2803 # sleep 3 seconds to make sure MDTs are failed after
2804 # lfs mkdir -i has finished on all of MDTs.
2809 fail mds${MDTIDX},mds$((MDTIDX + 1))
2811 wait $CLIENT_PID || error "remote dir creation failed"
2813 remote_dir_check_80 || error "remote dir check failed"
2814 rm -rf $DIR/$tdir || error "rmdir failed"
2818 run_test 80h "DNE: create remote dir, drop MDT1 rep, fail 2 MDTs"
2821 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2822 ([ $FAILURE_MODE == "HARD" ] &&
2823 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2824 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2828 local remote_dir=$DIR/$tdir/remote_dir
2830 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2831 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2833 touch $remote_dir || error "touch $remote_dir failed"
2834 # OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2835 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2840 fail mds$((MDTIDX + 1))
2842 wait $CLIENT_PID || error "rm remote dir failed"
2844 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2846 rm -rf $DIR/$tdir || error "rmdir failed"
2850 run_test 81a "DNE: unlink remote dir, drop MDT0 update rep, fail MDT1"
2853 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2854 ([ $FAILURE_MODE == "HARD" ] &&
2855 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2856 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2859 local remote_dir=$DIR/$tdir/remote_dir
2861 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2862 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2864 # OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2865 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2872 wait $CLIENT_PID || error "rm remote dir failed"
2874 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2876 rm -rf $DIR/$tdir || error "rmdir failed"
2880 run_test 81b "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0"
2883 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2884 ([ $FAILURE_MODE == "HARD" ] &&
2885 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2886 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2890 local remote_dir=$DIR/$tdir/remote_dir
2892 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2893 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2895 # OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2896 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2903 fail mds$((MDTIDX + 1))
2905 wait $CLIENT_PID || error "rm remote dir failed"
2907 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2909 rm -rf $DIR/$tdir || error "rmdir failed"
2913 run_test 81c "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0,MDT1"
2916 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2918 local remote_dir=$DIR/$tdir/remote_dir
2920 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2921 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2923 # OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2924 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2930 fail mds${MDTIDX},mds$((MDTIDX + 1))
2932 wait $CLIENT_PID || error "rm remote dir failed"
2934 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2936 rm -rf $DIR/$tdir || error "rmdir failed"
2940 run_test 81d "DNE: unlink remote dir, drop MDT0 update reply, fail 2 MDTs"
2943 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2944 ([ $FAILURE_MODE == "HARD" ] &&
2945 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2946 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2950 local remote_dir=$DIR/$tdir/remote_dir
2952 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2953 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2955 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2956 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2959 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
2964 wait $CLIENT_PID || error "rm remote dir failed"
2966 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2968 rm -rf $DIR/$tdir || error "rmdir failed"
2972 run_test 81e "DNE: unlink remote dir, drop MDT1 req reply, fail MDT0"
2975 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2976 ([ $FAILURE_MODE == "HARD" ] &&
2977 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2978 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2982 local remote_dir=$DIR/$tdir/remote_dir
2984 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2985 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2987 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2988 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2993 fail mds$((MDTIDX + 1))
2995 wait $CLIENT_PID || error "rm remote dir failed"
2997 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2999 rm -rf $DIR/$tdir || error "rmdir failed"
3003 run_test 81f "DNE: unlink remote dir, drop MDT1 req reply, fail MDT1"
3006 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3007 ([ $FAILURE_MODE == "HARD" ] &&
3008 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3009 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3013 local remote_dir=$DIR/$tdir/remote_dir
3015 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3016 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
3018 # OBD_FAIL_MDS_REINT_NET_REP 0x119
3019 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
3026 fail mds$((MDTIDX + 1))
3028 wait $CLIENT_PID || error "rm remote dir failed"
3030 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
3032 rm -rf $DIR/$tdir || error "rmdir failed"
3036 run_test 81g "DNE: unlink remote dir, drop req reply, fail M0, then M1"
3039 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3041 local remote_dir=$DIR/$tdir/remote_dir
3043 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3044 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
3046 # OBD_FAIL_MDS_REINT_NET_REP 0x119
3047 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
3053 fail mds${MDTIDX},mds$((MDTIDX + 1))
3055 wait $CLIENT_PID || error "rm remote dir failed"
3057 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
3059 rm -rf $DIR/$tdir || error "rmdir failed"
3063 run_test 81h "DNE: unlink remote dir, drop request reply, fail 2 MDTs"
3066 #define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x144
3067 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000144"
3068 createmany -o $DIR/$tfile- 1 &
3072 client_up || client_up || true # reconnect
3074 run_test 84a "stale open during export disconnect"
3076 test_85a() { #bug 16774
3077 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
3079 for i in $(seq 100); do
3080 echo "tag-$i" > $DIR/$tfile-$i
3081 grep -q "tag-$i" $DIR/$tfile-$i || error "f2-$i"
3084 lov_id=$(lctl dl | grep "clilov")
3085 addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $NF}')
3086 count=$(lctl get_param -n \
3087 ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
3088 echo "before recovery: unused locks count = $count"
3092 count2=$(lctl get_param -n \
3093 ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
3094 echo "after recovery: unused locks count = $count2"
3096 if [ $count2 -ge $count ]; then
3097 error "unused locks are not canceled"
3100 run_test 85a "check the cancellation of unused locks during recovery(IBITS)"
3102 test_85b() { #bug 16774
3103 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
3105 if ! combined_mgs_mds ; then
3109 create_pool $FSNAME.$TESTNAME ||
3110 error "unable to create pool $TESTNAME"
3111 do_facet mgs $LCTL pool_add $FSNAME.$TESTNAME $FSNAME-OST0000 ||
3112 error "unable to add pool $TESTNAME"
3114 $SETSTRIPE -c 1 -p $FSNAME.$TESTNAME $DIR
3116 for i in $(seq 100); do
3117 dd if=/dev/urandom of=$DIR/$tfile-$i bs=4096 \
3118 count=32 >/dev/null 2>&1
3121 cancel_lru_locks osc
3123 for i in $(seq 100); do
3124 dd if=$DIR/$tfile-$i of=/dev/null bs=4096 \
3125 count=32 >/dev/null 2>&1
3128 lov_id=$(lctl dl | grep "clilov")
3129 addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $NF}')
3130 count=$(lctl get_param -n \
3131 ldlm.namespaces.*OST0000*$addr.lock_unused_count)
3132 echo "before recovery: unused locks count = $count"
3133 [ $count -ne 0 ] || error "unused locks ($count) should be zero"
3137 count2=$(lctl get_param \
3138 -n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
3139 echo "after recovery: unused locks count = $count2"
3141 do_facet mgs $LCTL pool_remove $FSNAME.$TESTNAME $FSNAME-OST0000 ||
3142 error "unable to remove pool $TESTNAME"
3143 do_facet mgs $LCTL pool_destroy $FSNAME.$TESTNAME ||
3144 error "unable to destroy the pool $TESTNAME"
3146 if ! combined_mgs_mds ; then
3150 if [ $count2 -ge $count ]; then
3151 error "unused locks are not canceled"
3154 run_test 85b "check the cancellation of unused locks during recovery(EXTENT)"
3157 local clients=${CLIENTS:-$HOSTNAME}
3159 zconf_umount_clients $clients $MOUNT
3160 do_facet $SINGLEMDS lctl set_param mdt.${FSNAME}-MDT*.exports.clear=0
3161 remount_facet $SINGLEMDS
3162 zconf_mount_clients $clients $MOUNT
3164 run_test 86 "umount server after clear nid_stats should not hit LBUG"
3167 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
3170 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
3171 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
3172 error "dd to $DIR/$tfile failed"
3173 cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
3174 cancel_lru_locks osc
3176 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
3177 cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
3178 if [ $cksum != $cksum2 ] ; then
3179 error "New checksum $cksum2 does not match original $cksum"
3182 run_test 87a "write replay"
3185 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
3188 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
3189 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
3190 error "dd to $DIR/$tfile failed"
3191 sleep 1 # Give it a chance to flush dirty data
3192 echo TESTTEST | dd of=$DIR/$tfile bs=1 count=8 seek=64
3193 cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
3194 cancel_lru_locks osc
3196 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
3197 cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
3198 if [ $cksum != $cksum2 ] ; then
3199 error "New checksum $cksum2 does not match original $cksum"
3202 run_test 87b "write replay with changed data (checksum resend)"
3204 test_88() { #bug 17485
3205 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3206 mkdir -p $TMP/$tdir || error "mkdir $TMP/$tdir failed"
3208 $SETSTRIPE -i 0 -c 1 $DIR/$tdir || error "$SETSTRIPE"
3211 replay_barrier $SINGLEMDS
3213 # exhaust precreations on ost1
3214 local OST=$(ostname_from_index 0)
3215 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $OST)
3216 local last_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
3217 local next_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
3218 echo "before test: last_id = $last_id, next_id = $next_id"
3220 echo "Creating to objid $last_id on ost $OST..."
3221 createmany -o $DIR/$tdir/f-%d $next_id $((last_id - next_id + 2)) ||
3222 error "createmany create files to last_id failed"
3224 #create some files to use some uncommitted objids
3225 last_id=$(($last_id + 1))
3226 createmany -o $DIR/$tdir/f-%d $last_id 8 ||
3227 error "createmany create files with uncommitted objids failed"
3229 last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
3230 next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
3231 echo "before recovery: last_id = $last_id2, next_id = $next_id2"
3233 # if test uses shutdown_facet && reboot_facet instead of facet_failover ()
3234 # it has to take care about the affected facets, bug20407
3235 local affected_mds1=$(affected_facets mds1)
3236 local affected_ost1=$(affected_facets ost1)
3238 shutdown_facet $SINGLEMDS
3241 reboot_facet $SINGLEMDS
3242 change_active $affected_mds1
3243 wait_for_facet $affected_mds1
3244 mount_facets $affected_mds1 || error "Restart of mds failed"
3247 change_active $affected_ost1
3248 wait_for_facet $affected_ost1
3249 mount_facets $affected_ost1 || error "Restart of ost1 failed"
3253 last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
3254 next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
3255 echo "after recovery: last_id = $last_id2, next_id = $next_id2"
3257 # create new files, which should use new objids, and ensure the orphan
3258 # cleanup phase for ost1 is completed at the same time
3259 for i in $(seq 8); do
3260 file_id=$(($last_id + 10 + $i))
3261 dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
3264 # if the objids were not recreated, then "ls" will fail with -ENOENT
3265 ls -l $DIR/$tdir/* || error "can't get the status of precreated files"
3268 # write into previously created files
3269 for i in $(seq 8); do
3270 file_id=$(($last_id + $i))
3271 dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
3272 cp -f $DIR/$tdir/f-$file_id $TMP/$tdir/
3275 # compare the content
3276 for i in $(seq 8); do
3277 file_id=$(($last_id + $i))
3278 cmp $TMP/$tdir/f-$file_id $DIR/$tdir/f-$file_id ||
3279 error "the content of file is modified!"
3284 run_test 88 "MDS should not assign same objid to different files "
3287 cancel_lru_locks osc
3288 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3289 rm -f $DIR/$tdir/$tfile
3290 wait_mds_ost_sync || error "initial MDS-OST sync timed out"
3291 wait_delete_completed || error "initial wait delete timed out"
3292 local blocks1=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
3293 local write_size=$(fs_log_size)
3295 $SETSTRIPE -i 0 -c 1 $DIR/$tdir/$tfile
3296 [ $write_size -lt 1024 ] && write_size=1024
3297 dd if=/dev/zero bs=${write_size}k count=10 of=$DIR/$tdir/$tfile
3300 facet_failover $SINGLEMDS
3301 rm $DIR/$tdir/$tfile
3304 zconf_mount $(hostname) $MOUNT || error "mount fails"
3305 client_up || error "client_up failed"
3307 # wait for the remounted client to connect to ost1
3308 local target=$(get_osc_import_name client ost1)
3309 wait_import_state "FULL" "osc.${target}.ost_server_uuid" \
3310 $(max_recovery_time)
3312 wait_mds_ost_sync || error "MDS-OST sync timed out"
3313 wait_delete_completed || error "wait delete timed out"
3314 local blocks2=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
3316 [ $((blocks2 - blocks1)) -le $(fs_log_size) ] ||
3317 error $((blocks2 - blocks1)) blocks leaked
3319 run_test 89 "no disk space leak on late ost connection"
3326 change_active $facet
3327 wait_for_facet $facet
3328 mount_facet $facet || error "Restart of $facet failed"
3332 test_90() { # bug 19494
3333 local dir=$DIR/$tdir
3334 local ostfail=$(get_random_entry $(get_facets OST))
3336 if [[ $FAILURE_MODE = HARD ]]; then
3337 local affected=$(affected_facets $ostfail);
3338 if [[ "$affected" != $ostfail ]]; then
3339 skip not functional with FAILURE_MODE=$FAILURE_MODE, affected: $affected
3343 # ensure all OSTs are active to allow allocations
3346 mkdir $dir || error "mkdir $dir failed"
3348 echo "Create the files"
3350 # file "f${index}" striped over 1 OST
3351 # file "all" striped over all OSTs
3353 $LFS setstripe -c $OSTCOUNT $dir/all ||
3354 error "setstripe failed to create $dir/all"
3356 for ((i = 0; i < $OSTCOUNT; i++)); do
3359 $LFS setstripe -i $i -c 1 $f ||
3360 error "$LFS setstripe failed to create $f"
3362 # confirm setstripe actually created stripe on requested OST
3363 local uuid=$(ostuuid_from_index $i)
3365 for file in f$i all; do
3366 local found=$($LFS find --obd $uuid --name $file $dir)
3368 if [[ $dir/$file != $found ]]; then
3369 $LFS getstripe $dir/$file
3370 error "wrong stripe: $file, uuid: $uuid"
3375 # Before failing an OST, get its obd name and index
3376 local varsvc=${ostfail}_svc
3377 local obd=$(do_facet $ostfail lctl get_param \
3378 -n obdfilter.${!varsvc}.uuid)
3379 local index=$(($(facet_number $ostfail) - 1))
3381 echo "Fail $ostfail $obd, display the list of affected files"
3382 shutdown_facet $ostfail || error "shutdown_facet $ostfail failed"
3384 trap "cleanup_90 $ostfail" EXIT INT
3385 echo "General Query: lfs find $dir"
3386 local list=$($LFS find $dir)
3388 for (( i=0; i<$OSTCOUNT; i++ )); do
3389 list_member "$list" $dir/f$i ||
3390 error_noexit "lfs find $dir: no file f$i"
3392 list_member "$list" $dir/all ||
3393 error_noexit "lfs find $dir: no file all"
3395 # focus on the missing OST,
3396 # we expect to see only two files affected: "f$(index)" and "all"
3398 echo "Querying files on shutdown $ostfail: lfs find --obd $obd"
3399 list=$($LFS find --obd $obd $dir)
3401 for file in all f$index; do
3402 list_member "$list" $dir/$file ||
3403 error_noexit "lfs find does not report the affected $obd for $file"
3406 [[ $(echo $list | wc -w) -eq 2 ]] ||
3407 error_noexit "lfs find reports the wrong list of affected files ${#list[@]}"
3409 echo "Check getstripe: $GETSTRIPE -r --obd $obd"
3410 list=$($GETSTRIPE -r --obd $obd $dir)
3412 for file in all f$index; do
3413 echo "$list" | grep $dir/$file ||
3414 error_noexit "lfs getsripe does not report the affected $obd for $file"
3419 run_test 90 "lfs find identifies the missing striped file segments"
3422 local server_version=$(lustre_version_code $SINGLEMDS)
3423 [[ $server_version -ge $(version_code 2.6.90) ]] ||
3424 [[ $server_version -ge $(version_code 2.5.4) &&
3425 $server_version -lt $(version_code 2.5.50) ]] ||
3426 { skip "Need MDS version 2.5.4+ or 2.6.90+"; return; }
3428 cancel_lru_locks osc
3430 $SETSTRIPE -i 0 -c 1 $DIR/$tfile ||
3431 error "$SETSTRIPE $DIR/$tfile failed"
3432 dd if=/dev/zero of=$DIR/$tfile bs=1024 count=1 ||
3433 error "dd to $DIR/$tfile failed"
3434 #define OBD_FAIL_TGT_REPLAY_RECONNECT 0x715
3435 # We need to emulate a state that OST is waiting for other clients
3436 # not completing the recovery. Final ping is queued, but reply will be
3437 # sent on the recovery completion. It is done by sleep before
3438 # processing final pings
3439 do_facet ost1 "$LCTL set_param fail_val=40"
3440 do_facet ost1 "$LCTL set_param fail_loc=0x715"
3443 run_test 93a "replay + reconnect"
3446 local server_version=$(lustre_version_code $SINGLEMDS)
3447 [[ $server_version -ge $(version_code 2.7.90) ]] ||
3448 { skip "Need MDS version 2.7.90+"; return; }
3450 cancel_lru_locks mdc
3452 createmany -o $DIR/$tfile 20 ||
3453 error "createmany -o $DIR/$tfile failed"
3455 #define OBD_FAIL_TGT_REPLAY_RECONNECT 0x715
3456 # We need to emulate a state that MDT is waiting for other clients
3457 # not completing the recovery. Final ping is queued, but reply will be
3458 # sent on the recovery completion. It is done by sleep before
3459 # processing final pings
3460 do_facet mds1 "$LCTL set_param fail_val=80"
3461 do_facet mds1 "$LCTL set_param fail_loc=0x715"
3464 run_test 93b "replay + reconnect on mds"
3466 striped_dir_check_100() {
3467 local striped_dir=$DIR/$tdir/striped_dir
3468 local stripe_count=$($LFS getdirstripe -c $striped_dir)
3470 $LFS getdirstripe $striped_dir
3471 [ $stripe_count -eq 2 ] || error "$stripe_count != 2"
3473 createmany -o $striped_dir/f-%d 20 ||
3474 error "creation failed under striped dir"
3478 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3479 ([ $FAILURE_MODE == "HARD" ] &&
3480 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3481 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3484 local striped_dir=$DIR/$tdir/striped_dir
3487 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3489 #To make sure MDT1 and MDT0 are connected
3490 #otherwise it may create single stripe dir here
3491 $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
3493 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
3494 do_facet mds$((MDTIDX+1)) lctl set_param fail_loc=0x1701
3495 $LFS setdirstripe -i0 -c2 $striped_dir &
3498 fail mds$((MDTIDX + 1))
3500 wait $CLIENT_PID || error "striped dir creation failed"
3502 striped_dir_check_100 || error "striped dir check failed"
3503 rm -rf $DIR/$tdir || error "rmdir failed"
3505 run_test 100a "DNE: create striped dir, drop update rep from MDT1, fail MDT1"
3508 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3509 ([ $FAILURE_MODE == "HARD" ] &&
3510 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3511 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3514 local striped_dir=$DIR/$tdir/striped_dir
3517 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3519 #To make sure MDT1 and MDT0 are connected
3520 #otherwise it may create single stripe dir here
3521 $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
3523 # OBD_FAIL_MDS_REINT_NET_REP 0x119
3524 do_facet mds$MDTIDX lctl set_param fail_loc=0x119
3525 $LFS mkdir -i0 -c2 $striped_dir &
3530 wait $CLIENT_PID || error "striped dir creation failed"
3532 striped_dir_check_100 || error "striped dir check failed"
3533 rm -rf $DIR/$tdir || error "rmdir failed"
3535 run_test 100b "DNE: create striped dir, fail MDT0"
3537 test_101() { #LU-5648
3538 mkdir -p $DIR/$tdir/d1
3539 mkdir -p $DIR/$tdir/d2
3540 touch $DIR/$tdir/file0
3543 replay_barrier $SINGLEMDS
3544 for i in $(seq $num) ; do
3545 echo test$i > $DIR/$tdir/d1/file$i
3548 fail_abort $SINGLEMDS
3549 for i in $(seq $num) ; do
3550 touch $DIR/$tdir/d2/file$i
3551 test -s $DIR/$tdir/d2/file$i &&
3552 ls -al $DIR/$tdir/d2/file$i && error "file$i's size > 0"
3557 run_test 101 "Shouldn't reassign precreated objs to other files after recovery"
3566 [[ $(lctl get_param mdc.*.import |
3567 grep "connect_flags:.*multi_mod_rpc") ]] ||
3568 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3570 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3571 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3572 facet="mds$((0x$idx + 1))"
3574 # get current value of max_mod_rcps_in_flight
3575 num=$($LCTL get_param -n \
3576 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3577 # set default value if client does not support multi mod RPCs
3578 [ -z "$num" ] && num=1
3580 echo "creating $num files ..."
3582 for i in $(seq $num); do
3583 touch $DIR/$tdir/file-$i
3586 # drop request on MDT to force resend
3587 #define OBD_FAIL_MDS_REINT_MULTI_NET 0x159
3588 do_facet $facet "$LCTL set_param fail_loc=0x159"
3589 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3590 for i in $(seq $num); do
3591 chmod 0600 $DIR/$tdir/file-$i &
3595 do_facet $facet "$LCTL set_param fail_loc=0"
3596 for pid in $pids; do
3597 wait $pid || error "chmod failed"
3599 echo "done ($(date +%H:%M:%S))"
3601 # check chmod succeed
3602 for i in $(seq $num); do
3603 checkstat -vp 0600 $DIR/$tdir/file-$i
3608 run_test 102a "check resend (request lost) with multiple modify RPCs in flight"
3617 [[ $(lctl get_param mdc.*.import |
3618 grep "connect_flags:.*multi_mod_rpc") ]] ||
3619 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3621 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3622 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3623 facet="mds$((0x$idx + 1))"
3625 # get current value of max_mod_rcps_in_flight
3626 num=$($LCTL get_param -n \
3627 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3628 # set default value if client does not support multi mod RPCs
3629 [ -z "$num" ] && num=1
3631 echo "creating $num files ..."
3633 for i in $(seq $num); do
3634 touch $DIR/$tdir/file-$i
3637 # drop reply on MDT to force reconstruction
3638 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3639 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3640 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3641 for i in $(seq $num); do
3642 chmod 0600 $DIR/$tdir/file-$i &
3646 do_facet $facet "$LCTL set_param fail_loc=0"
3647 for pid in $pids; do
3648 wait $pid || error "chmod failed"
3650 echo "done ($(date +%H:%M:%S))"
3652 # check chmod succeed
3653 for i in $(seq $num); do
3654 checkstat -vp 0600 $DIR/$tdir/file-$i
3659 run_test 102b "check resend (reply lost) with multiple modify RPCs in flight"
3668 [[ $(lctl get_param mdc.*.import |
3669 grep "connect_flags:.*multi_mod_rpc") ]] ||
3670 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3672 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3673 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3674 facet="mds$((0x$idx + 1))"
3676 # get current value of max_mod_rcps_in_flight
3677 num=$($LCTL get_param -n \
3678 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3679 # set default value if client does not support multi mod RPCs
3680 [ -z "$num" ] && num=1
3682 echo "creating $num files ..."
3684 for i in $(seq $num); do
3685 touch $DIR/$tdir/file-$i
3688 replay_barrier $facet
3691 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3692 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3693 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3694 for i in $(seq $num); do
3695 chmod 0600 $DIR/$tdir/file-$i &
3699 do_facet $facet "$LCTL set_param fail_loc=0"
3704 for pid in $pids; do
3705 wait $pid || error "chmod failed"
3707 echo "done ($(date +%H:%M:%S))"
3709 # check chmod succeed
3710 for i in $(seq $num); do
3711 checkstat -vp 0600 $DIR/$tdir/file-$i
3716 run_test 102c "check replay w/o reconstruction with multiple mod RPCs in flight"
3725 [[ $(lctl get_param mdc.*.import |
3726 grep "connect_flags:.*multi_mod_rpc") ]] ||
3727 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3729 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3730 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3731 facet="mds$((0x$idx + 1))"
3733 # get current value of max_mod_rcps_in_flight
3734 num=$($LCTL get_param -n \
3735 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3736 # set default value if client does not support multi mod RPCs
3737 [ -z "$num" ] && num=1
3739 echo "creating $num files ..."
3741 for i in $(seq $num); do
3742 touch $DIR/$tdir/file-$i
3746 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3747 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3748 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3749 for i in $(seq $num); do
3750 chmod 0600 $DIR/$tdir/file-$i &
3755 # write MDT transactions to disk
3756 do_facet $facet "sync; sync; sync"
3758 do_facet $facet "$LCTL set_param fail_loc=0"
3763 for pid in $pids; do
3764 wait $pid || error "chmod failed"
3766 echo "done ($(date +%H:%M:%S))"
3768 # check chmod succeed
3769 for i in $(seq $num); do
3770 checkstat -vp 0600 $DIR/$tdir/file-$i
3775 run_test 102d "check replay & reconstruction with multiple mod RPCs in flight"
3778 remote_mds_nodsh && skip "remote MDS with nodsh" && return
3779 local mds_version=$(lustre_version_code $SINGLEMDS)
3780 [[ $mds_version -gt $(version_code 2.8.54) ]] ||
3781 { skip "Need MDS version 2.8.54+"; return; }
3783 #define OBD_FAIL_MDS_TRACK_OVERFLOW 0x162
3784 do_facet mds1 $LCTL set_param fail_loc=0x80000162
3787 createmany -o $DIR/$tdir/t- 30 ||
3788 error "create files on remote directory failed"
3790 rm -rf $DIR/$tdir/t-*
3792 #MDS should crash with tr->otr_next_id overflow
3795 run_test 103 "Check otr_next_id overflow"
3798 check_striped_dir_110()
3800 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
3801 error "create striped dir failed"
3802 local stripe_count=$($LFS getdirstripe -c $DIR/$tdir/striped_dir)
3803 [ $stripe_count -eq $MDSCOUNT ] ||
3804 error "$stripe_count != 2 after recovery"
3808 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3809 ([ $FAILURE_MODE == "HARD" ] &&
3810 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3811 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3816 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3819 check_striped_dir_110 || error "check striped_dir failed"
3820 rm -rf $DIR/$tdir || error "rmdir failed"
3824 run_test 110a "DNE: create striped dir, fail MDT1"
3827 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3828 ([ $FAILURE_MODE == "HARD" ] &&
3829 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3830 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3835 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3838 zconf_mount $(hostname) $MOUNT
3839 client_up || return 1
3841 check_striped_dir_110 || error "check striped_dir failed"
3843 rm -rf $DIR/$tdir || error "rmdir failed"
3847 run_test 110b "DNE: create striped dir, fail MDT1 and client"
3850 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3851 ([ $FAILURE_MODE == "HARD" ] &&
3852 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3853 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3858 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3861 check_striped_dir_110 || error "check striped_dir failed"
3863 rm -rf $DIR/$tdir || error "rmdir failed"
3867 run_test 110c "DNE: create striped dir, fail MDT2"
3870 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3871 ([ $FAILURE_MODE == "HARD" ] &&
3872 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3873 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3878 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3881 zconf_mount $(hostname) $MOUNT
3882 client_up || return 1
3884 check_striped_dir_110 || error "check striped_dir failed"
3886 rm -rf $DIR/$tdir || error "rmdir failed"
3890 run_test 110d "DNE: create striped dir, fail MDT2 and client"
3893 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3894 ([ $FAILURE_MODE == "HARD" ] &&
3895 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3896 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3901 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3905 zconf_mount $(hostname) $MOUNT
3906 client_up || return 1
3908 check_striped_dir_110 || error "check striped_dir failed"
3910 rm -rf $DIR/$tdir || error "rmdir failed"
3914 run_test 110e "DNE: create striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
3917 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3918 ([ $FAILURE_MODE == "HARD" ] &&
3919 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3920 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3926 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3929 check_striped_dir_110 || error "check striped_dir failed"
3931 rm -rf $DIR/$tdir || error "rmdir failed"
3935 run_test 110f "DNE: create striped dir, fail MDT1/MDT2"
3938 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3939 ([ $FAILURE_MODE == "HARD" ] &&
3940 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3941 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3946 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3950 zconf_mount $(hostname) $MOUNT
3951 client_up || return 1
3953 check_striped_dir_110 || error "check striped_dir failed"
3955 rm -rf $DIR/$tdir || error "rmdir failed"
3959 run_test 110g "DNE: create striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
3962 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3963 ([ $FAILURE_MODE == "HARD" ] &&
3964 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3965 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3969 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3971 rm -rf $DIR/$tdir/striped_dir
3974 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3975 error "striped dir still exists"
3978 run_test 111a "DNE: unlink striped dir, fail MDT1"
3981 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3982 ([ $FAILURE_MODE == "HARD" ] &&
3983 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3984 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3988 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3990 rm -rf $DIR/$tdir/striped_dir
3993 zconf_mount $(hostname) $MOUNT
3994 client_up || return 1
3996 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3997 error "striped dir still exists"
4000 run_test 111b "DNE: unlink striped dir, fail MDT2"
4003 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4004 ([ $FAILURE_MODE == "HARD" ] &&
4005 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4006 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4010 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
4012 rm -rf $DIR/$tdir/striped_dir
4016 zconf_mount $(hostname) $MOUNT
4017 client_up || return 1
4018 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
4019 error "striped dir still exists"
4022 run_test 111c "DNE: unlink striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
4025 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4026 ([ $FAILURE_MODE == "HARD" ] &&
4027 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4028 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4032 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
4034 rm -rf $DIR/$tdir/striped_dir
4038 zconf_mount $(hostname) $MOUNT
4039 client_up || return 1
4040 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
4041 error "striped dir still exists"
4045 run_test 111d "DNE: unlink striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
4048 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4049 ([ $FAILURE_MODE == "HARD" ] &&
4050 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4051 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4055 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
4057 rm -rf $DIR/$tdir/striped_dir
4060 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
4061 error "striped dir still exists"
4064 run_test 111e "DNE: unlink striped dir, uncommit on MDT2, fail MDT1/MDT2"
4067 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4068 ([ $FAILURE_MODE == "HARD" ] &&
4069 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4070 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4074 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
4076 rm -rf $DIR/$tdir/striped_dir
4079 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
4080 error "striped dir still exists"
4083 run_test 111f "DNE: unlink striped dir, uncommit on MDT1, fail MDT1/MDT2"
4086 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4087 ([ $FAILURE_MODE == "HARD" ] &&
4088 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4089 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4093 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
4096 rm -rf $DIR/$tdir/striped_dir
4098 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
4099 error "striped dir still exists"
4102 run_test 111g "DNE: unlink striped dir, fail MDT1/MDT2"
4104 test_112_rename_prepare() {
4105 mkdir -p $DIR/$tdir/src_dir
4106 $LFS mkdir -i 1 $DIR/$tdir/src_dir/src_child ||
4107 error "create remote source failed"
4109 touch $DIR/$tdir/src_dir/src_child/a
4111 $LFS mkdir -i 2 $DIR/$tdir/tgt_dir ||
4112 error "create remote target dir failed"
4114 $LFS mkdir -i 3 $DIR/$tdir/tgt_dir/tgt_child ||
4115 error "create remote target child failed"
4120 $CHECKSTAT -t dir $DIR/$tdir/src_dir/src_child &&
4121 error "src_child still exists after rename"
4123 $CHECKSTAT -t file $DIR/$tdir/tgt_dir/tgt_child/a ||
4124 error "missing file(a) after rename"
4128 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4129 ([ $FAILURE_MODE == "HARD" ] &&
4130 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4131 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4134 test_112_rename_prepare
4137 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4138 error "rename dir cross MDT failed!"
4142 rm -rf $DIR/$tdir || error "rmdir failed"
4144 run_test 112a "DNE: cross MDT rename, fail MDT1"
4147 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4148 ([ $FAILURE_MODE == "HARD" ] &&
4149 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4150 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4153 test_112_rename_prepare
4156 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4157 error "rename dir cross MDT failed!"
4162 rm -rf $DIR/$tdir || error "rmdir failed"
4164 run_test 112b "DNE: cross MDT rename, fail MDT2"
4167 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4168 ([ $FAILURE_MODE == "HARD" ] &&
4169 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4170 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4173 test_112_rename_prepare
4176 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4177 error "rename dir cross MDT failed!"
4182 rm -rf $DIR/$tdir || error "rmdir failed"
4184 run_test 112c "DNE: cross MDT rename, fail MDT3"
4187 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4188 ([ $FAILURE_MODE == "HARD" ] &&
4189 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4190 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4193 test_112_rename_prepare
4196 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4197 error "rename dir cross MDT failed!"
4202 rm -rf $DIR/$tdir || error "rmdir failed"
4204 run_test 112d "DNE: cross MDT rename, fail MDT4"
4207 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4208 ([ $FAILURE_MODE == "HARD" ] &&
4209 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4210 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4213 test_112_rename_prepare
4217 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4218 error "rename dir cross MDT failed!"
4223 rm -rf $DIR/$tdir || error "rmdir failed"
4225 run_test 112e "DNE: cross MDT rename, fail MDT1 and MDT2"
4228 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4229 ([ $FAILURE_MODE == "HARD" ] &&
4230 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4231 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4234 test_112_rename_prepare
4238 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4239 error "rename dir cross MDT failed!"
4244 rm -rf $DIR/$tdir || error "rmdir failed"
4246 run_test 112f "DNE: cross MDT rename, fail MDT1 and MDT3"
4249 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4250 ([ $FAILURE_MODE == "HARD" ] &&
4251 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4252 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4255 test_112_rename_prepare
4259 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4260 error "rename dir cross MDT failed!"
4265 rm -rf $DIR/$tdir || error "rmdir failed"
4267 run_test 112g "DNE: cross MDT rename, fail MDT1 and MDT4"
4270 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4271 ([ $FAILURE_MODE == "HARD" ] &&
4272 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4273 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4276 test_112_rename_prepare
4280 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4281 error "rename dir cross MDT failed!"
4286 rm -rf $DIR/$tdir || error "rmdir failed"
4288 run_test 112h "DNE: cross MDT rename, fail MDT2 and MDT3"
4291 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4292 ([ $FAILURE_MODE == "HARD" ] &&
4293 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4294 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4297 test_112_rename_prepare
4301 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4302 error "rename dir cross MDT failed!"
4307 rm -rf $DIR/$tdir || error "rmdir failed"
4309 run_test 112i "DNE: cross MDT rename, fail MDT2 and MDT4"
4312 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4313 ([ $FAILURE_MODE == "HARD" ] &&
4314 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4315 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4318 test_112_rename_prepare
4322 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4323 error "rename dir cross MDT failed!"
4328 rm -rf $DIR/$tdir || error "rmdir failed"
4330 run_test 112j "DNE: cross MDT rename, fail MDT3 and MDT4"
4333 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4334 ([ $FAILURE_MODE == "HARD" ] &&
4335 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4336 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4339 test_112_rename_prepare
4344 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4345 error "rename dir cross MDT failed!"
4350 rm -rf $DIR/$tdir || error "rmdir failed"
4352 run_test 112k "DNE: cross MDT rename, fail MDT1,MDT2,MDT3"
4355 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4356 ([ $FAILURE_MODE == "HARD" ] &&
4357 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4358 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4361 test_112_rename_prepare
4366 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4367 error "rename dir cross MDT failed!"
4372 rm -rf $DIR/$tdir || error "rmdir failed"
4374 run_test 112l "DNE: cross MDT rename, fail MDT1,MDT2,MDT4"
4377 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4378 ([ $FAILURE_MODE == "HARD" ] &&
4379 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4380 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4383 test_112_rename_prepare
4388 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4389 error "rename dir cross MDT failed!"
4394 rm -rf $DIR/$tdir || error "rmdir failed"
4396 run_test 112m "DNE: cross MDT rename, fail MDT1,MDT3,MDT4"
4399 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4400 ([ $FAILURE_MODE == "HARD" ] &&
4401 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4402 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4405 test_112_rename_prepare
4410 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4411 error "rename dir cross MDT failed!"
4416 rm -rf $DIR/$tdir || error "rmdir failed"
4418 run_test 112n "DNE: cross MDT rename, fail MDT2,MDT3,MDT4"
4421 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4422 ([ $FAILURE_MODE == "HARD" ] &&
4423 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4424 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4432 for ((j=0;j<$((MDSCOUNT));j++)); do
4433 fail_index=$((fail_index+1))
4434 index=$((fail_index % MDSCOUNT))
4435 replay_barrier mds$((index + 1))
4436 for ((i=0;i<5;i++)); do
4437 test_mkdir -i$index -c$MDSCOUNT $DIR/$tdir/test_$i ||
4438 error "create striped dir $DIR/$tdir/test_$i"
4441 fail mds$((index + 1))
4442 for ((i=0;i<5;i++)); do
4443 checkstat -t dir $DIR/$tdir/test_$i ||
4444 error "$DIR/$tdir/test_$i does not exist!"
4446 rm -rf $DIR/$tdir/test_* ||
4450 run_test 115 "failover for create/unlink striped directory"
4453 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4454 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.55) ] &&
4455 skip "Do not support large update log before 2.7.55" &&
4457 ([ $FAILURE_MODE == "HARD" ] &&
4458 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4459 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4466 # OBD_FAIL_SPLIT_UPDATE_REC 0x1702
4467 do_facet mds1 "lctl set_param fail_loc=0x80001702"
4468 $LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/striped_dir
4471 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
4472 error "stried_dir does not exists"
4474 run_test 116a "large update log master MDT recovery"
4477 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4478 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.55) ] &&
4479 skip "Do not support large update log before 2.7.55" &&
4482 ([ $FAILURE_MODE == "HARD" ] &&
4483 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4484 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4491 # OBD_FAIL_SPLIT_UPDATE_REC 0x1702
4492 do_facet mds2 "lctl set_param fail_loc=0x80001702"
4493 $LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/striped_dir
4496 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
4497 error "stried_dir does not exists"
4499 run_test 116b "large update log slave MDT recovery"
4502 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4503 ([ $FAILURE_MODE == "HARD" ] &&
4504 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4505 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4511 $LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/remote_dir
4512 $LFS setdirstripe -i1 -c$MDSCOUNT $DIR/$tdir/remote_dir_1
4515 # Let's set rdonly on all MDTs, so client will send
4516 # replay requests on all MDTs and replay these requests
4517 # at the same time. This test will verify the recovery
4518 # will not be deadlock in this case, LU-7531.
4519 for ((index = 0; index < $((MDSCOUNT)); index++)); do
4520 replay_barrier mds$((index + 1))
4521 if [ -z $mds_indexs ]; then
4522 mds_indexs="${mds_indexs}mds$((index+1))"
4524 mds_indexs="${mds_indexs},mds$((index+1))"
4528 rm -rf $DIR/$tdir/remote_dir
4529 rm -rf $DIR/$tdir/remote_dir_1
4533 rm -rf $DIR/$tdir || error "rmdir failed"
4535 run_test 117 "DNE: cross MDT unlink, fail MDT1 and MDT2"
4538 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4539 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
4540 skip "Do not support large update log before 2.7.64" &&
4545 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir ||
4546 error "setdirstripe fails"
4547 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir1 ||
4548 error "setdirstripe fails 1"
4549 rm -rf $DIR/$tdir/striped_dir* || error "rmdir fails"
4551 # OBD_FAIL_INVALIDATE_UPDATE 0x1705
4552 do_facet mds1 "lctl set_param fail_loc=0x1705"
4553 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir
4554 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir1
4555 do_facet mds1 "lctl set_param fail_loc=0x0"
4558 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir
4559 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir1
4564 run_test 118 "invalidate osp update will not cause update log corruption"
4567 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4568 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
4569 skip "Do not support large update log before 2.7.64" &&
4572 local hard_timeout=$(do_facet mds1 \
4573 "lctl get_param -n mdt.$FSNAME-MDT0000.recovery_time_hard")
4575 local clients=${CLIENTS:-$HOSTNAME}
4576 local time_min=$(recovery_time_min)
4579 mkdir $DIR/$tdir/tmp
4580 rmdir $DIR/$tdir/tmp
4583 mkdir $DIR/$tdir/dir_1
4584 for ((i = 0; i < 20; i++)); do
4585 $LFS setdirstripe -i0 -c2 $DIR/$tdir/stripe_dir-$i
4592 #define OBD_FAIL_TGT_REPLAY_DELAY 0x714
4593 do_facet mds1 $LCTL set_param fail_loc=0x80000714
4594 #sleep (timeout + 5), so mds will evict the client exports,
4595 #but DNE update recovery will keep going.
4596 do_facet mds1 $LCTL set_param fail_val=$((time_min + 5))
4598 mount_facet mds1 "-o recovery_time_hard=$time_min"
4600 wait_clients_import_state "$clients" mds1 FULL
4602 clients_up || clients_up || error "failover df: $?"
4604 #revert back the hard timeout
4605 do_facet mds1 $LCTL set_param \
4606 mdt.$FSNAME-MDT0000.recovery_time_hard=$hard_timeout
4608 for ((i = 0; i < 20; i++)); do
4609 stripe_count=$($LFS getdirstripe -c $DIR/$tdir/stripe_dir-$i)
4610 [ $stripe_count == 2 ] || {
4611 error "stripe_dir-$i creation replay fails"
4616 run_test 119 "timeout of normal replay does not cause DNE replay fails "
4619 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4620 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
4621 skip "Do not support large update log before 2.7.64" &&
4625 replay_barrier_nosync mds1
4626 for ((i = 0; i < 20; i++)); do
4627 mkdir $DIR/$tdir/dir-$i || {
4628 error "create dir-$i fails"
4631 $LFS setdirstripe -i0 -c2 $DIR/$tdir/stripe_dir-$i || {
4632 error "create stripe_dir-$i fails"
4639 for ((i = 0; i < 20; i++)); do
4640 [ ! -e "$DIR/$tdir/dir-$i" ] || {
4641 error "dir-$i still exists"
4644 [ ! -e "$DIR/$tdir/stripe_dir-$i" ] || {
4645 error "stripe_dir-$i still exists"
4650 run_test 120 "DNE fail abort should stop both normal and DNE replay"
4653 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.10.90) ] &&
4654 skip "Don't support it before 2.11" &&
4657 local at_max_saved=$(at_max_get mds)
4659 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
4660 cancel_lru_locks mdc
4662 multiop_bg_pause $DIR/$tfile s_s || error "multiop $DIR/$tfile failed"
4665 lctl set_param -n ldlm.cancel_unused_locks_before_replay "0"
4671 #define OBD_FAIL_TGT_RECOVERY_REQ_RACE 0x721
4672 do_facet $SINGLEMDS "lctl set_param fail_loc=0x721 fail_val=0"
4676 wait_clients_import_state "$clients" mds1 FULL
4677 clients_up || clients_up || error "failover df: $?"
4680 wait $mpid || error "multiop_bg_pause pid failed"
4682 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
4683 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
4684 at_max_set $at_max_saved mds
4687 run_test 121 "lock replay timed out and race"
4690 check_and_cleanup_lustre