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 $SINGLEMDS) = "zfs" ] &&
32 # bug number for skipped test: LU-5761
33 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 89"
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"
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 createmany -o $DIR/$tfile 20 20 ||
1159 error "createmany recraete $DIR/$tfile failed"
1160 unlinkmany $DIR/$tfile 40 || error "unlinkmany $DIR/$tfile failed"
1163 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1166 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $ost1_svc)
1167 local oscdev=$(do_facet $SINGLEMDS "lctl get_param -n devices" |
1168 grep $mdtosc | awk '{print $1}')
1169 [ "$oscdev" ] || error "could not find OSC device on MDS"
1170 do_facet $SINGLEMDS $LCTL --device $oscdev recover ||
1171 error "OSC device $oscdev recovery failed"
1172 do_facet $SINGLEMDS $LCTL --device $oscdev recover ||
1173 error "second OSC device $oscdev recovery failed"
1174 # give the mds_lov_sync threads a chance to run
1177 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1179 # b3764 timed out lock replay
1181 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ] &&
1182 skip "MDS prior to 2.6.90 handle LDLM_REPLY_NET incorrectly" &&
1185 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1186 cancel_lru_locks mdc
1188 multiop_bg_pause $DIR/$tfile s_s || error "multiop $DIR/$tfile failed"
1191 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
1192 lctl set_param -n ldlm.cancel_unused_locks_before_replay "0"
1193 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000157"
1195 fail $SINGLEMDS || error "fail $SINGLEMDS failed"
1197 wait $mpid || error "multiop_bg_pause pid failed"
1199 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1200 lctl set_param fail_loc=0x0
1201 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
1204 run_test 52 "time out lock replay (3764)"
1206 # bug 3462 - simultaneous MDC requests
1208 [[ $(lctl get_param mdc.*.import |
1209 grep "connect_flags:.*multi_mod_rpc") ]] ||
1210 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
1212 cancel_lru_locks mdc # cleanup locks from former test cases
1213 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1214 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1215 multiop $DIR/${tdir}-1/f O_c &
1217 # give multiop a change to open
1220 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1221 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1222 kill -USR1 $close_pid
1223 cancel_lru_locks mdc # force the close
1224 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1226 mcreate $DIR/${tdir}-2/f || error "mcreate $DIR/${tdir}-2/f failed"
1228 # close should still be here
1229 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1231 replay_barrier_nodf $SINGLEMDS
1233 wait $close_pid || error "close_pid $close_pid failed"
1235 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1236 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1237 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1238 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1239 rm -rf $DIR/${tdir}-*
1241 run_test 53a "|X| close request while two MDC requests in flight"
1244 cancel_lru_locks mdc # cleanup locks from former test cases
1246 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1247 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1248 multiop_bg_pause $DIR/${tdir}-1/f O_c ||
1249 error "multiop_bg_pause $DIR/${tdir}-1/f failed"
1252 #define OBD_FAIL_MDS_REINT_NET 0x107
1253 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1254 mcreate $DIR/${tdir}-2/f &
1258 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1259 kill -USR1 $close_pid
1260 cancel_lru_locks mdc # force the close
1261 wait $close_pid || error "close_pid $close_pid failed"
1262 # open should still be here
1263 [ -d /proc/$open_pid ] || error "open_pid doesn't exist"
1265 replay_barrier_nodf $SINGLEMDS
1267 wait $open_pid || error "open_pid failed"
1269 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1270 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1271 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1272 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1273 rm -rf $DIR/${tdir}-*
1275 run_test 53b "|X| open request while two MDC requests in flight"
1278 cancel_lru_locks mdc # cleanup locks from former test cases
1280 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1281 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1282 multiop $DIR/${tdir}-1/f O_c &
1285 #define OBD_FAIL_MDS_REINT_NET 0x107
1286 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1287 mcreate $DIR/${tdir}-2/f &
1291 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1292 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1293 kill -USR1 $close_pid
1294 cancel_lru_locks mdc # force the close
1296 #bz20647: make sure all pids exist before failover
1297 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1298 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1299 replay_barrier_nodf $SINGLEMDS
1300 fail_nodf $SINGLEMDS
1301 wait $open_pid || error "open_pid failed"
1303 # close should be gone
1304 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1305 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1307 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1308 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1309 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1310 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1311 rm -rf $DIR/${tdir}-*
1313 run_test 53c "|X| open request and close request while two MDC requests in flight"
1316 [[ $(lctl get_param mdc.*.import |
1317 grep "connect_flags:.*multi_mod_rpc") ]] ||
1318 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
1320 cancel_lru_locks mdc # cleanup locks from former test cases
1322 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1323 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1324 multiop $DIR/${tdir}-1/f O_c &
1326 # give multiop a chance to open
1329 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1330 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1331 kill -USR1 $close_pid
1332 cancel_lru_locks mdc # force the close
1333 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1334 mcreate $DIR/${tdir}-2/f || error "mcreate $DIR/${tdir}-2/f failed"
1336 # close should still be here
1337 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1339 wait $close_pid || error "close_pid failed"
1341 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1342 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1343 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1344 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1345 rm -rf $DIR/${tdir}-*
1347 run_test 53d "close reply while two MDC requests in flight"
1350 cancel_lru_locks mdc # cleanup locks from former test cases
1352 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1353 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1354 multiop $DIR/${tdir}-1/f O_c &
1357 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1358 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1359 mcreate $DIR/${tdir}-2/f &
1363 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1364 kill -USR1 $close_pid
1365 cancel_lru_locks mdc # force the close
1366 wait $close_pid || error "close_pid failed"
1367 # open should still be here
1368 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1370 replay_barrier_nodf $SINGLEMDS
1372 wait $open_pid || error "open_pid failed"
1374 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1375 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1376 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1377 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1378 rm -rf $DIR/${tdir}-*
1380 run_test 53e "|X| open reply while two MDC requests in flight"
1383 cancel_lru_locks mdc # cleanup locks from former test cases
1385 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1386 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1387 multiop $DIR/${tdir}-1/f O_c &
1390 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1391 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1392 mcreate $DIR/${tdir}-2/f &
1396 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1397 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1398 kill -USR1 $close_pid
1399 cancel_lru_locks mdc # force the close
1401 #bz20647: make sure all pids are exists before failover
1402 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1403 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1404 replay_barrier_nodf $SINGLEMDS
1405 fail_nodf $SINGLEMDS
1406 wait $open_pid || error "open_pid failed"
1408 # close should be gone
1409 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1410 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1412 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1413 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1414 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1415 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1416 rm -rf $DIR/${tdir}-*
1418 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1421 cancel_lru_locks mdc # cleanup locks from former test cases
1423 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1424 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1425 multiop $DIR/${tdir}-1/f O_c &
1428 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1429 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1430 mcreate $DIR/${tdir}-2/f &
1434 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1435 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1436 kill -USR1 $close_pid
1437 cancel_lru_locks mdc # force the close
1438 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1440 #bz20647: make sure all pids are exists before failover
1441 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1442 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1443 replay_barrier_nodf $SINGLEMDS
1444 fail_nodf $SINGLEMDS
1445 wait $open_pid || error "open_pid failed"
1447 # close should be gone
1448 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1450 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1451 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1452 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1453 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1454 rm -rf $DIR/${tdir}-*
1456 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1459 cancel_lru_locks mdc # cleanup locks from former test cases
1461 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1462 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1463 multiop $DIR/${tdir}-1/f O_c &
1466 #define OBD_FAIL_MDS_REINT_NET 0x107
1467 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1468 mcreate $DIR/${tdir}-2/f &
1472 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1473 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1474 kill -USR1 $close_pid
1475 cancel_lru_locks mdc # force the close
1478 #bz20647: make sure all pids are exists before failover
1479 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1480 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1481 replay_barrier_nodf $SINGLEMDS
1482 fail_nodf $SINGLEMDS
1483 wait $open_pid || error "open_pid failed"
1485 # close should be gone
1486 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1487 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1489 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1490 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1491 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1492 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1493 rm -rf $DIR/${tdir}-*
1495 run_test 53h "open request and close reply while two MDC requests in flight"
1497 #b3761 ASSERTION(hash != 0) failed
1499 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1500 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1502 # give touch a chance to run
1504 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1508 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1510 #b3440 ASSERTION(rec->ur_fid2->id) failed
1512 ln -s foo $DIR/$tfile
1513 replay_barrier $SINGLEMDS
1514 #drop_reply "cat $DIR/$tfile"
1518 run_test 56 "don't replay a symlink open request (3440)"
1520 #recovery one mds-ost setattr from llog
1522 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1523 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1524 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1525 replay_barrier $SINGLEMDS
1527 wait_recovery_complete $SINGLEMDS || error "MDS recovery is not done"
1528 wait_mds_ost_sync || error "wait_mds_ost_sync failed"
1529 $CHECKSTAT -t file $DIR/$tfile ||
1530 error "$CHECKSTAT $DIR/$tfile attribute check failed"
1531 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1534 run_test 57 "test recovery from llog for setattr op"
1537 zconf_umount $(hostname) $MOUNT2
1541 #recovery many mds-ost setattr from llog
1543 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1544 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1545 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1546 createmany -o $DIR/$tdir/$tfile-%d 2500
1547 replay_barrier $SINGLEMDS
1550 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null ||
1551 error "$CHECKSTAT $DIR/$tfile-* attribute check failed"
1552 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1553 unlinkmany $DIR/$tdir/$tfile-%d 2500 ||
1554 error "unlinkmany $DIR/$tfile failed"
1557 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
1563 trap cleanup_58 EXIT
1565 large_xattr_enabled &&
1566 orig="$(generate_string $(max_xattr_size))" || orig="bar"
1567 # Original extended attribute can be long. Print a small version of
1568 # attribute if an error occurs
1569 local sm_msg=$(printf "%.9s" $orig)
1571 mount_client $MOUNT2 || error "mount_client on $MOUNT2 failed"
1572 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1573 touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
1574 replay_barrier $SINGLEMDS
1575 setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile
1577 new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
1578 [[ "$new" = "$orig" ]] ||
1579 error "xattr set ($sm_msg...) differs from xattr get ($new)"
1580 rm -f $DIR/$tdir/$tfile
1583 wait_clients_import_state ${CLIENTS:-$HOSTNAME} "mgs" FULL
1585 run_test 58b "test replay of setxattr op"
1587 test_58c() { # bug 16570
1592 trap cleanup_58 EXIT
1594 if large_xattr_enabled; then
1595 local xattr_size=$(max_xattr_size)
1596 orig="$(generate_string $((xattr_size / 2)))"
1597 orig1="$(generate_string $xattr_size)"
1603 # PING_INTERVAL max(obd_timeout / 4, 1U)
1604 sleep $((TIMEOUT / 4))
1606 # Original extended attribute can be long. Print a small version of
1607 # attribute if an error occurs
1608 local sm_msg=$(printf "%.9s" $orig)
1609 local sm_msg1=$(printf "%.9s" $orig1)
1611 mount_client $MOUNT2 || error "mount_client on $MOUNT2 failed"
1612 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1613 touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
1614 drop_request "setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile" ||
1615 error "drop_request for setfattr failed"
1616 new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
1617 [[ "$new" = "$orig" ]] ||
1618 error "xattr set ($sm_msg...) differs from xattr get ($new)"
1619 drop_reint_reply "setfattr -n trusted.foo1 \
1620 -v $orig1 $DIR/$tdir/$tfile" ||
1621 error "drop_reint_reply for setfattr failed"
1622 new=$(get_xattr_value trusted.foo1 $MOUNT2/$tdir/$tfile)
1623 [[ "$new" = "$orig1" ]] ||
1624 error "second xattr set ($sm_msg1...) differs xattr get ($new)"
1625 rm -f $DIR/$tdir/$tfile
1629 run_test 58c "resend/reconstruct setxattr op"
1631 # log_commit_thread vs filter_destroy race used to lead to import use after free
1634 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1636 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1637 createmany -o $DIR/$tdir/$tfile-%d 200 ||
1638 error "createmany create files failed"
1640 unlinkmany $DIR/$tdir/$tfile-%d 200 ||
1641 error "unlinkmany $DIR/$tdir/$tfile failed"
1642 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1643 do_facet ost1 "lctl set_param fail_loc=0x507"
1646 do_facet ost1 "lctl set_param fail_loc=0x0"
1650 run_test 59 "test log_commit_thread vs filter_destroy race"
1652 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1653 # bug 12086: should no oops and No ctxt error for this test
1655 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1656 createmany -o $DIR/$tdir/$tfile-%d 200 ||
1657 error "createmany create files failed"
1658 replay_barrier $SINGLEMDS
1659 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1661 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1662 local no_ctxt=$(dmesg | grep "No ctxt")
1663 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1665 run_test 60 "test llog post recovery init vs llog unlink"
1667 #test race llog recovery thread vs llog cleanup
1668 test_61a() { # was test_61
1669 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1671 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1672 createmany -o $DIR/$tdir/$tfile-%d 800 ||
1673 error "createmany create files failed"
1675 unlinkmany $DIR/$tdir/$tfile-%d 800
1676 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1677 set_nodes_failloc "$(osts_nodes)" 0x80000221
1682 set_nodes_failloc "$(osts_nodes)" 0x0
1684 $CHECKSTAT -t file $DIR/$tdir/$tfile-* &&
1685 error "$CHECKSTAT $DIR/$tdir/$tfile attribute check should fail"
1688 run_test 61a "test race llog recovery vs llog cleanup"
1690 #test race mds llog sync vs llog cleanup
1692 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1693 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
1694 facet_failover $SINGLEMDS
1697 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 ||
1700 run_test 61b "test race mds llog sync vs llog cleanup"
1702 #test race cancel cookie cb vs llog cleanup
1704 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1706 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1707 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1708 set_nodes_failloc "$(osts_nodes)" 0x80000222
1712 set_nodes_failloc "$(osts_nodes)" 0x0
1714 run_test 61c "test race mds llog sync vs llog cleanup"
1716 test_61d() { # bug 16002 # bug 17466 # bug 22137
1717 # OBD_FAIL_OBD_LLOG_SETUP 0x605
1719 do_facet mgs "lctl set_param fail_loc=0x80000605"
1720 start mgs $(mgsdevname) $MGS_MOUNT_OPTS &&
1721 error "mgs start should have failed"
1722 do_facet mgs "lctl set_param fail_loc=0"
1723 start mgs $(mgsdevname) $MGS_MOUNT_OPTS || error "cannot restart mgs"
1725 run_test 61d "error in llog_setup should cleanup the llog context correctly"
1727 test_62() { # Bug 15756 - don't mis-drop resent replay
1728 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1729 replay_barrier $SINGLEMDS
1730 createmany -o $DIR/$tdir/$tfile- 25 ||
1731 error "createmany create files failed"
1732 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1733 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1735 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1736 unlinkmany $DIR/$tdir/$tfile- 25 ||
1737 error "unlinkmany $DIR/$tdir/$tfile failed"
1740 run_test 62 "don't mis-drop resent replay"
1742 #Adaptive Timeouts (bug 3055)
1750 echo "Cleaning up AT ..."
1751 if [ -n "$ATOLDBASE" ]; then
1752 local at_history=$($LCTL get_param -n at_history)
1753 do_facet $SINGLEMDS "lctl set_param at_history=$at_history" || true
1754 do_facet ost1 "lctl set_param at_history=$at_history" || true
1757 if [ $AT_MAX_SET -ne 0 ]; then
1758 for facet in mds client ost; do
1759 var=AT_MAX_SAVE_${facet}
1760 echo restore AT on $facet to saved value ${!var}
1761 at_max_set ${!var} $facet
1762 at_new=$(at_max_get $facet)
1763 echo Restored AT value on $facet $at_new
1764 [ $at_new -eq ${!var} ] ||
1765 error "AT value not restored SAVED ${!var} NEW $at_new"
1772 local at_max_new=600
1774 # Save at_max original values
1776 if [ $AT_MAX_SET -eq 0 ]; then
1777 # Suppose that all osts have the same at_max
1778 for facet in mds client ost; do
1779 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1783 for facet in mds client ost; do
1784 at_max=$(at_max_get $facet)
1785 if [ $at_max -ne $at_max_new ]; then
1786 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1787 at_max_set $at_max_new $facet
1792 if [ -z "$ATOLDBASE" ]; then
1793 ATOLDBASE=$(do_facet $SINGLEMDS "lctl get_param -n at_history")
1794 # speed up the timebase so we can check decreasing AT
1795 do_facet $SINGLEMDS "lctl set_param at_history=8" || true
1796 do_facet ost1 "lctl set_param at_history=8" || true
1798 # sleep for a while to cool down, should be > 8s and also allow
1799 # at least one ping to be sent. simply use TIMEOUT to be safe.
1804 test_65a() #bug 3055
1806 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1808 at_start || return 0
1809 $LCTL dk > /dev/null
1811 $LCTL set_param debug="other"
1812 # Slow down a request to the current service time, this is critical
1813 # because previous tests may have caused this value to increase.
1814 REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
1815 awk '/portal 12/ {print $5}'`
1816 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1818 do_facet $SINGLEMDS lctl set_param fail_val=$((${REQ_DELAY} * 1000))
1819 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1820 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x8000050a
1821 createmany -o $DIR/$tfile 10 > /dev/null
1822 unlinkmany $DIR/$tfile 10 > /dev/null
1823 # check for log message
1824 $LCTL dk | grep "Early reply #" || error "No early reply"
1826 # client should show REQ_DELAY estimates
1827 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1829 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1831 run_test 65a "AT: verify early replies"
1833 test_65b() #bug 3055
1835 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1837 at_start || return 0
1840 $LCTL set_param debug="other trace"
1841 $LCTL dk > /dev/null
1842 # Slow down a request to the current service time, this is critical
1843 # because previous tests may have caused this value to increase.
1844 $SETSTRIPE --stripe-index=0 --stripe-count=1 $DIR/$tfile ||
1845 error "$SETSTRIPE failed for $DIR/$tfile"
1847 multiop $DIR/$tfile Ow1yc
1848 REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
1849 awk '/portal 6/ {print $5}'`
1850 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1852 do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
1853 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1854 do_facet ost1 $LCTL set_param fail_loc=0x224
1857 $SETSTRIPE --stripe-index=0 --stripe-count=1 $DIR/$tfile ||
1858 error "$SETSTRIPE failed"
1859 # force some real bulk transfer
1860 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1862 do_facet ost1 $LCTL set_param fail_loc=0
1863 # check for log message
1864 $LCTL dk | grep "Early reply #" || error "No early reply"
1866 # client should show REQ_DELAY estimates
1867 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1869 run_test 65b "AT: verify early replies on packed reply / bulk"
1871 test_66a() #bug 3055
1873 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1875 at_start || return 0
1876 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1877 # adjust 5s at a time so no early reply is sent (within deadline)
1878 do_facet $SINGLEMDS "$LCTL set_param fail_val=5000"
1879 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1880 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x8000050a"
1881 createmany -o $DIR/$tfile 20 > /dev/null
1882 unlinkmany $DIR/$tfile 20 > /dev/null
1883 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1884 do_facet $SINGLEMDS "$LCTL set_param fail_val=10000"
1885 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x8000050a"
1886 createmany -o $DIR/$tfile 20 > /dev/null
1887 unlinkmany $DIR/$tfile 20 > /dev/null
1888 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1889 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0"
1891 createmany -o $DIR/$tfile 20 > /dev/null
1892 unlinkmany $DIR/$tfile 20 > /dev/null
1893 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1894 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1895 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1896 echo "Current MDT timeout $CUR, worst $WORST"
1897 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1899 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1901 test_66b() #bug 3055
1903 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1905 at_start || return 0
1906 ORIG=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1907 awk '/network/ {print $4}')
1908 $LCTL set_param fail_val=$(($ORIG + 5))
1909 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1910 $LCTL set_param fail_loc=0x50c
1911 ls $DIR/$tfile > /dev/null 2>&1
1912 $LCTL set_param fail_loc=0
1913 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1914 awk '/network/ {print $4}')
1915 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1916 awk '/network/ {print $6}')
1917 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1918 [ $WORST -gt $ORIG ] ||
1919 error "Worst $WORST should be worse than orig $ORIG"
1921 run_test 66b "AT: verify net latency adjusts"
1923 test_67a() #bug 3055
1925 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1927 at_start || return 0
1928 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1929 # sleeping threads may drive values above this
1930 do_facet ost1 "$LCTL set_param fail_val=400"
1931 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1932 do_facet ost1 "$LCTL set_param fail_loc=0x50a"
1933 createmany -o $DIR/$tfile 20 > /dev/null
1934 unlinkmany $DIR/$tfile 20 > /dev/null
1935 do_facet ost1 "$LCTL set_param fail_loc=0"
1936 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1937 ATTEMPTS=$(($CONN2 - $CONN1))
1938 echo "$ATTEMPTS osc reconnect attempts on gradual slow"
1939 [ $ATTEMPTS -gt 0 ] &&
1940 error_ignore bz13721 "AT should have prevented reconnect"
1943 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1945 test_67b() #bug 3055
1947 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1949 at_start || return 0
1950 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1952 # exhaust precreations on ost1
1953 local OST=$(ostname_from_index 0)
1954 local mdtosc=$(get_mdtosc_proc_path mds $OST)
1955 local last_id=$(do_facet $SINGLEMDS lctl get_param -n \
1956 osc.$mdtosc.prealloc_last_id)
1957 local next_id=$(do_facet $SINGLEMDS lctl get_param -n \
1958 osc.$mdtosc.prealloc_next_id)
1960 mkdir -p $DIR/$tdir/${OST} || error "mkdir $DIR/$tdir/${OST} failed"
1961 $SETSTRIPE -i 0 -c 1 $DIR/$tdir/${OST} || error "$SETSTRIPE failed"
1962 echo "Creating to objid $last_id on ost $OST..."
1963 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1964 do_facet ost1 "$LCTL set_param fail_val=20000"
1965 do_facet ost1 "$LCTL set_param fail_loc=0x80000223"
1966 createmany -o $DIR/$tdir/${OST}/f $next_id $((last_id - next_id + 2))
1969 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1971 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1972 ATTEMPTS=$(($CONN2 - $CONN1))
1973 echo "$ATTEMPTS osc reconnect attempts on instant slow"
1974 # do it again; should not timeout
1975 do_facet ost1 "$LCTL set_param fail_loc=0x80000223"
1976 cp /etc/profile $DIR/$tfile || error "cp failed"
1977 do_facet ost1 "$LCTL set_param fail_loc=0"
1979 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1980 CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1981 ATTEMPTS=$(($CONN3 - $CONN2))
1982 echo "$ATTEMPTS osc reconnect attempts on 2nd slow"
1983 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1986 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1988 test_68 () #bug 13813
1990 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1992 at_start || return 0
1993 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1994 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1995 local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
1996 [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
1997 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1998 local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
1999 echo $TIMEOUT >> $ldlm_enqueue_min
2000 do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
2002 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2003 $SETSTRIPE --stripe-index=0 -c 1 $DIR/$tdir ||
2004 error "$SETSTRIPE failed for $DIR/$tdir"
2005 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
2006 $LCTL set_param fail_val=$(($TIMEOUT - 1))
2007 $LCTL set_param fail_loc=0x80000312
2008 cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
2009 $LCTL set_param fail_val=$((TIMEOUT * 5 / 4))
2010 $LCTL set_param fail_loc=0x80000312
2011 cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
2012 $LCTL set_param fail_loc=0
2014 echo $ENQ_MIN >> $ldlm_enqueue_min
2015 do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
2019 run_test 68 "AT: verify slowing locks"
2022 # end of AT tests includes above lines
2024 # start multi-client tests
2026 [ -z "$CLIENTS" ] &&
2027 { skip "Need two or more clients." && return; }
2028 [ $CLIENTCOUNT -lt 2 ] &&
2029 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
2031 echo "mount clients $CLIENTS ..."
2032 zconf_mount_clients $CLIENTS $MOUNT
2034 local clients=${CLIENTS//,/ }
2035 echo "Write/read files on $DIR ; clients $CLIENTS ... "
2036 for CLIENT in $clients; do
2037 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
2038 of=$DIR/${tfile}_${CLIENT} 2>/dev/null ||
2039 error "dd failed on $CLIENT"
2042 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
2043 for C in ${CLIENTS//,/ }; do
2044 do_node $prev_client dd if=$DIR/${tfile}_${C} \
2045 of=/dev/null 2>/dev/null ||
2046 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
2052 run_test 70a "check multi client t-f"
2054 check_for_process () {
2059 killall_process $clients "$prog" -0
2063 local clients=${CLIENTS:-$HOSTNAME}
2065 zconf_mount_clients $clients $MOUNT
2068 [ "$SLOW" = "no" ] && duration=120
2069 # set duration to 900 because it takes some time to boot node
2070 [ "$FAILURE_MODE" = HARD ] && duration=900
2073 local start_ts=$(date +%s)
2074 local cmd="rundbench 1 -t $duration"
2076 if [ $MDSCOUNT -ge 2 ]; then
2077 test_mkdir -p -c$MDSCOUNT $DIR/$tdir
2078 $LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir
2080 do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
2081 PATH=\$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
2082 DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
2083 MOUNT=$MOUNT DIR=$DIR/$tdir/\\\$(hostname) LCTL=$LCTL $cmd" &
2086 #LU-1897 wait for all dbench copies to start
2087 while ! check_for_process $clients dbench; do
2088 elapsed=$(($(date +%s) - start_ts))
2089 if [ $elapsed -gt $duration ]; then
2090 killall_process $clients dbench
2091 error "dbench failed to start on $clients!"
2096 log "Started rundbench load pid=$pid ..."
2098 elapsed=$(($(date +%s) - start_ts))
2099 local num_failovers=0
2101 while [ $elapsed -lt $duration ]; do
2102 if ! check_for_process $clients dbench; then
2103 error_noexit "dbench stopped on some of $clients!"
2104 killall_process $clients dbench
2108 replay_barrier mds$fail_index
2109 sleep 1 # give clients a time to do operations
2110 # Increment the number of failovers
2111 num_failovers=$((num_failovers+1))
2112 log "$TESTNAME fail mds$fail_index $num_failovers times"
2114 elapsed=$(($(date +%s) - start_ts))
2115 if [ $fail_index -ge $MDSCOUNT ]; then
2118 fail_index=$((fail_index+1))
2122 wait $pid || error "rundbench load on $clients failed!"
2124 run_test 70b "dbench ${MDSCOUNT}mdts recovery; $CLIENTCOUNT clients"
2125 # end multi-client tests
2130 local monitor_pid=$3
2132 local start_ts=$(date +%s)
2133 local num_failovers=0
2136 elapsed=$(($(date +%s) - start_ts))
2137 while [ $elapsed -lt $duration ]; do
2138 fail_index=$((RANDOM%max_index+1))
2139 kill -0 $monitor_pid ||
2140 error "$monitor_pid stopped"
2142 replay_barrier mds$fail_index
2144 # Increment the number of failovers
2145 num_failovers=$((num_failovers+1))
2146 log "$TESTNAME fail mds$fail_index $num_failovers times"
2148 elapsed=$(($(date +%s) - start_ts))
2154 rm -f $DIR/replay-single.70c.lck
2159 local clients=${CLIENTS:-$HOSTNAME}
2162 zconf_mount_clients $clients $MOUNT
2165 [ "$SLOW" = "no" ] && duration=180
2166 # set duration to 900 because it takes some time to boot node
2167 [ "$FAILURE_MODE" = HARD ] && duration=600
2170 local start_ts=$(date +%s)
2172 trap cleanup_70c EXIT
2174 while [ ! -e $DIR/replay-single.70c.lck ]; do
2175 test_mkdir -p -c$MDSCOUNT $DIR/$tdir || break
2176 if [ $MDSCOUNT -ge 2 ]; then
2177 $LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir ||
2178 error "set default dirstripe failed"
2180 cd $DIR/$tdir || break
2181 tar cf - /etc | tar xf - || error "tar failed in loop"
2185 echo "Started tar $tar_70c_pid"
2187 random_fail_mdt $MDSCOUNT $duration $tar_70c_pid
2188 kill -0 $tar_70c_pid || error "tar $tar_70c_pid stopped"
2190 touch $DIR/replay-single.70c.lck
2191 wait $tar_70c_pid || error "$?: tar failed"
2196 run_test 70c "tar ${MDSCOUNT}mdts recovery"
2200 kill -9 $mkdir_70d_pid
2204 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2205 local clients=${CLIENTS:-$HOSTNAME}
2208 zconf_mount_clients $clients $MOUNT
2211 [ "$SLOW" = "no" ] && duration=180
2212 # set duration to 900 because it takes some time to boot node
2213 [ "$FAILURE_MODE" = HARD ] && duration=900
2218 local start_ts=$(date +%s)
2220 trap cleanup_70d EXIT
2223 $LFS mkdir -i0 -c2 $DIR/$tdir/test || {
2227 $LFS mkdir -i1 -c2 $DIR/$tdir/test1 || {
2232 touch $DIR/$tdir/test/a || {
2236 mkdir $DIR/$tdir/test/b || {
2240 rm -rf $DIR/$tdir/test || {
2245 touch $DIR/$tdir/test1/a || {
2249 mkdir $DIR/$tdir/test1/b || {
2254 rm -rf $DIR/$tdir/test1 || {
2261 echo "Started $mkdir_70d_pid"
2263 random_fail_mdt $MDSCOUNT $duration $mkdir_70d_pid
2264 kill -0 $mkdir_70d_pid || error "mkdir/rmdir $mkdir_70d_pid stopped"
2269 run_test 70d "mkdir/rmdir striped dir ${MDSCOUNT}mdts recovery"
2273 kill -9 $rename_70e_pid
2277 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2278 local clients=${CLIENTS:-$HOSTNAME}
2281 echo ha > /proc/sys/lnet/debug
2282 zconf_mount_clients $clients $MOUNT
2285 [ "$SLOW" = "no" ] && duration=180
2286 # set duration to 900 because it takes some time to boot node
2287 [ "$FAILURE_MODE" = HARD ] && duration=900
2290 $LFS mkdir -i0 $DIR/$tdir/test_0
2291 $LFS mkdir -i0 $DIR/$tdir/test_1
2292 touch $DIR/$tdir/test_0/a
2293 touch $DIR/$tdir/test_1/b
2294 trap cleanup_70e EXIT
2297 mrename $DIR/$tdir/test_0/a $DIR/$tdir/test_1/b > \
2303 checkstat $DIR/$tdir/test_0/a && {
2304 echo "a still exists"
2308 checkstat $DIR/$tdir/test_1/b || {
2309 echo "b still exists"
2313 touch $DIR/$tdir/test_0/a || {
2314 echo "touch a fails"
2318 mrename $DIR/$tdir/test_1/b $DIR/$tdir/test_0/a > \
2326 echo "Started $rename_70e_pid"
2328 random_fail_mdt 2 $duration $rename_70e_pid
2329 kill -0 $rename_70e_pid || error "rename $rename_70e_pid stopped"
2334 run_test 70e "rename cross-MDT with random fails"
2336 test_70f_write_and_read(){
2341 echo "Write/read files in: '$DIR/$tdir', clients: '$CLIENTS' ..."
2342 for client in ${CLIENTS//,/ }; do
2343 [ -f $stopflag ] || return
2345 local tgtfile=$DIR/$tdir/$tfile.$client
2346 do_node $client dd $DD_OPTS bs=1M count=10 if=$srcfile \
2347 of=$tgtfile 2>/dev/null ||
2348 error "dd $DD_OPTS bs=1M count=10 if=$srcfile " \
2349 "of=$tgtfile failed on $client, rc=$?"
2352 local prev_client=$(echo ${CLIENTS//,/ } | awk '{ print $NF }')
2355 for client in ${CLIENTS//,/ }; do
2356 [ -f $stopflag ] || return
2358 # flush client cache in case test is running on only one client
2359 # do_node $client cancel_lru_locks osc
2360 do_node $client $LCTL set_param ldlm.namespaces.*.lru_size=clear
2362 tgtfile=$DIR/$tdir/$tfile.$client
2363 local md5=$(do_node $prev_client "md5sum $tgtfile")
2364 [ ${checksum[$index]// */} = ${md5// */} ] ||
2365 error "$tgtfile: checksum doesn't match on $prev_client"
2366 index=$((index + 1))
2376 mkdir -p $DIR/$tdir || error "cannot create $DIR/$tdir directory"
2377 $SETSTRIPE -c -1 $DIR/$tdir || error "cannot $SETSTRIPE $DIR/$tdir"
2380 while [ -f $stopflag ]; do
2381 test_70f_write_and_read $srcfile $stopflag
2382 # use direct IO and buffer cache in turns if loop
2383 [ -n "$DD_OPTS" ] && DD_OPTS="" || DD_OPTS="oflag=direct"
2387 test_70f_cleanup() {
2389 rm -f $TMP/$tfile.stop
2390 do_nodes $CLIENTS rm -f $TMP/$tfile
2391 rm -f $DIR/$tdir/$tfile.*
2395 # [ x$ost1failover_HOST = x$ost_HOST ] &&
2396 # { skip "Failover host not defined" && return; }
2397 # [ -z "$CLIENTS" ] &&
2398 # { skip "CLIENTS are not specified." && return; }
2399 # [ $CLIENTCOUNT -lt 2 ] &&
2400 # { skip "Need 2 or more clients, have $CLIENTCOUNT" && return; }
2402 [[ $(lustre_version_code ost1) -lt $(version_code 2.9.53) ]] &&
2403 skip "Need server version at least 2.9.53" && return
2405 echo "mount clients $CLIENTS ..."
2406 zconf_mount_clients $CLIENTS $MOUNT
2408 local srcfile=$TMP/$tfile
2412 trap test_70f_cleanup EXIT
2413 # create a different source file local to each client node so we can
2414 # detect if the file wasn't written out properly after failover
2415 do_nodes $CLIENTS dd bs=1M count=10 if=/dev/urandom of=$srcfile \
2416 2>/dev/null || error "can't create $srcfile on $CLIENTS"
2417 for client in ${CLIENTS//,/ }; do
2418 checksum[$index]=$(do_node $client "md5sum $srcfile")
2419 index=$((index + 1))
2423 [ "$SLOW" = "no" ] && duration=60
2424 # set duration to 900 because it takes some time to boot node
2425 [ "$FAILURE_MODE" = HARD ] && duration=900
2427 local stopflag=$TMP/$tfile.stop
2428 test_70f_loop $srcfile $stopflag &
2432 local num_failovers=0
2433 local start_ts=$SECONDS
2434 while [ $elapsed -lt $duration ]; do
2438 num_failovers=$((num_failovers + 1))
2439 log "$TESTNAME failing OST $num_failovers times"
2442 elapsed=$((SECONDS - start_ts))
2449 run_test 70f "OSS O_DIRECT recovery with $CLIENTCOUNT clients"
2453 kill -9 $mkdir_71a_pid
2456 random_double_fail_mdt() {
2459 local monitor_pid=$3
2461 local start_ts=$(date +%s)
2462 local num_failovers=0
2466 elapsed=$(($(date +%s) - start_ts))
2467 while [ $elapsed -lt $duration ]; do
2468 fail_index=$((RANDOM%max_index + 1))
2469 if [ $fail_index -eq $max_index ]; then
2472 second_index=$((fail_index + 1))
2474 kill -0 $monitor_pid ||
2475 error "$monitor_pid stopped"
2477 replay_barrier mds$fail_index
2478 replay_barrier mds$second_index
2480 # Increment the number of failovers
2481 num_failovers=$((num_failovers+1))
2482 log "fail mds$fail_index mds$second_index $num_failovers times"
2483 fail mds${fail_index},mds${second_index}
2484 elapsed=$(($(date +%s) - start_ts))
2489 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2490 local clients=${CLIENTS:-$HOSTNAME}
2493 zconf_mount_clients $clients $MOUNT
2496 [ "$SLOW" = "no" ] && duration=180
2497 # set duration to 900 because it takes some time to boot node
2498 [ "$FAILURE_MODE" = HARD ] && duration=900
2503 local start_ts=$(date +%s)
2505 trap cleanup_71a EXIT
2508 $LFS mkdir -i0 -c2 $DIR/$tdir/test
2509 rmdir $DIR/$tdir/test
2513 echo "Started $mkdir_71a_pid"
2515 random_double_fail_mdt 2 $duration $mkdir_71a_pid
2516 kill -0 $mkdir_71a_pid || error "mkdir/rmdir $mkdir_71a_pid stopped"
2521 run_test 71a "mkdir/rmdir striped dir with 2 mdts recovery"
2524 multiop_bg_pause $DIR/$tfile O_tSc ||
2525 error "multiop_bg_pause $DIR/$tfile failed"
2529 replay_barrier $SINGLEMDS
2530 #define OBD_FAIL_LDLM_ENQUEUE_NET 0x302
2531 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
2534 wait $pid || error "multiop pid failed"
2535 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
2538 run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay, close"
2541 multiop_bg_pause $DIR/$tfile O_tSc ||
2542 error "multiop_bg_pause $DIR/$tfile failed"
2546 replay_barrier $SINGLEMDS
2547 #define OBD_FAIL_LDLM_REPLY 0x30c
2548 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
2551 wait $pid || error "multiop pid failed"
2552 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
2555 run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
2559 local clients=${CLIENTS:-$HOSTNAME}
2561 zconf_umount_clients $clients $MOUNT
2563 facet_failover $SINGLEMDS
2564 zconf_mount_clients $clients $MOUNT
2566 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
2567 rm $DIR/$tfile || error "rm $DIR/$tfile failed"
2568 clients_up || error "client evicted: $?"
2571 run_test 74 "Ensure applications don't fail waiting for OST recovery"
2573 remote_dir_check_80() {
2578 diridx=$($GETSTRIPE -M $remote_dir) ||
2579 error "$GETSTRIPE -M $remote_dir failed"
2580 [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
2582 createmany -o $remote_dir/f-%d 20 || error "creation failed"
2583 fileidx=$($GETSTRIPE -M $remote_dir/f-1) ||
2584 error "$GETSTRIPE -M $remote_dir/f-1 failed"
2585 [ $fileidx -eq $MDTIDX ] || error "$fileidx != $MDTIDX"
2591 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2592 ([ $FAILURE_MODE == "HARD" ] &&
2593 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2594 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2598 local remote_dir=$DIR/$tdir/remote_dir
2600 mkdir -p $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2601 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2602 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2603 $LFS mkdir -i $MDTIDX $remote_dir &
2609 wait $CLIENT_PID || error "remote creation failed"
2611 remote_dir_check_80 || error "remote dir check failed"
2612 rm -rf $DIR/$tdir || error "rmdir failed"
2616 run_test 80a "DNE: create remote dir, drop update rep from MDT0, fail MDT0"
2619 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2620 ([ $FAILURE_MODE == "HARD" ] &&
2621 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2622 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2626 local remote_dir=$DIR/$tdir/remote_dir
2628 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2629 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2630 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2631 $LFS mkdir -i $MDTIDX $remote_dir &
2636 fail mds$((MDTIDX + 1))
2638 wait $CLIENT_PID || error "remote creation failed"
2640 remote_dir_check_80 || error "remote dir check failed"
2641 rm -rf $DIR/$tdir || error "rmdir failed"
2645 run_test 80b "DNE: create remote dir, drop update rep from MDT0, fail MDT1"
2648 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2649 ([ $FAILURE_MODE == "HARD" ] &&
2650 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2651 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2655 local remote_dir=$DIR/$tdir/remote_dir
2657 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2658 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2659 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2660 $LFS mkdir -i $MDTIDX $remote_dir &
2666 fail mds$((MDTIDX + 1))
2668 wait $CLIENT_PID || error "remote creation failed"
2670 remote_dir_check_80 || error "remote dir check failed"
2671 rm -rf $DIR/$tdir || error "rmdir failed"
2675 run_test 80c "DNE: create remote dir, drop update rep from MDT1, fail MDT[0,1]"
2678 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2680 local remote_dir=$DIR/$tdir/remote_dir
2682 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2683 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2684 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2685 $LFS mkdir -i $MDTIDX $remote_dir &
2688 # sleep 3 seconds to make sure MDTs are failed after
2689 # lfs mkdir -i has finished on all of MDTs.
2694 fail mds${MDTIDX},mds$((MDTIDX + 1))
2696 wait $CLIENT_PID || error "remote creation failed"
2698 remote_dir_check_80 || error "remote dir check failed"
2699 rm -rf $DIR/$tdir || error "rmdir failed"
2703 run_test 80d "DNE: create remote dir, drop update rep from MDT1, fail 2 MDTs"
2706 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2707 ([ $FAILURE_MODE == "HARD" ] &&
2708 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2709 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2713 local remote_dir=$DIR/$tdir/remote_dir
2715 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2716 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2717 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2718 $LFS mkdir -i $MDTIDX $remote_dir &
2721 # sleep 3 seconds to make sure MDTs are failed after
2722 # lfs mkdir -i has finished on all of MDTs.
2728 wait $CLIENT_PID || error "remote creation failed"
2730 remote_dir_check_80 || error "remote dir check failed"
2731 rm -rf $DIR/$tdir || error "rmdir failed"
2735 run_test 80e "DNE: create remote dir, drop MDT1 rep, fail MDT0"
2738 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2739 ([ $FAILURE_MODE == "HARD" ] &&
2740 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2741 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2744 local remote_dir=$DIR/$tdir/remote_dir
2746 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2747 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2748 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2749 $LFS mkdir -i $MDTIDX $remote_dir &
2753 fail mds$((MDTIDX + 1))
2755 wait $CLIENT_PID || error "remote creation failed"
2757 remote_dir_check_80 || error "remote dir check failed"
2758 rm -rf $DIR/$tdir || error "rmdir failed"
2762 run_test 80f "DNE: create remote dir, drop MDT1 rep, fail MDT1"
2765 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2766 ([ $FAILURE_MODE == "HARD" ] &&
2767 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2768 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2772 local remote_dir=$DIR/$tdir/remote_dir
2774 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2775 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2776 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2777 $LFS mkdir -i $MDTIDX $remote_dir &
2780 # sleep 3 seconds to make sure MDTs are failed after
2781 # lfs mkdir -i has finished on all of MDTs.
2787 fail mds$((MDTIDX + 1))
2789 wait $CLIENT_PID || error "remote creation failed"
2791 remote_dir_check_80 || error "remote dir check failed"
2792 rm -rf $DIR/$tdir || error "rmdir failed"
2796 run_test 80g "DNE: create remote dir, drop MDT1 rep, fail MDT0, then MDT1"
2799 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2801 local remote_dir=$DIR/$tdir/remote_dir
2803 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2804 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2805 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2806 $LFS mkdir -i $MDTIDX $remote_dir &
2809 # sleep 3 seconds to make sure MDTs are failed after
2810 # lfs mkdir -i has finished on all of MDTs.
2815 fail mds${MDTIDX},mds$((MDTIDX + 1))
2817 wait $CLIENT_PID || error "remote dir creation failed"
2819 remote_dir_check_80 || error "remote dir check failed"
2820 rm -rf $DIR/$tdir || error "rmdir failed"
2824 run_test 80h "DNE: create remote dir, drop MDT1 rep, fail 2 MDTs"
2827 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2828 ([ $FAILURE_MODE == "HARD" ] &&
2829 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2830 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2834 local remote_dir=$DIR/$tdir/remote_dir
2836 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2837 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2839 touch $remote_dir || error "touch $remote_dir failed"
2840 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2841 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2846 fail mds$((MDTIDX + 1))
2848 wait $CLIENT_PID || error "rm remote dir failed"
2850 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2852 rm -rf $DIR/$tdir || error "rmdir failed"
2856 run_test 81a "DNE: unlink remote dir, drop MDT0 update rep, fail MDT1"
2859 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2860 ([ $FAILURE_MODE == "HARD" ] &&
2861 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2862 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2865 local remote_dir=$DIR/$tdir/remote_dir
2867 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2868 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2870 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2871 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2878 wait $CLIENT_PID || error "rm remote dir failed"
2880 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2882 rm -rf $DIR/$tdir || error "rmdir failed"
2886 run_test 81b "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0"
2889 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2890 ([ $FAILURE_MODE == "HARD" ] &&
2891 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2892 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2896 local remote_dir=$DIR/$tdir/remote_dir
2898 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2899 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2901 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2902 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2909 fail mds$((MDTIDX + 1))
2911 wait $CLIENT_PID || error "rm remote dir failed"
2913 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2915 rm -rf $DIR/$tdir || error "rmdir failed"
2919 run_test 81c "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0,MDT1"
2922 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2924 local remote_dir=$DIR/$tdir/remote_dir
2926 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2927 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2929 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2930 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2936 fail mds${MDTIDX},mds$((MDTIDX + 1))
2938 wait $CLIENT_PID || error "rm remote dir failed"
2940 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2942 rm -rf $DIR/$tdir || error "rmdir failed"
2946 run_test 81d "DNE: unlink remote dir, drop MDT0 update reply, fail 2 MDTs"
2949 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2950 ([ $FAILURE_MODE == "HARD" ] &&
2951 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2952 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2956 local remote_dir=$DIR/$tdir/remote_dir
2958 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2959 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2961 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2962 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2965 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
2970 wait $CLIENT_PID || error "rm remote dir failed"
2972 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
2974 rm -rf $DIR/$tdir || error "rmdir failed"
2978 run_test 81e "DNE: unlink remote dir, drop MDT1 req reply, fail MDT0"
2981 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2982 ([ $FAILURE_MODE == "HARD" ] &&
2983 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2984 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2988 local remote_dir=$DIR/$tdir/remote_dir
2990 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2991 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2993 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2994 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2999 fail mds$((MDTIDX + 1))
3001 wait $CLIENT_PID || error "rm remote dir failed"
3003 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
3005 rm -rf $DIR/$tdir || error "rmdir failed"
3009 run_test 81f "DNE: unlink remote dir, drop MDT1 req reply, fail MDT1"
3012 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3013 ([ $FAILURE_MODE == "HARD" ] &&
3014 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3015 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3019 local remote_dir=$DIR/$tdir/remote_dir
3021 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3022 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
3024 # OBD_FAIL_MDS_REINT_NET_REP 0x119
3025 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
3032 fail mds$((MDTIDX + 1))
3034 wait $CLIENT_PID || error "rm remote dir failed"
3036 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
3038 rm -rf $DIR/$tdir || error "rmdir failed"
3042 run_test 81g "DNE: unlink remote dir, drop req reply, fail M0, then M1"
3045 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3047 local remote_dir=$DIR/$tdir/remote_dir
3049 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3050 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
3052 # OBD_FAIL_MDS_REINT_NET_REP 0x119
3053 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
3059 fail mds${MDTIDX},mds$((MDTIDX + 1))
3061 wait $CLIENT_PID || error "rm remote dir failed"
3063 stat $remote_dir &>/dev/null && error "$remote_dir still exist!"
3065 rm -rf $DIR/$tdir || error "rmdir failed"
3069 run_test 81h "DNE: unlink remote dir, drop request reply, fail 2 MDTs"
3072 #define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x144
3073 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000144"
3074 createmany -o $DIR/$tfile- 1 &
3078 client_up || client_up || true # reconnect
3080 run_test 84a "stale open during export disconnect"
3082 test_85a() { #bug 16774
3083 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
3085 for i in $(seq 100); do
3086 echo "tag-$i" > $DIR/$tfile-$i
3087 grep -q "tag-$i" $DIR/$tfile-$i || error "f2-$i"
3090 lov_id=$(lctl dl | grep "clilov")
3091 addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $NF}')
3092 count=$(lctl get_param -n \
3093 ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
3094 echo "before recovery: unused locks count = $count"
3098 count2=$(lctl get_param -n \
3099 ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
3100 echo "after recovery: unused locks count = $count2"
3102 if [ $count2 -ge $count ]; then
3103 error "unused locks are not canceled"
3106 run_test 85a "check the cancellation of unused locks during recovery(IBITS)"
3108 test_85b() { #bug 16774
3109 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
3111 if ! combined_mgs_mds ; then
3115 create_pool $FSNAME.$TESTNAME ||
3116 error "unable to create pool $TESTNAME"
3117 do_facet mgs $LCTL pool_add $FSNAME.$TESTNAME $FSNAME-OST0000 ||
3118 error "unable to add pool $TESTNAME"
3120 $SETSTRIPE -c 1 -p $FSNAME.$TESTNAME $DIR
3122 for i in $(seq 100); do
3123 dd if=/dev/urandom of=$DIR/$tfile-$i bs=4096 \
3124 count=32 >/dev/null 2>&1
3127 cancel_lru_locks osc
3129 for i in $(seq 100); do
3130 dd if=$DIR/$tfile-$i of=/dev/null bs=4096 \
3131 count=32 >/dev/null 2>&1
3134 lov_id=$(lctl dl | grep "clilov")
3135 addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $NF}')
3136 count=$(lctl get_param -n \
3137 ldlm.namespaces.*OST0000*$addr.lock_unused_count)
3138 echo "before recovery: unused locks count = $count"
3139 [ $count -ne 0 ] || error "unused locks ($count) should be zero"
3143 count2=$(lctl get_param \
3144 -n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
3145 echo "after recovery: unused locks count = $count2"
3147 do_facet mgs $LCTL pool_remove $FSNAME.$TESTNAME $FSNAME-OST0000 ||
3148 error "unable to remove pool $TESTNAME"
3149 do_facet mgs $LCTL pool_destroy $FSNAME.$TESTNAME ||
3150 error "unable to destroy the pool $TESTNAME"
3152 if ! combined_mgs_mds ; then
3156 if [ $count2 -ge $count ]; then
3157 error "unused locks are not canceled"
3160 run_test 85b "check the cancellation of unused locks during recovery(EXTENT)"
3163 local clients=${CLIENTS:-$HOSTNAME}
3165 zconf_umount_clients $clients $MOUNT
3166 do_facet $SINGLEMDS lctl set_param mdt.${FSNAME}-MDT*.exports.clear=0
3167 remount_facet $SINGLEMDS
3168 zconf_mount_clients $clients $MOUNT
3170 run_test 86 "umount server after clear nid_stats should not hit LBUG"
3173 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
3176 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
3177 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
3178 error "dd to $DIR/$tfile failed"
3179 cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
3180 cancel_lru_locks osc
3182 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
3183 cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
3184 if [ $cksum != $cksum2 ] ; then
3185 error "New checksum $cksum2 does not match original $cksum"
3188 run_test 87a "write replay"
3191 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
3194 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
3195 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
3196 error "dd to $DIR/$tfile failed"
3197 sleep 1 # Give it a chance to flush dirty data
3198 echo TESTTEST | dd of=$DIR/$tfile bs=1 count=8 seek=64
3199 cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
3200 cancel_lru_locks osc
3202 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
3203 cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
3204 if [ $cksum != $cksum2 ] ; then
3205 error "New checksum $cksum2 does not match original $cksum"
3208 run_test 87b "write replay with changed data (checksum resend)"
3210 test_88() { #bug 17485
3211 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3212 mkdir -p $TMP/$tdir || error "mkdir $TMP/$tdir failed"
3214 $SETSTRIPE -i 0 -c 1 $DIR/$tdir || error "$SETSTRIPE"
3217 replay_barrier $SINGLEMDS
3219 # exhaust precreations on ost1
3220 local OST=$(ostname_from_index 0)
3221 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $OST)
3222 local last_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
3223 local next_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
3224 echo "before test: last_id = $last_id, next_id = $next_id"
3226 echo "Creating to objid $last_id on ost $OST..."
3227 createmany -o $DIR/$tdir/f-%d $next_id $((last_id - next_id + 2)) ||
3228 error "createmany create files to last_id failed"
3230 #create some files to use some uncommitted objids
3231 last_id=$(($last_id + 1))
3232 createmany -o $DIR/$tdir/f-%d $last_id 8 ||
3233 error "createmany create files with uncommitted objids failed"
3235 last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
3236 next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
3237 echo "before recovery: last_id = $last_id2, next_id = $next_id2"
3239 # if test uses shutdown_facet && reboot_facet instead of facet_failover ()
3240 # it has to take care about the affected facets, bug20407
3241 local affected_mds1=$(affected_facets mds1)
3242 local affected_ost1=$(affected_facets ost1)
3244 shutdown_facet $SINGLEMDS
3247 reboot_facet $SINGLEMDS
3248 change_active $affected_mds1
3249 wait_for_facet $affected_mds1
3250 mount_facets $affected_mds1 || error "Restart of mds failed"
3253 change_active $affected_ost1
3254 wait_for_facet $affected_ost1
3255 mount_facets $affected_ost1 || error "Restart of ost1 failed"
3259 last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
3260 next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
3261 echo "after recovery: last_id = $last_id2, next_id = $next_id2"
3263 # create new files, which should use new objids, and ensure the orphan
3264 # cleanup phase for ost1 is completed at the same time
3265 for i in $(seq 8); do
3266 file_id=$(($last_id + 10 + $i))
3267 dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
3270 # if the objids were not recreated, then "ls" will fail with -ENOENT
3271 ls -l $DIR/$tdir/* || error "can't get the status of precreated files"
3274 # write into previously created files
3275 for i in $(seq 8); do
3276 file_id=$(($last_id + $i))
3277 dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
3278 cp -f $DIR/$tdir/f-$file_id $TMP/$tdir/
3281 # compare the content
3282 for i in $(seq 8); do
3283 file_id=$(($last_id + $i))
3284 cmp $TMP/$tdir/f-$file_id $DIR/$tdir/f-$file_id ||
3285 error "the content of file is modified!"
3290 run_test 88 "MDS should not assign same objid to different files "
3293 cancel_lru_locks osc
3294 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3295 rm -f $DIR/$tdir/$tfile
3297 wait_delete_completed
3298 BLOCKS1=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
3299 $SETSTRIPE -i 0 -c 1 $DIR/$tdir/$tfile
3300 dd if=/dev/zero bs=1M count=10 of=$DIR/$tdir/$tfile
3303 facet_failover $SINGLEMDS
3304 rm $DIR/$tdir/$tfile
3307 zconf_mount $(hostname) $MOUNT || error "mount fails"
3308 client_up || error "client_up failed"
3310 wait_delete_completed
3311 BLOCKS2=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
3312 [ $((BLOCKS2 - BLOCKS1)) -le 4 ] ||
3313 error $((BLOCKS2 - BLOCKS1)) blocks leaked
3315 run_test 89 "no disk space leak on late ost connection"
3322 change_active $facet
3323 wait_for_facet $facet
3324 mount_facet $facet || error "Restart of $facet failed"
3328 test_90() { # bug 19494
3329 local dir=$DIR/$tdir
3330 local ostfail=$(get_random_entry $(get_facets OST))
3332 if [[ $FAILURE_MODE = HARD ]]; then
3333 local affected=$(affected_facets $ostfail);
3334 if [[ "$affected" != $ostfail ]]; then
3335 skip not functional with FAILURE_MODE=$FAILURE_MODE, affected: $affected
3339 # ensure all OSTs are active to allow allocations
3342 mkdir $dir || error "mkdir $dir failed"
3344 echo "Create the files"
3346 # file "f${index}" striped over 1 OST
3347 # file "all" striped over all OSTs
3349 $SETSTRIPE -c $OSTCOUNT $dir/all ||
3350 error "setstripe failed to create $dir/all"
3352 for (( i=0; i<$OSTCOUNT; i++ )); do
3354 $SETSTRIPE -i $i -c 1 $f || error "$SETSTRIPE failed to create $f"
3356 # confirm setstripe actually created the stripe on the requested OST
3357 local uuid=$(ostuuid_from_index $i)
3358 for file in f$i all; do
3359 if [[ $dir/$file != $($LFS find --obd $uuid --name $file $dir) ]]; then
3360 $GETSTRIPE $dir/$file
3361 error wrong stripe: $file, uuid: $uuid
3366 # Before failing an OST, get its obd name and index
3367 local varsvc=${ostfail}_svc
3368 local obd=$(do_facet $ostfail lctl get_param \
3369 -n obdfilter.${!varsvc}.uuid)
3370 local index=$(($(facet_number $ostfail) - 1))
3372 echo "Fail $ostfail $obd, display the list of affected files"
3373 shutdown_facet $ostfail || error "shutdown_facet $ostfail failed"
3375 trap "cleanup_90 $ostfail" EXIT INT
3376 echo "General Query: lfs find $dir"
3377 local list=$($LFS find $dir)
3379 for (( i=0; i<$OSTCOUNT; i++ )); do
3380 list_member "$list" $dir/f$i ||
3381 error_noexit "lfs find $dir: no file f$i"
3383 list_member "$list" $dir/all ||
3384 error_noexit "lfs find $dir: no file all"
3386 # focus on the missing OST,
3387 # we expect to see only two files affected: "f$(index)" and "all"
3389 echo "Querying files on shutdown $ostfail: lfs find --obd $obd"
3390 list=$($LFS find --obd $obd $dir)
3392 for file in all f$index; do
3393 list_member "$list" $dir/$file ||
3394 error_noexit "lfs find does not report the affected $obd for $file"
3397 [[ $(echo $list | wc -w) -eq 2 ]] ||
3398 error_noexit "lfs find reports the wrong list of affected files ${#list[@]}"
3400 echo "Check getstripe: $GETSTRIPE -r --obd $obd"
3401 list=$($GETSTRIPE -r --obd $obd $dir)
3403 for file in all f$index; do
3404 echo "$list" | grep $dir/$file ||
3405 error_noexit "lfs getsripe does not report the affected $obd for $file"
3410 run_test 90 "lfs find identifies the missing striped file segments"
3413 local server_version=$(lustre_version_code $SINGLEMDS)
3414 [[ $server_version -ge $(version_code 2.6.90) ]] ||
3415 [[ $server_version -ge $(version_code 2.5.4) &&
3416 $server_version -lt $(version_code 2.5.50) ]] ||
3417 { skip "Need MDS version 2.5.4+ or 2.6.90+"; return; }
3419 cancel_lru_locks osc
3421 $SETSTRIPE -i 0 -c 1 $DIR/$tfile ||
3422 error "$SETSTRIPE $DIR/$tfile failed"
3423 dd if=/dev/zero of=$DIR/$tfile bs=1024 count=1 ||
3424 error "dd to $DIR/$tfile failed"
3425 #define OBD_FAIL_TGT_REPLAY_RECONNECT 0x715
3426 # We need to emulate a state that OST is waiting for other clients
3427 # not completing the recovery. Final ping is queued, but reply will be
3428 # sent on the recovery completion. It is done by sleep before
3429 # processing final pings
3430 do_facet ost1 "$LCTL set_param fail_val=40"
3431 do_facet ost1 "$LCTL set_param fail_loc=0x715"
3434 run_test 93a "replay + reconnect"
3437 local server_version=$(lustre_version_code $SINGLEMDS)
3438 [[ $server_version -ge $(version_code 2.7.90) ]] ||
3439 { skip "Need MDS version 2.7.90+"; return; }
3441 cancel_lru_locks mdc
3443 createmany -o $DIR/$tfile 20 ||
3444 error "createmany -o $DIR/$tfile failed"
3446 #define OBD_FAIL_TGT_REPLAY_RECONNECT 0x715
3447 # We need to emulate a state that MDT is waiting for other clients
3448 # not completing the recovery. Final ping is queued, but reply will be
3449 # sent on the recovery completion. It is done by sleep before
3450 # processing final pings
3451 do_facet mds1 "$LCTL set_param fail_val=80"
3452 do_facet mds1 "$LCTL set_param fail_loc=0x715"
3455 run_test 93b "replay + reconnect on mds"
3457 striped_dir_check_100() {
3458 local striped_dir=$DIR/$tdir/striped_dir
3459 local stripe_count=$($LFS getdirstripe -c $striped_dir)
3461 $LFS getdirstripe $striped_dir
3462 [ $stripe_count -eq 2 ] || error "$stripe_count != 2"
3464 createmany -o $striped_dir/f-%d 20 ||
3465 error "creation failed under striped dir"
3469 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3470 ([ $FAILURE_MODE == "HARD" ] &&
3471 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3472 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3475 local striped_dir=$DIR/$tdir/striped_dir
3478 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3480 #To make sure MDT1 and MDT0 are connected
3481 #otherwise it may create single stripe dir here
3482 $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
3484 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
3485 do_facet mds$((MDTIDX+1)) lctl set_param fail_loc=0x1701
3486 $LFS setdirstripe -i0 -c2 $striped_dir &
3489 fail mds$((MDTIDX + 1))
3491 wait $CLIENT_PID || error "striped dir creation failed"
3493 striped_dir_check_100 || error "striped dir check failed"
3494 rm -rf $DIR/$tdir || error "rmdir failed"
3496 run_test 100a "DNE: create striped dir, drop update rep from MDT1, fail MDT1"
3499 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3500 ([ $FAILURE_MODE == "HARD" ] &&
3501 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3502 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3505 local striped_dir=$DIR/$tdir/striped_dir
3508 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3510 #To make sure MDT1 and MDT0 are connected
3511 #otherwise it may create single stripe dir here
3512 $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
3514 # OBD_FAIL_MDS_REINT_NET_REP 0x119
3515 do_facet mds$MDTIDX lctl set_param fail_loc=0x119
3516 $LFS mkdir -i0 -c2 $striped_dir &
3521 wait $CLIENT_PID || error "striped dir creation failed"
3523 striped_dir_check_100 || error "striped dir check failed"
3524 rm -rf $DIR/$tdir || error "rmdir failed"
3526 run_test 100b "DNE: create striped dir, fail MDT0"
3528 test_101() { #LU-5648
3529 mkdir -p $DIR/$tdir/d1
3530 mkdir -p $DIR/$tdir/d2
3531 touch $DIR/$tdir/file0
3534 replay_barrier $SINGLEMDS
3535 for i in $(seq $num) ; do
3536 echo test$i > $DIR/$tdir/d1/file$i
3539 fail_abort $SINGLEMDS
3540 for i in $(seq $num) ; do
3541 touch $DIR/$tdir/d2/file$i
3542 test -s $DIR/$tdir/d2/file$i &&
3543 ls -al $DIR/$tdir/d2/file$i && error "file$i's size > 0"
3548 run_test 101 "Shouldn't reassign precreated objs to other files after recovery"
3557 [[ $(lctl get_param mdc.*.import |
3558 grep "connect_flags:.*multi_mod_rpc") ]] ||
3559 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3561 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3562 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3563 facet="mds$((0x$idx + 1))"
3565 # get current value of max_mod_rcps_in_flight
3566 num=$($LCTL get_param -n \
3567 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3568 # set default value if client does not support multi mod RPCs
3569 [ -z "$num" ] && num=1
3571 echo "creating $num files ..."
3573 for i in $(seq $num); do
3574 touch $DIR/$tdir/file-$i
3577 # drop request on MDT to force resend
3578 #define OBD_FAIL_MDS_REINT_MULTI_NET 0x159
3579 do_facet $facet "$LCTL set_param fail_loc=0x159"
3580 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3581 for i in $(seq $num); do
3582 chmod 0600 $DIR/$tdir/file-$i &
3586 do_facet $facet "$LCTL set_param fail_loc=0"
3587 for pid in $pids; do
3588 wait $pid || error "chmod failed"
3590 echo "done ($(date +%H:%M:%S))"
3592 # check chmod succeed
3593 for i in $(seq $num); do
3594 checkstat -vp 0600 $DIR/$tdir/file-$i
3599 run_test 102a "check resend (request lost) with multiple modify RPCs in flight"
3608 [[ $(lctl get_param mdc.*.import |
3609 grep "connect_flags:.*multi_mod_rpc") ]] ||
3610 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3612 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3613 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3614 facet="mds$((0x$idx + 1))"
3616 # get current value of max_mod_rcps_in_flight
3617 num=$($LCTL get_param -n \
3618 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3619 # set default value if client does not support multi mod RPCs
3620 [ -z "$num" ] && num=1
3622 echo "creating $num files ..."
3624 for i in $(seq $num); do
3625 touch $DIR/$tdir/file-$i
3628 # drop reply on MDT to force reconstruction
3629 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3630 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3631 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3632 for i in $(seq $num); do
3633 chmod 0600 $DIR/$tdir/file-$i &
3637 do_facet $facet "$LCTL set_param fail_loc=0"
3638 for pid in $pids; do
3639 wait $pid || error "chmod failed"
3641 echo "done ($(date +%H:%M:%S))"
3643 # check chmod succeed
3644 for i in $(seq $num); do
3645 checkstat -vp 0600 $DIR/$tdir/file-$i
3650 run_test 102b "check resend (reply lost) with multiple modify RPCs in flight"
3659 [[ $(lctl get_param mdc.*.import |
3660 grep "connect_flags:.*multi_mod_rpc") ]] ||
3661 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3663 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3664 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3665 facet="mds$((0x$idx + 1))"
3667 # get current value of max_mod_rcps_in_flight
3668 num=$($LCTL get_param -n \
3669 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3670 # set default value if client does not support multi mod RPCs
3671 [ -z "$num" ] && num=1
3673 echo "creating $num files ..."
3675 for i in $(seq $num); do
3676 touch $DIR/$tdir/file-$i
3679 replay_barrier $facet
3682 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3683 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3684 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3685 for i in $(seq $num); do
3686 chmod 0600 $DIR/$tdir/file-$i &
3690 do_facet $facet "$LCTL set_param fail_loc=0"
3695 for pid in $pids; do
3696 wait $pid || error "chmod failed"
3698 echo "done ($(date +%H:%M:%S))"
3700 # check chmod succeed
3701 for i in $(seq $num); do
3702 checkstat -vp 0600 $DIR/$tdir/file-$i
3707 run_test 102c "check replay w/o reconstruction with multiple mod RPCs in flight"
3716 [[ $(lctl get_param mdc.*.import |
3717 grep "connect_flags:.*multi_mod_rpc") ]] ||
3718 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3720 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3721 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3722 facet="mds$((0x$idx + 1))"
3724 # get current value of max_mod_rcps_in_flight
3725 num=$($LCTL get_param -n \
3726 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3727 # set default value if client does not support multi mod RPCs
3728 [ -z "$num" ] && num=1
3730 echo "creating $num files ..."
3732 for i in $(seq $num); do
3733 touch $DIR/$tdir/file-$i
3737 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3738 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3739 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3740 for i in $(seq $num); do
3741 chmod 0600 $DIR/$tdir/file-$i &
3746 # write MDT transactions to disk
3747 do_facet $facet "sync; sync; sync"
3749 do_facet $facet "$LCTL set_param fail_loc=0"
3754 for pid in $pids; do
3755 wait $pid || error "chmod failed"
3757 echo "done ($(date +%H:%M:%S))"
3759 # check chmod succeed
3760 for i in $(seq $num); do
3761 checkstat -vp 0600 $DIR/$tdir/file-$i
3766 run_test 102d "check replay & reconstruction with multiple mod RPCs in flight"
3769 remote_mds_nodsh && skip "remote MDS with nodsh" && return
3770 local mds_version=$(lustre_version_code $SINGLEMDS)
3771 [[ $mds_version -gt $(version_code 2.8.54) ]] ||
3772 { skip "Need MDS version 2.8.54+"; return; }
3774 #define OBD_FAIL_MDS_TRACK_OVERFLOW 0x162
3775 do_facet mds1 $LCTL set_param fail_loc=0x80000162
3778 createmany -o $DIR/$tdir/t- 30 ||
3779 error "create files on remote directory failed"
3781 rm -rf $DIR/$tdir/t-*
3783 #MDS should crash with tr->otr_next_id overflow
3786 run_test 103 "Check otr_next_id overflow"
3789 check_striped_dir_110()
3791 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
3792 error "create striped dir failed"
3793 local stripe_count=$($LFS getdirstripe -c $DIR/$tdir/striped_dir)
3794 [ $stripe_count -eq $MDSCOUNT ] ||
3795 error "$stripe_count != 2 after recovery"
3799 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3800 ([ $FAILURE_MODE == "HARD" ] &&
3801 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3802 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3807 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3810 check_striped_dir_110 || error "check striped_dir failed"
3811 rm -rf $DIR/$tdir || error "rmdir failed"
3815 run_test 110a "DNE: create striped dir, fail MDT1"
3818 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3819 ([ $FAILURE_MODE == "HARD" ] &&
3820 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3821 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3826 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3829 zconf_mount $(hostname) $MOUNT
3830 client_up || return 1
3832 check_striped_dir_110 || error "check striped_dir failed"
3834 rm -rf $DIR/$tdir || error "rmdir failed"
3838 run_test 110b "DNE: create striped dir, fail MDT1 and client"
3841 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3842 ([ $FAILURE_MODE == "HARD" ] &&
3843 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3844 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3849 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3852 check_striped_dir_110 || error "check striped_dir failed"
3854 rm -rf $DIR/$tdir || error "rmdir failed"
3858 run_test 110c "DNE: create striped dir, fail MDT2"
3861 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3862 ([ $FAILURE_MODE == "HARD" ] &&
3863 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3864 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3869 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3872 zconf_mount $(hostname) $MOUNT
3873 client_up || return 1
3875 check_striped_dir_110 || error "check striped_dir failed"
3877 rm -rf $DIR/$tdir || error "rmdir failed"
3881 run_test 110d "DNE: create striped dir, fail MDT2 and client"
3884 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3885 ([ $FAILURE_MODE == "HARD" ] &&
3886 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3887 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3892 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3896 zconf_mount $(hostname) $MOUNT
3897 client_up || return 1
3899 check_striped_dir_110 || error "check striped_dir failed"
3901 rm -rf $DIR/$tdir || error "rmdir failed"
3905 run_test 110e "DNE: create striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
3908 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3909 ([ $FAILURE_MODE == "HARD" ] &&
3910 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3911 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3917 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3920 check_striped_dir_110 || error "check striped_dir failed"
3922 rm -rf $DIR/$tdir || error "rmdir failed"
3926 run_test 110f "DNE: create striped dir, fail MDT1/MDT2"
3929 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3930 ([ $FAILURE_MODE == "HARD" ] &&
3931 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3932 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3937 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3941 zconf_mount $(hostname) $MOUNT
3942 client_up || return 1
3944 check_striped_dir_110 || error "check striped_dir failed"
3946 rm -rf $DIR/$tdir || error "rmdir failed"
3950 run_test 110g "DNE: create striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
3953 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3954 ([ $FAILURE_MODE == "HARD" ] &&
3955 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3956 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3960 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3962 rm -rf $DIR/$tdir/striped_dir
3965 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3966 error "striped dir still exists"
3969 run_test 111a "DNE: unlink striped dir, fail MDT1"
3972 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3973 ([ $FAILURE_MODE == "HARD" ] &&
3974 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3975 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3979 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3981 rm -rf $DIR/$tdir/striped_dir
3984 zconf_mount $(hostname) $MOUNT
3985 client_up || return 1
3987 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3988 error "striped dir still exists"
3991 run_test 111b "DNE: unlink striped dir, fail MDT2"
3994 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3995 ([ $FAILURE_MODE == "HARD" ] &&
3996 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3997 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4001 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
4003 rm -rf $DIR/$tdir/striped_dir
4007 zconf_mount $(hostname) $MOUNT
4008 client_up || return 1
4009 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
4010 error "striped dir still exists"
4013 run_test 111c "DNE: unlink striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
4016 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4017 ([ $FAILURE_MODE == "HARD" ] &&
4018 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4019 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4023 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
4025 rm -rf $DIR/$tdir/striped_dir
4029 zconf_mount $(hostname) $MOUNT
4030 client_up || return 1
4031 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
4032 error "striped dir still exists"
4036 run_test 111d "DNE: unlink striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
4039 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4040 ([ $FAILURE_MODE == "HARD" ] &&
4041 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4042 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4046 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
4048 rm -rf $DIR/$tdir/striped_dir
4051 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
4052 error "striped dir still exists"
4055 run_test 111e "DNE: unlink striped dir, uncommit on MDT2, fail MDT1/MDT2"
4058 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4059 ([ $FAILURE_MODE == "HARD" ] &&
4060 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4061 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4065 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
4067 rm -rf $DIR/$tdir/striped_dir
4070 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
4071 error "striped dir still exists"
4074 run_test 111f "DNE: unlink striped dir, uncommit on MDT1, fail MDT1/MDT2"
4077 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4078 ([ $FAILURE_MODE == "HARD" ] &&
4079 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4080 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4084 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
4087 rm -rf $DIR/$tdir/striped_dir
4089 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
4090 error "striped dir still exists"
4093 run_test 111g "DNE: unlink striped dir, fail MDT1/MDT2"
4095 test_112_rename_prepare() {
4096 mkdir -p $DIR/$tdir/src_dir
4097 $LFS mkdir -i 1 $DIR/$tdir/src_dir/src_child ||
4098 error "create remote source failed"
4100 touch $DIR/$tdir/src_dir/src_child/a
4102 $LFS mkdir -i 2 $DIR/$tdir/tgt_dir ||
4103 error "create remote target dir failed"
4105 $LFS mkdir -i 3 $DIR/$tdir/tgt_dir/tgt_child ||
4106 error "create remote target child failed"
4111 $CHECKSTAT -t dir $DIR/$tdir/src_dir/src_child &&
4112 error "src_child still exists after rename"
4114 $CHECKSTAT -t file $DIR/$tdir/tgt_dir/tgt_child/a ||
4115 error "missing file(a) after rename"
4119 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4120 ([ $FAILURE_MODE == "HARD" ] &&
4121 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4122 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4125 test_112_rename_prepare
4128 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4129 error "rename dir cross MDT failed!"
4133 rm -rf $DIR/$tdir || error "rmdir failed"
4135 run_test 112a "DNE: cross MDT rename, fail MDT1"
4138 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4139 ([ $FAILURE_MODE == "HARD" ] &&
4140 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4141 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4144 test_112_rename_prepare
4147 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4148 error "rename dir cross MDT failed!"
4153 rm -rf $DIR/$tdir || error "rmdir failed"
4155 run_test 112b "DNE: cross MDT rename, fail MDT2"
4158 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4159 ([ $FAILURE_MODE == "HARD" ] &&
4160 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4161 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4164 test_112_rename_prepare
4167 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4168 error "rename dir cross MDT failed!"
4173 rm -rf $DIR/$tdir || error "rmdir failed"
4175 run_test 112c "DNE: cross MDT rename, fail MDT3"
4178 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4179 ([ $FAILURE_MODE == "HARD" ] &&
4180 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4181 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4184 test_112_rename_prepare
4187 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4188 error "rename dir cross MDT failed!"
4193 rm -rf $DIR/$tdir || error "rmdir failed"
4195 run_test 112d "DNE: cross MDT rename, fail MDT4"
4198 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4199 ([ $FAILURE_MODE == "HARD" ] &&
4200 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4201 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4204 test_112_rename_prepare
4208 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4209 error "rename dir cross MDT failed!"
4214 rm -rf $DIR/$tdir || error "rmdir failed"
4216 run_test 112e "DNE: cross MDT rename, fail MDT1 and MDT2"
4219 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4220 ([ $FAILURE_MODE == "HARD" ] &&
4221 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4222 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4225 test_112_rename_prepare
4229 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4230 error "rename dir cross MDT failed!"
4235 rm -rf $DIR/$tdir || error "rmdir failed"
4237 run_test 112f "DNE: cross MDT rename, fail MDT1 and MDT3"
4240 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4241 ([ $FAILURE_MODE == "HARD" ] &&
4242 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4243 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4246 test_112_rename_prepare
4250 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4251 error "rename dir cross MDT failed!"
4256 rm -rf $DIR/$tdir || error "rmdir failed"
4258 run_test 112g "DNE: cross MDT rename, fail MDT1 and MDT4"
4261 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4262 ([ $FAILURE_MODE == "HARD" ] &&
4263 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4264 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4267 test_112_rename_prepare
4271 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4272 error "rename dir cross MDT failed!"
4277 rm -rf $DIR/$tdir || error "rmdir failed"
4279 run_test 112h "DNE: cross MDT rename, fail MDT2 and MDT3"
4282 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4283 ([ $FAILURE_MODE == "HARD" ] &&
4284 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4285 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4288 test_112_rename_prepare
4292 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4293 error "rename dir cross MDT failed!"
4298 rm -rf $DIR/$tdir || error "rmdir failed"
4300 run_test 112i "DNE: cross MDT rename, fail MDT2 and MDT4"
4303 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4304 ([ $FAILURE_MODE == "HARD" ] &&
4305 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4306 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4309 test_112_rename_prepare
4313 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4314 error "rename dir cross MDT failed!"
4319 rm -rf $DIR/$tdir || error "rmdir failed"
4321 run_test 112j "DNE: cross MDT rename, fail MDT3 and MDT4"
4324 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4325 ([ $FAILURE_MODE == "HARD" ] &&
4326 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4327 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4330 test_112_rename_prepare
4335 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4336 error "rename dir cross MDT failed!"
4341 rm -rf $DIR/$tdir || error "rmdir failed"
4343 run_test 112k "DNE: cross MDT rename, fail MDT1,MDT2,MDT3"
4346 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4347 ([ $FAILURE_MODE == "HARD" ] &&
4348 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4349 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4352 test_112_rename_prepare
4357 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4358 error "rename dir cross MDT failed!"
4363 rm -rf $DIR/$tdir || error "rmdir failed"
4365 run_test 112l "DNE: cross MDT rename, fail MDT1,MDT2,MDT4"
4368 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4369 ([ $FAILURE_MODE == "HARD" ] &&
4370 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4371 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4374 test_112_rename_prepare
4379 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4380 error "rename dir cross MDT failed!"
4385 rm -rf $DIR/$tdir || error "rmdir failed"
4387 run_test 112m "DNE: cross MDT rename, fail MDT1,MDT3,MDT4"
4390 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4391 ([ $FAILURE_MODE == "HARD" ] &&
4392 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4393 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4396 test_112_rename_prepare
4401 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4402 error "rename dir cross MDT failed!"
4407 rm -rf $DIR/$tdir || error "rmdir failed"
4409 run_test 112n "DNE: cross MDT rename, fail MDT2,MDT3,MDT4"
4412 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4413 ([ $FAILURE_MODE == "HARD" ] &&
4414 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4415 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4423 for ((j=0;j<$((MDSCOUNT));j++)); do
4424 fail_index=$((fail_index+1))
4425 index=$((fail_index % MDSCOUNT))
4426 replay_barrier mds$((index + 1))
4427 for ((i=0;i<5;i++)); do
4428 test_mkdir -i$index -c$MDSCOUNT $DIR/$tdir/test_$i ||
4429 error "create striped dir $DIR/$tdir/test_$i"
4432 fail mds$((index + 1))
4433 for ((i=0;i<5;i++)); do
4434 checkstat -t dir $DIR/$tdir/test_$i ||
4435 error "$DIR/$tdir/test_$i does not exist!"
4437 rm -rf $DIR/$tdir/test_* ||
4441 run_test 115 "failover for create/unlink striped directory"
4444 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4445 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.55) ] &&
4446 skip "Do not support large update log before 2.7.55" &&
4448 ([ $FAILURE_MODE == "HARD" ] &&
4449 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4450 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4457 # OBD_FAIL_SPLIT_UPDATE_REC 0x1702
4458 do_facet mds1 "lctl set_param fail_loc=0x80001702"
4459 $LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
4462 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
4463 error "stried_dir does not exists"
4465 run_test 116a "large update log master MDT recovery"
4468 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4469 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.55) ] &&
4470 skip "Do not support large update log before 2.7.55" &&
4473 ([ $FAILURE_MODE == "HARD" ] &&
4474 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4475 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4482 # OBD_FAIL_SPLIT_UPDATE_REC 0x1702
4483 do_facet mds2 "lctl set_param fail_loc=0x80001702"
4484 $LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
4487 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
4488 error "stried_dir does not exists"
4490 run_test 116b "large update log slave MDT recovery"
4493 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4494 ([ $FAILURE_MODE == "HARD" ] &&
4495 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4496 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4502 $LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/remote_dir
4503 $LFS setdirstripe -i1 -c$MDSCOUNT $DIR/$tdir/remote_dir_1
4506 # Let's set rdonly on all MDTs, so client will send
4507 # replay requests on all MDTs and replay these requests
4508 # at the same time. This test will verify the recovery
4509 # will not be deadlock in this case, LU-7531.
4510 for ((index = 0; index < $((MDSCOUNT)); index++)); do
4511 replay_barrier mds$((index + 1))
4512 if [ -z $mds_indexs ]; then
4513 mds_indexs="${mds_indexs}mds$((index+1))"
4515 mds_indexs="${mds_indexs},mds$((index+1))"
4519 rm -rf $DIR/$tdir/remote_dir
4520 rm -rf $DIR/$tdir/remote_dir_1
4524 rm -rf $DIR/$tdir || error "rmdir failed"
4526 run_test 117 "DNE: cross MDT unlink, fail MDT1 and MDT2"
4529 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4530 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
4531 skip "Do not support large update log before 2.7.64" &&
4536 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir ||
4537 error "setdirstripe fails"
4538 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir1 ||
4539 error "setdirstripe fails 1"
4540 rm -rf $DIR/$tdir/striped_dir* || error "rmdir fails"
4542 # OBD_FAIL_INVALIDATE_UPDATE 0x1705
4543 do_facet mds1 "lctl set_param fail_loc=0x1705"
4544 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir
4545 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir1
4546 do_facet mds1 "lctl set_param fail_loc=0x0"
4549 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir
4550 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir1
4555 run_test 118 "invalidate osp update will not cause update log corruption"
4558 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4559 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
4560 skip "Do not support large update log before 2.7.64" &&
4563 local hard_timeout=$(do_facet mds1 \
4564 "lctl get_param -n mdt.$FSNAME-MDT0000.recovery_time_hard")
4566 local clients=${CLIENTS:-$HOSTNAME}
4567 local time_min=$(recovery_time_min)
4570 mkdir $DIR/$tdir/tmp
4571 rmdir $DIR/$tdir/tmp
4574 mkdir $DIR/$tdir/dir_1
4575 for ((i = 0; i < 20; i++)); do
4576 $LFS setdirstripe -c2 $DIR/$tdir/stripe_dir-$i
4583 #define OBD_FAIL_TGT_REPLAY_DELAY 0x714
4584 do_facet mds1 $LCTL set_param fail_loc=0x80000714
4585 #sleep (timeout + 5), so mds will evict the client exports,
4586 #but DNE update recovery will keep going.
4587 do_facet mds1 $LCTL set_param fail_val=$((time_min + 5))
4589 mount_facet mds1 "-o recovery_time_hard=$time_min"
4591 wait_clients_import_state "$clients" mds1 FULL
4593 clients_up || clients_up || error "failover df: $?"
4595 #revert back the hard timeout
4596 do_facet mds1 $LCTL set_param \
4597 mdt.$FSNAME-MDT0000.recovery_time_hard=$hard_timeout
4599 for ((i = 0; i < 20; i++)); do
4600 stripe_count=$($LFS getdirstripe -c $DIR/$tdir/stripe_dir-$i)
4601 [ $stripe_count == 2 ] || {
4602 error "stripe_dir-$i creation replay fails"
4607 run_test 119 "timeout of normal replay does not cause DNE replay fails "
4610 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4611 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
4612 skip "Do not support large update log before 2.7.64" &&
4616 replay_barrier_nosync mds1
4617 for ((i = 0; i < 20; i++)); do
4618 mkdir $DIR/$tdir/dir-$i || {
4619 error "create dir-$i fails"
4622 $LFS setdirstripe -c2 $DIR/$tdir/stripe_dir-$i || {
4623 error "create stripe_dir-$i fails"
4630 for ((i = 0; i < 20; i++)); do
4631 [ ! -e "$DIR/$tdir/dir-$i" ] || {
4632 error "dir-$i still exists"
4635 [ ! -e "$DIR/$tdir/stripe_dir-$i" ] || {
4636 error "stripe_dir-$i still exists"
4641 run_test 120 "DNE fail abort should stop both normal and DNE replay"
4644 check_and_cleanup_lustre