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=${GRANT_CHECK_LIST:-""}
21 require_dsh_mds || exit 0
24 # bug number for skipped tests:
26 ALWAYS_EXCEPT="$REPLAY_SINGLE_EXCEPT 61d"
27 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
29 case "$(lsb_release -sr)" in # only disable tests for el7
30 7*) # bug number: LU-6455-----
31 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 28"
35 # 63 min 7 min AT AT AT AT"
36 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a 44b 65 66 67 68"
38 [ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
39 # bug number for skipped test: LU-1867 LU-3127
40 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 89 73b"
44 check_and_setup_lustre
49 rm -rf $DIR/[df][0-9]* $DIR/f.$TESTSUITE.*
51 # LU-482 Avert LVM and VM inability to flush caches in pre .33 kernels
52 if [ $LINUX_VERSION_CODE -lt $(version_code 2.6.33) ]; then
54 do_facet $SINGLEMDS sync
57 test_0a() { # was test_0
58 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
59 replay_barrier $SINGLEMDS
63 run_test 0a "empty replay"
66 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
68 # this test attempts to trigger a race in the precreation code,
69 # and must run before any other objects are created on the filesystem
71 createmany -o $DIR/$tfile 20 || error "createmany -o $DIR/$tfile failed"
72 unlinkmany $DIR/$tfile 20 || error "unlinkmany $DIR/$tfile failed"
74 run_test 0b "ensure object created after recover exists. (3284)"
77 replay_barrier $SINGLEMDS
80 facet_failover $SINGLEMDS
81 zconf_mount $(hostname) $MOUNT || error "mount fails"
82 client_up || error "post-failover df failed"
83 # file shouldn't exist if replay-barrier works as expected
84 rm $DIR/$tfile && error "File exists and it shouldn't"
87 run_test 0c "check replay-barrier"
90 replay_barrier $SINGLEMDS
92 facet_failover $SINGLEMDS
93 zconf_mount $(hostname) $MOUNT || error "mount fails"
94 client_up || error "post-failover df failed"
96 run_test 0d "expired recovery with no clients"
99 replay_barrier $SINGLEMDS
102 $CHECKSTAT -t file $DIR/$tfile ||
103 error "$CHECKSTAT $DIR/$tfile attribute check failed"
106 run_test 1 "simple create"
109 replay_barrier $SINGLEMDS
112 $CHECKSTAT -t file $DIR/$tfile ||
113 error "$CHECKSTAT $DIR/$tfile attribute check failed"
119 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
120 replay_barrier $SINGLEMDS
123 $CHECKSTAT -t file $DIR/$tfile ||
124 error "$CHECKSTAT $DIR/$tfile attribute check failed"
130 local file=$DIR/$tfile
131 replay_barrier $SINGLEMDS
133 openfile -f O_DIRECTORY $file
135 $CHECKSTAT -t file $file ||
136 error "$CHECKSTAT $file attribute check failed"
139 run_test 3a "replay failed open(O_DIRECTORY)"
142 replay_barrier $SINGLEMDS
143 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
144 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000114"
146 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
148 $CHECKSTAT -t file $DIR/$tfile &&
149 error "$CHECKSTAT $DIR/$tfile attribute check should fail"
152 run_test 3b "replay failed open -ENOMEM"
155 replay_barrier $SINGLEMDS
156 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
157 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
159 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
162 $CHECKSTAT -t file $DIR/$tfile &&
163 error "$CHECKSTAT $DIR/$tfile attribute check should fail"
166 run_test 3c "replay failed open -ENOMEM"
168 test_4a() { # was test_4
169 replay_barrier $SINGLEMDS
170 for i in $(seq 10); do
171 echo "tag-$i" > $DIR/$tfile-$i
174 for i in $(seq 10); do
175 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
178 run_test 4a "|x| 10 open(O_CREAT)s"
181 for i in $(seq 10); do
182 echo "tag-$i" > $DIR/$tfile-$i
184 replay_barrier $SINGLEMDS
187 $CHECKSTAT -t file $DIR/$tfile-* &&
188 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
191 run_test 4b "|x| rm 10 files"
193 # The idea is to get past the first block of precreated files on both
194 # osts, and then replay.
196 replay_barrier $SINGLEMDS
197 for i in $(seq 220); do
198 echo "tag-$i" > $DIR/$tfile-$i
201 for i in $(seq 220); do
202 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
206 # waiting for commitment of removal
208 run_test 5 "|x| 220 open(O_CREAT)"
210 test_6a() { # was test_6
211 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
212 replay_barrier $SINGLEMDS
213 mcreate $DIR/$tdir/$tfile
215 $CHECKSTAT -t dir $DIR/$tdir ||
216 error "$CHECKSTAT $DIR/$tdir attribute check failed"
217 $CHECKSTAT -t file $DIR/$tdir/$tfile ||
218 error "$CHECKSTAT $DIR/$tdir/$tfile attribute check failed"
220 # waiting for log process thread
222 run_test 6a "mkdir + contained create"
225 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
226 replay_barrier $SINGLEMDS
229 $CHECKSTAT -t dir $DIR/$tdir &&
230 error "$CHECKSTAT $DIR/$tdir attribute check should fail" ||
233 run_test 6b "|X| rmdir"
236 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
237 replay_barrier $SINGLEMDS
238 mcreate $DIR/$tdir/$tfile
240 $CHECKSTAT -t dir $DIR/$tdir ||
241 error "$CHECKSTAT $DIR/$tdir attribute check failed"
242 $CHECKSTAT -t file $DIR/$tdir/$tfile ||
243 error "$CHECKSTAT $DIR/$tdir/$tfile attribute check failed"
246 run_test 7 "mkdir |X| contained create"
249 replay_barrier $SINGLEMDS
250 multiop_bg_pause $DIR/$tfile mo_c ||
251 error "multiop mknod $DIR/$tfile failed"
255 $CHECKSTAT -t file $DIR/$tfile ||
256 error "$CHECKSTAT $DIR/$tfile attribute check failed"
257 kill -USR1 $MULTIPID || error "multiop mknod $MULTIPID not running"
258 wait $MULTIPID || error "multiop mknod $MULTIPID failed"
261 run_test 8 "creat open |X| close"
264 replay_barrier $SINGLEMDS
266 local old_inum=$(ls -i $DIR/$tfile | awk '{print $1}')
268 local new_inum=$(ls -i $DIR/$tfile | awk '{print $1}')
270 echo " old_inum == $old_inum, new_inum == $new_inum"
271 if [ $old_inum -eq $new_inum ] ;
273 echo "old_inum and new_inum match"
275 echo " old_inum and new_inum do not match"
276 error "old index($old_inum) does not match new index($new_inum)"
280 run_test 9 "|X| create (same inum/gen)"
283 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
284 replay_barrier $SINGLEMDS
285 mv $DIR/$tfile $DIR/$tfile-2
288 $CHECKSTAT $DIR/$tfile &&
289 error "$CHECKSTAT $DIR/$tfile attribute check should fail"
290 $CHECKSTAT $DIR/$tfile-2 ||
291 error "$CHECKSTAT $DIR/$tfile-2 attribute check failed"
295 run_test 10 "create |X| rename unlink"
298 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
299 echo "old" > $DIR/$tfile
300 mv $DIR/$tfile $DIR/$tfile-2
301 replay_barrier $SINGLEMDS
302 echo "new" > $DIR/$tfile
304 grep old $DIR/$tfile-2
306 grep new $DIR/$tfile || error "grep $DIR/$tfile failed"
307 grep old $DIR/$tfile-2 || error "grep $DIR/$tfile-2 failed"
309 run_test 11 "create open write rename |X| create-old-name read"
312 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
313 multiop_bg_pause $DIR/$tfile o_tSc ||
314 error "multiop_bg_pause $DIR/$tfile failed"
317 replay_barrier $SINGLEMDS
318 kill -USR1 $pid || error "multiop $pid not running"
319 wait $pid || error "multiop $pid failed"
322 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
325 run_test 12 "open, unlink |X| close"
327 # 1777 - replay open after committed chmod that would make
328 # a regular open a failure
330 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
331 multiop_bg_pause $DIR/$tfile O_wc ||
332 error "multiop_bg_pause $DIR/$tfile failed"
335 $CHECKSTAT -p 0 $DIR/$tfile ||
336 error "$CHECKSTAT $DIR/$tfile attribute check failed"
337 replay_barrier $SINGLEMDS
339 kill -USR1 $pid || error "multiop $pid not running"
340 wait $pid || error "multiop $pid failed"
342 $CHECKSTAT -s 1 -p 0 $DIR/$tfile ||
343 error "second $CHECKSTAT $DIR/$tfile attribute check failed"
344 rm $DIR/$tfile || error "rm $DIR/$tfile failed"
347 run_test 13 "open chmod 0 |x| write close"
350 multiop_bg_pause $DIR/$tfile O_tSc ||
351 error "multiop_bg_pause $DIR/$tfile failed"
354 replay_barrier $SINGLEMDS
355 kill -USR1 $pid || error "multiop $pid not running"
356 wait $pid || error "multiop $pid failed"
359 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
362 run_test 14 "open(O_CREAT), unlink |X| close"
365 multiop_bg_pause $DIR/$tfile O_tSc ||
366 error "multiop_bg_pause $DIR/$tfile failed"
369 replay_barrier $SINGLEMDS
370 touch $DIR/$tfile-1 || error "touch $DIR/$tfile-1 failed"
371 kill -USR1 $pid || error "multiop $pid not running"
372 wait $pid || error "multiop $pid failed"
375 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
376 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
379 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
382 replay_barrier $SINGLEMDS
385 mcreate $DIR/$tfile-2
387 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
388 [ -e $DIR/$tfile-2 ] || error "file $DIR/$tfile-2 does not exist"
389 munlink $DIR/$tfile-2 || error "munlink $DIR/$tfile-2 failed"
391 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
394 replay_barrier $SINGLEMDS
395 multiop_bg_pause $DIR/$tfile O_c ||
396 error "multiop_bg_pause $DIR/$tfile failed"
399 kill -USR1 $pid || error "multiop $pid not running"
400 wait $pid || error "multiop $pid failed"
401 $CHECKSTAT -t file $DIR/$tfile ||
402 error "$CHECKSTAT $DIR/$tfile attribute check failed"
405 run_test 17 "|X| open(O_CREAT), |replay| close"
408 replay_barrier $SINGLEMDS
409 multiop_bg_pause $DIR/$tfile O_tSc ||
410 error "multiop_bg_pause $DIR/$tfile failed"
413 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
414 echo "pid: $pid will close"
415 kill -USR1 $pid || error "multiop $pid not running"
416 wait $pid || error "multiop $pid failed"
419 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
420 [ -e $DIR/$tfile-2 ] || error "file $DIR/$tfile-2 does not exist"
421 # this touch frequently fails
422 touch $DIR/$tfile-3 || error "touch $DIR/$tfile-3 failed"
423 munlink $DIR/$tfile-2 || error "munlink $DIR/$tfile-2 failed"
424 munlink $DIR/$tfile-3 || error "munlink $DIR/$tfile-3 failed"
427 run_test 18 "open(O_CREAT), unlink, touch new, close, touch, unlink"
429 # bug 1855 (a simpler form of test_11 above)
431 replay_barrier $SINGLEMDS
433 echo "old" > $DIR/$tfile
434 mv $DIR/$tfile $DIR/$tfile-2
435 grep old $DIR/$tfile-2
437 grep old $DIR/$tfile-2 || error "grep $DIR/$tfile-2 failed"
439 run_test 19 "mcreate, open, write, rename "
441 test_20a() { # was test_20
442 replay_barrier $SINGLEMDS
443 multiop_bg_pause $DIR/$tfile O_tSc ||
444 error "multiop_bg_pause $DIR/$tfile failed"
449 kill -USR1 $pid || error "multiop $pid not running"
450 wait $pid || error "multiop $pid failed"
451 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
454 run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
456 test_20b() { # bug 10480
457 local wait_timeout=$((TIMEOUT * 4))
461 BEFOREUSED=$(df -P $DIR | tail -1 | awk '{ print $3 }')
462 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
463 while [ ! -e $DIR/$tfile ] ; do
464 usleep 60 # give dd a chance to start
467 $GETSTRIPE $DIR/$tfile || error "$GETSTRIPE $DIR/$tfile failed"
469 rm -f $DIR/$tfile || error "rm -f $DIR/$tfile failed"
471 client_up || client_up || true # reconnect
473 do_facet $SINGLEMDS "lctl set_param -n osd*.*MDT*.force_sync 1"
475 fail $SINGLEMDS # start orphan recovery
476 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
477 wait_delete_completed_mds $wait_timeout ||
478 error "delete did not complete"
480 AFTERUSED=$(df -P $DIR | tail -1 | awk '{ print $3 }')
481 log "before $BEFOREUSED, after $AFTERUSED"
482 (( $AFTERUSED > $BEFOREUSED + $(fs_log_size) )) &&
483 error "after $AFTERUSED > before $BEFOREUSED"
486 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
488 test_20c() { # bug 10480
489 multiop_bg_pause $DIR/$tfile Ow_c ||
490 error "multiop_bg_pause $DIR/$tfile failed"
496 client_up || client_up || true # reconnect
498 kill -USR1 $pid || error "multiop $pid not running"
499 wait $pid || error "multiop $pid failed"
500 [ -s $DIR/$tfile ] || error "File was truncated"
504 run_test 20c "check that client eviction does not affect file content"
507 replay_barrier $SINGLEMDS
508 multiop_bg_pause $DIR/$tfile O_tSc ||
509 error "multiop_bg_pause $DIR/$tfile failed"
512 touch $DIR/$tfile-1 || error "touch $DIR/$tfile-1 failed"
515 kill -USR1 $pid || error "multiop $pid not running"
516 wait $pid || error "multiop $pid failed"
517 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
518 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
521 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
524 multiop_bg_pause $DIR/$tfile O_tSc ||
525 error "multiop_bg_pause $DIR/$tfile failed"
528 replay_barrier $SINGLEMDS
532 kill -USR1 $pid || error "multiop $pid not running"
533 wait $pid || error "multiop $pid failed"
534 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
537 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
540 multiop_bg_pause $DIR/$tfile O_tSc ||
541 error "multiop_bg_pause $DIR/$tfile failed"
544 replay_barrier $SINGLEMDS
546 touch $DIR/$tfile-1 || error "touch $DIR/$tfile-1 failed"
549 kill -USR1 $pid || error "multiop $pid not running"
550 wait $pid || error "multiop $pid failed"
551 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
552 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
555 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
558 multiop_bg_pause $DIR/$tfile O_tSc ||
559 error "multiop_bg_pause $DIR/$tfile failed"
562 replay_barrier $SINGLEMDS
565 kill -USR1 $pid || error "multiop $pid not running"
566 wait $pid || error "multiop $pid failed"
567 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
570 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
573 multiop_bg_pause $DIR/$tfile O_tSc ||
574 error "multiop_bg_pause $DIR/$tfile failed"
578 replay_barrier $SINGLEMDS
580 kill -USR1 $pid || error "multiop $pid not running"
581 wait $pid || error "multiop $pid failed"
582 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
585 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
588 replay_barrier $SINGLEMDS
589 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
590 error "multiop_bg_pause $DIR/$tfile-1 failed"
592 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
593 error "multiop_bg_pause $DIR/$tfile-2 failed"
597 kill -USR1 $pid2 || error "second multiop $pid2 not running"
598 wait $pid2 || error "second multiop $pid2 failed"
601 kill -USR1 $pid1 || error "multiop $pid1 not running"
602 wait $pid1 || error "multiop $pid1 failed"
603 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
604 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
607 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
610 replay_barrier $SINGLEMDS
611 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
612 error "multiop_bg_pause $DIR/$tfile-1 failed"
614 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
615 error "multiop_bg_pause $DIR/$tfile-2 failed"
621 kill -USR1 $pid1 || error "multiop $pid1 not running"
622 wait $pid1 || error "multiop $pid1 failed"
623 kill -USR1 $pid2 || error "second multiop $pid2 not running"
624 wait $pid2 || error "second multiop $pid2 failed"
625 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
626 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
629 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
632 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
633 error "multiop_bg_pause $DIR/$tfile-1 failed"
635 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
636 error "multiop_bg_pause $DIR/$tfile-2 failed"
638 replay_barrier $SINGLEMDS
641 kill -USR1 $pid2 || error "second multiop $pid2 not running"
642 wait $pid2 || error "second multiop $pid2 failed"
645 kill -USR1 $pid1 || error "multiop $pid1 not running"
646 wait $pid1 || error "multiop $pid1 failed"
647 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
648 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
651 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
654 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
655 error "multiop_bg_pause $DIR/$tfile-1 failed"
657 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
658 error "multiop_bg_pause $DIR/$tfile-2 failed"
660 replay_barrier $SINGLEMDS
665 kill -USR1 $pid1 || error "multiop $pid1 not running"
666 wait $pid1 || error "multiop $pid1 failed"
667 kill -USR1 $pid2 || error "second multiop $pid2 not running"
668 wait $pid2 || error "second multiop $pid2 failed"
669 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
670 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
673 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
676 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
677 error "multiop_bg_pause $DIR/$tfile-1 failed"
679 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
680 error "multiop_bg_pause $DIR/$tfile-2 failed"
685 replay_barrier $SINGLEMDS
687 kill -USR1 $pid1 || error "multiop $pid1 not running"
688 wait $pid1 || error "multiop $pid1 failed"
689 kill -USR1 $pid2 || error "second multiop $pid2 not running"
690 wait $pid2 || error "second multiop $pid2 failed"
691 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
692 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
695 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
698 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
699 error "multiop_bg_pause $DIR/$tfile-1 failed"
701 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
702 error "multiop_bg_pause $DIR/$tfile-2 failed"
706 replay_barrier $SINGLEMDS
709 kill -USR1 $pid1 || error "multiop $pid1 not running"
710 wait $pid1 || error "multiop $pid1 failed"
711 kill -USR1 $pid2 || error "second multiop $pid2 not running"
712 wait $pid2 || error "second multiop $pid2 failed"
713 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
714 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
717 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
719 # tests for bug 2104; completion without crashing is success. The close is
720 # stale, but we always return 0 for close, so the app never sees it.
722 multiop_bg_pause $DIR/$tfile O_c ||
723 error "multiop_bg_pause $DIR/$tfile failed"
725 multiop_bg_pause $DIR/$tfile O_c ||
726 error "second multiop_bg_pause $DIR/$tfile failed"
729 client_up || client_up || error "client_up failed"
730 kill -USR1 $pid1 || error "multiop $pid1 not running"
731 kill -USR1 $pid2 || error "second multiop $pid2 not running"
732 wait $pid1 || error "multiop $pid1 failed"
733 wait $pid2 || error "second multiop $pid2 failed"
736 run_test 32 "close() notices client eviction; close() after client eviction"
739 createmany -o $DIR/$tfile-%d 10 ||
740 error "createmany create $DIR/$tfile failed"
741 replay_barrier_nosync $SINGLEMDS
742 fail_abort $SINGLEMDS
743 # recreate shouldn't fail
744 createmany -o $DIR/$tfile--%d 10 ||
745 error "createmany recreate $DIR/$tfile failed"
749 run_test 33a "fid seq shouldn't be reused after abort recovery"
752 #define OBD_FAIL_SEQ_ALLOC 0x1311
753 do_facet $SINGLEMDS "lctl set_param fail_loc=0x1311"
755 createmany -o $DIR/$tfile-%d 10
756 replay_barrier_nosync $SINGLEMDS
757 fail_abort $SINGLEMDS
758 # recreate shouldn't fail
759 createmany -o $DIR/$tfile--%d 10 ||
760 error "createmany recreate $DIR/$tfile failed"
764 run_test 33b "test fid seq allocation"
767 multiop_bg_pause $DIR/$tfile O_c ||
768 error "multiop_bg_pause $DIR/$tfile failed"
772 replay_barrier $SINGLEMDS
773 fail_abort $SINGLEMDS
774 kill -USR1 $pid || error "multiop $pid not running"
775 wait $pid || error "multiop $pid failed"
776 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
780 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
782 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
784 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
786 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
787 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
792 # give a chance to remove from MDS
793 fail_abort $SINGLEMDS
794 $CHECKSTAT -t file $DIR/$tfile &&
795 error "$CHECKSTAT $DIR/$tfile attribute check should fail" ||
798 run_test 35 "test recovery from llog for unlink op"
800 # b=2432 resent cancel after replay uses wrong cookie,
801 # so don't resend cancels
803 replay_barrier $SINGLEMDS
805 checkstat $DIR/$tfile
806 facet_failover $SINGLEMDS
808 if dmesg | grep "unknown lock cookie"; then
809 error "cancel after replay failed"
812 run_test 36 "don't resend cancel"
815 # directory orphans can't be unlinked from PENDING directory
817 rmdir $DIR/$tfile 2>/dev/null
818 multiop_bg_pause $DIR/$tfile dD_c ||
819 error "multiop_bg_pause $DIR/$tfile failed"
823 replay_barrier $SINGLEMDS
824 # clear the dmesg buffer so we only see errors from this recovery
825 do_facet $SINGLEMDS dmesg -c >/dev/null
826 fail_abort $SINGLEMDS
827 kill -USR1 $pid || error "multiop $pid not running"
828 do_facet $SINGLEMDS dmesg | grep "error .* unlinking .* from PENDING" &&
829 error "error unlinking files"
830 wait $pid || error "multiop $pid failed"
834 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
837 createmany -o $DIR/$tfile-%d 800 ||
838 error "createmany -o $DIR/$tfile failed"
839 unlinkmany $DIR/$tfile-%d 0 400 || error "unlinkmany $DIR/$tfile failed"
840 replay_barrier $SINGLEMDS
842 unlinkmany $DIR/$tfile-%d 400 400 ||
843 error "unlinkmany $DIR/$tfile 400 failed"
845 $CHECKSTAT -t file $DIR/$tfile-* &&
846 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
849 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
851 test_39() { # bug 4176
852 createmany -o $DIR/$tfile-%d 800 ||
853 error "createmany -o $DIR/$tfile failed"
854 replay_barrier $SINGLEMDS
855 unlinkmany $DIR/$tfile-%d 0 400
857 unlinkmany $DIR/$tfile-%d 400 400 ||
858 error "unlinkmany $DIR/$tfile 400 failed"
860 $CHECKSTAT -t file $DIR/$tfile-* &&
861 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
864 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
867 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
872 # always need connection to MDS to verify layout during IO. LU-2628.
873 lctl get_param mdc.*.connect_flags | grep -q layout_lock &&
874 skip "layout_lock needs MDS connection for IO" && return 0
876 $LCTL mark multiop $MOUNT/$tfile OS_c
877 multiop $MOUNT/$tfile OS_c &
879 writeme -s $MOUNT/${tfile}-2 &
882 facet_failover $SINGLEMDS
883 #define OBD_FAIL_MDS_CONNECT_NET 0x117
884 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
886 stat1=$(count_ost_writes)
888 stat2=$(count_ost_writes)
889 echo "$stat1, $stat2"
890 if [ $stat1 -lt $stat2 ]; then
891 echo "writes continuing during recovery"
894 echo "writes not continuing during recovery, bug 2477"
897 echo "waiting for writeme $WRITE_PID"
901 echo "waiting for multiop $PID"
902 wait $PID || error "multiop $PID failed"
903 do_facet client munlink $MOUNT/$tfile ||
904 error "munlink $MOUNT/$tfile failed"
905 do_facet client munlink $MOUNT/${tfile}-2 ||
906 error "munlink $MOUNT/$tfile-2 failed"
909 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
912 # make sure that a read to one osc doesn't try to double-unlock its page just
913 # because another osc is invalid. trigger_group_io used to mistakenly return
914 # an error if any oscs were invalid even after having successfully put rpcs
915 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
916 # the page, guarnateeing that the unlock from the RPC completion would
917 # assert on trying to unlock the unlocked page.
919 [ $OSTCOUNT -lt 2 ] &&
920 skip_env "skipping test 41: we don't have a second OST to test with" &&
923 local f=$MOUNT/$tfile
924 # make sure the start of the file is ost1
925 $SETSTRIPE -S $((128 * 1024)) -i 0 $f
926 do_facet client dd if=/dev/zero of=$f bs=4k count=1 ||
927 error "dd on client failed"
929 # fail ost2 and read from ost1
930 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $ost2_svc)
931 local osc2dev=$(do_facet $SINGLEMDS "lctl get_param -n devices" |
932 grep $mdtosc | awk '{print $1}')
933 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" &&
934 lctl get_param -n devices &&
935 error "OST 2 $osc2dev does not exist"
936 do_facet $SINGLEMDS $LCTL --device $osc2dev deactivate ||
937 error "deactive device on $SINGLEMDS failed"
938 do_facet client dd if=$f of=/dev/null bs=4k count=1 ||
939 error "second dd on client failed"
940 do_facet $SINGLEMDS $LCTL --device $osc2dev activate ||
941 error "active device on $SINGLEMDS failed"
944 run_test 41 "read from a valid osc while other oscs are invalid"
946 # test MDS recovery after ost failure
948 blocks=$(df -P $MOUNT | tail -n 1 | awk '{ print $2 }')
949 createmany -o $DIR/$tfile-%d 800 ||
950 error "createmany -o $DIR/$tfile failed"
952 unlinkmany $DIR/$tfile-%d 0 400
954 lctl set_param debug=-1
957 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
958 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
959 #[ $blocks_after -lt $blocks ] || return 1
960 echo "wait for MDS to timeout and recover"
961 sleep $((TIMEOUT * 2))
963 unlinkmany $DIR/$tfile-%d 400 400 ||
964 error "unlinkmany $DIR/$tfile 400 failed"
965 $CHECKSTAT -t file $DIR/$tfile-* &&
966 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
969 run_test 42 "recovery after ost failure"
971 # timeout in MDS/OST recovery RPC will LBUG MDS
972 test_43() { # bug 2530
973 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
975 replay_barrier $SINGLEMDS
977 # OBD_FAIL_OST_CREATE_NET 0x204
978 do_facet ost1 "lctl set_param fail_loc=0x80000204"
984 run_test 43 "mds osc import failure during recovery; don't LBUG"
986 test_44a() { # was test_44
989 local mdcdev=$($LCTL dl |
990 awk "/${FSNAME}-MDT0000-mdc-/ {if (\$2 == \"UP\") {print \$1}}")
991 [ "$mdcdev" ] || error "${FSNAME}-MDT0000-mdc- not UP"
992 [ $(echo $mdcdev | wc -w) -eq 1 ] ||
993 { $LCTL dl; error "looking for mdcdev=$mdcdev"; }
995 # adaptive timeouts slow this way down
996 if at_is_enabled; then
997 at_max_saved=$(at_max_get mds)
1001 for i in $(seq 1 10); do
1002 echo "$i of 10 ($(date +%s))"
1003 do_facet $SINGLEMDS \
1004 "lctl get_param -n md[ts].*.mdt.timeouts | grep service"
1005 #define OBD_FAIL_TGT_CONN_RACE 0x701
1006 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
1007 # lctl below may fail, it is valid case
1008 $LCTL --device $mdcdev recover
1011 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1012 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
1015 run_test 44a "race in target handle connect"
1018 local mdcdev=$($LCTL dl |
1019 awk "/${FSNAME}-MDT0000-mdc-/ {if (\$2 == \"UP\") {print \$1}}")
1020 [ "$mdcdev" ] || error "${FSNAME}-MDT0000-mdc not up"
1021 [ $(echo $mdcdev | wc -w) -eq 1 ] ||
1022 { echo mdcdev=$mdcdev; $LCTL dl;
1023 error "more than one ${FSNAME}-MDT0000-mdc"; }
1025 for i in $(seq 1 10); do
1026 echo "$i of 10 ($(date +%s))"
1027 do_facet $SINGLEMDS \
1028 "lctl get_param -n md[ts].*.mdt.timeouts | grep service"
1029 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
1030 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
1031 # lctl below may fail, it is valid case
1032 $LCTL --device $mdcdev recover
1037 run_test 44b "race in target handle connect"
1040 replay_barrier $SINGLEMDS
1041 createmany -m $DIR/$tfile-%d 100 || error "failed to create directories"
1042 #define OBD_FAIL_TGT_RCVG_FLAG 0x712
1043 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000712"
1044 fail_abort $SINGLEMDS
1045 unlinkmany $DIR/$tfile-%d 100 && error "unliked after fail abort"
1047 unlinkmany $DIR/$tfile-%d 100 && error "unliked after fail"
1050 run_test 44c "race in target handle connect"
1052 # Handle failed close
1054 local mdcdev=$($LCTL get_param -n devices |
1055 awk "/ ${FSNAME}-MDT0000-mdc-/ {print \$1}")
1056 [ "$mdcdev" ] || error "${FSNAME}-MDT0000-mdc not up"
1057 [ $(echo $mdcdev | wc -w) -eq 1 ] ||
1058 { echo mdcdev=$mdcdev; $LCTL dl;
1059 error "more than one ${FSNAME}-MDT0000-mdc"; }
1061 $LCTL --device $mdcdev recover ||
1062 error "$LCTL --device $mdcdev recover failed"
1064 multiop_bg_pause $DIR/$tfile O_c ||
1065 error "multiop_bg_pause $DIR/$tfile failed"
1068 # This will cause the CLOSE to fail before even
1069 # allocating a reply buffer
1070 $LCTL --device $mdcdev deactivate ||
1071 error "$LCTL --device $mdcdev deactivate failed"
1074 kill -USR1 $pid || error "multiop $pid not running"
1075 wait $pid || error "multiop $pid failed"
1077 $LCTL --device $mdcdev activate ||
1078 error "$LCTL --device $mdcdev activate failed"
1081 $CHECKSTAT -t file $DIR/$tfile ||
1082 error "$CHECKSTAT $DIR/$tfile attribute check failed"
1085 run_test 45 "Handle failed close"
1089 drop_reply "touch $DIR/$tfile"
1091 # ironically, the previous test, 45, will cause a real forced close,
1092 # so just look for one for this test
1093 dmesg | grep -i "force closing client file handle for $tfile" &&
1094 error "found force closing in dmesg"
1097 run_test 46 "Don't leak file handle after open resend (3325)"
1099 test_47() { # bug 2824
1100 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1102 # create some files to make sure precreate has been done on all
1103 # OSTs. (just in case this test is run independently)
1104 createmany -o $DIR/$tfile 20 ||
1105 error "createmany create $DIR/$tfile failed"
1107 # OBD_FAIL_OST_CREATE_NET 0x204
1109 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1110 client_up || error "client_up failed"
1112 # let the MDS discover the OST failure, attempt to recover, fail
1113 # and recover again.
1114 sleep $((3 * TIMEOUT))
1116 # Without 2824, this createmany would hang
1117 createmany -o $DIR/$tfile 20 ||
1118 error "createmany recraete $DIR/$tfile failed"
1119 unlinkmany $DIR/$tfile 20 || error "unlinkmany $DIR/$tfile failed"
1123 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1126 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1127 [ "$OSTCOUNT" -lt "2" ] &&
1128 skip_env "$OSTCOUNT < 2 OSTs -- skipping" && return
1130 replay_barrier $SINGLEMDS
1131 createmany -o $DIR/$tfile 20 ||
1132 error "createmany -o $DIR/$tfile failed"
1133 # OBD_FAIL_OST_EROFS 0x216
1134 facet_failover $SINGLEMDS
1135 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1136 client_up || error "client_up failed"
1138 createmany -o $DIR/$tfile 20 20 ||
1139 error "createmany recraete $DIR/$tfile failed"
1140 unlinkmany $DIR/$tfile 40 || error "unlinkmany $DIR/$tfile failed"
1143 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1146 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $ost1_svc)
1147 local oscdev=$(do_facet $SINGLEMDS "lctl get_param -n devices" |
1148 grep $mdtosc | awk '{print $1}')
1149 [ "$oscdev" ] || error "could not find OSC device on MDS"
1150 do_facet $SINGLEMDS $LCTL --device $oscdev recover ||
1151 error "OSC device $oscdev recovery failed"
1152 do_facet $SINGLEMDS $LCTL --device $oscdev recover ||
1153 error "second OSC device $oscdev recovery failed"
1154 # give the mds_lov_sync threads a chance to run
1157 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1159 # b3764 timed out lock replay
1161 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ] &&
1162 skip "MDS prior to 2.6.90 handle LDLM_REPLY_NET incorrectly" &&
1165 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1166 cancel_lru_locks mdc
1168 multiop_bg_pause $DIR/$tfile s_s || error "multiop $DIR/$tfile failed"
1171 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
1172 lctl set_param -n ldlm.cancel_unused_locks_before_replay "0"
1173 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000157"
1175 fail $SINGLEMDS || error "fail $SINGLEMDS failed"
1177 wait $mpid || error "multiop_bg_pause pid failed"
1179 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1180 lctl set_param fail_loc=0x0
1181 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
1184 run_test 52 "time out lock replay (3764)"
1186 # bug 3462 - simultaneous MDC requests
1188 [[ $(lctl get_param mdc.*.import |
1189 grep "connect_flags:.*multi_mod_rpc") ]] ||
1190 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
1192 cancel_lru_locks mdc # cleanup locks from former test cases
1193 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1194 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1195 multiop $DIR/${tdir}-1/f O_c &
1197 # give multiop a change to open
1200 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1201 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1202 kill -USR1 $close_pid
1203 cancel_lru_locks mdc # force the close
1204 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1206 mcreate $DIR/${tdir}-2/f || error "mcreate $DIR/${tdir}-2/f failed"
1208 # close should still be here
1209 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1211 replay_barrier_nodf $SINGLEMDS
1213 wait $close_pid || error "close_pid $close_pid failed"
1215 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1216 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1217 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1218 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1219 rm -rf $DIR/${tdir}-*
1221 run_test 53a "|X| close request while two MDC requests in flight"
1224 cancel_lru_locks mdc # cleanup locks from former test cases
1226 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1227 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1228 multiop_bg_pause $DIR/${tdir}-1/f O_c ||
1229 error "multiop_bg_pause $DIR/${tdir}-1/f failed"
1232 #define OBD_FAIL_MDS_REINT_NET 0x107
1233 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1234 mcreate $DIR/${tdir}-2/f &
1238 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1239 kill -USR1 $close_pid
1240 cancel_lru_locks mdc # force the close
1241 wait $close_pid || error "close_pid $close_pid failed"
1242 # open should still be here
1243 [ -d /proc/$open_pid ] || error "open_pid doesn't exist"
1245 replay_barrier_nodf $SINGLEMDS
1247 wait $open_pid || error "open_pid failed"
1249 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1250 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1251 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1252 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1253 rm -rf $DIR/${tdir}-*
1255 run_test 53b "|X| open request while two MDC requests in flight"
1258 cancel_lru_locks mdc # cleanup locks from former test cases
1260 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1261 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1262 multiop $DIR/${tdir}-1/f O_c &
1265 #define OBD_FAIL_MDS_REINT_NET 0x107
1266 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1267 mcreate $DIR/${tdir}-2/f &
1271 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1272 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1273 kill -USR1 $close_pid
1274 cancel_lru_locks mdc # force the close
1276 #bz20647: make sure all pids exist before failover
1277 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1278 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1279 replay_barrier_nodf $SINGLEMDS
1280 fail_nodf $SINGLEMDS
1281 wait $open_pid || error "open_pid failed"
1283 # close should be gone
1284 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1285 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1287 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1288 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1289 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1290 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1291 rm -rf $DIR/${tdir}-*
1293 run_test 53c "|X| open request and close request while two MDC requests in flight"
1296 [[ $(lctl get_param mdc.*.import |
1297 grep "connect_flags:.*multi_mod_rpc") ]] ||
1298 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
1300 cancel_lru_locks mdc # cleanup locks from former test cases
1302 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1303 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1304 multiop $DIR/${tdir}-1/f O_c &
1306 # give multiop a chance to open
1309 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1310 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1311 kill -USR1 $close_pid
1312 cancel_lru_locks mdc # force the close
1313 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1314 mcreate $DIR/${tdir}-2/f || error "mcreate $DIR/${tdir}-2/f failed"
1316 # close should still be here
1317 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1319 wait $close_pid || error "close_pid failed"
1321 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1322 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1323 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1324 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1325 rm -rf $DIR/${tdir}-*
1327 run_test 53d "close reply while two MDC requests in flight"
1330 cancel_lru_locks mdc # cleanup locks from former test cases
1332 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1333 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1334 multiop $DIR/${tdir}-1/f O_c &
1337 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1338 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1339 mcreate $DIR/${tdir}-2/f &
1343 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1344 kill -USR1 $close_pid
1345 cancel_lru_locks mdc # force the close
1346 wait $close_pid || error "close_pid failed"
1347 # open should still be here
1348 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1350 replay_barrier_nodf $SINGLEMDS
1352 wait $open_pid || error "open_pid failed"
1354 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1355 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1356 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1357 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1358 rm -rf $DIR/${tdir}-*
1360 run_test 53e "|X| open reply while two MDC requests in flight"
1363 cancel_lru_locks mdc # cleanup locks from former test cases
1365 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1366 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1367 multiop $DIR/${tdir}-1/f O_c &
1370 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1371 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1372 mcreate $DIR/${tdir}-2/f &
1376 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1377 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1378 kill -USR1 $close_pid
1379 cancel_lru_locks mdc # force the close
1381 #bz20647: make sure all pids are exists before failover
1382 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1383 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1384 replay_barrier_nodf $SINGLEMDS
1385 fail_nodf $SINGLEMDS
1386 wait $open_pid || error "open_pid failed"
1388 # close should be gone
1389 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1390 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1392 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1393 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1394 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1395 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1396 rm -rf $DIR/${tdir}-*
1398 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1401 cancel_lru_locks mdc # cleanup locks from former test cases
1403 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1404 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1405 multiop $DIR/${tdir}-1/f O_c &
1408 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1409 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1410 mcreate $DIR/${tdir}-2/f &
1414 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1415 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1416 kill -USR1 $close_pid
1417 cancel_lru_locks mdc # force the close
1418 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1420 #bz20647: make sure all pids are exists before failover
1421 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1422 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1423 replay_barrier_nodf $SINGLEMDS
1424 fail_nodf $SINGLEMDS
1425 wait $open_pid || error "open_pid failed"
1427 # close should be gone
1428 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1430 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1431 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1432 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1433 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1434 rm -rf $DIR/${tdir}-*
1436 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1439 cancel_lru_locks mdc # cleanup locks from former test cases
1441 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1442 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1443 multiop $DIR/${tdir}-1/f O_c &
1446 #define OBD_FAIL_MDS_REINT_NET 0x107
1447 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1448 mcreate $DIR/${tdir}-2/f &
1452 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1453 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1454 kill -USR1 $close_pid
1455 cancel_lru_locks mdc # force the close
1458 #bz20647: make sure all pids are exists before failover
1459 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1460 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1461 replay_barrier_nodf $SINGLEMDS
1462 fail_nodf $SINGLEMDS
1463 wait $open_pid || error "open_pid failed"
1465 # close should be gone
1466 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1467 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1469 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1470 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1471 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1472 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1473 rm -rf $DIR/${tdir}-*
1475 run_test 53h "open request and close reply while two MDC requests in flight"
1477 #b3761 ASSERTION(hash != 0) failed
1479 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1480 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1482 # give touch a chance to run
1484 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1488 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1490 #b3440 ASSERTION(rec->ur_fid2->id) failed
1492 ln -s foo $DIR/$tfile
1493 replay_barrier $SINGLEMDS
1494 #drop_reply "cat $DIR/$tfile"
1498 run_test 56 "don't replay a symlink open request (3440)"
1500 #recovery one mds-ost setattr from llog
1502 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1503 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1504 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1505 replay_barrier $SINGLEMDS
1508 $CHECKSTAT -t file $DIR/$tfile ||
1509 error "$CHECKSTAT $DIR/$tfile attribute check failed"
1510 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1513 run_test 57 "test recovery from llog for setattr op"
1516 zconf_umount $(hostname) $MOUNT2
1520 #recovery many mds-ost setattr from llog
1522 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1523 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1524 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1525 createmany -o $DIR/$tdir/$tfile-%d 2500
1526 replay_barrier $SINGLEMDS
1529 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null ||
1530 error "$CHECKSTAT $DIR/$tfile-* attribute check failed"
1531 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1532 unlinkmany $DIR/$tdir/$tfile-%d 2500 ||
1533 error "unlinkmany $DIR/$tfile failed"
1536 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
1542 trap cleanup_58 EXIT
1544 large_xattr_enabled &&
1545 orig="$(generate_string $(max_xattr_size))" || orig="bar"
1546 # Original extended attribute can be long. Print a small version of
1547 # attribute if an error occurs
1548 local sm_msg=$(printf "%.9s" $orig)
1550 mount_client $MOUNT2 || error "mount_client on $MOUNT2 failed"
1551 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1552 touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
1553 replay_barrier $SINGLEMDS
1554 setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile
1556 new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
1557 [[ "$new" = "$orig" ]] ||
1558 error "xattr set ($sm_msg...) differs from xattr get ($new)"
1559 rm -f $DIR/$tdir/$tfile
1562 wait_clients_import_state ${CLIENTS:-$HOSTNAME} "mgs" FULL
1564 run_test 58b "test replay of setxattr op"
1566 test_58c() { # bug 16570
1571 trap cleanup_58 EXIT
1573 if large_xattr_enabled; then
1574 local xattr_size=$(max_xattr_size)
1575 orig="$(generate_string $((xattr_size / 2)))"
1576 orig1="$(generate_string $xattr_size)"
1582 # PING_INTERVAL max(obd_timeout / 4, 1U)
1583 sleep $((TIMEOUT / 4))
1585 # Original extended attribute can be long. Print a small version of
1586 # attribute if an error occurs
1587 local sm_msg=$(printf "%.9s" $orig)
1588 local sm_msg1=$(printf "%.9s" $orig1)
1590 mount_client $MOUNT2 || error "mount_client on $MOUNT2 failed"
1591 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1592 touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
1593 drop_request "setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile" ||
1594 error "drop_request for setfattr failed"
1595 new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
1596 [[ "$new" = "$orig" ]] ||
1597 error "xattr set ($sm_msg...) differs from xattr get ($new)"
1598 drop_reint_reply "setfattr -n trusted.foo1 \
1599 -v $orig1 $DIR/$tdir/$tfile" ||
1600 error "drop_reint_reply for setfattr failed"
1601 new=$(get_xattr_value trusted.foo1 $MOUNT2/$tdir/$tfile)
1602 [[ "$new" = "$orig1" ]] ||
1603 error "second xattr set ($sm_msg1...) differs xattr get ($new)"
1604 rm -f $DIR/$tdir/$tfile
1608 run_test 58c "resend/reconstruct setxattr op"
1610 # log_commit_thread vs filter_destroy race used to lead to import use after free
1613 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1615 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1616 createmany -o $DIR/$tdir/$tfile-%d 200 ||
1617 error "createmany create files failed"
1619 unlinkmany $DIR/$tdir/$tfile-%d 200 ||
1620 error "unlinkmany $DIR/$tdir/$tfile failed"
1621 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1622 do_facet ost1 "lctl set_param fail_loc=0x507"
1625 do_facet ost1 "lctl set_param fail_loc=0x0"
1629 run_test 59 "test log_commit_thread vs filter_destroy race"
1631 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1632 # bug 12086: should no oops and No ctxt error for this test
1634 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1635 createmany -o $DIR/$tdir/$tfile-%d 200 ||
1636 error "createmany create files failed"
1637 replay_barrier $SINGLEMDS
1638 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1640 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1641 local no_ctxt=$(dmesg | grep "No ctxt")
1642 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1644 run_test 60 "test llog post recovery init vs llog unlink"
1646 #test race llog recovery thread vs llog cleanup
1647 test_61a() { # was test_61
1648 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1650 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1651 createmany -o $DIR/$tdir/$tfile-%d 800 ||
1652 error "createmany create files failed"
1654 unlinkmany $DIR/$tdir/$tfile-%d 800
1655 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1656 set_nodes_failloc "$(osts_nodes)" 0x80000221
1661 set_nodes_failloc "$(osts_nodes)" 0x0
1663 $CHECKSTAT -t file $DIR/$tdir/$tfile-* &&
1664 error "$CHECKSTAT $DIR/$tdir/$tfile attribute check should fail"
1667 run_test 61a "test race llog recovery vs llog cleanup"
1669 #test race mds llog sync vs llog cleanup
1671 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1672 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
1673 facet_failover $SINGLEMDS
1676 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 ||
1679 run_test 61b "test race mds llog sync vs llog cleanup"
1681 #test race cancel cookie cb vs llog cleanup
1683 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1685 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1686 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1687 set_nodes_failloc "$(osts_nodes)" 0x80000222
1691 set_nodes_failloc "$(osts_nodes)" 0x0
1693 run_test 61c "test race mds llog sync vs llog cleanup"
1695 test_61d() { # bug 16002 # bug 17466 # bug 22137
1696 # OBD_FAIL_OBD_LLOG_SETUP 0x605
1698 do_facet mgs "lctl set_param fail_loc=0x80000605"
1699 start mgs $(mgsdevname) $MGS_MOUNT_OPTS &&
1700 error "mgs start should have failed"
1701 do_facet mgs "lctl set_param fail_loc=0"
1702 start mgs $(mgsdevname) $MGS_MOUNT_OPTS || error "cannot restart mgs"
1704 run_test 61d "error in llog_setup should cleanup the llog context correctly"
1706 test_62() { # Bug 15756 - don't mis-drop resent replay
1707 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1708 replay_barrier $SINGLEMDS
1709 createmany -o $DIR/$tdir/$tfile- 25 ||
1710 error "createmany create files failed"
1711 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1712 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1714 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1715 unlinkmany $DIR/$tdir/$tfile- 25 ||
1716 error "unlinkmany $DIR/$tdir/$tfile failed"
1719 run_test 62 "don't mis-drop resent replay"
1721 #Adaptive Timeouts (bug 3055)
1729 echo "Cleaning up AT ..."
1730 if [ -n "$ATOLDBASE" ]; then
1731 local at_history=$($LCTL get_param -n at_history)
1732 do_facet $SINGLEMDS "lctl set_param at_history=$at_history" || true
1733 do_facet ost1 "lctl set_param at_history=$at_history" || true
1736 if [ $AT_MAX_SET -ne 0 ]; then
1737 for facet in mds client ost; do
1738 var=AT_MAX_SAVE_${facet}
1739 echo restore AT on $facet to saved value ${!var}
1740 at_max_set ${!var} $facet
1741 at_new=$(at_max_get $facet)
1742 echo Restored AT value on $facet $at_new
1743 [ $at_new -eq ${!var} ] ||
1744 error "AT value not restored SAVED ${!var} NEW $at_new"
1751 local at_max_new=600
1753 # Save at_max original values
1755 if [ $AT_MAX_SET -eq 0 ]; then
1756 # Suppose that all osts have the same at_max
1757 for facet in mds client ost; do
1758 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1762 for facet in mds client ost; do
1763 at_max=$(at_max_get $facet)
1764 if [ $at_max -ne $at_max_new ]; then
1765 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1766 at_max_set $at_max_new $facet
1771 if [ -z "$ATOLDBASE" ]; then
1772 ATOLDBASE=$(do_facet $SINGLEMDS "lctl get_param -n at_history")
1773 # speed up the timebase so we can check decreasing AT
1774 do_facet $SINGLEMDS "lctl set_param at_history=8" || true
1775 do_facet ost1 "lctl set_param at_history=8" || true
1777 # sleep for a while to cool down, should be > 8s and also allow
1778 # at least one ping to be sent. simply use TIMEOUT to be safe.
1783 test_65a() #bug 3055
1785 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1787 at_start || return 0
1788 $LCTL dk > /dev/null
1790 $LCTL set_param debug="other"
1791 # Slow down a request to the current service time, this is critical
1792 # because previous tests may have caused this value to increase.
1793 REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
1794 awk '/portal 12/ {print $5}'`
1795 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1797 do_facet $SINGLEMDS lctl set_param fail_val=$((${REQ_DELAY} * 1000))
1798 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1799 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x8000050a
1800 createmany -o $DIR/$tfile 10 > /dev/null
1801 unlinkmany $DIR/$tfile 10 > /dev/null
1802 # check for log message
1803 $LCTL dk | grep "Early reply #" || error "No early reply"
1805 # client should show REQ_DELAY estimates
1806 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1808 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1810 run_test 65a "AT: verify early replies"
1812 test_65b() #bug 3055
1814 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1816 at_start || return 0
1819 $LCTL set_param debug="other trace"
1820 $LCTL dk > /dev/null
1821 # Slow down a request to the current service time, this is critical
1822 # because previous tests may have caused this value to increase.
1823 $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tfile
1824 multiop $DIR/$tfile Ow1yc
1825 REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
1826 awk '/portal 6/ {print $5}'`
1827 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1829 do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
1830 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1831 do_facet ost1 $LCTL set_param fail_loc=0x224
1834 $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tfile
1835 # force some real bulk transfer
1836 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1838 do_facet ost1 $LCTL set_param fail_loc=0
1839 # check for log message
1840 $LCTL dk | grep "Early reply #" || error "No early reply"
1842 # client should show REQ_DELAY estimates
1843 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1845 run_test 65b "AT: verify early replies on packed reply / bulk"
1847 test_66a() #bug 3055
1849 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1851 at_start || return 0
1852 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1853 # adjust 5s at a time so no early reply is sent (within deadline)
1854 do_facet $SINGLEMDS "$LCTL set_param fail_val=5000"
1855 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1856 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x8000050a"
1857 createmany -o $DIR/$tfile 20 > /dev/null
1858 unlinkmany $DIR/$tfile 20 > /dev/null
1859 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1860 do_facet $SINGLEMDS "$LCTL set_param fail_val=10000"
1861 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x8000050a"
1862 createmany -o $DIR/$tfile 20 > /dev/null
1863 unlinkmany $DIR/$tfile 20 > /dev/null
1864 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1865 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0"
1867 createmany -o $DIR/$tfile 20 > /dev/null
1868 unlinkmany $DIR/$tfile 20 > /dev/null
1869 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1870 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1871 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1872 echo "Current MDT timeout $CUR, worst $WORST"
1873 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1875 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1877 test_66b() #bug 3055
1879 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1881 at_start || return 0
1882 ORIG=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1883 awk '/network/ {print $4}')
1884 $LCTL set_param fail_val=$(($ORIG + 5))
1885 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1886 $LCTL set_param fail_loc=0x50c
1887 ls $DIR/$tfile > /dev/null 2>&1
1888 $LCTL set_param fail_loc=0
1889 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1890 awk '/network/ {print $4}')
1891 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1892 awk '/network/ {print $6}')
1893 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1894 [ $WORST -gt $ORIG ] ||
1895 error "Worst $WORST should be worse than orig $ORIG"
1897 run_test 66b "AT: verify net latency adjusts"
1899 test_67a() #bug 3055
1901 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1903 at_start || return 0
1904 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1905 # sleeping threads may drive values above this
1906 do_facet ost1 "$LCTL set_param fail_val=400"
1907 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1908 do_facet ost1 "$LCTL set_param fail_loc=0x50a"
1909 createmany -o $DIR/$tfile 20 > /dev/null
1910 unlinkmany $DIR/$tfile 20 > /dev/null
1911 do_facet ost1 "$LCTL set_param fail_loc=0"
1912 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1913 ATTEMPTS=$(($CONN2 - $CONN1))
1914 echo "$ATTEMPTS osc reconnect attempts on gradual slow"
1915 [ $ATTEMPTS -gt 0 ] &&
1916 error_ignore bz13721 "AT should have prevented reconnect"
1919 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1921 test_67b() #bug 3055
1923 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1925 at_start || return 0
1926 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1928 # exhaust precreations on ost1
1929 local OST=$(ostname_from_index 0)
1930 local mdtosc=$(get_mdtosc_proc_path mds $OST)
1931 local last_id=$(do_facet $SINGLEMDS lctl get_param -n \
1932 osc.$mdtosc.prealloc_last_id)
1933 local next_id=$(do_facet $SINGLEMDS lctl get_param -n \
1934 osc.$mdtosc.prealloc_next_id)
1936 mkdir -p $DIR/$tdir/${OST} || error "mkdir $DIR/$tdir/${OST} failed"
1937 $SETSTRIPE -i 0 -c 1 $DIR/$tdir/${OST} || error "$SETSTRIPE failed"
1938 echo "Creating to objid $last_id on ost $OST..."
1939 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1940 do_facet ost1 "$LCTL set_param fail_val=20000"
1941 do_facet ost1 "$LCTL set_param fail_loc=0x80000223"
1942 createmany -o $DIR/$tdir/${OST}/f $next_id $((last_id - next_id + 2))
1945 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1947 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1948 ATTEMPTS=$(($CONN2 - $CONN1))
1949 echo "$ATTEMPTS osc reconnect attempts on instant slow"
1950 # do it again; should not timeout
1951 do_facet ost1 "$LCTL set_param fail_loc=0x80000223"
1952 cp /etc/profile $DIR/$tfile || error "cp failed"
1953 do_facet ost1 "$LCTL set_param fail_loc=0"
1955 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1956 CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1957 ATTEMPTS=$(($CONN3 - $CONN2))
1958 echo "$ATTEMPTS osc reconnect attempts on 2nd slow"
1959 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1962 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1964 test_68 () #bug 13813
1966 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1968 at_start || return 0
1969 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1970 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1971 local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
1972 [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
1973 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1974 local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
1975 echo $TIMEOUT >> $ldlm_enqueue_min
1976 do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
1978 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1979 $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tdir
1980 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1981 $LCTL set_param fail_val=$(($TIMEOUT - 1))
1982 $LCTL set_param fail_loc=0x80000312
1983 cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
1984 $LCTL set_param fail_val=$((TIMEOUT * 5 / 4))
1985 $LCTL set_param fail_loc=0x80000312
1986 cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
1987 $LCTL set_param fail_loc=0
1989 echo $ENQ_MIN >> $ldlm_enqueue_min
1990 do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
1994 run_test 68 "AT: verify slowing locks"
1997 # end of AT tests includes above lines
1999 # start multi-client tests
2001 [ -z "$CLIENTS" ] &&
2002 { skip "Need two or more clients." && return; }
2003 [ $CLIENTCOUNT -lt 2 ] &&
2004 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
2006 echo "mount clients $CLIENTS ..."
2007 zconf_mount_clients $CLIENTS $MOUNT
2009 local clients=${CLIENTS//,/ }
2010 echo "Write/read files on $DIR ; clients $CLIENTS ... "
2011 for CLIENT in $clients; do
2012 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
2013 of=$DIR/${tfile}_${CLIENT} 2>/dev/null ||
2014 error "dd failed on $CLIENT"
2017 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
2018 for C in ${CLIENTS//,/ }; do
2019 do_node $prev_client dd if=$DIR/${tfile}_${C} \
2020 of=/dev/null 2>/dev/null ||
2021 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
2027 run_test 70a "check multi client t-f"
2029 check_for_process () {
2034 killall_process $clients "$prog" -0
2038 local clients=${CLIENTS:-$HOSTNAME}
2040 zconf_mount_clients $clients $MOUNT
2043 [ "$SLOW" = "no" ] && duration=120
2044 # set duration to 900 because it takes some time to boot node
2045 [ "$FAILURE_MODE" = HARD ] && duration=900
2048 local start_ts=$(date +%s)
2049 local cmd="rundbench 1 -t $duration"
2051 if [ $MDSCOUNT -ge 2 ]; then
2052 test_mkdir -p -c$MDSCOUNT $DIR/$tdir
2053 $LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir
2055 do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
2056 PATH=\$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
2057 DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
2058 MOUNT=$MOUNT DIR=$DIR/$tdir/\\\$(hostname) LCTL=$LCTL $cmd" &
2061 #LU-1897 wait for all dbench copies to start
2062 while ! check_for_process $clients dbench; do
2063 elapsed=$(($(date +%s) - start_ts))
2064 if [ $elapsed -gt $duration ]; then
2065 killall_process $clients dbench
2066 error "dbench failed to start on $clients!"
2071 log "Started rundbench load pid=$pid ..."
2073 elapsed=$(($(date +%s) - start_ts))
2074 local num_failovers=0
2076 while [ $elapsed -lt $duration ]; do
2077 if ! check_for_process $clients dbench; then
2078 error_noexit "dbench stopped on some of $clients!"
2079 killall_process $clients dbench
2083 replay_barrier mds$fail_index
2084 sleep 1 # give clients a time to do operations
2085 # Increment the number of failovers
2086 num_failovers=$((num_failovers+1))
2087 log "$TESTNAME fail mds$fail_index $num_failovers times"
2089 elapsed=$(($(date +%s) - start_ts))
2090 if [ $fail_index -ge $MDSCOUNT ]; then
2093 fail_index=$((fail_index+1))
2097 wait $pid || error "rundbench load on $clients failed!"
2099 run_test 70b "dbench ${MDSCOUNT}mdts recovery; $CLIENTCOUNT clients"
2100 # end multi-client tests
2103 multiop_bg_pause $DIR/$tfile O_tSc ||
2104 error "multiop_bg_pause $DIR/$tfile failed"
2108 replay_barrier $SINGLEMDS
2109 #define OBD_FAIL_LDLM_ENQUEUE_NET 0x302
2110 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
2113 wait $pid || error "multiop pid failed"
2114 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
2117 run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay, close"
2120 multiop_bg_pause $DIR/$tfile O_tSc ||
2121 error "multiop_bg_pause $DIR/$tfile failed"
2125 replay_barrier $SINGLEMDS
2126 #define OBD_FAIL_LDLM_REPLY 0x30c
2127 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
2130 wait $pid || error "multiop pid failed"
2131 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
2134 run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
2138 local clients=${CLIENTS:-$HOSTNAME}
2140 zconf_umount_clients $clients $MOUNT
2142 facet_failover $SINGLEMDS
2143 zconf_mount_clients $clients $MOUNT
2145 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
2146 rm $DIR/$tfile || error "rm $DIR/$tfile failed"
2147 clients_up || error "client evicted: $?"
2150 run_test 74 "Ensure applications don't fail waiting for OST recovery"
2152 remote_dir_check_80() {
2157 diridx=$($GETSTRIPE -M $remote_dir) ||
2158 error "$GETSTRIPE -M $remote_dir failed"
2159 [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
2161 createmany -o $remote_dir/f-%d 20 || error "creation failed"
2162 fileidx=$($GETSTRIPE -M $remote_dir/f-1) ||
2163 error "$GETSTRIPE -M $remote_dir/f-1 failed"
2164 [ $fileidx -eq $MDTIDX ] || error "$fileidx != $MDTIDX"
2170 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2171 ([ $FAILURE_MODE == "HARD" ] &&
2172 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2173 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2177 local remote_dir=$DIR/$tdir/remote_dir
2179 mkdir -p $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2180 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2181 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2182 $LFS mkdir -i $MDTIDX $remote_dir &
2188 wait $CLIENT_PID || error "remote creation failed"
2190 remote_dir_check_80 || error "remote dir check failed"
2191 rm -rf $DIR/$tdir || error "rmdir failed"
2195 run_test 80a "DNE: create remote dir, drop update rep from MDT0, fail MDT0"
2198 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2199 ([ $FAILURE_MODE == "HARD" ] &&
2200 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2201 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2205 local remote_dir=$DIR/$tdir/remote_dir
2207 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2208 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2209 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2210 $LFS mkdir -i $MDTIDX $remote_dir &
2215 fail mds$((MDTIDX + 1))
2217 wait $CLIENT_PID || error "remote creation failed"
2219 remote_dir_check_80 || error "remote dir check failed"
2220 rm -rf $DIR/$tdir || error "rmdir failed"
2224 run_test 80b "DNE: create remote dir, drop update rep from MDT0, fail MDT1"
2227 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2228 ([ $FAILURE_MODE == "HARD" ] &&
2229 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2230 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2234 local remote_dir=$DIR/$tdir/remote_dir
2236 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2237 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2238 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2239 $LFS mkdir -i $MDTIDX $remote_dir &
2245 fail mds$((MDTIDX + 1))
2247 wait $CLIENT_PID || error "remote creation failed"
2249 remote_dir_check_80 || error "remote dir check failed"
2250 rm -rf $DIR/$tdir || error "rmdir failed"
2254 run_test 80c "DNE: create remote dir, drop update rep from MDT1, fail MDT[0,1]"
2257 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2259 local remote_dir=$DIR/$tdir/remote_dir
2261 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2262 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2263 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2264 $LFS mkdir -i $MDTIDX $remote_dir &
2267 # sleep 3 seconds to make sure MDTs are failed after
2268 # lfs mkdir -i has finished on all of MDTs.
2273 fail mds${MDTIDX},mds$((MDTIDX + 1))
2275 wait $CLIENT_PID || error "remote creation failed"
2277 remote_dir_check_80 || error "remote dir check failed"
2278 rm -rf $DIR/$tdir || error "rmdir failed"
2282 run_test 80d "DNE: create remote dir, drop update rep from MDT1, fail 2 MDTs"
2285 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2286 ([ $FAILURE_MODE == "HARD" ] &&
2287 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2288 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2292 local remote_dir=$DIR/$tdir/remote_dir
2294 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2295 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2296 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2297 $LFS mkdir -i $MDTIDX $remote_dir &
2300 # sleep 3 seconds to make sure MDTs are failed after
2301 # lfs mkdir -i has finished on all of MDTs.
2307 wait $CLIENT_PID || error "remote creation failed"
2309 remote_dir_check_80 || error "remote dir check failed"
2310 rm -rf $DIR/$tdir || error "rmdir failed"
2314 run_test 80e "DNE: create remote dir, drop MDT1 rep, fail MDT0"
2317 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2318 ([ $FAILURE_MODE == "HARD" ] &&
2319 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2320 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2323 local remote_dir=$DIR/$tdir/remote_dir
2325 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2326 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2327 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2328 $LFS mkdir -i $MDTIDX $remote_dir &
2332 fail mds$((MDTIDX + 1))
2334 wait $CLIENT_PID || error "remote creation failed"
2336 remote_dir_check_80 || error "remote dir check failed"
2337 rm -rf $DIR/$tdir || error "rmdir failed"
2341 run_test 80f "DNE: create remote dir, drop MDT1 rep, fail MDT1"
2344 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2345 ([ $FAILURE_MODE == "HARD" ] &&
2346 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2347 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2351 local remote_dir=$DIR/$tdir/remote_dir
2353 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2354 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2355 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2356 $LFS mkdir -i $MDTIDX $remote_dir &
2359 # sleep 3 seconds to make sure MDTs are failed after
2360 # lfs mkdir -i has finished on all of MDTs.
2366 fail mds$((MDTIDX + 1))
2368 wait $CLIENT_PID || error "remote creation failed"
2370 remote_dir_check_80 || error "remote dir check failed"
2371 rm -rf $DIR/$tdir || error "rmdir failed"
2375 run_test 80g "DNE: create remote dir, drop MDT1 rep, fail MDT0, then MDT1"
2378 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2380 local remote_dir=$DIR/$tdir/remote_dir
2382 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2383 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2384 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2385 $LFS mkdir -i $MDTIDX $remote_dir &
2388 # sleep 3 seconds to make sure MDTs are failed after
2389 # lfs mkdir -i has finished on all of MDTs.
2394 fail mds${MDTIDX},mds$((MDTIDX + 1))
2396 wait $CLIENT_PID || error "remote dir creation failed"
2398 remote_dir_check_80 || error "remote dir check failed"
2399 rm -rf $DIR/$tdir || error "rmdir failed"
2403 run_test 80h "DNE: create remote dir, drop MDT1 rep, fail 2 MDTs"
2406 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2407 ([ $FAILURE_MODE == "HARD" ] &&
2408 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2409 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2413 local remote_dir=$DIR/$tdir/remote_dir
2415 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2416 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2418 touch $remote_dir || error "touch $remote_dir failed"
2419 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2420 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2425 fail mds$((MDTIDX + 1))
2427 wait $CLIENT_PID || error "rm remote dir failed"
2429 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2431 rm -rf $DIR/$tdir || error "rmdir failed"
2435 run_test 81a "DNE: unlink remote dir, drop MDT0 update rep, fail MDT1"
2438 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2439 ([ $FAILURE_MODE == "HARD" ] &&
2440 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2441 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2444 local remote_dir=$DIR/$tdir/remote_dir
2446 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2447 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2449 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2450 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2457 wait $CLIENT_PID || error "rm remote dir failed"
2459 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2461 rm -rf $DIR/$tdir || error "rmdir failed"
2465 run_test 81b "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0"
2468 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2469 ([ $FAILURE_MODE == "HARD" ] &&
2470 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2471 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2475 local remote_dir=$DIR/$tdir/remote_dir
2477 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2478 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2480 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2481 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2488 fail mds$((MDTIDX + 1))
2490 wait $CLIENT_PID || error "rm remote dir failed"
2492 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2494 rm -rf $DIR/$tdir || error "rmdir failed"
2498 run_test 81c "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0,MDT1"
2501 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2503 local remote_dir=$DIR/$tdir/remote_dir
2505 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2506 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2508 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2509 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2515 fail mds${MDTIDX},mds$((MDTIDX + 1))
2517 wait $CLIENT_PID || error "rm remote dir failed"
2519 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2521 rm -rf $DIR/$tdir || error "rmdir failed"
2525 run_test 81d "DNE: unlink remote dir, drop MDT0 update reply, fail 2 MDTs"
2528 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2529 ([ $FAILURE_MODE == "HARD" ] &&
2530 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2531 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2535 local remote_dir=$DIR/$tdir/remote_dir
2537 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2538 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2540 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2541 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2544 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
2549 wait $CLIENT_PID || error "rm remote dir failed"
2551 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2553 rm -rf $DIR/$tdir || error "rmdir failed"
2557 run_test 81e "DNE: unlink remote dir, drop MDT1 req reply, fail MDT0"
2560 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2561 ([ $FAILURE_MODE == "HARD" ] &&
2562 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2563 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2567 local remote_dir=$DIR/$tdir/remote_dir
2569 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2570 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2572 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2573 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2578 fail mds$((MDTIDX + 1))
2580 wait $CLIENT_PID || error "rm remote dir failed"
2582 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2584 rm -rf $DIR/$tdir || error "rmdir failed"
2588 run_test 81f "DNE: unlink remote dir, drop MDT1 req reply, fail MDT1"
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 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2601 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2603 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2604 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2611 fail mds$((MDTIDX + 1))
2613 wait $CLIENT_PID || error "rm remote dir failed"
2615 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2617 rm -rf $DIR/$tdir || error "rmdir failed"
2621 run_test 81g "DNE: unlink remote dir, drop req reply, fail M0, then M1"
2624 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2626 local remote_dir=$DIR/$tdir/remote_dir
2628 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2629 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2631 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2632 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2638 fail mds${MDTIDX},mds$((MDTIDX + 1))
2640 wait $CLIENT_PID || error "rm remote dir failed"
2642 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2644 rm -rf $DIR/$tdir || error "rmdir failed"
2648 run_test 81h "DNE: unlink remote dir, drop request reply, fail 2 MDTs"
2651 #define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x144
2652 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000144"
2653 createmany -o $DIR/$tfile- 1 &
2657 client_up || client_up || true # reconnect
2659 run_test 84a "stale open during export disconnect"
2661 test_85a() { #bug 16774
2662 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
2664 for i in $(seq 100); do
2665 echo "tag-$i" > $DIR/$tfile-$i
2666 grep -q "tag-$i" $DIR/$tfile-$i || error "f2-$i"
2669 lov_id=$(lctl dl | grep "clilov")
2670 addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}')
2671 count=$(lctl get_param -n \
2672 ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
2673 echo "before recovery: unused locks count = $count"
2677 count2=$(lctl get_param -n \
2678 ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
2679 echo "after recovery: unused locks count = $count2"
2681 if [ $count2 -ge $count ]; then
2682 error "unused locks are not canceled"
2685 run_test 85a "check the cancellation of unused locks during recovery(IBITS)"
2687 test_85b() { #bug 16774
2688 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
2690 do_facet mgs $LCTL pool_new $FSNAME.$TESTNAME ||
2691 error "unable to create pool $TESTNAME"
2692 do_facet mgs $LCTL pool_add $FSNAME.$TESTNAME $FSNAME-OST0000 ||
2693 error "unable to add pool $TESTNAME"
2695 $SETSTRIPE -c 1 -p $FSNAME.$TESTNAME $DIR
2697 for i in $(seq 100); do
2698 dd if=/dev/urandom of=$DIR/$tfile-$i bs=4096 \
2699 count=32 >/dev/null 2>&1
2702 cancel_lru_locks osc
2704 for i in $(seq 100); do
2705 dd if=$DIR/$tfile-$i of=/dev/null bs=4096 \
2706 count=32 >/dev/null 2>&1
2709 lov_id=$(lctl dl | grep "clilov")
2710 addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}')
2711 count=$(lctl get_param \
2712 -n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
2713 echo "before recovery: unused locks count = $count"
2714 [ $count != 0 ] || error "unused locks ($count) should be zero"
2718 count2=$(lctl get_param \
2719 -n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
2720 echo "after recovery: unused locks count = $count2"
2722 do_facet mgs $LCTL pool_remove $FSNAME.$TESTNAME $FSNAME-OST0000 ||
2723 error "unable to remove pool $TESTNAME"
2724 do_facet mgs $LCTL pool_destroy $FSNAME.$TESTNAME ||
2725 error "unable to destroy the pool $TESTNAME"
2727 if [ $count2 -ge $count ]; then
2728 error "unused locks are not canceled"
2731 run_test 85b "check the cancellation of unused locks during recovery(EXTENT)"
2734 local clients=${CLIENTS:-$HOSTNAME}
2736 zconf_umount_clients $clients $MOUNT
2737 do_facet $SINGLEMDS lctl set_param mdt.${FSNAME}-MDT*.exports.clear=0
2738 remount_facet $SINGLEMDS
2739 zconf_mount_clients $clients $MOUNT
2741 run_test 86 "umount server after clear nid_stats should not hit LBUG"
2744 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
2747 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
2748 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
2749 error "dd to $DIR/$tfile failed"
2750 cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
2751 cancel_lru_locks osc
2753 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
2754 cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
2755 if [ $cksum != $cksum2 ] ; then
2756 error "New checksum $cksum2 does not match original $cksum"
2759 run_test 87 "write replay"
2762 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
2765 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
2766 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
2767 error "dd to $DIR/$tfile failed"
2768 sleep 1 # Give it a chance to flush dirty data
2769 echo TESTTEST | dd of=$DIR/$tfile bs=1 count=8 seek=64
2770 cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
2771 cancel_lru_locks osc
2773 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
2774 cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
2775 if [ $cksum != $cksum2 ] ; then
2776 error "New checksum $cksum2 does not match original $cksum"
2779 run_test 87b "write replay with changed data (checksum resend)"
2781 test_88() { #bug 17485
2782 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2783 mkdir -p $TMP/$tdir || error "mkdir $TMP/$tdir failed"
2785 $SETSTRIPE -i 0 -c 1 $DIR/$tdir || error "$SETSTRIPE"
2788 replay_barrier $SINGLEMDS
2790 # exhaust precreations on ost1
2791 local OST=$(ostname_from_index 0)
2792 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $OST)
2793 local last_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
2794 local next_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
2795 echo "before test: last_id = $last_id, next_id = $next_id"
2797 echo "Creating to objid $last_id on ost $OST..."
2798 createmany -o $DIR/$tdir/f-%d $next_id $((last_id - next_id + 2)) ||
2799 error "createmany create files to last_id failed"
2801 #create some files to use some uncommitted objids
2802 last_id=$(($last_id + 1))
2803 createmany -o $DIR/$tdir/f-%d $last_id 8 ||
2804 error "createmany create files with uncommitted objids failed"
2806 last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
2807 next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
2808 echo "before recovery: last_id = $last_id2, next_id = $next_id2"
2810 # if test uses shutdown_facet && reboot_facet instead of facet_failover ()
2811 # it has to take care about the affected facets, bug20407
2812 local affected_mds1=$(affected_facets mds1)
2813 local affected_ost1=$(affected_facets ost1)
2815 shutdown_facet $SINGLEMDS
2818 reboot_facet $SINGLEMDS
2819 change_active $affected_mds1
2820 wait_for_facet $affected_mds1
2821 mount_facets $affected_mds1 || error "Restart of mds failed"
2824 change_active $affected_ost1
2825 wait_for_facet $affected_ost1
2826 mount_facets $affected_ost1 || error "Restart of ost1 failed"
2830 last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
2831 next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
2832 echo "after recovery: last_id = $last_id2, next_id = $next_id2"
2834 # create new files, which should use new objids, and ensure the orphan
2835 # cleanup phase for ost1 is completed at the same time
2836 for i in $(seq 8); do
2837 file_id=$(($last_id + 10 + $i))
2838 dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
2841 # if the objids were not recreated, then "ls" will fail with -ENOENT
2842 ls -l $DIR/$tdir/* || error "can't get the status of precreated files"
2845 # write into previously created files
2846 for i in $(seq 8); do
2847 file_id=$(($last_id + $i))
2848 dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
2849 cp -f $DIR/$tdir/f-$file_id $TMP/$tdir/
2852 # compare the content
2853 for i in $(seq 8); do
2854 file_id=$(($last_id + $i))
2855 cmp $TMP/$tdir/f-$file_id $DIR/$tdir/f-$file_id ||
2856 error "the content of file is modified!"
2861 run_test 88 "MDS should not assign same objid to different files "
2864 cancel_lru_locks osc
2865 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2866 rm -f $DIR/$tdir/$tfile
2868 wait_delete_completed
2869 BLOCKS1=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
2870 $SETSTRIPE -i 0 -c 1 $DIR/$tdir/$tfile
2871 dd if=/dev/zero bs=1M count=10 of=$DIR/$tdir/$tfile
2874 facet_failover $SINGLEMDS
2875 rm $DIR/$tdir/$tfile
2878 zconf_mount $(hostname) $MOUNT || error "mount fails"
2879 client_up || error "client_up failed"
2881 wait_delete_completed
2882 BLOCKS2=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
2883 [ $((BLOCKS2 - BLOCKS1)) -le 4 ] ||
2884 error $((BLOCKS2 - BLOCKS1)) blocks leaked
2887 run_test 89 "no disk space leak on late ost connection"
2893 change_active $facet
2894 wait_for_facet $facet
2895 mount_facet $facet || error "Restart of $facet failed"
2899 test_90() { # bug 19494
2900 local dir=$DIR/$tdir
2901 local ostfail=$(get_random_entry $(get_facets OST))
2903 if [[ $FAILURE_MODE = HARD ]]; then
2904 local affected=$(affected_facets $ostfail);
2905 if [[ "$affected" != $ostfail ]]; then
2906 skip not functional with FAILURE_MODE=$FAILURE_MODE, affected: $affected
2911 mkdir $dir || error "mkdir $dir failed"
2913 echo "Create the files"
2915 # file "f${index}" striped over 1 OST
2916 # file "all" striped over all OSTs
2918 $SETSTRIPE -c $OSTCOUNT $dir/all ||
2919 error "setstripe failed to create $dir/all"
2921 for (( i=0; i<$OSTCOUNT; i++ )); do
2923 $SETSTRIPE -i $i -c 1 $f || error "$SETSTRIPE failed to create $f"
2925 # confirm setstripe actually created the stripe on the requested OST
2926 local uuid=$(ostuuid_from_index $i)
2927 for file in f$i all; do
2928 if [[ $dir/$file != $($LFS find --obd $uuid --name $file $dir) ]]; then
2929 $GETSTRIPE $dir/$file
2930 error wrong stripe: $file, uuid: $uuid
2935 # Before failing an OST, get its obd name and index
2936 local varsvc=${ostfail}_svc
2937 local obd=$(do_facet $ostfail lctl get_param \
2938 -n obdfilter.${!varsvc}.uuid)
2939 local index=$(($(facet_number $ostfail) - 1))
2941 echo "Fail $ostfail $obd, display the list of affected files"
2942 shutdown_facet $ostfail || error "shutdown_facet $ostfail failed"
2944 trap "cleanup_90 $ostfail" EXIT INT
2945 echo "General Query: lfs find $dir"
2946 local list=$($LFS find $dir)
2948 for (( i=0; i<$OSTCOUNT; i++ )); do
2949 list_member "$list" $dir/f$i ||
2950 error_noexit "lfs find $dir: no file f$i"
2952 list_member "$list" $dir/all ||
2953 error_noexit "lfs find $dir: no file all"
2955 # focus on the missing OST,
2956 # we expect to see only two files affected: "f$(index)" and "all"
2958 echo "Querying files on shutdown $ostfail: lfs find --obd $obd"
2959 list=$($LFS find --obd $obd $dir)
2961 for file in all f$index; do
2962 list_member "$list" $dir/$file ||
2963 error_noexit "lfs find does not report the affected $obd for $file"
2966 [[ $(echo $list | wc -w) -eq 2 ]] ||
2967 error_noexit "lfs find reports the wrong list of affected files ${#list[@]}"
2969 echo "Check getstripe: $GETSTRIPE -r --obd $obd"
2970 list=$($GETSTRIPE -r --obd $obd $dir)
2972 for file in all f$index; do
2973 echo "$list" | grep $dir/$file ||
2974 error_noexit "lfs getsripe does not report the affected $obd for $file"
2979 run_test 90 "lfs find identifies the missing striped file segments"
2982 local server_version=$(lustre_version_code $SINGLEMDS)
2983 [[ $server_version -ge $(version_code 2.6.90) ]] ||
2984 [[ $server_version -ge $(version_code 2.5.4) &&
2985 $server_version -lt $(version_code 2.5.50) ]] ||
2986 { skip "Need MDS version 2.5.4+ or 2.6.90+"; return; }
2988 cancel_lru_locks osc
2990 $SETSTRIPE -i 0 -c 1 $DIR/$tfile ||
2991 error "$SETSTRIPE $DIR/$tfile failed"
2992 dd if=/dev/zero of=$DIR/$tfile bs=1024 count=1 ||
2993 error "dd to $DIR/$tfile failed"
2994 #define OBD_FAIL_TGT_REPLAY_RECONNECT 0x715
2995 # We need to emulate a state that OST is waiting for other clients
2996 # not completing the recovery. Final ping is queued, but reply will be
2997 # sent on the recovery completion. It is done by sleep before
2998 # processing final pings
2999 do_facet ost1 "$LCTL set_param fail_val=40"
3000 do_facet ost1 "$LCTL set_param fail_loc=0x715"
3003 run_test 93 "replay + reconnect"
3005 striped_dir_check_100() {
3006 local striped_dir=$DIR/$tdir/striped_dir
3007 local stripe_count=$($LFS getdirstripe -c $striped_dir)
3009 $LFS getdirstripe $striped_dir
3010 [ $stripe_count -eq 2 ] || error "$stripe_count != 2"
3012 createmany -o $striped_dir/f-%d 20 ||
3013 error "creation failed under striped dir"
3017 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3018 ([ $FAILURE_MODE == "HARD" ] &&
3019 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3020 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3023 local striped_dir=$DIR/$tdir/striped_dir
3026 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3028 #To make sure MDT1 and MDT0 are connected
3029 #otherwise it may create single stripe dir here
3030 $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
3032 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
3033 do_facet mds$((MDTIDX+1)) lctl set_param fail_loc=0x1701
3034 $LFS setdirstripe -i0 -c2 $striped_dir &
3037 fail mds$((MDTIDX + 1))
3039 wait $CLIENT_PID || error "striped dir creation failed"
3041 striped_dir_check_100 || error "striped dir check failed"
3042 rm -rf $DIR/$tdir || error "rmdir failed"
3044 run_test 100a "DNE: create striped dir, drop update rep from MDT1, fail MDT1"
3047 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3048 ([ $FAILURE_MODE == "HARD" ] &&
3049 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3050 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3053 local striped_dir=$DIR/$tdir/striped_dir
3056 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3058 #To make sure MDT1 and MDT0 are connected
3059 #otherwise it may create single stripe dir here
3060 $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
3062 # OBD_FAIL_MDS_REINT_NET_REP 0x119
3063 do_facet mds$MDTIDX lctl set_param fail_loc=0x119
3064 $LFS mkdir -i0 -c2 $striped_dir &
3069 wait $CLIENT_PID || error "striped dir creation failed"
3071 striped_dir_check_100 || error "striped dir check failed"
3072 rm -rf $DIR/$tdir || error "rmdir failed"
3074 run_test 100b "DNE: create striped dir, fail MDT0"
3076 test_101() { #LU-5648
3077 mkdir -p $DIR/$tdir/d1
3078 mkdir -p $DIR/$tdir/d2
3079 touch $DIR/$tdir/file0
3082 replay_barrier $SINGLEMDS
3083 for i in $(seq $num) ; do
3084 echo test$i > $DIR/$tdir/d1/file$i
3087 fail_abort $SINGLEMDS
3088 for i in $(seq $num) ; do
3089 touch $DIR/$tdir/d2/file$i
3090 test -s $DIR/$tdir/d2/file$i &&
3091 ls -al $DIR/$tdir/d2/file$i && error "file$i's size > 0"
3096 run_test 101 "Shouldn't reassign precreated objs to other files after recovery"
3105 [[ $(lctl get_param mdc.*.import |
3106 grep "connect_flags:.*multi_mod_rpc") ]] ||
3107 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3109 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3110 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3111 facet="mds$((0x$idx + 1))"
3113 # get current value of max_mod_rcps_in_flight
3114 num=$($LCTL get_param -n \
3115 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3116 # set default value if client does not support multi mod RPCs
3117 [ -z "$num" ] && num=1
3119 echo "creating $num files ..."
3121 for i in $(seq $num); do
3122 touch $DIR/$tdir/file-$i
3125 # drop request on MDT to force resend
3126 #define OBD_FAIL_MDS_REINT_MULTI_NET 0x159
3127 do_facet $facet "$LCTL set_param fail_loc=0x159"
3128 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3129 for i in $(seq $num); do
3130 chmod 0600 $DIR/$tdir/file-$i &
3134 do_facet $facet "$LCTL set_param fail_loc=0"
3135 for pid in $pids; do
3136 wait $pid || error "chmod failed"
3138 echo "done ($(date +%H:%M:%S))"
3140 # check chmod succeed
3141 for i in $(seq $num); do
3142 checkstat -vp 0600 $DIR/$tdir/file-$i
3147 run_test 102a "check resend (request lost) with multiple modify RPCs in flight"
3156 [[ $(lctl get_param mdc.*.import |
3157 grep "connect_flags:.*multi_mod_rpc") ]] ||
3158 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3160 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3161 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3162 facet="mds$((0x$idx + 1))"
3164 # get current value of max_mod_rcps_in_flight
3165 num=$($LCTL get_param -n \
3166 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3167 # set default value if client does not support multi mod RPCs
3168 [ -z "$num" ] && num=1
3170 echo "creating $num files ..."
3172 for i in $(seq $num); do
3173 touch $DIR/$tdir/file-$i
3176 # drop reply on MDT to force reconstruction
3177 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3178 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3179 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3180 for i in $(seq $num); do
3181 chmod 0600 $DIR/$tdir/file-$i &
3185 do_facet $facet "$LCTL set_param fail_loc=0"
3186 for pid in $pids; do
3187 wait $pid || error "chmod failed"
3189 echo "done ($(date +%H:%M:%S))"
3191 # check chmod succeed
3192 for i in $(seq $num); do
3193 checkstat -vp 0600 $DIR/$tdir/file-$i
3198 run_test 102b "check resend (reply lost) with multiple modify RPCs in flight"
3207 [[ $(lctl get_param mdc.*.import |
3208 grep "connect_flags:.*multi_mod_rpc") ]] ||
3209 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3211 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3212 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3213 facet="mds$((0x$idx + 1))"
3215 # get current value of max_mod_rcps_in_flight
3216 num=$($LCTL get_param -n \
3217 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3218 # set default value if client does not support multi mod RPCs
3219 [ -z "$num" ] && num=1
3221 echo "creating $num files ..."
3223 for i in $(seq $num); do
3224 touch $DIR/$tdir/file-$i
3227 replay_barrier $facet
3230 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3231 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3232 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3233 for i in $(seq $num); do
3234 chmod 0600 $DIR/$tdir/file-$i &
3238 do_facet $facet "$LCTL set_param fail_loc=0"
3243 for pid in $pids; do
3244 wait $pid || error "chmod failed"
3246 echo "done ($(date +%H:%M:%S))"
3248 # check chmod succeed
3249 for i in $(seq $num); do
3250 checkstat -vp 0600 $DIR/$tdir/file-$i
3255 run_test 102c "check replay w/o reconstruction with multiple mod RPCs in flight"
3264 [[ $(lctl get_param mdc.*.import |
3265 grep "connect_flags:.*multi_mod_rpc") ]] ||
3266 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3268 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3269 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3270 facet="mds$((0x$idx + 1))"
3272 # get current value of max_mod_rcps_in_flight
3273 num=$($LCTL get_param -n \
3274 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3275 # set default value if client does not support multi mod RPCs
3276 [ -z "$num" ] && num=1
3278 echo "creating $num files ..."
3280 for i in $(seq $num); do
3281 touch $DIR/$tdir/file-$i
3285 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3286 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3287 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3288 for i in $(seq $num); do
3289 chmod 0600 $DIR/$tdir/file-$i &
3294 # write MDT transactions to disk
3295 do_facet $facet "sync; sync; sync"
3297 do_facet $facet "$LCTL set_param fail_loc=0"
3302 for pid in $pids; do
3303 wait $pid || error "chmod failed"
3305 echo "done ($(date +%H:%M:%S))"
3307 # check chmod succeed
3308 for i in $(seq $num); do
3309 checkstat -vp 0600 $DIR/$tdir/file-$i
3314 run_test 102d "check replay & reconstruction with multiple mod RPCs in flight"
3316 check_striped_dir_110()
3318 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
3319 error "create striped dir failed"
3320 local stripe_count=$($LFS getdirstripe -c $DIR/$tdir/striped_dir)
3321 [ $stripe_count -eq $MDSCOUNT ] ||
3322 error "$stripe_count != 2 after recovery"
3326 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3327 ([ $FAILURE_MODE == "HARD" ] &&
3328 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3329 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3334 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3337 check_striped_dir_110 || error "check striped_dir failed"
3338 rm -rf $DIR/$tdir || error "rmdir failed"
3342 run_test 110a "DNE: create striped dir, fail MDT1"
3345 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3346 ([ $FAILURE_MODE == "HARD" ] &&
3347 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3348 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3353 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3356 zconf_mount $(hostname) $MOUNT
3357 client_up || return 1
3359 check_striped_dir_110 || error "check striped_dir failed"
3361 rm -rf $DIR/$tdir || error "rmdir failed"
3365 run_test 110b "DNE: create striped dir, fail MDT1 and client"
3368 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3369 ([ $FAILURE_MODE == "HARD" ] &&
3370 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3371 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3376 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3379 check_striped_dir_110 || error "check striped_dir failed"
3381 rm -rf $DIR/$tdir || error "rmdir failed"
3385 run_test 110c "DNE: create striped dir, fail MDT2"
3388 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3389 ([ $FAILURE_MODE == "HARD" ] &&
3390 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3391 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3396 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3399 zconf_mount $(hostname) $MOUNT
3400 client_up || return 1
3402 check_striped_dir_110 || error "check striped_dir failed"
3404 rm -rf $DIR/$tdir || error "rmdir failed"
3408 run_test 110d "DNE: create striped dir, fail MDT2 and client"
3411 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3412 ([ $FAILURE_MODE == "HARD" ] &&
3413 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3414 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3419 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3423 zconf_mount $(hostname) $MOUNT
3424 client_up || return 1
3426 check_striped_dir_110 || error "check striped_dir failed"
3428 rm -rf $DIR/$tdir || error "rmdir failed"
3432 run_test 110e "DNE: create striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
3435 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3436 ([ $FAILURE_MODE == "HARD" ] &&
3437 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3438 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3444 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3447 check_striped_dir_110 || error "check striped_dir failed"
3449 rm -rf $DIR/$tdir || error "rmdir failed"
3453 run_test 110f "DNE: create striped dir, fail MDT1/MDT2"
3456 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3457 ([ $FAILURE_MODE == "HARD" ] &&
3458 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3459 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3464 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3468 zconf_mount $(hostname) $MOUNT
3469 client_up || return 1
3471 check_striped_dir_110 || error "check striped_dir failed"
3473 rm -rf $DIR/$tdir || error "rmdir failed"
3477 run_test 110g "DNE: create striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
3480 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3481 ([ $FAILURE_MODE == "HARD" ] &&
3482 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3483 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3487 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3489 rm -rf $DIR/$tdir/striped_dir
3492 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3493 error "striped dir still exists"
3496 run_test 111a "DNE: unlink striped dir, 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" &&
3506 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3508 rm -rf $DIR/$tdir/striped_dir
3511 zconf_mount $(hostname) $MOUNT
3512 client_up || return 1
3514 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3515 error "striped dir still exists"
3518 run_test 111b "DNE: unlink striped dir, fail MDT2"
3521 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3522 ([ $FAILURE_MODE == "HARD" ] &&
3523 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3524 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3528 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3530 rm -rf $DIR/$tdir/striped_dir
3534 zconf_mount $(hostname) $MOUNT
3535 client_up || return 1
3536 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3537 error "striped dir still exists"
3540 run_test 111c "DNE: unlink striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
3543 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3544 ([ $FAILURE_MODE == "HARD" ] &&
3545 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3546 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3550 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3552 rm -rf $DIR/$tdir/striped_dir
3556 zconf_mount $(hostname) $MOUNT
3557 client_up || return 1
3558 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3559 error "striped dir still exists"
3563 run_test 111d "DNE: unlink striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
3566 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3567 ([ $FAILURE_MODE == "HARD" ] &&
3568 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3569 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3573 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3575 rm -rf $DIR/$tdir/striped_dir
3578 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3579 error "striped dir still exists"
3582 run_test 111e "DNE: unlink striped dir, uncommit on MDT2, fail MDT1/MDT2"
3585 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3586 ([ $FAILURE_MODE == "HARD" ] &&
3587 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3588 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3592 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3594 rm -rf $DIR/$tdir/striped_dir
3597 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3598 error "striped dir still exists"
3601 run_test 111f "DNE: unlink striped dir, uncommit on MDT1, fail MDT1/MDT2"
3604 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3605 ([ $FAILURE_MODE == "HARD" ] &&
3606 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3607 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3611 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3614 rm -rf $DIR/$tdir/striped_dir
3616 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3617 error "striped dir still exists"
3620 run_test 111g "DNE: unlink striped dir, fail MDT1/MDT2"
3622 test_112_rename_prepare() {
3623 mkdir -p $DIR/$tdir/src_dir
3624 $LFS mkdir -i 1 $DIR/$tdir/src_dir/src_child ||
3625 error "create remote source failed"
3627 touch $DIR/$tdir/src_dir/src_child/a
3629 $LFS mkdir -i 2 $DIR/$tdir/tgt_dir ||
3630 error "create remote target dir failed"
3632 $LFS mkdir -i 3 $DIR/$tdir/tgt_dir/tgt_child ||
3633 error "create remote target child failed"
3638 $CHECKSTAT -t dir $DIR/$tdir/src_dir/src_child &&
3639 error "src_child still exists after rename"
3641 $CHECKSTAT -t file $DIR/$tdir/tgt_dir/tgt_child/a ||
3642 error "missing file(a) after rename"
3646 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3647 ([ $FAILURE_MODE == "HARD" ] &&
3648 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3649 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3652 test_112_rename_prepare
3655 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3656 error "rename dir cross MDT failed!"
3660 rm -rf $DIR/$tdir || error "rmdir failed"
3662 run_test 112a "DNE: cross MDT rename, fail MDT1"
3665 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3666 ([ $FAILURE_MODE == "HARD" ] &&
3667 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3668 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3671 test_112_rename_prepare
3674 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3675 error "rename dir cross MDT failed!"
3680 rm -rf $DIR/$tdir || error "rmdir failed"
3682 run_test 112b "DNE: cross MDT rename, fail MDT2"
3685 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3686 ([ $FAILURE_MODE == "HARD" ] &&
3687 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3688 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3691 test_112_rename_prepare
3694 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3695 error "rename dir cross MDT failed!"
3700 rm -rf $DIR/$tdir || error "rmdir failed"
3702 run_test 112c "DNE: cross MDT rename, fail MDT3"
3705 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3706 ([ $FAILURE_MODE == "HARD" ] &&
3707 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3708 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3711 test_112_rename_prepare
3714 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3715 error "rename dir cross MDT failed!"
3720 rm -rf $DIR/$tdir || error "rmdir failed"
3722 run_test 112d "DNE: cross MDT rename, fail MDT4"
3725 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3726 ([ $FAILURE_MODE == "HARD" ] &&
3727 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3728 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3731 test_112_rename_prepare
3735 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3736 error "rename dir cross MDT failed!"
3741 rm -rf $DIR/$tdir || error "rmdir failed"
3743 run_test 112e "DNE: cross MDT rename, fail MDT1 and MDT2"
3746 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3747 ([ $FAILURE_MODE == "HARD" ] &&
3748 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3749 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3752 test_112_rename_prepare
3756 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3757 error "rename dir cross MDT failed!"
3762 rm -rf $DIR/$tdir || error "rmdir failed"
3764 run_test 112f "DNE: cross MDT rename, fail MDT1 and MDT3"
3767 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3768 ([ $FAILURE_MODE == "HARD" ] &&
3769 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3770 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3773 test_112_rename_prepare
3777 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3778 error "rename dir cross MDT failed!"
3783 rm -rf $DIR/$tdir || error "rmdir failed"
3785 run_test 112g "DNE: cross MDT rename, fail MDT1 and MDT4"
3788 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3789 ([ $FAILURE_MODE == "HARD" ] &&
3790 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3791 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3794 test_112_rename_prepare
3798 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3799 error "rename dir cross MDT failed!"
3804 rm -rf $DIR/$tdir || error "rmdir failed"
3806 run_test 112h "DNE: cross MDT rename, fail MDT2 and MDT3"
3809 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3810 ([ $FAILURE_MODE == "HARD" ] &&
3811 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3812 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3815 test_112_rename_prepare
3819 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3820 error "rename dir cross MDT failed!"
3825 rm -rf $DIR/$tdir || error "rmdir failed"
3827 run_test 112i "DNE: cross MDT rename, fail MDT2 and MDT4"
3830 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3831 ([ $FAILURE_MODE == "HARD" ] &&
3832 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3833 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3836 test_112_rename_prepare
3840 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3841 error "rename dir cross MDT failed!"
3846 rm -rf $DIR/$tdir || error "rmdir failed"
3848 run_test 112j "DNE: cross MDT rename, fail MDT3 and MDT4"
3851 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3852 ([ $FAILURE_MODE == "HARD" ] &&
3853 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3854 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3857 test_112_rename_prepare
3862 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3863 error "rename dir cross MDT failed!"
3868 rm -rf $DIR/$tdir || error "rmdir failed"
3870 run_test 112k "DNE: cross MDT rename, fail MDT1,MDT2,MDT3"
3873 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3874 ([ $FAILURE_MODE == "HARD" ] &&
3875 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3876 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3879 test_112_rename_prepare
3884 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3885 error "rename dir cross MDT failed!"
3890 rm -rf $DIR/$tdir || error "rmdir failed"
3892 run_test 112l "DNE: cross MDT rename, fail MDT1,MDT2,MDT4"
3895 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3896 ([ $FAILURE_MODE == "HARD" ] &&
3897 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3898 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3901 test_112_rename_prepare
3906 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3907 error "rename dir cross MDT failed!"
3912 rm -rf $DIR/$tdir || error "rmdir failed"
3914 run_test 112m "DNE: cross MDT rename, fail MDT1,MDT3,MDT4"
3917 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3918 ([ $FAILURE_MODE == "HARD" ] &&
3919 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3920 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3923 test_112_rename_prepare
3928 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3929 error "rename dir cross MDT failed!"
3934 rm -rf $DIR/$tdir || error "rmdir failed"
3936 run_test 112n "DNE: cross MDT rename, fail MDT2,MDT3,MDT4"
3939 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3940 ([ $FAILURE_MODE == "HARD" ] &&
3941 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3942 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3950 for ((j=0;j<$((MDSCOUNT));j++)); do
3951 fail_index=$((fail_index+1))
3952 index=$((fail_index % MDSCOUNT))
3953 replay_barrier mds$((index + 1))
3954 for ((i=0;i<5;i++)); do
3955 test_mkdir -i$index -c$MDSCOUNT $DIR/$tdir/test_$i ||
3956 error "create striped dir $DIR/$tdir/test_$i"
3959 fail mds$((index + 1))
3960 for ((i=0;i<5;i++)); do
3961 checkstat -t dir $DIR/$tdir/test_$i ||
3962 error "$DIR/$tdir/test_$i does not exist!"
3964 rm -rf $DIR/$tdir/test_* ||
3968 run_test 115 "failover for create/unlink striped directory"
3971 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3972 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.55) ] &&
3973 skip "Do not support large update log before 2.7.55" &&
3975 ([ $FAILURE_MODE == "HARD" ] &&
3976 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3977 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3984 # OBD_FAIL_SPLIT_UPDATE_REC 0x1702
3985 do_facet mds1 "lctl set_param fail_loc=0x80001702"
3986 $LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
3989 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
3990 error "stried_dir does not exists"
3992 run_test 116a "large update log master MDT recovery"
3995 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3996 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.55) ] &&
3997 skip "Do not support large update log before 2.7.55" &&
4000 ([ $FAILURE_MODE == "HARD" ] &&
4001 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4002 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4009 # OBD_FAIL_SPLIT_UPDATE_REC 0x1702
4010 do_facet mds2 "lctl set_param fail_loc=0x80001702"
4011 $LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
4014 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
4015 error "stried_dir does not exists"
4017 run_test 116b "large update log slave MDT recovery"
4021 check_and_cleanup_lustre