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}
2039 local mdscount=$MDSCOUNT
2041 # until LU-6844 is fixed, run on one MDT instead of disabling test
2044 zconf_mount_clients $clients $MOUNT
2047 [ "$SLOW" = "no" ] && duration=120
2048 # set duration to 900 because it takes some time to boot node
2049 [ "$FAILURE_MODE" = HARD ] && duration=900
2052 local start_ts=$(date +%s)
2053 local cmd="rundbench 1 -t $duration"
2055 if [ $mdscount -ge 2 ]; then
2056 test_mkdir -p -c$mdscount $DIR/$tdir
2057 $LFS setdirstripe -D -c$mdscount $DIR/$tdir
2059 do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
2060 PATH=\$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
2061 DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
2062 MOUNT=$MOUNT DIR=$DIR/$tdir/\\\$(hostname) LCTL=$LCTL $cmd" &
2065 #LU-1897 wait for all dbench copies to start
2066 while ! check_for_process $clients dbench; do
2067 elapsed=$(($(date +%s) - start_ts))
2068 if [ $elapsed -gt $duration ]; then
2069 killall_process $clients dbench
2070 error "dbench failed to start on $clients!"
2075 log "Started rundbench load pid=$pid ..."
2077 elapsed=$(($(date +%s) - start_ts))
2078 local num_failovers=0
2080 while [ $elapsed -lt $duration ]; do
2081 if ! check_for_process $clients dbench; then
2082 error_noexit "dbench stopped on some of $clients!"
2083 killall_process $clients dbench
2087 replay_barrier mds$fail_index
2088 sleep 1 # give clients a time to do operations
2089 # Increment the number of failovers
2090 num_failovers=$((num_failovers+1))
2091 log "$TESTNAME fail mds$fail_index $num_failovers times"
2093 elapsed=$(($(date +%s) - start_ts))
2094 if [ $fail_index -ge $mdscount ]; then
2097 fail_index=$((fail_index+1))
2101 wait $pid || error "rundbench load on $clients failed!"
2103 run_test 70b "dbench ${MDSCOUNT}mdts recovery; $CLIENTCOUNT clients"
2104 # end multi-client tests
2109 local monitor_pid=$3
2111 local start_ts=$(date +%s)
2112 local num_failovers=0
2115 elapsed=$(($(date +%s) - start_ts))
2116 while [ $elapsed -lt $duration ]; do
2117 fail_index=$((RANDOM%max_index+1))
2118 kill -0 $monitor_pid ||
2119 error "$monitor_pid stopped"
2121 replay_barrier mds$fail_index
2123 # Increment the number of failovers
2124 num_failovers=$((num_failovers+1))
2125 log "$TESTNAME fail mds$fail_index $num_failovers times"
2127 elapsed=$(($(date +%s) - start_ts))
2133 kill -9 $tar_70c_pid
2136 local clients=${CLIENTS:-$HOSTNAME}
2139 zconf_mount_clients $clients $MOUNT
2142 [ "$SLOW" = "no" ] && duration=180
2143 # set duration to 900 because it takes some time to boot node
2144 [ "$FAILURE_MODE" = HARD ] && duration=600
2147 local start_ts=$(date +%s)
2149 trap cleanup_70c EXIT
2152 test_mkdir -p -c$MDSCOUNT $DIR/$tdir || break
2153 if [ $MDSCOUNT -ge 2 ]; then
2154 $LFS setdirstripe -D -c$MDSCOUNT $DIR/$tdir ||
2155 error "set default dirstripe failed"
2157 cd $DIR/$tdir || break
2158 tar cf - /etc | tar xf - || error "tar failed"
2160 rm -rf $DIR/$tdir || break
2164 echo "Started tar $tar_70c_pid"
2166 random_fail_mdt $MDSCOUNT $duration $tar_70c_pid
2167 kill -0 $tar_70c_pid || error "tar $tar_70c_pid stopped"
2172 run_test 70c "tar ${MDSCOUNT}mdts recovery"
2176 kill -9 $mkdir_70d_pid
2180 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2181 local clients=${CLIENTS:-$HOSTNAME}
2184 zconf_mount_clients $clients $MOUNT
2187 [ "$SLOW" = "no" ] && duration=180
2188 # set duration to 900 because it takes some time to boot node
2189 [ "$FAILURE_MODE" = HARD ] && duration=900
2194 local start_ts=$(date +%s)
2196 trap cleanup_70d EXIT
2199 $LFS mkdir -i0 -c2 $DIR/$tdir/test || {
2203 $LFS mkdir -i1 -c2 $DIR/$tdir/test1 || {
2208 touch $DIR/$tdir/test/a || {
2212 mkdir $DIR/$tdir/test/b || {
2216 rm -rf $DIR/$tdir/test || {
2221 touch $DIR/$tdir/test1/a || {
2225 mkdir $DIR/$tdir/test1/b || {
2230 rm -rf $DIR/$tdir/test1 || {
2237 echo "Started $mkdir_70d_pid"
2239 random_fail_mdt $MDSCOUNT $duration $mkdir_70d_pid
2240 kill -0 $mkdir_70d_pid || error "mkdir/rmdir $mkdir_70d_pid stopped"
2245 run_test 70d "mkdir/rmdir striped dir ${MDSCOUNT}mdts recovery"
2249 kill -9 $rename_70e_pid
2253 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2254 local clients=${CLIENTS:-$HOSTNAME}
2257 echo ha > /proc/sys/lnet/debug
2258 zconf_mount_clients $clients $MOUNT
2261 [ "$SLOW" = "no" ] && duration=180
2262 # set duration to 900 because it takes some time to boot node
2263 [ "$FAILURE_MODE" = HARD ] && duration=900
2266 $LFS mkdir -i0 $DIR/$tdir/test_0
2267 $LFS mkdir -i0 $DIR/$tdir/test_1
2268 touch $DIR/$tdir/test_0/a
2269 touch $DIR/$tdir/test_1/b
2270 trap cleanup_70e EXIT
2273 mrename $DIR/$tdir/test_0/a $DIR/$tdir/test_1/b > \
2279 checkstat $DIR/$tdir/test_0/a && {
2280 echo "a still exists"
2284 checkstat $DIR/$tdir/test_1/b || {
2285 echo "b still exists"
2289 touch $DIR/$tdir/test_0/a || {
2290 echo "touch a fails"
2294 mrename $DIR/$tdir/test_1/b $DIR/$tdir/test_0/a > \
2302 echo "Started $rename_70e_pid"
2304 random_fail_mdt 2 $duration $rename_70e_pid
2305 kill -0 $rename_70e_pid || error "rename $rename_70e_pid stopped"
2310 run_test 70e "rename cross-MDT with random fails"
2314 kill -9 $mkdir_71a_pid
2317 random_double_fail_mdt() {
2320 local monitor_pid=$3
2322 local start_ts=$(date +%s)
2323 local num_failovers=0
2327 elapsed=$(($(date +%s) - start_ts))
2328 while [ $elapsed -lt $duration ]; do
2329 fail_index=$((RANDOM%max_index + 1))
2330 if [ $fail_index -eq $max_index ]; then
2333 second_index=$((fail_index + 1))
2335 kill -0 $monitor_pid ||
2336 error "$monitor_pid stopped"
2338 replay_barrier mds$fail_index
2339 replay_barrier mds$second_index
2341 # Increment the number of failovers
2342 num_failovers=$((num_failovers+1))
2343 log "fail mds$fail_index mds$second_index $num_failovers times"
2344 fail mds${fail_index},mds${second_index}
2345 elapsed=$(($(date +%s) - start_ts))
2350 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2351 local clients=${CLIENTS:-$HOSTNAME}
2354 zconf_mount_clients $clients $MOUNT
2357 [ "$SLOW" = "no" ] && duration=180
2358 # set duration to 900 because it takes some time to boot node
2359 [ "$FAILURE_MODE" = HARD ] && duration=900
2364 local start_ts=$(date +%s)
2366 trap cleanup_71a EXIT
2369 $LFS mkdir -i0 -c2 $DIR/$tdir/test
2370 rmdir $DIR/$tdir/test
2374 echo "Started $mkdir_71a_pid"
2376 random_double_fail_mdt 2 $duration $mkdir_71a_pid
2377 kill -0 $mkdir_71a_pid || error "mkdir/rmdir $mkdir_71a_pid stopped"
2382 run_test 71a "mkdir/rmdir striped dir with 2 mdts recovery"
2385 multiop_bg_pause $DIR/$tfile O_tSc ||
2386 error "multiop_bg_pause $DIR/$tfile failed"
2390 replay_barrier $SINGLEMDS
2391 #define OBD_FAIL_LDLM_ENQUEUE_NET 0x302
2392 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
2395 wait $pid || error "multiop pid failed"
2396 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
2399 run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay, close"
2402 multiop_bg_pause $DIR/$tfile O_tSc ||
2403 error "multiop_bg_pause $DIR/$tfile failed"
2407 replay_barrier $SINGLEMDS
2408 #define OBD_FAIL_LDLM_REPLY 0x30c
2409 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
2412 wait $pid || error "multiop pid failed"
2413 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
2416 run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
2420 local clients=${CLIENTS:-$HOSTNAME}
2422 zconf_umount_clients $clients $MOUNT
2424 facet_failover $SINGLEMDS
2425 zconf_mount_clients $clients $MOUNT
2427 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
2428 rm $DIR/$tfile || error "rm $DIR/$tfile failed"
2429 clients_up || error "client evicted: $?"
2432 run_test 74 "Ensure applications don't fail waiting for OST recovery"
2434 remote_dir_check_80() {
2439 diridx=$($GETSTRIPE -M $remote_dir) ||
2440 error "$GETSTRIPE -M $remote_dir failed"
2441 [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
2443 createmany -o $remote_dir/f-%d 20 || error "creation failed"
2444 fileidx=$($GETSTRIPE -M $remote_dir/f-1) ||
2445 error "$GETSTRIPE -M $remote_dir/f-1 failed"
2446 [ $fileidx -eq $MDTIDX ] || error "$fileidx != $MDTIDX"
2452 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2453 ([ $FAILURE_MODE == "HARD" ] &&
2454 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2455 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2459 local remote_dir=$DIR/$tdir/remote_dir
2461 mkdir -p $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2462 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2463 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2464 $LFS mkdir -i $MDTIDX $remote_dir &
2470 wait $CLIENT_PID || error "remote creation failed"
2472 remote_dir_check_80 || error "remote dir check failed"
2473 rm -rf $DIR/$tdir || error "rmdir failed"
2477 run_test 80a "DNE: create remote dir, drop update rep from MDT0, fail MDT0"
2480 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2481 ([ $FAILURE_MODE == "HARD" ] &&
2482 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2483 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2487 local remote_dir=$DIR/$tdir/remote_dir
2489 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2490 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2491 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2492 $LFS mkdir -i $MDTIDX $remote_dir &
2497 fail mds$((MDTIDX + 1))
2499 wait $CLIENT_PID || error "remote creation failed"
2501 remote_dir_check_80 || error "remote dir check failed"
2502 rm -rf $DIR/$tdir || error "rmdir failed"
2506 run_test 80b "DNE: create remote dir, drop update rep from MDT0, fail MDT1"
2509 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2510 ([ $FAILURE_MODE == "HARD" ] &&
2511 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2512 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2516 local remote_dir=$DIR/$tdir/remote_dir
2518 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2519 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2520 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2521 $LFS mkdir -i $MDTIDX $remote_dir &
2527 fail mds$((MDTIDX + 1))
2529 wait $CLIENT_PID || error "remote creation failed"
2531 remote_dir_check_80 || error "remote dir check failed"
2532 rm -rf $DIR/$tdir || error "rmdir failed"
2536 run_test 80c "DNE: create remote dir, drop update rep from MDT1, fail MDT[0,1]"
2539 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2541 local remote_dir=$DIR/$tdir/remote_dir
2543 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2544 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2545 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2546 $LFS mkdir -i $MDTIDX $remote_dir &
2549 # sleep 3 seconds to make sure MDTs are failed after
2550 # lfs mkdir -i has finished on all of MDTs.
2555 fail mds${MDTIDX},mds$((MDTIDX + 1))
2557 wait $CLIENT_PID || error "remote creation failed"
2559 remote_dir_check_80 || error "remote dir check failed"
2560 rm -rf $DIR/$tdir || error "rmdir failed"
2564 run_test 80d "DNE: create remote dir, drop update rep from MDT1, fail 2 MDTs"
2567 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2568 ([ $FAILURE_MODE == "HARD" ] &&
2569 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2570 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2574 local remote_dir=$DIR/$tdir/remote_dir
2576 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2577 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2578 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2579 $LFS mkdir -i $MDTIDX $remote_dir &
2582 # sleep 3 seconds to make sure MDTs are failed after
2583 # lfs mkdir -i has finished on all of MDTs.
2589 wait $CLIENT_PID || error "remote creation failed"
2591 remote_dir_check_80 || error "remote dir check failed"
2592 rm -rf $DIR/$tdir || error "rmdir failed"
2596 run_test 80e "DNE: create remote dir, drop MDT1 rep, fail MDT0"
2599 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2600 ([ $FAILURE_MODE == "HARD" ] &&
2601 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2602 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2605 local remote_dir=$DIR/$tdir/remote_dir
2607 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2608 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2609 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2610 $LFS mkdir -i $MDTIDX $remote_dir &
2614 fail mds$((MDTIDX + 1))
2616 wait $CLIENT_PID || error "remote creation failed"
2618 remote_dir_check_80 || error "remote dir check failed"
2619 rm -rf $DIR/$tdir || error "rmdir failed"
2623 run_test 80f "DNE: create remote dir, drop MDT1 rep, fail MDT1"
2626 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2627 ([ $FAILURE_MODE == "HARD" ] &&
2628 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2629 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2633 local remote_dir=$DIR/$tdir/remote_dir
2635 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2636 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2637 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2638 $LFS mkdir -i $MDTIDX $remote_dir &
2641 # sleep 3 seconds to make sure MDTs are failed after
2642 # lfs mkdir -i has finished on all of MDTs.
2648 fail mds$((MDTIDX + 1))
2650 wait $CLIENT_PID || error "remote creation failed"
2652 remote_dir_check_80 || error "remote dir check failed"
2653 rm -rf $DIR/$tdir || error "rmdir failed"
2657 run_test 80g "DNE: create remote dir, drop MDT1 rep, fail MDT0, then MDT1"
2660 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2662 local remote_dir=$DIR/$tdir/remote_dir
2664 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2665 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2666 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2667 $LFS mkdir -i $MDTIDX $remote_dir &
2670 # sleep 3 seconds to make sure MDTs are failed after
2671 # lfs mkdir -i has finished on all of MDTs.
2676 fail mds${MDTIDX},mds$((MDTIDX + 1))
2678 wait $CLIENT_PID || error "remote dir creation failed"
2680 remote_dir_check_80 || error "remote dir check failed"
2681 rm -rf $DIR/$tdir || error "rmdir failed"
2685 run_test 80h "DNE: create remote dir, drop MDT1 rep, fail 2 MDTs"
2688 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2689 ([ $FAILURE_MODE == "HARD" ] &&
2690 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2691 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2695 local remote_dir=$DIR/$tdir/remote_dir
2697 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2698 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2700 touch $remote_dir || error "touch $remote_dir failed"
2701 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2702 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2707 fail mds$((MDTIDX + 1))
2709 wait $CLIENT_PID || error "rm remote dir failed"
2711 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2713 rm -rf $DIR/$tdir || error "rmdir failed"
2717 run_test 81a "DNE: unlink remote dir, drop MDT0 update rep, fail MDT1"
2720 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2721 ([ $FAILURE_MODE == "HARD" ] &&
2722 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2723 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2726 local remote_dir=$DIR/$tdir/remote_dir
2728 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2729 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2731 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2732 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2739 wait $CLIENT_PID || error "rm remote dir failed"
2741 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2743 rm -rf $DIR/$tdir || error "rmdir failed"
2747 run_test 81b "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0"
2750 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2751 ([ $FAILURE_MODE == "HARD" ] &&
2752 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2753 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2757 local remote_dir=$DIR/$tdir/remote_dir
2759 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2760 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2762 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2763 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2770 fail mds$((MDTIDX + 1))
2772 wait $CLIENT_PID || error "rm remote dir failed"
2774 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2776 rm -rf $DIR/$tdir || error "rmdir failed"
2780 run_test 81c "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0,MDT1"
2783 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2785 local remote_dir=$DIR/$tdir/remote_dir
2787 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2788 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2790 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2791 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2797 fail mds${MDTIDX},mds$((MDTIDX + 1))
2799 wait $CLIENT_PID || error "rm remote dir failed"
2801 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2803 rm -rf $DIR/$tdir || error "rmdir failed"
2807 run_test 81d "DNE: unlink remote dir, drop MDT0 update reply, fail 2 MDTs"
2810 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2811 ([ $FAILURE_MODE == "HARD" ] &&
2812 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2813 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2817 local remote_dir=$DIR/$tdir/remote_dir
2819 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2820 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2822 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2823 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2826 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
2831 wait $CLIENT_PID || error "rm remote dir failed"
2833 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2835 rm -rf $DIR/$tdir || error "rmdir failed"
2839 run_test 81e "DNE: unlink remote dir, drop MDT1 req reply, fail MDT0"
2842 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2843 ([ $FAILURE_MODE == "HARD" ] &&
2844 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2845 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2849 local remote_dir=$DIR/$tdir/remote_dir
2851 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2852 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2854 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2855 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2860 fail mds$((MDTIDX + 1))
2862 wait $CLIENT_PID || error "rm remote dir failed"
2864 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2866 rm -rf $DIR/$tdir || error "rmdir failed"
2870 run_test 81f "DNE: unlink remote dir, drop MDT1 req reply, fail MDT1"
2873 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2874 ([ $FAILURE_MODE == "HARD" ] &&
2875 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2876 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2880 local remote_dir=$DIR/$tdir/remote_dir
2882 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2883 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2885 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2886 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2893 fail mds$((MDTIDX + 1))
2895 wait $CLIENT_PID || error "rm remote dir failed"
2897 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2899 rm -rf $DIR/$tdir || error "rmdir failed"
2903 run_test 81g "DNE: unlink remote dir, drop req reply, fail M0, then M1"
2906 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2908 local remote_dir=$DIR/$tdir/remote_dir
2910 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2911 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2913 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2914 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2920 fail mds${MDTIDX},mds$((MDTIDX + 1))
2922 wait $CLIENT_PID || error "rm remote dir failed"
2924 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2926 rm -rf $DIR/$tdir || error "rmdir failed"
2930 run_test 81h "DNE: unlink remote dir, drop request reply, fail 2 MDTs"
2933 #define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x144
2934 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000144"
2935 createmany -o $DIR/$tfile- 1 &
2939 client_up || client_up || true # reconnect
2941 run_test 84a "stale open during export disconnect"
2943 test_85a() { #bug 16774
2944 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
2946 for i in $(seq 100); do
2947 echo "tag-$i" > $DIR/$tfile-$i
2948 grep -q "tag-$i" $DIR/$tfile-$i || error "f2-$i"
2951 lov_id=$(lctl dl | grep "clilov")
2952 addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}')
2953 count=$(lctl get_param -n \
2954 ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
2955 echo "before recovery: unused locks count = $count"
2959 count2=$(lctl get_param -n \
2960 ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
2961 echo "after recovery: unused locks count = $count2"
2963 if [ $count2 -ge $count ]; then
2964 error "unused locks are not canceled"
2967 run_test 85a "check the cancellation of unused locks during recovery(IBITS)"
2969 test_85b() { #bug 16774
2970 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
2972 do_facet mgs $LCTL pool_new $FSNAME.$TESTNAME ||
2973 error "unable to create pool $TESTNAME"
2974 do_facet mgs $LCTL pool_add $FSNAME.$TESTNAME $FSNAME-OST0000 ||
2975 error "unable to add pool $TESTNAME"
2977 $SETSTRIPE -c 1 -p $FSNAME.$TESTNAME $DIR
2979 for i in $(seq 100); do
2980 dd if=/dev/urandom of=$DIR/$tfile-$i bs=4096 \
2981 count=32 >/dev/null 2>&1
2984 cancel_lru_locks osc
2986 for i in $(seq 100); do
2987 dd if=$DIR/$tfile-$i of=/dev/null bs=4096 \
2988 count=32 >/dev/null 2>&1
2991 lov_id=$(lctl dl | grep "clilov")
2992 addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}')
2993 count=$(lctl get_param \
2994 -n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
2995 echo "before recovery: unused locks count = $count"
2996 [ $count != 0 ] || error "unused locks ($count) should be zero"
3000 count2=$(lctl get_param \
3001 -n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
3002 echo "after recovery: unused locks count = $count2"
3004 do_facet mgs $LCTL pool_remove $FSNAME.$TESTNAME $FSNAME-OST0000 ||
3005 error "unable to remove pool $TESTNAME"
3006 do_facet mgs $LCTL pool_destroy $FSNAME.$TESTNAME ||
3007 error "unable to destroy the pool $TESTNAME"
3009 if [ $count2 -ge $count ]; then
3010 error "unused locks are not canceled"
3013 run_test 85b "check the cancellation of unused locks during recovery(EXTENT)"
3016 local clients=${CLIENTS:-$HOSTNAME}
3018 zconf_umount_clients $clients $MOUNT
3019 do_facet $SINGLEMDS lctl set_param mdt.${FSNAME}-MDT*.exports.clear=0
3020 remount_facet $SINGLEMDS
3021 zconf_mount_clients $clients $MOUNT
3023 run_test 86 "umount server after clear nid_stats should not hit LBUG"
3026 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
3029 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
3030 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
3031 error "dd to $DIR/$tfile failed"
3032 cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
3033 cancel_lru_locks osc
3035 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
3036 cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
3037 if [ $cksum != $cksum2 ] ; then
3038 error "New checksum $cksum2 does not match original $cksum"
3041 run_test 87 "write replay"
3044 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
3047 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
3048 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
3049 error "dd to $DIR/$tfile failed"
3050 sleep 1 # Give it a chance to flush dirty data
3051 echo TESTTEST | dd of=$DIR/$tfile bs=1 count=8 seek=64
3052 cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
3053 cancel_lru_locks osc
3055 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
3056 cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
3057 if [ $cksum != $cksum2 ] ; then
3058 error "New checksum $cksum2 does not match original $cksum"
3061 run_test 87b "write replay with changed data (checksum resend)"
3063 test_88() { #bug 17485
3064 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3065 mkdir -p $TMP/$tdir || error "mkdir $TMP/$tdir failed"
3067 $SETSTRIPE -i 0 -c 1 $DIR/$tdir || error "$SETSTRIPE"
3070 replay_barrier $SINGLEMDS
3072 # exhaust precreations on ost1
3073 local OST=$(ostname_from_index 0)
3074 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $OST)
3075 local last_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
3076 local next_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
3077 echo "before test: last_id = $last_id, next_id = $next_id"
3079 echo "Creating to objid $last_id on ost $OST..."
3080 createmany -o $DIR/$tdir/f-%d $next_id $((last_id - next_id + 2)) ||
3081 error "createmany create files to last_id failed"
3083 #create some files to use some uncommitted objids
3084 last_id=$(($last_id + 1))
3085 createmany -o $DIR/$tdir/f-%d $last_id 8 ||
3086 error "createmany create files with uncommitted objids failed"
3088 last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
3089 next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
3090 echo "before recovery: last_id = $last_id2, next_id = $next_id2"
3092 # if test uses shutdown_facet && reboot_facet instead of facet_failover ()
3093 # it has to take care about the affected facets, bug20407
3094 local affected_mds1=$(affected_facets mds1)
3095 local affected_ost1=$(affected_facets ost1)
3097 shutdown_facet $SINGLEMDS
3100 reboot_facet $SINGLEMDS
3101 change_active $affected_mds1
3102 wait_for_facet $affected_mds1
3103 mount_facets $affected_mds1 || error "Restart of mds failed"
3106 change_active $affected_ost1
3107 wait_for_facet $affected_ost1
3108 mount_facets $affected_ost1 || error "Restart of ost1 failed"
3112 last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
3113 next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
3114 echo "after recovery: last_id = $last_id2, next_id = $next_id2"
3116 # create new files, which should use new objids, and ensure the orphan
3117 # cleanup phase for ost1 is completed at the same time
3118 for i in $(seq 8); do
3119 file_id=$(($last_id + 10 + $i))
3120 dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
3123 # if the objids were not recreated, then "ls" will fail with -ENOENT
3124 ls -l $DIR/$tdir/* || error "can't get the status of precreated files"
3127 # write into previously created files
3128 for i in $(seq 8); do
3129 file_id=$(($last_id + $i))
3130 dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
3131 cp -f $DIR/$tdir/f-$file_id $TMP/$tdir/
3134 # compare the content
3135 for i in $(seq 8); do
3136 file_id=$(($last_id + $i))
3137 cmp $TMP/$tdir/f-$file_id $DIR/$tdir/f-$file_id ||
3138 error "the content of file is modified!"
3143 run_test 88 "MDS should not assign same objid to different files "
3146 cancel_lru_locks osc
3147 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3148 rm -f $DIR/$tdir/$tfile
3150 wait_delete_completed
3151 BLOCKS1=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
3152 $SETSTRIPE -i 0 -c 1 $DIR/$tdir/$tfile
3153 dd if=/dev/zero bs=1M count=10 of=$DIR/$tdir/$tfile
3156 facet_failover $SINGLEMDS
3157 rm $DIR/$tdir/$tfile
3160 zconf_mount $(hostname) $MOUNT || error "mount fails"
3161 client_up || error "client_up failed"
3163 wait_delete_completed
3164 BLOCKS2=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
3165 [ $((BLOCKS2 - BLOCKS1)) -le 4 ] ||
3166 error $((BLOCKS2 - BLOCKS1)) blocks leaked
3169 run_test 89 "no disk space leak on late ost connection"
3175 change_active $facet
3176 wait_for_facet $facet
3177 mount_facet $facet || error "Restart of $facet failed"
3181 test_90() { # bug 19494
3182 local dir=$DIR/$tdir
3183 local ostfail=$(get_random_entry $(get_facets OST))
3185 if [[ $FAILURE_MODE = HARD ]]; then
3186 local affected=$(affected_facets $ostfail);
3187 if [[ "$affected" != $ostfail ]]; then
3188 skip not functional with FAILURE_MODE=$FAILURE_MODE, affected: $affected
3193 mkdir $dir || error "mkdir $dir failed"
3195 echo "Create the files"
3197 # file "f${index}" striped over 1 OST
3198 # file "all" striped over all OSTs
3200 $SETSTRIPE -c $OSTCOUNT $dir/all ||
3201 error "setstripe failed to create $dir/all"
3203 for (( i=0; i<$OSTCOUNT; i++ )); do
3205 $SETSTRIPE -i $i -c 1 $f || error "$SETSTRIPE failed to create $f"
3207 # confirm setstripe actually created the stripe on the requested OST
3208 local uuid=$(ostuuid_from_index $i)
3209 for file in f$i all; do
3210 if [[ $dir/$file != $($LFS find --obd $uuid --name $file $dir) ]]; then
3211 $GETSTRIPE $dir/$file
3212 error wrong stripe: $file, uuid: $uuid
3217 # Before failing an OST, get its obd name and index
3218 local varsvc=${ostfail}_svc
3219 local obd=$(do_facet $ostfail lctl get_param \
3220 -n obdfilter.${!varsvc}.uuid)
3221 local index=$(($(facet_number $ostfail) - 1))
3223 echo "Fail $ostfail $obd, display the list of affected files"
3224 shutdown_facet $ostfail || error "shutdown_facet $ostfail failed"
3226 trap "cleanup_90 $ostfail" EXIT INT
3227 echo "General Query: lfs find $dir"
3228 local list=$($LFS find $dir)
3230 for (( i=0; i<$OSTCOUNT; i++ )); do
3231 list_member "$list" $dir/f$i ||
3232 error_noexit "lfs find $dir: no file f$i"
3234 list_member "$list" $dir/all ||
3235 error_noexit "lfs find $dir: no file all"
3237 # focus on the missing OST,
3238 # we expect to see only two files affected: "f$(index)" and "all"
3240 echo "Querying files on shutdown $ostfail: lfs find --obd $obd"
3241 list=$($LFS find --obd $obd $dir)
3243 for file in all f$index; do
3244 list_member "$list" $dir/$file ||
3245 error_noexit "lfs find does not report the affected $obd for $file"
3248 [[ $(echo $list | wc -w) -eq 2 ]] ||
3249 error_noexit "lfs find reports the wrong list of affected files ${#list[@]}"
3251 echo "Check getstripe: $GETSTRIPE -r --obd $obd"
3252 list=$($GETSTRIPE -r --obd $obd $dir)
3254 for file in all f$index; do
3255 echo "$list" | grep $dir/$file ||
3256 error_noexit "lfs getsripe does not report the affected $obd for $file"
3261 run_test 90 "lfs find identifies the missing striped file segments"
3264 local server_version=$(lustre_version_code $SINGLEMDS)
3265 [[ $server_version -ge $(version_code 2.6.90) ]] ||
3266 [[ $server_version -ge $(version_code 2.5.4) &&
3267 $server_version -lt $(version_code 2.5.50) ]] ||
3268 { skip "Need MDS version 2.5.4+ or 2.6.90+"; return; }
3270 cancel_lru_locks osc
3272 $SETSTRIPE -i 0 -c 1 $DIR/$tfile ||
3273 error "$SETSTRIPE $DIR/$tfile failed"
3274 dd if=/dev/zero of=$DIR/$tfile bs=1024 count=1 ||
3275 error "dd to $DIR/$tfile failed"
3276 #define OBD_FAIL_TGT_REPLAY_RECONNECT 0x715
3277 # We need to emulate a state that OST is waiting for other clients
3278 # not completing the recovery. Final ping is queued, but reply will be
3279 # sent on the recovery completion. It is done by sleep before
3280 # processing final pings
3281 do_facet ost1 "$LCTL set_param fail_val=40"
3282 do_facet ost1 "$LCTL set_param fail_loc=0x715"
3285 run_test 93 "replay + reconnect"
3287 striped_dir_check_100() {
3288 local striped_dir=$DIR/$tdir/striped_dir
3289 local stripe_count=$($LFS getdirstripe -c $striped_dir)
3291 $LFS getdirstripe $striped_dir
3292 [ $stripe_count -eq 2 ] || error "$stripe_count != 2"
3294 createmany -o $striped_dir/f-%d 20 ||
3295 error "creation failed under striped dir"
3299 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3300 ([ $FAILURE_MODE == "HARD" ] &&
3301 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3302 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3305 local striped_dir=$DIR/$tdir/striped_dir
3308 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3310 #To make sure MDT1 and MDT0 are connected
3311 #otherwise it may create single stripe dir here
3312 $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
3314 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
3315 do_facet mds$((MDTIDX+1)) lctl set_param fail_loc=0x1701
3316 $LFS setdirstripe -i0 -c2 $striped_dir &
3319 fail mds$((MDTIDX + 1))
3321 wait $CLIENT_PID || error "striped dir creation failed"
3323 striped_dir_check_100 || error "striped dir check failed"
3324 rm -rf $DIR/$tdir || error "rmdir failed"
3326 run_test 100a "DNE: create striped dir, drop update rep from MDT1, fail MDT1"
3329 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3330 ([ $FAILURE_MODE == "HARD" ] &&
3331 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3332 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3335 local striped_dir=$DIR/$tdir/striped_dir
3338 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3340 #To make sure MDT1 and MDT0 are connected
3341 #otherwise it may create single stripe dir here
3342 $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
3344 # OBD_FAIL_MDS_REINT_NET_REP 0x119
3345 do_facet mds$MDTIDX lctl set_param fail_loc=0x119
3346 $LFS mkdir -i0 -c2 $striped_dir &
3351 wait $CLIENT_PID || error "striped dir creation failed"
3353 striped_dir_check_100 || error "striped dir check failed"
3354 rm -rf $DIR/$tdir || error "rmdir failed"
3356 run_test 100b "DNE: create striped dir, fail MDT0"
3358 test_101() { #LU-5648
3359 mkdir -p $DIR/$tdir/d1
3360 mkdir -p $DIR/$tdir/d2
3361 touch $DIR/$tdir/file0
3364 replay_barrier $SINGLEMDS
3365 for i in $(seq $num) ; do
3366 echo test$i > $DIR/$tdir/d1/file$i
3369 fail_abort $SINGLEMDS
3370 for i in $(seq $num) ; do
3371 touch $DIR/$tdir/d2/file$i
3372 test -s $DIR/$tdir/d2/file$i &&
3373 ls -al $DIR/$tdir/d2/file$i && error "file$i's size > 0"
3378 run_test 101 "Shouldn't reassign precreated objs to other files after recovery"
3387 [[ $(lctl get_param mdc.*.import |
3388 grep "connect_flags:.*multi_mod_rpc") ]] ||
3389 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3391 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3392 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3393 facet="mds$((0x$idx + 1))"
3395 # get current value of max_mod_rcps_in_flight
3396 num=$($LCTL get_param -n \
3397 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3398 # set default value if client does not support multi mod RPCs
3399 [ -z "$num" ] && num=1
3401 echo "creating $num files ..."
3403 for i in $(seq $num); do
3404 touch $DIR/$tdir/file-$i
3407 # drop request on MDT to force resend
3408 #define OBD_FAIL_MDS_REINT_MULTI_NET 0x159
3409 do_facet $facet "$LCTL set_param fail_loc=0x159"
3410 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3411 for i in $(seq $num); do
3412 chmod 0600 $DIR/$tdir/file-$i &
3416 do_facet $facet "$LCTL set_param fail_loc=0"
3417 for pid in $pids; do
3418 wait $pid || error "chmod failed"
3420 echo "done ($(date +%H:%M:%S))"
3422 # check chmod succeed
3423 for i in $(seq $num); do
3424 checkstat -vp 0600 $DIR/$tdir/file-$i
3429 run_test 102a "check resend (request lost) with multiple modify RPCs in flight"
3438 [[ $(lctl get_param mdc.*.import |
3439 grep "connect_flags:.*multi_mod_rpc") ]] ||
3440 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3442 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3443 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3444 facet="mds$((0x$idx + 1))"
3446 # get current value of max_mod_rcps_in_flight
3447 num=$($LCTL get_param -n \
3448 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3449 # set default value if client does not support multi mod RPCs
3450 [ -z "$num" ] && num=1
3452 echo "creating $num files ..."
3454 for i in $(seq $num); do
3455 touch $DIR/$tdir/file-$i
3458 # drop reply on MDT to force reconstruction
3459 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3460 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3461 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3462 for i in $(seq $num); do
3463 chmod 0600 $DIR/$tdir/file-$i &
3467 do_facet $facet "$LCTL set_param fail_loc=0"
3468 for pid in $pids; do
3469 wait $pid || error "chmod failed"
3471 echo "done ($(date +%H:%M:%S))"
3473 # check chmod succeed
3474 for i in $(seq $num); do
3475 checkstat -vp 0600 $DIR/$tdir/file-$i
3480 run_test 102b "check resend (reply lost) with multiple modify RPCs in flight"
3489 [[ $(lctl get_param mdc.*.import |
3490 grep "connect_flags:.*multi_mod_rpc") ]] ||
3491 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3493 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3494 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3495 facet="mds$((0x$idx + 1))"
3497 # get current value of max_mod_rcps_in_flight
3498 num=$($LCTL get_param -n \
3499 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3500 # set default value if client does not support multi mod RPCs
3501 [ -z "$num" ] && num=1
3503 echo "creating $num files ..."
3505 for i in $(seq $num); do
3506 touch $DIR/$tdir/file-$i
3509 replay_barrier $facet
3512 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3513 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3514 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3515 for i in $(seq $num); do
3516 chmod 0600 $DIR/$tdir/file-$i &
3520 do_facet $facet "$LCTL set_param fail_loc=0"
3525 for pid in $pids; do
3526 wait $pid || error "chmod failed"
3528 echo "done ($(date +%H:%M:%S))"
3530 # check chmod succeed
3531 for i in $(seq $num); do
3532 checkstat -vp 0600 $DIR/$tdir/file-$i
3537 run_test 102c "check replay w/o reconstruction with multiple mod RPCs in flight"
3546 [[ $(lctl get_param mdc.*.import |
3547 grep "connect_flags:.*multi_mod_rpc") ]] ||
3548 { skip "Need MDC with 'multi_mod_rpcs' feature"; return 0; }
3550 $LFS mkdir -c1 $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3551 idx=$(printf "%04x" $($LFS getdirstripe -i $DIR/$tdir))
3552 facet="mds$((0x$idx + 1))"
3554 # get current value of max_mod_rcps_in_flight
3555 num=$($LCTL get_param -n \
3556 mdc.$FSNAME-MDT$idx-mdc-*.max_mod_rpcs_in_flight)
3557 # set default value if client does not support multi mod RPCs
3558 [ -z "$num" ] && num=1
3560 echo "creating $num files ..."
3562 for i in $(seq $num); do
3563 touch $DIR/$tdir/file-$i
3567 #define OBD_FAIL_MDS_REINT_MULTI_NET_REP 0x15a
3568 do_facet $facet "$LCTL set_param fail_loc=0x15a"
3569 echo "launch $num chmod in parallel ($(date +%H:%M:%S)) ..."
3570 for i in $(seq $num); do
3571 chmod 0600 $DIR/$tdir/file-$i &
3576 # write MDT transactions to disk
3577 do_facet $facet "sync; sync; sync"
3579 do_facet $facet "$LCTL set_param fail_loc=0"
3584 for pid in $pids; do
3585 wait $pid || error "chmod failed"
3587 echo "done ($(date +%H:%M:%S))"
3589 # check chmod succeed
3590 for i in $(seq $num); do
3591 checkstat -vp 0600 $DIR/$tdir/file-$i
3596 run_test 102d "check replay & reconstruction with multiple mod RPCs in flight"
3598 check_striped_dir_110()
3600 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
3601 error "create striped dir failed"
3602 local stripe_count=$($LFS getdirstripe -c $DIR/$tdir/striped_dir)
3603 [ $stripe_count -eq $MDSCOUNT ] ||
3604 error "$stripe_count != 2 after recovery"
3608 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3609 ([ $FAILURE_MODE == "HARD" ] &&
3610 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3611 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3616 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3619 check_striped_dir_110 || error "check striped_dir failed"
3620 rm -rf $DIR/$tdir || error "rmdir failed"
3624 run_test 110a "DNE: create striped dir, fail MDT1"
3627 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3628 ([ $FAILURE_MODE == "HARD" ] &&
3629 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3630 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3635 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3638 zconf_mount $(hostname) $MOUNT
3639 client_up || return 1
3641 check_striped_dir_110 || error "check striped_dir failed"
3643 rm -rf $DIR/$tdir || error "rmdir failed"
3647 run_test 110b "DNE: create striped dir, fail MDT1 and client"
3650 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3651 ([ $FAILURE_MODE == "HARD" ] &&
3652 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3653 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3658 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3661 check_striped_dir_110 || error "check striped_dir failed"
3663 rm -rf $DIR/$tdir || error "rmdir failed"
3667 run_test 110c "DNE: create striped dir, fail MDT2"
3670 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3671 ([ $FAILURE_MODE == "HARD" ] &&
3672 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3673 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3678 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3681 zconf_mount $(hostname) $MOUNT
3682 client_up || return 1
3684 check_striped_dir_110 || error "check striped_dir failed"
3686 rm -rf $DIR/$tdir || error "rmdir failed"
3690 run_test 110d "DNE: create striped dir, fail MDT2 and client"
3693 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3694 ([ $FAILURE_MODE == "HARD" ] &&
3695 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3696 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3701 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3705 zconf_mount $(hostname) $MOUNT
3706 client_up || return 1
3708 check_striped_dir_110 || error "check striped_dir failed"
3710 rm -rf $DIR/$tdir || error "rmdir failed"
3714 run_test 110e "DNE: create striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
3717 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3718 ([ $FAILURE_MODE == "HARD" ] &&
3719 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3720 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3726 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3729 check_striped_dir_110 || error "check striped_dir failed"
3731 rm -rf $DIR/$tdir || error "rmdir failed"
3735 run_test 110f "DNE: create striped dir, fail MDT1/MDT2"
3738 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3739 ([ $FAILURE_MODE == "HARD" ] &&
3740 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3741 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3746 $LFS mkdir -i1 -c$MDSCOUNT $DIR/$tdir/striped_dir
3750 zconf_mount $(hostname) $MOUNT
3751 client_up || return 1
3753 check_striped_dir_110 || error "check striped_dir failed"
3755 rm -rf $DIR/$tdir || error "rmdir failed"
3759 run_test 110g "DNE: create striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
3762 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3763 ([ $FAILURE_MODE == "HARD" ] &&
3764 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3765 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3769 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3771 rm -rf $DIR/$tdir/striped_dir
3774 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3775 error "striped dir still exists"
3778 run_test 111a "DNE: unlink striped dir, fail MDT1"
3781 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3782 ([ $FAILURE_MODE == "HARD" ] &&
3783 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3784 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3788 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3790 rm -rf $DIR/$tdir/striped_dir
3793 zconf_mount $(hostname) $MOUNT
3794 client_up || return 1
3796 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3797 error "striped dir still exists"
3800 run_test 111b "DNE: unlink striped dir, fail MDT2"
3803 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3804 ([ $FAILURE_MODE == "HARD" ] &&
3805 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3806 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3810 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3812 rm -rf $DIR/$tdir/striped_dir
3816 zconf_mount $(hostname) $MOUNT
3817 client_up || return 1
3818 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3819 error "striped dir still exists"
3822 run_test 111c "DNE: unlink striped dir, uncommit on MDT1, fail client/MDT1/MDT2"
3825 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3826 ([ $FAILURE_MODE == "HARD" ] &&
3827 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3828 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3832 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3834 rm -rf $DIR/$tdir/striped_dir
3838 zconf_mount $(hostname) $MOUNT
3839 client_up || return 1
3840 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3841 error "striped dir still exists"
3845 run_test 111d "DNE: unlink striped dir, uncommit on MDT2, fail client/MDT1/MDT2"
3848 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3849 ([ $FAILURE_MODE == "HARD" ] &&
3850 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3851 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3855 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3857 rm -rf $DIR/$tdir/striped_dir
3860 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3861 error "striped dir still exists"
3864 run_test 111e "DNE: unlink striped dir, uncommit on MDT2, fail MDT1/MDT2"
3867 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3868 ([ $FAILURE_MODE == "HARD" ] &&
3869 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3870 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3874 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3876 rm -rf $DIR/$tdir/striped_dir
3879 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3880 error "striped dir still exists"
3883 run_test 111f "DNE: unlink striped dir, uncommit on MDT1, fail MDT1/MDT2"
3886 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
3887 ([ $FAILURE_MODE == "HARD" ] &&
3888 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3889 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3893 $LFS mkdir -i1 -c2 $DIR/$tdir/striped_dir
3896 rm -rf $DIR/$tdir/striped_dir
3898 $CHECKSTAT -t dir $DIR/$tdir/striped_dir &&
3899 error "striped dir still exists"
3902 run_test 111g "DNE: unlink striped dir, fail MDT1/MDT2"
3904 test_112_rename_prepare() {
3905 mkdir -p $DIR/$tdir/src_dir
3906 $LFS mkdir -i 1 $DIR/$tdir/src_dir/src_child ||
3907 error "create remote source failed"
3909 touch $DIR/$tdir/src_dir/src_child/a
3911 $LFS mkdir -i 2 $DIR/$tdir/tgt_dir ||
3912 error "create remote target dir failed"
3914 $LFS mkdir -i 3 $DIR/$tdir/tgt_dir/tgt_child ||
3915 error "create remote target child failed"
3920 $CHECKSTAT -t dir $DIR/$tdir/src_dir/src_child &&
3921 error "src_child still exists after rename"
3923 $CHECKSTAT -t file $DIR/$tdir/tgt_dir/tgt_child/a ||
3924 error "missing file(a) after rename"
3928 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3929 ([ $FAILURE_MODE == "HARD" ] &&
3930 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3931 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3934 test_112_rename_prepare
3937 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3938 error "rename dir cross MDT failed!"
3942 rm -rf $DIR/$tdir || error "rmdir failed"
3944 run_test 112a "DNE: cross MDT rename, fail MDT1"
3947 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3948 ([ $FAILURE_MODE == "HARD" ] &&
3949 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3950 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3953 test_112_rename_prepare
3956 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3957 error "rename dir cross MDT failed!"
3962 rm -rf $DIR/$tdir || error "rmdir failed"
3964 run_test 112b "DNE: cross MDT rename, fail MDT2"
3967 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3968 ([ $FAILURE_MODE == "HARD" ] &&
3969 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3970 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3973 test_112_rename_prepare
3976 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3977 error "rename dir cross MDT failed!"
3982 rm -rf $DIR/$tdir || error "rmdir failed"
3984 run_test 112c "DNE: cross MDT rename, fail MDT3"
3987 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
3988 ([ $FAILURE_MODE == "HARD" ] &&
3989 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3990 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3993 test_112_rename_prepare
3996 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
3997 error "rename dir cross MDT failed!"
4002 rm -rf $DIR/$tdir || error "rmdir failed"
4004 run_test 112d "DNE: cross MDT rename, fail MDT4"
4007 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4008 ([ $FAILURE_MODE == "HARD" ] &&
4009 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4010 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4013 test_112_rename_prepare
4017 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4018 error "rename dir cross MDT failed!"
4023 rm -rf $DIR/$tdir || error "rmdir failed"
4025 run_test 112e "DNE: cross MDT rename, fail MDT1 and MDT2"
4028 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4029 ([ $FAILURE_MODE == "HARD" ] &&
4030 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4031 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4034 test_112_rename_prepare
4038 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4039 error "rename dir cross MDT failed!"
4044 rm -rf $DIR/$tdir || error "rmdir failed"
4046 run_test 112f "DNE: cross MDT rename, fail MDT1 and MDT3"
4049 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4050 ([ $FAILURE_MODE == "HARD" ] &&
4051 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4052 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4055 test_112_rename_prepare
4059 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4060 error "rename dir cross MDT failed!"
4065 rm -rf $DIR/$tdir || error "rmdir failed"
4067 run_test 112g "DNE: cross MDT rename, fail MDT1 and MDT4"
4070 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4071 ([ $FAILURE_MODE == "HARD" ] &&
4072 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4073 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4076 test_112_rename_prepare
4080 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4081 error "rename dir cross MDT failed!"
4086 rm -rf $DIR/$tdir || error "rmdir failed"
4088 run_test 112h "DNE: cross MDT rename, fail MDT2 and MDT3"
4091 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4092 ([ $FAILURE_MODE == "HARD" ] &&
4093 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4094 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4097 test_112_rename_prepare
4101 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4102 error "rename dir cross MDT failed!"
4107 rm -rf $DIR/$tdir || error "rmdir failed"
4109 run_test 112i "DNE: cross MDT rename, fail MDT2 and MDT4"
4112 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4113 ([ $FAILURE_MODE == "HARD" ] &&
4114 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4115 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4118 test_112_rename_prepare
4122 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4123 error "rename dir cross MDT failed!"
4128 rm -rf $DIR/$tdir || error "rmdir failed"
4130 run_test 112j "DNE: cross MDT rename, fail MDT3 and MDT4"
4133 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4134 ([ $FAILURE_MODE == "HARD" ] &&
4135 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4136 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4139 test_112_rename_prepare
4144 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4145 error "rename dir cross MDT failed!"
4150 rm -rf $DIR/$tdir || error "rmdir failed"
4152 run_test 112k "DNE: cross MDT rename, fail MDT1,MDT2,MDT3"
4155 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4156 ([ $FAILURE_MODE == "HARD" ] &&
4157 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4158 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4161 test_112_rename_prepare
4166 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4167 error "rename dir cross MDT failed!"
4172 rm -rf $DIR/$tdir || error "rmdir failed"
4174 run_test 112l "DNE: cross MDT rename, fail MDT1,MDT2,MDT4"
4177 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4178 ([ $FAILURE_MODE == "HARD" ] &&
4179 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4180 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4183 test_112_rename_prepare
4188 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4189 error "rename dir cross MDT failed!"
4194 rm -rf $DIR/$tdir || error "rmdir failed"
4196 run_test 112m "DNE: cross MDT rename, fail MDT1,MDT3,MDT4"
4199 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4200 ([ $FAILURE_MODE == "HARD" ] &&
4201 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4202 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4205 test_112_rename_prepare
4210 mrename $DIR/$tdir/src_dir/src_child $DIR/$tdir/tgt_dir/tgt_child ||
4211 error "rename dir cross MDT failed!"
4216 rm -rf $DIR/$tdir || error "rmdir failed"
4218 run_test 112n "DNE: cross MDT rename, fail MDT2,MDT3,MDT4"
4221 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4222 ([ $FAILURE_MODE == "HARD" ] &&
4223 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4224 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4232 for ((j=0;j<$((MDSCOUNT));j++)); do
4233 fail_index=$((fail_index+1))
4234 index=$((fail_index % MDSCOUNT))
4235 replay_barrier mds$((index + 1))
4236 for ((i=0;i<5;i++)); do
4237 test_mkdir -i$index -c$MDSCOUNT $DIR/$tdir/test_$i ||
4238 error "create striped dir $DIR/$tdir/test_$i"
4241 fail mds$((index + 1))
4242 for ((i=0;i<5;i++)); do
4243 checkstat -t dir $DIR/$tdir/test_$i ||
4244 error "$DIR/$tdir/test_$i does not exist!"
4246 rm -rf $DIR/$tdir/test_* ||
4250 run_test 115 "failover for create/unlink striped directory"
4253 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4254 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.55) ] &&
4255 skip "Do not support large update log before 2.7.55" &&
4257 ([ $FAILURE_MODE == "HARD" ] &&
4258 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4259 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4266 # OBD_FAIL_SPLIT_UPDATE_REC 0x1702
4267 do_facet mds1 "lctl set_param fail_loc=0x80001702"
4268 $LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
4271 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
4272 error "stried_dir does not exists"
4274 run_test 116a "large update log master MDT recovery"
4277 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4278 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.55) ] &&
4279 skip "Do not support large update log before 2.7.55" &&
4282 ([ $FAILURE_MODE == "HARD" ] &&
4283 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4284 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4291 # OBD_FAIL_SPLIT_UPDATE_REC 0x1702
4292 do_facet mds2 "lctl set_param fail_loc=0x80001702"
4293 $LFS setdirstripe -c$MDSCOUNT $DIR/$tdir/striped_dir
4296 $CHECKSTAT -t dir $DIR/$tdir/striped_dir ||
4297 error "stried_dir does not exists"
4299 run_test 116b "large update log slave MDT recovery"
4302 [ $MDSCOUNT -lt 4 ] && skip "needs >= 4 MDTs" && return 0
4303 ([ $FAILURE_MODE == "HARD" ] &&
4304 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
4305 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
4311 $LFS setdirstripe -i0 -c$MDSCOUNT $DIR/$tdir/remote_dir
4312 $LFS setdirstripe -i1 -c$MDSCOUNT $DIR/$tdir/remote_dir_1
4315 # Let's set rdonly on all MDTs, so client will send
4316 # replay requests on all MDTs and replay these requests
4317 # at the same time. This test will verify the recovery
4318 # will not be deadlock in this case, LU-7531.
4319 for ((index = 0; index < $((MDSCOUNT)); index++)); do
4320 replay_barrier mds$((index + 1))
4321 if [ -z $mds_indexs ]; then
4322 mds_indexs="${mds_indexs}mds$((index+1))"
4324 mds_indexs="${mds_indexs},mds$((index+1))"
4328 rm -rf $DIR/$tdir/remote_dir
4329 rm -rf $DIR/$tdir/remote_dir_1
4333 rm -rf $DIR/$tdir || error "rmdir failed"
4335 run_test 117 "DNE: cross MDT unlink, fail MDT1 and MDT2"
4338 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4339 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
4340 skip "Do not support large update log before 2.7.64" &&
4345 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir ||
4346 error "setdirstripe fails"
4347 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir1 ||
4348 error "setdirstripe fails 1"
4349 rm -rf $DIR/$tdir/striped_dir* || error "rmdir fails"
4351 # OBD_FAIL_INVALIDATE_UPDATE 0x1705
4352 do_facet mds1 "lctl set_param fail_loc=0x1705"
4353 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir
4354 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir1
4355 do_facet mds1 "lctl set_param fail_loc=0x0"
4358 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir
4359 $LFS setdirstripe -c2 $DIR/$tdir/striped_dir1
4364 run_test 118 "invalidate osp update will not cause update log corruption"
4367 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4368 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
4369 skip "Do not support large update log before 2.7.64" &&
4372 local hard_timeout=$(do_facet mds1 \
4373 "lctl get_param -n mdt.$FSNAME-MDT0000.recovery_time_hard")
4375 local clients=${CLIENTS:-$HOSTNAME}
4376 local time_min=$(recovery_time_min)
4379 mkdir $DIR/$tdir/tmp
4380 rmdir $DIR/$tdir/tmp
4383 mkdir $DIR/$tdir/dir_1
4384 for ((i = 0; i < 20; i++)); do
4385 $LFS setdirstripe -c2 $DIR/$tdir/stripe_dir-$i
4392 #define OBD_FAIL_TGT_REPLAY_DELAY 0x714
4393 do_facet mds1 $LCTL set_param fail_loc=0x80000714
4394 #sleep (timeout + 5), so mds will evict the client exports,
4395 #but DNE update recovery will keep going.
4396 do_facet mds1 $LCTL set_param fail_val=$((time_min + 5))
4398 mount_facet mds1 "-o recovery_time_hard=$time_min"
4400 wait_clients_import_state "$clients" mds1 FULL
4402 clients_up || clients_up || error "failover df: $?"
4404 #revert back the hard timeout
4405 do_facet mds1 $LCTL set_param \
4406 mdt.$FSNAME-MDT0000.recovery_time_hard=$hard_timeout
4408 for ((i = 0; i < 20; i++)); do
4409 stripe_count=$($LFS getdirstripe -c $DIR/$tdir/stripe_dir-$i)
4410 [ $stripe_count == 2 ] || {
4411 error "stripe_dir-$i creation replay fails"
4416 run_test 119 "timeout of normal replay does not cause DNE replay fails "
4419 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
4420 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.7.64) ] &&
4421 skip "Do not support large update log before 2.7.64" &&
4425 replay_barrier_nosync mds1
4426 for ((i = 0; i < 20; i++)); do
4427 mkdir $DIR/$tdir/dir-$i || {
4428 error "create dir-$i fails"
4431 $LFS setdirstripe -c2 $DIR/$tdir/stripe_dir-$i || {
4432 error "create stripe_dir-$i fails"
4439 for ((i = 0; i < 20; i++)); do
4440 [ ! -e "$DIR/$tdir/dir-$i" ] || {
4441 error "dir-$i still exists"
4444 [ ! -e "$DIR/$tdir/stripe_dir-$i" ] || {
4445 error "stripe_dir-$i still exists"
4450 run_test 120 "DNE fail abort should stop both normal and DNE replay"
4453 check_and_cleanup_lustre