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: b=17466/LU-472
25 ALWAYS_EXCEPT=" 61d $REPLAY_SINGLE_EXCEPT"
26 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
28 # 63 min 7 min AT AT AT AT"
29 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a 44b 65 66 67 68"
31 [ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
32 # bug number for skipped test: LU-1867 LU-3127
33 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 89 73b"
37 check_and_setup_lustre
42 rm -rf $DIR/[df][0-9]* $DIR/f.$TESTSUITE.*
44 # LU-482 Avert LVM and VM inability to flush caches in pre .33 kernels
45 if [ $LINUX_VERSION_CODE -lt $(version_code 2.6.33) ]; then
47 do_facet $SINGLEMDS sync
50 test_0a() { # was test_0
51 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
52 replay_barrier $SINGLEMDS
56 run_test 0a "empty replay"
59 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
61 # this test attempts to trigger a race in the precreation code,
62 # and must run before any other objects are created on the filesystem
64 createmany -o $DIR/$tfile 20 || error "createmany -o $DIR/$tfile failed"
65 unlinkmany $DIR/$tfile 20 || error "unlinkmany $DIR/$tfile failed"
67 run_test 0b "ensure object created after recover exists. (3284)"
70 replay_barrier $SINGLEMDS
73 facet_failover $SINGLEMDS
74 zconf_mount $(hostname) $MOUNT || error "mount fails"
75 client_up || error "post-failover df failed"
76 # file shouldn't exist if replay-barrier works as expected
77 rm $DIR/$tfile && error "File exists and it shouldn't"
80 run_test 0c "check replay-barrier"
83 replay_barrier $SINGLEMDS
85 facet_failover $SINGLEMDS
86 zconf_mount $(hostname) $MOUNT || error "mount fails"
87 client_up || error "post-failover df failed"
89 run_test 0d "expired recovery with no clients"
92 replay_barrier $SINGLEMDS
95 $CHECKSTAT -t file $DIR/$tfile ||
96 error "$CHECKSTAT $DIR/$tfile attribute check failed"
99 run_test 1 "simple create"
102 replay_barrier $SINGLEMDS
105 $CHECKSTAT -t file $DIR/$tfile ||
106 error "$CHECKSTAT $DIR/$tfile attribute check failed"
112 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
113 replay_barrier $SINGLEMDS
116 $CHECKSTAT -t file $DIR/$tfile ||
117 error "$CHECKSTAT $DIR/$tfile attribute check failed"
123 local file=$DIR/$tfile
124 replay_barrier $SINGLEMDS
126 openfile -f O_DIRECTORY $file
128 $CHECKSTAT -t file $file ||
129 error "$CHECKSTAT $file attribute check failed"
132 run_test 3a "replay failed open(O_DIRECTORY)"
135 replay_barrier $SINGLEMDS
136 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
137 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000114"
139 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
141 $CHECKSTAT -t file $DIR/$tfile &&
142 error "$CHECKSTAT $DIR/$tfile attribute check should fail"
145 run_test 3b "replay failed open -ENOMEM"
148 replay_barrier $SINGLEMDS
149 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
150 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000128"
152 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
155 $CHECKSTAT -t file $DIR/$tfile &&
156 error "$CHECKSTAT $DIR/$tfile attribute check should fail"
159 run_test 3c "replay failed open -ENOMEM"
161 test_4a() { # was test_4
162 replay_barrier $SINGLEMDS
163 for i in $(seq 10); do
164 echo "tag-$i" > $DIR/$tfile-$i
167 for i in $(seq 10); do
168 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
171 run_test 4a "|x| 10 open(O_CREAT)s"
174 for i in $(seq 10); do
175 echo "tag-$i" > $DIR/$tfile-$i
177 replay_barrier $SINGLEMDS
180 $CHECKSTAT -t file $DIR/$tfile-* &&
181 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
184 run_test 4b "|x| rm 10 files"
186 # The idea is to get past the first block of precreated files on both
187 # osts, and then replay.
189 replay_barrier $SINGLEMDS
190 for i in $(seq 220); do
191 echo "tag-$i" > $DIR/$tfile-$i
194 for i in $(seq 220); do
195 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
199 # waiting for commitment of removal
201 run_test 5 "|x| 220 open(O_CREAT)"
203 test_6a() { # was test_6
204 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
205 replay_barrier $SINGLEMDS
206 mcreate $DIR/$tdir/$tfile
208 $CHECKSTAT -t dir $DIR/$tdir ||
209 error "$CHECKSTAT $DIR/$tdir attribute check failed"
210 $CHECKSTAT -t file $DIR/$tdir/$tfile ||
211 error "$CHECKSTAT $DIR/$tdir/$tfile attribute check failed"
213 # waiting for log process thread
215 run_test 6a "mkdir + contained create"
218 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
219 replay_barrier $SINGLEMDS
222 $CHECKSTAT -t dir $DIR/$tdir &&
223 error "$CHECKSTAT $DIR/$tdir attribute check should fail" ||
226 run_test 6b "|X| rmdir"
229 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
230 replay_barrier $SINGLEMDS
231 mcreate $DIR/$tdir/$tfile
233 $CHECKSTAT -t dir $DIR/$tdir ||
234 error "$CHECKSTAT $DIR/$tdir attribute check failed"
235 $CHECKSTAT -t file $DIR/$tdir/$tfile ||
236 error "$CHECKSTAT $DIR/$tdir/$tfile attribute check failed"
239 run_test 7 "mkdir |X| contained create"
242 replay_barrier $SINGLEMDS
243 multiop_bg_pause $DIR/$tfile mo_c ||
244 error "multiop mknod $DIR/$tfile failed"
248 $CHECKSTAT -t file $DIR/$tfile ||
249 error "$CHECKSTAT $DIR/$tfile attribute check failed"
250 kill -USR1 $MULTIPID || error "multiop mknod $MULTIPID not running"
251 wait $MULTIPID || error "multiop mknod $MULTIPID failed"
254 run_test 8 "creat open |X| close"
257 replay_barrier $SINGLEMDS
259 local old_inum=$(ls -i $DIR/$tfile | awk '{print $1}')
261 local new_inum=$(ls -i $DIR/$tfile | awk '{print $1}')
263 echo " old_inum == $old_inum, new_inum == $new_inum"
264 if [ $old_inum -eq $new_inum ] ;
266 echo "old_inum and new_inum match"
268 echo " old_inum and new_inum do not match"
269 error "old index($old_inum) does not match new index($new_inum)"
273 run_test 9 "|X| create (same inum/gen)"
276 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
277 replay_barrier $SINGLEMDS
278 mv $DIR/$tfile $DIR/$tfile-2
281 $CHECKSTAT $DIR/$tfile &&
282 error "$CHECKSTAT $DIR/$tfile attribute check should fail"
283 $CHECKSTAT $DIR/$tfile-2 ||
284 error "$CHECKSTAT $DIR/$tfile-2 attribute check failed"
288 run_test 10 "create |X| rename unlink"
291 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
292 echo "old" > $DIR/$tfile
293 mv $DIR/$tfile $DIR/$tfile-2
294 replay_barrier $SINGLEMDS
295 echo "new" > $DIR/$tfile
297 grep old $DIR/$tfile-2
299 grep new $DIR/$tfile || error "grep $DIR/$tfile failed"
300 grep old $DIR/$tfile-2 || error "grep $DIR/$tfile-2 failed"
302 run_test 11 "create open write rename |X| create-old-name read"
305 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
306 multiop_bg_pause $DIR/$tfile o_tSc ||
307 error "multiop_bg_pause $DIR/$tfile failed"
310 replay_barrier $SINGLEMDS
311 kill -USR1 $pid || error "multiop $pid not running"
312 wait $pid || error "multiop $pid failed"
315 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
318 run_test 12 "open, unlink |X| close"
320 # 1777 - replay open after committed chmod that would make
321 # a regular open a failure
323 mcreate $DIR/$tfile || error "mcreate $DIR/$tfile failed"
324 multiop_bg_pause $DIR/$tfile O_wc ||
325 error "multiop_bg_pause $DIR/$tfile failed"
328 $CHECKSTAT -p 0 $DIR/$tfile ||
329 error "$CHECKSTAT $DIR/$tfile attribute check failed"
330 replay_barrier $SINGLEMDS
332 kill -USR1 $pid || error "multiop $pid not running"
333 wait $pid || error "multiop $pid failed"
335 $CHECKSTAT -s 1 -p 0 $DIR/$tfile ||
336 error "second $CHECKSTAT $DIR/$tfile attribute check failed"
337 rm $DIR/$tfile || error "rm $DIR/$tfile failed"
340 run_test 13 "open chmod 0 |x| write close"
343 multiop_bg_pause $DIR/$tfile O_tSc ||
344 error "multiop_bg_pause $DIR/$tfile failed"
347 replay_barrier $SINGLEMDS
348 kill -USR1 $pid || error "multiop $pid not running"
349 wait $pid || error "multiop $pid failed"
352 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
355 run_test 14 "open(O_CREAT), unlink |X| close"
358 multiop_bg_pause $DIR/$tfile O_tSc ||
359 error "multiop_bg_pause $DIR/$tfile failed"
362 replay_barrier $SINGLEMDS
363 touch $DIR/$tfile-1 || error "touch $DIR/$tfile-1 failed"
364 kill -USR1 $pid || error "multiop $pid not running"
365 wait $pid || error "multiop $pid failed"
368 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
369 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
372 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
375 replay_barrier $SINGLEMDS
378 mcreate $DIR/$tfile-2
380 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
381 [ -e $DIR/$tfile-2 ] || error "file $DIR/$tfile-2 does not exist"
382 munlink $DIR/$tfile-2 || error "munlink $DIR/$tfile-2 failed"
384 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
387 replay_barrier $SINGLEMDS
388 multiop_bg_pause $DIR/$tfile O_c ||
389 error "multiop_bg_pause $DIR/$tfile failed"
392 kill -USR1 $pid || error "multiop $pid not running"
393 wait $pid || error "multiop $pid failed"
394 $CHECKSTAT -t file $DIR/$tfile ||
395 error "$CHECKSTAT $DIR/$tfile attribute check failed"
398 run_test 17 "|X| open(O_CREAT), |replay| close"
401 replay_barrier $SINGLEMDS
402 multiop_bg_pause $DIR/$tfile O_tSc ||
403 error "multiop_bg_pause $DIR/$tfile failed"
406 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
407 echo "pid: $pid will close"
408 kill -USR1 $pid || error "multiop $pid not running"
409 wait $pid || error "multiop $pid failed"
412 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
413 [ -e $DIR/$tfile-2 ] || error "file $DIR/$tfile-2 does not exist"
414 # this touch frequently fails
415 touch $DIR/$tfile-3 || error "touch $DIR/$tfile-3 failed"
416 munlink $DIR/$tfile-2 || error "munlink $DIR/$tfile-2 failed"
417 munlink $DIR/$tfile-3 || error "munlink $DIR/$tfile-3 failed"
420 run_test 18 "open(O_CREAT), unlink, touch new, close, touch, unlink"
422 # bug 1855 (a simpler form of test_11 above)
424 replay_barrier $SINGLEMDS
426 echo "old" > $DIR/$tfile
427 mv $DIR/$tfile $DIR/$tfile-2
428 grep old $DIR/$tfile-2
430 grep old $DIR/$tfile-2 || error "grep $DIR/$tfile-2 failed"
432 run_test 19 "mcreate, open, write, rename "
434 test_20a() { # was test_20
435 replay_barrier $SINGLEMDS
436 multiop_bg_pause $DIR/$tfile O_tSc ||
437 error "multiop_bg_pause $DIR/$tfile failed"
442 kill -USR1 $pid || error "multiop $pid not running"
443 wait $pid || error "multiop $pid failed"
444 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
447 run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
449 test_20b() { # bug 10480
450 local wait_timeout=$((TIMEOUT * 4))
454 BEFOREUSED=$(df -P $DIR | tail -1 | awk '{ print $3 }')
455 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
456 while [ ! -e $DIR/$tfile ] ; do
457 usleep 60 # give dd a chance to start
460 $GETSTRIPE $DIR/$tfile || error "$GETSTRIPE $DIR/$tfile failed"
462 rm -f $DIR/$tfile || error "rm -f $DIR/$tfile failed"
464 client_up || client_up || true # reconnect
466 do_facet $SINGLEMDS "lctl set_param -n osd*.*MDT*.force_sync 1"
468 fail $SINGLEMDS # start orphan recovery
469 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
470 wait_delete_completed_mds $wait_timeout ||
471 error "delete did not complete"
473 AFTERUSED=$(df -P $DIR | tail -1 | awk '{ print $3 }')
474 log "before $BEFOREUSED, after $AFTERUSED"
475 (( $AFTERUSED > $BEFOREUSED + $(fs_log_size) )) &&
476 error "after $AFTERUSED > before $BEFOREUSED"
479 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
481 test_20c() { # bug 10480
482 multiop_bg_pause $DIR/$tfile Ow_c ||
483 error "multiop_bg_pause $DIR/$tfile failed"
489 client_up || client_up || true # reconnect
491 kill -USR1 $pid || error "multiop $pid not running"
492 wait $pid || error "multiop $pid failed"
493 [ -s $DIR/$tfile ] || error "File was truncated"
497 run_test 20c "check that client eviction does not affect file content"
500 replay_barrier $SINGLEMDS
501 multiop_bg_pause $DIR/$tfile O_tSc ||
502 error "multiop_bg_pause $DIR/$tfile failed"
505 touch $DIR/$tfile-1 || error "touch $DIR/$tfile-1 failed"
508 kill -USR1 $pid || error "multiop $pid not running"
509 wait $pid || error "multiop $pid failed"
510 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
511 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
514 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
517 multiop_bg_pause $DIR/$tfile O_tSc ||
518 error "multiop_bg_pause $DIR/$tfile failed"
521 replay_barrier $SINGLEMDS
525 kill -USR1 $pid || error "multiop $pid not running"
526 wait $pid || error "multiop $pid failed"
527 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
530 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
533 multiop_bg_pause $DIR/$tfile O_tSc ||
534 error "multiop_bg_pause $DIR/$tfile failed"
537 replay_barrier $SINGLEMDS
539 touch $DIR/$tfile-1 || error "touch $DIR/$tfile-1 failed"
542 kill -USR1 $pid || error "multiop $pid not running"
543 wait $pid || error "multiop $pid failed"
544 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
545 touch $DIR/$tfile-2 || error "touch $DIR/$tfile-2 failed"
548 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
551 multiop_bg_pause $DIR/$tfile O_tSc ||
552 error "multiop_bg_pause $DIR/$tfile failed"
555 replay_barrier $SINGLEMDS
558 kill -USR1 $pid || error "multiop $pid not running"
559 wait $pid || error "multiop $pid failed"
560 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
563 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
566 multiop_bg_pause $DIR/$tfile O_tSc ||
567 error "multiop_bg_pause $DIR/$tfile failed"
571 replay_barrier $SINGLEMDS
573 kill -USR1 $pid || error "multiop $pid not running"
574 wait $pid || error "multiop $pid failed"
575 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
578 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
581 replay_barrier $SINGLEMDS
582 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
583 error "multiop_bg_pause $DIR/$tfile-1 failed"
585 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
586 error "multiop_bg_pause $DIR/$tfile-2 failed"
590 kill -USR1 $pid2 || error "second multiop $pid2 not running"
591 wait $pid2 || error "second multiop $pid2 failed"
594 kill -USR1 $pid1 || error "multiop $pid1 not running"
595 wait $pid1 || error "multiop $pid1 failed"
596 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
597 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
600 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
603 replay_barrier $SINGLEMDS
604 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
605 error "multiop_bg_pause $DIR/$tfile-1 failed"
607 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
608 error "multiop_bg_pause $DIR/$tfile-2 failed"
614 kill -USR1 $pid1 || error "multiop $pid1 not running"
615 wait $pid1 || error "multiop $pid1 failed"
616 kill -USR1 $pid2 || error "second multiop $pid2 not running"
617 wait $pid2 || error "second multiop $pid2 failed"
618 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
619 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
622 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
625 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
626 error "multiop_bg_pause $DIR/$tfile-1 failed"
628 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
629 error "multiop_bg_pause $DIR/$tfile-2 failed"
631 replay_barrier $SINGLEMDS
634 kill -USR1 $pid2 || error "second multiop $pid2 not running"
635 wait $pid2 || error "second multiop $pid2 failed"
638 kill -USR1 $pid1 || error "multiop $pid1 not running"
639 wait $pid1 || error "multiop $pid1 failed"
640 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
641 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
644 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
647 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
648 error "multiop_bg_pause $DIR/$tfile-1 failed"
650 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
651 error "multiop_bg_pause $DIR/$tfile-2 failed"
653 replay_barrier $SINGLEMDS
658 kill -USR1 $pid1 || error "multiop $pid1 not running"
659 wait $pid1 || error "multiop $pid1 failed"
660 kill -USR1 $pid2 || error "second multiop $pid2 not running"
661 wait $pid2 || error "second multiop $pid2 failed"
662 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
663 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
666 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
669 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
670 error "multiop_bg_pause $DIR/$tfile-1 failed"
672 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
673 error "multiop_bg_pause $DIR/$tfile-2 failed"
678 replay_barrier $SINGLEMDS
680 kill -USR1 $pid1 || error "multiop $pid1 not running"
681 wait $pid1 || error "multiop $pid1 failed"
682 kill -USR1 $pid2 || error "second multiop $pid2 not running"
683 wait $pid2 || error "second multiop $pid2 failed"
684 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
685 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
688 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
691 multiop_bg_pause $DIR/$tfile-1 O_tSc ||
692 error "multiop_bg_pause $DIR/$tfile-1 failed"
694 multiop_bg_pause $DIR/$tfile-2 O_tSc ||
695 error "multiop_bg_pause $DIR/$tfile-2 failed"
699 replay_barrier $SINGLEMDS
702 kill -USR1 $pid1 || error "multiop $pid1 not running"
703 wait $pid1 || error "multiop $pid1 failed"
704 kill -USR1 $pid2 || error "second multiop $pid2 not running"
705 wait $pid2 || error "second multiop $pid2 failed"
706 [ -e $DIR/$tfile-1 ] && error "file $DIR/$tfile-1 should not exist"
707 [ -e $DIR/$tfile-2 ] && error "file $DIR/$tfile-2 should not exist"
710 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
712 # tests for bug 2104; completion without crashing is success. The close is
713 # stale, but we always return 0 for close, so the app never sees it.
715 multiop_bg_pause $DIR/$tfile O_c ||
716 error "multiop_bg_pause $DIR/$tfile failed"
718 multiop_bg_pause $DIR/$tfile O_c ||
719 error "second multiop_bg_pause $DIR/$tfile failed"
722 client_up || client_up || error "client_up failed"
723 kill -USR1 $pid1 || error "multiop $pid1 not running"
724 kill -USR1 $pid2 || error "second multiop $pid2 not running"
725 wait $pid1 || error "multiop $pid1 failed"
726 wait $pid2 || error "second multiop $pid2 failed"
729 run_test 32 "close() notices client eviction; close() after client eviction"
732 createmany -o $DIR/$tfile-%d 10 ||
733 error "createmany create $DIR/$tfile failed"
734 replay_barrier_nosync $SINGLEMDS
735 fail_abort $SINGLEMDS
736 # recreate shouldn't fail
737 createmany -o $DIR/$tfile--%d 10 ||
738 error "createmany recreate $DIR/$tfile failed"
742 run_test 33a "fid seq shouldn't be reused after abort recovery"
745 #define OBD_FAIL_SEQ_ALLOC 0x1311
746 do_facet $SINGLEMDS "lctl set_param fail_loc=0x1311"
748 createmany -o $DIR/$tfile-%d 10
749 replay_barrier_nosync $SINGLEMDS
750 fail_abort $SINGLEMDS
751 # recreate shouldn't fail
752 createmany -o $DIR/$tfile--%d 10 ||
753 error "createmany recreate $DIR/$tfile failed"
757 run_test 33b "test fid seq allocation"
760 multiop_bg_pause $DIR/$tfile O_c ||
761 error "multiop_bg_pause $DIR/$tfile failed"
765 replay_barrier $SINGLEMDS
766 fail_abort $SINGLEMDS
767 kill -USR1 $pid || error "multiop $pid not running"
768 wait $pid || error "multiop $pid failed"
769 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
773 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
775 # bug 2278 - generate one orphan on OST, then destroy it during recovery from llog
777 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
779 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
780 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
785 # give a chance to remove from MDS
786 fail_abort $SINGLEMDS
787 $CHECKSTAT -t file $DIR/$tfile &&
788 error "$CHECKSTAT $DIR/$tfile attribute check should fail" ||
791 run_test 35 "test recovery from llog for unlink op"
793 # b=2432 resent cancel after replay uses wrong cookie,
794 # so don't resend cancels
796 replay_barrier $SINGLEMDS
798 checkstat $DIR/$tfile
799 facet_failover $SINGLEMDS
801 if dmesg | grep "unknown lock cookie"; then
802 error "cancel after replay failed"
805 run_test 36 "don't resend cancel"
808 # directory orphans can't be unlinked from PENDING directory
810 rmdir $DIR/$tfile 2>/dev/null
811 multiop_bg_pause $DIR/$tfile dD_c ||
812 error "multiop_bg_pause $DIR/$tfile failed"
816 replay_barrier $SINGLEMDS
817 # clear the dmesg buffer so we only see errors from this recovery
818 do_facet $SINGLEMDS dmesg -c >/dev/null
819 fail_abort $SINGLEMDS
820 kill -USR1 $pid || error "multiop $pid not running"
821 do_facet $SINGLEMDS dmesg | grep "error .* unlinking .* from PENDING" &&
822 error "error unlinking files"
823 wait $pid || error "multiop $pid failed"
827 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
830 createmany -o $DIR/$tfile-%d 800 ||
831 error "createmany -o $DIR/$tfile failed"
832 unlinkmany $DIR/$tfile-%d 0 400 || error "unlinkmany $DIR/$tfile failed"
833 replay_barrier $SINGLEMDS
835 unlinkmany $DIR/$tfile-%d 400 400 ||
836 error "unlinkmany $DIR/$tfile 400 failed"
838 $CHECKSTAT -t file $DIR/$tfile-* &&
839 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
842 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
844 test_39() { # bug 4176
845 createmany -o $DIR/$tfile-%d 800 ||
846 error "createmany -o $DIR/$tfile failed"
847 replay_barrier $SINGLEMDS
848 unlinkmany $DIR/$tfile-%d 0 400
850 unlinkmany $DIR/$tfile-%d 400 400 ||
851 error "unlinkmany $DIR/$tfile 400 failed"
853 $CHECKSTAT -t file $DIR/$tfile-* &&
854 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
857 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
860 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
865 # always need connection to MDS to verify layout during IO. LU-2628.
866 lctl get_param mdc.*.connect_flags | grep -q layout_lock &&
867 skip "layout_lock needs MDS connection for IO" && return 0
869 $LCTL mark multiop $MOUNT/$tfile OS_c
870 multiop $MOUNT/$tfile OS_c &
872 writeme -s $MOUNT/${tfile}-2 &
875 facet_failover $SINGLEMDS
876 #define OBD_FAIL_MDS_CONNECT_NET 0x117
877 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000117"
879 stat1=$(count_ost_writes)
881 stat2=$(count_ost_writes)
882 echo "$stat1, $stat2"
883 if [ $stat1 -lt $stat2 ]; then
884 echo "writes continuing during recovery"
887 echo "writes not continuing during recovery, bug 2477"
890 echo "waiting for writeme $WRITE_PID"
894 echo "waiting for multiop $PID"
895 wait $PID || error "multiop $PID failed"
896 do_facet client munlink $MOUNT/$tfile ||
897 error "munlink $MOUNT/$tfile failed"
898 do_facet client munlink $MOUNT/${tfile}-2 ||
899 error "munlink $MOUNT/$tfile-2 failed"
902 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
905 # make sure that a read to one osc doesn't try to double-unlock its page just
906 # because another osc is invalid. trigger_group_io used to mistakenly return
907 # an error if any oscs were invalid even after having successfully put rpcs
908 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
909 # the page, guarnateeing that the unlock from the RPC completion would
910 # assert on trying to unlock the unlocked page.
912 [ $OSTCOUNT -lt 2 ] &&
913 skip_env "skipping test 41: we don't have a second OST to test with" &&
916 local f=$MOUNT/$tfile
917 # make sure the start of the file is ost1
918 $SETSTRIPE -S $((128 * 1024)) -i 0 $f
919 do_facet client dd if=/dev/zero of=$f bs=4k count=1 ||
920 error "dd on client failed"
922 # fail ost2 and read from ost1
923 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $ost2_svc)
924 local osc2dev=$(do_facet $SINGLEMDS "lctl get_param -n devices" |
925 grep $mdtosc | awk '{print $1}')
926 [ -z "$osc2dev" ] && echo "OST: $ost2_svc" &&
927 lctl get_param -n devices &&
928 error "OST 2 $osc2dev does not exist"
929 do_facet $SINGLEMDS $LCTL --device $osc2dev deactivate ||
930 error "deactive device on $SINGLEMDS failed"
931 do_facet client dd if=$f of=/dev/null bs=4k count=1 ||
932 error "second dd on client failed"
933 do_facet $SINGLEMDS $LCTL --device $osc2dev activate ||
934 error "active device on $SINGLEMDS failed"
937 run_test 41 "read from a valid osc while other oscs are invalid"
939 # test MDS recovery after ost failure
941 blocks=$(df -P $MOUNT | tail -n 1 | awk '{ print $2 }')
942 createmany -o $DIR/$tfile-%d 800 ||
943 error "createmany -o $DIR/$tfile failed"
945 unlinkmany $DIR/$tfile-%d 0 400
947 lctl set_param debug=-1
950 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
951 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
952 #[ $blocks_after -lt $blocks ] || return 1
953 echo "wait for MDS to timeout and recover"
954 sleep $((TIMEOUT * 2))
956 unlinkmany $DIR/$tfile-%d 400 400 ||
957 error "unlinkmany $DIR/$tfile 400 failed"
958 $CHECKSTAT -t file $DIR/$tfile-* &&
959 error "$CHECKSTAT $DIR/$tfile-* attribute check should fail" ||
962 run_test 42 "recovery after ost failure"
964 # timeout in MDS/OST recovery RPC will LBUG MDS
965 test_43() { # bug 2530
966 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
968 replay_barrier $SINGLEMDS
970 # OBD_FAIL_OST_CREATE_NET 0x204
971 do_facet ost1 "lctl set_param fail_loc=0x80000204"
977 run_test 43 "mds osc import failure during recovery; don't LBUG"
979 test_44a() { # was test_44
982 local mdcdev=$($LCTL dl |
983 awk "/${FSNAME}-MDT0000-mdc-/ {if (\$2 == \"UP\") {print \$1}}")
984 [ "$mdcdev" ] || error "${FSNAME}-MDT0000-mdc- not UP"
985 [ $(echo $mdcdev | wc -w) -eq 1 ] ||
986 { $LCTL dl; error "looking for mdcdev=$mdcdev"; }
988 # adaptive timeouts slow this way down
989 if at_is_enabled; then
990 at_max_saved=$(at_max_get mds)
994 for i in $(seq 1 10); do
995 echo "$i of 10 ($(date +%s))"
996 do_facet $SINGLEMDS \
997 "lctl get_param -n md[ts].*.mdt.timeouts | grep service"
998 #define OBD_FAIL_TGT_CONN_RACE 0x701
999 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000701"
1000 # lctl below may fail, it is valid case
1001 $LCTL --device $mdcdev recover
1004 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1005 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
1008 run_test 44a "race in target handle connect"
1011 local mdcdev=$($LCTL dl |
1012 awk "/${FSNAME}-MDT0000-mdc-/ {if (\$2 == \"UP\") {print \$1}}")
1013 [ "$mdcdev" ] || error "${FSNAME}-MDT0000-mdc not up"
1014 [ $(echo $mdcdev | wc -w) -eq 1 ] ||
1015 { echo mdcdev=$mdcdev; $LCTL dl;
1016 error "more than one ${FSNAME}-MDT0000-mdc"; }
1018 for i in $(seq 1 10); do
1019 echo "$i of 10 ($(date +%s))"
1020 do_facet $SINGLEMDS \
1021 "lctl get_param -n md[ts].*.mdt.timeouts | grep service"
1022 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
1023 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000704"
1024 # lctl below may fail, it is valid case
1025 $LCTL --device $mdcdev recover
1030 run_test 44b "race in target handle connect"
1033 replay_barrier $SINGLEMDS
1034 createmany -m $DIR/$tfile-%d 100 || error "failed to create directories"
1035 #define OBD_FAIL_TGT_RCVG_FLAG 0x712
1036 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000712"
1037 fail_abort $SINGLEMDS
1038 unlinkmany $DIR/$tfile-%d 100 && error "unliked after fail abort"
1040 unlinkmany $DIR/$tfile-%d 100 && error "unliked after fail"
1043 run_test 44c "race in target handle connect"
1045 # Handle failed close
1047 local mdcdev=$($LCTL get_param -n devices |
1048 awk "/ ${FSNAME}-MDT0000-mdc-/ {print \$1}")
1049 [ "$mdcdev" ] || error "${FSNAME}-MDT0000-mdc not up"
1050 [ $(echo $mdcdev | wc -w) -eq 1 ] ||
1051 { echo mdcdev=$mdcdev; $LCTL dl;
1052 error "more than one ${FSNAME}-MDT0000-mdc"; }
1054 $LCTL --device $mdcdev recover ||
1055 error "$LCTL --device $mdcdev recover failed"
1057 multiop_bg_pause $DIR/$tfile O_c ||
1058 error "multiop_bg_pause $DIR/$tfile failed"
1061 # This will cause the CLOSE to fail before even
1062 # allocating a reply buffer
1063 $LCTL --device $mdcdev deactivate ||
1064 error "$LCTL --device $mdcdev deactivate failed"
1067 kill -USR1 $pid || error "multiop $pid not running"
1068 wait $pid || error "multiop $pid failed"
1070 $LCTL --device $mdcdev activate ||
1071 error "$LCTL --device $mdcdev activate failed"
1074 $CHECKSTAT -t file $DIR/$tfile ||
1075 error "$CHECKSTAT $DIR/$tfile attribute check failed"
1078 run_test 45 "Handle failed close"
1082 drop_reply "touch $DIR/$tfile"
1084 # ironically, the previous test, 45, will cause a real forced close,
1085 # so just look for one for this test
1086 dmesg | grep -i "force closing client file handle for $tfile" &&
1087 error "found force closing in dmesg"
1090 run_test 46 "Don't leak file handle after open resend (3325)"
1092 test_47() { # bug 2824
1093 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1095 # create some files to make sure precreate has been done on all
1096 # OSTs. (just in case this test is run independently)
1097 createmany -o $DIR/$tfile 20 ||
1098 error "createmany create $DIR/$tfile failed"
1100 # OBD_FAIL_OST_CREATE_NET 0x204
1102 do_facet ost1 "lctl set_param fail_loc=0x80000204"
1103 client_up || error "client_up failed"
1105 # let the MDS discover the OST failure, attempt to recover, fail
1106 # and recover again.
1107 sleep $((3 * TIMEOUT))
1109 # Without 2824, this createmany would hang
1110 createmany -o $DIR/$tfile 20 ||
1111 error "createmany recraete $DIR/$tfile failed"
1112 unlinkmany $DIR/$tfile 20 || error "unlinkmany $DIR/$tfile failed"
1116 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
1119 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1120 [ "$OSTCOUNT" -lt "2" ] &&
1121 skip_env "$OSTCOUNT < 2 OSTs -- skipping" && return
1123 replay_barrier $SINGLEMDS
1124 createmany -o $DIR/$tfile 20 ||
1125 error "createmany -o $DIR/$tfile failed"
1126 # OBD_FAIL_OST_EROFS 0x216
1127 facet_failover $SINGLEMDS
1128 do_facet ost1 "lctl set_param fail_loc=0x80000216"
1129 client_up || error "client_up failed"
1131 createmany -o $DIR/$tfile 20 20 ||
1132 error "createmany recraete $DIR/$tfile failed"
1133 unlinkmany $DIR/$tfile 40 || error "unlinkmany $DIR/$tfile failed"
1136 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
1139 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $ost1_svc)
1140 local oscdev=$(do_facet $SINGLEMDS "lctl get_param -n devices" |
1141 grep $mdtosc | awk '{print $1}')
1142 [ "$oscdev" ] || error "could not find OSC device on MDS"
1143 do_facet $SINGLEMDS $LCTL --device $oscdev recover ||
1144 error "OSC device $oscdev recovery failed"
1145 do_facet $SINGLEMDS $LCTL --device $oscdev recover ||
1146 error "second OSC device $oscdev recovery failed"
1147 # give the mds_lov_sync threads a chance to run
1150 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1152 # b3764 timed out lock replay
1154 [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ] &&
1155 skip "MDS prior to 2.6.90 handle LDLM_REPLY_NET incorrectly" &&
1158 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1159 cancel_lru_locks mdc
1161 multiop_bg_pause $DIR/$tfile s_s || error "multiop $DIR/$tfile failed"
1164 #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
1165 lctl set_param -n ldlm.cancel_unused_locks_before_replay "0"
1166 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000157"
1168 fail $SINGLEMDS || error "fail $SINGLEMDS failed"
1170 wait $mpid || error "multiop_bg_pause pid failed"
1172 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1173 lctl set_param fail_loc=0x0
1174 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
1177 run_test 52 "time out lock replay (3764)"
1179 # bug 3462 - simultaneous MDC requests
1181 cancel_lru_locks mdc # cleanup locks from former test cases
1182 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1183 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1184 multiop $DIR/${tdir}-1/f O_c &
1186 # give multiop a change to open
1189 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1190 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1191 kill -USR1 $close_pid
1192 cancel_lru_locks mdc # force the close
1193 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1195 mcreate $DIR/${tdir}-2/f || error "mcreate $DIR/${tdir}-2/f failed"
1197 # close should still be here
1198 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1200 replay_barrier_nodf $SINGLEMDS
1202 wait $close_pid || error "close_pid $close_pid failed"
1204 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1205 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1206 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1207 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1208 rm -rf $DIR/${tdir}-*
1210 run_test 53a "|X| close request while two MDC requests in flight"
1213 cancel_lru_locks mdc # cleanup locks from former test cases
1215 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1216 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1217 multiop_bg_pause $DIR/${tdir}-1/f O_c ||
1218 error "multiop_bg_pause $DIR/${tdir}-1/f failed"
1221 #define OBD_FAIL_MDS_REINT_NET 0x107
1222 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1223 mcreate $DIR/${tdir}-2/f &
1227 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1228 kill -USR1 $close_pid
1229 cancel_lru_locks mdc # force the close
1230 wait $close_pid || error "close_pid $close_pid failed"
1231 # open should still be here
1232 [ -d /proc/$open_pid ] || error "open_pid doesn't exist"
1234 replay_barrier_nodf $SINGLEMDS
1236 wait $open_pid || error "open_pid failed"
1238 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1239 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1240 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1241 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1242 rm -rf $DIR/${tdir}-*
1244 run_test 53b "|X| open request while two MDC requests in flight"
1247 cancel_lru_locks mdc # cleanup locks from former test cases
1249 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1250 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1251 multiop $DIR/${tdir}-1/f O_c &
1254 #define OBD_FAIL_MDS_REINT_NET 0x107
1255 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1256 mcreate $DIR/${tdir}-2/f &
1260 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1261 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1262 kill -USR1 $close_pid
1263 cancel_lru_locks mdc # force the close
1265 #bz20647: make sure all pids exist before failover
1266 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1267 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1268 replay_barrier_nodf $SINGLEMDS
1269 fail_nodf $SINGLEMDS
1270 wait $open_pid || error "open_pid failed"
1272 # close should be gone
1273 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1274 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1276 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1277 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1278 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1279 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1280 rm -rf $DIR/${tdir}-*
1282 run_test 53c "|X| open request and close request while two MDC requests in flight"
1285 cancel_lru_locks mdc # cleanup locks from former test cases
1287 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1288 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1289 multiop $DIR/${tdir}-1/f O_c &
1291 # give multiop a chance to open
1294 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1295 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1296 kill -USR1 $close_pid
1297 cancel_lru_locks mdc # force the close
1298 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1299 mcreate $DIR/${tdir}-2/f || error "mcreate $DIR/${tdir}-2/f failed"
1301 # close should still be here
1302 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1304 wait $close_pid || error "close_pid failed"
1306 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1307 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1308 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1309 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1310 rm -rf $DIR/${tdir}-*
1312 run_test 53d "close reply while two MDC requests in flight"
1315 cancel_lru_locks mdc # cleanup locks from former test cases
1317 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1318 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1319 multiop $DIR/${tdir}-1/f O_c &
1322 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1323 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1324 mcreate $DIR/${tdir}-2/f &
1328 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1329 kill -USR1 $close_pid
1330 cancel_lru_locks mdc # force the close
1331 wait $close_pid || error "close_pid failed"
1332 # open should still be here
1333 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1335 replay_barrier_nodf $SINGLEMDS
1337 wait $open_pid || error "open_pid failed"
1339 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1340 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1341 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1342 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1343 rm -rf $DIR/${tdir}-*
1345 run_test 53e "|X| open reply while two MDC requests in flight"
1348 cancel_lru_locks mdc # cleanup locks from former test cases
1350 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1351 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1352 multiop $DIR/${tdir}-1/f O_c &
1355 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1356 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1357 mcreate $DIR/${tdir}-2/f &
1361 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1362 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1363 kill -USR1 $close_pid
1364 cancel_lru_locks mdc # force the close
1366 #bz20647: make sure all pids are exists before failover
1367 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1368 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1369 replay_barrier_nodf $SINGLEMDS
1370 fail_nodf $SINGLEMDS
1371 wait $open_pid || error "open_pid failed"
1373 # close should be gone
1374 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1375 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1377 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1378 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1379 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1380 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1381 rm -rf $DIR/${tdir}-*
1383 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1386 cancel_lru_locks mdc # cleanup locks from former test cases
1388 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1389 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1390 multiop $DIR/${tdir}-1/f O_c &
1393 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1394 do_facet $SINGLEMDS "lctl set_param fail_loc=0x119"
1395 mcreate $DIR/${tdir}-2/f &
1399 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1400 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000115"
1401 kill -USR1 $close_pid
1402 cancel_lru_locks mdc # force the close
1403 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1405 #bz20647: make sure all pids are exists before failover
1406 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1407 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1408 replay_barrier_nodf $SINGLEMDS
1409 fail_nodf $SINGLEMDS
1410 wait $open_pid || error "open_pid failed"
1412 # close should be gone
1413 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1415 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1416 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1417 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1418 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1419 rm -rf $DIR/${tdir}-*
1421 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1424 cancel_lru_locks mdc # cleanup locks from former test cases
1426 mkdir $DIR/${tdir}-1 || error "mkdir $DIR/${tdir}-1 failed"
1427 mkdir $DIR/${tdir}-2 || error "mkdir $DIR/${tdir}-2 failed"
1428 multiop $DIR/${tdir}-1/f O_c &
1431 #define OBD_FAIL_MDS_REINT_NET 0x107
1432 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000107"
1433 mcreate $DIR/${tdir}-2/f &
1437 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1438 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013b"
1439 kill -USR1 $close_pid
1440 cancel_lru_locks mdc # force the close
1443 #bz20647: make sure all pids are exists before failover
1444 [ -d /proc/$close_pid ] || error "close_pid doesn't exist"
1445 [ -d /proc/$open_pid ] || error "open_pid doesn't exists"
1446 replay_barrier_nodf $SINGLEMDS
1447 fail_nodf $SINGLEMDS
1448 wait $open_pid || error "open_pid failed"
1450 # close should be gone
1451 [ -d /proc/$close_pid ] && error "close_pid should not exist"
1452 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1454 $CHECKSTAT -t file $DIR/${tdir}-1/f ||
1455 error "$CHECKSTAT $DIR/${tdir}-1/f attribute check failed"
1456 $CHECKSTAT -t file $DIR/${tdir}-2/f ||
1457 error "$CHECKSTAT $DIR/${tdir}-2/f attribute check failed"
1458 rm -rf $DIR/${tdir}-*
1460 run_test 53h "open request and close reply while two MDC requests in flight"
1462 #b3761 ASSERTION(hash != 0) failed
1464 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1465 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012b"
1467 # give touch a chance to run
1469 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1473 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1475 #b3440 ASSERTION(rec->ur_fid2->id) failed
1477 ln -s foo $DIR/$tfile
1478 replay_barrier $SINGLEMDS
1479 #drop_reply "cat $DIR/$tfile"
1483 run_test 56 "don't replay a symlink open request (3440)"
1485 #recovery one mds-ost setattr from llog
1487 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1488 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1489 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1490 replay_barrier $SINGLEMDS
1493 $CHECKSTAT -t file $DIR/$tfile ||
1494 error "$CHECKSTAT $DIR/$tfile attribute check failed"
1495 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1498 run_test 57 "test recovery from llog for setattr op"
1501 zconf_umount $(hostname) $MOUNT2
1505 #recovery many mds-ost setattr from llog
1507 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1508 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1509 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000012c"
1510 createmany -o $DIR/$tdir/$tfile-%d 2500
1511 replay_barrier $SINGLEMDS
1514 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null ||
1515 error "$CHECKSTAT $DIR/$tfile-* attribute check failed"
1516 do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
1517 unlinkmany $DIR/$tdir/$tfile-%d 2500 ||
1518 error "unlinkmany $DIR/$tfile failed"
1521 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
1527 trap cleanup_58 EXIT
1529 large_xattr_enabled &&
1530 orig="$(generate_string $(max_xattr_size))" || orig="bar"
1532 mount_client $MOUNT2
1533 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1534 touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
1535 replay_barrier $SINGLEMDS
1536 setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile
1538 new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
1539 [[ "$new" = "$orig" ]] ||
1540 error "xattr set ($orig) is not what was returned ($new)"
1541 rm -f $DIR/$tdir/$tfile
1544 wait_clients_import_state ${CLIENTS:-$HOSTNAME} "mgs" FULL
1546 run_test 58b "test replay of setxattr op"
1548 test_58c() { # bug 16570
1553 trap cleanup_58 EXIT
1555 if large_xattr_enabled; then
1556 local xattr_size=$(max_xattr_size)
1557 orig="$(generate_string $((xattr_size / 2)))"
1558 orig1="$(generate_string $xattr_size)"
1564 # PING_INTERVAL max(obd_timeout / 4, 1U)
1565 sleep $((TIMEOUT / 4))
1567 mount_client $MOUNT2
1568 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1569 touch $DIR/$tdir/$tfile || error "touch $DIR/$tdir/$tfile failed"
1570 drop_request "setfattr -n trusted.foo -v $orig $DIR/$tdir/$tfile" ||
1571 error "drop_request for setfattr failed"
1572 new=$(get_xattr_value trusted.foo $MOUNT2/$tdir/$tfile)
1573 [[ "$new" = "$orig" ]] ||
1574 error "xattr set ($orig) is not what was returned ($new)"
1575 drop_reint_reply "setfattr -n trusted.foo1 \
1576 -v $orig1 $DIR/$tdir/$tfile" ||
1577 error "drop_request for setfattr failed"
1578 new=$(get_xattr_value trusted.foo1 $MOUNT2/$tdir/$tfile)
1579 [[ "$new" = "$orig1" ]] ||
1580 error "second xattr set ($orig1) not what was returned ($new)"
1581 rm -f $DIR/$tdir/$tfile
1585 run_test 58c "resend/reconstruct setxattr op"
1587 # log_commit_thread vs filter_destroy race used to lead to import use after free
1590 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1592 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1593 createmany -o $DIR/$tdir/$tfile-%d 200 ||
1594 error "createmany create files failed"
1596 unlinkmany $DIR/$tdir/$tfile-%d 200 ||
1597 error "unlinkmany $DIR/$tdir/$tfile failed"
1598 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1599 do_facet ost1 "lctl set_param fail_loc=0x507"
1602 do_facet ost1 "lctl set_param fail_loc=0x0"
1606 run_test 59 "test log_commit_thread vs filter_destroy race"
1608 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1609 # bug 12086: should no oops and No ctxt error for this test
1611 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1612 createmany -o $DIR/$tdir/$tfile-%d 200 ||
1613 error "createmany create files failed"
1614 replay_barrier $SINGLEMDS
1615 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1617 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1618 local no_ctxt=$(dmesg | grep "No ctxt")
1619 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1621 run_test 60 "test llog post recovery init vs llog unlink"
1623 #test race llog recovery thread vs llog cleanup
1624 test_61a() { # was test_61
1625 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1627 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1628 createmany -o $DIR/$tdir/$tfile-%d 800 ||
1629 error "createmany create files failed"
1631 unlinkmany $DIR/$tdir/$tfile-%d 800
1632 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1633 set_nodes_failloc "$(osts_nodes)" 0x80000221
1638 set_nodes_failloc "$(osts_nodes)" 0x0
1640 $CHECKSTAT -t file $DIR/$tdir/$tfile-* &&
1641 error "$CHECKSTAT $DIR/$tdir/$tfile attribute check should fail"
1644 run_test 61a "test race llog recovery vs llog cleanup"
1646 #test race mds llog sync vs llog cleanup
1648 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1649 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000013a"
1650 facet_failover $SINGLEMDS
1653 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 ||
1656 run_test 61b "test race mds llog sync vs llog cleanup"
1658 #test race cancel cookie cb vs llog cleanup
1660 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1662 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1663 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
1664 set_nodes_failloc "$(osts_nodes)" 0x80000222
1668 set_nodes_failloc "$(osts_nodes)" 0x0
1670 run_test 61c "test race mds llog sync vs llog cleanup"
1672 test_61d() { # bug 16002 # bug 17466 # bug 22137
1673 # OBD_FAIL_OBD_LLOG_SETUP 0x605
1675 do_facet mgs "lctl set_param fail_loc=0x80000605"
1676 start mgs $(mgsdevname) $MGS_MOUNT_OPTS &&
1677 error "mgs start should have failed"
1678 do_facet mgs "lctl set_param fail_loc=0"
1679 start mgs $(mgsdevname) $MGS_MOUNT_OPTS || error "cannot restart mgs"
1681 run_test 61d "error in llog_setup should cleanup the llog context correctly"
1683 test_62() { # Bug 15756 - don't mis-drop resent replay
1684 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1685 replay_barrier $SINGLEMDS
1686 createmany -o $DIR/$tdir/$tfile- 25 ||
1687 error "createmany create files failed"
1688 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1689 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000707"
1691 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
1692 unlinkmany $DIR/$tdir/$tfile- 25 ||
1693 error "unlinkmany $DIR/$tdir/$tfile failed"
1696 run_test 62 "don't mis-drop resent replay"
1698 #Adaptive Timeouts (bug 3055)
1706 echo "Cleaning up AT ..."
1707 if [ -n "$ATOLDBASE" ]; then
1708 local at_history=$($LCTL get_param -n at_history)
1709 do_facet $SINGLEMDS "lctl set_param at_history=$at_history" || true
1710 do_facet ost1 "lctl set_param at_history=$at_history" || true
1713 if [ $AT_MAX_SET -ne 0 ]; then
1714 for facet in mds client ost; do
1715 var=AT_MAX_SAVE_${facet}
1716 echo restore AT on $facet to saved value ${!var}
1717 at_max_set ${!var} $facet
1718 at_new=$(at_max_get $facet)
1719 echo Restored AT value on $facet $at_new
1720 [ $at_new -eq ${!var} ] ||
1721 error "AT value not restored SAVED ${!var} NEW $at_new"
1728 local at_max_new=600
1730 # Save at_max original values
1732 if [ $AT_MAX_SET -eq 0 ]; then
1733 # Suppose that all osts have the same at_max
1734 for facet in mds client ost; do
1735 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1739 for facet in mds client ost; do
1740 at_max=$(at_max_get $facet)
1741 if [ $at_max -ne $at_max_new ]; then
1742 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1743 at_max_set $at_max_new $facet
1748 if [ -z "$ATOLDBASE" ]; then
1749 ATOLDBASE=$(do_facet $SINGLEMDS "lctl get_param -n at_history")
1750 # speed up the timebase so we can check decreasing AT
1751 do_facet $SINGLEMDS "lctl set_param at_history=8" || true
1752 do_facet ost1 "lctl set_param at_history=8" || true
1754 # sleep for a while to cool down, should be > 8s and also allow
1755 # at least one ping to be sent. simply use TIMEOUT to be safe.
1760 test_65a() #bug 3055
1762 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1764 at_start || return 0
1765 $LCTL dk > /dev/null
1767 $LCTL set_param debug="other"
1768 # Slow down a request to the current service time, this is critical
1769 # because previous tests may have caused this value to increase.
1770 REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
1771 awk '/portal 12/ {print $5}'`
1772 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1774 do_facet $SINGLEMDS lctl set_param fail_val=$((${REQ_DELAY} * 1000))
1775 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1776 do_facet $SINGLEMDS $LCTL set_param fail_loc=0x8000050a
1777 createmany -o $DIR/$tfile 10 > /dev/null
1778 unlinkmany $DIR/$tfile 10 > /dev/null
1779 # check for log message
1780 $LCTL dk | grep "Early reply #" || error "No early reply"
1782 # client should show REQ_DELAY estimates
1783 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1785 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1787 run_test 65a "AT: verify early replies"
1789 test_65b() #bug 3055
1791 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1793 at_start || return 0
1796 $LCTL set_param debug="other trace"
1797 $LCTL dk > /dev/null
1798 # Slow down a request to the current service time, this is critical
1799 # because previous tests may have caused this value to increase.
1800 $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tfile
1801 multiop $DIR/$tfile Ow1yc
1802 REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
1803 awk '/portal 6/ {print $5}'`
1804 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1806 do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
1807 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1808 do_facet ost1 $LCTL set_param fail_loc=0x224
1811 $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tfile
1812 # force some real bulk transfer
1813 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1815 do_facet ost1 $LCTL set_param fail_loc=0
1816 # check for log message
1817 $LCTL dk | grep "Early reply #" || error "No early reply"
1819 # client should show REQ_DELAY estimates
1820 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1822 run_test 65b "AT: verify early replies on packed reply / bulk"
1824 test_66a() #bug 3055
1826 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1828 at_start || return 0
1829 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1830 # adjust 5s at a time so no early reply is sent (within deadline)
1831 do_facet $SINGLEMDS "$LCTL set_param fail_val=5000"
1832 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1833 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x8000050a"
1834 createmany -o $DIR/$tfile 20 > /dev/null
1835 unlinkmany $DIR/$tfile 20 > /dev/null
1836 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1837 do_facet $SINGLEMDS "$LCTL set_param fail_val=10000"
1838 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0x8000050a"
1839 createmany -o $DIR/$tfile 20 > /dev/null
1840 unlinkmany $DIR/$tfile 20 > /dev/null
1841 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1842 do_facet $SINGLEMDS "$LCTL set_param fail_loc=0"
1844 createmany -o $DIR/$tfile 20 > /dev/null
1845 unlinkmany $DIR/$tfile 20 > /dev/null
1846 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1847 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1848 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1849 echo "Current MDT timeout $CUR, worst $WORST"
1850 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1852 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1854 test_66b() #bug 3055
1856 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1858 at_start || return 0
1859 ORIG=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1860 awk '/network/ {print $4}')
1861 $LCTL set_param fail_val=$(($ORIG + 5))
1862 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1863 $LCTL set_param fail_loc=0x50c
1864 ls $DIR/$tfile > /dev/null 2>&1
1865 $LCTL set_param fail_loc=0
1866 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1867 awk '/network/ {print $4}')
1868 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000*.timeouts |
1869 awk '/network/ {print $6}')
1870 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1871 [ $WORST -gt $ORIG ] ||
1872 error "Worst $WORST should be worse than orig $ORIG"
1874 run_test 66b "AT: verify net latency adjusts"
1876 test_67a() #bug 3055
1878 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1880 at_start || return 0
1881 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1882 # sleeping threads may drive values above this
1883 do_facet ost1 "$LCTL set_param fail_val=400"
1884 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1885 do_facet ost1 "$LCTL set_param fail_loc=0x50a"
1886 createmany -o $DIR/$tfile 20 > /dev/null
1887 unlinkmany $DIR/$tfile 20 > /dev/null
1888 do_facet ost1 "$LCTL set_param fail_loc=0"
1889 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1890 ATTEMPTS=$(($CONN2 - $CONN1))
1891 echo "$ATTEMPTS osc reconnect attempts on gradual slow"
1892 [ $ATTEMPTS -gt 0 ] &&
1893 error_ignore bz13721 "AT should have prevented reconnect"
1896 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1898 test_67b() #bug 3055
1900 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1902 at_start || return 0
1903 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1905 # exhaust precreations on ost1
1906 local OST=$(ostname_from_index 0)
1907 local mdtosc=$(get_mdtosc_proc_path mds $OST)
1908 local last_id=$(do_facet $SINGLEMDS lctl get_param -n \
1909 osc.$mdtosc.prealloc_last_id)
1910 local next_id=$(do_facet $SINGLEMDS lctl get_param -n \
1911 osc.$mdtosc.prealloc_next_id)
1913 mkdir -p $DIR/$tdir/${OST} || error "mkdir $DIR/$tdir/${OST} failed"
1914 $SETSTRIPE -i 0 -c 1 $DIR/$tdir/${OST} || error "$SETSTRIPE failed"
1915 echo "Creating to objid $last_id on ost $OST..."
1916 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1917 do_facet ost1 "$LCTL set_param fail_val=20000"
1918 do_facet ost1 "$LCTL set_param fail_loc=0x80000223"
1919 createmany -o $DIR/$tdir/${OST}/f $next_id $((last_id - next_id + 2))
1922 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1924 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1925 ATTEMPTS=$(($CONN2 - $CONN1))
1926 echo "$ATTEMPTS osc reconnect attempts on instant slow"
1927 # do it again; should not timeout
1928 do_facet ost1 "$LCTL set_param fail_loc=0x80000223"
1929 cp /etc/profile $DIR/$tfile || error "cp failed"
1930 do_facet ost1 "$LCTL set_param fail_loc=0"
1932 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1933 CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1934 ATTEMPTS=$(($CONN3 - $CONN2))
1935 echo "$ATTEMPTS osc reconnect attempts on 2nd slow"
1936 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1939 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1941 test_68 () #bug 13813
1943 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1945 at_start || return 0
1946 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1947 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1948 local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
1949 [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
1950 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1951 local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
1952 echo $TIMEOUT >> $ldlm_enqueue_min
1953 do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
1955 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
1956 $SETSTRIPE --stripe-index=0 --count=1 $DIR/$tdir
1957 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1958 $LCTL set_param fail_val=$(($TIMEOUT - 1))
1959 $LCTL set_param fail_loc=0x80000312
1960 cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
1961 $LCTL set_param fail_val=$((TIMEOUT * 5 / 4))
1962 $LCTL set_param fail_loc=0x80000312
1963 cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
1964 $LCTL set_param fail_loc=0
1966 echo $ENQ_MIN >> $ldlm_enqueue_min
1967 do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
1971 run_test 68 "AT: verify slowing locks"
1974 # end of AT tests includes above lines
1976 # start multi-client tests
1978 [ -z "$CLIENTS" ] &&
1979 { skip "Need two or more clients." && return; }
1980 [ $CLIENTCOUNT -lt 2 ] &&
1981 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1983 echo "mount clients $CLIENTS ..."
1984 zconf_mount_clients $CLIENTS $MOUNT
1986 local clients=${CLIENTS//,/ }
1987 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1988 for CLIENT in $clients; do
1989 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1990 of=$DIR/${tfile}_${CLIENT} 2>/dev/null ||
1991 error "dd failed on $CLIENT"
1994 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
1995 for C in ${CLIENTS//,/ }; do
1996 do_node $prev_client dd if=$DIR/${tfile}_${C} \
1997 of=/dev/null 2>/dev/null ||
1998 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
2004 run_test 70a "check multi client t-f"
2006 check_for_process () {
2011 killall_process $clients "$prog" -0
2014 killall_process () {
2015 local clients=${1:-$(hostname)}
2020 do_nodes $clients "killall $signal $name"
2024 local clients=${CLIENTS:-$HOSTNAME}
2026 zconf_mount_clients $clients $MOUNT
2029 [ "$SLOW" = "no" ] && duration=120
2030 # set duration to 900 because it takes some time to boot node
2031 [ "$FAILURE_MODE" = HARD ] && duration=900
2034 local start_ts=$(date +%s)
2035 local cmd="rundbench 1 -t $duration"
2037 do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
2038 PATH=\$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
2039 DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
2040 MOUNT=$MOUNT DIR=$DIR/$tdir/\\\$(hostname) LCTL=$LCTL $cmd" &
2043 #LU-1897 wait for all dbench copies to start
2044 while ! check_for_process $clients dbench; do
2045 elapsed=$(($(date +%s) - start_ts))
2046 if [ $elapsed -gt $duration ]; then
2047 killall_process $clients dbench
2048 error "dbench failed to start on $clients!"
2053 log "Started rundbench load pid=$pid ..."
2055 elapsed=$(($(date +%s) - start_ts))
2056 local num_failovers=0
2057 while [ $elapsed -lt $duration ]; do
2058 if ! check_for_process $clients dbench; then
2059 error_noexit "dbench stopped on some of $clients!"
2060 killall_process $clients dbench
2064 replay_barrier $SINGLEMDS
2065 sleep 1 # give clients a time to do operations
2066 # Increment the number of failovers
2067 num_failovers=$((num_failovers+1))
2068 log "$TESTNAME fail $SINGLEMDS $num_failovers times"
2070 elapsed=$(($(date +%s) - start_ts))
2073 wait $pid || error "rundbench load on $clients failed!"
2075 run_test 70b "mds recovery; $CLIENTCOUNT clients"
2076 # end multi-client tests
2079 multiop_bg_pause $DIR/$tfile O_tSc ||
2080 error "multiop_bg_pause $DIR/$tfile failed"
2084 replay_barrier $SINGLEMDS
2085 #define OBD_FAIL_LDLM_ENQUEUE_NET 0x302
2086 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000302"
2089 wait $pid || error "multiop pid failed"
2090 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
2093 run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay, close"
2096 multiop_bg_pause $DIR/$tfile O_tSc ||
2097 error "multiop_bg_pause $DIR/$tfile failed"
2101 replay_barrier $SINGLEMDS
2102 #define OBD_FAIL_LDLM_REPLY 0x30c
2103 do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
2106 wait $pid || error "multiop pid failed"
2107 [ -e $DIR/$tfile ] && error "file $DIR/$tfile should not exist"
2110 run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
2114 local clients=${CLIENTS:-$HOSTNAME}
2116 zconf_umount_clients $clients $MOUNT
2118 facet_failover $SINGLEMDS
2119 zconf_mount_clients $clients $MOUNT
2121 touch $DIR/$tfile || error "touch $DIR/$tfile failed"
2122 rm $DIR/$tfile || error "rm $DIR/$tfile failed"
2123 clients_up || error "client evicted: $?"
2126 run_test 74 "Ensure applications don't fail waiting for OST recovery"
2128 remote_dir_check_80() {
2133 diridx=$($GETSTRIPE -M $remote_dir) ||
2134 error "$GETSTRIPE -M $remote_dir failed"
2135 [ $diridx -eq $MDTIDX ] || error "$diridx != $MDTIDX"
2137 createmany -o $remote_dir/f-%d 20 || error "creation failed"
2138 fileidx=$($GETSTRIPE -M $remote_dir/f-1) ||
2139 error "$GETSTRIPE -M $remote_dir/f-1 failed"
2140 [ $fileidx -eq $MDTIDX ] || error "$fileidx != $MDTIDX"
2146 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2147 ([ $FAILURE_MODE == "HARD" ] &&
2148 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2149 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2153 local remote_dir=$DIR/$tdir/remote_dir
2155 mkdir -p $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2156 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2157 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2158 $LFS mkdir -i $MDTIDX $remote_dir &
2163 wait $CLIENT_PID || error "remote creation failed"
2165 remote_dir_check_80 || error "remote dir check failed"
2166 rm -rf $DIR/$tdir || error "rmdir failed"
2170 run_test 80a "DNE: create remote dir, drop update rep from MDT0, fail MDT0"
2173 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2174 ([ $FAILURE_MODE == "HARD" ] &&
2175 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2176 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2180 local remote_dir=$DIR/$tdir/remote_dir
2182 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2183 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2184 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2185 $LFS mkdir -i $MDTIDX $remote_dir &
2188 fail mds$((MDTIDX + 1))
2190 wait $CLIENT_PID || error "remote creation failed"
2192 remote_dir_check_80 || error "remote dir check failed"
2193 rm -rf $DIR/$tdir || error "rmdir failed"
2197 run_test 80b "DNE: create remote dir, drop update rep from MDT0, fail MDT1"
2200 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2201 ([ $FAILURE_MODE == "HARD" ] &&
2202 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2203 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2207 local remote_dir=$DIR/$tdir/remote_dir
2209 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2210 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2211 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2212 $LFS mkdir -i $MDTIDX $remote_dir &
2216 fail mds$((MDTIDX + 1))
2218 wait $CLIENT_PID || error "remote creation failed"
2220 remote_dir_check_80 || error "remote dir check failed"
2221 rm -rf $DIR/$tdir || error "rmdir failed"
2225 run_test 80c "DNE: create remote dir, drop update rep from MDT1, fail MDT[0,1]"
2228 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2230 local remote_dir=$DIR/$tdir/remote_dir
2232 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2233 #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
2234 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2235 $LFS mkdir -i $MDTIDX $remote_dir &
2238 # sleep 3 seconds to make sure MDTs are failed after
2239 # lfs mkdir -i has finished on all of MDTs.
2241 fail mds${MDTIDX},mds$((MDTIDX + 1))
2243 wait $CLIENT_PID || error "remote creation failed"
2245 remote_dir_check_80 || error "remote dir check failed"
2246 rm -rf $DIR/$tdir || error "rmdir failed"
2250 run_test 80d "DNE: create remote dir, drop update rep from MDT1, fail 2 MDTs"
2253 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2254 ([ $FAILURE_MODE == "HARD" ] &&
2255 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2256 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2260 local remote_dir=$DIR/$tdir/remote_dir
2262 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2263 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2264 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2265 $LFS mkdir -i $MDTIDX $remote_dir &
2268 # sleep 3 seconds to make sure MDTs are failed after
2269 # lfs mkdir -i has finished on all of MDTs.
2274 wait $CLIENT_PID || error "remote creation failed"
2276 remote_dir_check_80 || error "remote dir check failed"
2277 rm -rf $DIR/$tdir || error "rmdir failed"
2281 run_test 80e "DNE: create remote dir, drop MDT1 rep, fail MDT0"
2284 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2285 ([ $FAILURE_MODE == "HARD" ] &&
2286 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2287 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2290 local remote_dir=$DIR/$tdir/remote_dir
2292 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2293 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2294 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2295 $LFS mkdir -i $MDTIDX $remote_dir &
2298 fail mds$((MDTIDX + 1))
2300 wait $CLIENT_PID || error "remote creation failed"
2302 remote_dir_check_80 || error "remote dir check failed"
2303 rm -rf $DIR/$tdir || error "rmdir failed"
2307 run_test 80f "DNE: create remote dir, drop MDT1 rep, fail MDT1"
2310 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2311 ([ $FAILURE_MODE == "HARD" ] &&
2312 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2313 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2317 local remote_dir=$DIR/$tdir/remote_dir
2319 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2320 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2321 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2322 $LFS mkdir -i $MDTIDX $remote_dir &
2325 # sleep 3 seconds to make sure MDTs are failed after
2326 # lfs mkdir -i has finished on all of MDTs.
2330 fail mds$((MDTIDX + 1))
2332 wait $CLIENT_PID || error "remote creation failed"
2334 remote_dir_check_80 || error "remote dir check failed"
2335 rm -rf $DIR/$tdir || error "rmdir failed"
2339 run_test 80g "DNE: create remote dir, drop MDT1 rep, fail MDT0, then MDT1"
2342 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2344 local remote_dir=$DIR/$tdir/remote_dir
2346 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2347 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2348 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2349 $LFS mkdir -i $MDTIDX $remote_dir &
2352 # sleep 3 seconds to make sure MDTs are failed after
2353 # lfs mkdir -i has finished on all of MDTs.
2356 fail mds${MDTIDX},mds$((MDTIDX + 1))
2358 wait $CLIENT_PID || error "remote dir creation failed"
2360 remote_dir_check_80 || error "remote dir check failed"
2361 rm -rf $DIR/$tdir || error "rmdir failed"
2365 run_test 80h "DNE: create remote dir, drop MDT1 rep, fail 2 MDTs"
2368 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2369 ([ $FAILURE_MODE == "HARD" ] &&
2370 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2371 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2375 local remote_dir=$DIR/$tdir/remote_dir
2377 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2378 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2380 touch $remote_dir || error "touch $remote_dir failed"
2381 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2382 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2386 fail mds$((MDTIDX + 1))
2388 wait $CLIENT_PID || error "rm remote dir failed"
2390 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2392 rm -rf $DIR/$tdir || error "rmdir failed"
2396 run_test 81a "DNE: unlink remote dir, drop MDT0 update rep, fail MDT1"
2399 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2400 ([ $FAILURE_MODE == "HARD" ] &&
2401 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2402 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2405 local remote_dir=$DIR/$tdir/remote_dir
2407 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2408 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2410 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2411 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2417 wait $CLIENT_PID || error "rm remote dir failed"
2419 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2421 rm -rf $DIR/$tdir || error "rmdir failed"
2425 run_test 81b "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0"
2428 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2429 ([ $FAILURE_MODE == "HARD" ] &&
2430 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2431 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2435 local remote_dir=$DIR/$tdir/remote_dir
2437 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2438 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2440 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2441 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2446 fail mds$((MDTIDX + 1))
2448 wait $CLIENT_PID || error "rm remote dir failed"
2450 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2452 rm -rf $DIR/$tdir || error "rmdir failed"
2456 run_test 81c "DNE: unlink remote dir, drop MDT0 update reply, fail MDT0,MDT1"
2459 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2461 local remote_dir=$DIR/$tdir/remote_dir
2463 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2464 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2466 # OBD_FAIL_OBJ_UPDATE_NET_REP 0x1701
2467 do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
2471 fail mds${MDTIDX},mds$((MDTIDX + 1))
2473 wait $CLIENT_PID || error "rm remote dir failed"
2475 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2477 rm -rf $DIR/$tdir || error "rmdir failed"
2481 run_test 81d "DNE: unlink remote dir, drop MDT0 update reply, fail 2 MDTs"
2484 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2485 ([ $FAILURE_MODE == "HARD" ] &&
2486 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2487 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2491 local remote_dir=$DIR/$tdir/remote_dir
2493 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2494 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2496 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2497 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2500 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
2504 wait $CLIENT_PID || error "rm remote dir failed"
2506 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2508 rm -rf $DIR/$tdir || error "rmdir failed"
2512 run_test 81e "DNE: unlink remote dir, drop MDT1 req reply, fail MDT0"
2515 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2516 ([ $FAILURE_MODE == "HARD" ] &&
2517 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2518 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2522 local remote_dir=$DIR/$tdir/remote_dir
2524 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2525 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2527 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2528 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2532 fail mds$((MDTIDX + 1))
2534 wait $CLIENT_PID || error "rm remote dir failed"
2536 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2538 rm -rf $DIR/$tdir || error "rmdir failed"
2542 run_test 81f "DNE: unlink remote dir, drop MDT1 req reply, fail MDT1"
2545 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2546 ([ $FAILURE_MODE == "HARD" ] &&
2547 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2548 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2552 local remote_dir=$DIR/$tdir/remote_dir
2554 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2555 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2557 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2558 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2563 fail mds$((MDTIDX + 1))
2565 wait $CLIENT_PID || error "rm remote dir failed"
2567 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2569 rm -rf $DIR/$tdir || error "rmdir failed"
2573 run_test 81g "DNE: unlink remote dir, drop req reply, fail M0, then M1"
2576 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2578 local remote_dir=$DIR/$tdir/remote_dir
2580 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2581 $LFS mkdir -i $MDTIDX $remote_dir || error "lfs mkdir failed"
2583 # OBD_FAIL_MDS_REINT_NET_REP 0x119
2584 do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
2588 fail mds${MDTIDX},mds$((MDTIDX + 1))
2590 wait $CLIENT_PID || error "rm remote dir failed"
2592 stat $remote_dir 2&>/dev/null && error "$remote_dir still exist!"
2594 rm -rf $DIR/$tdir || error "rmdir failed"
2598 run_test 81h "DNE: unlink remote dir, drop request reply, fail 2 MDTs"
2601 #define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x144
2602 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000144"
2603 createmany -o $DIR/$tfile- 1 &
2607 client_up || client_up || true # reconnect
2609 run_test 84a "stale open during export disconnect"
2611 test_85a() { #bug 16774
2612 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
2614 for i in $(seq 100); do
2615 echo "tag-$i" > $DIR/$tfile-$i
2616 grep -q "tag-$i" $DIR/$tfile-$i || error "f2-$i"
2619 lov_id=$(lctl dl | grep "clilov")
2620 addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}')
2621 count=$(lctl get_param -n \
2622 ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
2623 echo "before recovery: unused locks count = $count"
2627 count2=$(lctl get_param -n \
2628 ldlm.namespaces.*MDT0000*$addr.lock_unused_count)
2629 echo "after recovery: unused locks count = $count2"
2631 if [ $count2 -ge $count ]; then
2632 error "unused locks are not canceled"
2635 run_test 85a "check the cancellation of unused locks during recovery(IBITS)"
2637 test_85b() { #bug 16774
2638 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
2640 do_facet mgs $LCTL pool_new $FSNAME.$TESTNAME ||
2641 error "unable to create pool $TESTNAME"
2642 do_facet mgs $LCTL pool_add $FSNAME.$TESTNAME $FSNAME-OST0000 ||
2643 error "unable to add pool $TESTNAME"
2645 $SETSTRIPE -c 1 -p $FSNAME.$TESTNAME $DIR
2647 for i in $(seq 100); do
2648 dd if=/dev/urandom of=$DIR/$tfile-$i bs=4096 \
2649 count=32 >/dev/null 2>&1
2652 cancel_lru_locks osc
2654 for i in $(seq 100); do
2655 dd if=$DIR/$tfile-$i of=/dev/null bs=4096 \
2656 count=32 >/dev/null 2>&1
2659 lov_id=$(lctl dl | grep "clilov")
2660 addr=$(echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}')
2661 count=$(lctl get_param \
2662 -n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
2663 echo "before recovery: unused locks count = $count"
2664 [ $count != 0 ] || error "unused locks ($count) should be zero"
2668 count2=$(lctl get_param \
2669 -n ldlm.namespaces.*OST0000*$addr.lock_unused_count)
2670 echo "after recovery: unused locks count = $count2"
2672 do_facet mgs $LCTL pool_remove $FSNAME.$TESTNAME $FSNAME-OST0000 ||
2673 error "unable to remove pool $TESTNAME"
2674 do_facet mgs $LCTL pool_destroy $FSNAME.$TESTNAME ||
2675 error "unable to destroy the pool $TESTNAME"
2677 if [ $count2 -ge $count ]; then
2678 error "unused locks are not canceled"
2681 run_test 85b "check the cancellation of unused locks during recovery(EXTENT)"
2684 local clients=${CLIENTS:-$HOSTNAME}
2686 zconf_umount_clients $clients $MOUNT
2687 do_facet $SINGLEMDS lctl set_param mdt.${FSNAME}-MDT*.exports.clear=0
2688 remount_facet $SINGLEMDS
2689 zconf_mount_clients $clients $MOUNT
2691 run_test 86 "umount server after clear nid_stats should not hit LBUG"
2694 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
2697 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
2698 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
2699 error "dd to $DIR/$tfile failed"
2700 cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
2701 cancel_lru_locks osc
2703 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
2704 cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
2705 if [ $cksum != $cksum2 ] ; then
2706 error "New checksum $cksum2 does not match original $cksum"
2709 run_test 87 "write replay"
2712 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
2715 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
2716 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 ||
2717 error "dd to $DIR/$tfile failed"
2718 sleep 1 # Give it a chance to flush dirty data
2719 echo TESTTEST | dd of=$DIR/$tfile bs=1 count=8 seek=64
2720 cksum=$(md5sum $DIR/$tfile | awk '{print $1}')
2721 cancel_lru_locks osc
2723 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
2724 cksum2=$(md5sum $DIR/$tfile | awk '{print $1}')
2725 if [ $cksum != $cksum2 ] ; then
2726 error "New checksum $cksum2 does not match original $cksum"
2729 run_test 87b "write replay with changed data (checksum resend)"
2731 test_88() { #bug 17485
2732 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2733 mkdir -p $TMP/$tdir || error "mkdir $TMP/$tdir failed"
2735 $SETSTRIPE -i 0 -c 1 $DIR/$tdir || error "$SETSTRIPE"
2738 replay_barrier $SINGLEMDS
2740 # exhaust precreations on ost1
2741 local OST=$(ostname_from_index 0)
2742 local mdtosc=$(get_mdtosc_proc_path $SINGLEMDS $OST)
2743 local last_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
2744 local next_id=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
2745 echo "before test: last_id = $last_id, next_id = $next_id"
2747 echo "Creating to objid $last_id on ost $OST..."
2748 createmany -o $DIR/$tdir/f-%d $next_id $((last_id - next_id + 2)) ||
2749 error "createmany create files to last_id failed"
2751 #create some files to use some uncommitted objids
2752 last_id=$(($last_id + 1))
2753 createmany -o $DIR/$tdir/f-%d $last_id 8 ||
2754 error "createmany create files with uncommitted objids failed"
2756 last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
2757 next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
2758 echo "before recovery: last_id = $last_id2, next_id = $next_id2"
2760 # if test uses shutdown_facet && reboot_facet instead of facet_failover ()
2761 # it has to take care about the affected facets, bug20407
2762 local affected_mds1=$(affected_facets mds1)
2763 local affected_ost1=$(affected_facets ost1)
2765 shutdown_facet $SINGLEMDS
2768 reboot_facet $SINGLEMDS
2769 change_active $affected_mds1
2770 wait_for_facet $affected_mds1
2771 mount_facets $affected_mds1 || error "Restart of mds failed"
2774 change_active $affected_ost1
2775 wait_for_facet $affected_ost1
2776 mount_facets $affected_ost1 || error "Restart of ost1 failed"
2780 last_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_last_id)
2781 next_id2=$(do_facet $SINGLEMDS lctl get_param -n osc.$mdtosc.prealloc_next_id)
2782 echo "after recovery: last_id = $last_id2, next_id = $next_id2"
2784 # create new files, which should use new objids, and ensure the orphan
2785 # cleanup phase for ost1 is completed at the same time
2786 for i in $(seq 8); do
2787 file_id=$(($last_id + 10 + $i))
2788 dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
2791 # if the objids were not recreated, then "ls" will fail with -ENOENT
2792 ls -l $DIR/$tdir/* || error "can't get the status of precreated files"
2795 # write into previously created files
2796 for i in $(seq 8); do
2797 file_id=$(($last_id + $i))
2798 dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
2799 cp -f $DIR/$tdir/f-$file_id $TMP/$tdir/
2802 # compare the content
2803 for i in $(seq 8); do
2804 file_id=$(($last_id + $i))
2805 cmp $TMP/$tdir/f-$file_id $DIR/$tdir/f-$file_id ||
2806 error "the content of file is modified!"
2811 run_test 88 "MDS should not assign same objid to different files "
2814 cancel_lru_locks osc
2815 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2816 rm -f $DIR/$tdir/$tfile
2818 wait_delete_completed
2819 BLOCKS1=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
2820 $SETSTRIPE -i 0 -c 1 $DIR/$tdir/$tfile
2821 dd if=/dev/zero bs=1M count=10 of=$DIR/$tdir/$tfile
2824 facet_failover $SINGLEMDS
2825 rm $DIR/$tdir/$tfile
2828 zconf_mount $(hostname) $MOUNT || error "mount fails"
2829 client_up || error "client_up failed"
2831 wait_delete_completed
2832 BLOCKS2=$(df -P $MOUNT | tail -n 1 | awk '{ print $3 }')
2833 [ $((BLOCKS2 - BLOCKS1)) -le 4 ] ||
2834 error $((BLOCKS2 - BLOCKS1)) blocks leaked
2837 run_test 89 "no disk space leak on late ost connection"
2843 change_active $facet
2844 wait_for_facet $facet
2845 mount_facet $facet || error "Restart of $facet failed"
2849 test_90() { # bug 19494
2850 local dir=$DIR/$tdir
2851 local ostfail=$(get_random_entry $(get_facets OST))
2853 if [[ $FAILURE_MODE = HARD ]]; then
2854 local affected=$(affected_facets $ostfail);
2855 if [[ "$affected" != $ostfail ]]; then
2856 skip not functional with FAILURE_MODE=$FAILURE_MODE, affected: $affected
2861 mkdir $dir || error "mkdir $dir failed"
2863 echo "Create the files"
2865 # file "f${index}" striped over 1 OST
2866 # file "all" striped over all OSTs
2868 $SETSTRIPE -c $OSTCOUNT $dir/all ||
2869 error "setstripe failed to create $dir/all"
2871 for (( i=0; i<$OSTCOUNT; i++ )); do
2873 $SETSTRIPE -i $i -c 1 $f || error "$SETSTRIPE failed to create $f"
2875 # confirm setstripe actually created the stripe on the requested OST
2876 local uuid=$(ostuuid_from_index $i)
2877 for file in f$i all; do
2878 if [[ $dir/$file != $($LFS find --obd $uuid --name $file $dir) ]]; then
2879 $GETSTRIPE $dir/$file
2880 error wrong stripe: $file, uuid: $uuid
2885 # Before failing an OST, get its obd name and index
2886 local varsvc=${ostfail}_svc
2887 local obd=$(do_facet $ostfail lctl get_param \
2888 -n obdfilter.${!varsvc}.uuid)
2889 local index=$(($(facet_number $ostfail) - 1))
2891 echo "Fail $ostfail $obd, display the list of affected files"
2892 shutdown_facet $ostfail || error "shutdown_facet $ostfail failed"
2894 trap "cleanup_90 $ostfail" EXIT INT
2895 echo "General Query: lfs find $dir"
2896 local list=$($LFS find $dir)
2898 for (( i=0; i<$OSTCOUNT; i++ )); do
2899 list_member "$list" $dir/f$i ||
2900 error_noexit "lfs find $dir: no file f$i"
2902 list_member "$list" $dir/all ||
2903 error_noexit "lfs find $dir: no file all"
2905 # focus on the missing OST,
2906 # we expect to see only two files affected: "f$(index)" and "all"
2908 echo "Querying files on shutdown $ostfail: lfs find --obd $obd"
2909 list=$($LFS find --obd $obd $dir)
2911 for file in all f$index; do
2912 list_member "$list" $dir/$file ||
2913 error_noexit "lfs find does not report the affected $obd for $file"
2916 [[ $(echo $list | wc -w) -eq 2 ]] ||
2917 error_noexit "lfs find reports the wrong list of affected files ${#list[@]}"
2919 echo "Check getstripe: $GETSTRIPE -r --obd $obd"
2920 list=$($GETSTRIPE -r --obd $obd $dir)
2922 for file in all f$index; do
2923 echo "$list" | grep $dir/$file ||
2924 error_noexit "lfs getsripe does not report the affected $obd for $file"
2929 run_test 90 "lfs find identifies the missing striped file segments"
2932 local server_version=$(lustre_version_code $SINGLEMDS)
2933 [[ $server_version -ge $(version_code 2.6.90) ]] ||
2934 [[ $server_version -ge $(version_code 2.5.4) &&
2935 $server_version -lt $(version_code 2.5.50) ]] ||
2936 { skip "Need MDS version 2.5.4+ or 2.6.90+"; return; }
2938 cancel_lru_locks osc
2940 $SETSTRIPE -i 0 -c 1 $DIR/$tfile ||
2941 error "$SETSTRIPE $DIR/$tfile failed"
2942 dd if=/dev/zero of=$DIR/$tfile bs=1024 count=1 ||
2943 error "dd to $DIR/$tfile failed"
2944 #define OBD_FAIL_TGT_REPLAY_RECONNECT 0x715
2945 # We need to emulate a state that OST is waiting for other clients
2946 # not completing the recovery. Final ping is queued, but reply will be
2947 # sent on the recovery completion. It is done by sleep before
2948 # processing final pings
2949 do_facet ost1 "$LCTL set_param fail_val=40"
2950 do_facet ost1 "$LCTL set_param fail_loc=0x715"
2953 run_test 93 "replay + reconnect"
2955 striped_dir_check_100() {
2956 local striped_dir=$DIR/$tdir/striped_dir
2957 local stripe_count=$($LFS getdirstripe -c $striped_dir)
2959 $LFS getdirstripe $striped_dir
2960 [ $stripe_count -eq 2 ] || error "$stripe_count != 2"
2962 createmany -o $striped_dir/f-%d 20 ||
2963 error "creation failed under striped dir"
2967 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2968 ([ $FAILURE_MODE == "HARD" ] &&
2969 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
2970 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
2973 local striped_dir=$DIR/$tdir/striped_dir
2976 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
2978 #To make sure MDT1 and MDT0 are connected
2979 #otherwise it may create single stripe dir here
2980 $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
2982 #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
2983 do_facet mds$((MDTIDX+1)) lctl set_param fail_loc=0x1701
2984 $LFS setdirstripe -i0 -c2 $striped_dir &
2987 fail mds$((MDTIDX + 1))
2989 wait $CLIENT_PID || error "striped dir creation failed"
2991 striped_dir_check_100 || error "striped dir check failed"
2992 rm -rf $DIR/$tdir || error "rmdir failed"
2994 run_test 100a "DNE: create striped dir, drop update rep from MDT1, fail MDT1"
2997 [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
2998 ([ $FAILURE_MODE == "HARD" ] &&
2999 [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
3000 skip "MDTs needs to be on diff hosts for HARD fail mode" &&
3003 local striped_dir=$DIR/$tdir/striped_dir
3006 mkdir $DIR/$tdir || error "mkdir $DIR/$tdir failed"
3008 #To make sure MDT1 and MDT0 are connected
3009 #otherwise it may create single stripe dir here
3010 $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
3012 # OBD_FAIL_MDS_REINT_NET_REP 0x119
3013 do_facet mds$MDTIDX lctl set_param fail_loc=0x119
3014 $LFS mkdir -i0 -c2 $striped_dir &
3019 wait $CLIENT_PID || error "striped dir creation failed"
3021 striped_dir_check_100 || error "striped dir check failed"
3022 rm -rf $DIR/$tdir || error "rmdir failed"
3024 run_test 100b "DNE: create striped dir, fail MDT0"
3026 test_101() { #LU-5648
3027 mkdir -p $DIR/$tdir/d1
3028 mkdir -p $DIR/$tdir/d2
3029 touch $DIR/$tdir/file0
3032 replay_barrier $SINGLEMDS
3033 for i in $(seq $num) ; do
3034 echo test$i > $DIR/$tdir/d1/file$i
3037 fail_abort $SINGLEMDS
3038 for i in $(seq $num) ; do
3039 touch $DIR/$tdir/d2/file$i
3040 test -s $DIR/$tdir/d2/file$i &&
3041 ls -al $DIR/$tdir/d2/file$i && error "file$i's size > 0"
3046 run_test 101 "Shouldn't reassign precreated objs to other files after recovery"
3049 check_and_cleanup_lustre