7 # This test needs to be run on the client
10 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
13 . $LUSTRE/tests/test-framework.sh
15 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
16 CHECK_GRANT=${CHECK_GRANT:-"yes"}
17 GRANT_CHECK_LIST=${GRANT_CHECK_LIST:-""}
19 remote_mds_nodsh && log "SKIP: remote MDS with nodsh" && exit 0
23 ALWAYS_EXCEPT="$REPLAY_SINGLE_EXCEPT"
25 if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
26 CONFIG_EXCEPTIONS="0b 42 47 61a 61c"
27 echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
28 echo "Except the tests: $CONFIG_EXCEPTIONS"
29 ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
32 # 63 min 7 min AT AT AT AT"
33 [ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a 44b 65 66 67 68"
37 check_and_setup_lustre
42 rm -rf $DIR/[df][0-9]*
44 test_0a() { # was test_0
48 run_test 0a "empty replay"
51 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
53 # this test attempts to trigger a race in the precreation code,
54 # and must run before any other objects are created on the filesystem
56 createmany -o $DIR/$tfile 20 || return 1
57 unlinkmany $DIR/$tfile 20 || return 2
59 run_test 0b "ensure object created after recover exists. (3284)"
65 zconf_mount `hostname` $DIR || error "mount fails"
66 clients_up || error "post-failover df failed"
68 run_test 0c "expired recovery with no clients"
74 $CHECKSTAT -t file $DIR/$tfile || return 1
77 run_test 1 "simple create"
83 $CHECKSTAT -t file $DIR/$tfile || return 1
93 $CHECKSTAT -t file $DIR/$tfile || return 1
99 local file=$DIR/$tfile
102 openfile -f O_DIRECTORY $file
104 $CHECKSTAT -t file $file || return 2
107 run_test 3a "replay failed open(O_DIRECTORY)"
111 #define OBD_FAIL_MDS_OPEN_PACK | OBD_FAIL_ONCE
112 do_facet mds "lctl set_param fail_loc=0x80000114"
114 do_facet mds "lctl set_param fail_loc=0"
116 $CHECKSTAT -t file $DIR/$tfile && return 2
119 run_test 3b "replay failed open -ENOMEM"
123 #define OBD_FAIL_MDS_ALLOC_OBDO | OBD_FAIL_ONCE
124 do_facet mds "lctl set_param fail_loc=0x80000128"
126 do_facet mds "lctl set_param fail_loc=0"
129 $CHECKSTAT -t file $DIR/$tfile && return 2
132 run_test 3c "replay failed open -ENOMEM"
134 test_4a() { # was test_4
136 for i in `seq 10`; do
137 echo "tag-$i" > $DIR/$tfile-$i
140 for i in `seq 10`; do
141 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
144 run_test 4a "|x| 10 open(O_CREAT)s"
150 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
152 run_test 4b "|x| rm 10 files"
154 # The idea is to get past the first block of precreated files on both
155 # osts, and then replay.
158 for i in `seq 220`; do
159 echo "tag-$i" > $DIR/$tfile-$i
162 for i in `seq 220`; do
163 grep -q "tag-$i" $DIR/$tfile-$i || error "$tfile-$i"
167 # waiting for commitment of removal
169 run_test 5 "|x| 220 open(O_CREAT)"
172 test_6a() { # was test_6
175 mcreate $DIR/$tdir/$tfile
177 $CHECKSTAT -t dir $DIR/$tdir || return 1
178 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
180 # waiting for log process thread
182 run_test 6a "mkdir + contained create"
189 $CHECKSTAT -t dir $DIR/$tdir && return 1 || true
191 run_test 6b "|X| rmdir"
196 mcreate $DIR/$tdir/$tfile
198 $CHECKSTAT -t dir $DIR/$tdir || return 1
199 $CHECKSTAT -t file $DIR/$tdir/$tfile || return 2
202 run_test 7 "mkdir |X| contained create"
206 multiop_bg_pause $DIR/$tfile mo_c || return 4
210 $CHECKSTAT -t file $DIR/$tfile || return 1
211 kill -USR1 $MULTIPID || return 2
212 wait $MULTIPID || return 3
215 run_test 8 "creat open |X| close"
220 local old_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
222 local new_inum=`ls -i $DIR/$tfile | awk '{print $1}'`
224 echo " old_inum == $old_inum, new_inum == $new_inum"
225 if [ $old_inum -eq $new_inum ] ;
227 echo " old_inum and new_inum match"
229 echo "!!!! old_inum and new_inum NOT match"
234 run_test 9 "|X| create (same inum/gen)"
239 mv $DIR/$tfile $DIR/$tfile-2
242 $CHECKSTAT $DIR/$tfile && return 1
243 $CHECKSTAT $DIR/$tfile-2 ||return 2
247 run_test 10 "create |X| rename unlink"
251 echo "old" > $DIR/$tfile
252 mv $DIR/$tfile $DIR/$tfile-2
254 echo "new" > $DIR/$tfile
256 grep old $DIR/$tfile-2
258 grep new $DIR/$tfile || return 1
259 grep old $DIR/$tfile-2 || return 2
261 run_test 11 "create open write rename |X| create-old-name read"
265 multiop_bg_pause $DIR/$tfile o_tSc || return 3
270 wait $pid || return 1
273 [ -e $DIR/$tfile ] && return 2
276 run_test 12 "open, unlink |X| close"
279 # 1777 - replay open after committed chmod that would make
280 # a regular open a failure
283 multiop_bg_pause $DIR/$tfile O_wc || return 3
286 $CHECKSTAT -p 0 $DIR/$tfile
290 wait $pid || return 1
292 $CHECKSTAT -s 1 -p 0 $DIR/$tfile || return 2
293 rm $DIR/$tfile || return 4
296 run_test 13 "open chmod 0 |x| write close"
299 multiop_bg_pause $DIR/$tfile O_tSc || return 4
303 kill -USR1 $pid || return 1
304 wait $pid || return 2
307 [ -e $DIR/$tfile ] && return 3
310 run_test 14 "open(O_CREAT), unlink |X| close"
313 multiop_bg_pause $DIR/$tfile O_tSc || return 5
317 touch $DIR/g11 || return 1
319 wait $pid || return 2
322 [ -e $DIR/$tfile ] && return 3
323 touch $DIR/h11 || return 4
326 run_test 15 "open(O_CREAT), unlink |X| touch new, close"
332 mcreate $DIR/$tfile-2
334 [ -e $DIR/$tfile ] && return 1
335 [ -e $DIR/$tfile-2 ] || return 2
336 munlink $DIR/$tfile-2 || return 3
338 run_test 16 "|X| open(O_CREAT), unlink, touch new, unlink new"
342 multiop_bg_pause $DIR/$tfile O_c || return 4
345 kill -USR1 $pid || return 1
346 wait $pid || return 2
347 $CHECKSTAT -t file $DIR/$tfile || return 3
350 run_test 17 "|X| open(O_CREAT), |replay| close"
354 multiop_bg_pause $DIR/$tfile O_tSc || return 8
357 touch $DIR/$tfile-2 || return 1
358 echo "pid: $pid will close"
360 wait $pid || return 2
363 [ -e $DIR/$tfile ] && return 3
364 [ -e $DIR/$tfile-2 ] || return 4
365 # this touch frequently fails
366 touch $DIR/$tfile-3 || return 5
367 munlink $DIR/$tfile-2 || return 6
368 munlink $DIR/$tfile-3 || return 7
371 run_test 18 "|X| open(O_CREAT), unlink, touch new, close, touch, unlink"
373 # bug 1855 (a simpler form of test_11 above)
377 echo "old" > $DIR/$tfile
378 mv $DIR/$tfile $DIR/$tfile-2
379 grep old $DIR/$tfile-2
381 grep old $DIR/$tfile-2 || return 2
383 run_test 19 "|X| mcreate, open, write, rename "
385 test_20a() { # was test_20
387 multiop_bg_pause $DIR/$tfile O_tSc || return 3
393 wait $pid || return 1
394 [ -e $DIR/$tfile ] && return 2
397 run_test 20a "|X| open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
399 test_20b() { # bug 10480
400 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
402 dd if=/dev/zero of=$DIR/$tfile bs=4k count=10000 &
404 while [ ! -e $DIR/$tfile ] ; do
405 sleep 0.060s # give dd a chance to start
408 lfs getstripe $DIR/$tfile || return 1
409 rm -f $DIR/$tfile || return 2 # make it an orphan
411 client_up || client_up || true # reconnect
413 fail mds # start orphan recovery
414 wait_recovery_complete mds || error "MDS recovery isn't done"
415 wait_mds_ost_sync || return 3
416 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
417 log "before $BEFOREUSED, after $AFTERUSED"
418 [ $AFTERUSED -gt $((BEFOREUSED + 20)) ] && \
419 error "after $AFTERUSED > before $BEFOREUSED" && return 5
422 run_test 20b "write, unlink, eviction, replay, (test mds_cleanup_orphans)"
424 test_20c() { # bug 10480
425 multiop $DIR/$tfile Ow_c &
427 # give multiop a chance to open
433 client_up || client_up || true # reconnect
436 wait $pid || return 1
437 [ -s $DIR/$tfile ] || error "File was truncated"
441 run_test 20c "check that client eviction does not affect file content"
445 multiop_bg_pause $DIR/$tfile O_tSc || return 5
448 touch $DIR/g11 || return 1
452 wait $pid || return 2
453 [ -e $DIR/$tfile ] && return 3
454 touch $DIR/h11 || return 4
457 run_test 21 "|X| open(O_CREAT), unlink touch new, replay, close (test mds_cleanup_orphans)"
460 multiop_bg_pause $DIR/$tfile O_tSc || return 3
468 wait $pid || return 1
469 [ -e $DIR/$tfile ] && return 2
472 run_test 22 "open(O_CREAT), |X| unlink, replay, close (test mds_cleanup_orphans)"
475 multiop_bg_pause $DIR/$tfile O_tSc || return 5
480 touch $DIR/g11 || return 1
484 wait $pid || return 2
485 [ -e $DIR/$tfile ] && return 3
486 touch $DIR/h11 || return 4
489 run_test 23 "open(O_CREAT), |X| unlink touch new, replay, close (test mds_cleanup_orphans)"
492 multiop_bg_pause $DIR/$tfile O_tSc || return 3
499 wait $pid || return 1
500 [ -e $DIR/$tfile ] && return 2
503 run_test 24 "open(O_CREAT), replay, unlink, close (test mds_cleanup_orphans)"
506 multiop_bg_pause $DIR/$tfile O_tSc || return 3
513 wait $pid || return 1
514 [ -e $DIR/$tfile ] && return 2
517 run_test 25 "open(O_CREAT), unlink, replay, close (test mds_cleanup_orphans)"
521 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
523 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
528 wait $pid2 || return 1
532 wait $pid1 || return 2
533 [ -e $DIR/$tfile-1 ] && return 3
534 [ -e $DIR/$tfile-2 ] && return 4
537 run_test 26 "|X| open(O_CREAT), unlink two, close one, replay, close one (test mds_cleanup_orphans)"
541 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
543 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
550 wait $pid1 || return 1
552 wait $pid2 || return 2
553 [ -e $DIR/$tfile-1 ] && return 3
554 [ -e $DIR/$tfile-2 ] && return 4
557 run_test 27 "|X| open(O_CREAT), unlink two, replay, close two (test mds_cleanup_orphans)"
560 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
562 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
568 wait $pid2 || return 1
572 wait $pid1 || return 2
573 [ -e $DIR/$tfile-1 ] && return 3
574 [ -e $DIR/$tfile-2 ] && return 4
577 run_test 28 "open(O_CREAT), |X| unlink two, close one, replay, close one (test mds_cleanup_orphans)"
580 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
582 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
590 wait $pid1 || return 1
592 wait $pid2 || return 2
593 [ -e $DIR/$tfile-1 ] && return 3
594 [ -e $DIR/$tfile-2 ] && return 4
597 run_test 29 "open(O_CREAT), |X| unlink two, replay, close two (test mds_cleanup_orphans)"
600 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
602 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
610 wait $pid1 || return 1
612 wait $pid2 || return 2
613 [ -e $DIR/$tfile-1 ] && return 3
614 [ -e $DIR/$tfile-2 ] && return 4
617 run_test 30 "open(O_CREAT) two, unlink two, replay, close two (test mds_cleanup_orphans)"
620 multiop_bg_pause $DIR/$tfile-1 O_tSc || return 5
622 multiop_bg_pause $DIR/$tfile-2 O_tSc || return 6
630 wait $pid1 || return 1
632 wait $pid2 || return 2
633 [ -e $DIR/$tfile-1 ] && return 3
634 [ -e $DIR/$tfile-2 ] && return 4
637 run_test 31 "open(O_CREAT) two, unlink one, |X| unlink one, close two (test mds_cleanup_orphans)"
639 # tests for bug 2104; completion without crashing is success. The close is
640 # stale, but we always return 0 for close, so the app never sees it.
642 multiop_bg_pause $DIR/$tfile O_c || return 2
644 multiop_bg_pause $DIR/$tfile O_c || return 3
647 client_up || client_up || return 1
650 wait $pid1 || return 4
651 wait $pid2 || return 5
654 run_test 32 "close() notices client eviction; close() after client eviction"
656 # Abort recovery before client complete
657 test_33a() { # was test_33
659 createmany -o $DIR/$tfile-%d 100
661 # this file should be gone, because the replay was aborted
662 $CHECKSTAT -t file $DIR/$tfile-* && return 3
663 unlinkmany $DIR/$tfile-%d 0 100
666 run_test 33a "abort recovery before client does replay"
669 multiop_bg_pause $DIR/$tfile O_c || return 2
676 wait $pid || return 3
677 [ -e $DIR/$tfile ] && return 1
681 run_test 34 "abort recovery before client does replay (test mds_cleanup_orphans)"
683 # bug 2278 - generate one orphan on OST, then destroy it during llog recovery
687 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
688 do_facet mds "lctl set_param fail_loc=0x80000119"
693 # give a chance to remove from MDS
695 $CHECKSTAT -t file $DIR/$tfile && return 1 || true
697 run_test 35 "test recovery from llog for unlink op"
699 # b=2432 resent cancel after replay uses wrong cookie,
700 # so don't resend cancels
704 checkstat $DIR/$tfile
707 if dmesg | grep "unknown lock cookie"; then
708 echo "cancel after replay failed"
712 run_test 36 "don't resend cancel"
715 # directory orphans can't be unlinked from PENDING directory
717 rmdir $DIR/$tfile 2>/dev/null
718 multiop_bg_pause $DIR/$tfile dD_c || return 2
723 # clear the dmesg buffer so we only see errors from this recovery
727 dmesg | grep "mds_unlink_orphan.*error .* unlinking orphan" && return 1
728 wait $pid || return 3
732 run_test 37 "abort recovery before client does replay (test mds_cleanup_orphans for directories)"
735 createmany -o $DIR/$tfile-%d 800
736 unlinkmany $DIR/$tfile-%d 0 400
739 unlinkmany $DIR/$tfile-%d 400 400
741 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
743 run_test 38 "test recovery from unlink llog (test llog_gen_rec) "
745 test_39() { # bug 4176
746 createmany -o $DIR/$tfile-%d 800
748 unlinkmany $DIR/$tfile-%d 0 400
750 unlinkmany $DIR/$tfile-%d 400 400
753 $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
755 run_test 39 "test recovery from unlink llog (test llog_gen_rec) "
758 lctl get_param -n osc.*.stats | awk -vwrites=0 '/ost_write/ { writes += $2 } END { print writes; }'
763 $LCTL mark multiop $MOUNT/$tfile OS_c
764 multiop $MOUNT/$tfile OS_c &
766 writeme -s $MOUNT/${tfile}-2 &
770 #define OBD_FAIL_MDS_CONNECT_NET 0x117
771 do_facet mds "lctl set_param fail_loc=0x80000117"
773 stat1=`count_ost_writes`
775 stat2=`count_ost_writes`
776 echo "$stat1, $stat2"
777 if [ $stat1 -lt $stat2 ]; then
778 echo "writes continuing during recovery"
781 echo "writes not continuing during recovery, bug 2477"
784 echo "waiting for writeme $WRITE_PID"
788 echo "waiting for multiop $PID"
789 wait $PID || return 2
790 do_facet client munlink $MOUNT/$tfile || return 3
791 do_facet client munlink $MOUNT/${tfile}-2 || return 3
794 run_test 40 "cause recovery in ptlrpc, ensure IO continues"
797 # make sure that a read to one osc doesn't try to double-unlock its page just
798 # because another osc is invalid. trigger_group_io used to mistakenly return
799 # an error if any oscs were invalid even after having successfully put rpcs
800 # on valid oscs. This was fatal if the caller was ll_readpage who unlocked
801 # the page, guarnateeing that the unlock from the RPC completion would
802 # assert on trying to unlock the unlocked page.
804 [ $OSTCOUNT -lt 2 ] && \
805 skip_env "skipping test 41: we don't have a second OST to test with" && \
808 local f=$MOUNT/$tfile
809 # make sure the start of the file is ost1
810 lfs setstripe $f -s $((128 * 1024)) -i 0
811 do_facet client dd if=/dev/zero of=$f bs=4k count=1 || return 3
813 # fail ost2 and read from ost1
814 local osc2dev=`lctl get_param -n devices | grep ${ost2_svc}-osc- | awk '{print $1}'`
815 [ "$osc2dev" ] || return 4
816 $LCTL --device $osc2dev deactivate || return 1
817 do_facet client dd if=$f of=/dev/null bs=4k count=1 || return 3
818 $LCTL --device $osc2dev activate || return 2
821 run_test 41 "read from a valid osc while other oscs are invalid"
823 # test MDS recovery after ost failure
825 blocks=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
826 createmany -o $DIR/$tfile-%d 800
828 unlinkmany $DIR/$tfile-%d 0 400
830 lctl set_param debug=-1
833 # osc is evicted, fs is smaller (but only with failout OSTs (bug 7287)
834 #blocks_after=`df -P $MOUNT | tail -n 1 | awk '{ print $2 }'`
835 #[ $blocks_after -lt $blocks ] || return 1
836 echo wait for MDS to timeout and recover
837 sleep $((TIMEOUT * 2))
839 unlinkmany $DIR/$tfile-%d 400 400
840 $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
842 run_test 42 "recovery after ost failure"
844 # timeout in MDS/OST recovery RPC will LBUG MDS
845 test_43() { # bug 2530
846 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
850 # OBD_FAIL_OST_CREATE_NET 0x204
851 do_facet ost1 "lctl set_param fail_loc=0x80000204"
854 do_facet ost1 "lctl set_param fail_loc=0"
858 run_test 43 "mds osc import failure during recovery; don't LBUG"
860 test_44a() { # was test_44
863 mdcdev=`lctl get_param -n devices | awk '/-mdc-/ {print $1}'`
864 [ "$mdcdev" ] || return 2
865 [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
867 # adaptive timeouts slow this way down
868 if at_is_enabled; then
869 at_max_saved=$(at_max_get mds)
873 local dev=$(get_mds_mdt_device_proc_path)
874 for i in `seq 1 10`; do
875 echo "$i of 10 ($(date +%s))"
876 do_facet mds "lctl get_param -n mdt.*.$dev.timeouts | grep service"
877 #define OBD_FAIL_TGT_CONN_RACE 0x701
878 do_facet mds "lctl set_param fail_loc=0x80000701"
879 # lctl below may fail, it is valid case
880 $LCTL --device $mdcdev recover
884 do_facet mds "lctl set_param fail_loc=0"
885 [ $at_max_saved -ne 0 ] && at_max_set $at_max_saved mds
888 run_test 44a "race in target handle connect"
891 local mdcdev=`lctl get_param -n devices | awk '/-mdc-/ {print $1}'`
892 [ "$mdcdev" ] || return 2
893 [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
895 local dev=$(get_mds_mdt_device_proc_path)
896 for i in `seq 1 10`; do
897 echo "$i of 10 ($(date +%s))"
898 do_facet mds "lctl get_param -n mdt.*.$dev.timeouts | grep service"
899 #define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
900 do_facet mds "lctl set_param fail_loc=0x80000704"
901 # lctl below may fail, it is valid case
902 $LCTL --device $mdcdev recover
905 do_facet mds "lctl set_param fail_loc=0"
908 run_test 44b "race in target handle connect"
910 # Handle failed close
912 local mdcdev=`lctl get_param -n devices | awk '/-mdc-/ {print $1}'`
913 [ "$mdcdev" ] || return 2
914 [ $(echo $mdcdev | wc -w) -eq 1 ] || { echo $mdcdev=$mdcdev && return 3; }
916 $LCTL --device $mdcdev recover || return 6
918 multiop_bg_pause $DIR/$tfile O_c || return 1
921 # This will cause the CLOSE to fail before even
922 # allocating a reply buffer
923 $LCTL --device $mdcdev deactivate || return 4
927 wait $pid || return 1
929 $LCTL --device $mdcdev activate || return 5
932 $CHECKSTAT -t file $DIR/$tfile || return 2
935 run_test 45 "Handle failed close"
939 drop_reply "touch $DIR/$tfile"
941 # ironically, the previous test, 45, will cause a real forced close,
942 # so just look for one for this test
943 dmesg | grep -i "force closing client file handle for $tfile" && return 1
946 run_test 46 "Don't leak file handle after open resend (3325)"
948 test_47() { # bug 2824
949 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
951 # create some files to make sure precreate has been done on all
952 # OSTs. (just in case this test is run independently)
953 createmany -o $DIR/$tfile 20 || return 1
955 # OBD_FAIL_OST_CREATE_NET 0x204
957 do_facet ost1 "lctl set_param fail_loc=0x80000204"
958 client_up || return 2
960 # let the MDS discover the OST failure, attempt to recover, fail
962 sleep $((3 * TIMEOUT))
964 # Without 2824, this createmany would hang
965 createmany -o $DIR/$tfile 20 || return 3
966 unlinkmany $DIR/$tfile 20 || return 4
968 do_facet ost1 "lctl set_param fail_loc=0"
971 run_test 47 "MDS->OSC failure during precreate cleanup (2824)"
974 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
975 [ "$OSTCOUNT" -lt "2" ] && skip_env "$OSTCOUNT < 2 OSTs -- skipping" && return
978 createmany -o $DIR/$tfile 20 || return 1
980 #define OBD_FAIL_OST_EROFS 0x216
981 do_facet ost1 "lctl set_param fail_loc=0x80000216"
982 client_up || return 2
984 createmany -o $DIR/$tfile 20 20 || return 2
985 unlinkmany $DIR/$tfile 40 || return 3
988 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
991 multiop $DIR/$tfile O_c &
996 #define OBD_FAIL_MDS_ALL_REPLY_NET 0x122
997 do_facet mds "lctl set_param fail_loc=0x80000122"
999 do_facet mds "lctl set_param fail_loc=0"
1001 replay_barrier_nodf mds
1004 wait $pid || return 1
1005 $CHECKSTAT -t file $DIR/$tfile && return 2
1008 run_test 49a "mds fail after close reply is dropped: open|create "
1012 multiop $DIR/$tfile o_c &
1017 #define OBD_FAIL_MDS_ALL_REPLY_NET 0x122
1018 do_facet mds "lctl set_param fail_loc=0x80000122"
1020 do_facet mds "lctl set_param fail_loc=0"
1022 replay_barrier_nodf mds
1025 wait $pid || return 1
1026 $CHECKSTAT -t file $DIR/$tfile && return 2
1029 run_test 49c "mds fail after close reply is dropped: open"
1032 local oscdev=`do_facet mds lctl get_param -n devices | grep ${ost1_svc}-osc | awk '{print $1}' | head -1`
1033 [ "$oscdev" ] || return 1
1034 do_facet mds $LCTL --device $oscdev recover || return 2
1035 do_facet mds $LCTL --device $oscdev recover || return 3
1036 # give the mds_lov_sync threads a chance to run
1039 run_test 50 "Double OSC recovery, don't LASSERT (3812)"
1041 # b3764 timed out lock replay
1044 cancel_lru_locks mdc
1046 multiop $DIR/$tfile s || return 1
1048 #define OBD_FAIL_LDLM_REPLY 0x30c
1049 do_facet mds "lctl set_param fail_loc=0x8000030c"
1050 fail mds || return 2
1051 do_facet mds "lctl set_param fail_loc=0x0"
1053 $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
1055 run_test 52 "time out lock replay (3764)"
1057 # bug 3462 - simultaneous MDC requests
1059 cancel_lru_locks mdc # lock cleanup
1060 mkdir -p $DIR/${tdir}-1
1061 mkdir -p $DIR/${tdir}-2
1062 multiop $DIR/${tdir}-1/f O_c &
1064 # give multiop a chance to open
1067 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1068 do_facet mds "lctl set_param fail_loc=0x80000115"
1069 kill -USR1 $close_pid
1070 cancel_lru_locks mdc # force the close
1071 do_facet mds "lctl set_param fail_loc=0"
1072 mcreate $DIR/${tdir}-2/f || return 1
1074 # close should still be here
1075 [ -d /proc/$close_pid ] || return 2
1076 replay_barrier_nodf mds
1078 wait $close_pid || return 3
1080 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1081 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1082 rm -rf $DIR/${tdir}-*
1084 run_test 53a "|X| close request while two MDC requests in flight"
1087 cancel_lru_locks mdc # lock cleanup
1088 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1090 mkdir -p $DIR/$tdir-1
1091 mkdir -p $DIR/$tdir-2
1093 multiop_bg_pause $DIR/$tdir-1/f O_c || return 6
1096 #define OBD_FAIL_MDS_REINT_NET 0x107
1097 do_facet mds "lctl set_param fail_loc=0x80000107"
1098 mcreate $DIR/${tdir}-2/f &
1102 do_facet mds "lctl set_param fail_loc=0"
1103 kill -USR1 $close_pid
1104 cancel_lru_locks mdc # force the close
1105 wait $close_pid || return 1
1106 # open should still be here
1107 [ -d /proc/$open_pid ] || return 2
1109 replay_barrier_nodf mds
1111 wait $open_pid || return 3
1113 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1114 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1115 rm -rf $DIR/${tdir}-*
1117 run_test 53b "|X| open request while two MDC requests in flight"
1120 cancel_lru_locks mdc # lock cleanup
1121 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1123 mkdir -p $DIR/${tdir}-1
1124 mkdir -p $DIR/${tdir}-2
1125 multiop $DIR/${tdir}-1/f O_c &
1128 #define OBD_FAIL_MDS_REINT_NET 0x107
1129 do_facet mds "lctl set_param fail_loc=0x80000107"
1130 mcreate $DIR/${tdir}-2/f &
1134 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1135 do_facet mds "lctl set_param fail_loc=0x80000115"
1136 kill -USR1 $close_pid
1137 cancel_lru_locks mdc # force the close
1139 replay_barrier_nodf mds
1141 wait $open_pid || return 1
1143 # close should be gone
1144 [ -d /proc/$close_pid ] && return 2
1145 do_facet mds "lctl set_param fail_loc=0"
1147 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1148 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1149 rm -rf $DIR/${tdir}-*
1151 run_test 53c "|X| open request and close request while two MDC requests in flight"
1154 cancel_lru_locks mdc # lock cleanup
1155 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1157 mkdir -p $DIR/${tdir}-1
1158 mkdir -p $DIR/${tdir}-2
1159 multiop $DIR/${tdir}-1/f O_c &
1161 # give multiop a chance to open
1164 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1165 do_facet mds "lctl set_param fail_loc=0x8000013b"
1166 kill -USR1 $close_pid
1167 cancel_lru_locks mdc # force the close
1168 do_facet mds "lctl set_param fail_loc=0"
1169 mcreate $DIR/${tdir}-2/f || return 1
1171 # close should still be here
1172 [ -d /proc/$close_pid ] || return 2
1173 replay_barrier_nodf mds
1175 wait $close_pid || return 3
1177 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1178 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1179 rm -rf $DIR/${tdir}-*
1181 run_test 53d "|X| close reply while two MDC requests in flight"
1184 cancel_lru_locks mdc # lock cleanup
1185 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1187 mkdir -p $DIR/$tdir-1
1188 mkdir -p $DIR/$tdir-2
1189 multiop $DIR/$tdir-1/f O_c &
1192 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1193 do_facet mds "lctl set_param fail_loc=0x80000119"
1194 mcreate $DIR/${tdir}-2/f &
1198 do_facet mds "lctl set_param fail_loc=0"
1199 kill -USR1 $close_pid
1200 cancel_lru_locks mdc # force the close
1201 wait $close_pid || return 1
1202 # open should still be here
1203 [ -d /proc/$open_pid ] || return 2
1205 replay_barrier_nodf mds
1207 wait $open_pid || return 3
1209 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 4
1210 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 5
1211 rm -rf $DIR/${tdir}-*
1213 run_test 53e "|X| open reply while two MDC requests in flight"
1216 cancel_lru_locks mdc # lock cleanup
1217 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1219 mkdir -p $DIR/${tdir}-1
1220 mkdir -p $DIR/${tdir}-2
1221 multiop $DIR/${tdir}-1/f O_c &
1224 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1225 do_facet mds "lctl set_param fail_loc=0x80000119"
1226 mcreate $DIR/${tdir}-2/f &
1230 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1231 do_facet mds "lctl set_param fail_loc=0x8000013b"
1232 kill -USR1 $close_pid
1233 cancel_lru_locks mdc
1235 replay_barrier_nodf mds
1237 wait $open_pid || return 1
1239 #close should be gone
1240 [ -d /proc/$close_pid ] && return 2
1241 do_facet mds "lctl set_param fail_loc=0"
1243 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1244 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1245 rm -rf $DIR/${tdir}-*
1247 run_test 53f "|X| open reply and close reply while two MDC requests in flight"
1250 cancel_lru_locks mdc # lock cleanup
1251 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1253 mkdir -p $DIR/${tdir}-1
1254 mkdir -p $DIR/${tdir}-2
1255 multiop $DIR/${tdir}-1/f O_c &
1258 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
1259 do_facet mds "lctl set_param fail_loc=0x80000119"
1260 mcreate $DIR/${tdir}-2/f &
1264 #define OBD_FAIL_MDS_CLOSE_NET 0x115
1265 do_facet mds "lctl set_param fail_loc=0x80000115"
1266 kill -USR1 $close_pid
1267 cancel_lru_locks mdc # force the close
1269 do_facet mds "lctl set_param fail_loc=0"
1270 replay_barrier_nodf mds
1272 wait $open_pid || return 1
1274 # close should be gone
1275 [ -d /proc/$close_pid ] && return 2
1277 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1278 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1279 rm -rf $DIR/${tdir}-*
1281 run_test 53g "|X| drop open reply and close request while close and open are both in flight"
1284 cancel_lru_locks mdc # lock cleanup
1285 rm -rf $DIR/${tdir}-1 $DIR/${tdir}-2
1287 mkdir -p $DIR/${tdir}-1
1288 mkdir -p $DIR/${tdir}-2
1289 multiop $DIR/${tdir}-1/f O_c &
1292 #define OBD_FAIL_MDS_REINT_NET 0x107
1293 do_facet mds "lctl set_param fail_loc=0x80000107"
1294 mcreate $DIR/${tdir}-2/f &
1298 #define OBD_FAIL_MDS_CLOSE_NET_REP 0x13b
1299 do_facet mds "lctl set_param fail_loc=0x8000013b"
1300 kill -USR1 $close_pid
1301 cancel_lru_locks mdc # force the close
1304 replay_barrier_nodf mds
1306 wait $open_pid || return 1
1308 # close should be gone
1309 [ -d /proc/$close_pid ] && return 2
1310 do_facet mds "lctl set_param fail_loc=0"
1312 $CHECKSTAT -t file $DIR/${tdir}-1/f || return 3
1313 $CHECKSTAT -t file $DIR/${tdir}-2/f || return 4
1314 rm -rf $DIR/${tdir}-*
1316 run_test 53h "|X| open request and close reply while two MDC requests in flight"
1318 #b3761 ASSERTION(hash != 0) failed
1320 # OBD_FAIL_MDS_OPEN_CREATE | OBD_FAIL_ONCE
1321 do_facet mds "lctl set_param fail_loc=0x8000012b"
1323 # give touch a chance to run
1325 do_facet mds "lctl set_param fail_loc=0x0"
1329 run_test 55 "let MDS_CHECK_RESENT return the original return code instead of 0"
1331 #b3440 ASSERTION(rec->ur_fid2->id) failed
1333 ln -s foo $DIR/$tfile
1335 #drop_reply "cat $DIR/$tfile"
1339 run_test 56 "don't replay a symlink open request (3440)"
1341 #recovery one mds-ost setattr from llog
1343 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1344 do_facet mds "lctl set_param fail_loc=0x8000012c"
1349 $CHECKSTAT -t file $DIR/$tfile || return 1
1350 do_facet mds "lctl set_param fail_loc=0x0"
1353 run_test 57 "test recovery from llog for setattr op"
1355 #recovery many mds-ost setattr from llog
1358 #define OBD_FAIL_MDS_OST_SETATTR 0x12c
1359 do_facet mds "lctl set_param fail_loc=0x8000012c"
1360 createmany -o $DIR/$tdir/$tfile-%d 2500
1364 $CHECKSTAT -t file $DIR/$tdir/$tfile-* >/dev/null || return 1
1365 do_facet mds "lctl set_param fail_loc=0x0"
1366 unlinkmany $DIR/$tdir/$tfile-%d 2500
1369 run_test 58a "test recovery from llog for setattr op (test llog_gen_rec)"
1372 mount_client $MOUNT2
1374 touch $DIR/$tdir/$tfile
1376 setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile
1378 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1379 [ x$VAL = x"bar" ] || return 1
1380 rm -f $DIR/$tdir/$tfile
1382 zconf_umount `hostname` $MOUNT2
1384 run_test 58b "test replay of setxattr op"
1386 test_58c() { # bug 16570
1387 mount_client $MOUNT2
1389 touch $DIR/$tdir/$tfile
1390 drop_request "setfattr -n trusted.foo -v bar $DIR/$tdir/$tfile" || \
1392 VAL=`getfattr --absolute-names --only-value -n trusted.foo $MOUNT2/$tdir/$tfile`
1393 [ x$VAL = x"bar" ] || return 2
1394 drop_reint_reply "setfattr -n trusted.foo1 -v bar1 $DIR/$tdir/$tfile" || \
1396 VAL=`getfattr --absolute-names --only-value -n trusted.foo1 $MOUNT2/$tdir/$tfile`
1397 [ x$VAL = x"bar1" ] || return 4
1398 rm -f $DIR/$tdir/$tfile
1400 zconf_umount `hostname` $MOUNT2
1402 run_test 58c "resend/reconstruct setxattr op"
1404 # log_commit_thread vs filter_destroy race used to lead to import use after free
1407 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1410 createmany -o $DIR/$tdir/$tfile-%d 200
1412 unlinkmany $DIR/$tdir/$tfile-%d 200
1413 #define OBD_FAIL_PTLRPC_DELAY_RECOV 0x507
1414 do_facet ost1 "lctl set_param fail_loc=0x507"
1417 do_facet ost1 "lctl set_param fail_loc=0x0"
1421 run_test 59 "test log_commit_thread vs filter_destroy race"
1423 # race between add unlink llog vs cat log init in post_recovery (only for b1_6)
1424 # bug 12086: should no oops and No ctxt error for this test
1427 createmany -o $DIR/$tdir/$tfile-%d 200
1429 unlinkmany $DIR/$tdir/$tfile-%d 0 100
1431 unlinkmany $DIR/$tdir/$tfile-%d 100 100
1432 local no_ctxt=`dmesg | grep "No ctxt"`
1433 [ -z "$no_ctxt" ] || error "ctxt is not initialized in recovery"
1435 run_test 60 "test llog post recovery init vs llog unlink"
1437 #test race llog recovery thread vs llog cleanup
1439 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1442 createmany -o $DIR/$tdir/$tfile-%d 800
1444 # OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
1445 unlinkmany $DIR/$tdir/$tfile-%d 800
1446 do_facet ost "lctl set_param fail_loc=0x80000221"
1451 do_facet ost "lctl set_param fail_loc=0x0"
1452 $CHECKSTAT -t file $DIR/$tdir/$tfile-* && return 1
1455 run_test 61a "test race llog recovery vs llog cleanup"
1457 #test race mds llog sync vs llog cleanup
1459 # OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x13a
1460 do_facet mds "lctl set_param fail_loc=0x8000013a"
1464 do_facet client dd if=/dev/zero of=$DIR/$tfile bs=4k count=1 || return 1
1466 run_test 61b "test race mds llog sync vs llog cleanup"
1468 #test race cancel cookie cb vs llog cleanup
1470 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1472 # OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
1474 do_facet ost "lctl set_param fail_loc=0x80000222"
1479 run_test 61c "test race mds llog sync vs llog cleanup"
1481 test_61d() { # bug 16002
1482 #define OBD_FAIL_OBD_LLOG_SETUP 0x605
1484 do_facet mgs "lctl set_param fail_loc=0x80000605"
1485 start mgs $MGSDEV $mgs_MOUNT_OPTS && error "mgs start should have failed"
1486 do_facet mgs "lctl set_param fail_loc=0"
1487 start mgs $MGSDEV $mgs_MOUNT_OPTS || error "cannot restart mgs"
1489 run_test 61d "error in llog_setup should cleanup the llog context correctly"
1491 test_62() { # Bug 15756 - don't mis-drop resent replay
1494 createmany -o $DIR/$tdir/$tfile- 25
1495 #define OBD_FAIL_TGT_REPLAY_DROP 0x707
1496 do_facet mds "lctl set_param fail_loc=0x80000707"
1498 do_facet mds "lctl set_param fail_loc=0"
1499 unlinkmany $DIR/$tdir/$tfile- 25 || return 2
1502 run_test 62 "don't mis-drop resent replay"
1504 #Adaptive Timeouts (bug 3055)
1512 echo "Cleaning up AT ..."
1513 if [ -n "$ATOLDBASE" ]; then
1514 local at_history=$($LCTL get_param -n at_history)
1515 do_facet mds "lctl set_param at_history=$at_history" || true
1516 do_facet ost1 "lctl set_param at_history=$at_history" || true
1519 if [ $AT_MAX_SET -ne 0 ]; then
1520 for facet in mds client ost; do
1521 var=AT_MAX_SAVE_${facet}
1522 echo restore AT on $facet to saved value ${!var}
1523 at_max_set ${!var} $facet
1524 at_new=$(at_max_get $facet)
1525 echo Restored AT value on $facet $at_new
1526 [ $at_new -eq ${!var} ] || \
1527 error "$facet : AT value was not restored SAVED ${!var} NEW $at_new"
1534 local at_max_new=600
1536 # Save at_max original values
1538 if [ $AT_MAX_SET -eq 0 ]; then
1539 # Suppose that all osts have the same at_max
1540 for facet in mds client ost; do
1541 eval AT_MAX_SAVE_${facet}=$(at_max_get $facet)
1545 for facet in mds client ost; do
1546 at_max=$(at_max_get $facet)
1547 if [ $at_max -ne $at_max_new ]; then
1548 echo "AT value on $facet is $at_max, set it by force temporarily to $at_max_new"
1549 at_max_set $at_max_new $facet
1554 if [ -z "$ATOLDBASE" ]; then
1555 ATOLDBASE=$(do_facet mds "lctl get_param -n at_history")
1556 # speed up the timebase so we can check decreasing AT
1557 do_facet mds "lctl set_param at_history=8" || true
1558 do_facet ost1 "lctl set_param at_history=8" || true
1560 # sleep for a while to cool down, should be > 8s and also allow
1561 # at least one ping to be sent. simply use TIMEOUT to be safe.
1566 test_65a() #bug 3055
1568 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1570 at_start || return 0
1571 $LCTL dk > /dev/null
1573 lctl set_param debug="+other"
1574 # Slow down a request to the current service time, this is critical
1575 # because previous tests may have caused this value to increase.
1576 REQ_DELAY=`lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts |
1577 awk '/portal 12/ {print $5}'`
1578 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1580 do_facet mds lctl set_param fail_val=$((${REQ_DELAY} * 1000))
1581 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1582 do_facet mds lctl set_param fail_loc=0x8000050a
1583 createmany -o $DIR/$tfile 10 > /dev/null
1584 unlinkmany $DIR/$tfile 10 > /dev/null
1585 # check for log message
1586 $LCTL dk | grep "Early reply #" || error "No early reply"
1587 # client should show REQ_DELAY estimates
1588 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1590 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal
1592 run_test 65a "AT: verify early replies"
1594 test_65b() #bug 3055
1596 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1598 at_start || return 0
1601 lctl set_param debug="+other"
1602 $LCTL dk > /dev/null
1603 # Slow down a request to the current service time, this is critical
1604 # because previous tests may have caused this value to increase.
1605 lfs setstripe $DIR/$tfile --index=0 --count=1
1606 multiop $DIR/$tfile Ow1yc
1607 REQ_DELAY=`lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts |
1608 awk '/portal 6/ {print $5}'`
1609 REQ_DELAY=$((${REQ_DELAY} + ${REQ_DELAY} / 4 + 5))
1611 do_facet ost1 lctl set_param fail_val=${REQ_DELAY}
1612 #define OBD_FAIL_OST_BRW_PAUSE_PACK 0x224
1613 do_facet ost1 lctl set_param fail_loc=0x224
1616 lfs setstripe $DIR/$tfile --index=0 --count=1
1617 # force some real bulk transfer
1618 multiop $DIR/$tfile oO_CREAT:O_RDWR:O_SYNC:w4096c
1620 do_facet ost1 lctl set_param fail_loc=0
1621 # check for log message
1622 $LCTL dk | grep "Early reply #" || error "No early reply"
1624 # client should show REQ_DELAY estimates
1625 lctl get_param -n osc.${FSNAME}-OST0000-osc-*.timeouts | grep portal
1627 run_test 65b "AT: verify early replies on packed reply / bulk"
1629 test_66a() #bug 3055
1631 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1633 at_start || return 0
1634 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1635 # adjust 5s at a time so no early reply is sent (within deadline)
1636 do_facet mds "lctl set_param fail_val=5000"
1637 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1638 do_facet mds "lctl set_param fail_loc=0x8000050a"
1639 createmany -o $DIR/$tfile 20 > /dev/null
1640 unlinkmany $DIR/$tfile 20 > /dev/null
1641 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1642 do_facet mds "lctl set_param fail_val=10000"
1643 do_facet mds "lctl set_param fail_loc=0x8000050a"
1644 createmany -o $DIR/$tfile 20 > /dev/null
1645 unlinkmany $DIR/$tfile 20 > /dev/null
1646 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep "portal 12"
1647 do_facet mds "lctl set_param fail_loc=0"
1649 createmany -o $DIR/$tfile 20 > /dev/null
1650 unlinkmany $DIR/$tfile 20 > /dev/null
1651 lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | grep portal | grep "portal 12"
1652 CUR=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $5}')
1653 WORST=$(lctl get_param -n mdc.${FSNAME}-MDT0000-mdc-*.timeouts | awk '/portal 12/ {print $7}')
1654 echo "Current MDT timeout $CUR, worst $WORST"
1655 [ $CUR -lt $WORST ] || error "Current $CUR should be less than worst $WORST"
1657 run_test 66a "AT: verify MDT service time adjusts with no early replies"
1659 test_66b() #bug 3055
1661 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1663 at_start || return 0
1664 ORIG=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1665 lctl set_param fail_val=$(($ORIG + 5))
1666 #define OBD_FAIL_PTLRPC_PAUSE_REP 0x50c
1667 lctl set_param fail_loc=0x50c
1668 ls $DIR/$tfile > /dev/null 2>&1
1669 lctl set_param fail_loc=0
1670 CUR=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $4}')
1671 WORST=$(lctl get_param -n mdc.${FSNAME}-*.timeouts | awk '/network/ {print $6}')
1672 echo "network timeout orig $ORIG, cur $CUR, worst $WORST"
1673 [ $WORST -gt $ORIG ] || error "Worst $WORST should be worse than orig $ORIG"
1675 run_test 66b "AT: verify net latency adjusts"
1677 test_67a() #bug 3055
1679 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1681 at_start || return 0
1682 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1683 # sleeping threads may drive values above this
1684 do_facet ost1 "lctl set_param fail_val=400"
1685 #define OBD_FAIL_PTLRPC_PAUSE_REQ 0x50a
1686 do_facet ost1 "lctl set_param fail_loc=0x50a"
1687 createmany -o $DIR/$tfile 20 > /dev/null
1688 unlinkmany $DIR/$tfile 20 > /dev/null
1689 do_facet ost1 "lctl set_param fail_loc=0"
1690 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1691 ATTEMPTS=$(($CONN2 - $CONN1))
1692 echo "$ATTEMPTS osc reconnect attemps on gradual slow"
1693 [ $ATTEMPTS -gt 0 ] && error_ignore 13721 "AT should have prevented reconnect"
1696 run_test 67a "AT: verify slow request processing doesn't induce reconnects"
1698 test_67b() #bug 3055
1700 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1702 at_start || return 0
1703 CONN1=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1705 # exhaust precreations on ost1
1706 local OST=$(lfs osts | grep ^0": " | awk '{print $2}' | sed -e 's/_UUID$//')
1707 local mdtosc=$(get_mdtosc_proc_path $OST)
1708 local last_id=$(do_facet mds lctl get_param -n osc.$mdtosc.prealloc_last_id)
1709 local next_id=$(do_facet mds lctl get_param -n osc.$mdtosc.prealloc_next_id)
1711 mkdir -p $DIR/$tdir/${OST}
1712 lfs setstripe $DIR/$tdir/${OST} -o 0 -c 1 || error "setstripe"
1713 echo "Creating to objid $last_id on ost $OST..."
1714 #define OBD_FAIL_OST_PAUSE_CREATE 0x223
1715 do_facet ost1 "lctl set_param fail_val=20000"
1716 do_facet ost1 "lctl set_param fail_loc=0x80000223"
1717 createmany -o $DIR/$tdir/${OST}/f $next_id $((last_id - next_id + 2))
1720 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1722 CONN2=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1723 ATTEMPTS=$(($CONN2 - $CONN1))
1724 echo "$ATTEMPTS osc reconnect attemps on instant slow"
1725 # do it again; should not timeout
1726 do_facet ost1 "lctl set_param fail_loc=0x80000223"
1727 cp /etc/profile $DIR/$tfile || error "cp failed"
1728 do_facet ost1 "lctl set_param fail_loc=0"
1730 do_facet ost1 "lctl get_param -n ost.OSS.ost_create.timeouts"
1731 CONN3=$(lctl get_param -n osc.*.stats | awk '/_connect/ {total+=$2} END {print total}')
1732 ATTEMPTS=$(($CONN3 - $CONN2))
1733 echo "$ATTEMPTS osc reconnect attemps on 2nd slow"
1734 [ $ATTEMPTS -gt 0 ] && error "AT should have prevented reconnect"
1737 run_test 67b "AT: verify instant slowdown doesn't induce reconnects"
1739 test_68 () #bug 13813
1741 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
1743 at_start || return 0
1744 local ldlm_enqueue_min=$(find /sys -name ldlm_enqueue_min)
1745 [ -z "$ldlm_enqueue_min" ] && skip "missing /sys/.../ldlm_enqueue_min" && return 0
1746 local ldlm_enqueue_min_r=$(do_facet ost1 "find /sys -name ldlm_enqueue_min")
1747 [ -z "$ldlm_enqueue_min_r" ] && skip "missing /sys/.../ldlm_enqueue_min in the ost1" && return 0
1748 local ENQ_MIN=$(cat $ldlm_enqueue_min)
1749 local ENQ_MIN_R=$(do_facet ost1 "cat $ldlm_enqueue_min_r")
1750 echo $TIMEOUT >> $ldlm_enqueue_min
1751 do_facet ost1 "echo $TIMEOUT >> $ldlm_enqueue_min_r"
1755 lfs setstripe $DIR/$tdir --index=0 --count=1
1756 #define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
1757 lctl set_param fail_val=$(($TIMEOUT - 1))
1758 lctl set_param fail_loc=0x80000312
1759 cp /etc/profile $DIR/$tdir/${tfile}_1 || error "1st cp failed $?"
1760 lctl set_param fail_val=$((TIMEOUT * 5 / 4))
1761 lctl set_param fail_loc=0x80000312
1762 cp /etc/profile $DIR/$tdir/${tfile}_2 || error "2nd cp failed $?"
1763 lctl set_param fail_loc=0
1765 echo $ENQ_MIN >> $ldlm_enqueue_min
1766 do_facet ost1 "echo $ENQ_MIN_R >> $ldlm_enqueue_min_r"
1770 run_test 68 "AT: verify slowing locks"
1773 # end of AT tests includes above lines
1775 # start multi-client tests
1777 [ -z "$CLIENTS" ] && \
1778 { skip "Need two or more clients." && return; }
1779 [ $CLIENTCOUNT -lt 2 ] && \
1780 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
1782 echo "mount clients $CLIENTS ..."
1783 zconf_mount_clients $CLIENTS $DIR
1785 local clients=${CLIENTS//,/ }
1786 echo "Write/read files on $DIR ; clients $CLIENTS ... "
1787 for CLIENT in $clients; do
1788 do_node $CLIENT dd bs=1M count=10 if=/dev/zero \
1789 of=$DIR/${tfile}_${CLIENT} 2>/dev/null || \
1790 error "dd failed on $CLIENT"
1793 local prev_client=$(echo $clients | sed 's/^.* \(.\+\)$/\1/')
1794 for C in ${CLIENTS//,/ }; do
1795 do_node $prev_client dd if=$DIR/${tfile}_${C} of=/dev/null 2>/dev/null || \
1796 error "dd if=$DIR/${tfile}_${C} failed on $prev_client"
1802 run_test 70a "check multi client t-f"
1804 check_dbench_load () {
1805 local clients=${1//,/ }
1808 for client in $clients; do
1809 if ! do_node $client "ps ax | grep -v grep | awk '{ print $6 }' | grep -q rundbench"; then
1810 error_noexit "rundbench load on $client failed!"
1817 kill_dbench_load () {
1818 local clients=${1:-$(hostname)}
1819 do_nodes $clients "killall dbench"
1823 local clients=${CLIENTS:-$HOSTNAME}
1825 zconf_mount_clients $clients $DIR
1828 [ "$SLOW" = "no" ] && duration=60
1829 local cmd="rundbench 1 -t $duration"
1831 do_nodesv $clients "set -x; MISSING_DBENCH_OK=$MISSING_DBENCH_OK \
1832 PATH=:$PATH:$LUSTRE/utils:$LUSTRE/tests/:$DBENCH_LIB \
1833 DBENCH_LIB=$DBENCH_LIB TESTSUITE=$TESTSUITE TESTNAME=$TESTNAME \
1836 log "Started rundbench load PID=$PID ..."
1840 START_TS=$(date +%s)
1841 CURRENT_TS=$START_TS
1842 while [ $ELAPSED -lt $duration ]; do
1843 if ! check_dbench_load $clients; then
1844 kill_dbench_load $clients
1849 sleep 1 # give clients a time to do operations
1850 # Increment the number of failovers
1851 NUM_FAILOVERS=$((NUM_FAILOVERS+1))
1852 log "$TESTNAME fail mds1 $NUM_FAILOVERS times"
1854 CURRENT_TS=$(date +%s)
1855 ELAPSED=$((CURRENT_TS - START_TS))
1858 wait $PID || error "rundbench load on $CLIENTS failed!"
1860 run_test 70b "mds recovery; $CLIENTCOUNT clients"
1861 # end multi-client tests
1863 # vbr export handling
1865 delayed_recovery_enabled || { skip "No delayed recovery support"; return 0; }
1867 UUID=$(lctl dl | awk '/mdc.*-mdc-/ { print $5 }')
1868 echo "Client UUID is $UUID"
1872 zconf_mount `hostname` $DIR || error "mount fails"
1873 client_up || error "post-failover df failed"
1874 do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|grep $UUID" || \
1875 error "no delayed exports"
1876 OLD_AGE=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_export_age")
1878 do_facet mds "lctl set_param mds.${mds_svc}.stale_export_age=$NEW_AGE"
1879 sleep $((NEW_AGE + 2))
1880 do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|grep \"$UUID.*EXPIRED\"" || \
1881 error "exports didn't expire"
1882 do_facet mds "lctl set_param mds.${mds_svc}.evict_client=$UUID"
1883 do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|grep $UUID" && \
1884 error "Export wasn't removed manually"
1885 do_facet mds "lctl set_param mds.${mds_svc}.stale_export_age=$OLD_AGE"
1888 run_test 71a "lost client export is kept"
1891 delayed_recovery_enabled || { skip "No delayed recovery support"; return 0; }
1894 create_fake_exports mds $FAKE_NUM
1895 NUM=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|wc -l")
1896 [ $NUM -eq 0 ] && error "no fake exports $NUM - $FAKE_NUM"
1897 OLD_AGE=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_export_age")
1899 do_facet mds "lctl set_param mds.${mds_svc}.stale_export_age=$NEW_AGE"
1900 sleep $((NEW_AGE + 2))
1901 EX_NUM=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|grep -c EXPIRED")
1902 [ "$EX_NUM" -eq "$NUM" ] || error "not all exports are expired $EX_NUM != $NUM"
1903 do_facet mds "lctl set_param mds.${mds_svc}.flush_stale_exports=1"
1904 do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|grep EXPIRED" && \
1905 error "Exports weren't flushed"
1906 do_facet mds "lctl set_param mds.${mds_svc}.stale_export_age=$OLD_AGE"
1909 run_test 71b "stale exports are expired, lctl flushes them"
1912 delayed_recovery_enabled || { skip "No delayed recovery support"; return 0; }
1915 create_fake_exports mds $FAKE_NUM
1916 NUM=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|wc -l")
1917 [ "$NUM" -eq "$FAKE_NUM" ] || error "no fake exports $NUM - $FAKE_NUM"
1918 OLD_AGE=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_export_age")
1920 do_facet mds "lctl set_param mds.${mds_svc}.stale_export_age=$NEW_AGE"
1921 sleep $((NEW_AGE + 2))
1922 EX_NUM=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|grep -c EXPIRED")
1923 [ "$EX_NUM" -eq "$NUM" ] || error "not all exports are expired $EX_NUM != $NUM"
1926 zconf_mount `hostname` $DIR || error "mount fails"
1928 NUM=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|wc -l")
1929 [ $NUM -eq 0 ] || error "$NUM fake exports are still exists"
1930 do_facet mds "lctl set_param mds.${mds_svc}.stale_export_age=$OLD_AGE"
1933 run_test 71c "stale exports are expired, new client connection flush them"
1936 delayed_recovery_enabled || { skip "No delayed recovery support"; return 0; }
1939 create_fake_exports mds $FAKE_NUM
1940 NUM=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|wc -l")
1941 [ "$NUM" -eq "$FAKE_NUM" ] || error "no fake exports $NUM - $FAKE_NUM"
1942 OLD_AGE=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_export_age")
1944 do_facet mds "lctl conf_param ${mds_svc}.mdt.stale_export_age=$NEW_AGE"
1945 sleep $((NEW_AGE + 2))
1946 EX_NUM=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|grep -c EXPIRED")
1947 [ "$EX_NUM" -eq "$NUM" ] || error "not all exports are expired $EX_NUM != $NUM"
1951 FAIL_AGE=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_export_age")
1952 [ $FAIL_AGE -eq $NEW_AGE ] || error "new age wasn't set after recovery"
1953 NUM=$(do_facet mds "lctl get_param -n mds.${mds_svc}.stale_exports|wc -l")
1954 [ $NUM -eq 0 ] || error "$NUM fake exports are still exists"
1955 do_facet mds "lctl conf_param ${mds_svc}.mdt.stale_export_age=$OLD_AGE"
1958 run_test 71d "expired exports, server init removes them, conf_param works"
1960 # end vbr exports tests
1962 test_72() { #bug 16711
1964 multiop_bg_pause $DIR/$tfile O_c || return 4
1966 #define OBD_FAIL_TGT_REPLAY_DELAY 0x709
1967 do_facet mds "lctl set_param fail_loc=0x80000709"
1969 kill -USR1 $pid || return 1
1970 wait $pid || return 2
1971 $CHECKSTAT -t file $DIR/$tfile || return 3
1973 run_test 72 "target_finish_recovery vs process_recovery_queue race"
1976 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1981 #define OBD_FAIL_LDLM_ENQUEUE 0x302
1982 do_facet mds "lctl set_param fail_loc=0x80000302"
1985 wait $pid || return 1
1986 [ -e $DIR/$tfile ] && return 2
1989 run_test 73a "open(O_CREAT), unlink, replay, reconnect before open replay , close"
1992 multiop_bg_pause $DIR/$tfile O_tSc || return 3
1997 #define OBD_FAIL_LDLM_REPLY 0x30c
1998 do_facet mds "lctl set_param fail_loc=0x8000030c"
2001 wait $pid || return 1
2002 [ -e $DIR/$tfile ] && return 2
2005 run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
2008 multiop_bg_pause $DIR/$tfile O_tSc || return 3
2013 #define OBD_FAIL_TGT_LAST_REPLAY 0x710
2014 do_facet mds "lctl set_param fail_loc=0x80000710"
2017 wait $pid || return 1
2018 [ -e $DIR/$tfile ] && return 2
2021 run_test 73c "open(O_CREAT), unlink, replay, reconnect at last_replay, close"
2024 do_facet ost1 "lctl get_param obdfilter.${ost1_svc}.sync_journal" ||
2025 { skip "sync journal is not implemeted" && return; }
2027 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
2030 lfs setstripe -i 0 -c 1 $DIR/$tfile
2031 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 || error "Cannot write"
2032 cksum=`md5sum $DIR/$tfile | awk '{print $1}'`
2033 cancel_lru_locks osc
2035 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
2036 cksum2=`md5sum $DIR/$tfile | awk '{print $1}'`
2037 if [ $cksum != $cksum2 ] ; then
2038 error "New checksum $cksum2 does not match original $cksum"
2041 run_test 80a "write replay"
2044 do_facet ost1 "lctl get_param obdfilter.${ost1_svc}.sync_journal" ||
2045 { skip "sync journal is not implemeted" && return; }
2047 do_facet ost1 "lctl set_param -n obdfilter.${ost1_svc}.sync_journal 0"
2050 lfs setstripe -i 0 -c 1 $DIR/$tfile
2051 dd if=/dev/urandom of=$DIR/$tfile bs=1024k count=8 || error "Cannot write"
2052 sleep 1 # Give it a chance to flush dirty data
2053 echo TESTTEST | dd of=$DIR/$tfile bs=1 count=8 seek=64
2054 cksum=`md5sum $DIR/$tfile | awk '{print $1}'`
2055 cancel_lru_locks osc
2057 dd if=$DIR/$tfile of=/dev/null bs=1024k count=8 || error "Cannot read"
2058 cksum2=`md5sum $DIR/$tfile | awk '{print $1}'`
2059 if [ $cksum != $cksum2 ] ; then
2060 error "New checksum $cksum2 does not match original $cksum"
2063 run_test 80b "write replay with changed data (checksum resend)"
2067 createmany -o $DIR/$tdir/$tfile- 10 || return 1
2068 #define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD 0x140
2069 do_facet mds "lctl set_param fail_loc=0x80000140"
2070 unlinkmany $DIR/$tdir/$tfile- 10 || return 2
2072 run_test 81a "fail log_add during unlink recovery"
2076 createmany -o $DIR/$tdir/$tfile- 10 || return 1
2078 unlinkmany $DIR/$tdir/$tfile- 10 || return 2
2079 #define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD 0x140
2080 do_facet mds "lctl set_param fail_loc=0x80000140"
2083 run_test 81b "fail log_add during unlink recovery"
2085 test_82() { #bug 18927
2086 multiop_bg_pause $DIR/$tfile O_c || return 1
2088 multiop_bg_pause $DIR/$tfile o:O_CREAT:O_RDONLY:_c || return 2
2094 wait $pid1 || return 3
2096 wait $pid2 || return 4
2097 [ -e $DIR/$tfile ] && return 5
2100 run_test 82 "second open|creat in replay with open orphan"
2102 test_83() { #bug 19224
2103 #define OBD_FAIL_MDS_SPLIT_OPEN 0x142
2104 do_facet mds "lctl set_param fail_loc=0x80000142"
2105 # open will sleep after first transaction
2109 # set barrier between open transactions
2110 replay_barrier_nodf mds
2111 createmany -o $DIR/$tfile- 10
2112 # open should finish now
2113 wait $PID || return 1
2115 rm $DIR/$tfile || return 2
2116 unlinkmany $DIR/$tfile- 10 || return 3
2119 run_test 83 "open replay with barrier between transactions"
2122 #define OBD_FAIL_OBD_FAIL_MDS_SPLIT_OPEN 0x142
2123 do_facet mds "lctl set_param fail_loc=0x80000142"
2124 createmany -o $DIR/$tfile- 1 &
2128 client_up || client_up || true # reconnect
2130 run_test 84 "stale open during export disconnect"
2132 test_85a() { #bug 16774
2133 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
2135 for i in `seq 100`; do
2136 echo "tag-$i" > $DIR/$tfile-$i
2137 grep -q "tag-$i" $DIR/$tfile-$i || error "f2-$i"
2140 lov_id=`lctl dl | grep "clilov"`
2141 addr=`echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}'`
2142 count=`lctl get_param -n ldlm.namespaces.*MDT0000*$addr.lock_unused_count`
2143 echo "before recovery: unused locks count = $count"
2147 count2=`lctl get_param -n ldlm.namespaces.*MDT0000*$addr.lock_unused_count`
2148 echo "after recovery: unused locks count = $count2"
2150 if [ $count2 -ge $count ]; then
2151 error "unused locks are not canceled"
2154 run_test 85a "check the cancellation of unused locks during recovery(IBITS)"
2156 test_85b() { #bug 16774
2157 lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
2159 lfs setstripe -o 0 -c 1 $DIR
2161 for i in `seq 100`; do
2162 dd if=/dev/urandom of=$DIR/$tfile-$i bs=4096 count=32 >/dev/null 2>&1
2165 cancel_lru_locks osc
2167 for i in `seq 100`; do
2168 dd if=$DIR/$tfile-$i of=/dev/null bs=4096 count=32 >/dev/null 2>&1
2171 lov_id=`lctl dl | grep "clilov"`
2172 addr=`echo $lov_id | awk '{print $4}' | awk -F '-' '{print $3}'`
2173 count=`lctl get_param -n ldlm.namespaces.*OST0000*$addr.lock_unused_count`
2174 echo "before recovery: unused locks count = $count"
2178 count2=`lctl get_param -n ldlm.namespaces.*OST0000*$addr.lock_unused_count`
2179 echo "after recovery: unused locks count = $count2"
2181 if [ $count2 -ge $count ]; then
2182 error "unused locks are not canceled"
2185 run_test 85b "check the cancellation of unused locks during recovery(EXTENT)"
2188 local dev=$(get_mds_mdt_device_proc_path)
2190 do_facet mds lctl set_param $dev.${FSNAME}-MDT*.exports.clear=0
2193 run_test 86 "umount server after clear nid_stats should not hit LBUG"
2195 equals_msg `basename $0`: test complete, cleaning up
2196 check_and_cleanup_lustre
2197 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true