6 ALWAYS_EXCEPT="15c $REPLAY_DUAL_EXCEPT"
9 SETSTRIPE=${SETSTRIPE:-"$LFS setstripe"}
10 GETSTRIPE=${GETSTRIPE:-"$LFS getstripe"}
12 PTLDEBUG=${PTLDEBUG:--1}
13 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
15 CLEANUP=${CLEANUP:-""}
16 MOUNT_2=${MOUNT_2:-"yes"}
17 . $LUSTRE/tests/test-framework.sh
19 if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
20 CONFIG_EXCEPTIONS="17"
21 echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
22 echo "Except the tests: $CONFIG_EXCEPTIONS"
23 ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
27 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
30 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
32 [ "$SLOW" = "no" ] && EXCEPT_SLOW="21b"
36 check_and_setup_lustre
37 MOUNTED=$(mounted_lustre_filesystems)
38 if ! $(echo $MOUNTED | grep -w -q $MOUNT2); then
39 zconf_mount $HOSTNAME $MOUNT2
44 rm -rf $DIR/[df][0-9]*
46 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
50 replay_barrier $SINGLEMDS
54 checkstat $MOUNT2/a || return 1
55 checkstat $MOUNT1/b || return 2
56 rm $MOUNT2/a $MOUNT1/b
57 checkstat $MOUNT1/a && return 3
58 checkstat $MOUNT2/b && return 4
62 run_test 1 "|X| simple create"
66 replay_barrier $SINGLEMDS
70 checkstat $MOUNT2/adir || return 1
72 checkstat $MOUNT2/adir && return 2
75 run_test 2 "|X| mkdir adir"
78 replay_barrier $SINGLEMDS
80 mkdir $MOUNT2/adir/bdir
83 checkstat $MOUNT2/adir || return 1
84 checkstat $MOUNT1/adir/bdir || return 2
85 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
86 checkstat $MOUNT1/adir && return 3
87 checkstat $MOUNT2/adir/bdir && return 4
90 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
94 replay_barrier $SINGLEMDS
95 mkdir $MOUNT1/adir && return 1
96 mkdir $MOUNT2/adir/bdir
99 checkstat $MOUNT2/adir || return 2
100 checkstat $MOUNT1/adir/bdir || return 3
102 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
103 checkstat $MOUNT1/adir && return 4
104 checkstat $MOUNT2/adir/bdir && return 5
107 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
111 # multiclient version of replay_single.sh/test_8
113 multiop_bg_pause $MOUNT2/a o_tSc || return 1
116 replay_barrier $SINGLEMDS
118 wait $pid || return 1
121 [ -e $MOUNT2/a ] && return 2
124 run_test 5 "open, unlink |X| close"
129 multiop_bg_pause $MOUNT2/a o_c || return 1
131 multiop_bg_pause $MOUNT1/a o_c || return 1
134 replay_barrier $SINGLEMDS
136 wait $pid1 || return 1
140 wait $pid2 || return 1
141 [ -e $MOUNT2/a ] && return 2
144 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
147 replay_barrier $SINGLEMDS
148 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
150 checkstat $MOUNT2/$tfile || return 2
151 rm $MOUNT1/$tfile || return 3
155 run_test 8 "replay of resent request"
158 replay_barrier $SINGLEMDS
159 mcreate $MOUNT1/$tfile-1
160 mcreate $MOUNT2/$tfile-2
161 # drop first reint reply
162 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
164 do_facet $SINGLEMDS lctl set_param fail_loc=0
166 rm $MOUNT1/$tfile-[1,2] || return 1
170 run_test 9 "resending a replayed create"
173 mcreate $MOUNT1/$tfile-1
174 replay_barrier $SINGLEMDS
175 munlink $MOUNT1/$tfile-1
176 mcreate $MOUNT2/$tfile-2
177 # drop first reint reply
178 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
180 do_facet $SINGLEMDS lctl set_param fail_loc=0
182 checkstat $MOUNT1/$tfile-1 && return 1
183 checkstat $MOUNT1/$tfile-2 || return 2
188 run_test 10 "resending a replayed unlink"
191 replay_barrier $SINGLEMDS
192 mcreate $MOUNT1/$tfile-1
193 mcreate $MOUNT2/$tfile-2
194 mcreate $MOUNT1/$tfile-3
195 mcreate $MOUNT2/$tfile-4
196 mcreate $MOUNT1/$tfile-5
197 # drop all reint replies for a while
198 do_facet $SINGLEMDS lctl set_param fail_loc=0x0119
199 # note that with this fail_loc set, facet_failover df will fail
200 facet_failover $SINGLEMDS
201 #sleep for while, let both clients reconnect and timeout
202 sleep $((TIMEOUT * 2))
203 do_facet $SINGLEMDS lctl set_param fail_loc=0
205 rm $MOUNT1/$tfile-[1-5] || return 1
209 run_test 11 "both clients timeout during replay"
212 replay_barrier $SINGLEMDS
214 multiop_bg_pause $DIR/$tfile mo_c || return 1
217 #define OBD_FAIL_LDLM_ENQUEUE 0x302
218 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000302
219 facet_failover $SINGLEMDS
220 do_facet $SINGLEMDS lctl set_param fail_loc=0
221 clients_up || return 1
224 kill -USR1 $MULTIPID || return 3
225 wait $MULTIPID || return 4
226 $CHECKSTAT -t file $DIR/$tfile || return 2
231 run_test 12 "open resend timeout"
234 multiop_bg_pause $DIR/$tfile mo_c || return 1
237 replay_barrier $SINGLEMDS
239 kill -USR1 $MULTIPID || return 3
240 wait $MULTIPID || return 4
243 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000115
244 facet_failover $SINGLEMDS
245 do_facet $SINGLEMDS lctl set_param fail_loc=0
246 clients_up || return 1
249 $CHECKSTAT -t file $DIR/$tfile || return 2
254 run_test 13 "close resend timeout"
257 replay_barrier $SINGLEMDS
258 createmany -o $MOUNT1/$tfile- 25
259 createmany -o $MOUNT2/$tfile-2- 1
260 createmany -o $MOUNT1/$tfile-3- 25
263 facet_failover $SINGLEMDS
264 # expect failover to fail due to missing client 2
265 client_evicted || return 1
268 # first 25 files should have been replayed
269 unlinkmany $MOUNT1/$tfile- 25 || return 2
271 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
274 run_test 14a "timeouts waiting for lost client during replay"
278 wait_destroy_complete
279 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
280 mkdir -p $MOUNT1/$tdir
281 $SETSTRIPE -o 0 $MOUNT1/$tdir
282 replay_barrier $SINGLEMDS
283 createmany -o $MOUNT1/$tdir/$tfile- 5
285 $SETSTRIPE -o 0 $MOUNT2/f14b-3
286 echo "data" > $MOUNT2/f14b-3
287 createmany -o $MOUNT1/$tdir/$tfile-3- 5
291 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
293 # first 25 files should have been replayed
294 unlinkmany $MOUNT1/$tdir/$tfile- 5 || return 2
295 unlinkmany $MOUNT1/$tdir/$tfile-3- 5 || return 3
297 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
299 wait_mds_ost_sync || return 4
300 wait_destroy_complete || return 5
302 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
303 log "before $BEFOREUSED, after $AFTERUSED"
304 [ $AFTERUSED -ne $BEFOREUSED ] && \
305 error "after $AFTERUSED > before $BEFOREUSED" && return 4
308 run_test 14b "delete ost orphans if gap occured in objids due to VBR"
310 test_15a() { # was test_15
311 replay_barrier $SINGLEMDS
312 createmany -o $MOUNT1/$tfile- 25
313 createmany -o $MOUNT2/$tfile-2- 1
318 unlinkmany $MOUNT1/$tfile- 25 || return 2
319 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
321 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
324 run_test 15a "timeout waiting for lost client during replay, 1 client completes"
327 replay_barrier $SINGLEMDS
328 for ((i = 0; i < 2000; i++)); do
329 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
335 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
338 run_test 15c "remove multiple OST orphans"
341 replay_barrier $SINGLEMDS
342 createmany -o $MOUNT1/$tfile- 25
343 createmany -o $MOUNT2/$tfile-2- 1
346 facet_failover $SINGLEMDS
350 unlinkmany $MOUNT1/$tfile- 25 || return 2
352 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
356 run_test 16 "fail MDS during recovery (3571)"
359 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
361 createmany -o $MOUNT1/$tfile- 25
362 createmany -o $MOUNT2/$tfile-2- 1
364 # Make sure the disconnect is lost
372 unlinkmany $MOUNT1/$tfile- 25 || return 2
374 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
378 run_test 17 "fail OST during recovery (3571)"
380 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
383 test_18() { # bug 3822 - evicting client with enqueued lock
385 mkdir -p $MOUNT1/$tdir
386 touch $MOUNT1/$tdir/f0
387 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
388 statmany -s $MOUNT1/$tdir/f 1 500 &
391 do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b # hold enqueue
393 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
394 do_facet client lctl set_param fail_loc=0x80000305 # drop cb, evict
396 usleep 500 # wait to ensure first client is one that will be evicted
397 openfile -f O_RDONLY $MOUNT2/$tdir/f0
399 dmesg | grep "entering recovery in server" && \
400 error "client not evicted" || true
401 do_facet client "lctl set_param fail_loc=0"
402 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
404 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
406 test_19() { # Bug 10991 - resend of open request does not fail assertion.
407 replay_barrier $SINGLEMDS
408 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
410 checkstat $DIR2/${tfile}0 || return 2
411 rm $DIR/${tfile}0 || return 3
415 run_test 19 "resend of open request"
419 replay_barrier $SINGLEMDS
425 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
426 TIER1=$((`date +%s` - BEFORE))
428 replay_barrier $SINGLEMDS
434 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
435 TIER2=$((`date +%s` - BEFORE))
436 [ $TIER2 -ge $((TIER1 * 2)) ] && \
437 error "recovery time is growing $TIER2 > $TIER1"
440 run_test 20 "recovery time is not increasing"
442 # commit on sharing tests
444 local param_file=$TMP/$tfile-params
446 save_lustre_params $(facet_active_host $SINGLEMDS) "mdt.*.commit_on_sharing" > $param_file
447 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=1
448 touch $MOUNT1/$tfile-1
449 mv $MOUNT2/$tfile-1 $MOUNT2/$tfile-2
450 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
451 replay_barrier_nosync $SINGLEMDS
454 facet_failover $SINGLEMDS
456 # all renames are replayed
457 unlink $MOUNT1/$tfile-3 || return 2
459 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
461 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=0
462 rm -rf $MOUNT1/$tfile-*
463 restore_lustre_params < $param_file
467 run_test 21a "commit on sharing"
471 do_node $CLIENT1 rm -f $MOUNT1/$tfile-*
474 do_node $CLIENT1 touch $MOUNT1/$tfile-1
475 do_node $CLIENT2 mv $MOUNT1/$tfile-1 $MOUNT1/$tfile-2
476 do_node $CLIENT1 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
478 replay_barrier_nosync $mds
479 shutdown_client $CLIENT2 $MOUNT1
483 # were renames replayed?
485 echo UNLINK $MOUNT1/$tfile-3
486 do_node $CLIENT1 unlink $MOUNT1/$tfile-3 || \
487 { echo "unlink $tfile-3 fail!" && rc=1; }
490 zconf_mount_clients $CLIENT2 $MOUNT1 || error "mount $CLIENT2 $MOUNT1 fail"
496 [ -z "$CLIENTS" ] && skip "Need two or more clients." && return
497 [ $CLIENTCOUNT -lt 2 ] && \
498 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
500 if [ "$FAILURE_MODE" = "HARD" ] && mixed_mdt_devs; then
501 skip "Several mdt services on one mds node are used with FAILURE_MODE=$FAILURE_MODE. "
506 zconf_umount_clients $CLIENTS $MOUNT2
507 zconf_mount_clients $CLIENTS $MOUNT1
509 local param_file=$TMP/$tfile-params
511 local num=$(get_mds_dir $MOUNT1)
513 save_lustre_params $(facet_active_host mds$num) "mdt.*.commit_on_sharing" > $param_file
517 do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
519 test_21b_sub mds$num || error "Not all renames are replayed. COS=$COS"
521 # COS disabled (should fail)
523 do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
525 # there is still a window when transactions may be written to disk before
526 # the mds device is set R/O. To avoid such a rare test failure, the check
527 # is repeated several times.
530 test_21b_sub mds$num || break;
531 let n_attempts=n_attempts+1
532 [ $n_attemtps -gt 3] &&
533 error "The test cannot check whether COS works or not: all renames are replied w/o COS"
535 restore_lustre_params < $param_file
539 run_test 21b "commit on sharing, two clients"
541 # end commit on sharing tests
543 equals_msg `basename $0`: test complete, cleaning up
544 SLEEP=$((`date +%s` - $NOW))
545 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
546 [ "$MOUNTED2" = yes ] && zconf_umount $HOSTNAME $MOUNT2 || true
547 check_and_cleanup_lustre
548 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true