6 ALWAYS_EXCEPT="15c $REPLAY_DUAL_EXCEPT"
9 PTLDEBUG=${PTLDEBUG:--1}
10 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
12 CLEANUP=${CLEANUP:-""}
13 MOUNT_2=${MOUNT_2:-"yes"}
14 . $LUSTRE/tests/test-framework.sh
16 if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
17 CONFIG_EXCEPTIONS="17"
18 echo -n "Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
19 echo "Except the tests: $CONFIG_EXCEPTIONS"
20 ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
24 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
27 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
29 [ "$SLOW" = "no" ] && EXCEPT_SLOW="21b"
33 check_and_setup_lustre
34 MOUNTED=$(mounted_lustre_filesystems)
35 if ! $(echo $MOUNTED | grep -w -q $MOUNT2); then
36 zconf_mount $HOSTNAME $MOUNT2
41 rm -rf $DIR/[df][0-9]*
43 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
47 replay_barrier $SINGLEMDS
51 checkstat $MOUNT2/a || return 1
52 checkstat $MOUNT1/b || return 2
53 rm $MOUNT2/a $MOUNT1/b
54 checkstat $MOUNT1/a && return 3
55 checkstat $MOUNT2/b && return 4
59 run_test 1 "|X| simple create"
63 replay_barrier $SINGLEMDS
67 checkstat $MOUNT2/adir || return 1
69 checkstat $MOUNT2/adir && return 2
72 run_test 2 "|X| mkdir adir"
75 replay_barrier $SINGLEMDS
77 mkdir $MOUNT2/adir/bdir
80 checkstat $MOUNT2/adir || return 1
81 checkstat $MOUNT1/adir/bdir || return 2
82 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
83 checkstat $MOUNT1/adir && return 3
84 checkstat $MOUNT2/adir/bdir && return 4
87 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
91 replay_barrier $SINGLEMDS
92 mkdir $MOUNT1/adir && return 1
93 mkdir $MOUNT2/adir/bdir
96 checkstat $MOUNT2/adir || return 2
97 checkstat $MOUNT1/adir/bdir || return 3
99 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
100 checkstat $MOUNT1/adir && return 4
101 checkstat $MOUNT2/adir/bdir && return 5
104 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
108 # multiclient version of replay_single.sh/test_8
110 multiop_bg_pause $MOUNT2/a o_tSc || return 1
113 replay_barrier $SINGLEMDS
115 wait $pid || return 1
118 [ -e $MOUNT2/a ] && return 2
121 run_test 5 "open, unlink |X| close"
126 multiop_bg_pause $MOUNT2/a o_c || return 1
128 multiop_bg_pause $MOUNT1/a o_c || return 1
131 replay_barrier $SINGLEMDS
133 wait $pid1 || return 1
137 wait $pid2 || return 1
138 [ -e $MOUNT2/a ] && return 2
141 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
144 replay_barrier $SINGLEMDS
145 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
147 checkstat $MOUNT2/$tfile || return 2
148 rm $MOUNT1/$tfile || return 3
152 run_test 8 "replay of resent request"
155 replay_barrier $SINGLEMDS
156 mcreate $MOUNT1/$tfile-1
157 mcreate $MOUNT2/$tfile-2
158 # drop first reint reply
159 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
161 do_facet $SINGLEMDS lctl set_param fail_loc=0
163 rm $MOUNT1/$tfile-[1,2] || return 1
167 run_test 9 "resending a replayed create"
170 mcreate $MOUNT1/$tfile-1
171 replay_barrier $SINGLEMDS
172 munlink $MOUNT1/$tfile-1
173 mcreate $MOUNT2/$tfile-2
174 # drop first reint reply
175 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
177 do_facet $SINGLEMDS lctl set_param fail_loc=0
179 checkstat $MOUNT1/$tfile-1 && return 1
180 checkstat $MOUNT1/$tfile-2 || return 2
185 run_test 10 "resending a replayed unlink"
188 replay_barrier $SINGLEMDS
189 mcreate $MOUNT1/$tfile-1
190 mcreate $MOUNT2/$tfile-2
191 mcreate $MOUNT1/$tfile-3
192 mcreate $MOUNT2/$tfile-4
193 mcreate $MOUNT1/$tfile-5
194 # drop all reint replies for a while
195 do_facet $SINGLEMDS lctl set_param fail_loc=0x0119
196 # note that with this fail_loc set, facet_failover df will fail
197 facet_failover $SINGLEMDS
198 #sleep for while, let both clients reconnect and timeout
199 sleep $((TIMEOUT * 2))
200 do_facet $SINGLEMDS lctl set_param fail_loc=0
202 rm $MOUNT1/$tfile-[1-5] || return 1
206 run_test 11 "both clients timeout during replay"
209 replay_barrier $SINGLEMDS
211 multiop_bg_pause $DIR/$tfile mo_c || return 1
214 #define OBD_FAIL_LDLM_ENQUEUE 0x302
215 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000302
216 facet_failover $SINGLEMDS
217 do_facet $SINGLEMDS lctl set_param fail_loc=0
218 clients_up || return 1
221 kill -USR1 $MULTIPID || return 3
222 wait $MULTIPID || return 4
223 $CHECKSTAT -t file $DIR/$tfile || return 2
228 run_test 12 "open resend timeout"
231 multiop_bg_pause $DIR/$tfile mo_c || return 1
234 replay_barrier $SINGLEMDS
236 kill -USR1 $MULTIPID || return 3
237 wait $MULTIPID || return 4
240 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000115
241 facet_failover $SINGLEMDS
242 do_facet $SINGLEMDS lctl set_param fail_loc=0
243 clients_up || return 1
246 $CHECKSTAT -t file $DIR/$tfile || return 2
251 run_test 13 "close resend timeout"
253 # test 14a removed after 18143 because it shouldn't fail anymore and do the same
257 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
258 mkdir -p $MOUNT1/$tdir
259 replay_barrier $SINGLEMDS
260 createmany -o $MOUNT1/$tfile- 5
261 echo "data" > $MOUNT2/$tdir/$tfile-2
262 createmany -o $MOUNT1/$tfile-3- 5
266 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
268 # first 25 files should have been replayed
269 unlinkmany $MOUNT1/$tfile- 5 || return 2
270 unlinkmany $MOUNT1/$tfile-3- 5 || return 3
272 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
274 wait_mds_ost_sync || return 4
275 wait_destroy_complete || return 5
277 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
278 log "before $BEFOREUSED, after $AFTERUSED"
279 [ $AFTERUSED -ne $BEFOREUSED ] && \
280 error "after $AFTERUSED > before $BEFOREUSED" && return 4
283 run_test 14b "delete ost orphans if gap occured in objids due to VBR"
285 test_15a() { # was test_15
286 replay_barrier $SINGLEMDS
287 createmany -o $MOUNT1/$tfile- 25
288 createmany -o $MOUNT2/$tfile-2- 1
293 unlinkmany $MOUNT1/$tfile- 25 || return 2
294 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
296 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
299 run_test 15a "timeout waiting for lost client during replay, 1 client completes"
302 replay_barrier $SINGLEMDS
303 for ((i = 0; i < 2000; i++)); do
304 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
310 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
313 run_test 15c "remove multiple OST orphans"
316 replay_barrier $SINGLEMDS
317 createmany -o $MOUNT1/$tfile- 25
318 createmany -o $MOUNT2/$tfile-2- 1
321 facet_failover $SINGLEMDS
325 unlinkmany $MOUNT1/$tfile- 25 || return 2
327 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
331 run_test 16 "fail MDS during recovery (3571)"
334 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
336 createmany -o $MOUNT1/$tfile- 25
337 createmany -o $MOUNT2/$tfile-2- 1
339 # Make sure the disconnect is lost
347 unlinkmany $MOUNT1/$tfile- 25 || return 2
349 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
353 run_test 17 "fail OST during recovery (3571)"
355 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
358 test_18() { # bug 3822 - evicting client with enqueued lock
360 mkdir -p $MOUNT1/$tdir
361 touch $MOUNT1/$tdir/f0
362 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
363 statmany -s $MOUNT1/$tdir/f 1 500 &
366 do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b # hold enqueue
368 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
369 do_facet client lctl set_param fail_loc=0x80000305 # drop cb, evict
371 usleep 500 # wait to ensure first client is one that will be evicted
372 openfile -f O_RDONLY $MOUNT2/$tdir/f0
374 dmesg | grep "entering recovery in server" && \
375 error "client not evicted" || true
376 do_facet client "lctl set_param fail_loc=0"
377 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
379 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
381 test_19() { # Bug 10991 - resend of open request does not fail assertion.
382 replay_barrier $SINGLEMDS
383 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
385 checkstat $DIR2/${tfile}0 || return 2
386 rm $DIR/${tfile}0 || return 3
390 run_test 19 "resend of open request"
394 replay_barrier $SINGLEMDS
400 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
401 TIER1=$((`date +%s` - BEFORE))
403 replay_barrier $SINGLEMDS
409 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
410 TIER2=$((`date +%s` - BEFORE))
411 [ $TIER2 -ge $((TIER1 * 2)) ] && \
412 error "recovery time is growing $TIER2 > $TIER1"
415 run_test 20 "recovery time is not increasing"
417 # commit on sharing tests
419 local param_file=$TMP/$tfile-params
421 save_lustre_params $(facet_active_host $SINGLEMDS) "mdt.*.commit_on_sharing" > $param_file
422 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=1
423 touch $MOUNT1/$tfile-1
424 mv $MOUNT2/$tfile-1 $MOUNT2/$tfile-2
425 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
426 replay_barrier_nosync $SINGLEMDS
429 facet_failover $SINGLEMDS
431 # all renames are replayed
432 unlink $MOUNT1/$tfile-3 || return 2
434 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
436 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=0
437 rm -rf $MOUNT1/$tfile-*
438 restore_lustre_params < $param_file
442 run_test 21a "commit on sharing"
446 do_node $CLIENT1 rm -f $MOUNT1/$tfile-*
449 do_node $CLIENT1 touch $MOUNT1/$tfile-1
450 do_node $CLIENT2 mv $MOUNT1/$tfile-1 $MOUNT1/$tfile-2
451 do_node $CLIENT1 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
453 replay_barrier_nosync $mds
454 shutdown_client $CLIENT2 $MOUNT1
458 # were renames replayed?
460 echo UNLINK $MOUNT1/$tfile-3
461 do_node $CLIENT1 unlink $MOUNT1/$tfile-3 || \
462 { echo "unlink $tfile-3 fail!" && rc=1; }
465 zconf_mount_clients $CLIENT2 $MOUNT1 || error "mount $CLIENT2 $MOUNT1 fail"
471 [ -z "$CLIENTS" ] && skip "Need two or more clients." && return
472 [ $CLIENTCOUNT -lt 2 ] && \
473 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
475 if [ "$FAILURE_MODE" = "HARD" ] && mixed_mdt_devs; then
476 skip "Several mdt services on one mds node are used with FAILURE_MODE=$FAILURE_MODE. "
481 zconf_umount_clients $CLIENTS $MOUNT2
482 zconf_mount_clients $CLIENTS $MOUNT1
484 local param_file=$TMP/$tfile-params
486 local num=$(get_mds_dir $MOUNT1)
488 save_lustre_params $(facet_active_host mds$num) "mdt.*.commit_on_sharing" > $param_file
492 do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
494 test_21b_sub mds$num || error "Not all renames are replayed. COS=$COS"
496 # COS disabled (should fail)
498 do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
500 # there is still a window when transactions may be written to disk before
501 # the mds device is set R/O. To avoid such a rare test failure, the check
502 # is repeated several times.
505 test_21b_sub mds$num || break;
506 let n_attempts=n_attempts+1
507 [ $n_attempts -gt 3 ] &&
508 error "The test cannot check whether COS works or not: all renames are replied w/o COS"
510 restore_lustre_params < $param_file
514 run_test 21b "commit on sharing, two clients"
516 # end commit on sharing tests
518 equals_msg `basename $0`: test complete, cleaning up
519 SLEEP=$((`date +%s` - $NOW))
520 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
521 [ "$MOUNTED2" = yes ] && zconf_umount $HOSTNAME $MOUNT2 || true
522 check_and_cleanup_lustre
523 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true