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"
256 # test 14a removed after 18143 because it shouldn't fail anymore and do the same
261 wait_destroy_complete
262 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
263 mkdir -p $MOUNT1/$tdir
264 $SETSTRIPE -o 0 $MOUNT1/$tdir
265 replay_barrier $SINGLEMDS
266 createmany -o $MOUNT1/$tdir/$tfile- 5
268 $SETSTRIPE -o 0 $MOUNT2/f14b-3
269 echo "data" > $MOUNT2/f14b-3
270 createmany -o $MOUNT1/$tdir/$tfile-3- 5
274 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
276 # first 25 files should have been replayed
277 unlinkmany $MOUNT1/$tdir/$tfile- 5 || return 2
278 unlinkmany $MOUNT1/$tdir/$tfile-3- 5 || return 3
280 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
282 wait_mds_ost_sync || return 4
283 wait_destroy_complete || return 5
285 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
286 log "before $BEFOREUSED, after $AFTERUSED"
287 [ $AFTERUSED -ne $BEFOREUSED ] && \
288 error "after $AFTERUSED > before $BEFOREUSED" && return 4
291 run_test 14b "delete ost orphans if gap occured in objids due to VBR"
293 test_15a() { # was test_15
294 replay_barrier $SINGLEMDS
295 createmany -o $MOUNT1/$tfile- 25
296 createmany -o $MOUNT2/$tfile-2- 1
301 unlinkmany $MOUNT1/$tfile- 25 || return 2
302 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
304 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
307 run_test 15a "timeout waiting for lost client during replay, 1 client completes"
310 replay_barrier $SINGLEMDS
311 for ((i = 0; i < 2000; i++)); do
312 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
318 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
321 run_test 15c "remove multiple OST orphans"
324 replay_barrier $SINGLEMDS
325 createmany -o $MOUNT1/$tfile- 25
326 createmany -o $MOUNT2/$tfile-2- 1
329 facet_failover $SINGLEMDS
333 unlinkmany $MOUNT1/$tfile- 25 || return 2
335 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
339 run_test 16 "fail MDS during recovery (3571)"
342 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
344 createmany -o $MOUNT1/$tfile- 25
345 createmany -o $MOUNT2/$tfile-2- 1
347 # Make sure the disconnect is lost
355 unlinkmany $MOUNT1/$tfile- 25 || return 2
357 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
361 run_test 17 "fail OST during recovery (3571)"
363 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
366 test_18() { # bug 3822 - evicting client with enqueued lock
368 mkdir -p $MOUNT1/$tdir
369 touch $MOUNT1/$tdir/f0
370 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
371 statmany -s $MOUNT1/$tdir/f 1 500 &
374 do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b # hold enqueue
376 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
377 do_facet client lctl set_param fail_loc=0x80000305 # drop cb, evict
379 usleep 500 # wait to ensure first client is one that will be evicted
380 openfile -f O_RDONLY $MOUNT2/$tdir/f0
382 dmesg | grep "entering recovery in server" && \
383 error "client not evicted" || true
384 do_facet client "lctl set_param fail_loc=0"
385 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
387 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
389 test_19() { # Bug 10991 - resend of open request does not fail assertion.
390 replay_barrier $SINGLEMDS
391 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
393 checkstat $DIR2/${tfile}0 || return 2
394 rm $DIR/${tfile}0 || return 3
398 run_test 19 "resend of open request"
402 replay_barrier $SINGLEMDS
408 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
409 TIER1=$((`date +%s` - BEFORE))
411 replay_barrier $SINGLEMDS
417 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
418 TIER2=$((`date +%s` - BEFORE))
419 [ $TIER2 -ge $((TIER1 * 2)) ] && \
420 error "recovery time is growing $TIER2 > $TIER1"
423 run_test 20 "recovery time is not increasing"
425 # commit on sharing tests
427 local param_file=$TMP/$tfile-params
429 save_lustre_params $(facet_active_host $SINGLEMDS) "mdt.*.commit_on_sharing" > $param_file
430 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=1
431 touch $MOUNT1/$tfile-1
432 mv $MOUNT2/$tfile-1 $MOUNT2/$tfile-2
433 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
434 replay_barrier_nosync $SINGLEMDS
437 facet_failover $SINGLEMDS
439 # all renames are replayed
440 unlink $MOUNT1/$tfile-3 || return 2
442 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
444 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=0
445 rm -rf $MOUNT1/$tfile-*
446 restore_lustre_params < $param_file
450 run_test 21a "commit on sharing"
454 do_node $CLIENT1 rm -f $MOUNT1/$tfile-*
457 do_node $CLIENT1 touch $MOUNT1/$tfile-1
458 do_node $CLIENT2 mv $MOUNT1/$tfile-1 $MOUNT1/$tfile-2
459 do_node $CLIENT1 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
461 replay_barrier_nosync $mds
462 shutdown_client $CLIENT2 $MOUNT1
466 # were renames replayed?
468 echo UNLINK $MOUNT1/$tfile-3
469 do_node $CLIENT1 unlink $MOUNT1/$tfile-3 || \
470 { echo "unlink $tfile-3 fail!" && rc=1; }
473 zconf_mount_clients $CLIENT2 $MOUNT1 || error "mount $CLIENT2 $MOUNT1 fail"
479 [ -z "$CLIENTS" ] && skip "Need two or more clients." && return
480 [ $CLIENTCOUNT -lt 2 ] && \
481 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
483 if [ "$FAILURE_MODE" = "HARD" ] && mixed_mdt_devs; then
484 skip "Several mdt services on one mds node are used with FAILURE_MODE=$FAILURE_MODE. "
489 zconf_umount_clients $CLIENTS $MOUNT2
490 zconf_mount_clients $CLIENTS $MOUNT1
492 local param_file=$TMP/$tfile-params
494 local num=$(get_mds_dir $MOUNT1)
496 save_lustre_params $(facet_active_host mds$num) "mdt.*.commit_on_sharing" > $param_file
500 do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
502 test_21b_sub mds$num || error "Not all renames are replayed. COS=$COS"
504 # COS disabled (should fail)
506 do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
508 # there is still a window when transactions may be written to disk before
509 # the mds device is set R/O. To avoid such a rare test failure, the check
510 # is repeated several times.
513 test_21b_sub mds$num || break;
514 let n_attempts=n_attempts+1
515 [ $n_attempts -gt 3 ] &&
516 error "The test cannot check whether COS works or not: all renames are replied w/o COS"
518 restore_lustre_params < $param_file
522 run_test 21b "commit on sharing, two clients"
524 # end commit on sharing tests
526 equals_msg `basename $0`: test complete, cleaning up
527 SLEEP=$((`date +%s` - $NOW))
528 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
529 [ "$MOUNTED2" = yes ] && zconf_umount $HOSTNAME $MOUNT2 || true
530 check_and_cleanup_lustre
531 [ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true