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
20 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
23 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
25 [ "$SLOW" = "no" ] && EXCEPT_SLOW="21b"
29 check_and_setup_lustre
30 MOUNTED=$(mounted_lustre_filesystems)
31 if ! $(echo $MOUNTED | grep -w -q $MOUNT2); then
32 zconf_mount $HOSTNAME $MOUNT2
37 rm -rf $DIR/[df][0-9]*
39 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
41 # LU-482 Avert LVM and VM inability to flush caches in pre .33 kernels
42 if [ $LINUX_VERSION_CODE -lt $(kernel_version 2 6 33) ]; then
43 sync; sleep 5; sync; sleep 5; sync; sleep 5
47 touch $MOUNT2/$tfile-A # force sync FLD/SEQ update before barrier
48 replay_barrier $SINGLEMDS
49 #define OBD_FAIL_PTLRPC_FINISH_REPLAY | OBD_FAIL_ONCE
51 createmany -o $MOUNT1/$tfile- 50
52 $LCTL set_param fail_loc=0x80000514
53 facet_failover $SINGLEMDS
57 zconf_mount `hostname` $MOUNT2 || error "mount2 fais"
58 unlinkmany $MOUNT1/$tfile- 50 || return 2
59 rm $MOUNT2/$tfile || return 3
60 rm $MOUNT2/$tfile-A || return 4
62 run_test 0a "expired recovery with lost client"
65 replay_barrier $SINGLEMDS
67 touch $MOUNT1/$tfile-2
69 facet_failover $SINGLEMDS
71 zconf_mount `hostname` $MOUNT1 || error "mount1 fais"
72 zconf_mount `hostname` $MOUNT2 || error "mount2 fais"
73 checkstat $MOUNT1/$tfile-2 && return 1
74 checkstat $MOUNT2/$tfile && return 2
77 run_test 0b "lost client during waiting for next transno"
81 replay_barrier $SINGLEMDS
85 checkstat $MOUNT2/a || return 1
86 checkstat $MOUNT1/b || return 2
87 rm $MOUNT2/a $MOUNT1/b
88 checkstat $MOUNT1/a && return 3
89 checkstat $MOUNT2/b && return 4
93 run_test 1 "|X| simple create"
97 replay_barrier $SINGLEMDS
101 checkstat $MOUNT2/adir || return 1
103 checkstat $MOUNT2/adir && return 2
106 run_test 2 "|X| mkdir adir"
109 replay_barrier $SINGLEMDS
111 mkdir $MOUNT2/adir/bdir
114 checkstat $MOUNT2/adir || return 1
115 checkstat $MOUNT1/adir/bdir || return 2
116 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
117 checkstat $MOUNT1/adir && return 3
118 checkstat $MOUNT2/adir/bdir && return 4
121 run_test 3 "|X| mkdir adir, mkdir adir/bdir "
125 replay_barrier $SINGLEMDS
126 mkdir $MOUNT1/adir && return 1
127 mkdir $MOUNT2/adir/bdir
130 checkstat $MOUNT2/adir || return 2
131 checkstat $MOUNT1/adir/bdir || return 3
133 rmdir $MOUNT2/adir/bdir $MOUNT1/adir
134 checkstat $MOUNT1/adir && return 4
135 checkstat $MOUNT2/adir/bdir && return 5
138 run_test 4 "|X| mkdir adir (-EEXIST), mkdir adir/bdir "
142 # multiclient version of replay_single.sh/test_8
144 multiop_bg_pause $MOUNT2/a o_tSc || return 1
147 replay_barrier $SINGLEMDS
149 wait $pid || return 1
152 [ -e $MOUNT2/a ] && return 2
155 run_test 5 "open, unlink |X| close"
160 multiop_bg_pause $MOUNT2/a o_c || return 1
162 multiop_bg_pause $MOUNT1/a o_c || return 1
165 replay_barrier $SINGLEMDS
167 wait $pid1 || return 1
171 wait $pid2 || return 1
172 [ -e $MOUNT2/a ] && return 2
175 run_test 6 "open1, open2, unlink |X| close1 [fail $SINGLEMDS] close2"
178 replay_barrier $SINGLEMDS
179 drop_reint_reply "mcreate $MOUNT1/$tfile" || return 1
181 checkstat $MOUNT2/$tfile || return 2
182 rm $MOUNT1/$tfile || return 3
186 run_test 8 "replay of resent request"
189 replay_barrier $SINGLEMDS
190 mcreate $MOUNT1/$tfile-1
191 mcreate $MOUNT2/$tfile-2
192 # drop first reint reply
193 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
195 do_facet $SINGLEMDS lctl set_param fail_loc=0
197 rm $MOUNT1/$tfile-[1,2] || return 1
201 run_test 9 "resending a replayed create"
204 mcreate $MOUNT1/$tfile-1
205 replay_barrier $SINGLEMDS
206 munlink $MOUNT1/$tfile-1
207 mcreate $MOUNT2/$tfile-2
208 # drop first reint reply
209 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000119
211 do_facet $SINGLEMDS lctl set_param fail_loc=0
213 checkstat $MOUNT1/$tfile-1 && return 1
214 checkstat $MOUNT1/$tfile-2 || return 2
219 run_test 10 "resending a replayed unlink"
222 replay_barrier $SINGLEMDS
223 mcreate $MOUNT1/$tfile-1
224 mcreate $MOUNT2/$tfile-2
225 mcreate $MOUNT1/$tfile-3
226 mcreate $MOUNT2/$tfile-4
227 mcreate $MOUNT1/$tfile-5
228 # drop all reint replies for a while
229 do_facet $SINGLEMDS lctl set_param fail_loc=0x0119
230 # note that with this fail_loc set, facet_failover df will fail
231 facet_failover $SINGLEMDS
232 #sleep for while, let both clients reconnect and timeout
233 sleep $((TIMEOUT * 2))
234 do_facet $SINGLEMDS lctl set_param fail_loc=0
236 rm $MOUNT1/$tfile-[1-5] || return 1
240 run_test 11 "both clients timeout during replay"
243 replay_barrier $SINGLEMDS
245 multiop_bg_pause $DIR/$tfile mo_c || return 1
248 #define OBD_FAIL_LDLM_ENQUEUE 0x302
249 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000302
250 facet_failover $SINGLEMDS
251 do_facet $SINGLEMDS lctl set_param fail_loc=0
252 clients_up || return 1
255 kill -USR1 $MULTIPID || return 3
256 wait $MULTIPID || return 4
257 $CHECKSTAT -t file $DIR/$tfile || return 2
262 run_test 12 "open resend timeout"
265 multiop_bg_pause $DIR/$tfile mo_c || return 1
268 replay_barrier $SINGLEMDS
270 kill -USR1 $MULTIPID || return 3
271 wait $MULTIPID || return 4
274 do_facet $SINGLEMDS lctl set_param fail_loc=0x80000115
275 facet_failover $SINGLEMDS
276 do_facet $SINGLEMDS lctl set_param fail_loc=0
277 clients_up || return 1
280 $CHECKSTAT -t file $DIR/$tfile || return 2
285 run_test 13 "close resend timeout"
287 # test 14a removed after 18143 because it shouldn't fail anymore and do the same
292 wait_destroy_complete
293 BEFOREUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
294 mkdir -p $MOUNT1/$tdir
295 $SETSTRIPE -o 0 $MOUNT1/$tdir
296 replay_barrier $SINGLEMDS
297 createmany -o $MOUNT1/$tdir/$tfile- 5
299 $SETSTRIPE -o 0 $MOUNT2/f14b-3
300 echo "data" > $MOUNT2/f14b-3
301 createmany -o $MOUNT1/$tdir/$tfile-3- 5
305 wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
307 # first 25 files should have been replayed
308 unlinkmany $MOUNT1/$tdir/$tfile- 5 || return 2
309 unlinkmany $MOUNT1/$tdir/$tfile-3- 5 || return 3
311 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
313 wait_mds_ost_sync || return 4
314 wait_destroy_complete || return 5
316 AFTERUSED=`df -P $DIR | tail -1 | awk '{ print $3 }'`
317 log "before $BEFOREUSED, after $AFTERUSED"
318 [ $AFTERUSED -ne $BEFOREUSED ] && \
319 error "after $AFTERUSED > before $BEFOREUSED" && return 4
322 run_test 14b "delete ost orphans if gap occured in objids due to VBR"
324 test_15a() { # was test_15
325 replay_barrier $SINGLEMDS
326 createmany -o $MOUNT1/$tfile- 25
327 createmany -o $MOUNT2/$tfile-2- 1
332 unlinkmany $MOUNT1/$tfile- 25 || return 2
333 [ -e $MOUNT1/$tfile-2-0 ] && error "$tfile-2-0 exists"
335 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
338 run_test 15a "timeout waiting for lost client during replay, 1 client completes"
341 replay_barrier $SINGLEMDS
342 for ((i = 0; i < 2000; i++)); do
343 echo "data" > "$MOUNT2/${tfile}-$i" || error "create ${tfile}-$i failed"
349 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
352 run_test 15c "remove multiple OST orphans"
355 replay_barrier $SINGLEMDS
356 createmany -o $MOUNT1/$tfile- 25
357 createmany -o $MOUNT2/$tfile-2- 1
360 facet_failover $SINGLEMDS
364 unlinkmany $MOUNT1/$tfile- 25 || return 2
366 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
370 run_test 16 "fail MDS during recovery (3571)"
373 remote_ost_nodsh && skip "remote OST with nodsh" && return 0
375 createmany -o $MOUNT1/$tfile- 25
376 createmany -o $MOUNT2/$tfile-2- 1
378 # Make sure the disconnect is lost
386 unlinkmany $MOUNT1/$tfile- 25 || return 2
388 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
392 run_test 17 "fail OST during recovery (3571)"
394 # cleanup with blocked enqueue fails until timer elapses (MDS busy), wait for it
397 test_18() { # bug 3822 - evicting client with enqueued lock
399 mkdir -p $MOUNT1/$tdir
400 touch $MOUNT1/$tdir/f0
401 #define OBD_FAIL_LDLM_ENQUEUE_BLOCKED 0x30b
402 statmany -s $MOUNT1/$tdir/f 1 500 &
405 do_facet $SINGLEMDS lctl set_param fail_loc=0x8000030b # hold enqueue
407 #define OBD_FAIL_LDLM_BL_CALLBACK 0x305
408 do_facet client lctl set_param fail_loc=0x80000305 # drop cb, evict
410 usleep 500 # wait to ensure first client is one that will be evicted
411 openfile -f O_RDONLY $MOUNT2/$tdir/f0
413 dmesg | grep "entering recovery in server" && \
414 error "client not evicted" || true
415 do_facet client "lctl set_param fail_loc=0"
416 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
418 run_test 18 "ldlm_handle_enqueue succeeds on evicted export (3822)"
420 test_19() { # Bug 10991 - resend of open request does not fail assertion.
421 replay_barrier $SINGLEMDS
422 drop_ldlm_reply "createmany -o $DIR/$tfile 1" || return 1
424 checkstat $DIR2/${tfile}0 || return 2
425 rm $DIR/${tfile}0 || return 3
429 run_test 19 "resend of open request"
433 replay_barrier $SINGLEMDS
439 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
440 TIER1=$((`date +%s` - BEFORE))
442 replay_barrier $SINGLEMDS
448 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
449 TIER2=$((`date +%s` - BEFORE))
450 [ $TIER2 -ge $((TIER1 * 2)) ] && \
451 error "recovery time is growing $TIER2 > $TIER1"
454 run_test 20 "recovery time is not increasing"
456 # commit on sharing tests
458 local param_file=$TMP/$tfile-params
460 save_lustre_params $(facet_active_host $SINGLEMDS) "mdt.*.commit_on_sharing" > $param_file
461 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=1
462 touch $MOUNT1/$tfile-1
463 mv $MOUNT2/$tfile-1 $MOUNT2/$tfile-2
464 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
465 replay_barrier_nosync $SINGLEMDS
468 facet_failover $SINGLEMDS
470 # all renames are replayed
471 unlink $MOUNT1/$tfile-3 || return 2
473 zconf_mount `hostname` $MOUNT2 || error "mount $MOUNT2 fail"
475 do_facet $SINGLEMDS lctl set_param mdt.*.commit_on_sharing=0
476 rm -rf $MOUNT1/$tfile-*
477 restore_lustre_params < $param_file
481 run_test 21a "commit on sharing"
485 do_node $CLIENT1 rm -f $MOUNT1/$tfile-*
488 do_node $CLIENT1 touch $MOUNT1/$tfile-1
489 do_node $CLIENT2 mv $MOUNT1/$tfile-1 $MOUNT1/$tfile-2
490 do_node $CLIENT1 mv $MOUNT1/$tfile-2 $MOUNT1/$tfile-3
492 replay_barrier_nosync $mds
493 shutdown_client $CLIENT2 $MOUNT1
497 # were renames replayed?
499 echo UNLINK $MOUNT1/$tfile-3
500 do_node $CLIENT1 unlink $MOUNT1/$tfile-3 || \
501 { echo "unlink $tfile-3 fail!" && rc=1; }
504 zconf_mount_clients $CLIENT2 $MOUNT1 || error "mount $CLIENT2 $MOUNT1 fail"
510 [ -z "$CLIENTS" ] && skip "Need two or more clients." && return
511 [ $CLIENTCOUNT -lt 2 ] && \
512 { skip "Need two or more clients, have $CLIENTCOUNT" && return; }
514 if [ "$FAILURE_MODE" = "HARD" ] && mixed_mdt_devs; then
515 skip "Several mdt services on one mds node are used with FAILURE_MODE=$FAILURE_MODE. "
520 zconf_umount_clients $CLIENTS $MOUNT2
521 zconf_mount_clients $CLIENTS $MOUNT1
523 local param_file=$TMP/$tfile-params
525 local num=$(get_mds_dir $MOUNT1)
527 save_lustre_params $(facet_active_host mds$num) "mdt.*.commit_on_sharing" > $param_file
531 do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
533 test_21b_sub mds$num || error "Not all renames are replayed. COS=$COS"
535 # COS disabled (should fail)
537 do_facet mds$num lctl set_param mdt.*.commit_on_sharing=$COS
539 # there is still a window when transactions may be written to disk before
540 # the mds device is set R/O. To avoid such a rare test failure, the check
541 # is repeated several times.
544 test_21b_sub mds$num || break;
545 let n_attempts=n_attempts+1
546 [ $n_attempts -gt 3 ] &&
547 error "The test cannot check whether COS works or not: all renames are replied w/o COS"
549 restore_lustre_params < $param_file
553 run_test 21b "commit on sharing, two clients"
555 # end commit on sharing tests
557 complete $(basename $0) $SECONDS
558 SLEEP=$((`date +%s` - $NOW))
559 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
560 [ "$MOUNTED2" = yes ] && zconf_umount $HOSTNAME $MOUNT2 || true
561 check_and_cleanup_lustre