9 LUSTRE=${LUSTRE:-`dirname $0`/..}
11 . $LUSTRE/tests/test-framework.sh
15 . ${CONFIG:=$LUSTRE/tests/cfg/local.sh}
20 # Allow us to override the setup if we already have a mounted system by
21 # setting SETUP=" " and CLEANUP=" "
22 SETUP=${SETUP:-"setup"}
23 CLEANUP=${CLEANUP:-"cleanup"}
24 FORCE=${FORCE:-"--force"}
28 add_mds mds --dev $MDSDEV --size $MDSSIZE
29 add_lov lov1 mds --stripe_sz $STRIPE_BYTES \
30 --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
31 add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE
32 add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE
33 add_client client mds --lov lov1 --path $MOUNT
38 start ost --reformat $OSTLCONFARGS
39 start ost2 --reformat $OSTLCONFARGS
40 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
41 start mds $MDSLCONFARGS --reformat
42 grep " $MOUNT " /proc/mounts || zconf_mount `hostname` $MOUNT
46 zconf_umount `hostname` $MOUNT
47 stop mds ${FORCE} $MDSLCONFARGS
49 stop ost ${FORCE} --dump $TMP/recovery-small-`hostname`.log
54 do_mds 'echo -e "device \$mds1\\nprobe\\nnotransno\\nreadonly" | lctl'
59 do_client "df -h $MOUNT" # trigger failover, if we haven't already
62 if [ ! -z "$EVAL" ]; then
67 if [ "$ONLY" == "cleanup" ]; then
68 sysctl -w portals.debug=0 || true
73 REFORMAT=--reformat $SETUP
76 [ "$ONLY" == "setup" ] && exit
79 drop_request "mcreate $MOUNT/1" || return 1
80 drop_reint_reply "mcreate $MOUNT/2" || return 2
82 run_test 1 "mcreate: drop req, drop rep"
85 drop_request "tchmod 111 $MOUNT/2" || return 1
86 drop_reint_reply "tchmod 666 $MOUNT/2" || return 2
88 run_test 2 "chmod: drop req, drop rep"
91 drop_request "statone $MOUNT/2" || return 1
92 drop_reply "statone $MOUNT/2" || return 2
94 run_test 3 "stat: drop req, drop rep"
97 do_facet client "cp /etc/resolv.conf $MOUNT/resolv.conf" || return 1
98 drop_request "cat $MOUNT/resolv.conf > /dev/null" || return 2
99 drop_reply "cat $MOUNT/resolv.conf > /dev/null" || return 3
101 run_test 4 "open: drop req, drop rep"
104 drop_request "mv $MOUNT/resolv.conf $MOUNT/renamed" || return 1
105 drop_reint_reply "mv $MOUNT/renamed $MOUNT/renamed-again" || return 2
106 do_facet client "checkstat -v $MOUNT/renamed-again" || return 3
108 run_test 5 "rename: drop req, drop rep"
111 drop_request "mlink $MOUNT/renamed-again $MOUNT/link1" || return 1
112 drop_reint_reply "mlink $MOUNT/renamed-again $MOUNT/link2" || return 2
114 run_test 6 "link: drop req, drop rep"
117 drop_request "munlink $MOUNT/link1" || return 1
118 drop_reint_reply "munlink $MOUNT/link2" || return 2
120 run_test 7 "unlink: drop req, drop rep"
124 drop_reint_reply "touch $MOUNT/renamed" || return 1
126 run_test 8 "touch: drop rep (bug 1423)"
130 pause_bulk "cp /etc/profile $MOUNT" || return 1
131 do_facet client "cp /etc/termcap $MOUNT" || return 2
132 do_facet client "sync"
133 do_facet client "rm $MOUNT/termcap $MOUNT/profile" || return 3
135 run_test 9 "pause bulk on OST (bug 1420)"
139 do_facet client mcreate $MOUNT/$tfile || return 1
140 drop_bl_callback "chmod 0777 $MOUNT/$tfile" || return 2
141 # wait for the mds to evict the client
142 #echo "sleep $(($TIMEOUT*2))"
143 #sleep $(($TIMEOUT*2))
144 do_facet client touch $MOUNT/$tfile || echo "touch failed, evicted"
145 do_facet client checkstat -v -p 0777 $MOUNT/$tfile || return 3
146 do_facet client "munlink $MOUNT/$tfile"
148 run_test 10 "finish request on server after client eviction (bug 1521)"
151 # wake up a thread waiting for completion after eviction
153 do_facet client multiop $MOUNT/$tfile Ow || return 1
154 do_facet client multiop $MOUNT/$tfile or || return 2
158 do_facet client multiop $MOUNT/$tfile or || return 3
159 drop_bl_callback multiop $MOUNT/$tfile Ow ||
160 echo "client evicted, as expected"
162 do_facet client munlink $MOUNT/$tfile || return 4
164 run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
168 $LCTL mark multiop $MOUNT/$tfile OS_c
169 do_facet mds "sysctl -w lustre.fail_loc=0x115"
170 clear_failloc mds $((TIMEOUT * 2)) &
171 multiop $MOUNT/$tfile OS_c &
173 #define OBD_FAIL_MDS_CLOSE_NET 0x115
176 echo "waiting for multiop $PID"
177 wait $PID || return 2
178 do_facet client munlink $MOUNT/$tfile || return 3
180 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
182 # Bug 113, check that readdir lost recv timeout works.
184 mkdir /mnt/lustre/readdir || return 1
185 touch /mnt/lustre/readdir/newentry || return
186 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
187 do_facet mds "sysctl -w lustre.fail_loc=0x80000104"
188 ls /mnt/lustre/readdir || return 3
189 do_facet mds "sysctl -w lustre.fail_loc=0"
190 rm -rf /mnt/lustre/readdir || return 4
192 run_test 13 "mdc_readpage restart test (bug 1138)"
194 # Bug 113, check that readdir lost send timeout works.
196 mkdir /mnt/lustre/readdir
197 touch /mnt/lustre/readdir/newentry
198 # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
199 do_facet mds "sysctl -w lustre.fail_loc=0x80000106"
200 ls /mnt/lustre/readdir || return 1
201 do_facet mds "sysctl -w lustre.fail_loc=0"
203 run_test 14 "mdc_readpage resend test (bug 1138)"
206 do_facet mds "sysctl -w lustre.fail_loc=0x80000128"
207 touch $DIR/$tfile && return 1
210 run_test 15 "failed open (-ENOMEM)"
212 READ_AHEAD=`cat /proc/fs/lustre/llite/*/max_read_ahead_mb | head -n 1`
214 for f in /proc/fs/lustre/llite/*/max_read_ahead_mb; do
220 for f in /proc/fs/lustre/llite/*/max_read_ahead_mb; do
221 echo $READ_AHEAD > $f
226 do_facet client cp /etc/termcap $MOUNT
230 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
231 sysctl -w lustre.fail_loc=0x80000504
233 # will get evicted here
234 do_facet client "cmp /etc/termcap $MOUNT/termcap" && return 1
235 sysctl -w lustre.fail_loc=0
236 # give recovery a chance to finish (shouldn't take long)
238 do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 2
241 run_test 16 "timeout bulk put, evict client (2732)"
244 # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
245 # client will get evicted here
246 sysctl -w lustre.fail_loc=0x80000503
247 # need to ensure we send an RPC
248 do_facet client cp /etc/termcap $DIR/$tfile
252 sysctl -w lustre.fail_loc=0
253 do_facet client "df $DIR"
255 do_facet client "cmp /etc/termcap $DIR/$tfile" && return 1
256 do_facet client "rm $DIR/$tfile" || return 2
259 run_test 17 "timeout bulk get, evict client (2732)"
262 do_facet client mkdir -p $MOUNT/$tdir
263 f=$MOUNT/$tdir/$tfile
266 pgcache_empty || return 1
269 lfs setstripe $f $((128 * 1024)) 1 1
271 do_facet client cp /etc/termcap $f
273 local osc2_dev=`$LCTL device_list | \
274 awk '(/ost2.*client_facet/){print $4}' `
275 $LCTL --device %$osc2_dev deactivate
276 # my understanding is that there should be nothing in the page
277 # cache after the client reconnects?
279 pgcache_empty || rc=2
280 $LCTL --device %$osc2_dev activate
284 run_test 18a "manual ost invalidate clears page cache immediately"
287 do_facet client mkdir -p $MOUNT/$tdir
288 f=$MOUNT/$tdir/$tfile
289 f2=$MOUNT/$tdir/${tfile}-2
292 pgcache_empty || return 1
294 # shouldn't have to set stripe size of count==1
295 lfs setstripe $f $((128 * 1024)) 0 1
296 lfs setstripe $f2 $((128 * 1024)) 0 1
298 do_facet client cp /etc/termcap $f
300 # just use this write to trigger the client's eviction from the ost
301 # OBD_FAIL_PTLRPC_BULK_GET_NET|OBD_FAIL_ONCE
302 sysctl -w lustre.fail_loc=0x80000503
303 do_facet client dd if=/dev/zero of=$f2 bs=4k count=1
305 sysctl -w lustre.fail_loc=0
306 # allow recovery to complete
307 sleep $((TIMEOUT + 2))
308 # my understanding is that there should be nothing in the page
309 # cache after the client reconnects?
311 pgcache_empty || rc=2
315 run_test 18b "eviction and reconnect clears page cache (2766)"
319 do_facet client mcreate $f || return 1
320 drop_ldlm_cancel "chmod 0777 $f" || echo evicted
322 do_facet client checkstat -v -p 0777 $f || echo evicted
323 # let the client reconnect
325 do_facet client "munlink $f"
327 run_test 19a "test expired_lock_main on mds (2867)"
331 do_facet client multiop $f Ow || return 1
332 do_facet client multiop $f or || return 2
336 do_facet client multiop $f or || return 3
337 drop_ldlm_cancel multiop $f Ow || echo "client evicted, as expected"
339 do_facet client munlink $f || return 4
341 run_test 19b "test expired_lock_main on ost (2867)"
343 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
345 multiop $DIR/$tdir/${tfile} O_wc &
349 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
350 do_facet ost sysctl -w lustre.fail_loc=0x80000308
351 kill -USR1 $MULTI_PID
354 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
356 run_test 20a "ldlm_handle_enqueue error (should return error)"
358 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
360 touch $DIR/$tdir/${tfile}
362 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
363 do_facet ost sysctl -w lustre.fail_loc=0x80000308
364 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
365 error "didn't fail open enqueue" || true
367 run_test 20b "ldlm_handle_enqueue error (should return error)"
369 #b_cray run_test 21a "drop close request while close and open are both in flight"
370 #b_cray run_test 21b "drop open request while close and open are both in flight"
371 #b_cray run_test 21c "drop both request while close and open are both in flight"
372 #b_cray run_test 21d "drop close reply while close and open are both in flight"
373 #b_cray run_test 21e "drop open reply while close and open are both in flight"
374 #b_cray run_test 21f "drop both reply while close and open are both in flight"
375 #b_cray run_test 21g "drop open reply and close request while close and open are both in flight"
376 #b_cray run_test 21h "drop open request and close reply while close and open are both in flight"
377 #b_cray run_test 22 "drop close request and do mknod"
378 #b_cray run_test 23 "client hang when close a file after mds crash"
380 test_24() { # bug 2248 - eviction fails writeback but app doesn't see it
383 multiop $DIR/$tdir/$tfile Owy_wyc &
386 # OBD_FAIL_PTLRPC_BULK_GET_NET|OBD_FAIL_ONCE
387 sysctl -w lustre.fail_loc=0x80000503
389 kill -USR1 $MULTI_PID
392 sysctl -w lustre.fail_loc=0x0
394 [ $rc -eq 0 ] && error "multiop didn't fail fsync: rc $rc" || true
396 run_test 24 "fsync error (should return error)"
398 test_26() { # bug 5921 - evict dead exports
399 # this test can only run from a client on a separate node.
400 [ "`lsmod | grep obdfilter`" ] && \
401 echo "skipping test 26 (local OST)" && return
402 [ "`lsmod | grep mds`" ] && \
403 echo "skipping test 26 (local MDS)" && return
404 OST_FILE=/proc/fs/lustre/obdfilter/ost_svc/num_exports
405 OST_EXP="`do_facet ost cat $OST_FILE`"
406 OST_NEXP1=`echo $OST_EXP | cut -d' ' -f2`
407 echo starting with $OST_NEXP1 OST exports
408 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
409 do_facet client sysctl -w lustre.fail_loc=0x505
410 # evictor takes up to 2.25x to evict. But if there's a
411 # race to start the evictor from various obds, the loser
412 # might have to wait for the next ping.
413 echo Waiting for $(($TIMEOUT * 4)) secs
414 sleep $(($TIMEOUT * 4))
415 OST_EXP="`do_facet ost cat $OST_FILE`"
416 OST_NEXP2=`echo $OST_EXP | cut -d' ' -f2`
417 echo ending with $OST_NEXP2 OST exports
418 do_facet client sysctl -w lustre.fail_loc=0x0
419 [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted"
422 run_test 26 "evict dead exports"
425 [ "`lsmod | grep mds`" ] || \
426 { echo "skipping test 27 (non-local MDS)" && return 0; }
428 writemany -q -a $DIR/$tdir/$tfile 0 5 &
433 #define OBD_FAIL_OSC_SHUTDOWN 0x407
434 sysctl -w lustre.fail_loc=0x80000407
435 # need to wait for reconnect
436 echo -n waiting for fail_loc
437 while [ `sysctl -n lustre.fail_loc` -eq -2147482617 ]; do
443 kill -USR1 $CLIENT_PID
447 run_test 27 "fail LOV while using OSC's"
449 test_28() { # bug 6086 - error adding new clients
450 do_facet client mcreate $MOUNT/$tfile || return 1
451 drop_bl_callback "chmod 0777 $MOUNT/$tfile" || return 2
452 #define OBD_FAIL_MDS_ADD_CLIENT 0x12f
453 do_facet mds sysctl -w lustre.fail_loc=0x8000012f
454 # fail once (evicted), reconnect fail (fail_loc), ok
455 df || (sleep 1; df) || (sleep 1; df) || error "reconnect failed"
457 fail mds # verify MDS last_rcvd can be loaded
459 run_test 28 "handle error adding new clients (bug 6086)"
463 # put a load of file creates/writes/deletes
464 writemany -q $DIR/$tdir/$tfile 0 5 &
466 echo writemany pid $CLIENT_PID
470 # wait for client to reconnect to MDS
475 # client process should see no problems even though MDS went down
477 kill -USR1 $CLIENT_PID
480 echo writemany returned $rc
481 #these may fail because of eviction due to slow AST response.
484 run_test 50 "failover MDS under load"
488 # put a load of file creates/writes/deletes
489 writemany -q $DIR/$tdir/$tfile 0 5 &
494 # failover at various points during recovery
495 SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
496 echo will failover at $SEQ
499 echo failover in $i sec
503 # client process should see no problems even though MDS went down
504 # and recovery was interrupted
506 kill -USR1 $CLIENT_PID
509 echo writemany returned $rc
512 run_test 51 "failover MDS during recovery"
515 do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
517 echo writemany pid $CLIENT_PID
522 wait $CLIENT_PID || rc=$?
523 # active client process should see an EIO for down OST
524 [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
525 # but timing or failover setup may allow success
526 [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
527 echo "writemany returned $rc"
535 [ $rc -ne 0 ] && { return $rc; }
536 # wait for client to reconnect to OST
540 [ $rc -ne 0 ] && { return $rc; }
547 run_test 52 "failover OST under load"
550 FORCE=--force $CLEANUP