6 ALWAYS_EXCEPT="20b 24 27 $RECOVERY_SMALL_EXCEPT"
8 LUSTRE=${LUSTRE:-`dirname $0`/..}
10 . $LUSTRE/tests/test-framework.sh
14 . ${CONFIG:=$LUSTRE/tests/cfg/local.sh}
19 # Allow us to override the setup if we already have a mounted system by
20 # setting SETUP=" " and CLEANUP=" "
21 SETUP=${SETUP:-"setup"}
22 CLEANUP=${CLEANUP:-"cleanup"}
27 add_mds mds --dev $MDSDEV --size $MDSSIZE
28 add_lov lov1 mds --stripe_sz $STRIPE_BYTES \
29 --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
30 add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE
31 add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE
32 add_client client mds --lov lov1 --path $MOUNT
37 start ost --reformat $OSTLCONFARGS
38 start ost2 --reformat $OSTLCONFARGS
39 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
40 start mds $MDSLCONFARGS --reformat
41 grep " $MOUNT " /proc/mounts || zconf_mount `hostname` $MOUNT
45 zconf_umount `hostname` $MOUNT
46 stop mds ${FORCE} $MDSLCONFARGS
48 stop ost ${FORCE} --dump $TMP/recovery-small-`hostname`.log
53 do_mds 'echo -e "device \$mds1\\nprobe\\nnotransno\\nreadonly" | lctl'
58 do_client "df -h $MOUNT" # trigger failover, if we haven't already
61 if [ ! -z "$EVAL" ]; then
66 if [ "$ONLY" == "cleanup" ]; then
67 sysctl -w lnet.debug=0 || true
72 REFORMAT=--reformat $SETUP
75 [ "$ONLY" == "setup" ] && exit
78 drop_request "mcreate $MOUNT/1" || return 1
79 drop_reint_reply "mcreate $MOUNT/2" || return 2
81 run_test 1 "mcreate: drop req, drop rep"
84 drop_request "tchmod 111 $MOUNT/2" || return 1
85 drop_reint_reply "tchmod 666 $MOUNT/2" || return 2
87 run_test 2 "chmod: drop req, drop rep"
90 drop_request "statone $MOUNT/2" || return 1
91 drop_reply "statone $MOUNT/2" || return 2
93 run_test 3 "stat: drop req, drop rep"
96 do_facet client "cp /etc/resolv.conf $MOUNT/resolv.conf" || return 1
97 drop_request "cat $MOUNT/resolv.conf > /dev/null" || return 2
98 drop_reply "cat $MOUNT/resolv.conf > /dev/null" || return 3
100 run_test 4 "open: drop req, drop rep"
103 drop_request "mv $MOUNT/resolv.conf $MOUNT/renamed" || return 1
104 drop_reint_reply "mv $MOUNT/renamed $MOUNT/renamed-again" || return 2
105 do_facet client "checkstat -v $MOUNT/renamed-again" || return 3
107 run_test 5 "rename: drop req, drop rep"
110 drop_request "mlink $MOUNT/renamed-again $MOUNT/link1" || return 1
111 drop_reint_reply "mlink $MOUNT/renamed-again $MOUNT/link2" || return 2
113 run_test 6 "link: drop req, drop rep"
116 drop_request "munlink $MOUNT/link1" || return 1
117 drop_reint_reply "munlink $MOUNT/link2" || return 2
119 run_test 7 "unlink: drop req, drop rep"
123 drop_reint_reply "touch $MOUNT/renamed" || return 1
125 run_test 8 "touch: drop rep (bug 1423)"
129 pause_bulk "cp /etc/profile $MOUNT" || return 1
130 do_facet client "cp /etc/termcap $MOUNT" || return 2
131 do_facet client "sync"
132 do_facet client "rm $MOUNT/termcap $MOUNT/profile" || return 3
134 run_test 9 "pause bulk on OST (bug 1420)"
138 do_facet client mcreate $MOUNT/$tfile || return 1
139 drop_bl_callback "chmod 0777 $MOUNT/$tfile" || return 2
140 # wait for the mds to evict the client
141 #echo "sleep $(($TIMEOUT*2))"
142 #sleep $(($TIMEOUT*2))
143 do_facet client touch $MOUNT/$tfile || echo "touch failed, evicted"
144 do_facet client checkstat -v -p 0777 $MOUNT/$tfile || return 3
145 do_facet client "munlink $MOUNT/$tfile"
147 run_test 10 "finish request on server after client eviction (bug 1521)"
150 # wake up a thread waiting for completion after eviction
152 do_facet client multiop $MOUNT/$tfile Ow || return 1
153 do_facet client multiop $MOUNT/$tfile or || return 2
157 do_facet client multiop $MOUNT/$tfile or || return 3
158 drop_bl_callback multiop $MOUNT/$tfile Ow ||
159 echo "client evicted, as expected"
161 do_facet client munlink $MOUNT/$tfile || return 4
163 run_test 11 "wake up a thread waiting for completion after eviction (b=2460)"
167 $LCTL mark multiop $MOUNT/$tfile OS_c
168 do_facet mds "sysctl -w lustre.fail_loc=0x115"
169 clear_failloc mds $((TIMEOUT * 2)) &
170 multiop $MOUNT/$tfile OS_c &
172 #define OBD_FAIL_MDS_CLOSE_NET 0x115
175 echo "waiting for multiop $PID"
176 wait $PID || return 2
177 do_facet client munlink $MOUNT/$tfile || return 3
179 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
181 # Bug 113, check that readdir lost recv timeout works.
183 mkdir /mnt/lustre/readdir || return 1
184 touch /mnt/lustre/readdir/newentry || return
185 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
186 do_facet mds "sysctl -w lustre.fail_loc=0x80000104"
187 ls /mnt/lustre/readdir || return 3
188 do_facet mds "sysctl -w lustre.fail_loc=0"
189 rm -rf /mnt/lustre/readdir || return 4
191 run_test 13 "mdc_readpage restart test (bug 1138)"
193 # Bug 113, check that readdir lost send timeout works.
195 mkdir /mnt/lustre/readdir
196 touch /mnt/lustre/readdir/newentry
197 # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
198 do_facet mds "sysctl -w lustre.fail_loc=0x80000106"
199 ls /mnt/lustre/readdir || return 1
200 do_facet mds "sysctl -w lustre.fail_loc=0"
202 run_test 14 "mdc_readpage resend test (bug 1138)"
205 do_facet mds "sysctl -w lustre.fail_loc=0x80000128"
206 touch $DIR/$tfile && return 1
209 run_test 15 "failed open (-ENOMEM)"
211 READ_AHEAD=`cat /proc/fs/lustre/llite/*/max_read_ahead_mb | head -n 1`
213 for f in /proc/fs/lustre/llite/*/max_read_ahead_mb; do
219 for f in /proc/fs/lustre/llite/*/max_read_ahead_mb; do
220 echo $READ_AHEAD > $f
225 do_facet client cp /etc/termcap $MOUNT
229 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
230 sysctl -w lustre.fail_loc=0x80000504
232 # will get evicted here
233 do_facet client "cmp /etc/termcap $MOUNT/termcap" && return 1
234 sysctl -w lustre.fail_loc=0
235 # give recovery a chance to finish (shouldn't take long)
237 do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 2
240 run_test 16 "timeout bulk put, evict client (2732)"
243 # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
244 # client will get evicted here
245 sysctl -w lustre.fail_loc=0x80000503
246 # need to ensure we send an RPC
247 do_facet client cp /etc/termcap $DIR/$tfile
251 sysctl -w lustre.fail_loc=0
252 do_facet client "df $DIR"
254 do_facet client "cmp /etc/termcap $DIR/$tfile" && return 3
255 do_facet client "rm $DIR/$tfile" || return 4
258 run_test 17 "timeout bulk get, evict client (2732)"
261 do_facet client mkdir -p $MOUNT/$tdir
262 f=$MOUNT/$tdir/$tfile
265 pgcache_empty || return 1
268 lfs setstripe $f $((128 * 1024)) 1 1
270 do_facet client cp /etc/termcap $f
272 local osc2_dev=`$LCTL device_list | \
273 awk '(/ost2.*client_facet/){print $4}' `
274 $LCTL --device %$osc2_dev deactivate
275 # my understanding is that there should be nothing in the page
276 # cache after the client reconnects?
278 pgcache_empty || rc=2
279 $LCTL --device %$osc2_dev activate
283 run_test 18a "manual ost invalidate clears page cache immediately"
286 do_facet client mkdir -p $MOUNT/$tdir
287 f=$MOUNT/$tdir/$tfile
288 f2=$MOUNT/$tdir/${tfile}-2
291 pgcache_empty || return 1
293 # shouldn't have to set stripe size of count==1
294 lfs setstripe $f $((128 * 1024)) 0 1
295 lfs setstripe $f2 $((128 * 1024)) 0 1
297 do_facet client cp /etc/termcap $f
299 # just use this write to trigger the client's eviction from the ost
300 # OBD_FAIL_PTLRPC_BULK_GET_NET|OBD_FAIL_ONCE
301 sysctl -w lustre.fail_loc=0x80000503
302 do_facet client dd if=/dev/zero of=$f2 bs=4k count=1
304 sysctl -w lustre.fail_loc=0
305 # allow recovery to complete
306 sleep $((TIMEOUT + 2))
307 # my understanding is that there should be nothing in the page
308 # cache after the client reconnects?
310 pgcache_empty || rc=2
314 run_test 18b "eviction and reconnect clears page cache (2766)"
318 do_facet client mcreate $f || return 1
319 drop_ldlm_cancel "chmod 0777 $f" || echo evicted
321 do_facet client checkstat -v -p 0777 $f || echo evicted
322 # let the client reconnect
324 do_facet client "munlink $f"
326 run_test 19a "test expired_lock_main on mds (2867)"
330 do_facet client multiop $f Ow || return 1
331 do_facet client multiop $f or || return 2
335 do_facet client multiop $f or || return 3
336 drop_ldlm_cancel multiop $f Ow || echo "client evicted, as expected"
338 do_facet client munlink $f || return 4
340 run_test 19b "test expired_lock_main on ost (2867)"
342 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
344 multiop $DIR/$tdir/${tfile} O_wc &
348 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
349 do_facet ost sysctl -w lustre.fail_loc=0x80000308
350 kill -USR1 $MULTI_PID
353 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
355 run_test 20a "ldlm_handle_enqueue error (should return error)"
357 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
359 touch $DIR/$tdir/${tfile}
361 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
362 do_facet ost sysctl -w lustre.fail_loc=0x80000308
363 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
364 error "didn't fail open enqueue" || true
366 run_test 20b "ldlm_handle_enqueue error (should return error)"
368 #b_cray run_test 21a "drop close request while close and open are both in flight"
369 #b_cray run_test 21b "drop open request while close and open are both in flight"
370 #b_cray run_test 21c "drop both request while close and open are both in flight"
371 #b_cray run_test 21d "drop close reply while close and open are both in flight"
372 #b_cray run_test 21e "drop open reply while close and open are both in flight"
373 #b_cray run_test 21f "drop both reply while close and open are both in flight"
374 #b_cray run_test 21g "drop open reply and close request while close and open are both in flight"
375 #b_cray run_test 21h "drop open request and close reply while close and open are both in flight"
376 #b_cray run_test 22 "drop close request and do mknod"
377 #b_cray run_test 23 "client hang when close a file after mds crash"
379 test_24() { # bug 2248 - eviction fails writeback but app doesn't see it
382 multiop $DIR/$tdir/$tfile Owy_wyc &
385 # OBD_FAIL_PTLRPC_BULK_GET_NET|OBD_FAIL_ONCE
386 sysctl -w lustre.fail_loc=0x80000503
388 kill -USR1 $MULTI_PID
391 sysctl -w lustre.fail_loc=0x0
393 [ $rc -eq 0 ] && error "multiop didn't fail fsync: rc $rc" || true
395 run_test 24 "fsync error (should return error)"
397 test_26() { # bug 5921 - evict dead exports
398 # this test can only run from a client on a separate node.
399 [ "`lsmod | grep obdfilter`" ] && \
400 echo "skipping test 26 (local OST)" && return
401 [ "`lsmod | grep mds`" ] && \
402 echo "skipping test 26 (local MDS)" && return
403 OST_FILE=/proc/fs/lustre/obdfilter/ost_svc/num_exports
404 OST_EXP="`do_facet ost cat $OST_FILE`"
405 OST_NEXP1=`echo $OST_EXP | cut -d' ' -f2`
406 echo starting with $OST_NEXP1 OST exports
407 # OBD_FAIL_PTLRPC_DROP_RPC 0x505
408 do_facet client sysctl -w lustre.fail_loc=0x505
409 # evictor takes up to 2.25x to evict. But if there's a
410 # race to start the evictor from various obds, the loser
411 # might have to wait for the next ping.
412 echo Waiting for $(($TIMEOUT * 4)) secs
413 sleep $(($TIMEOUT * 4))
414 OST_EXP="`do_facet ost cat $OST_FILE`"
415 OST_NEXP2=`echo $OST_EXP | cut -d' ' -f2`
416 echo ending with $OST_NEXP2 OST exports
417 do_facet client sysctl -w lustre.fail_loc=0x0
418 [ $OST_NEXP1 -le $OST_NEXP2 ] && error "client not evicted"
421 run_test 26 "evict dead exports"
424 [ "`lsmod | grep mds`" ] || \
425 { echo "skipping test 27 (non-local MDS)" && return 0; }
427 writemany -q -a $DIR/$tdir/$tfile 0 5 &
432 #define OBD_FAIL_OSC_SHUTDOWN 0x407
433 sysctl -w lustre.fail_loc=0x80000407
434 # need to wait for reconnect
435 echo -n waiting for fail_loc
436 while [ `sysctl -n lustre.fail_loc` -eq -2147482617 ]; do
442 kill -USR1 $CLIENT_PID
446 run_test 27 "fail LOV while using OSC's"
448 test_28() { # bug 6086 - error adding new clients
449 do_facet client mcreate $MOUNT/$tfile || return 1
450 drop_bl_callback "chmod 0777 $MOUNT/$tfile" || return 2
451 #define OBD_FAIL_MDS_ADD_CLIENT 0x12f
452 do_facet mds sysctl -w lustre.fail_loc=0x8000012f
453 # fail once (evicted), reconnect fail (fail_loc), ok
454 df || (sleep 1; df) || (sleep 1; df) || error "reconnect failed"
456 fail mds # verify MDS last_rcvd can be loaded
458 run_test 28 "handle error adding new clients (bug 6086)"
462 # put a load of file creates/writes/deletes
463 writemany -q $DIR/$tdir/$tfile 0 5 &
465 echo writemany pid $CLIENT_PID
469 # wait for client to reconnect to MDS
474 # client process should see no problems even though MDS went down
476 kill -USR1 $CLIENT_PID
479 echo writemany returned $rc
480 #these may fail because of eviction due to slow AST response.
483 run_test 50 "failover MDS under load"
487 # put a load of file creates/writes/deletes
488 writemany -q $DIR/$tdir/$tfile 0 5 &
493 # failover at various points during recovery
494 SEQ="1 5 10 $(seq $TIMEOUT 5 $(($TIMEOUT+10)))"
495 echo will failover at $SEQ
498 echo failover in $i sec
502 # client process should see no problems even though MDS went down
503 # and recovery was interrupted
505 kill -USR1 $CLIENT_PID
508 echo writemany returned $rc
511 run_test 51 "failover MDS during recovery"
514 do_facet client "writemany -q -a $DIR/$tdir/$tfile 300 5" &
516 echo writemany pid $CLIENT_PID
521 wait $CLIENT_PID || rc=$?
522 # active client process should see an EIO for down OST
523 [ $rc -eq 5 ] && { echo "writemany correctly failed $rc" && return 0; }
524 # but timing or failover setup may allow success
525 [ $rc -eq 0 ] && { echo "writemany succeeded" && return 0; }
526 echo "writemany returned $rc"
534 [ $rc -ne 0 ] && { return $rc; }
535 # wait for client to reconnect to OST
539 [ $rc -ne 0 ] && { return $rc; }
546 run_test 52 "failover OST under load"
549 FORCE=--force $CLEANUP