2 # -*- mode: Bash; tab-width: 4; indent-tabs-mode: t; -*-
3 # vim:shiftwidth=4:softtabstop=4:tabstop=4:
7 LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
10 . $LUSTRE/tests/test-framework.sh
12 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
15 # While we do not use OSTCOUNT=1 setup anymore,
16 # ost1failover_HOST is used
17 #ostfailover_HOST=${ostfailover_HOST:-$ost_HOST}
18 #failover= must be defined in OST_MKFS_OPTIONS if ostfailover_HOST != ost_HOST
20 require_dsh_ost || exit 0
24 ALWAYS_EXCEPT="$REPLAY_OST_SINGLE_EXCEPT"
27 [ "$SLOW" = "no" ] && EXCEPT_SLOW="5"
31 check_and_setup_lustre
33 rm -rf $DIR/[df][0-9]*
35 TDIR=$DIR/d0.${TESTSUITE}
37 $LFS setstripe $TDIR -i 0 -c 1
41 zconf_umount `hostname` $MOUNT -f
42 # needs to run during initial client->OST connection
43 #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
44 do_facet ost1 "lctl set_param fail_loc=0x80000211"
45 zconf_mount `hostname` $MOUNT && df $MOUNT || error "0a mount fail"
47 run_test 0a "target handle mismatch (bug 5317) `date +%H:%M:%S`"
51 cp /etc/profile $TDIR/$tfile
53 diff /etc/profile $TDIR/$tfile
56 run_test 0b "empty replay"
59 date > $TDIR/$tfile || error "error creating $TDIR/$tfile"
61 $CHECKSTAT -t file $TDIR/$tfile || return 1
68 echo "tag-$i" > $TDIR/$tfile-$i || error "create $TDIR/$tfile-$i"
72 grep -q "tag-$i" $TDIR/$tfile-$i || error "grep $TDIR/$tfile-$i"
76 run_test 2 "|x| 10 open(O_CREAT)s"
79 verify=$ROOT/tmp/verify-$$
80 dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile &
84 wait $ddpid || return 1
85 cmp $verify $TDIR/$tfile || return 2
86 rm -f $verify $TDIR/$tfile
88 run_test 3 "Fail OST during write, with verification"
91 verify=$ROOT/tmp/verify-$$
92 dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile
93 # invalidate cache, so that we're reading over the wire
95 cmp $verify $TDIR/$tfile &
98 wait $cmppid || return 1
99 rm -f $verify $TDIR/$tfile
101 run_test 4 "Fail OST during read, with verification"
106 local tmppipe=$TMP/${TESTSUITE}.${TESTNAME}.pipe
109 echo "+ iozone $args"
110 iozone $args > $tmppipe &
114 echo "tmppipe=$tmppipe"
117 # iozone exit code is 0 even if iozone is not completed
118 # need to check iozone output on "complete"
119 local iozonelog=$TMP/${TESTSUITE}.iozone.log
121 cat $tmppipe | while read line ; do
123 echo "$line" >>$iozonelog
129 if ! $(tail -1 $iozonelog | grep -q complete); then
139 [ -z "`which iozone 2> /dev/null`" ] && skip_env "iozone missing" && return 0
141 # striping is -c 1, get min of available
142 local minavail=$(lctl get_param -n osc.*[oO][sS][cC][-_]*.kbytesavail | sort -n | head -1)
143 local size=$(( minavail * 3/4 ))
144 local GB=1048576 # 1048576KB == 1GB
146 if (( size > GB )); then
149 local iozone_opts="-i 0 -i 1 -i 2 -+d -r 4 -s $size -f $TDIR/$tfile"
151 iozone_bg $iozone_opts &
154 echo iozone bg pid=$pid
163 wait_delete_completed_mds
164 [ $rc -eq 0 ] || error "iozone failed"
167 run_test 5 "Fail OST during iozone"
170 calc_osc_kbytes kbytesfree
174 remote_mds_nodsh && skip "remote MDS with nodsh" && return 0
178 sync && sleep 2 && sync # wait for delete thread
180 # wait till space is returned, following
181 # (( $before > $after_dd)) test counting on that
182 wait_mds_ost_sync || return 4
183 wait_destroy_complete || return 5
186 dd if=/dev/urandom bs=4096 count=1280 of=$f || return 28
188 stripe_index=$(lfs getstripe -i $f)
191 sleep 2 # ensure we have a fresh statfs
194 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
195 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
196 after_dd=`kbytesfree`
197 log "before: $before after_dd: $after_dd"
198 (( $before > $after_dd )) || return 1
200 fail ost$((stripe_index + 1))
201 wait_recovery_complete ost$((stripe_index + 1)) ||
202 error "OST$((stripe_index + 1)) recovery not completed"
203 $CHECKSTAT -t file $f && return 2 || true
205 # let the delete happen
206 wait_mds_ost_sync || return 4
207 wait_delete_completed || return 5
209 log "before: $before after: $after"
210 (( $before <= $after + 40 )) || return 3 # take OST logs into account
212 run_test 6 "Fail OST before obd_destroy"
217 sync && sleep 5 && sync # wait for delete thread
219 # wait till space is returned, following
220 # (( $before > $after_dd)) test counting on that
221 wait_mds_ost_sync || return 4
222 wait_destroy_complete || return 5
225 dd if=/dev/urandom bs=4096 count=1280 of=$f || return 4
227 sleep 2 # ensure we have a fresh statfs
229 after_dd=`kbytesfree`
230 log "before: $before after_dd: $after_dd"
231 (( $before > $after_dd )) || return 1
235 wait_recovery_complete ost1 || error "OST recovery not done"
236 $CHECKSTAT -t file $f && return 2 || true
238 # let the delete happen
239 wait_mds_ost_sync || return 4
240 wait_delete_completed || return 5
242 log "before: $before after: $after"
243 (( $before <= $after + 40 )) || return 3 # take OST logs into account
245 run_test 7 "Fail OST before obd_destroy"
248 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
249 { skip "Need MDS version at least 2.3.0"; return; }
250 verify=$ROOT/tmp/verify-$$
251 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
252 error "Create verify file failed"
253 #define OBD_FAIL_OST_DQACQ_NET 0x230
254 do_facet ost1 "lctl set_param fail_loc=0x230"
255 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
257 sleep $TIMEOUT # wait for the io to become redo io
258 if ! ps -p $ddpid > /dev/null 2>&1; then
259 error "redo io finished incorrectly"
262 do_facet ost1 "lctl set_param fail_loc=0"
265 cmp $verify $TDIR/$tfile || return 2
266 rm -f $verify $TDIR/$tfile
267 message=`dmesg | grep "redo for recoverable error -115"`
268 [ -z "$message" ] || error "redo error messages found in dmesg"
270 run_test 8a "Verify redo io: redo io when get -EINPROGRESS error"
273 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
274 { skip "Need MDS version at least 2.3.0"; return; }
275 verify=$ROOT/tmp/verify-$$
276 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
277 error "Create verify file failed"
278 #define OBD_FAIL_OST_DQACQ_NET 0x230
279 do_facet ost1 "lctl set_param fail_loc=0x230"
280 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
282 sleep $TIMEOUT # wait for the io to become redo io
284 do_facet ost1 "lctl set_param fail_loc=0"
285 wait $ddpid || return 1
287 cmp $verify $TDIR/$tfile || return 2
288 rm -f $verify $TDIR/$tfile
290 run_test 8b "Verify redo io: redo io should success after recovery"
293 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
294 { skip "Need MDS version at least 2.3.0"; return; }
295 verify=$ROOT/tmp/verify-$$
296 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
297 error "Create verify file failed"
298 #define OBD_FAIL_OST_DQACQ_NET 0x230
299 do_facet ost1 "lctl set_param fail_loc=0x230"
300 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
302 sleep $TIMEOUT # wait for the io to become redo io
304 # allow recovery to complete
305 sleep $((TIMEOUT + 2))
306 do_facet ost1 "lctl set_param fail_loc=0"
309 cmp $verify $TDIR/$tfile && return 2
310 rm -f $verify $TDIR/$tfile
312 run_test 8c "Verify redo io: redo io should fail after eviction"
315 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
316 { skip "Need MDS version at least 2.3.0"; return; }
317 #define OBD_FAIL_MDS_DQACQ_NET 0x187
318 do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
319 # test the non-intent create path
320 mcreate $TDIR/$tfile &
323 if ! ps -p $cpid > /dev/null 2>&1; then
324 error "mknod finished incorrectly"
327 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
328 wait $cpid || return 2
329 stat $TDIR/$tfile || error "mknod failed"
333 #define OBD_FAIL_MDS_DQACQ_NET 0x187
334 do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
335 # test the intent create path
336 openfile -f O_RDWR:O_CREAT $TDIR/$tfile &
339 if ! ps -p $cpid > /dev/null 2>&1; then
340 error "open finished incorrectly"
343 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
344 wait $cpid || return 4
345 stat $TDIR/$tfile || error "open failed"
347 run_test 8d "Verify redo creation on -EINPROGRESS"
350 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
351 { skip "Need MDS version at least 2.3.0"; return; }
352 sleep 1 # ensure we have a fresh statfs
353 #define OBD_FAIL_OST_STATFS_EINPROGRESS 0x231
354 do_facet ost1 "lctl set_param fail_loc=0x231"
358 if ! ps -p $dfpid > /dev/null 2>&1; then
359 do_facet ost1 "lctl set_param fail_loc=0"
360 error "df shouldn't have completed!"
363 do_facet ost1 "lctl set_param fail_loc=0"
365 run_test 8e "Verify that ptlrpc resends request on -EINPROGRESS"
368 check_and_cleanup_lustre