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"
29 if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then
30 # bug number for skipped test: LU-2285
31 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 3"
32 # bug number for slowed tests: LU-2874
33 [ "$SLOW" = "no" ] && EXCEPT_SLOW="$EXCEPT_SLOW 8a"
38 check_and_setup_lustre
40 rm -rf $DIR/[df][0-9]*
42 TDIR=$DIR/d0.${TESTSUITE}
44 $LFS setstripe $TDIR -i 0 -c 1
48 zconf_umount `hostname` $MOUNT -f
49 # needs to run during initial client->OST connection
50 #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
51 do_facet ost1 "lctl set_param fail_loc=0x80000211"
52 zconf_mount `hostname` $MOUNT && df $MOUNT || error "0a mount fail"
54 run_test 0a "target handle mismatch (bug 5317) `date +%H:%M:%S`"
58 cp /etc/profile $TDIR/$tfile
60 diff /etc/profile $TDIR/$tfile
63 run_test 0b "empty replay"
66 date > $TDIR/$tfile || error "error creating $TDIR/$tfile"
68 $CHECKSTAT -t file $TDIR/$tfile || return 1
75 echo "tag-$i" > $TDIR/$tfile-$i || error "create $TDIR/$tfile-$i"
79 grep -q "tag-$i" $TDIR/$tfile-$i || error "grep $TDIR/$tfile-$i"
83 run_test 2 "|x| 10 open(O_CREAT)s"
86 verify=$ROOT/tmp/verify-$$
87 dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile &
91 wait $ddpid || return 1
92 cmp $verify $TDIR/$tfile || return 2
93 rm -f $verify $TDIR/$tfile
95 run_test 3 "Fail OST during write, with verification"
98 verify=$ROOT/tmp/verify-$$
99 dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile
100 # invalidate cache, so that we're reading over the wire
102 cmp $verify $TDIR/$tfile &
105 wait $cmppid || return 1
106 rm -f $verify $TDIR/$tfile
108 run_test 4 "Fail OST during read, with verification"
113 local tmppipe=$TMP/${TESTSUITE}.${TESTNAME}.pipe
116 echo "+ iozone $args"
117 iozone $args > $tmppipe &
121 echo "tmppipe=$tmppipe"
124 # iozone exit code is 0 even if iozone is not completed
125 # need to check iozone output on "complete"
126 local iozonelog=$TMP/${TESTSUITE}.iozone.log
128 cat $tmppipe | while read line ; do
130 echo "$line" >>$iozonelog
136 if ! $(tail -1 $iozonelog | grep -q complete); then
146 [ -z "`which iozone 2> /dev/null`" ] && skip_env "iozone missing" && return 0
148 # striping is -c 1, get min of available
149 local minavail=$(lctl get_param -n osc.*[oO][sS][cC][-_]*.kbytesavail | sort -n | head -1)
150 local size=$(( minavail * 3/4 ))
151 local GB=1048576 # 1048576KB == 1GB
153 if (( size > GB )); then
156 local iozone_opts="-i 0 -i 1 -i 2 -+d -r 4 -s $size -f $TDIR/$tfile"
158 iozone_bg $iozone_opts &
161 echo iozone bg pid=$pid
170 wait_delete_completed_mds
171 [ $rc -eq 0 ] || error "iozone failed"
174 run_test 5 "Fail OST during iozone"
177 calc_osc_kbytes kbytesfree
181 remote_mds_nodsh && skip "remote MDS with nodsh" && return 0
185 sync && sleep 2 && sync # wait for delete thread
187 # wait till space is returned, following
188 # (( $before > $after_dd)) test counting on that
189 wait_mds_ost_sync || return 4
190 wait_destroy_complete || return 5
193 dd if=/dev/urandom bs=4096 count=1280 of=$f || return 28
195 stripe_index=$(lfs getstripe -i $f)
198 sleep 2 # ensure we have a fresh statfs
201 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
202 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
203 after_dd=`kbytesfree`
204 log "before: $before after_dd: $after_dd"
205 (( $before > $after_dd )) || return 1
207 fail ost$((stripe_index + 1))
208 wait_recovery_complete ost$((stripe_index + 1)) ||
209 error "OST$((stripe_index + 1)) recovery not completed"
210 $CHECKSTAT -t file $f && return 2 || true
212 # let the delete happen
213 wait_mds_ost_sync || return 4
214 wait_delete_completed || return 5
216 log "before: $before after: $after"
217 (( $before <= $after + 40 )) || return 3 # take OST logs into account
219 run_test 6 "Fail OST before obd_destroy"
224 sync && sleep 5 && sync # wait for delete thread
226 # wait till space is returned, following
227 # (( $before > $after_dd)) test counting on that
228 wait_mds_ost_sync || return 4
229 wait_destroy_complete || return 5
232 dd if=/dev/urandom bs=4096 count=1280 of=$f || return 4
234 sleep 2 # ensure we have a fresh statfs
236 after_dd=`kbytesfree`
237 log "before: $before after_dd: $after_dd"
238 (( $before > $after_dd )) || return 1
242 wait_recovery_complete ost1 || error "OST recovery not done"
243 $CHECKSTAT -t file $f && return 2 || true
245 # let the delete happen
246 wait_mds_ost_sync || return 4
247 wait_delete_completed || return 5
249 log "before: $before after: $after"
250 (( $before <= $after + 40 )) || return 3 # take OST logs into account
252 run_test 7 "Fail OST before obd_destroy"
255 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
256 { skip "Need MDS version at least 2.3.0"; return; }
257 verify=$ROOT/tmp/verify-$$
258 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
259 error "Create verify file failed"
260 #define OBD_FAIL_OST_DQACQ_NET 0x230
261 do_facet ost1 "lctl set_param fail_loc=0x230"
262 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
264 sleep $TIMEOUT # wait for the io to become redo io
265 if ! ps -p $ddpid > /dev/null 2>&1; then
266 error "redo io finished incorrectly"
269 do_facet ost1 "lctl set_param fail_loc=0"
272 cmp $verify $TDIR/$tfile || return 2
273 rm -f $verify $TDIR/$tfile
274 message=`dmesg | grep "redo for recoverable error -115"`
275 [ -z "$message" ] || error "redo error messages found in dmesg"
277 run_test 8a "Verify redo io: redo io when get -EINPROGRESS error"
280 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
281 { skip "Need MDS version at least 2.3.0"; return; }
282 verify=$ROOT/tmp/verify-$$
283 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
284 error "Create verify file failed"
285 #define OBD_FAIL_OST_DQACQ_NET 0x230
286 do_facet ost1 "lctl set_param fail_loc=0x230"
287 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
289 sleep $TIMEOUT # wait for the io to become redo io
291 do_facet ost1 "lctl set_param fail_loc=0"
292 wait $ddpid || return 1
294 cmp $verify $TDIR/$tfile || return 2
295 rm -f $verify $TDIR/$tfile
297 run_test 8b "Verify redo io: redo io should success after recovery"
300 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
301 { skip "Need MDS version at least 2.3.0"; return; }
302 verify=$ROOT/tmp/verify-$$
303 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
304 error "Create verify file failed"
305 #define OBD_FAIL_OST_DQACQ_NET 0x230
306 do_facet ost1 "lctl set_param fail_loc=0x230"
307 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
309 sleep $TIMEOUT # wait for the io to become redo io
311 # allow recovery to complete
312 sleep $((TIMEOUT + 2))
313 do_facet ost1 "lctl set_param fail_loc=0"
316 cmp $verify $TDIR/$tfile && return 2
317 rm -f $verify $TDIR/$tfile
319 run_test 8c "Verify redo io: redo io should fail after eviction"
322 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
323 { skip "Need MDS version at least 2.3.0"; return; }
324 #define OBD_FAIL_MDS_DQACQ_NET 0x187
325 do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
326 # test the non-intent create path
327 mcreate $TDIR/$tfile &
330 if ! ps -p $cpid > /dev/null 2>&1; then
331 error "mknod finished incorrectly"
334 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
335 wait $cpid || return 2
336 stat $TDIR/$tfile || error "mknod failed"
340 #define OBD_FAIL_MDS_DQACQ_NET 0x187
341 do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
342 # test the intent create path
343 openfile -f O_RDWR:O_CREAT $TDIR/$tfile &
346 if ! ps -p $cpid > /dev/null 2>&1; then
347 error "open finished incorrectly"
350 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
351 wait $cpid || return 4
352 stat $TDIR/$tfile || error "open failed"
354 run_test 8d "Verify redo creation on -EINPROGRESS"
357 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
358 { skip "Need MDS version at least 2.3.0"; return; }
359 sleep 1 # ensure we have a fresh statfs
360 #define OBD_FAIL_OST_STATFS_EINPROGRESS 0x231
361 do_facet ost1 "lctl set_param fail_loc=0x231"
365 if ! ps -p $dfpid > /dev/null 2>&1; then
366 do_facet ost1 "lctl set_param fail_loc=0"
367 error "df shouldn't have completed!"
370 do_facet ost1 "lctl set_param fail_loc=0"
372 run_test 8e "Verify that ptlrpc resends request on -EINPROGRESS"
375 check_and_cleanup_lustre