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 [ $rc -eq 0 ] || error "iozone failed"
166 run_test 5 "Fail OST during iozone"
169 calc_osc_kbytes kbytesfree
173 remote_mds_nodsh && skip "remote MDS with nodsh" && return 0
177 sync && sleep 2 && sync # wait for delete thread
179 # wait till space is returned, following
180 # (( $before > $after_dd)) test counting on that
181 wait_mds_ost_sync || return 4
182 wait_destroy_complete || return 5
185 dd if=/dev/urandom bs=4096 count=1280 of=$f || return 28
187 stripe_index=$(lfs getstripe -i $f)
190 sleep 2 # ensure we have a fresh statfs
193 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
194 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
195 after_dd=`kbytesfree`
196 log "before: $before after_dd: $after_dd"
197 (( $before > $after_dd )) || return 1
199 fail ost$((stripe_index + 1))
200 wait_recovery_complete ost$((stripe_index + 1)) ||
201 error "OST$((stripe_index + 1)) recovery not completed"
202 $CHECKSTAT -t file $f && return 2 || true
204 # let the delete happen
205 wait_mds_ost_sync || return 4
206 wait_delete_completed || return 5
208 log "before: $before after: $after"
209 (( $before <= $after + 40 )) || return 3 # take OST logs into account
211 run_test 6 "Fail OST before obd_destroy"
216 sync && sleep 5 && sync # wait for delete thread
218 # wait till space is returned, following
219 # (( $before > $after_dd)) test counting on that
220 wait_mds_ost_sync || return 4
221 wait_destroy_complete || return 5
224 dd if=/dev/urandom bs=4096 count=1280 of=$f || return 4
226 sleep 2 # ensure we have a fresh statfs
228 after_dd=`kbytesfree`
229 log "before: $before after_dd: $after_dd"
230 (( $before > $after_dd )) || return 1
234 wait_recovery_complete ost1 || error "OST recovery not done"
235 $CHECKSTAT -t file $f && return 2 || true
237 # let the delete happen
238 wait_mds_ost_sync || return 4
239 wait_delete_completed || return 5
241 log "before: $before after: $after"
242 (( $before <= $after + 40 )) || return 3 # take OST logs into account
244 run_test 7 "Fail OST before obd_destroy"
247 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
248 { skip "Need MDS version at least 2.3.0"; return; }
249 verify=$ROOT/tmp/verify-$$
250 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
251 error "Create verify file failed"
252 #define OBD_FAIL_OST_DQACQ_NET 0x230
253 do_facet ost1 "lctl set_param fail_loc=0x230"
254 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
256 sleep $TIMEOUT # wait for the io to become redo io
257 if ! ps -p $ddpid > /dev/null 2>&1; then
258 error "redo io finished incorrectly"
261 do_facet ost1 "lctl set_param fail_loc=0"
264 cmp $verify $TDIR/$tfile || return 2
265 rm -f $verify $TDIR/$tfile
266 message=`dmesg | grep "redo for recoverable error -115"`
267 [ -z "$message" ] || error "redo error messages found in dmesg"
269 run_test 8a "Verify redo io: redo io when get -EINPROGRESS error"
272 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
273 { skip "Need MDS version at least 2.3.0"; return; }
274 verify=$ROOT/tmp/verify-$$
275 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
276 error "Create verify file failed"
277 #define OBD_FAIL_OST_DQACQ_NET 0x230
278 do_facet ost1 "lctl set_param fail_loc=0x230"
279 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
281 sleep $TIMEOUT # wait for the io to become redo io
283 do_facet ost1 "lctl set_param fail_loc=0"
284 wait $ddpid || return 1
286 cmp $verify $TDIR/$tfile || return 2
287 rm -f $verify $TDIR/$tfile
289 run_test 8b "Verify redo io: redo io should success after recovery"
292 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
293 { skip "Need MDS version at least 2.3.0"; return; }
294 verify=$ROOT/tmp/verify-$$
295 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
296 error "Create verify file failed"
297 #define OBD_FAIL_OST_DQACQ_NET 0x230
298 do_facet ost1 "lctl set_param fail_loc=0x230"
299 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
301 sleep $TIMEOUT # wait for the io to become redo io
303 # allow recovery to complete
304 sleep $((TIMEOUT + 2))
305 do_facet ost1 "lctl set_param fail_loc=0"
308 cmp $verify $TDIR/$tfile && return 2
309 rm -f $verify $TDIR/$tfile
311 run_test 8c "Verify redo io: redo io should fail after eviction"
314 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
315 { skip "Need MDS version at least 2.3.0"; return; }
316 #define OBD_FAIL_MDS_DQACQ_NET 0x187
317 do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
318 # test the non-intent create path
319 mcreate $TDIR/$tfile &
322 if ! ps -p $cpid > /dev/null 2>&1; then
323 error "mknod finished incorrectly"
326 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
327 wait $cpid || return 2
328 stat $TDIR/$tfile || error "mknod failed"
332 #define OBD_FAIL_MDS_DQACQ_NET 0x187
333 do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
334 # test the intent create path
335 openfile -f O_RDWR:O_CREAT $TDIR/$tfile &
338 if ! ps -p $cpid > /dev/null 2>&1; then
339 error "open finished incorrectly"
342 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
343 wait $cpid || return 4
344 stat $TDIR/$tfile || error "open failed"
346 run_test 8d "Verify redo creation on -EINPROGRESS"
349 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
350 { skip "Need MDS version at least 2.3.0"; return; }
351 sleep 1 # ensure we have a fresh statfs
352 #define OBD_FAIL_OST_STATFS_EINPROGRESS 0x231
353 do_facet ost1 "lctl set_param fail_loc=0x231"
357 if ! ps -p $dfpid > /dev/null 2>&1; then
358 do_facet ost1 "lctl set_param fail_loc=0"
359 error "df shouldn't have completed!"
362 do_facet ost1 "lctl set_param fail_loc=0"
364 run_test 8e "Verify that ptlrpc resends request on -EINPROGRESS"
367 check_and_cleanup_lustre