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 [ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
30 # bug number for skipped test: LU-2285
31 ALWAYS_EXCEPT="$ALWAYS_EXCEPT 3"
35 check_and_setup_lustre
37 rm -rf $DIR/[df][0-9]*
39 TDIR=$DIR/d0.${TESTSUITE}
41 $LFS setstripe $TDIR -i 0 -c 1
45 zconf_umount `hostname` $MOUNT -f
46 # needs to run during initial client->OST connection
47 #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
48 do_facet ost1 "lctl set_param fail_loc=0x80000211"
49 zconf_mount `hostname` $MOUNT && df $MOUNT || error "0a mount fail"
51 run_test 0a "target handle mismatch (bug 5317) `date +%H:%M:%S`"
55 cp /etc/profile $TDIR/$tfile
57 diff /etc/profile $TDIR/$tfile
60 run_test 0b "empty replay"
63 date > $TDIR/$tfile || error "error creating $TDIR/$tfile"
65 $CHECKSTAT -t file $TDIR/$tfile || return 1
72 echo "tag-$i" > $TDIR/$tfile-$i || error "create $TDIR/$tfile-$i"
76 grep -q "tag-$i" $TDIR/$tfile-$i || error "grep $TDIR/$tfile-$i"
80 run_test 2 "|x| 10 open(O_CREAT)s"
83 verify=$ROOT/tmp/verify-$$
84 dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile &
88 wait $ddpid || return 1
89 cmp $verify $TDIR/$tfile || return 2
90 rm -f $verify $TDIR/$tfile
92 run_test 3 "Fail OST during write, with verification"
95 verify=$ROOT/tmp/verify-$$
96 dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile
97 # invalidate cache, so that we're reading over the wire
99 cmp $verify $TDIR/$tfile &
102 wait $cmppid || return 1
103 rm -f $verify $TDIR/$tfile
105 run_test 4 "Fail OST during read, with verification"
110 local tmppipe=$TMP/${TESTSUITE}.${TESTNAME}.pipe
113 echo "+ iozone $args"
114 iozone $args > $tmppipe &
118 echo "tmppipe=$tmppipe"
121 # iozone exit code is 0 even if iozone is not completed
122 # need to check iozone output on "complete"
123 local iozonelog=$TMP/${TESTSUITE}.iozone.log
125 cat $tmppipe | while read line ; do
127 echo "$line" >>$iozonelog
133 if ! $(tail -1 $iozonelog | grep -q complete); then
143 [ -z "`which iozone 2> /dev/null`" ] && skip_env "iozone missing" && return 0
145 # striping is -c 1, get min of available
146 local minavail=$(lctl get_param -n osc.*[oO][sS][cC][-_]*.kbytesavail | sort -n | head -1)
147 local size=$(( minavail * 3/4 ))
148 local GB=1048576 # 1048576KB == 1GB
150 if (( size > GB )); then
153 local iozone_opts="-i 0 -i 1 -i 2 -+d -r 4 -s $size -f $TDIR/$tfile"
155 iozone_bg $iozone_opts &
158 echo iozone bg pid=$pid
167 wait_delete_completed_mds
168 [ $rc -eq 0 ] || error "iozone failed"
171 run_test 5 "Fail OST during iozone"
174 calc_osc_kbytes kbytesfree
178 remote_mds_nodsh && skip "remote MDS with nodsh" && return 0
182 sync && sleep 2 && sync # wait for delete thread
184 # wait till space is returned, following
185 # (( $before > $after_dd)) test counting on that
186 wait_mds_ost_sync || return 4
187 wait_destroy_complete || return 5
190 dd if=/dev/urandom bs=4096 count=1280 of=$f || return 28
192 stripe_index=$(lfs getstripe -i $f)
195 sleep 2 # ensure we have a fresh statfs
198 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
199 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
200 after_dd=`kbytesfree`
201 log "before: $before after_dd: $after_dd"
202 (( $before > $after_dd )) || return 1
204 fail ost$((stripe_index + 1))
205 wait_recovery_complete ost$((stripe_index + 1)) ||
206 error "OST$((stripe_index + 1)) recovery not completed"
207 $CHECKSTAT -t file $f && return 2 || true
209 # let the delete happen
210 wait_mds_ost_sync || return 4
211 wait_delete_completed || return 5
213 log "before: $before after: $after"
214 (( $before <= $after + 40 )) || return 3 # take OST logs into account
216 run_test 6 "Fail OST before obd_destroy"
221 sync && sleep 5 && sync # wait for delete thread
223 # wait till space is returned, following
224 # (( $before > $after_dd)) test counting on that
225 wait_mds_ost_sync || return 4
226 wait_destroy_complete || return 5
229 dd if=/dev/urandom bs=4096 count=1280 of=$f || return 4
231 sleep 2 # ensure we have a fresh statfs
233 after_dd=`kbytesfree`
234 log "before: $before after_dd: $after_dd"
235 (( $before > $after_dd )) || return 1
239 wait_recovery_complete ost1 || error "OST recovery not done"
240 $CHECKSTAT -t file $f && return 2 || true
242 # let the delete happen
243 wait_mds_ost_sync || return 4
244 wait_delete_completed || return 5
246 log "before: $before after: $after"
247 (( $before <= $after + 40 )) || return 3 # take OST logs into account
249 run_test 7 "Fail OST before obd_destroy"
252 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
253 { skip "Need MDS version at least 2.3.0"; return; }
254 verify=$ROOT/tmp/verify-$$
255 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
256 error "Create verify file failed"
257 #define OBD_FAIL_OST_DQACQ_NET 0x230
258 do_facet ost1 "lctl set_param fail_loc=0x230"
259 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
261 sleep $TIMEOUT # wait for the io to become redo io
262 if ! ps -p $ddpid > /dev/null 2>&1; then
263 error "redo io finished incorrectly"
266 do_facet ost1 "lctl set_param fail_loc=0"
269 cmp $verify $TDIR/$tfile || return 2
270 rm -f $verify $TDIR/$tfile
271 message=`dmesg | grep "redo for recoverable error -115"`
272 [ -z "$message" ] || error "redo error messages found in dmesg"
274 run_test 8a "Verify redo io: redo io when get -EINPROGRESS error"
277 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
278 { skip "Need MDS version at least 2.3.0"; return; }
279 verify=$ROOT/tmp/verify-$$
280 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
281 error "Create verify file failed"
282 #define OBD_FAIL_OST_DQACQ_NET 0x230
283 do_facet ost1 "lctl set_param fail_loc=0x230"
284 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
286 sleep $TIMEOUT # wait for the io to become redo io
288 do_facet ost1 "lctl set_param fail_loc=0"
289 wait $ddpid || return 1
291 cmp $verify $TDIR/$tfile || return 2
292 rm -f $verify $TDIR/$tfile
294 run_test 8b "Verify redo io: redo io should success after recovery"
297 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
298 { skip "Need MDS version at least 2.3.0"; return; }
299 verify=$ROOT/tmp/verify-$$
300 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
301 error "Create verify file failed"
302 #define OBD_FAIL_OST_DQACQ_NET 0x230
303 do_facet ost1 "lctl set_param fail_loc=0x230"
304 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
306 sleep $TIMEOUT # wait for the io to become redo io
308 # allow recovery to complete
309 sleep $((TIMEOUT + 2))
310 do_facet ost1 "lctl set_param fail_loc=0"
313 cmp $verify $TDIR/$tfile && return 2
314 rm -f $verify $TDIR/$tfile
316 run_test 8c "Verify redo io: redo io should fail after eviction"
319 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
320 { skip "Need MDS version at least 2.3.0"; return; }
321 #define OBD_FAIL_MDS_DQACQ_NET 0x187
322 do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
323 # test the non-intent create path
324 mcreate $TDIR/$tfile &
327 if ! ps -p $cpid > /dev/null 2>&1; then
328 error "mknod finished incorrectly"
331 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
332 wait $cpid || return 2
333 stat $TDIR/$tfile || error "mknod failed"
337 #define OBD_FAIL_MDS_DQACQ_NET 0x187
338 do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
339 # test the intent create path
340 openfile -f O_RDWR:O_CREAT $TDIR/$tfile &
343 if ! ps -p $cpid > /dev/null 2>&1; then
344 error "open finished incorrectly"
347 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
348 wait $cpid || return 4
349 stat $TDIR/$tfile || error "open failed"
351 run_test 8d "Verify redo creation on -EINPROGRESS"
354 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
355 { skip "Need MDS version at least 2.3.0"; return; }
356 sleep 1 # ensure we have a fresh statfs
357 #define OBD_FAIL_OST_STATFS_EINPROGRESS 0x231
358 do_facet ost1 "lctl set_param fail_loc=0x231"
362 if ! ps -p $dfpid > /dev/null 2>&1; then
363 do_facet ost1 "lctl set_param fail_loc=0"
364 error "df shouldn't have completed!"
367 do_facet ost1 "lctl set_param fail_loc=0"
369 run_test 8e "Verify that ptlrpc resends request on -EINPROGRESS"
372 check_and_cleanup_lustre