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-2887
33 [ "$SLOW" = "no" ] && EXCEPT_SLOW="$EXCEPT_SLOW 8a 8b"
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 $iozonelog | grep -q complete); then
146 if [ -z "`which iozone 2> /dev/null`" ]; then
147 skip_env "iozone missing"
151 # striping is -c 1, get min of available
152 local minavail=$(lctl get_param -n osc.*[oO][sS][cC][-_]*.kbytesavail |
154 local size=$(( minavail * 3/4 ))
155 local GB=1048576 # 1048576KB == 1GB
157 if (( size > GB )); then
160 local iozone_opts="-i 0 -i 1 -i 2 -+d -r 4 -s $size -f $TDIR/$tfile"
162 iozone_bg $iozone_opts &
165 echo iozone bg pid=$pid
174 wait_delete_completed_mds
175 [ $rc -eq 0 ] || error "iozone failed"
178 run_test 5 "Fail OST during iozone"
181 calc_osc_kbytes kbytesfree
185 remote_mds_nodsh && skip "remote MDS with nodsh" && return 0
189 sync && sleep 5 && sync # wait for delete thread
191 # wait till space is returned, following
192 # (( $before > $after_dd)) test counting on that
193 wait_mds_ost_sync || return 4
194 wait_destroy_complete || return 5
196 local before=$(kbytesfree)
197 dd if=/dev/urandom bs=4096 count=1280 of=$f || return 28
199 local stripe_index=$(lfs getstripe -i $f)
202 sleep 2 # ensure we have a fresh statfs
205 #define OBD_FAIL_MDS_REINT_NET_REP 0x119
206 do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
208 # retry till statfs returns useful results
209 local after_dd=$(kbytesfree)
211 while (( $before <= $after_dd && $i < 20 )); do
215 after_dd=$(kbytesfree)
218 log "before: $before after_dd: $after_dd took $i seconds"
219 (( $before > $after_dd )) ||
220 error "space grew after dd: before:$before after_dd:$after_dd"
222 fail ost$((stripe_index + 1))
223 wait_recovery_complete ost$((stripe_index + 1)) ||
224 error "OST$((stripe_index + 1)) recovery not completed"
225 $CHECKSTAT -t file $f && return 2 || true
227 # let the delete happen
228 wait_mds_ost_sync || return 4
229 wait_delete_completed || return 5
230 local after=$(kbytesfree)
231 log "before: $before after: $after"
232 (( $before <= $after + $(fs_log_size) )) ||
233 error "$before > $after + logsize $(fs_log_size)"
235 run_test 6 "Fail OST before obd_destroy"
240 sync && sleep 5 && sync # wait for delete thread
242 # wait till space is returned, following
243 # (( $before > $after_dd)) test counting on that
244 wait_mds_ost_sync || return 4
245 wait_destroy_complete || return 5
247 local before=$(kbytesfree)
248 dd if=/dev/urandom bs=4096 count=1280 of=$f || error "dd to file failed: $?"
251 local after_dd=$(kbytesfree)
253 while (( $before <= $after_dd && $i < 10 )); do
257 after_dd=$(kbytesfree)
260 log "before: $before after_dd: $after_dd took $i seconds"
261 (( $before > $after_dd )) ||
262 error "space grew after dd: before:$before after_dd:$after_dd"
266 wait_recovery_complete ost1 || error "OST recovery not done"
267 $CHECKSTAT -t file $f && return 2 || true
269 # let the delete happen
270 wait_mds_ost_sync || return 4
271 wait_delete_completed || return 5
272 local after=$(kbytesfree)
273 log "before: $before after: $after"
274 (( $before <= $after + $(fs_log_size) )) ||
275 error "$before > $after + logsize $(fs_log_size)"
277 run_test 7 "Fail OST before obd_destroy"
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-$$
283 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
284 error "Create verify file failed"
285 #define OBD_FAIL_OST_DQACQ_NET 0x230
286 do_facet ost1 "lctl set_param fail_loc=0x230"
287 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
289 sleep $TIMEOUT # wait for the io to become redo io
290 if ! ps -p $ddpid > /dev/null 2>&1; then
291 error "redo io finished incorrectly"
294 do_facet ost1 "lctl set_param fail_loc=0"
297 cmp $verify $TDIR/$tfile || return 2
298 rm -f $verify $TDIR/$tfile
299 message=`dmesg | grep "redo for recoverable error -115"`
300 [ -z "$message" ] || error "redo error messages found in dmesg"
302 run_test 8a "Verify redo io: redo io when get -EINPROGRESS error"
305 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
306 { skip "Need MDS version at least 2.3.0"; return; }
307 verify=$ROOT/tmp/verify-$$
308 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
309 error "Create verify file failed"
310 #define OBD_FAIL_OST_DQACQ_NET 0x230
311 do_facet ost1 "lctl set_param fail_loc=0x230"
312 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
314 sleep $TIMEOUT # wait for the io to become redo io
316 do_facet ost1 "lctl set_param fail_loc=0"
317 wait $ddpid || return 1
319 cmp $verify $TDIR/$tfile || return 2
320 rm -f $verify $TDIR/$tfile
322 run_test 8b "Verify redo io: redo io should success after recovery"
325 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
326 { skip "Need MDS version at least 2.3.0"; return; }
327 verify=$ROOT/tmp/verify-$$
328 dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
329 error "Create verify file failed"
330 #define OBD_FAIL_OST_DQACQ_NET 0x230
331 do_facet ost1 "lctl set_param fail_loc=0x230"
332 dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
334 sleep $TIMEOUT # wait for the io to become redo io
336 # allow recovery to complete
337 sleep $((TIMEOUT + 2))
338 do_facet ost1 "lctl set_param fail_loc=0"
341 cmp $verify $TDIR/$tfile && return 2
342 rm -f $verify $TDIR/$tfile
344 run_test 8c "Verify redo io: redo io should fail after eviction"
347 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
348 { skip "Need MDS version at least 2.3.0"; return; }
349 #define OBD_FAIL_MDS_DQACQ_NET 0x187
350 do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
351 # test the non-intent create path
352 mcreate $TDIR/$tfile &
355 if ! ps -p $cpid > /dev/null 2>&1; then
356 error "mknod finished incorrectly"
359 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
360 wait $cpid || return 2
361 stat $TDIR/$tfile || error "mknod failed"
365 #define OBD_FAIL_MDS_DQACQ_NET 0x187
366 do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
367 # test the intent create path
368 openfile -f O_RDWR:O_CREAT $TDIR/$tfile &
371 if ! ps -p $cpid > /dev/null 2>&1; then
372 error "open finished incorrectly"
375 do_facet $SINGLEMDS "lctl set_param fail_loc=0"
376 wait $cpid || return 4
377 stat $TDIR/$tfile || error "open failed"
379 run_test 8d "Verify redo creation on -EINPROGRESS"
382 [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
383 { skip "Need MDS version at least 2.3.0"; return; }
384 sleep 1 # ensure we have a fresh statfs
385 #define OBD_FAIL_OST_STATFS_EINPROGRESS 0x231
386 do_facet ost1 "lctl set_param fail_loc=0x231"
390 if ! ps -p $dfpid > /dev/null 2>&1; then
391 do_facet ost1 "lctl set_param fail_loc=0"
392 error "df shouldn't have completed!"
395 do_facet ost1 "lctl set_param fail_loc=0"
397 run_test 8e "Verify that ptlrpc resends request on -EINPROGRESS"
400 [ $(lustre_version_code ost1) -ge $(version_code 2.6.54) ] ||
401 { skip "Need OST version at least 2.6.54"; return; }
402 $SETSTRIPE -i 0 -c 1 $DIR/$tfile
405 dd if=/dev/zero of=$DIR/$tfile count=1 bs=1M > /dev/null ||
406 error "failed to write"
407 # failover, replay and resend replayed waiting request
408 #define OBD_FAIL_TGT_REPLAY_DELAY2 0x714
409 do_facet ost1 $LCTL set_param fail_loc=0x00000714
410 do_facet ost1 $LCTL set_param fail_val=$TIMEOUT
412 do_facet ost1 $LCTL set_param fail_loc=0
413 do_facet ost1 "dmesg | tail -n 100" |\
414 sed -n '/no req deadline/,$ p' | grep -q 'Already past' &&
418 run_test 9 "Verify that no req deadline happened during recovery"
421 check_and_cleanup_lustre