-#!/bin/sh
+#!/bin/bash
set -e
-LUSTRE=${LUSTRE:-`dirname $0`/..}
+LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
+SETUP=${SETUP:-""}
+CLEANUP=${CLEANUP:-""}
. $LUSTRE/tests/test-framework.sh
-
init_test_env $@
+. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
-. ${CONFIG:=$LUSTRE/tests/cfg/local.sh}
+# While we do not use OSTCOUNT=1 setup anymore,
+# ost1failover_HOST is used
+#ostfailover_HOST=${ostfailover_HOST:-$ost_HOST}
+#failover= must be defined in OST_MKFS_OPTIONS if ostfailover_HOST != ost_HOST
-ostfailover_HOST=${ostfailover_HOST:-$ost_HOST}
+require_dsh_ost || exit 0
-# Skip these tests
-ALWAYS_EXCEPT=""
+# bug number for skipped test:
+ALWAYS_EXCEPT="$REPLAY_OST_SINGLE_EXCEPT"
+# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
-gen_config() {
- rm -f $XMLCONFIG
- add_mds mds --dev $MDSDEV --size $MDSSIZE
- add_lov lov1 mds --stripe_sz $STRIPE_BYTES\
- --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
- add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE --failover
- if [ ! -z "$ostfailover_HOST" ]; then
- add_ostfailover ost --dev $OSTDEV --size $OSTSIZE
- fi
- add_client client mds --lov lov1 --path $MOUNT
-}
-
-cleanup() {
- # make sure we are using the primary MDS, so the config log will
- # be able to clean up properly.
- activeost=`facet_active ost`
- if [ $activeost != "ost" ]; then
- fail ost
- fi
- zconf_umount $MOUNT
- stop mds ${FORCE} $MDSLCONFARGS
- stop ost ${FORCE} --dump cleanup.log
-}
+# bug number for SLOW test:
+# time in minutes: 40"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="5"
-if [ "$ONLY" == "cleanup" ]; then
- sysctl -w portals.debug=0
- cleanup
- exit
+if [ $(facet_fstype $SINGLEMDS) = "zfs" ]; then
+# bug number for slow tests: LU-2887
+# time in minutes: 32 12.5"
+ [ "$SLOW" = "no" ] && EXCEPT_SLOW+=" 8a 8b"
+ if [ $MDSCOUNT -gt 1 ]; then
+# bug number for skipped test:
+ ALWAYS_EXCEPT+=""
+ fi
fi
build_test_filter
-rm -f ostactive
-
-gen_config
+check_and_setup_lustre
+assert_DIR
+rm -rf $DIR/[df][0-9]*
-start ost --reformat $OSTLCONFARGS
-PINGER=`cat /proc/fs/lustre/pinger`
+TDIR=$DIR/d0.${TESTSUITE}
+mkdir -p $TDIR
+$LFS setstripe $TDIR -i 0 -c 1
+$LFS getstripe $TDIR
-if [ "$PINGER" != "on" ]; then
- echo "ERROR: Lustre must be built with --enable-pinger for this test."
- stop ost
- exit 1
-fi
-[ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
-start mds --reformat $MDSLCONFARGS
-zconf_mount $MOUNT
-
-mkdir -p $DIR
+test_0a() {
+ zconf_umount $(hostname) $MOUNT -f
+ # needs to run during initial client->OST connection
+ #define OBD_FAIL_OST_ALL_REPLY_NET 0x211
+ do_facet ost1 "lctl set_param fail_loc=0x80000211"
+ zconf_mount $(hostname) $MOUNT && $LFS df $MOUNT || error "mount fail"
+}
+run_test 0a "target handle mismatch (bug 5317)"
-test_0() {
- replay_barrier ost
- fail ost
- cp /etc/profile $DIR/$tfile
- sync
- diff /etc/profile $DIR/$tfile
+test_0b() {
+ fail ost1
+ cp /etc/profile $TDIR/$tfile
+ sync
+ diff /etc/profile $TDIR/$tfile
+ rm -f $TDIR/$tfile
}
-run_test 0 "empty replay"
+run_test 0b "empty replay"
test_1() {
- replay_barrier ost
- date > $DIR/$tfile
- fail ost
- $CHECKSTAT -t file $DIR/$tfile || return 1
+ date > $TDIR/$tfile || error "error creating $TDIR/$tfile"
+ fail ost1
+ $CHECKSTAT -t file $TDIR/$tfile || error "check for file failed"
+ rm -f $TDIR/$tfile
}
run_test 1 "touch"
test_2() {
- replay_barrier ost
- for i in `seq 10`; do
- echo "tag-$i" > $DIR/$tfile-$i
- done
- fail ost
- for i in `seq 10`; do
- grep -q "tag-$i" $DIR/$tfile-$i || error "f1c-$i"
- done
+ for i in $(seq 10); do
+ echo "tag-$i" > $TDIR/$tfile-$i ||
+ error "create $TDIR/$tfile-$i failed"
+ done
+ fail ost1
+ for i in $(seq 10); do
+ grep -q "tag-$i" $TDIR/$tfile-$i ||
+ error "grep $TDIR/$tfile-$i failed"
+ done
+ rm -f $TDIR/$tfile-*
}
run_test 2 "|x| 10 open(O_CREAT)s"
test_3() {
- verify=$ROOT/tmp/verify-$$
- dd if=/dev/urandom bs=1024 count=5120 | tee $verify > $DIR/$tfile &
- ddpid=$!
- sync &
- fail ost
- wait $ddpid || return 1
- cmp $verify $DIR/$tfile || return 2
- rm $verify
+ verify=$ROOT/tmp/verify-$$
+ dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile &
+ ddpid=$!
+ sync &
+ fail ost1
+ wait $ddpid || error "wait for dd failed"
+ cmp $verify $TDIR/$tfile || error "compare $verify $TDIR/$tfile failed"
+ rm -f $verify $TDIR/$tfile
}
run_test 3 "Fail OST during write, with verification"
test_4() {
- verify=$ROOT/tmp/verify-$$
- dd if=/dev/urandom bs=1024 count=5120 | tee $verify > $DIR/$tfile
- # invalidate cache, so that we're reading over the wire
- for i in /proc/fs/lustre/ldlm/namespaces/OSC_*MNT*; do
- echo -n clear > $i/lru_size
- done
- cmp $verify $DIR/$tfile &
- cmppid=$!
- fail ost
- wait $cmppid || return 1
- rm $verify
+ verify=$ROOT/tmp/verify-$$
+ dd if=/dev/urandom bs=4096 count=1280 | tee $verify > $TDIR/$tfile
+ # invalidate cache, so that we're reading over the wire
+ cancel_lru_locks osc
+ cmp $verify $TDIR/$tfile &
+ cmppid=$!
+ fail ost1
+ wait $cmppid || error "wait on cmp failed"
+ rm -f $verify $TDIR/$tfile
}
run_test 4 "Fail OST during read, with verification"
-equals_msg test complete, cleaning up
-cleanup
+iozone_bg () {
+ local args=$@
+
+ local tmppipe=$TMP/${TESTSUITE}.${TESTNAME}.pipe
+ mkfifo $tmppipe
+
+ echo "+ iozone $args"
+ iozone $args > $tmppipe &
+
+ local pid=$!
+
+ echo "tmppipe=$tmppipe"
+ echo iozone pid=$pid
+
+ # iozone exit code is 0 even if iozone is not completed
+ # need to check iozone output on "complete"
+ local iozonelog=$TMP/${TESTSUITE}.iozone.log
+ rm -f $iozonelog
+ cat $tmppipe | while read line ; do
+ echo "$line"
+ echo "$line" >>$iozonelog
+ done;
+
+ local rc=0
+ wait $pid
+ rc=$?
+ if ! $(tail -1 $iozonelog | grep -q complete); then
+ echo iozone failed!
+ rc=1
+ fi
+ rm -f $tmppipe
+ rm -f $iozonelog
+ return $rc
+}
+
+test_5() {
+ if [ -z "$(which iozone 2> /dev/null)" ]; then
+ skip_env "iozone missing"
+ return 0
+ fi
+
+ # striping is -c 1, get min of available
+ local minavail=$(lctl get_param -n osc.*[oO][sS][cC][-_]*.kbytesavail |
+ sort -n | head -n1)
+ local size=$(( minavail * 3/4 ))
+ local GB=1048576 # 1048576KB == 1GB
+
+ if (( size > GB )); then
+ size=$GB
+ fi
+ # no random I/O (-i 2) as it's very slow with ZFS
+ local iozone_opts="-i 0 -i 1 -+d -r 4 -s $size -f $TDIR/$tfile"
+
+ iozone_bg $iozone_opts &
+ local pid=$!
+
+ echo iozone bg pid=$pid
+
+ sleep 8
+ fail ost1
+ local rc=0
+ wait $pid || error "wait on iozone failed"
+ rc=$?
+ log "iozone rc=$rc"
+ rm -f $TDIR/$tfile
+ wait_delete_completed_mds
+ [ $rc -eq 0 ] || error "iozone failed"
+ return $rc
+}
+run_test 5 "Fail OST during iozone"
+
+kbytesfree() {
+ calc_osc_kbytes kbytesfree
+}
+
+test_6() {
+ remote_mds_nodsh && skip "remote MDS with nodsh" && return 0
+
+ local f=$TDIR/$tfile
+ sync && sleep 5 && sync # wait for delete thread
+
+ # wait till space is returned, following
+ # (( $before > $after_dd)) test counting on that
+ wait_mds_ost_sync || error "first wait_mds_ost_sync failed"
+ wait_destroy_complete || error "first wait_destroy_complete failed"
+
+ local before=$(kbytesfree)
+ dd if=/dev/urandom bs=4096 count=1280 of=$f || error "dd failed"
+ $LFS getstripe $f || error "$LFS getstripe $f failed"
+ local stripe_index=$(lfs getstripe -i $f)
+
+ sync
+ sleep 2 # ensure we have a fresh statfs
+ sync
+
+ #define OBD_FAIL_MDS_REINT_NET_REP 0x119
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
+
+ # retry till statfs returns useful results
+ local after_dd=$(kbytesfree)
+ local i=0
+ while (( $before <= $after_dd && $i < 20 )); do
+ sync
+ sleep 1
+ let ++i
+ after_dd=$(kbytesfree)
+ done
+
+ log "before: $before after_dd: $after_dd took $i seconds"
+ (( $before > $after_dd )) ||
+ error "space grew after dd: before:$before after_dd:$after_dd"
+ rm -f $f
+ fail ost$((stripe_index + 1))
+ wait_recovery_complete ost$((stripe_index + 1)) ||
+ error "OST$((stripe_index + 1)) recovery not completed"
+ $CHECKSTAT -t file $f && return 2 || true
+ sync
+ # let the delete happen
+ wait_mds_ost_sync || error "second wait_mds_ost_sync failed"
+ wait_delete_completed || error "second wait_delete_completed failed"
+ local after=$(kbytesfree)
+ log "before: $before after: $after"
+ (( $before <= $after + $(fs_log_size) )) ||
+ error "$before > $after + logsize $(fs_log_size)"
+}
+run_test 6 "Fail OST before obd_destroy"
+
+test_7() {
+ local f=$TDIR/$tfile
+ sync && sleep 5 && sync # wait for delete thread
+
+ # wait till space is returned, following
+ # (( $before > $after_dd)) test counting on that
+ wait_mds_ost_sync || error "wait_mds_ost_sync failed"
+ wait_destroy_complete || error "wait_destroy_complete failed"
+
+ local before=$(kbytesfree)
+ dd if=/dev/urandom bs=4096 count=1280 of=$f ||
+ error "dd to file failed: $?"
+
+ sync
+ local after_dd=$(kbytesfree)
+ local i=0
+ while (( $before <= $after_dd && $i < 10 )); do
+ sync
+ sleep 1
+ let ++i
+ after_dd=$(kbytesfree)
+ done
+
+ log "before: $before after_dd: $after_dd took $i seconds"
+ (( $before > $after_dd )) ||
+ error "space grew after dd: before:$before after_dd:$after_dd"
+ replay_barrier ost1
+ rm -f $f
+ fail ost1
+ wait_recovery_complete ost1 || error "OST recovery not done"
+ $CHECKSTAT -t file $f && return 2 || true
+ sync
+ # let the delete happen
+ wait_mds_ost_sync || error "wait_mds_ost_sync failed"
+ wait_delete_completed || error "wait_delete_completed failed"
+ local after=$(kbytesfree)
+ log "before: $before after: $after"
+ (( $before <= $after + $(fs_log_size) )) ||
+ error "$before > $after + logsize $(fs_log_size)"
+}
+run_test 7 "Fail OST before obd_destroy"
+
+test_8a() {
+ [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
+ { skip "Need MDS version at least 2.3.0"; return; }
+ verify=$ROOT/tmp/verify-$$
+ dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
+ error "Create verify file failed"
+ #define OBD_FAIL_OST_DQACQ_NET 0x230
+ do_facet ost1 "lctl set_param fail_loc=0x230"
+ dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
+ ddpid=$!
+ sleep $TIMEOUT # wait for the io to become redo io
+ if ! ps -p $ddpid > /dev/null 2>&1; then
+ error "redo io finished incorrectly"
+ fi
+ do_facet ost1 "lctl set_param fail_loc=0"
+ wait $ddpid || true
+ cancel_lru_locks osc
+ cmp $verify $TDIR/$tfile || error "compare $verify $TDIR/$tfile failed"
+ rm -f $verify $TDIR/$tfile
+ message=$(dmesg | grep "redo for recoverable error -115")
+ [ -z "$message" ] || error "redo error messages found in dmesg"
+}
+run_test 8a "Verify redo io: redo io when get -EINPROGRESS error"
+
+test_8b() {
+ [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
+ { skip "Need MDS version at least 2.3.0"; return; }
+ verify=$ROOT/tmp/verify-$$
+ dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
+ error "Create verify file failed"
+ #define OBD_FAIL_OST_DQACQ_NET 0x230
+ do_facet ost1 "lctl set_param fail_loc=0x230"
+ dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
+ ddpid=$!
+ sleep $TIMEOUT # wait for the io to become redo io
+ fail ost1
+ do_facet ost1 "lctl set_param fail_loc=0"
+ wait $ddpid || error "dd did not complete"
+ cancel_lru_locks osc
+ cmp $verify $TDIR/$tfile || error "compare $verify $TDIR/$tfile failed"
+ rm -f $verify $TDIR/$tfile
+}
+run_test 8b "Verify redo io: redo io should success after recovery"
+
+test_8c() {
+ [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
+ { skip "Need MDS version at least 2.3.0"; return; }
+ verify=$ROOT/tmp/verify-$$
+ dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
+ error "Create verify file failed"
+ #define OBD_FAIL_OST_DQACQ_NET 0x230
+ do_facet ost1 "lctl set_param fail_loc=0x230"
+ dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
+ ddpid=$!
+ sleep $TIMEOUT # wait for the io to become redo io
+ ost_evict_client
+ # allow recovery to complete
+ sleep $((TIMEOUT + 2))
+ do_facet ost1 "lctl set_param fail_loc=0"
+ wait $ddpid
+ cancel_lru_locks osc
+ cmp $verify $TDIR/$tfile && error "compare files should fail"
+ rm -f $verify $TDIR/$tfile
+}
+run_test 8c "Verify redo io: redo io should fail after eviction"
+
+test_8d() {
+ [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
+ { skip "Need MDS version at least 2.3.0"; return; }
+ #define OBD_FAIL_MDS_DQACQ_NET 0x187
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
+ # test the non-intent create path
+ mcreate $TDIR/$tfile &
+ cpid=$!
+ sleep $TIMEOUT
+ if ! ps -p $cpid > /dev/null 2>&1; then
+ error "mknod finished incorrectly"
+ fi
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+ wait $cpid || error "mcreate did not complete"
+ stat $TDIR/$tfile || error "mknod failed"
+
+ rm $TDIR/$tfile
+
+ #define OBD_FAIL_MDS_DQACQ_NET 0x187
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x187"
+ # test the intent create path
+ openfile -f O_RDWR:O_CREAT $TDIR/$tfile &
+ cpid=$!
+ sleep $TIMEOUT
+ if ! ps -p $cpid > /dev/null 2>&1; then
+ error "open finished incorrectly"
+ fi
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0"
+ wait $cpid || error "openfile failed"
+ stat $TDIR/$tfile || error "open failed"
+}
+run_test 8d "Verify redo creation on -EINPROGRESS"
+
+test_8e() {
+ [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.0) ]] ||
+ { skip "Need MDS version at least 2.3.0"; return; }
+ sleep 1 # ensure we have a fresh statfs
+ #define OBD_FAIL_OST_STATFS_EINPROGRESS 0x231
+ do_facet ost1 "lctl set_param fail_loc=0x231"
+ $LFS df $MOUNT &
+ dfpid=$!
+ sleep $TIMEOUT
+ if ! ps -p $dfpid > /dev/null 2>&1; then
+ do_facet ost1 "lctl set_param fail_loc=0"
+ error "df shouldn't have completed!"
+ fi
+}
+run_test 8e "Verify that ptlrpc resends request on -EINPROGRESS"
+
+test_9() {
+ [ $(lustre_version_code ost1) -ge $(version_code 2.6.54) ] ||
+ { skip "Need OST version at least 2.6.54"; return; }
+ $LFS setstripe -i 0 -c 1 $DIR/$tfile || error "setstripe failed"
+
+ # LU-1573 - Add duplicate write to generate grants
+ dd if=/dev/zero of=$DIR/$tfile count=1 bs=1M > /dev/null ||
+ error "First write failed"
+ replay_barrier ost1
+ # do IO
+ dd if=/dev/zero of=$DIR/$tfile count=1 bs=1M > /dev/null ||
+ error "failed to write"
+ # failover, replay and resend replayed waiting request
+ #define OBD_FAIL_TGT_REPLAY_DELAY2 0x714
+ do_facet ost1 $LCTL set_param fail_loc=0x00000714
+ do_facet ost1 $LCTL set_param fail_val=$TIMEOUT
+ fail ost1
+ do_facet ost1 $LCTL set_param fail_loc=0
+ do_facet ost1 "dmesg | tail -n 100" |
+ sed -n '/no req deadline/,$ p' | grep -q 'Already past' &&
+ return 1
+ return 0
+}
+run_test 9 "Verify that no req deadline happened during recovery"
+
+test_10() {
+ rm -f $TDIR/$tfile
+
+ dd if=/dev/zero of=$TDIR/$tfile count=10 || error "dd failed"
+
+ #define OBD_FAIL_OSC_DELAY_IO 0x414
+ $LCTL set_param fail_val=60 fail_loc=0x414
+ cancel_lru_locks OST0000-osc &
+ sleep 2
+ facet_failover ost1 || error "failover: $?"
+
+ #define OBD_FAIL_LDLM_GRANT_CHECK 0x32a
+ $LCTL set_param fail_loc=0x32a
+ stat $TDIR/$tfile
+
+ wait
+}
+run_test 10 "conflicting PW & PR locks on a client"
+
+test_12() {
+ [ $FAILURE_MODE != "HARD" ] &&
+ skip "Test needs FAILURE_MODE HARD" && return 0
+ remote_ost || { skip "need remote OST" && return 0; }
+
+ local tmp=$TMP/$tdir
+ local dir=$DIR/$tdir
+ declare -a pids
+
+
+ mkdir -p $tmp || error "can't create $tmp"
+ mkdir -p $dir || error "can't create $dir"
+
+ $LFS setstripe -c 1 -i 0 $dir
+
+ for i in `seq 1 10`; do mkdir $dir/d$i; done
+
+ #define OBD_FAIL_OST_DELAY_TRANS 0x245
+ do_facet ost1 "$LCTL set_param fail_loc=0x245" ||
+ error "can't set fail_loc"
+
+ for i in `seq 1 10`;
+ do
+ createmany -o $dir/d$i/$(openssl rand -base64 12) 500 &
+ pids+=($!)
+ done
+ echo "Waiting createmany pids"
+ wait ${pids[@]}
+
+ ls -lR $dir > $tmp/ls_r_out 2>&1&
+ local ls_pid=$!
+
+ facet_failover ost1
+
+ echo "starting wait for ls -l"
+ wait $ls_pid
+ grep "?\|No such file or directory" $tmp/ls_r_out &&
+ error "Found file without object on OST"
+ rm -rf $tmp
+ rm -rf $dir
+}
+run_test 12 "check stat after OST failover"
+
+complete $SECONDS
+check_and_cleanup_lustre
+exit_status
git://git.whamcloud.com / fs / lustre-release.git / blobdiff