set -e
-# bug 5493 LU2034
-ALWAYS_EXCEPT="52 $RECOVERY_SMALL_EXCEPT"
-
-export MULTIOP=${MULTIOP:-multiop}
PTLDEBUG=${PTLDEBUG:--1}
-LUSTRE=${LUSTRE:-`dirname $0`/..}
+LUSTRE=${LUSTRE:-$(dirname $0)/..}
. $LUSTRE/tests/test-framework.sh
init_test_env $@
-. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
init_logging
-require_dsh_mds || exit 0
-
-# also long tests: 19, 21a, 21e, 21f, 23, 27
-
-[ "$SLOW" = "no" ] && EXCEPT_SLOW=""
+ALWAYS_EXCEPT="$RECOVERY_SMALL_EXCEPT "
+if $SHARED_KEY; then
+ # bug number for skipped test: LU-12896
+ ALWAYS_EXCEPT+=" 110k"
+ # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
+fi
build_test_filter
+require_dsh_mds || exit 0
+
# Allow us to override the setup if we already have a mounted system by
# setting SETUP=" " and CLEANUP=" "
SETUP=${SETUP:-""}
local before=$(date +%s)
local evict
- [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.53) ]] &&
- skip "Need MDS version at least 2.6.53" && return
+ [[ "$MDS1_VERSION" -lt $(version_code 2.6.53) ]] &&
+ skip "Need MDS version at least 2.6.53"
do_facet client "stat $DIR > /dev/null" ||
error "failed to stat $DIR: $?"
drop_bl_callback_once "chmod 0777 $DIR" ||
local before=$(date +%s)
local evict
- [[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ]] &&
- skip "Need MDS version at least 2.6.90" && return
+ [[ "$MDS1_VERSION" -lt $(version_code 2.6.90) ]] &&
+ skip "Need MDS version at least 2.6.90"
# sleep 1 is to make sure that BEFORE is not equal to EVICTED below
sleep 1
test_10e()
{
- [[ $(lustre_version_code ost1) -le $(version_code 2.8.58) ]] &&
- skip "Need OST version at least 2.8.59" && return 0
- [ $CLIENTCOUNT -lt 2 ] && skip "need two clients" && return 0
+ [[ "$OST1_VERSION" -le $(version_code 2.8.58) ]] &&
+ skip "Need OST version at least 2.8.59"
+ [ $CLIENTCOUNT -lt 2 ] && skip "need two clients"
[ $(facet_host client) == $(facet_host ost1) ] &&
- skip "need ost1 and client on different nodes" && return 0
+ skip "need ost1 and client on different nodes"
local -a clients=(${CLIENTS//,/ })
local client1=${clients[0]}
local client2=${clients[1]}
for i in $SEQ
do
#echo failover in $i sec
- log "test_$testnum: failover in $i sec"
+ log "$TESTNAME: failover in $i sec"
sleep $i
facet_failover $SINGLEMDS
done
run_test 53 "touch: drop rep"
test_54() {
- zconf_mount `hostname` $MOUNT2
- touch $DIR/$tfile
- touch $DIR2/$tfile.1
- sleep 10
- cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
- fail $SINGLEMDS
- umount $MOUNT2
- ERROR=`dmesg | egrep "(test 54|went back in time)" | tail -n1 | grep "went back in time"`
- [ x"$ERROR" == x ] || error "back in time occured"
+ zconf_mount $(hostname) $MOUNT2
+ touch $DIR/$tfile
+ touch $DIR2/$tfile.1
+ sleep 10
+ cat $DIR2/$tfile.missing # save transno = 0, rc != 0 into last_rcvd
+ fail $SINGLEMDS
+ umount $MOUNT2
+ ERROR=$(dmesg | egrep "(test 54|went back in time)" | tail -n1 |
+ grep "went back in time")
+ [ x"$ERROR" == x ] || error "back in time occured"
}
run_test 54 "back in time"
# Minimum pass speed is 2MBps
local ddtimeout=64
# LU-2887/LU-3089 - set min pass speed to 500KBps
- [ "$(facet_fstype ost1)" = "zfs" ] && ddtimeout=256
+ [ "$ost1_FSTYPE" = zfs ] && ddtimeout=256
# first dd should be finished quickly
$LFS setstripe -c 1 -i 0 $DIR/$tdir/$tfile-1
mount_client $DIR2
#grant lock1, export2
- $SETSTRIPE -i -0 $DIR2/$tfile || return 1
- $MULTIOP $DIR2/$tfile Ow || return 2
+ $LFS setstripe -i -0 $DIR2/$tfile || error "setstripe failed"
+ $MULTIOP $DIR2/$tfile Ow || error "multiop failed"
#define OBD_FAIL_LDLM_BL_EVICT 0x31e
do_facet ost $LCTL set_param fail_loc=0x31e
test_66()
{
- [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.7.51) ]] ||
- { skip "Need MDS version at least 2.7.51"; return 0; }
+ [[ "$MDS1_VERSION" -ge $(version_code 2.7.51) ]] ||
+ skip "Need MDS version at least 2.7.51"
local list=$(comma_list $(osts_nodes))
do_nodes $list $LCTL set_param fail_loc=0x80000136
#initiate the re-connect & re-send
- local mdccli=$($LCTL dl | awk '/-MDT0000-mdc-/ {print $4;}')
+ local mdtname="MDT0000"
+ local mdccli=$($LCTL dl | grep "${mdtname}-mdc" | awk '{print $4;}')
local conn_uuid=$($LCTL get_param -n mdc.${mdccli}.conn_uuid)
$LCTL set_param "mdc.${mdccli}.import=connection=${conn_uuid}"
sleep 2
# Since the client just mounted, its last_rcvd entry is not on disk.
# Send an RPC so exp_need_sync forces last_rcvd to commit this export
# so the client can reconnect during OST recovery (LU-924, LU-1582)
- $SETSTRIPE -i 0 $DIR/$tfile
+ $LFS setstripe -i 0 $DIR/$tfile
dd if=/dev/zero of=$DIR/$tfile bs=1M count=1 conv=sync
# make sure MGS's state is Partial
}
test_106() { # LU-1789
- [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.50) ]] ||
- { skip "Need MDS version at least 2.3.50"; return 0; }
+ [[ "$MDS1_VERSION" -ge $(version_code 2.3.50) ]] ||
+ skip "Need MDS version at least 2.3.50"
#define OBD_FAIL_MDC_LIGHTWEIGHT 0x805
$LCTL set_param fail_loc=0x805
# lightweight goes through LUSTRE_IMP_RECOVER during failover
touch -c $DIR2/$tfile || true
$LCTL dk $TMP/lustre-log-$TESTNAME.log
- recovered=`awk '/MDT0000-mdc-[0-9a-f]*: lwp recover/ {
- print;
- }' $TMP/lustre-log-$TESTNAME.log`
+ recovered=$(awk '/MDT0000-mdc-[0-9a-f]*. lwp recover/ { print }' \
+ $TMP/lustre-log-$TESTNAME.log)
[ -z "$recovered" ] && error "lightweight client was not recovered"
# and all operations performed by lightweight client should be
test_108() {
mkdir -p $DIR/$tdir
- $SETSTRIPE -c 1 -i 0 $DIR/$tdir
+ $LFS setstripe -c 1 -i 0 $DIR/$tdir
dd if=/dev/zero of=$DIR/$tdir/$tfile bs=1M count=256 &
local dd_pid=$!
drop_update_reply $mdtidx "$LFS mkdir -i $mdtidx -c2 $remote_dir" ||
error "lfs mkdir failed"
- diridx=$($GETSTRIPE -m $remote_dir)
+ diridx=$($LFS getstripe -m $remote_dir)
[ $diridx -eq $mdtidx ] || error "$diridx != $mdtidx"
rm -rf $DIR/$tdir || error "rmdir failed"
run_test 110f "remove remote directory: drop slave rep"
test_110g () {
- [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.11.0) ]] ||
- { skip "Need MDS version at least 2.11.0"; return 0; }
+ [[ "$MDS1_VERSION" -ge $(version_code 2.11.0) ]] ||
+ skip "Need MDS version at least 2.11.0"
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
mkdir -p $DIR/$tdir
touch $DIR/$tdir/$tfile
run_test 110g "drop reply during migration"
test_110h () {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
- local server_version=$(lustre_version_code mds1)
- [[ $server_version -ge $(version_code 2.7.56) ]] ||
- { skip "Need MDS version at least 2.7.56"; return 0; }
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ [[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
+ skip "Need MDS version at least 2.7.56"
local src_dir=$DIR/$tdir/source_dir
local tgt_dir=$DIR/$tdir/target_dir
run_test 110h "drop update reply during cross-MDT file rename"
test_110i () {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
- local server_version=$(lustre_version_code mds1)
- [[ $server_version -ge $(version_code 2.7.56) ]] ||
- { skip "Need MDS version at least 2.7.56"; return 0; }
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ [[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
+ skip "Need MDS version at least 2.7.56"
local src_dir=$DIR/$tdir/source_dir
local tgt_dir=$DIR/$tdir/target_dir
run_test 110i "drop update reply during cross-MDT dir rename"
test_110j () {
- [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
- local server_version=$(lustre_version_code mds1)
- [[ $server_version -ge $(version_code 2.7.56) ]] ||
- { skip "Need MDS version at least 2.7.56"; return 0; }
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs"
+ [[ "$MDS1_VERSION" -ge $(version_code 2.7.56) ]] ||
+ skip "Need MDS version at least 2.7.56"
local remote_dir=$DIR/$tdir/remote_dir
local local_dir=$DIR/$tdir/local_dir
}
run_test 110j "drop update reply during cross-MDT ln"
+test_110k() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTS"
+ [[ "$MDS1_VERSION" -ge $(version_code 2.12.55) ]] ||
+ skip "Need MDS version at least 2.12.55"
+
+ stop mds2 || error "stop mds2 failed"
+ umount $MOUNT
+
+#define OBD_FAIL_FLD_QUERY_REQ 0x1103
+ do_facet mds2 lctl set_param fail_loc=0x1103
+ start mds2 $(mdsdevname 2) -o abort_recovery ||
+ error "start MDS with abort_recovery should succeed"
+ do_facet mds2 lctl set_param fail_loc=0
+
+ # cleanup
+ stop mds2 || error "cleanup: stop mds2 failed"
+ start mds2 $(mdsdevname 2) || error "cleanup: start mds2 failed"
+ zconf_mount $(hostname) $MOUNT || error "cleanup: mount failed"
+ client_up || error "post-failover df failed"
+}
+run_test 110k "FID_QUERY failed during recovery"
+
# LU-2844 mdt prepare fail should not cause umount oops
test_111 ()
{
- [[ $(lustre_version_code $SINGLEMDS) -ge $(version_code 2.3.62) ]] ||
- { skip "Need MDS version at least 2.3.62"; return 0; }
+ [[ "$MDS1_VERSION" -ge $(version_code 2.3.62) ]] ||
+ skip "Need MDS version at least 2.3.62"
#define OBD_FAIL_MDS_CHANGELOG_INIT 0x151
do_facet $SINGLEMDS lctl set_param fail_loc=0x151
}
test_115a() {
- [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
- skip "need at least 2.8.50 on OST" && return 0
+ [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST"
#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
#define OBD_FAIL_PTLRPC_DROP_BULK 0x51a
run_test 115a "read: late REQ MDunlink and no bulk"
test_115b() {
- [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
- skip "need at least 2.8.50 on OST" && return 0
+ [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST"
#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
#define OBD_FAIL_OST_ENOSPC 0x215
run_test 115b "write: late REQ MDunlink and no bulk"
test_115c() {
- [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
- skip "need at least 2.8.50 on OST" && return 0
+ [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST"
#define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
#define OBD_FAIL_PTLRPC_DROP_BULK 0x51a
run_test 115c "read: late Reply MDunlink and no bulk"
test_115d() {
- [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
- skip "need at least 2.8.50 on OST" && return 0
+ [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST"
#define OBD_FAIL_PTLRPC_LONG_REPL_UNLINK 0x50f
#define OBD_FAIL_OST_ENOSPC 0x215
run_test 115d "write: late Reply MDunlink and no bulk"
test_115e() {
- [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
- skip "need at least 2.8.50 on OST" && return 0
+ [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST"
#define OBD_FAIL_PTLRPC_LONG_BULK_UNLINK 0x510
#define OBD_FAIL_OST_ALL_REPLY_NET 0x211
run_test 115e "read: late Bulk MDunlink and no reply"
test_115f() {
- [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
- skip "need at least 2.8.50 on OST" && return 0
+ [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST"
#define OBD_FAIL_PTLRPC_LONG_REQ_UNLINK 0x51b
#define OBD_FAIL_OST_ALL_REPLY_NET 0x211
run_test 115f "read: late REQ MDunlink and no reply"
test_115g() {
- [ $(lustre_version_code ost1) -lt $(version_code 2.8.50) ] &&
- skip "need at least 2.8.50 on OST" && return 0
+ [ "$OST1_VERSION" -lt $(version_code 2.8.50) ] &&
+ skip "need at least 2.8.50 on OST"
#define OBD_FAIL_PTLRPC_LONG_BOTH_UNLINK 0x51c
test_115_read 0x8000051c 0
}
test_130a() {
- remote_mds_nodsh && skip "remote MDS with nodsh" && return
- local server_version=$(lustre_version_code $SINGLEMDS)
- [[ $server_version -ge $(version_code 2.7.2) ]] ||
- { skip "Need server version newer than 2.7.1"; return 0; }
+ remote_mds_nodsh && skip "remote MDS with nodsh"
+ [[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
+ skip "Need server version newer than 2.7.1"
test_130_base
run_test 130a "enqueue resend on not existing file"
test_130b() {
- remote_mds_nodsh && skip "remote MDS with nodsh" && return
- local server_version=$(lustre_version_code $SINGLEMDS)
- [[ $server_version -ge $(version_code 2.7.2) ]] ||
- { skip "Need server version newer than 2.7.1"; return 0; }
+ remote_mds_nodsh && skip "remote MDS with nodsh"
+ [[ "$MDS1_VERSION" -ge $(version_code 2.7.2) ]] ||
+ skip "Need server version newer than 2.7.1"
test_130_base
# let the reply to be dropped
rm -f $DIR/$tfile
# get a lock on client so that export would reach the stale list
- $SETSTRIPE -i 0 $DIR/$tfile || error "setstripe failed"
+ $LFS setstripe -i 0 $DIR/$tfile || error "setstripe failed"
dd if=/dev/zero of=$DIR/$tfile bs=4096 count=1 conv=fsync ||
error "dd failed"
rm -f $DIR/$tfile
# get a lock on client so that export would reach the stale list
- $SETSTRIPE -i 0 $DIR/$tfile || error "setstripe failed"
+ $LFS setstripe -i 0 $DIR/$tfile || error "setstripe failed"
dd if=/dev/zero of=$DIR/$tfile count=1 || error "dd failed"
# another IO under the same lock
run_test 134 "race between failover and search for reply data free slot"
test_135() {
- [ $MDS1_VERSION -lt $(version_code 2.12.51) ] &&
+ [ "$MDS1_VERSION" -lt $(version_code 2.12.51) ] &&
skip "Need MDS version at least 2.12.51"
mkdir -p $DIR/$tdir
run_test 135 "DOM: open/create resend to return size"
test_136() {
- remote_mds_nodsh && skip "remote MDS with nodsh" && return
- [[ $MDS1_VERSION -ge $(version_code 2.12.52) ]] ||
+ remote_mds_nodsh && skip "remote MDS with nodsh"
+ [[ "$MDS1_VERSION" -ge $(version_code 2.12.52) ]] ||
skip "Need MDS version at least 2.12.52"
local mdts=$(comma_list $(mdts_nodes))
}
run_test 136 "changelog_deregister leaving pending records"
+test_137() {
+ df $DIR
+ mkdir -p $DIR/d1
+ mkdir -p $DIR/d2
+ dd if=/dev/zero of=$DIR/d1/$tfile bs=4096 count=1
+ dd if=/dev/zero of=$DIR/d2/$tfile bs=4096 count=1
+ cancel_lru_locks osc
+
+ #define OBD_FAIL_PTLRPC_RESEND_RACE 0x525
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000525"
+
+ # RPC1: any reply is to be delayed to disable last_xid logic
+ ln $DIR/d1/$tfile $DIR/d1/f2 &
+ sleep 1
+
+ # RPC2: setattr1 reply is delayed & resent
+ # original reply comes to client; the resend get asleep
+ chmod 666 $DIR/d2/$tfile
+
+ # RPC3: setattr2 on the same file; run ahead of RPC2 resend
+ chmod 777 $DIR/d2/$tfile
+
+ # RPC2 resend wakes up
+ sleep 5
+ [ $(stat -c "%a" $DIR/d2/$tfile) == 777 ] || error "resend got applied"
+}
+run_test 137 "late resend must be skipped if already applied"
+
+test_138() {
+ remote_mds_nodsh && skip "remote MDS with nodsh"
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ [[ "$MDS1_VERSION" -ge $(version_code 2.12.59) ]] ||
+ skip "Need server version newer than 2.12.59"
+
+ zconf_umount_clients $CLIENTS $MOUNT
+
+#define OBD_FAIL_TGT_RECOVERY_CONNECT 0x724
+ #delay a first step of recovey when MDS waiting clients
+ #and failing to get osp logs
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0x724 fail_val=5
+
+ facet_failover $SINGLEMDS
+
+ #waiting failover and recovery timer
+ #the valuse is based on target_recovery_overseer() wait_event timeout
+ sleep 55
+ stop $SINGLEMDS || error "stop MDS failed"
+ do_facet $SINGLEMDS $LCTL set_param fail_loc=0
+ start $SINGLEMDS $(mdsdevname ${SINGLEMDS//mds/}) ||
+ error "start MDS failed"
+ zconf_mount_clients $CLIENTS $MOUNT
+}
+run_test 138 "Umount MDT during recovery"
+
+test_139() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ [ $MDS1_VERSION -lt $(version_code 2.13.50) ] &&
+ skip "Need MDS version at least 2.13.50"
+
+ mdt_dev=$(mdsdevname 1)
+
+ stop $SINGLEMDS || error "stop $SINGLEMDS failed"
+
+#define OBD_FAIL_OSP_INVALID_LOGID 0x2106
+ do_facet $SINGLEMDS $LCTL set_param fail_val=0x68 fail_loc=0x80002106
+ start $SINGLEMDS $mdt_dev $MDS_MOUNT_OPTS || error "Fail to start MDT"
+}
+run_test 139 "corrupted catid won't cause crash"
+
complete $SECONDS
check_and_cleanup_lustre
exit_status
git://git.whamcloud.com / fs / lustre-release.git / blobdiff