mds_evict_client
client_up || client_up || true # reconnect
+ do_facet $SINGLEMDS "lctl set_param -n osd*.*MDT*.force_sync 1"
+
fail $SINGLEMDS # start orphan recovery
wait_recovery_complete $SINGLEMDS || error "MDS recovery not done"
wait_delete_completed_mds $wait_timeout || return 3
+
AFTERUSED=$(df -P $DIR | tail -1 | awk '{ print $3 }')
log "before $BEFOREUSED, after $AFTERUSED"
(( $AFTERUSED > $BEFOREUSED + $(fs_log_size) )) &&
# b3764 timed out lock replay
test_52() {
- touch $DIR/$tfile
- cancel_lru_locks mdc
+ [ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.90) ] &&
+ skip "MDS prior to 2.6.90 handle LDLM_REPLY_NET incorrectly" &&
+ return 0
- multiop $DIR/$tfile s || return 1
- replay_barrier $SINGLEMDS
-#define OBD_FAIL_LDLM_REPLY 0x30c
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x8000030c"
- fail $SINGLEMDS || return 2
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
+ touch $DIR/$tfile
+ cancel_lru_locks mdc
+
+ multiop_bg_pause $DIR/$tfile s_s || return 1
+ mpid=$!
+
+ #define OBD_FAIL_MDS_LDLM_REPLY_NET 0x157
+ lctl set_param -n ldlm.cancel_unused_locks_before_replay "0"
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000157"
- $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
+ fail $SINGLEMDS || return 2
+ kill -USR1 $mpid
+ wait $mpid || return 3
+
+ do_facet $SINGLEMDS "lctl set_param fail_loc=0x0"
+ lctl set_param fail_loc=0x0
+ lctl set_param -n ldlm.cancel_unused_locks_before_replay "1"
+ rm -f $DIR/$tfile
}
run_test 52 "time out lock replay (3764)"
}
run_test 73b "open(O_CREAT), unlink, replay, reconnect at open_replay reply, close"
-test_73c() {
- multiop_bg_pause $DIR/$tfile O_tSc || return 3
- pid=$!
- rm -f $DIR/$tfile
-
- replay_barrier $SINGLEMDS
-#define OBD_FAIL_TGT_LAST_REPLAY 0x710
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000710"
- fail $SINGLEMDS
- kill -USR1 $pid
- wait $pid || return 1
- [ -e $DIR/$tfile ] && return 2
- return 0
-}
-run_test 73c "open(O_CREAT), unlink, replay, reconnect at last_replay, close"
-
# bug 18554
test_74() {
local clients=${CLIENTS:-$HOSTNAME}
local remote_dir=$DIR/$tdir/remote_dir
mkdir -p $DIR/$tdir
- #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
+ #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
$LFS mkdir -i $MDTIDX $remote_dir &
local CLIENT_PID=$!
}
run_test 81h "DNE: unlink remote dir, drop request reply, fail 2 MDTs"
-test_83a() {
- mkdir -p $DIR/$tdir
- createmany -o $DIR/$tdir/$tfile- 10 || return 1
-#define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD 0x140
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
- unlinkmany $DIR/$tdir/$tfile- 10 || return 2
-}
-run_test 83a "fail log_add during unlink recovery"
-
-test_83b() {
- mkdir -p $DIR/$tdir
- createmany -o $DIR/$tdir/$tfile- 10 || return 1
- replay_barrier $SINGLEMDS
- unlinkmany $DIR/$tdir/$tfile- 10 || return 2
-#define OBD_FAIL_MDS_FAIL_LOV_LOG_ADD 0x140
- do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000140"
- fail $SINGLEMDS
-}
-run_test 83b "fail log_add during unlink recovery"
-
test_84a() {
#define OBD_FAIL_MDS_OPEN_WAIT_CREATE 0x144
do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000144"
dd if=/dev/urandom of=$DIR/$tdir/f-$file_id bs=4096 count=128
done
- # if the objids were not recreated, then "ls" will failed for -ENOENT
+ # if the objids were not recreated, then "ls" will fail with -ENOENT
ls -l $DIR/$tdir/* || error "can't get the status of precreated files"
local file_id
}
run_test 90 "lfs find identifies the missing striped file segments"
+test_93() {
+ local server_version=$(lustre_version_code $SINGLEMDS)
+ [[ $server_version -ge $(version_code 2.6.90) ]] ||
+ [[ $server_version -ge $(version_code 2.5.4) &&
+ $server_version -lt $(version_code 2.5.50) ]] ||
+ { skip "Need MDS version 2.5.4+ or 2.6.90+"; return; }
+
+ cancel_lru_locks osc
+
+ $SETSTRIPE -i 0 -c 1 $DIR/$tfile
+ dd if=/dev/zero of=$DIR/$tfile bs=1024 count=1
+#define OBD_FAIL_TGT_REPLAY_RECONNECT 0x715
+ # We need to emulate a state that OST is waiting for other clients
+ # not completing the recovery. Final ping is queued, but reply will be sent
+ # on the recovery completion. It is done by sleep before processing final
+ # pings
+ do_facet ost1 "$LCTL set_param fail_val=40"
+ do_facet ost1 "$LCTL set_param fail_loc=0x715"
+ fail ost1
+}
+run_test 93 "replay + reconnect"
+
striped_dir_check_100() {
local striped_dir=$DIR/$tdir/striped_dir
local stripe_count=$($LFS getdirstripe -c $striped_dir)
#otherwise it may create single stripe dir here
$LFS setdirstripe -i1 $DIR/$tdir/remote_dir
- #define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
+ #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
do_facet mds$((MDTIDX+1)) lctl set_param fail_loc=0x1701
$LFS setdirstripe -i0 -c2 $striped_dir &
local CLIENT_PID=$!
}
run_test 100b "DNE: create striped dir, fail MDT0"
+test_101() { #LU-5648
+ mkdir -p $DIR/$tdir/d1
+ mkdir -p $DIR/$tdir/d2
+ touch $DIR/$tdir/file0
+ num=1000
+
+ replay_barrier $SINGLEMDS
+ for i in $(seq $num) ; do
+ echo test$i > $DIR/$tdir/d1/file$i
+ done
+
+ fail_abort $SINGLEMDS
+ for i in $(seq $num) ; do
+ touch $DIR/$tdir/d2/file$i
+ test -s $DIR/$tdir/d2/file$i &&
+ ls -al $DIR/$tdir/d2/file$i && error "file$i's size > 0"
+ done
+
+ rm -rf $DIR/$tdir
+}
+run_test 101 "Shouldn't reassign precreated objs to other files after recovery"
+
complete $SECONDS
check_and_cleanup_lustre
exit_status
git://git.whamcloud.com / fs / lustre-release.git / blobdiff