require_dsh_mds || exit 0
# Skip these tests
-# bug number: 17466 18857 LU-1867 LU-1473
-ALWAYS_EXCEPT="61d 33a 33b 89 62 $REPLAY_SINGLE_EXCEPT"
-
-[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test: LU-951
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 73a"
+# bug number: 17466 18857 LU-1473
+ALWAYS_EXCEPT="61d 33a 33b 62 $REPLAY_SINGLE_EXCEPT"
# 63 min 7 min AT AT AT AT"
[ "$SLOW" = "no" ] && EXCEPT_SLOW="1 2 3 4 6 12 16 44a 44b 65 66 67 68"
[ $(facet_fstype $SINGLEMDS) = "zfs" ] &&
-# bug number for skipped test: LU-3127
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 73b"
+# bug number for skipped test: LU-1867 LU-3127
+ ALWAYS_EXCEPT="$ALWAYS_EXCEPT 89 73b"
build_test_filter
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=$!
- $CHECKSTAT -t file $DIR/$tfile-* && return 3 || true
+ #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"
+
+ 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
- do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x1701
+ #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
$LFS mkdir -i $MDTIDX $remote_dir &
local CLIENT_PID=$!
- fail mds$((MDTIDX + 1))
+ fail mds${MDTIDX}
wait $CLIENT_PID || error "remote creation failed"
return 0
}
-run_test 80a "DNE: create remote dir, drop update rep from MDT1, fail MDT1"
+run_test 80a "DNE: create remote dir, drop update rep from MDT0, fail MDT0"
test_80b() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
mkdir -p $DIR/$tdir
#define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
- do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x1701
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
$LFS mkdir -i $MDTIDX $remote_dir &
local CLIENT_PID=$!
- fail mds${MDTIDX}
+ fail mds$((MDTIDX + 1))
wait $CLIENT_PID || error "remote creation failed"
return 0
}
-run_test 80b "DNE: create remote dir, drop update rep from MDT1, fail MDT0"
+run_test 80b "DNE: create remote dir, drop update rep from MDT0, fail MDT1"
test_80c() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
mkdir -p $DIR/$tdir
#define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
- do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x1701
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
$LFS mkdir -i $MDTIDX $remote_dir &
local CLIENT_PID=$!
mkdir -p $DIR/$tdir
#define OBD_FAIL_UPDATE_OBJ_NET_REP 0x1701
- do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x1701
+ do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
$LFS mkdir -i $MDTIDX $remote_dir &
local CLIENT_PID=$!
+ # sleep 3 seconds to make sure MDTs are failed after
+ # lfs mkdir -i has finished on all of MDTs.
+ sleep 3
fail mds${MDTIDX},mds$((MDTIDX + 1))
wait $CLIENT_PID || error "remote creation failed"
mkdir -p $DIR/$tdir
# OBD_FAIL_MDS_REINT_NET_REP 0x119
- do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
$LFS mkdir -i $MDTIDX $remote_dir &
local CLIENT_PID=$!
+ # sleep 3 seconds to make sure MDTs are failed after
+ # lfs mkdir -i has finished on all of MDTs.
+ sleep 3
+
fail mds${MDTIDX}
wait $CLIENT_PID || error "remote creation failed"
return 0
}
-run_test 80e "DNE: create remote dir, drop MDT0 rep, fail MDT0"
+run_test 80e "DNE: create remote dir, drop MDT1 rep, fail MDT0"
test_80f() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
mkdir -p $DIR/$tdir
# OBD_FAIL_MDS_REINT_NET_REP 0x119
- do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
$LFS mkdir -i $MDTIDX $remote_dir &
local CLIENT_PID=$!
return 0
}
-run_test 80f "DNE: create remote dir, drop MDT0 rep, fail MDT1"
+run_test 80f "DNE: create remote dir, drop MDT1 rep, fail MDT1"
test_80g() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
mkdir -p $DIR/$tdir
# OBD_FAIL_MDS_REINT_NET_REP 0x119
- do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
$LFS mkdir -i $MDTIDX $remote_dir &
local CLIENT_PID=$!
+ # sleep 3 seconds to make sure MDTs are failed after
+ # lfs mkdir -i has finished on all of MDTs.
+ sleep 3
+
fail mds${MDTIDX}
fail mds$((MDTIDX + 1))
return 0
}
-run_test 80g "DNE: create remote dir, drop MDT0 rep, fail MDT0, then MDT1"
+run_test 80g "DNE: create remote dir, drop MDT1 rep, fail MDT0, then MDT1"
test_80h() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
mkdir -p $DIR/$tdir
# OBD_FAIL_MDS_REINT_NET_REP 0x119
- do_facet mds${MDTIDX} lctl set_param fail_loc=0x119
+ do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
$LFS mkdir -i $MDTIDX $remote_dir &
local CLIENT_PID=$!
+ # sleep 3 seconds to make sure MDTs are failed after
+ # lfs mkdir -i has finished on all of MDTs.
+ sleep 3
+
fail mds${MDTIDX},mds$((MDTIDX + 1))
wait $CLIENT_PID || return 1
return 0
}
-run_test 80h "DNE: create remote dir, drop MDT0 rep, fail 2 MDTs"
+run_test 80h "DNE: create remote dir, drop MDT1 rep, fail 2 MDTs"
test_81a() {
[ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
}
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)
+
+ $LFS getdirstripe $striped_dir
+ [ $stripe_count -eq 2 ] || error "$stripe_count != 2"
+
+ createmany -o $striped_dir/f-%d 20 ||
+ error "creation failed under striped dir"
+}
+
+test_100a() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ ([ $FAILURE_MODE == "HARD" ] &&
+ [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+ skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+ return 0
+
+ local striped_dir=$DIR/$tdir/striped_dir
+ local MDTIDX=1
+
+ mkdir $DIR/$tdir
+
+ #To make sure MDT1 and MDT0 are connected
+ #otherwise it may create single stripe dir here
+ $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
+
+ #define OBD_FAIL_OUT_UPDATE_NET_REP 0x1701
+ do_facet mds$((MDTIDX+1)) lctl set_param fail_loc=0x1701
+ $LFS setdirstripe -i0 -c2 $striped_dir &
+ local CLIENT_PID=$!
+
+ fail mds$((MDTIDX + 1))
+
+ wait $CLIENT_PID || error "striped dir creation failed"
+
+ striped_dir_check_100 || error "striped dir check failed"
+ rm -rf $DIR/$tdir || error "rmdir failed"
+}
+run_test 100a "DNE: create striped dir, drop update rep from MDT1, fail MDT1"
+
+test_100b() {
+ [ $MDSCOUNT -lt 2 ] && skip "needs >= 2 MDTs" && return 0
+ ([ $FAILURE_MODE == "HARD" ] &&
+ [ "$(facet_host mds1)" == "$(facet_host mds2)" ]) &&
+ skip "MDTs needs to be on diff hosts for HARD fail mode" &&
+ return 0
+
+ local striped_dir=$DIR/$tdir/striped_dir
+ local MDTIDX=1
+
+ mkdir $DIR/$tdir
+
+ #To make sure MDT1 and MDT0 are connected
+ #otherwise it may create single stripe dir here
+ $LFS setdirstripe -i1 $DIR/$tdir/remote_dir
+
+ # OBD_FAIL_MDS_REINT_NET_REP 0x119
+ do_facet mds$MDTIDX lctl set_param fail_loc=0x119
+ $LFS mkdir -i0 -c2 $striped_dir &
+
+ local CLIENT_PID=$!
+ fail mds$MDTIDX
+
+ wait $CLIENT_PID || error "striped dir creation failed"
+
+ striped_dir_check_100 || error "striped dir check failed"
+ rm -rf $DIR/$tdir || error "rmdir failed"
+}
+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