LU-14182 lov: cancel layout lock on replay deadlock

[fs/lustre-release.git] / lustre / tests / replay-dual.sh

diff --git a/lustre/tests/replay-dual.sh b/lustre/tests/replay-dual.sh
index 2114600..b92df6b 100755 (executable)
--- a/lustre/tests/replay-dual.sh
+++ b/lustre/tests/replay-dual.sh
@@ -2,43 +2,35 @@
 
 set -e
 
-# bug number:  LU-2012 LU-8333 LU-7372
-ALWAYS_EXCEPT="14b     21b     26      $REPLAY_DUAL_EXCEPT"
-# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
-
-SAVE_PWD=$PWD
 PTLDEBUG=${PTLDEBUG:--1}
-LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
-SETUP=${SETUP:-""}
-CLEANUP=${CLEANUP:-""}
 MOUNT_2=${MOUNT_2:-"yes"}
-export MULTIOP=${MULTIOP:-multiop}
-. $LUSTRE/tests/test-framework.sh
 
+LUSTRE=${LUSTRE:-$(cd $(dirname $0)/..; echo $PWD)}
+. $LUSTRE/tests/test-framework.sh
 init_test_env $@
-. ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
 init_logging
 
 remote_mds_nodsh && skip "remote MDS with nodsh" && exit 0
 
-#                                   7  (min)"
-[ "$SLOW" = "no" ] && EXCEPT_SLOW="21b"
+ALWAYS_EXCEPT="$REPLAY_DUAL_EXCEPT "
+# bug number for skipped test:  LU-2012 LU-8333 LU-7372
+ALWAYS_EXCEPT+="                14b     21b     26 "
+# UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
 
 [[ "$mds1_FSTYPE" == zfs ]] &&
-# bug number for skipped test:       LU-2230
+# bug number for skipped test:        LU-2230
        ALWAYS_EXCEPT="$ALWAYS_EXCEPT 21b"
-if $SHARED_KEY; then
-# bug number for skipped tests:        LU-9795 LU-9795
-       ALWAYS_EXCEPT="         0a      0b      $ALWAYS_EXCEPT"
-fi
 
-build_test_filter
+#                                   7  (min)"
+[ "$SLOW" = "no" ] && EXCEPT_SLOW="21b "
 
+build_test_filter
 check_and_setup_lustre
+
 MOUNTED=$(mounted_lustre_filesystems)
 if ! $(echo $MOUNTED' ' | grep -w -q $MOUNT2' '); then
-    zconf_mount $HOSTNAME $MOUNT2
-    MOUNTED2=yes
+       zconf_mount $HOSTNAME $MOUNT2
+       MOUNTED2=yes
 fi
 
 assert_DIR
@@ -454,7 +446,7 @@ test_18() { # bug 3822 - evicting client with enqueued lock
        do_facet client lctl set_param ldlm.namespaces.*.early_lock_cancel=0
        do_facet client lctl set_param fail_loc=0x80000305  # drop cb, evict
        cancel_lru_locks mdc
-       usleep 500 # wait to ensure first client is one that will be evicted
+       sleep 0.1 # wait to ensure first client is one that will be evicted
        openfile -f O_RDONLY $MOUNT2/$tdir/$tfile
        wait $OPENPID
        do_facet client lctl set_param ldlm.namespaces.*.early_lock_cancel=1
@@ -637,6 +629,7 @@ test_22a () {
        do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
        do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
        CLIENT_PID=$!
+       sleep 1
 
        fail mds$((MDTIDX + 1))
        wait $CLIENT_PID || error "lfs mkdir failed"
@@ -663,6 +656,7 @@ test_22b () {
        do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
        do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
        CLIENT_PID=$!
+       sleep 1
 
        fail mds${MDTIDX},mds$((MDTIDX + 1))
        wait $CLIENT_PID || error "lfs mkdir failed"
@@ -693,6 +687,7 @@ test_22c () {
        do_facet mds$MDTIDX lctl set_param fail_loc=0x1701
        do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
        CLIENT_PID=$!
+       sleep 1
        do_facet mds$MDTIDX lctl set_param fail_loc=0
 
        fail mds$MDTIDX
@@ -720,6 +715,7 @@ test_22d () {
        do_facet mds$MDTIDX lctl set_param fail_loc=0x1701
        do_node $CLIENT1 $LFS mkdir -i $MDTIDX $MOUNT1/$remote_dir &
        CLIENT_PID=$!
+       sleep 1
        do_facet mds$MDTIDX lctl set_param fail_loc=0
 
        fail mds${MDTIDX},mds$((MDTIDX + 1))
@@ -764,6 +760,7 @@ test_23a () {
        do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
        do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
        local CLIENT_PID=$!
+       sleep 1
        do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
 
        fail mds$((MDTIDX + 1))
@@ -793,6 +790,7 @@ test_23b () {
        do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0x119
        do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
        local CLIENT_PID=$!
+       sleep 1
        do_facet mds$((MDTIDX + 1)) lctl set_param fail_loc=0
 
        fail mds${MDTIDX},mds$((MDTIDX + 1))
@@ -827,6 +825,7 @@ test_23c () {
        do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
        do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
        CLIENT_PID=$!
+       sleep 1
        do_facet mds${MDTIDX} lctl set_param fail_loc=0
 
        fail mds${MDTIDX}
@@ -856,6 +855,7 @@ test_23d () {
        do_facet mds${MDTIDX} lctl set_param fail_loc=0x1701
        do_node $CLIENT1 rmdir $MOUNT1/$remote_dir &
        CLIENT_PID=$!
+       sleep 1
        do_facet mds${MDTIDX} lctl set_param fail_loc=0
 
        fail mds${MDTIDX},mds$((MDTIDX + 1))
@@ -1060,6 +1060,26 @@ test_29() {
 }
 run_test 29 "replay vs update with the same xid"
 
+test_30() {
+       $LFS setstripe -E 1m -L mdt -E -1 $DIR/$tfile
+       #first write to have no problems with grants
+       dd if=/dev/zero of=$DIR/$tfile bs=4k count=10 ||
+               error "dd on client failed"
+       dd if=/dev/zero of=$DIR/$tfile bs=4k count=10 seek=10 ||
+               error "dd on client failed"
+
+       #define OBD_FAIL_LDLM_REPLAY_PAUSE       0x32e
+       lctl set_param fail_loc=0x32e fail_val=4
+       dd of=/dev/null if=$DIR2/$tfile &
+       local pid=$!
+       sleep 1
+
+       fail $SINGLEMDS
+
+       wait $pid || error "dd on client failed"
+}
+run_test 30 "layout lock replay is not blocked on IO"
+
 complete $SECONDS
 SLEEP=$((SECONDS - $NOW))
 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP