Land b_smallfix onto HEAD (20040423_1603)

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

diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index 67595fc..fbf8d19 100755 (executable)
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -63,6 +63,7 @@ setup() {
     [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
     start mds $MDSLCONFARGS --reformat
     zconf_mount `hostname` $MOUNT
+    echo 0x3f0410 > /proc/sys/portals/debug
 }
 
 $SETUP
@@ -378,6 +379,7 @@ test_18() {
     sleep 1 
     rm -f $DIR/$tfile
     touch $DIR/$tfile-2 || return 1
+    echo "pid: $pid will close"
     kill -USR1 $pid
     wait $pid || return 2
 
@@ -823,20 +825,71 @@ run_test 41 "read from a valid osc while other oscs are invalid"
 
 # test MDS recovery after ost failure
 test_42() {
+    blocks=`df $MOUNT | tail -1 | awk '{ print $1 }'`
     createmany -o $DIR/$tfile-%d 800
     replay_barrier ost
     unlinkmany $DIR/$tfile-%d 0 400
     facet_failover ost
     
-    # osc is evicted after
-    df $MOUNT && return 1
-    df $MOUNT || return 2
+    # osc is evicted, fs is smaller
+    blocks_after=`df $MOUNT | tail -1 | awk '{ print $1 }'`
+    [ $blocks_after -lt $blocks ] || return 1
     echo wait for MDS to timeout and recover
     sleep $((TIMEOUT * 2))
     unlinkmany $DIR/$tfile-%d 400 400
-    $CHECKSTAT -t file $DIR/$tfile-* && return 1 || true
+    $CHECKSTAT -t file $DIR/$tfile-* && return 2 || true
+}
+run_test 42 "recovery after ost failure"
+
+# b=2530
+# directory orphans can't be unlinked from PENDING directory
+test_43() {
+    replay_barrier mds
+
+    # OBD_FAIL_OST_CREATE_NET 0x204
+    do_facet ost "sysctl -w lustre.fail_loc=0x80000204"
+    facet_failover mds
+    df $MOUNT || return 1
+    sleep 10
+    do_facet ost "sysctl -w lustre.fail_loc=0"
+
+    return 0
+}
+run_test 43 "mds osc import failure during recovery; don't LBUG"
+
+test_44() {
+    mdcdev=`awk '/mds_svc_MNT/ {print $1}' < /proc/fs/lustre/devices`
+    do_facet mds "sysctl -w lustre.fail_loc=0x80000701"
+    $LCTL --device $mdcdev recover
+    df $MOUNT
+    do_facet mds "sysctl -w lustre.fail_loc=0"
+    return 0
+}
+run_test 44 "race in target handle connect"
+
+# Handle failed close
+test_45() {
+    mdcdev=`awk '/mds_svc_MNT/ {print $1}' < /proc/fs/lustre/devices`
+    $LCTL --device $mdcdev recover
+
+    multiop $DIR/$tfile O_c &
+    pid=$!
+    sleep 1
+
+    # This will cause the CLOSE to fail before even 
+    # allocating a reply buffer
+    $LCTL --device $mdcdev deactivate
+
+    # try the close
+    kill -USR1 $pid
+    wait $pid || return 1
+
+    $LCTL --device $mdcdev activate
+
+    $CHECKSTAT -t file $DIR/$tfile || return 2
+    return 0
 }
-run_test 42 "recoery after ost failure"
+run_test 45 "Handle failed close"
 
 equals_msg test complete, cleaning up
 $CLEANUP