LU-904 ptlrpc: redo io on -EINPROGRESS

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

diff --git a/lustre/tests/replay-ost-single.sh b/lustre/tests/replay-ost-single.sh
index 1bdbcbf..036750b 100755 (executable)
--- a/lustre/tests/replay-ost-single.sh
+++ b/lustre/tests/replay-ost-single.sh
@@ -8,18 +8,14 @@ CLEANUP=${CLEANUP:-""}
 . $LUSTRE/tests/test-framework.sh
 init_test_env $@
 . ${CONFIG:=$LUSTRE/tests/cfg/$NAME.sh}
+init_logging
 
 # While we do not use OSTCOUNT=1 setup anymore,
 # ost1failover_HOST is used
 #ostfailover_HOST=${ostfailover_HOST:-$ost_HOST}
 #failover= must be defined in OST_MKFS_OPTIONS if ostfailover_HOST != ost_HOST
 
-remote_ost_nodsh && skip "remote OST with nodsh" && exit 0
-
-if [ "$FAILURE_MODE" = "HARD" ] && mixed_ost_devs; then
-    skip_env "$0: Several ost services on one ost node are used with FAILURE_MODE=$FAILURE_MODE. "
-    exit 0
-fi
+require_dsh_ost || exit 0
 
 # Tests that fail on uml
 CPU=`awk '/model/ {print $4}' /proc/cpuinfo`
@@ -180,23 +176,31 @@ test_6() {
 
     f=$TDIR/$tfile
     rm -f $f
-    sync && sleep 2 && sync    # wait for delete thread
+    sync && sleep 2 && sync  # wait for delete thread
+
+    # wait till space is returned, following
+    # (( $before > $after_dd)) test counting on that
+    wait_mds_ost_sync || return 4
+    wait_destroy_complete || return 5
+
     before=`kbytesfree`
     dd if=/dev/urandom bs=4096 count=1280 of=$f || return 28
     lfs getstripe $f
-    get_stripe_info client $f
+    stripe_index=$(lfs getstripe -i $f)
 
     sync
     sleep 2 # ensure we have a fresh statfs
     sync
-#define OBD_FAIL_MDS_REINT_NET_REP       0x119
-    do_facet mds "lctl set_param fail_loc=0x80000119"
+
+    #define OBD_FAIL_MDS_REINT_NET_REP       0x119
+    do_facet $SINGLEMDS "lctl set_param fail_loc=0x80000119"
     after_dd=`kbytesfree`
     log "before: $before after_dd: $after_dd"
     (( $before > $after_dd )) || return 1
     rm -f $f
     fail ost$((stripe_index + 1))
-    wait_recovery_complete ost$((stripe_index + 1)) || error "OST recovery not done"
+    wait_recovery_complete ost$((stripe_index + 1)) ||
+        error "OST$((stripe_index + 1)) recovery not completed"
     $CHECKSTAT -t file $f && return 2 || true
     sync
     # let the delete happen
@@ -204,7 +208,7 @@ test_6() {
     wait_destroy_complete || return 5
     after=`kbytesfree`
     log "before: $before after: $after"
-    (( $before <= $after + 40 )) || return 3   # take OST logs into account
+    (( $before <= $after + 40 )) || return 3   # take OST logs into account
 }
 run_test 6 "Fail OST before obd_destroy"
 
@@ -212,6 +216,12 @@ test_7() {
     f=$TDIR/$tfile
     rm -f $f
     sync && sleep 5 && sync    # wait for delete thread
+
+    # wait till space is returned, following
+    # (( $before > $after_dd)) test counting on that
+    wait_mds_ost_sync || return 4
+    wait_destroy_complete || return 5
+
     before=`kbytesfree`
     dd if=/dev/urandom bs=4096 count=1280 of=$f || return 4
     sync
@@ -235,6 +245,66 @@ test_7() {
 }
 run_test 7 "Fail OST before obd_destroy"
 
-equals_msg `basename $0`: test complete, cleaning up
+test_8a() {
+    verify=$ROOT/tmp/verify-$$
+    dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
+        error "Create verify file failed"
+#define OBD_FAIL_OST_DQACQ_NET           0x230
+    do_facet ost1 "lctl set_param fail_loc=0x230"
+    dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
+    ddpid=$!
+    sleep $TIMEOUT  # wait for the io to become redo io
+    if ! ps -p $ddpid  > /dev/null 2>&1; then
+            error "redo io finished incorrectly"
+            return 1
+    fi
+    do_facet ost1 "lctl set_param fail_loc=0"
+    wait $ddpid || return 1
+    cancel_lru_locks osc
+    cmp $verify $TDIR/$tfile || return 2
+    rm -f $verify $TDIR/$tfile
+}
+run_test 8a "Verify redo io: redo io when get -EINPROGRESS error"
+
+test_8b() {
+    verify=$ROOT/tmp/verify-$$
+    dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
+        error "Create verify file failed"
+#define OBD_FAIL_OST_DQACQ_NET           0x230
+    do_facet ost1 "lctl set_param fail_loc=0x230"
+    dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
+    ddpid=$!
+    sleep $TIMEOUT  # wait for the io to become redo io
+    fail ost1
+    do_facet ost1 "lctl set_param fail_loc=0"
+    wait $ddpid || return 1
+    cancel_lru_locks osc
+    cmp $verify $TDIR/$tfile || return 2
+    rm -f $verify $TDIR/$tfile
+}
+run_test 8b "Verify redo io: redo io should success after recovery"
+
+test_8c() {
+    verify=$ROOT/tmp/verify-$$
+    dd if=/dev/urandom of=$verify bs=4096 count=1280 ||
+        error "Create verify file failed"
+#define OBD_FAIL_OST_DQACQ_NET           0x230
+    do_facet ost1 "lctl set_param fail_loc=0x230"
+    dd if=$verify of=$TDIR/$tfile bs=4096 count=1280 oflag=sync &
+    ddpid=$!
+    sleep $TIMEOUT  # wait for the io to become redo io
+    ost_evict_client
+    # allow recovery to complete
+    sleep $((TIMEOUT + 2))
+    do_facet ost1 "lctl set_param fail_loc=0"
+    wait $ddpid
+    cancel_lru_locks osc
+    cmp $verify $TDIR/$tfile && return 2
+    rm -f $verify $TDIR/$tfile
+}
+run_test 8c "Verify redo io: redo io should fail after eviction"
+
+
+complete $(basename $0) $SECONDS
 check_and_cleanup_lustre
-[ -f "$TESTSUITELOG" ] && cat $TESTSUITELOG && grep -q FAIL $TESTSUITELOG && exit 1 || true
+exit_status