ctdb-mutex: Test the lock by locking a 2nd byte range

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>

diff --git a/ctdb/server/ctdb_mutex_fcntl_helper.c b/ctdb/server/ctdb_mutex_fcntl_helper.c
index 07f111a0d6b692cef071b23b965860bcf30b5582..84d3790db57cc5518739d52b071a237ef0ceae04 100644 (file)
--- a/ctdb/server/ctdb_mutex_fcntl_helper.c
+++ b/ctdb/server/ctdb_mutex_fcntl_helper.c
@@ -265,6 +265,23 @@ static void lock_io_check_loop(struct tevent_req *subreq)
                goto done;
        }
 
+       /*
+        * Attempt to lock a 2nd byte range.  Using a blocking lock
+        * encourages ping timeouts if the cluster filesystem is in a
+        * bad state.  It also makes testing easier.
+        */
+       ret = fcntl_lock_fd(fd, true, 1);
+       if (ret != 0) {
+               fprintf(stderr,
+                       "%s: "
+                       "lock fail - lock file \"%s\" test lock error (%d)\n",
+                       progname,
+                       state->lock_file,
+                       ret);
+               goto done;
+       }
+
+       /* Unlock occurs on close */
        close(fd);
 
        subreq = tevent_wakeup_send(