recoverd: Don't release and re-take the recovery lock

Just continue to hold it, otherwise a broken node might win an
election and grab the lock.

Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
(Imported from commit 48c91407abd5e34463d3a10cb6fce47ec4a0d5f6)

diff --git a/server/ctdb_recoverd.c b/server/ctdb_recoverd.c
index 2abf19b4415a2841289cbda60991fa5002cf12c0..a4daf090afdb3db6a40df8a8ca8fd7d990c360b2 100644 (file)
--- a/server/ctdb_recoverd.c
+++ b/server/ctdb_recoverd.c
@@ -1809,29 +1809,35 @@ static int do_recovery(struct ctdb_recoverd *rec,
        }
 
         if (ctdb->recovery_lock_file != NULL) {
-               DEBUG(DEBUG_ERR, ("Taking out recovery lock from recovery daemon (%s)\n", ctdb->recovery_lock_file));
-               start_time = timeval_current();
-               ctdb_recovery_unlock(ctdb);
-               DEBUG(DEBUG_NOTICE, ("Attempting to take recovery lock\n"));
-               if (!ctdb_recovery_lock(ctdb)) {
-                       if (ctdb->runstate == CTDB_RUNSTATE_FIRST_RECOVERY) {
-                               /* If ctdb is trying first recovery, it's
-                                * possible that current node does not know yet
-                                * who the recmaster is.
-                                */
-                               DEBUG(DEBUG_ERR, ("Unable to get recovery lock"
-                                               " - retrying recovery\n"));
+               if (ctdb_recovery_have_lock(ctdb)) {
+                       DEBUG(DEBUG_NOTICE, ("Already holding recovery lock\n"));
+               } else {
+                       start_time = timeval_current();
+                       DEBUG(DEBUG_NOTICE, ("Attempting to take recovery lock (%s)\n",
+                                            ctdb->recovery_lock_file));
+                       if (!ctdb_recovery_lock(ctdb)) {
+                               if (ctdb->runstate == CTDB_RUNSTATE_FIRST_RECOVERY) {
+                                       /* If ctdb is trying first recovery, it's
+                                        * possible that current node does not know
+                                        * yet who the recmaster is.
+                                        */
+                                       DEBUG(DEBUG_ERR, ("Unable to get recovery lock"
+                                                         " - retrying recovery\n"));
+                                       return -1;
+                               }
+
+                               DEBUG(DEBUG_ERR,("Unable to get recovery lock - aborting recovery "
+                                                "and ban ourself for %u seconds\n",
+                                                ctdb->tunable.recovery_ban_period));
+                               ctdb_ban_node(rec, pnn, ctdb->tunable.recovery_ban_period);
                                return -1;
                        }
-
-                       DEBUG(DEBUG_ERR,("Unable to get recovery lock - aborting recovery "
-                                        "and ban ourself for %u seconds\n",
-                                        ctdb->tunable.recovery_ban_period));
-                       ctdb_ban_node(rec, pnn, ctdb->tunable.recovery_ban_period);
-                       return -1;
+                       ctdb_ctrl_report_recd_lock_latency(ctdb,
+                                                          CONTROL_TIMEOUT(),
+                                                          timeval_elapsed(&start_time));
+                       DEBUG(DEBUG_NOTICE,
+                             ("Recovery lock taken successfully by recovery daemon\n"));
                }
-               ctdb_ctrl_report_recd_lock_latency(ctdb, CONTROL_TIMEOUT(), timeval_elapsed(&start_time));
-               DEBUG(DEBUG_NOTICE,("Recovery lock taken successfully by recovery daemon\n"));
        }
 
        DEBUG(DEBUG_NOTICE, (__location__ " Recovery initiated due to problem with node %u\n", rec->last_culprit_node));