Ticket #8: greyfix-8.patch

greyfix.c

@@ -79 +79 @@
 static DBT dbkey = { 0 };
 static DBT dbdata = { 0 };
 static struct triplet_data triplet_data;
+static int deadlock_detect = 0;
 
 static unsigned long greylist_delay = DELAY_MAIL_SECS;
 static unsigned long bloc_max_idle = AUTO_RECORD_LIFE_SECS;
@@ -161 +162 @@
         log_db_error("db_env_create", rc);
     else {
         dbenv->set_errcall(dbenv, db_errcall_fcn);
+        rc = dbenv->set_lk_detect(dbenv, DB_LOCK_YOUNGEST);
+        if (rc)
+            log_db_error("dbenv->set_lk_detect DB_LOCK_YOUNGEST, expired triplets will not be deleted", rc);
+        else
+            deadlock_detect = 1;
         rc = dbenv->open(dbenv, db_home,
                          DB_INIT_LOCK | DB_INIT_MPOOL | DB_CREATE, 0);
         if (rc) {
@@ -269 +275 @@
     int rc;
     time_t now;
     unsigned int count = 0;
-    if (db == 0)
+    /* Cursor operations can hold several locks and therefore deadlock
+       so don't run expiry if deadlock detection does not work
+       http://www.oracle.com/technology/documentation/berkeley-db/db/ref/lock/notxn.html */
+    if (db == 0 || deadlock_detect == 0)
         return;
     if (time(&now) == (time_t)-1) {
         syslog(LOG_ERR, "time failed during run_expiry");
@@ -303 +312 @@
                     count++;
             }
         }
-        if (rc != DB_NOTFOUND)
+        if (rc == DB_LOCK_DEADLOCK)
+            syslog(LOG_DEBUG, "skipping concurrent expiry avoids "
+                   "deadlocks and unnecessary work");
+        else if (rc != DB_NOTFOUND)
             log_db_error("dbcp->c_get failed", rc);
         if (rc = dbcp->c_close(dbcp))
             log_db_error("dbcp->c_close failed", rc);