Whamcloud - gitweb
LU-4913 mgc: mgc import reconnect race
[fs/lustre-release.git] / lustre / ptlrpc / import.c
index 81cf710..c18855d 100644 (file)
@@ -281,6 +281,7 @@ void ptlrpc_invalidate_import(struct obd_import *imp)
        if (!imp->imp_invalid || imp->imp_obd->obd_no_recov)
                ptlrpc_deactivate_import(imp);
 
+       CFS_FAIL_TIMEOUT(OBD_FAIL_MGS_CONNECT_NET, 3 * cfs_fail_val / 2);
        LASSERT(imp->imp_invalid);
 
         /* Wait forever until inflight == 0. We really can't do it another
@@ -396,6 +397,19 @@ void ptlrpc_activate_import(struct obd_import *imp)
 }
 EXPORT_SYMBOL(ptlrpc_activate_import);
 
+static void ptlrpc_pinger_force(struct obd_import *imp)
+{
+       CDEBUG(D_HA, "%s: waking up pinger s:%s\n", obd2cli_tgt(imp->imp_obd),
+              ptlrpc_import_state_name(imp->imp_state));
+
+       spin_lock(&imp->imp_lock);
+       imp->imp_force_verify = 1;
+       spin_unlock(&imp->imp_lock);
+
+       if (imp->imp_state != LUSTRE_IMP_CONNECTING)
+               ptlrpc_pinger_wake_up();
+}
+
 void ptlrpc_fail_import(struct obd_import *imp, __u32 conn_cnt)
 {
         ENTRY;
@@ -412,14 +426,7 @@ void ptlrpc_fail_import(struct obd_import *imp, __u32 conn_cnt)
                         ptlrpc_deactivate_import(imp);
                 }
 
-                CDEBUG(D_HA, "%s: waking up pinger\n",
-                       obd2cli_tgt(imp->imp_obd));
-
-               spin_lock(&imp->imp_lock);
-               imp->imp_force_verify = 1;
-               spin_unlock(&imp->imp_lock);
-
-               ptlrpc_pinger_wake_up();
+               ptlrpc_pinger_force(imp);
        }
        EXIT;
 }
@@ -427,6 +434,23 @@ EXPORT_SYMBOL(ptlrpc_fail_import);
 
 int ptlrpc_reconnect_import(struct obd_import *imp)
 {
+#ifdef ENABLE_PINGER
+       struct l_wait_info lwi;
+       int secs = cfs_time_seconds(obd_timeout);
+       int rc;
+
+       ptlrpc_pinger_force(imp);
+
+       CDEBUG(D_HA, "%s: recovery started, waiting %u seconds\n",
+              obd2cli_tgt(imp->imp_obd), secs);
+
+       lwi = LWI_TIMEOUT(secs, NULL, NULL);
+       rc = l_wait_event(imp->imp_recovery_waitq,
+                         !ptlrpc_import_in_recovery(imp), &lwi);
+       CDEBUG(D_HA, "%s: recovery finished s:%s\n", obd2cli_tgt(imp->imp_obd),
+              ptlrpc_import_state_name(imp->imp_state));
+       return rc;
+#else
        ptlrpc_set_import_discon(imp, 0);
        /* Force a new connect attempt */
        ptlrpc_invalidate_import(imp);
@@ -452,6 +476,7 @@ int ptlrpc_reconnect_import(struct obd_import *imp)
        /* Attempt a new connect */
        ptlrpc_recover_import(imp, NULL, 0);
        return 0;
+#endif
 }
 EXPORT_SYMBOL(ptlrpc_reconnect_import);