if (imp->imp_state == LUSTRE_IMP_FULL) {
CERROR("%s: connection lost to %s@%s\n",
- imp->imp_obd->obd_name,
+ imp->imp_obd->obd_name,
imp->imp_target_uuid.uuid,
imp->imp_connection->c_remote_uuid.uuid);
IMPORT_SET_STATE_NOLOCK(imp, LUSTRE_IMP_DISCON);
EXIT;
}
+#define ATTEMPT_TOO_SOON(last) \
+ ((last) && ((long)(jiffies - (last)) <= (long)(obd_timeout * 2 * HZ)))
+
+static int import_select_connection(struct obd_import *imp)
+{
+ struct obd_import_conn *imp_conn, *tmp;
+ struct obd_export *dlmexp;
+ int found = 0;
+ ENTRY;
+
+ spin_lock(&imp->imp_lock);
+
+ if (list_empty(&imp->imp_conn_list)) {
+ CERROR("no available connections on imp %p@%s\n",
+ imp, imp->imp_obd->obd_name);
+ spin_unlock(&imp->imp_lock);
+ RETURN(-EINVAL);
+ }
+
+ list_for_each_entry(imp_conn, &imp->imp_conn_list, oic_item) {
+ if (!ATTEMPT_TOO_SOON(imp_conn->oic_last_attempt)) {
+ found = 1;
+ break;
+ }
+ }
+
+ /* if not found, simply choose the current one */
+ if (!found) {
+ CWARN("obd %s imp 0x%p: all connections have been "
+ "tried recently\n", imp->imp_obd->obd_name, imp);
+ LASSERT(imp->imp_conn_current);
+ imp_conn = imp->imp_conn_current;
+ }
+ LASSERT(imp_conn->oic_conn);
+
+ imp_conn->oic_last_attempt = jiffies;
+
+ /* move the items ahead of the selected one to list tail */
+ while (1) {
+ tmp= list_entry(imp->imp_conn_list.next,
+ struct obd_import_conn, oic_item);
+ if (tmp == imp_conn)
+ break;
+ list_del(&tmp->oic_item);
+ list_add_tail(&tmp->oic_item, &imp->imp_conn_list);
+ }
+
+ /* switch connection, don't mind if it's same as the current one */
+ if (imp->imp_connection)
+ ptlrpc_put_connection(imp->imp_connection);
+ imp->imp_connection = ptlrpc_connection_addref(imp_conn->oic_conn);
+
+ dlmexp = class_conn2export(&imp->imp_dlm_handle);
+ LASSERT(dlmexp != NULL);
+ if (dlmexp->exp_connection)
+ ptlrpc_put_connection(imp->imp_connection);
+ dlmexp->exp_connection = ptlrpc_connection_addref(imp_conn->oic_conn);
+ class_export_put(dlmexp);
+
+ imp->imp_conn_current = imp_conn;
+ CWARN("obd %s imp 0x%p: select conn %s\n",
+ imp->imp_obd->obd_name, imp,
+ imp_conn->oic_uuid.uuid);
+ spin_unlock(&imp->imp_lock);
+
+ RETURN(0);
+}
+
+
+
int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid)
{
struct obd_device *obd = imp->imp_obd;
spin_unlock_irqrestore(&imp->imp_lock, flags);
if (new_uuid) {
- struct ptlrpc_connection *conn;
struct obd_uuid uuid;
- struct obd_export *dlmexp;
obd_str2uuid(&uuid, new_uuid);
- conn = ptlrpc_uuid_to_connection(&uuid);
- if (!conn)
- GOTO(out, rc = -ENOENT);
-
- CDEBUG(D_HA, "switching import %s/%s from %s to %s\n",
- imp->imp_target_uuid.uuid, imp->imp_obd->obd_name,
- imp->imp_connection->c_remote_uuid.uuid,
- conn->c_remote_uuid.uuid);
-
- /* Switch the import's connection and the DLM export's
- * connection (which are almost certainly the same, but we
- * keep distinct refs just to make things clearer. I think. */
- if (imp->imp_connection)
- ptlrpc_put_connection(imp->imp_connection);
- /* We hand off the ref from ptlrpc_get_connection. */
- imp->imp_connection = conn;
-
- dlmexp = class_conn2export(&imp->imp_dlm_handle);
-
- LASSERT(dlmexp != NULL);
-
- if (dlmexp->exp_connection)
- ptlrpc_put_connection(dlmexp->exp_connection);
- dlmexp->exp_connection = ptlrpc_connection_addref(conn);
- class_export_put(dlmexp);
-
+ rc = import_set_conn_priority(imp, &uuid);
+ if (rc)
+ GOTO(out, rc);
}
+ rc = import_select_connection(imp);
+ if (rc)
+ GOTO(out, rc);
request = ptlrpc_prep_req(imp, LUSTRE_OBD_VERSION,
imp->imp_connect_op, 4, size, tmp);
if (rc)
GOTO(out, rc);
+ LASSERT(imp->imp_conn_current);
+ imp->imp_conn_current->oic_last_attempt = 0;
msg_flags = lustre_msg_get_op_flags(request->rq_repmsg);
out:
if (rc != 0) {
IMPORT_SET_STATE(imp, LUSTRE_IMP_DISCON);
- if (aa->pcaa_initial_connect && !imp->imp_initial_recov) {
+ if (aa->pcaa_initial_connect && !imp->imp_initial_recov)
ptlrpc_deactivate_import(imp);
- }
- if (rc == -ETIMEDOUT && (jiffies - imp->imp_connect_start) > HZ) {
- CDEBUG(D_ERROR, "recovery of %s on %s failed (timeout)\n",
- imp->imp_target_uuid.uuid,
- (char *)imp->imp_connection->c_remote_uuid.uuid);
- ptlrpc_connect_import(imp, NULL);
- RETURN(0);
- }
- CDEBUG(D_ERROR, "recovery of %s on %s failed (%d)\n",
+
+ CDEBUG(D_HA, "recovery of %s on %s failed (%d)\n",
imp->imp_target_uuid.uuid,
(char *)imp->imp_connection->c_remote_uuid.uuid, rc);
}