- int rc;
- unsigned long flags;
- int in_recovery = 0;
- int was_invalid = 0;
- ENTRY;
-
- spin_lock_irqsave(&imp->imp_lock, flags);
- if (imp->imp_state == LUSTRE_IMP_FULL) {
- CDEBUG(D_HA, "%s: new state: DISCON\n",
- imp->imp_client->cli_name);
- imp->imp_state = LUSTRE_IMP_DISCON;
- }
-
- if (imp->imp_state != LUSTRE_IMP_DISCON) {
- in_recovery = 1;
- } else if (imp->imp_invalid) {
- imp->imp_invalid = 0;
- was_invalid = 1;
- }
-
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- if (in_recovery == 1)
- RETURN(-EALREADY);
-
- down(&imp->imp_recovery_sem);
- /* If recovery happened while we waited, we're done. */
- if (imp->imp_state == LUSTRE_IMP_FULL)
- GOTO(out, rc = 0);
-
- LASSERT (imp->imp_state == LUSTRE_IMP_DISCON);
-
- if (new_uuid) {
- struct ptlrpc_connection *conn;
- struct obd_uuid uuid;
- struct ptlrpc_peer peer;
- struct obd_export *dlmexp;
-
- obd_str2uuid(&uuid, new_uuid);
- if (ptlrpc_uuid_to_peer(&uuid, &peer)) {
- CERROR("no connection found for UUID %s\n", new_uuid);
- GOTO(out, rc = -EINVAL);
- }
-
- conn = ptlrpc_get_connection(&peer, &uuid);
- if (!conn)
- GOTO(out, rc = -ENOMEM);
-
- 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);
- if (dlmexp->exp_connection)
- ptlrpc_put_connection(dlmexp->exp_connection);
- dlmexp->exp_connection = ptlrpc_connection_addref(conn);
- class_export_put(dlmexp);
-
- }
-
- connect:
- rc = ptlrpc_connect_import(imp);
-
- if (rc < 0) {
- CERROR("failed to reconnect to %s@%s: %d\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid, rc);
- GOTO(out, rc);
- }
-
- if (imp->imp_state == LUSTRE_IMP_EVICTED) {
- CDEBUG(D_HA, "evicted from %s@%s; invalidating\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid);
- ptlrpc_set_import_active(imp, 0);
- CDEBUG(D_HA, "%s: new state: RECOVER\n",
- imp->imp_client->cli_name);
- imp->imp_state = LUSTRE_IMP_RECOVER;
- }
-
- if (imp->imp_state == LUSTRE_IMP_REPLAY) {
- CDEBUG(D_HA, "replay requested by %s\n",
- imp->imp_target_uuid.uuid);
- rc = ptlrpc_replay(imp);
- if (rc)
- GOTO(out, rc);
-
- rc = ldlm_replay_locks(imp);
- if (rc)
- GOTO(out, rc);
-
- rc = signal_completed_replay(imp);
- if (rc)
- GOTO(out, rc);
- CDEBUG(D_HA, "%s: new state: RECOVER\n",
- imp->imp_client->cli_name);
- imp->imp_state = LUSTRE_IMP_RECOVER;
- }
-
- if (imp->imp_state == LUSTRE_IMP_RECOVER) {
- CDEBUG(D_HA, "reconnected to %s@%s\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid);
-
- ptlrpc_set_import_active(imp, 1);
- ptlrpc_resend(imp);
- spin_lock_irqsave(&imp->imp_lock, flags);
- CDEBUG(D_HA, "%s: new state: FULL\n",
- imp->imp_client->cli_name);
- imp->imp_state = LUSTRE_IMP_FULL;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
- ptlrpc_wake_delayed(imp);
- }
-
-
- LASSERT(imp->imp_state == LUSTRE_IMP_FULL);
-
- out:
- if (rc != 0) {
- spin_lock_irqsave(&imp->imp_lock, flags);
- imp->imp_state = LUSTRE_IMP_DISCON;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- if (rc == -ENOTCONN) {
- CDEBUG(D_HA, "evicted/aborted by %s@%s during recovery;"
- "invalidating and reconnecting\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid);
- GOTO(connect, -ENOTCONN);
- } else if (was_invalid) {
- ptlrpc_set_import_active(imp, 0);
- }
- }
- up(&imp->imp_recovery_sem);
- RETURN(rc);
+ int rc = 0;
+ ENTRY;
+
+ spin_lock(&imp->imp_lock);
+ if (imp->imp_state == LUSTRE_IMP_NEW || imp->imp_deactive ||
+ atomic_read(&imp->imp_inval_count))
+ rc = -EINVAL;
+ spin_unlock(&imp->imp_lock);
+ if (rc)
+ GOTO(out, rc);
+
+ /* force import to be disconnected. */
+ ptlrpc_set_import_discon(imp, 0, false);
+
+ if (new_uuid) {
+ struct obd_uuid uuid;
+
+ /* intruct import to use new uuid */
+ obd_str2uuid(&uuid, new_uuid);
+ rc = import_set_conn_priority(imp, &uuid);
+ if (rc)
+ GOTO(out, rc);
+ }
+
+ /* Check if reconnect is already in progress */
+ spin_lock(&imp->imp_lock);
+ if (imp->imp_state != LUSTRE_IMP_DISCON) {
+ imp->imp_force_verify = 1;
+ rc = -EALREADY;
+ }
+ spin_unlock(&imp->imp_lock);
+ if (rc)
+ GOTO(out, rc);
+
+ OBD_RACE(OBD_FAIL_PTLRPC_CONNECT_RACE);
+
+ rc = ptlrpc_connect_import(imp);
+ if (rc)
+ GOTO(out, rc);
+
+ if (!async) {
+ long timeout = cfs_time_seconds(obd_timeout);
+
+ CDEBUG(D_HA, "%s: recovery started, waiting %u jiffies\n",
+ obd2cli_tgt(imp->imp_obd), obd_timeout);
+
+ rc = wait_event_idle_timeout(imp->imp_recovery_waitq,
+ !ptlrpc_import_in_recovery(imp),
+ timeout);
+ if (rc == 0)
+ rc = -ETIMEDOUT;
+ else
+ rc = 0;
+ CDEBUG(D_HA, "%s: recovery finished\n",
+ obd2cli_tgt(imp->imp_obd));
+ }
+ EXIT;
+
+out:
+ return rc;