Whamcloud - gitweb
file xnu_sync.c was initially added on branch b_porting.
[fs/lustre-release.git] / lustre / ptlrpc / import.c
index aa4a6a9..5fdf812 100644 (file)
@@ -100,6 +100,10 @@ int ptlrpc_set_import_discon(struct obd_import *imp)
         spin_lock_irqsave(&imp->imp_lock, flags);
 
         if (imp->imp_state == LUSTRE_IMP_FULL) {
+                CERROR("%s: connection lost to %s@%s\n",
+                       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);
                 spin_unlock_irqrestore(&imp->imp_lock, flags);
                 obd_import_event(imp->imp_obd, imp, IMP_EVENT_DISCON);
@@ -180,7 +184,7 @@ void ptlrpc_invalidate_import(struct obd_import *imp, int in_rpc)
         obd_import_event(imp->imp_obd, imp, IMP_EVENT_INVALIDATE);
 }
 
-static void ptlrpc_activate_import(struct obd_import *imp)
+void ptlrpc_activate_import(struct obd_import *imp)
 {
         struct obd_device *obd = imp->imp_obd;
         unsigned long flags;
@@ -231,10 +235,12 @@ int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid)
         struct ptlrpc_request *request;
         int size[] = {sizeof(imp->imp_target_uuid),
                                  sizeof(obd->obd_uuid),
-                                 sizeof(imp->imp_dlm_handle)};
+                                 sizeof(imp->imp_dlm_handle),
+                                 sizeof(unsigned long)};
         char *tmp[] = {imp->imp_target_uuid.uuid,
                        obd->obd_uuid.uuid,
-                       (char *)&imp->imp_dlm_handle};
+                       (char *)&imp->imp_dlm_handle,
+                       (char *)&imp->imp_connect_flags}; /* XXX: make this portable! */
         struct ptlrpc_connect_async_args *aa;
         unsigned long flags;
 
@@ -255,14 +261,13 @@ int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid)
 
         IMPORT_SET_STATE_NOLOCK(imp, LUSTRE_IMP_CONNECTING);
 
-        imp->imp_conn_cnt++;
-        imp->imp_last_replay_transno = 0;
+        imp->imp_resend_replay = 0;
 
         if (imp->imp_remote_handle.cookie == 0) {
                 initial_connect = 1;
         } else {
                 committed_before_reconnect = imp->imp_peer_committed_transno;;
-
+                imp->imp_conn_cnt++;
         }
 
 
@@ -303,7 +308,8 @@ int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid)
 
         }
 
-        request = ptlrpc_prep_req(imp, imp->imp_connect_op, 3, size, tmp);
+        request = ptlrpc_prep_req(imp, LUSTRE_OBD_VERSION,
+                                  imp->imp_connect_op, 4, size, tmp);
         if (!request)
                 GOTO(out, rc = -ENOMEM);
 
@@ -322,11 +328,15 @@ int ptlrpc_connect_import(struct obd_import *imp, char * new_uuid)
         aa->pcaa_peer_committed = committed_before_reconnect;
         aa->pcaa_initial_connect = initial_connect;
 
-        if (aa->pcaa_initial_connect)
-                imp->imp_replayable = 1;
+        if (aa->pcaa_initial_connect) {
+                lustre_msg_add_op_flags(request->rq_reqmsg, 
+                                        MSG_CONNECT_INITIAL);
+                imp->imp_replayable = 1; 
+        }
 
         ptlrpcd_add_req(request);
         rc = 0;
+        imp->imp_connect_start = jiffies;
 out:
         if (rc != 0) {
                 IMPORT_SET_STATE(imp, LUSTRE_IMP_DISCON);
@@ -365,6 +375,10 @@ static int ptlrpc_connect_interpret(struct ptlrpc_request *request,
                 } else {
                         imp->imp_replayable = 0;
                 }
+                LASSERTF(imp->imp_conn_cnt < request->rq_repmsg->conn_cnt,
+                         "imp conn_cnt %d req conn_cnt %d", 
+                         imp->imp_conn_cnt, request->rq_repmsg->conn_cnt);
+                imp->imp_conn_cnt = request->rq_repmsg->conn_cnt;
                 imp->imp_remote_handle = request->rq_repmsg->handle;
                 IMPORT_SET_STATE(imp, LUSTRE_IMP_FULL);
                 GOTO(finish, rc = 0);
@@ -392,20 +406,29 @@ static int ptlrpc_connect_interpret(struct ptlrpc_request *request,
                                request->rq_repmsg->handle.cookie);
                         imp->imp_remote_handle = request->rq_repmsg->handle;
                 } else {
-                        CERROR("reconnected to %s@%s after partition\n",
+                        CDEBUG(D_HA, "reconnected to %s@%s after partition\n",
                                imp->imp_target_uuid.uuid,
                                imp->imp_connection->c_remote_uuid.uuid);
                 }
 
-                if (imp->imp_invalid)
+                if (imp->imp_invalid) {
                         IMPORT_SET_STATE(imp, LUSTRE_IMP_EVICTED);
-                else
+                } else if (MSG_CONNECT_RECOVERING & msg_flags) {
+                        CDEBUG(D_HA, "%s: reconnected to %s during replay\n",
+                               imp->imp_obd->obd_name, 
+                               imp->imp_target_uuid.uuid);
+                        imp->imp_resend_replay = 1;
+                        IMPORT_SET_STATE(imp, LUSTRE_IMP_REPLAY);
+                } else {
                         IMPORT_SET_STATE(imp, LUSTRE_IMP_RECOVER);
+                }
         } else if ((MSG_CONNECT_RECOVERING & msg_flags) && !imp->imp_invalid) {
                 LASSERT(imp->imp_replayable);
                 imp->imp_remote_handle = request->rq_repmsg->handle;
+                imp->imp_last_replay_transno = 0;
                 IMPORT_SET_STATE(imp, LUSTRE_IMP_REPLAY);
         } else {
+                CDEBUG(D_HA, "oops! we get evicted from %s\n", imp->imp_target_uuid.uuid);
                 imp->imp_remote_handle = request->rq_repmsg->handle;
                 IMPORT_SET_STATE(imp, LUSTRE_IMP_EVICTED);
         }
@@ -443,13 +466,13 @@ finish:
                 if (aa->pcaa_initial_connect && !imp->imp_initial_recov) {
                         ptlrpc_deactivate_import(imp);
                 }
-                /*if (rc == -ETIMEDOUT) {
+                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",
                        imp->imp_target_uuid.uuid,
                        (char *)imp->imp_connection->c_remote_uuid.uuid, rc);
@@ -463,7 +486,15 @@ static int completed_replay_interpret(struct ptlrpc_request *req,
                                     void * data, int rc)
 {
         atomic_dec(&req->rq_import->imp_replay_inflight);
-        ptlrpc_import_recovery_state_machine(req->rq_import);
+        if (req->rq_status == 0) {
+                ptlrpc_import_recovery_state_machine(req->rq_import);
+        } else {
+                CDEBUG(D_HA, "%s: LAST_REPLAY message error: %d, "
+                       "reconnecting\n", 
+                       req->rq_import->imp_obd->obd_name, req->rq_status);
+                ptlrpc_connect_import(req->rq_import, NULL);
+        }
+
         RETURN(0);
 }
 
@@ -475,7 +506,7 @@ static int signal_completed_replay(struct obd_import *imp)
         LASSERT(atomic_read(&imp->imp_replay_inflight) == 0);
         atomic_inc(&imp->imp_replay_inflight);
 
-        req = ptlrpc_prep_req(imp, OBD_PING, 0, NULL, NULL);
+        req = ptlrpc_prep_req(imp, LUSTRE_OBD_VERSION, OBD_PING, 0, NULL, NULL);
         if (!req)
                 RETURN(-ENOMEM);
 
@@ -544,6 +575,10 @@ int ptlrpc_import_recovery_state_machine(struct obd_import *imp)
                         GOTO(out, rc);
                 IMPORT_SET_STATE(imp, LUSTRE_IMP_FULL);
                 ptlrpc_activate_import(imp);
+                CERROR("%s: connection restored to %s@%s\n",
+                       imp->imp_obd->obd_name, 
+                       imp->imp_target_uuid.uuid,
+                       imp->imp_connection->c_remote_uuid.uuid);
         }
 
         if (imp->imp_state == LUSTRE_IMP_FULL) {
@@ -599,7 +634,8 @@ int ptlrpc_disconnect_import(struct obd_import *imp)
         }
         spin_unlock_irqrestore(&imp->imp_lock, flags);
 
-        request = ptlrpc_prep_req(imp, rq_opc, 0, NULL, NULL);
+        request = ptlrpc_prep_req(imp, LUSTRE_OBD_VERSION, rq_opc,
+                                  0, NULL, NULL);
         if (request) {
                 /* For non-replayable connections, don't attempt
                    reconnect if this fails */
@@ -617,6 +653,7 @@ int ptlrpc_disconnect_import(struct obd_import *imp)
 out:
         IMPORT_SET_STATE_NOLOCK(imp, LUSTRE_IMP_CLOSED);
         memset(&imp->imp_remote_handle, 0, sizeof(imp->imp_remote_handle));
+        imp->imp_conn_cnt = 0;
         spin_unlock_irqrestore(&imp->imp_lock, flags);
 
         RETURN(rc);