nfs4: Correctly handle timeouts and other rpc errors

src/add-ons/kernel/file_systems/nfs4/RPCServer.cpp

@@ -154,15 +154,28 @@ Server::SendCallAsync(Call* call, Reply** reply, Request** request)
 
 	fRequests.AddRequest(req);
 
+	*request = req;
+	return ResendCallAsync(call, req);
+}
+
+
+status_t
+Server::ResendCallAsync(Call* call, Request* req)
+{
+	if (fThreadError != B_OK) {
+		fRequests.FindRequest(req->fXID);
+		delete req;
+		return fThreadError;
+	}
+
 	XDR::WriteStream& stream = call->Stream();
 	status_t result = fConnection->Send(stream.Buffer(), stream.Size());
 	if (result != B_OK) {
-		fRequests.FindRequest(xid);
+		fRequests.FindRequest(req->fXID);
 		delete req;
 		return result;
 	}
 
-	*request = req;
 	return B_OK;
 }
 
@@ -198,7 +211,9 @@ Server::_Listener()
 
 	while (!fThreadCancel) {
 		result = fConnection->Receive(&buffer, &size);
-		if (result != B_OK) {
+		if (result == B_NO_MEMORY)
+			continue;
+		else if (result != B_OK) {
 			fThreadError = result;
 			return result;
 		}
@@ -206,8 +221,7 @@ Server::_Listener()
 		Reply* reply = new(std::nothrow) Reply(buffer, size);
 		if (reply == NULL) {
 			free(buffer);
-			fThreadError = result;
-			return B_NO_MEMORY;
+			continue;
 		}
 
 		Request* req = fRequests.FindRequest(reply->GetXID());

src/add-ons/kernel/file_systems/nfs4/RPCServer.h

@@ -52,6 +52,7 @@ public:
 
 			status_t				SendCallAsync(Call* call, Reply** reply,
 										Request** request);
+			status_t				ResendCallAsync(Call* call, Request* req);
 	inline	status_t				WaitCall(Request* request,
 										bigtime_t time = kWaitTime);
 	inline	status_t				CancelCall(Request* request);

src/add-ons/kernel/file_systems/nfs4/Request.cpp

@@ -13,12 +13,17 @@
 status_t
 Request::Send()
 {
-	return _TrySend();
+	switch (fServer->ID().fProtocol) {
+		case ProtocolUDP:	return _SendUDP();
+		case ProtocolTCP:	return _SendTCP();
+	}
+
+	return B_BAD_VALUE;
 }
 
 
 status_t
-Request::_TrySend()
+Request::_SendUDP()
 {
 	RPC::Reply *rpl;
 	RPC::Request *rpc;
@@ -29,15 +34,51 @@ Request::_TrySend()
 
 	result = fServer->WaitCall(rpc);
 	if (result != B_OK) {
-		fServer->CancelCall(rpc);
-		delete rpc;
-		return result;
+		int attempts = 1;
+		while (result != B_OK && attempts++ < kRetryLimit)
+			result = fServer->ResendCallAsync(fBuilder.Request(), rpc);
+
+		if (attempts == kRetryLimit) {
+			fServer->CancelCall(rpc);
+			delete rpc;
+			return result;
+		}
 	}
 
 	return fReply.SetTo(rpl);
 }
 
 
+status_t
+Request::_SendTCP()
+{
+	RPC::Reply *rpl;
+	RPC::Request *rpc;
+
+	status_t result;
+	int attempts = 0;
+	do {
+		result = fServer->SendCallAsync(fBuilder.Request(), &rpl, &rpc);
+		if (result == B_NO_MEMORY)
+			return result;
+		else if (result != B_OK) {
+			fServer->Repair();
+			continue;
+		}
+
+		result = fServer->WaitCall(rpc);
+		if (result != B_OK) {
+			fServer->CancelCall(rpc);
+			delete rpc;
+
+			fServer->Repair();	
+		}
+	} while (result != B_OK && attempts++ < kRetryLimit);
+
+	return fReply.SetTo(rpl);
+}
+
+
 void
 Request::Reset()
 {

src/add-ons/kernel/file_systems/nfs4/Request.h

@@ -25,12 +25,15 @@ public:
 			void				Reset();
 
 private:
-			status_t			_TrySend();
+			status_t			_SendUDP();
+			status_t			_SendTCP();
 
 			RPC::Server*		fServer;
 
 			RequestBuilder		fBuilder;
 			ReplyInterpreter	fReply;
+
+	static	const int			kRetryLimit = 5;
 };