I'm using C# Socket class to send/receive TCP messages to/from our server by short connection. The pseudo-code (exclude some false tolerant logic to make it clear) is as followings.
class MyTCPClient {
private Socket mSocket;
MyTCPClient() {
mSocket = new Socket(AddressFamily.InterNetwork, SocketType.Stream, ProtocolType.Tcp);
}
void Connect(ipEndpoint, sendTimeOut, receiveTimeOut) {
mSocket.Connect(ipEndpoint);
mSocket.SetSocketOption(SocketOptionLevel.Socket, SocketOptionName.SendTimeout, sendTimeOut);
mSocket.SetSocketOption(SocketOptionLevel.Socket, SocketOptionName.ReceiveTimeout, receiveTimeOut);
}
void Send(msg) {
mSocket.Send(msg);
}
void Receive(data) {
ByteStreamWriter bsw = new ByteStreamWriter();
int totalLen = 0;
while (true) {
byte[] temp = new byte[1024];
int len = mSocket.Receive(temp, SocketFlags.None);
System.Threading.Thread.Sleep(50); // This is the weirdest part
if (len <= 0) break;
totalLen += len;
bsw.WriteBytes(temp);
}
byte[] buff = bsw.GetBuffer();
data = new byte[totalLen];
Array.Copy(buff, 0, data, 0, totalLen);
}
~MyTCPClient() {
mSocket.Close();
}
}
I was using this class to request the same message from our server several times by short connection, and the following things occurred (only when the message size was larger than one MTU -- 1500 bytes).
If the "Sleep(50)" was commented out, the most of the times (90%) I received wrong "data", specifically to say, the "totalLen" was right, but the "data" was wrong.
If I replaced "Sleep(50)" to "Sleep(10)", about half of the times I received wrong "data", and "totalLen" was also right always.
If I used "Sleep(50)", occasionally, I received wrong "data".
I can guarantee that, every time, our server sent the right data, and the client received the right data too at TCP layer (I used WireShark to monitor all messages through the port I used). Is there anyone who can help to answer why my C# code cannot get the right data?
Also, if I use mSocket.Available instead of the return value of Socket.Receive() to judge the while loop, I would always get the right data and data length...
You don't truncate temp to len bytes before writing it. All the bytes after len in temp have not been filled with current data and thus contain nonsensical, stale date.
When copying to another stream you could use stream.Write(temp, 0, len)
Related
I'm trying to send multiple files over TCP using C# TcpClient, for a file below 64kB it works great, but when I have more it throws an exeption that host-computer disconnected.
Here is my code:
Server side(sending).
1) SocketServer class
public abstract class SocketServer
{
private Socket serverSocket;
public SocketServer(IPEndPoint localEndPoint)
{
serverSocket = new Socket(AddressFamily.InterNetwork, SocketType.Stream, ProtocolType.Tcp);
serverSocket.Bind(localEndPoint);
serverSocket.Listen(0);
serverSocket.BeginAccept(BeginAcceptCallback, null);
}
private void BeginAcceptCallback(IAsyncResult ar)
{
Socket clientSocket = serverSocket.EndAccept(ar);
Console.WriteLine("Client connected");
ClientConnection clientConnection = new ClientConnection(this, clientSocket);
Thread clientThread = new Thread(new ThreadStart(clientConnection.Process));
clientThread.Start();
serverSocket.BeginAccept(BeginAcceptCallback, null);
}
internal abstract void OnReceiveMessage(ClientConnection client, byte header, byte[] data);
}
Here goes ClientConnection:
public class ClientConnection
{
private SocketServer server;
private Socket clientSocket;
private byte[] buffer;
private readonly int BUFFER_SIZE = 8192;
public ClientConnection(SocketServer server, Socket clientSocket)
{
this.server = server;
this.clientSocket = clientSocket;
buffer = new byte[BUFFER_SIZE];
}
public void Process()
{
clientSocket.BeginReceive(buffer, 0, buffer.Length, SocketFlags.Peek, BeginReceiveCallback, null);
}
private void BeginReceiveCallback(IAsyncResult ar)
{
int bytesReceived = clientSocket.EndReceive(ar);
if (bytesReceived >= 4)
{
clientSocket.Receive(buffer, 0, 4, SocketFlags.None);
// message size
int size = BitConverter.ToInt32(buffer, 0);
// read message
int read = clientSocket.Receive(buffer, 0, size, SocketFlags.None);
// if data still fragmented, wait for it
while (read < size)
{
read += clientSocket.Receive(buffer, read, size - read, SocketFlags.None);
}
ProcessReceivedData(size);
}
clientSocket.BeginReceive(buffer, 0, buffer.Length, SocketFlags.Peek, BeginReceiveCallback, null);
}
private void ProcessReceivedData(int size)
{
using (PacketReader pr = new PacketReader(buffer))
{
// message header = 1 byte
byte header = pr.ReadByte();
// next message data
byte[] data = pr.ReadBytes(size - 1);
server.OnReceiveMessage(this, header, data);
}
}
public void Send(byte[] data)
{
// first of all, send message length
clientSocket.Send(BitConverter.GetBytes(data.Length), 0, 4, SocketFlags.None);
// and then message
clientSocket.Send(data, 0, data.Length, SocketFlags.None);
}
}
Implementation of FileServer class:
public class FileServer : SocketServer
{
string BinaryPath;
public FileServer(IPEndPoint localEndPoint) : base(localEndPoint)
{
BinaryPath = Path.GetDirectoryName(System.Reflection.Assembly.GetExecutingAssembly().Location);
}
internal override void OnReceiveMessage(ClientConnection client, byte hdr, byte[] data)
{
// Convert header byte to ENUM
Headers header = (Headers)hdr;
switch(header)
{
case Headers.Queue:
Queue(client, data);
break;
default:
Console.WriteLine("Wrong header received {0}", header);
break;
}
}
private void Queue(ClientConnection client, byte[] data)
{
// this message contains fileName
string fileName = Encoding.ASCII.GetString(data, 1, data.Length - 1);
// combine path with assembly location
fileName = Path.Combine(BinaryPath, fileName);
if (File.Exists(fileName))
{
FileInfo fileInfo = new FileInfo(fileName);
long fileLength = fileInfo.Length;
// pass the message that is now start a file transfer, contains:
// 1 byte = header
// 16 bytes = file length
using (PacketWriter pw = new PacketWriter())
{
pw.Write((byte)Headers.Start);
pw.Write(fileLength);
client.Send(pw.GetBytes());
}
//
using (FileStream fs = new FileStream(fileName, FileMode.Open, FileAccess.Read))
{
int read = 0, offset = 0;
byte[] fileChunk = new byte[8191];
while (offset < fileLength)
{
// pass message with file chunks, conatins
// 1 byte = header
// 8195 bytes = file chunk
using (PacketWriter pw = new PacketWriter())
{
fs.Position = offset;
read = fs.Read(fileChunk, 0, fileChunk.Length);
pw.Write((byte)Headers.Chunk);
pw.Write(fileChunk, 0, read);
client.Send(pw.GetBytes());
}
offset += read;
}
}
}
}
}
And helper classes:
public class PacketWriter : BinaryWriter
{
private MemoryStream memoryStream;
public PacketWriter() : base()
{
memoryStream = new MemoryStream();
OutStream = memoryStream;
}
public byte[] GetBytes()
{
byte[] data = memoryStream.ToArray();
return data;
}
}
public class PacketReader : BinaryReader
{
public PacketReader(byte[] data) : base(new MemoryStream(data))
{
binaryFormatter = new BinaryFormatter();
}
}
Client have the almost same code. Except receive:
internal override void OnReceiveMessage(ServerConnection client, byte hdr, byte[] data)
{
Headers header = (Headers)hdr;
switch(header)
{
case Headers.Start:
Start(data); // save length of file and prepare for receiving it
break;
case Headers.Chunk:
Chunk(data);
break;
default:
Console.WriteLine("Wrong header received {0}", header);
break;
}
}
private void Chunk(byte[] data)
{
// Process reveived data, write it into a file
}
Do TCP sockets automatically close after 64kB send?
No, TCP sockets do not automatically close after 64 KB of sent data, nor after any amount of sent data. They remain open until either end closes the connection or a network error of some sort happens.
Unfortunately, your question is extremely vague, providing practically no context. But I will point out one potentially serious bug in your client code: you are not restricting the data read by the client to that which you expect for the current file. If the server is sending more than one file, it is entirely possible and likely that on reaching the end of the data for the current file, a read containing that final sequence of bytes will also contain the initial sequence of bytes for the next file.
Thus, the expression totalBytesRead == fileLenght will fail to ever be true and your code will attempt to read all of the data the server sends as if it's all part of the first file.
One easy way to check for this scenario is to look at the size of the file that was written by the client. It will be much larger than expected, if the above is the issue.
You can fix the code by only ever receiving at most the number of bytes you actually expect to be remaining:
while ((bytesRead = binaryReader.Read(
buffer, 0, Math.Min(fileLenght - totalBytesRead, buffer.Length))) > 0)
{
fs.Write(buffer, 0, bytesRead);
totalBytesRead += bytesRead;
if (totalBytesRead == fileLenght)
{
break;
}
}
While you're at it, you might fix the spelling of the variable named fileLenght. I know Intellisense makes it easy to type the variable name whether it's spelled right or not, but if you one day have to grep through the source code looking for code that involves itself with "length" values, you'll miss that variable.
EDIT:
Having pointed out the problem in the code you posted with your question originally, I now find myself looking at completely different code, after your recent edit to the question. The example is still not a good, minimal, complete code example and so advice will still have to be of a limited nature. But I can see at least two major problems in your receive handling:
You are mixing Receive() and BeginReceive(). This is probably not too terrible in general; what makes it really bad in your case is that in the completion handler for the initial asynchronous receive, you are blocking the thread while you use the synchronous Receive() to (attempt to) receive the rest of the data, and then to make matters worse, in the FileServer case you continue to block that thread while you transmit the file data to the client.I/O completion handlers must not block the thread for any significant amount of time; they must retrieve the data for the completed operation, and return as quickly as is feasible. There's a lot of wiggle room, but not enough to justify tying up the IOCP thread for the entire file transfer.
While the first issue is bad, it's probably not bad enough to actually cause the connection to reset. In a low-volume scenario, and especially if you're just dealing with one client at a time at the moment, I doubt blocking the IOCP thread that's being used to call your I/O completion callback will cause any easily observable problems.What's much worse is that you allocate the buffer array once, with a length of 8192, and then proceed to attempt to stuff the entire counted-byte stream of bytes being sent. On the assumption that the client will be sending short messages, this is probably fine on the server side. But on the client side, the value of size can easily be greater than the 8192 bytes allocated in the buffer array.I find it very likely that your BeginReceiveCallback() method is throwing an ArgumentOutOfRangeException after receiving the first 8192 bytes, due to the fact that you are passing the Receive() method a size value that exceeds the length of the buffer array. I can't prove that, because the code example is incomplete. You'll have to follow up on that theory yourself.
As long as I'm reviewing the code you posted, some other irregularities stood out as I scanned through it:
It's a bit odd to be using a buffer length of 8191 when transmitting the file.
Your BeginReceiveCallback() method discards whatever bytes were sent initially. You call EndReceive(), which returns the count of bytes received, but the code that follows assumes that the size value in the stream is still waiting to be read, when it very easily could have been sent in the first block of bytes. Indeed, your comparison bytesReceived >= 4 seems to be intended to refrain from processing received data until you have that length value. Which is wrong for another reason:
If your header is 1 byte followed by a 4-byte count-of-bytes value, then you really need bytesReceived > 4. But worse, TCP is allowed to drip out a single byte at a time if it really wants to (it won't, but if you write your code correctly, it won't matter). Since you only process received data if you get at least 4 bytes, and since if you don't get 4 bytes, you just ignore whatever was sent so far, you could theoretically wind up ignoring all of the data that was sent.
That you have the risk that any data could be ignored at all is terrible. But #2 above guarantees it, and #3 just adds insult to injury.
When receiving the size value, you are reading a 32-bit integer, but when sending the size value (i.e. for the file itself), you are writing a 64-bit integer. You'll get the right number on the receiving end, because BinaryWriter uses little-endian and that means that the first four bytes of any 64-bit integer less than 2^31 are identical to the four bytes of the 32-bit integer representation of the same number. But you'll still wind up with four extra zero bytes after that; if treated as file data as they likely would, that would corrupt your file.
I had never used UDP before, so I gave it a go. To see what would happen, I had the 'server' send data every half a second, and the client receive data every 3 seconds. So even though the server is sending data much faster than the client can receive, the client still receives it all neatly one by one.
Can anyone explain why/how this happens? Where is the data buffered exactly?
Send
class CSimpleSend
{
CSomeObjServer obj = new CSomeObjServer();
public CSimpleSend()
{
obj.changedVar = varUpdated;
obj.threadedChangeSomeVar();
}
private void varUpdated(int var)
{
string send = var.ToString();
byte[] packetData = System.Text.UTF8Encoding.UTF8.GetBytes(send);
string ip = "127.0.0.1";
int port = 11000;
IPEndPoint ep = new IPEndPoint(IPAddress.Parse(ip), port);
Socket client = new Socket(AddressFamily.InterNetwork, SocketType.Dgram, ProtocolType.Udp);
client.SendTo(packetData, ep);
Console.WriteLine("Sent Message: " + send);
Thread.Sleep(100);
}
}
All CSomeObjServer does is increment an integer by one every half second
Receive
class CSimpleReceive
{
CSomeObjClient obj = new CSomeObjClient();
public Action<string> showMessage;
Int32 port = 11000;
UdpClient udpClient;
public CSimpleReceive()
{
udpClient = new UdpClient(port);
showMessage = Console.WriteLine;
Thread t = new Thread(() => ReceiveMessage());
t.Start();
}
private void ReceiveMessage()
{
while (true)
{
//Thread.Sleep(1000);
IPEndPoint remoteIPEndPoint = new IPEndPoint(IPAddress.Any, port);
byte[] content = udpClient.Receive(ref remoteIPEndPoint);
if (content.Length > 0)
{
string message = Encoding.UTF8.GetString(content);
if (showMessage != null)
showMessage("Recv:" + message);
int var_out = -1;
bool succ = Int32.TryParse(message, out var_out);
if (succ)
{
obj.alterSomeVar(var_out);
Console.WriteLine("Altered var to :" + var_out);
}
}
Thread.Sleep(3000);
}
}
}
CSomeObjClient stores the variable and has one function (alterSomeVar) to update it
Ouput:
Sent Message: 1
Recv:1
Altered var to :1
Sent Message: 2
Sent Message: 3
Sent Message: 4
Sent Message: 5
Recv:2
Altered var to :2
Sent Message: 6
Sent Message: 7
Sent Message: 8
Sent Message: 9
Sent Message: 10
Recv:3
Altered var to :3
The operating system kernel maintains separate send and receive buffers for each UDP and TCP socket. If you google SO_SNDBUF and SO_RCVBUF you'll find lots of information about them.
When you send data, it is copied from your application space into the send buffer. From there it is copied to the network interface card, and then onto the wire. The receive side is the reverse: NIC to receive buffer, where it waits until you read it. Additionally copies and buffering can also occur, depending on the OS.
It is critical to note that the sizes of these buffers can vary radically. Some systems might default to as little as 4 kilobytes, while others give you 2 megabytes. You can find the current size using getsockopt() with SO_SNDBUF or SO_RCVBUF and likewise set it using setsockopt(). But many systems limit the size of the buffer, sometimes to arbitrarily small amounts. This is typically a kernel value like net.core.wmem_max or net.core.rmem_max, but the exact reference will vary by system.
Also note that setsockopt() can fail even if you request an amount less than the supposed limit. So to actually get a desired size, you need to repeatedly call setsockopt() using decreasing amounts until it finally succeeds.
The following page is a Tech Note from my company which touches on this topic a little bit and provides references for some common systems: http://www.dataexpedition.com/support/notes/tn0024.html
It looks to me like the UdpClient-Class provides a buffer for received data. Try using a socket directly. You might also want to set that sockets ReceiveBufferSize to zero, even though I believe it is only used for TCP connections.
I'm sending sensor data from an android device to a TCP server in C#. The android client sends the data in fixed size chunks of 32 bytes.
In the server I read the data expecting it would come in full packs, but since TCP is a stream protocol some messages arrive at the server split in two parts. I know that because I can watch it with a software called SocketSniff.
The problem is that I don't know how to handle it on my server.
All examples I found use the NetworkStream.Read(), in this method I have to pass a array of bytes to store the data read, an offset and the number of bytes to read. This array of bytes must have a know size, in my case 32.
I don't know the real size of the message that arrived on my server, but it could be one of the following situations.
If the received data size is 32 bytes, it's all OK.
If the received data size if greater than 32 bytes, I think I'm loosing data.
If the received data size is less than 32 bytes, lets say 20 bytes, these bytes are stored in my array and the last 12 bytes of the array remain with the value of zero. Since I may be really receiving some zeros there's no way to know the size I really received, so I can't merge it with the remaining data which should come in the next reading.
My code which handles the receiving is the following:
int buffer = 32;
...
private void HandleClientComm(object client)
{
TcpClient tcpClient = (TcpClient)client;
NetworkStream clientStream = tcpClient.GetStream();
byte[] message = new byte[buffer];
int bytesRead;
while (true)
{
bytesRead = 0;
try
{
bytesRead = clientStream.Read(message, 0, message.Length);
}
catch
{
break;
}
if (bytesRead == 0)
{
// Connection closed
break;
}
SensorData sensorData = ProcessTcpPacket(message);
}
tcpClient.Close();
}
Is there any way to know the size of the data I'm receiving in the socket?
Well, yes, you have the bytesRead variable - it holds the number of bytes read from the stream. You will read at most message.Length bytes, but you may read less.
Note that if there are more bytes available, you will not lose them by reading just message.Length bytes. Rather, they will be available for you next time you read from the stream.
What you need to do is add another while loop liked this:
int messageRead = 0;
while(messageRead < message.Length)
{
int bytesRead = clientStream.Read(message, messageRead, message.Length - messageRead);
messageRead += bytesRead;
if(bytesRead==0)
return; // The socket was closed
}
// Here you have a full message
I have a simple TCP server that communicates with some devices via GPRS. It works correctly without any problem for a long time. Nowdays there is a change in the devices (client devices, that send data) and they send lot more data than earlier and TCP message size can be 10 times bigger than earlier. Earlier it was max 1024 bytes/data message, now it can be more than 10 thousand bytes/message.
As I see - as the nature of TCP predicted - there are a lot of messages that are 'broken', so instead of a complete message only a part of that arrives, and later can come the other part etc.
I think my server - that is based on asynchronous pattern - handles this situation in a wrong way and I think the bug is in my RecieveCallBack function. I could not find to correct way to handle this situation. How can I do this?
Here is my RecieveCallBack function:
public void RecieveCallBack(IAsyncResult ar)
{
StateObject state = (StateObject)ar.AsyncState;
try
{
Socket handler = state.WorkSocket;
int read = handler.EndReceive(ar);
if (read > 0)
{
var data = new byte[read];
Array.Copy(state.Buffer, 0, data, 0, read);
}
else
{
if ((handler.Connected == false) || (handler.Available == 0))
{
Close(state);
return;
}
}
}
catch (System.Net.Sockets.SocketException)
{
Close(state);
}
catch (System.Exception exc)
{
Debug.Assert(false, exc.Message);
HandleSocketError(state, exc);
}
}
.net4/c#/vs2010
Thanks
UPDATE:
Clients are connected as they can be connected, so it is up to the GSM network. They can be connected even for days and they send data at every minutes. The protocol of the client devices is different, there are different kind of devices with different kind of protocols. One of them has delimiter and CRC at the end of the message, others does not have.
You need message framing. The protocol must specify how large the messages are - usually either a constant well-known size, a length prefix, or using message delimiters.
The is no such thing as packets in TCP. It is a stream oriented protocol. That means when you read the socket you can get less data than you're expecting. You need to check the read size and if you got a short read then read again for the rest of the data.
For an example in C:
int read_data(int sock, int size, unsigned char *buf) {
int bytes_read = 0, len = 0;
while (bytes_read < size &&
((len = recv(sock, buf + bytes_read,size-bytes_read, 0)) > 0)) {
bytes_read += len;
}
if (len == 0 || len < 0) doerror();
return bytes_read;
}
I am not sure how the rest of the application looks but I think the problem is with the following line :-
Array.Copy(state.Buffer, 0, data, 0, read);
If a packet is fragmented i.e the receive callback is called multiple times for the same connection the state.Buffer will be overwritten with the last received packet, this as you copy data to offset 0 in state.Buffer.
Depending on what your needs are you need to probably need more state in StateObject :) so you can append the data rather than overwrite.
Hope this hint helps.
Problem:
When I do something like this:
for (int i = 0; i < 100; i++)
{
SendMessage( sometSocket, i.ToString());
Thread.Sleep(250); // works with this, doesn't work without
}
With or without the sleep the server logs sending of separate messages. However without the sleep the client ends up receiving multiple messages in single OnDataReceived so the client will receive messages like:
0,
1,
2,
34,
5,
678,
9 ....
Server sending Code:
private void SendMessage(Socket socket, string message)
{
logger.Info("SendMessage: Preparing to send message:" + message);
byte[] byteData = Encoding.ASCII.GetBytes(message);
if (socket == null) return;
if (!socket.Connected) return;
logger.Info("SendMessage: Sending message to non " +
"null and connected socket with ip:" + socket.RemoteEndPoint);
// Record this message so unit testing can very this works.
socket.Send(byteData);
}
Client receiving code:
private void OnDataReceived(IAsyncResult asyn)
{
logger.Info("OnDataReceived: Data received.");
try
{
SocketPacket theSockId = (SocketPacket)asyn.AsyncState;
int iRx = theSockId.Socket.EndReceive(asyn);
char[] chars = new char[iRx + 1];
System.Text.Decoder d = System.Text.Encoding.UTF8.GetDecoder();
int charLen = d.GetChars(theSockId.DataBuffer, 0, iRx, chars, 0);
System.String szData = new System.String(chars);
logger.Info("OnDataReceived: Received message:" + szData);
InvokeMessageReceived(new SocketMessageEventArgs(szData));
WaitForData(); // .....
Socket Packet:
public class SocketPacket
{
private Socket _socket;
private readonly int _clientNumber;
private byte[] _dataBuffer = new byte[1024]; ....
My hunch is it's something to do with the buffer size or its just the between the OnDataReceived and EndReceive we're getting multiple messages.
Update: It turns out when I put a Thread.Sleep at the start of OnDataReceived it gets every message. Is the only solution to this wrapping my message in a prefix of length and an string to signify the end?
This is expected behaviour. A TCP socket represents a linear stream of bytes, not a sequence of well-delimited “packets”. You must not assume that the data you receive is chunked the same way it was when it was sent.
Notice that this has two consequences:
Two messages may get merged into a single callback call. (You noticed this one.)
A single message may get split up (at any point) into two separate callback calls.
Your code must be written to handle both of these cases, otherwise it has a bug.
There is no need to abandon Tcp because it is stream oriented.
You can fix the problems that you are having by implementing message framing.
See
http://blogs.msdn.com/malarch/archive/2006/06/26/647993.aspx
also:
http://nitoprograms.blogspot.com/2009/04/message-framing.html
TCP sockets don't always send data right away -- in order to minimize network traffic, TCP/IP implementations will often buffer the data for a bit and send it when it sees there's a lull (or when the buffer's full).
If you want to ensure that the messages are processed one by one, you'll need to either set socket.NoDelay = true (which might not help much, since data received may still be bunched up together in the receive buffer), implement some protocol to separate messages in the stream (like prefixing each message with its length, or perhaps using CR/LF to separate them), or use a message-oriented protocol like SCTP (which might not be supported without additional software) or UDP (if you can deal with losing messages).