Introduction

As of version 3.4, libtins will introduce a new set of classes which allow reassembling TCP streams in a very simple, yet powerful, way. Before this classes were introduced, there was a TCPStreamFollower class which sort of did this, but in a not-so-extensible not usable way. Note that this code is currently available on the master branch of the repository.

These new classes aim to provide a really simple way to follow streams, process their data, get their attributes and more, using a simple callback based interface. The streams will handle out-of-order data, reassemble it and let the user process it without having to deal with packets, payloads, sequence numbers, etc.

All of this classes require using C++11, as they use std::function as a way to specify callbacks. Therefore, you should use some fairly recent compiler to use it. If you're using GCC, 4.6 is probably enough, maybe even older versions.

StreamFollower

The main class you should know about is StreamFollower. This class will process TCP packets, looking at the IP addresses and ports used in them. Whenever a new 4-tuple (client address, client port, server address, server port) is seen, it will create some context for that TCP stream and execute a user-provided callback to notify the creation of it. After that, all packets that belong to that stream will be forwarded to the right object, letting it process data and update its internal state.

The other responsibility of the StreamFollower is detecting that something's going wrong in a stream. Let's say you have a high packet loss (e.g. your program fails to process the packets fast enough), you don't want to keep buffering data for streams that will never be reassembled, or store state and data for a stream that was actually closed but the FIN/RST packets weren't captured. For this reason, this class will detect these events (too many buffered packets, stream timeouts, etc) and delete their state if this happens.

As a simple example of how to create a StreamFollower and set up some callbacks:

#include <tins/tcp_ip/stream_follower.h>

using Tins::TCPIP::Stream;
using Tins::TCPIP::StreamFollower;

// New stream is seen
void on_new_stream(Stream& stream) {

}

// A stream was terminated. The second argument is the reason why it was terminated
void on_stream_terminated(Stream& stream, StreamFollower::TerminationReason reason) {

}


// Create our follower
Tins::TCPIP::StreamFollower follower;

// Set the callback for new streams. Note that this is a <i>std::function</i>, so you
// could use <i>std::bind</i> and use a member function for this
follower.new_stream_callback(&on_new_stream);

// Now set up the termination callback. This will be called whenever a stream is 
// stopped being followed for some of the reasons explained above
follower.stream_termination_callback(&on_stream_terminated);

// Now create some sniffer
Sniffer sniffer = ...;

// And start sniffing, forwarding all packets to our follower
sniffer.sniff_loop([&](PDU& pdu) {
    follower.process_packed(pdu);
    return true;
})

Note that StreamFollower::process_packet has another overload that takes a Packet. You should try to use this overload as it will make streams timeout at the actual packet time rather than using the system clock.

So that is the first step into reassembling TCP streams. In the next sections, we'll see how to do something useful with that.

Using Stream

Once you've configured the callback for new streams on the StreamFollower, you probably want to do something with that new stream. Streams allow you to configure callbacks for different events that occur on the stream.

Data events are generated whenever there's new, ready to be processed, data. This means that a packet with the next expected sequence number arrived and its payload is available plus all of the out-of-order payload that could have been received before and couldn't be processed because that first packet's data was missing.

You can optionally subscribe to both client and server data events. This means that you can be notified on a different callback whenever there's new data from the client or from the server on each stream. Let's use this in a short example:

 
// This will be called when there's new client data
void on_client_data(Stream& stream) {
    // Get the client's payload, this is a vector<uint8_t>
    const Stream::payload_type& payload = stream.client_payload();

    // Now do something with it!
}

// This will be called when there's new server data
void on_server_data(Stream& stream) {
    // Process the server's data
}

// New stream is seen
void on_new_stream(Stream& stream) {
    // Configure the client and server data callbacks
    stream.client_data_callback(&on_client_data);
    stream.server_data_callback(&on_server_data);

    // Done!
}

So that's it, the StreamFollower we constructed before will keep processing packets and forwarding them to the right Stream objects, which will execute those callbacks when appropriate.

You can also subscribe to other events on each Stream. One of them is the close event, which is executed whenever the stream is properly closed. You can do this by calling Stream::stream_closed_callback.

Handling Streams' data

So now that we've seen the basics on how to use streams, lets see some other features.

By default, whenever new data is available for a stream, that data will be moved to the stream's payload, the data callback is going to be executed, and then that data is going to be erased. This is done so that data doesn't start buffering, making memory usage go up until the stream is closed (or you run out of memory). In case you want to buffer the data and use your own handling of it, then you should call the following functions:

 
// New stream is seen
void on_new_stream(Stream& stream) {
    // Disables auto-deleting the client's data after the callback is executed
    stream.auto_cleanup_client_data(true);

    // Same thing for the server's data
    stream.auto_cleanup_server_data(true);

    // Or a shortcut to doing this for both:
    stream.auto_cleanup_payloads(true);
}

If you're only planning on processing say the client's data and not the server's then you should call ignore_client/server_data. Otherwise, even if you don't set a callback, the data will still be buffered and re-ordered as needed:

 
// New stream is seen
void on_new_stream(Stream& stream) {
    // We don't even want to buffer the client's data
    stream.ignore_client_data():
}

Conclusion

This should have given you a fairly good introduction on how to use the StreamFollower and Stream classes. You can check out the HTTP requests example to see a pretty simple example of this being put into practice.

Previous part: Sending
Next part: Protocols