[Vadim Smirnov]

Windows Packet Filter demo package contains a sample named GRETunnel, which demonstrates how to attach/remove new headers to/from the network packets. Add GRE payload encryption and you get a simple VPN tunnel.

This another sample which with a different approach. It redirects selected TCP connections to the local proxy and then forwards theese through the SOCKS5 proxy. Just add an SSH client with SOCKS support (PuTTY, an example) and the result is VPN over SSH tunnel.

So, basically with winpkfilter you have everything needed to implement any type of VPN. The details depend on your concrete needs. An example, for Wireguard implementation you should insert/remove UDP headers (plus some protocol specific data) instead of GRE, but the idea is the same.

[Vadim Smirnov]

Yes, SendPacket/s return immediately.

[Vadim Smirnov]

Yes, entire library is thread safe.

The only thing to note about it is that there is only one instance of each network interface (setting mode, event and etc..) and only one copy of each packet (if one thread taken packet others threads won’t be able to get it). Though for one customer we created a special build with multiply filtering layers (e.g. packet injected on one layer can be picked up again on the next layer).

[Vadim Smirnov]

From the experiments I did, the only way I found to redirect some packets and let everything else pass is to use MSTCP_FLAG_TUNNEL for the adapter mode and then specify 2 filters: the first with the action FILTER_PACKET_REDIRECT that intercepts the packets I’m interested in, and then a second filter with action FILTER_PACKET_PASS to let everything else pass through. Am I correct?

Yes, you are right! There is also an alternative approach, when adapter is in tunnel mode then REDIRECT is a default action, so you can load one or more filters to pass selected traffic over and everything else will be redirected to your application automatically.

Or is there a way for example to set the adapter mode in something like “let everything pass” and then use a single filter with the action FILTER_PACKET_REDIRECT?

No, it won’t work. Adapter mode defines if network interface is filtered or not (independently in each direction). If it is not then loaded filters are not applied and all the traffic is passed over.

[Vadim Smirnov]

Below sf assigned a copy of the ft.StaticFilters[0]

var sf = ft.StaticFilters[0];

and then the copy is initialized. So, you should assign it back after initialization or define sf as a reference to ft.StaticFilters[0]

ref var sf = ref ft.StaticFilters[0];

[Vadim Smirnov]

That depends of which installer you have downloaded. Two of them (MSI ones) install driver only (x64 or x86 depending on the platform), the third one contains more demo binaries and includes ndisapi.dll.

The source code for ndisapi can be found here. You can use as a static or dynamic library (or even .net class library) depending on your requirements.

[Vadim Smirnov]

Support of NDIS 3.0 was removed from Windows 10 starting 1809, so the NDIS 3.0 of VirtNet can’t be used anymore. You can check this thread for the details and temporary NDIS 6.0 VirtNet driver replacement:

VirtNet does not install on Windows 10 version 1809 x64

However, this problem is not related to Windows Packet Filter, so if you have experienced any problems about it then could please provide the details.

[Vadim Smirnov]

No, in fact NDIS 6.x LWF is a direct replacement for NDIS 5.1 IM drivers.

[Vadim Smirnov]

In two words, NDIS IM is a NDIS 5.1 driver (though, it can be used in Vista, but in fact this is a compatibility mode) while LWF is NDIS 6.x and has a native support.

[Vadim Smirnov]

Yes, sure!

[Vadim Smirnov]

P.S. I was a little bit confused about your claim regarding code signing certificates costs, but yes, if you tried to order directly from the main page then it is a kind of expensive. However, if you try this link then you might be pleasantly surprised. 🙂

[Vadim Smirnov]

Well, yes, this is an option. However, it won’t protect if any other applications using standard winpkfilter build are running.

By the way we can sign your custom build with our code-signing certificate. Some of the customers prefer this option not only because of certificate costs but mostly because a relatively complex driver signing process.

[Vadim Smirnov]

Hmm, interesting question and I’m afraid I don’t have a quick answer. Inspecting all active processes for the open driver handle does not look a good idea. However, I think such functionality could be added to the driver itself, an example store the driver opened handles counter in the registry.

Still, I’m not sure that this type of functionality is really needed though, normally you should tie your NAT application with custom driver build. In this situation your NAT application always knows if it uses the driver or not while no other application are aware about the custom driver build therefore can’t use the driver.

[Vadim Smirnov]

Yes, you are right, it is bug. You should add UDP checksum recalculation in two places:

//DNS hook
//If we receive DNS packet on the NAT client adapter then we redirect it 
//to this system configured DNS server
if((pDlg->m_DNSIp.S_un.S_addr != INADDR_ANY) && (pDlg->m_DNSIp.S_un.S_addr != INADDR_NONE))
{
    if ((hAdapters[dwIndex]->m_NATState == CLIENT)&&
    (PacketBuffer.m_dwDeviceFlags == PACKET_FLAG_ON_RECEIVE))
    {
        if (ntohs(pUdpHeader->th_dport) == 53/*DNS port*/)
        {
            // Save the DNS IP used by the NAT client system
            hAdapters[dwIndex]->m_LocalDNS.S_un.S_addr = ntohl(pIpHeader->ip_dst.S_un.S_addr);
            
            pIpHeader->ip_dst.S_un.S_addr = pDlg->m_DNSIp.S_un.S_addr;
        
            if(bForceRouting) 
            {
                bNeedToBeRouted = pDlg->IsNeedToForceRouting(pEthHeader->h_dest, pIpHeader->ip_dst.S_un.S_addr, pProviderCard->m_Index);
            }

            RecalculateUDPChecksum(&PacketBuffer);
            RecalculateIPChecksum (&PacketBuffer);
        }
    }

    // DNS reply came, substitute source IP back to the original DNS address
    if ((hAdapters[dwIndex]->m_NATState == CLIENT)&&
        (PacketBuffer.m_dwDeviceFlags == PACKET_FLAG_ON_SEND))
    {
        if (ntohs(pUdpHeader->th_sport) == 53/*DNS port*/)
        {
            pIpHeader->ip_src.S_un.S_addr = htonl(hAdapters[dwIndex]->m_LocalDNS.S_un.S_addr);
            RecalculateUDPChecksum(&PacketBuffer);
            RecalculateIPChecksum (&PacketBuffer);
        }
    }
}

Internet Gateway is a very old sample and I have a newer NAT library implementation (not available for public though) where surprisingly this bug was already fixed.

[Vadim Smirnov]

Although sendto() can be called on unbound socket according MSDN “If the socket is unbound, unique values are assigned to the local association by the system, and the socket is then marked as bound”. So, I suspect that this is equivalent to calling bind explicitly. However, worth to test to ensure.

[Author]

Posts