Re: [lwip-users] Possible bug in src/api/tcpip.c

[Rishi Khan]

On May 4, 2008, at 6:42 AM, address@hidden wrote:

> > >  P.S. I've written a unix driver that actually binds to eth0 (or  
> > > other
> > > device) and operates there. This allows you to test inter-computer
> > > communication in unix (tap/tun seem to only work on the machine  
> > > running the
> > > program and no other machine can access this). Is this something  
> > > that is
> > > useful to the community?
> > You need to bridge the tun/tap interface with for instance eth0  
> > for it
> > to be available on the "real" network. This is my sh script that does
> > bridging:
>
> There you go, I _thought_ I had already used the unix port from  
> inside a VM to communicate with the guest OS (which means it is  
> visible to the outside network). Even so, I quite like the idea of  
> the other driver as it is more straightforward (and bridge-utils  
> aren't necessary). Is there any advandage of the tun/tap over the  
> raw socket approach?
>
> Simon

One issue about the direct ethernet driver is that you have to get an  
IP that exists on the current network that doesn't conflict. The tun/ 
tap driver may have similar problems if you're on the 192.168.0 network.

One disadvantage of my driver is that it attaches to the same device  
as the 'real' IP stack, so it gets first dibs at looking at the  
packet before passing it on (in the kernel) to my driver. This causes  
a little latency (not sure how much).

An advantage of my driver is that it will work with any OS with BSD  
type sockets (i.e. it supports E_P_ALL). The tun/tap is an extension  
that needs to be compiled as on in the kernel.

        -rishi