Re: [PATCH] Always list SSL in case any SSL connection succeeded

[Kristof Boeynaems <kristof.boeynaems () gmail com>]

David Fifield wrote:
> On Sun, Feb 22, 2009 at 12:04:53AM +0100, Kristof Boeynaems wrote:
>   
> > A quick patch for the 'bug' described in  
> > http://seclists.org/nmap-dev/2009/q1/0484.html.
> >
> > "There is still a 'bug', where all detected SSL information will be
> > thrown away in case the SSL server does not respond at all, beyond the
> > SSL handshake. However, because the SSL connections succeeded, Nmap
> > should list these situations as 'ssl/unknown'.
> >
> > This bug can very easily be reproduced by setting up your own OpenSSL  
> > server as follows:
> >       openssl s_server -cert 
> > /usr/share/doc/libssl-dev/demos/sign/cert.pem -key 
> > /usr/share/doc/libssl-dev/demos/sign/key.pem
> >
> > This OpenSSL server will listen at 4433 by default, and will not return  
> > anything beyond an SSL connection."
> >     
>
> <snip>
>
> I think the bug could be fixed in a better way, though. The code already
> passes (*svc)->tunnel in the case of no match, so the fact that SSL was
> detected is already recorded. I think you should rather patch
> getNmapServiceName in output.cc. That's the function that builds up a
> name from the service info.
>   

Thanks for looking into this.
In fact, trying to adapt getNmapServiceName() was the first thing I 
tried, but that does not work, as apparently that function never gets 
called in the case described above. It seems that Nmap gives up on the 
ssl tunnel completely, even before that function.

I followed the code a bit further, and it turns out that the tunnel that 
getNmapServiceName() ultimately receives is set by 
getServiceProbeProbeResults() in portlist.cc.
This function in turn gets it from 'serviceprobe_tunnel', which is 
initialized with the value of SERVICE_TUNNEL_NONE.
It seems that the only way that 'serviceprobe_tunnel' (and hence 
ultimately the tunnel description in getNmapServiceName()) can be 
changed, is through a call to setServiceProbeResults().
This function is called from nse_nmaplib.cc, and from service_scan.cc. 
But, this function is *never* called in the test case of above. Thus the 
tunnel in these cases defaults to SERVICE_TUNNEL_NONE.

As far as I can see, processResults() in service_scan.cc is the best 
place to solve this, by calling setServiceProbeResults() there, even in 
the test case above. It is in that place that the other calls to 
setServiceProbeResults() happen, so it seemed logical to add it there.

Nevertheless, if you have other suggestions, I am willing to try them as 
well.

> One more thing: In the test you described, the output should be
> "ssl/unknown?", not "ssl/unknown". Leaving off the question mark makes
> it look as if the port was positively identified. It's confusing in this
> case because the port is named "unknown", but that name comes from the
> nmap-services file. If you repeat the s_server experiment with port 80
> you'll see what I mean. The output should be "ssl/http?", not "ssl/http"
> or "http?".
>   

Good point. I'll look into that once we have agreed on where we should 
make the change :)

Thanks,

Kristof

_______________________________________________
Sent through the nmap-dev mailing list
http://cgi.insecure.org/mailman/listinfo/nmap-dev
Archived at http://SecLists.Org