Re: [NSE] HUGE ssl-enum-ciphers speed improvement

[Daniel Miller <bonsaiviking () gmail com>]

On Thu, Jul 12, 2012 at 6:25 PM, David Fifield <david () bamsoftware com> wrote:
> On Thu, Jul 12, 2012 at 04:37:56PM -0500, Daniel Miller wrote:
> > Previously, the script took this algorithm: For each of the 4
> > protocol versions (SSLv3, TLSv1.0, TLSv1.1, TLSv1.2), start 213
> > threads, one for each Cipher Suite. When all threads for one
> > protocol have finished, process them. If the protocol was not
> > supported, discard results. If at least one Cipher Suite was
> > supported, do the same thing for the 3 compressors. Besides the
> > primary improvement, this means 212 extra connections per
> > unsupported protocol (since support for the protocol should be
> > detectable in one connection).
> >
> > The SSL protocol allows clients to specify up to 2^16-1 cipher
> > suites in one Client Hello, and the server must choose one that it
> > supports. I chose to reimplement the script to remove the one suite
> > chosen by the server and retry. If the server supports 7 cipher
> > suites, then the 8th Client Hello will be rejected, and the script
> > can stop trying. To retain some of the multithreading benefits, I
> > chose to implement each protocol in a separate thread, retaining a
> > linear execution within each protocol. This lets me evaluate after
> > each Hello whether to keep trying or to finish.
>
> You should check if this is the same change Mak Kolybabi tried in 2010:
> http://seclists.org/nmap-dev/2010/q1/650
> (Look for "...it starts by offering all ciphers at once...".)
>
> There was some problem with this method, the details of which I don't
> remember, but you should try some of the test cases in this thread:
> http://seclists.org/nmap-dev/2010/q1/859
>
> David Fifield

David,

You are right, this is essentially the same method he was using. I
don't see very well-defined test cases there, but I did discover that
windowsupdate.microsoft.com shows more ciphers with the slow method
than the fast one. I'll work on getting that one down, and then see if
anyone else can find example servers. Perhaps I'll hit shodanhq.com
and get a cross-section of servers to test against. FWIW, ncat --ssl
and openssl s_server both respond the same to both methods.

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