Re: Not so random TCP initial sequence numbers

[Florian Weimer <Florian.Weimer () RUS UNI-STUTTGART DE>]

Elias Levy <aleph1 () SECURITYFOCUS COM> writes:

> It seems the vulnerability lies in the implementation of some TCP/IP
> stacks that attempt to randomize TCP's initial sequence numbers - ironically
> for the purpose of not generating predictable ISNs to stop blind IP spoofing
> of TCP connections. While the ISNs generated by these implementations appear
> random they apparently are statistically predictable.

I think this is the cause.  For example, Solaris 2.6 uses a PRNG when
the "tcp_strong_iss" sysctl has the value 1.  The PRNG output (and the
ISN derived from it) appears pretty random to the casual observer
(e.g. nmap), but with a more sophisticated approach, it should be
possible to recover the internal state of the PRNG.

If "tcp_strong_iss" is set to 2, the RFC 1948 approach is implemented,
which is probably secure.

For an example how to set "tcp_strong_iss" properly, see 'Example G'

http://www.enteract.com/~lspitz/example.html

--
Florian Weimer                    Florian.Weimer () RUS Uni-Stuttgart DE
University of Stuttgart           http://cert.uni-stuttgart.de/
RUS-CERT                          +49-711-685-5973/fax +49-711-685-5898