I think it's pretty well understood at this point that 0.9.8l does
the wrong thing when a client asks for a renegotiation (hangs the
negotiation, basically).  But it looks to me like 0.9.8-stable also
gets it wrong.

AFAICT by code inspection 0.9.8-stable (yesterday's snapshot) sends
a Fatal INVALID PARAMETER alert and ends the session.  That seems
wrong.

I believe a Warning NO RENEGOTIATION alert should be sent, and if the
client then wants to close, it can do so -- or not.  The code now seems
to disconnect clients in violation of the standard; is there really a
security reason to do this?

I started to implement this but then discovered that there's other code
which suppresses any attempt to send the NO RENEGOTIATION alert (commented
with "don't send :-)") -- what's with that?

It seems like, though the quick implementation of the new renegotiation
extension was very impressive, the handling of clients which try to do
unsafe renegotiations is either still quite broken, or I am suffering
from a serious misunderstanding of either the spec or the code.

Could someone from the OpenSSL team please comment on this?

Thor
______________________________________________________________________
OpenSSL Project                                 http://www.openssl.org
Development Mailing List                       [email protected]
Automated List Manager                           [email protected]

Reply via email to