Re: OpenSSL development (Heartbleed)

[Natasha Kerensikova <lithiumcat@instinctive.eu>]

On 2014-04-22, Dmitry A. Kazakov <mailbox@dmitry-kazakov.de> wrote:
> On Tue, 22 Apr 2014 16:57:28 +0000 (UTC), Simon Clubley wrote:
>> No, properly _implemented_ standards are what is required.
>> 
>> Heartbleed came about because a boundary check was missing which allowed
>> a invalid request to be processed instead of being rejected and, because
>> of the _implementation_, was allowed access to memory that had nothing to
>> do with the request.
>> 
>> This was a failure in the implementation of the standard, not a failure
>> of the standard itself.
>
> Boundary checks or not, the transport layer shall have no access to the
> server data.
>
> A tightly coupled system is vulnerable. If compromising just one component
> opens all gates wide, that is a bad standard and bad design. The effects of
> errors and faults must be bounded per design. 

How would you design a transport layer that has no access to whatever is
supposed to be transported?

"Heartbleed" didn't leak any data that ins't legitimataly needed by
OpenSSL (i.e. transported data and/or transport parameters (like keys))