From: Bill Kelly <billk@...>
Date: 2014-01-26T21:51:05-08:00
Subject: [ruby-core:60131] Re: [ruby-trunk - Bug #9424] ruby 1.9 & 2.x has insecure SSL/TLS client defaults

shyouhei@ruby-lang.org wrote:
>
> Alright, I see you are not blaming Debian people.  Thank you.
> 
> But I see "I don't know what these lines of code in OpenSSL do, but Valgrind
> complains." is a completely valid reason to fix something.  In fact I have
> just read the patch denbian introduced and still see no problem on it.  So I
> can't draw a line between #1 and #2.  They are equally true.  And the history
> tells the patch was wrong; how can you say #2 is OK?

To me, the difference hinges on the premise that the defaults function by
selecting from a set of features which are intended to be enabled or disabled
/by design/.

For example, the first time my servers were scanned with tools like Nessus or
OpenVAS, I received reports similar to the following:

  Your https is vulnerable due to old protocols and weak ciphers.

  Remove vulnerable Apache SSL defaults as follows:

  # enable SSLv3 and TLSv1, but not SSLv2
  SSLProtocol all -SSLv2

  # exclude weak ciphers
  SSLCipherSuite !EXPORT40:!EXPORT56:!LOW:!ADH:DHE-RSA-AES256-SHA:EDH-RSA-DES-CBC3-SHA:DHE-RSA-AES128-SHA:AES256-SHA:DES-CBC3-SHA:AES128-SHA:RC4-SHA

This seems to me very different than if the report were to tell me instead:
"Apply the following ad hoc patch to source code that affects the OpenSSL
internals."

Rather, the protocol list and cipher suite configuration seems intended to be
customizable by design.


> Recap.  Ruby is not just requested to have a particular patch.  Ruby is
> requested to act as a sanity proxy over OpenSSL to prevent it from going mad.
> Without any patch you CAN operate Ruby safely already, right?  But you say
> that's not sufficient.  You request us to provide Ruby that you CANT fail.

I believe my own position to be less extreme: I have not been arguing for a
Ruby that can't fail; just a Ruby whose default configuration excludes already-
known weak ciphers or protocol versions.

However, since I'm not a security expert, my argument has been based on the
assumption that information provided by tools like Nessus (and various
security blogs) is correct.

My presumption had been that customizing Ruby's OpenSSL defaults is something
which could be accomplished in approximately as simple a manner as the Apache
SSL customization above.

(It sounds like this has mostly been true, apart from some difficulty arising
from supporting older OpenSSL versions.)


> Those patches proposed here SEEMS to provide adequate defaults to
> OpenSSL and so what?  I think that doesn't finish this story.  Because no one
> can say those patches are ultimate solutions.  And the request here is for
> us to provide ultimate solution for users.

I'm not sure which posts have been advocating an ultimate solution?

My understanding is that once specific protocol versions or ciphers have been
identified by security experts as weak or exploitable, there's no plausible
future in which this will cease to be true.

So my reasoning is, if specific ciphers or protocol versions are known now
to be weak, and will continue to be weak until the end of time, then it would
seem to benefit Ruby users if these were disabled by default.

(Of course, my viewpoint is the same should also be true for Apache users, but
I've never joined any Apache development mailing lists.)


Regards,

Bill