On Sat, Nov 16, 2013 at 5:08 PM, Theodore Ts'o <tytso@xxxxxxx> wrote:
On Sat, Nov 16, 2013 at 03:11:34PM -0500, Phillip Hallam-Baker wrote:Whether the number is 100, or 600, or a thousand (and note that even
> But as is well known, many CAs own multiple embedded roots, typically three
> or four brands per large CA and each brand often has several roots. The
> conclusion that the EFF has been peddling is that there are 600 parties
> that can introduce spurious certs, this is not what their evidence
> demonstrates.
>
> My point was that like the 'Gore claimed to invent internet meme' this has
> become a zombie lie that is repeated by people despite the fact that it has
> been repeatedly shown to be false. People like to believe it because it
> reinforces their prejudices but that does not make it true.
>
> We should not be making policy decisions on the basis of zombie lies.
if there are dozens or even hundreds of CA's being run by one
"organization", those CA's may be run by different personnel, and have
different policies, and have their certificate signing keys stored in
different ways (i.e., some may be stored on some minimum wage worker's
laptop; others may be stored in some Tempest Shielded fancy-shamancy
BBN secure signing box requiring multiple crypto ignitition keys
before certificates can be issued) --- how you count the CA's or
organizations I don't think is really all that important. Even if
it's only 100 organizations, or even 50, do you really believe they
can all be trusted, and are you willing to assert that they will all
never having any process or technological failures?
Well first I think that it is important to use facts when making decisions. When someone misrepresents the facts by an order of magnitude and refuses to correct it after it has been demonstrated to be false, that is a data point.
If you are interested in the technical criteria for running a public CA then go look at the Basic Requirements on the CABForum Web site and they will answer all the hypotheticals you raise above.
Nobody uses the BBN Tempest shielded box any more as it hasn't been made in a decade. They do use FIPS certified hardware though.
I am far more certain of the ability of CAs to do their job than any other potential point of failure. I don't have absolute trust but I trust them more than I trust the end-points. I am also rather concerned about the black box crypto hardware we have to use. But right now there isn't a better alternative.
And if you believe that all dozen, or 50, or 100 certificate
organizations can be considered trustworthy, care to explain some of
the more spectacular failures (i.e., Comodo, Diginotar, etc.)?
How many CERT reports of significant security holes are filed on Linux every month? Care to explain them?
You think you are making a strong argument but the only reason the CA issues received notice is that they are so very unusual. Software security disasters are a daily occurrence.
Quibbling over numbers doesn't change the the fundamental premise,
which is that the certificate signing architecture for the web is
considered by some (including myself), to be pretty badly broken.
Pointing out that you are repeating a propaganda point that has been proved false is not quibbling over details. If you want to make the case against my industry then you need to check the figures you use as carefully as you would check a PGP key before signing it.
Website: http://hallambaker.com/