Re: [keyassure] Another comment from the mic

[Eric Rescorla <[e-mail]>]

On Wed, Mar 30, 2011 at 1:16 PM, Stephen Kent <[e-mail]> wrote:
> At 10:57 AM +0200 3/30/11, Eric Rescorla wrote:
>>
>> As I said at the mic, the vast majority of the certificate warnings
>> you see on the network
>> are not because of attacks, but rather are due to:
>>
>> - Self-signed certs
>> - Certificates from legitimate CAs which are uncompromised but invalid
>> for some technical
>>   reason (expired certs, trivial name mismatches, etc.)
>>
>> One of the purposes of my "permissive" case-2 model in my previous
>> email is to allow those
>> self-signed certificate servers to have verifiable credentials.
>> However, anything we do that
>> has the consequence that certificates which should verify don't for
>> mostly-irrelevant technical
>> reasons (e.g., certs which are validated by DANE but are expired or
>> have the wrong keyusage
>> bits) will defeat this purpose to some extent. Perhaps that's worth
>> doing in service of the correctness
>> of the validation chain, but it does need to be considered.
>>
>> -Ekr
>
> I fully agree with the goal of avoiding error messages that confuse a user,
> especially ones that make it hard to tell the difference between a sloppy
> admin and an attacker.  But, it may be hard to tell the difference in some
> cases :-). Maybe we can reduce the incidence of bad self-signed certs by
> providing better tools to generate these certs, and including defaults to
> make it easy to avoid the typical errors.

This seems like a very worthy goal.

-Ekr