Re: [pkix] fyi: Sovereign Keys: an EFF proposal for more secure TLS

[Stephen Kent <[e-mail]>]

At 9:51 AM +0000 12/19/11, Rob Stradling wrote:
>On Sunday 18 Dec 2011 20:40:06 Stephen Kent wrote:
>>  At 9:06 AM +0000 12/16/11, Rob Stradling wrote:
>>  >O...
>>  >
>>  >Martin, are you saying that it is never sensible to apply the robustness
>>  >principle when decoding ASN.1 data?
>>  >
>>  >http://en.wikipedia.org/wiki/Robustness_principle
>>
>>  NO is a good answer here, and in most security contexts.
>>
>>  ASN.1 decoding should, under the best circumstances, identify the
>>  error so that an RP can try to "persuade" the issuer of broken certs,
>>  CRLs, etc. to fix them. The decoded ought not core dump. But it also
>>  should reject invalid syntax. Failign to do so merely encourages the
>>  issuance of broken certs etc., and eventually imposes a burden on all
>>  RPs to accommodate errors by some issuers.
>>
>>  Steve
>
>I was actually thinking of PKCS#10 CSRs when I asked that question.  When we
>launched our CA ~10 years ago, my ASN.1 CSR decoder only accepted CSRs that
>were encoded 100% correctly.  Over time, I've had to relax the rules several
>times to accommodate various varieties of broken encoding.  I've never been
>able to suggest a business case for turning away customers just because their
>CSR generation software contains bugs!
>
>CAs are the "terminal endpoint" for CSRs, so hopefully Martin will 
>forgive me. 
>;-)
>
>Rob Stradling
>Senior Research & Development Scientist
>COMODO - Creating Trust Online

I understand the commercial rationale for such behavior, but I think
we agree that it results in precisely the problem I noted.

Steve