On 08/03/17 16:36, Ryan Sleevi wrote: > That is precisely the goal. We could define a set of process and procedures > specific to DTPs, which is effectively duplicitive with the handling of > subordinate CAs, or we could strive to align the two both conceptually and > materially, since, as you note below, there's a number of similarities in > the risk profile. > > The concern with the approach of both DTPs and subCAs is that it's very > easy for nuanced and subtle distinctions to be introduced, and as such, it > seems better to avoid that when possible by aligning on the majority-common > portion.
That seems to make sense to me. Given that the BRs have the concept of a DTP, how can we best align the two in practice? Does requiring every RA to have its own subCA do that? > Note: It does not appear you've updated > https://www.mozilla.org/en-US/about/governance/policies/security-group/certs/policy/ > - do you plan to? https://bugzilla.mozilla.org/show_bug.cgi?id=1343200 . It is a bit sad that a content push to our website requires the work to be scheduled in a sprint, but I digress. The fix is now on their master branch; I don't know how often they push to production. Looking at https://github.com/mozilla/bedrock/commits/prod , it may be as infrequently as once a week :-(( > I'm surprised by that reading, because Item 3 of that section states > > By "competent party" we mean a person or other entity who is authorized to > perform audits according to the stated criteria (e.g., by the organization > responsible for the criteria or by a relevant government agency) or for > whom there is sufficient public information available to determine that the > party is competent to judge the CA’s conformance to the stated criteria. > > In the absence of a proper license, such parties are not "authorized to > perform audits according to the stated criteria", so the only question is > whether "there is sufficient public information available to determine that > the party is competent to judge the CA's conformance to the stated > criteria". Yep, with you so far. > I recognize that Item 2 "replaces" the criteria for Section 8.2, but such a > replacement is neither reflected within the audit report produced (when > complying with the BRs) with respect to the issuing CA's oversight of the > DTP - that is, you might reasonably expect a qualification, but for Mozilla > to ignore said qualification, consistent with Item 2 of "Audit > Requirements". Can an audit be qualified (in the audit sense) by virtue of the person _doing_ the audit not being formally qualified (in the other sense!) to use those criteria? I would expect audit qualifications to relate to the audit subject, not the auditor. > "The burden is on the CA to prove that it has met the above >> requirements. However the CA may request a preliminary determination >> from us regarding the acceptability of the criteria and/or the competent >> independent party or parties by which it proposes to meet the >> requirements of this policy." >> >> I think a reasonable person might interpret this to mean that they >> needed to pick auditors who fulfilled the requirements in our policy, >> but don't need to _prove_ it unless asked. And they are not obliged to >> seek our determination. And I think that if we did ask Symantec to prove >> that the various bits of E&Y met the criteria in the policy at the time, >> I think they could probably do that. > > I find this an interesting and surprising interpretation, because I long > believed the intent and letter of Mozilla policy is that Mozilla required > such determination in the cases of accepting subordinate material, and the > burden of proof rests with them when presenting such material to Mozilla > during inclusion. It would, you are right. But these audits were not submitted to us (until recently) because they did not have to be, and so the question did not arise. >> Yes, I would expect externally-operated sub CAs to have the correct >> audits from a Mozilla-qualified auditor. > > Just to be clear: Given the definitions above, you believe it's acceptable > for sub-CAs to be issued to parties on the basis of the CA's judgement as > to whether there is "sufficient public information available to determine > that the party is competent to judge the CA's conformance to the stated > criteria", and that so long as they do so, it does not represent any form > of violation of Mozilla Policy, even if the CA makes an error in that > judgement? No, because in the case of a sub-CA, we require audits. And when we receive them, if they were done by unqualified parties, the CA would need to flag that, and we would make a judgement about that party's suitability at the time. The issue here arises that, because of the way things are set up, these RA's audits were not submitted to Mozilla, and so Symantec didn't have to resolve the Schrodinger's Cat of (qualified|not qualified and need us to make a judgement). Having danced enough angels for sufficiently long on the head of this pin, though, it's clear we should fix this. I propose we switch our auditor requirements to requiring qualified auditors, and saying that exceptions can be applied for in writing to Mozilla in advance of the audit starting, in which case Mozilla will make its own determination as to the suitability of the suggested party or parties. This would involve removing bullets 3-6 in the Audit section of 2.4, and rewording bullet 2 to say something like the above. >> If they didn't have control of the private key of their issuing >> intermediate(s) (as it seems they did not), then I do think this is a >> practical distinction from them being an unconstrained subordinate CA, >> in that they would e.g. not be audited for things like key management. > > So the extent of your concern is whether sufficient audit logs were > maintained, since key management is simply a subset of ensuring an > appropriate trail related to issuance and services. Well, and making sure it's in an HSM, and physically secure, and... right? > I highlight this, because at least one of these DTPs failed to maintain > sufficient audit logs, and Symantec has stated it plans to continue using > this information - information improperly secured, improperly maintained, > and with improper access controls - for the issuance of certificates. Hmm. Good point. >> In terms of the power they have, it's close - and, if the overseeing CA >> is not watching, basically identical. And there's the rub. As there is a >> distinction, then the CA should have been watching. If the CA were >> permitted not to watch, there would or should be no distinction in the >> BRs. And there is. Make sense? > > Unfortunately, I'm not sure I follow. Would you be able to rephrase? An unwatched RA and a sub-CA are effectively equivalent in power. A watched RA and a sub-CA might not be, if the CA is exercising effective control and limits on their issuance. There is a distinction in the BRs between an RA and a sub-CA, with the RA having less onerous requirements. One could say that the very existence of that distinction implies that CAs have a duty to carefully watch their RAs. Gerv _______________________________________________ dev-security-policy mailing list [email protected] https://lists.mozilla.org/listinfo/dev-security-policy

