Ryan,
In your counter proposal, could you list your proposed milestone dates and then for each one specify the max validity period, domain re-use period and Org validation associated with those dates? As it stands, Org validation requires CA to verify that address is the Applicant’s address and that typically involves a direct exchange with a person at the organization via a Reliable Method of Communication. It’s not clear how we address that if we move to anything below a year. From: Ryan Sleevi <r...@sleevi.com> Sent: Friday, March 13, 2020 9:23 PM To: Doug Beattie <doug.beat...@globalsign.com> Cc: Kathleen Wilson <kwil...@mozilla.com>; mozilla-dev-security-pol...@lists.mozilla.org Subject: Re: About upcoming limits on trusted certificates On Fri, Mar 13, 2020 at 2:38 PM Doug Beattie via dev-security-policy <dev-security-policy@lists.mozilla.org <mailto:dev-security-policy@lists.mozilla.org> > wrote: When we moved to SHA2 knew of security risks so the timeline could be justified, however, I don’t see the same pressing need to move to annual domain revalidation and 1 year max validity for that matter. I can understand, and despite several years of effort, it appears that we will be just as unlikely to make forward progress. When we think about the issuance models, we need to keep the Enterprise approach in mind where domains are validated against a specific account or profile within an account and then issuance can happen using any valid domain or subdomain of those registered with the account. Splitting the domain validation from issuance permits different teams to handle this and to manage the overall policy. Domains can be validated at any time by anyone and not tied to the issuance of a specific certificate which makes issuance less prone to errors. This is a security risk, not a benefit. It creates significant risk that the CA systems, rather than strongly authenticating a request, move to a model of weakly authenticating a user or account. I can understand why CAs would prefer this, and potentially why Subscribers would too: it's convenient for them, and they're willing to accept the risk individually. However, we need to keep in mind the User approach in mind when thinking about whether these are good. For users, this introduces yet more risk into the system. For example, if an account on a CA system is compromised, the attacker can issue any certificate for any of the authorized domains. Compare this to a model of fresh authentication for requests, in which the only certificates that can be issued are the ones that can be technically verified. Similarly, users must accept that a CA that deploys a weak authentication system, any domains which use that CA are now at risk if the authentication method used is weak. When we put the user first, we can see that those Enterprise needs simply shift the risk/complexity from the Enterprise and to the User. It's understandable why the Enterprise might prefer that, but we must not fool ourselves into thinking the risk is not there. Root Stores exist to balance the risk (collectively) to users, and to reject such attempts to shift the cost or burden onto them. If your driving requirement to reduce the domain validation reuse is the BygoneSSL, then the security analysis is flawed. There are so many things have to align to exploit domain ownership change that it's impactable, imo. Has this ever been exploited? Yes, this was covered in BygoneSSL. If you meant to say impractical, then your risk analysis is flawed, but I suspect we'll disagree. This sort of concern has been the forefront of a number of new technologies, such as HTTP/2, Signed Exchanges, and the ORIGIN frame. Heck, even APNIC has seen these as _highly practical_ concerns: https://blog.apnic.net/2019/01/09/be-careful-where-you-point-to-the-dangers-of-stale-dns-records/ . Search PowerDNS. Would it make sense (if even possible) to track the level of automation and set a threshold for when the periods are changed? Mozilla and Google are tracking HTTPS adoption and plan to hard block HTTP when it reaches a certain threshold. Is there a way we can track issuance automation? I'm guessing not, but that would be a good way to reduce validity based on web site administrators embrace of automation tools. I'm glad you appreciated the efforts of Google and Mozilla, as well as others, to make effort here. However, I think suggesting transparency alone is likely to have much impact is to ignore what actually happened. That transparency was accompanied by meaningful change to promote HTTPS and discourage HTTP, and included making the necessary, sometimes controversial, changes to prioritize user security over enterprise needs. For example, browsers would not launch new features over insecure HTTP, insecure traffic was flagged and increasingly alerted on, and ultimately blocked. I think if that's the suggestion, then the quickest solution is to have the CA indicate, within the certificate, that it was issued freshly and automated. UAs can then make similar efforts as we saw with HTTPS, as can CAs and other participants in the ecosystem. You'll recall I previously proposed this in the CA/Browser Forum in 2017, and CAs were opposed then. It'd be useful to understand what changed, so that future efforts aren't stymied for three years. If this is not possible, then I think I'd (begrudgingly) agree with some comments Ryan made several years ago (at SwissSign's F2F?) that we need to set a longer term plan for these changes, document the reasons/security threats, and publish the schedule (rip the band aid off). These incremental changes, one after another by BRs or the Root programs are painful for everyone, and it seems that the changes are coming weather the CABF agrees or not. You're absolutely correct that, in 2015, I outlined a plan for 1 year certificates, with annual revalidation - https://cabforum.org/2015/06/24/2015-06-24-face-to-face-meeting-35-minutes/ Thus, it's somewhat frustrating to see it suggested in 2020 that a plan was not outlined. Your proposed approach, to get everyone to agree on a plan or timeline, is exactly what browsers have been doing for years. That 2015 conversation was itself the culmination of over a year's effort to attempt to a plan. What happens then is the same thing that happens now: a reasonable timeframe is set forth, some participants drag out their favorite saws and complaints, the discussion takes forever, now the timeline is no longer reasonable, it gets updated, and we rinse repeat until the timeframe is something approximating the heat death of the universe and CAs are still concerned Enterprises may not have time to adjust. As such, I think at this point we can say with full confidence, the probability of setting a "longer term plan for these changes, document the reasons/security threats, and publish a schedule" is failed, because every attempt results in the same circular discussion. Earnest, good faith attempts were made, for years, and it vacillates between offering reasons to be opposed (which are shown to be demonstrably false) or complaining there's too much information about why and it needs to be summarized to be understood. It's clear which side of the pendulum this is. If (AND ONLY IF) the resulting security analysis shows the need to get to some minimum validity period and domain re-validation interval, then, in my personal capacity and not necessarily that of my employer, I toss out these dates: Counter proposal: April 2021: 395 day domain validation max April 2021: 366 day organization validation max April 2022: 92 day domain validation max September 2022: 31 day domain validation max April 2023: 3 day domain validation max April 2023: 31 day organization validation max September 2023: 6 hour domain validation max This sets an entirely timeline that encourages automation of domain validation, reduces the risk of stale organization data (which many jurisdictions require annual review), and eventually moves to a system where request-based authentication is the norm, and automated systems for organization data is used. If there are jurisdictions that don't provide their data in a machine readable format, yes, they're precluded. If there are organizations that don't freshly authenticate their domains, yes, they're precluded. Now, it's always possible to consider shifting from an account-authenticated model to a key-authenticated-model (i.e. has this key been used with this domain), since that's a core objective of domain revalidation, but I can't see wanting to get to an end state where that duration is greater than 30 days, at most, of reuse, because of the practical risks and realities of key compromises. Indeed, if you look at the IETF efforts, such as Delegated Credentials or STAR, the industry evaluation of risk suggests 7 days is likely a more realistic upper bound for authorization of binding a key to a domain before requiring a fresh challenge.
smime.p7s
Description: S/MIME cryptographic signature
_______________________________________________ dev-security-policy mailing list dev-security-policy@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-security-policy