I'll be the first to admit that the example I put together is far from ideal. Perhaps a shortcoming is the lack of any explicit mention regarding the knowledge, skill, competence, etc. of the cert requester--or "system administrator" if you will. The ability of individual people to navigate the minefield of security practices and exploits is at least as important as the relative size of the Internet presence a given domain may have. There should also be some mention of attack vectors as they are an important consideration in evaluating the relative disadvantages of wildcard certs. (I think the advantages are widely understood and not disputed, so I'll skip that side of the debate.)


I'll talk about an attack vector first:

Suppose I want to set up a system to be used for spam, malware distribution, and phishing but, naturally, I want to operate undetected. First step is to find a (legitimate) server that is already set up and is not well secured. Without getting bogged down in the details, let's just assume I can find such a server and I'm able to obtain access to the admin panel or a command line/shell that controls it. With this access, let's also just assume I'm able to obtain the certificate and private key data that the legitimate site owner is using.

If the cert is not a wildcard, my options are limited. I could use the server and its domain name to send out my spam. In practice, though, it's probably easier to just commandeer the domain name and find some spam-as-a-service provider in the underground market. I could probably copy my malware objects into some obscurely-named directory and serve them up to my victims from there. The benefit is that I don't need my own hardware somewhere else and worry about DNS settings but there is a much higher risk that someone will eventually figure out what I'm doing. Someone could be monitoring network bandwidth or server logs or the file system capacity or some other data point that prompts a closer look.

If, instead, the cert is a wildcard, I have more freedom to ‎do what I want while reducing the chances of getting caught. I can exploit the domain and it's cert without tying myself to the server that is handling the legitimate traffic. I do need to manipulate DNS lookups in the right manner, but let's again just assume I can do that just fine. With DNS pointing my nefarious subdomains to my own nefarious server and with a wildcard cert and private key ready to validate with whatever names I choose, I am ready to go. I can send spam, offer up a variety of malware and ransomware, and host the phishing sites/pages all while keeping a very low profile--at least as far as the legitimate site owner is aware.

Granted, there is a healthy amount of hand waving in this illustration and frankly there are situations where other attack methods are more advantageous for any number of reasons. That said, the point I am hoping to make is that a wildcard certificate opens up possibilities for me as the bad guy that I might not have otherwise.

With that I'll (briefly) address the role of the system administrator's skill level:

In order for my wildcard attack to work, I need to find the server. Perhaps I'll skulk around the Internet looking for servers with unpatched vulnerabilities. Perhaps then I'll look up ownership information on that domain and cross-reference that with lists of accounts/passwords that have been compromised. Since it's common for people to use the same password on multiple sites I might get lucky.

From there, I'm mostly in the free and clear. In some situations I might want to tweak the DNS settings in the legit name servers so perhaps I will have direct access to it once I pop the server. Maybe the control panel also can modify the DNS records or maybe the same password will work on the name servers?

The point here is that there are important security measures that a system administrator must take. I think we can all agree that a competent administrator will take those steps while a lesser administrator might not. When this lack of skill (or perhaps "diligence" is a better word?) are combined with wildcard certs, the consequences can be very bad for everyone else.

Again, I'll be the first to admit this is perhaps not the best illustration of the risks posed by wildcard certs but hopefully it's at least good enough. I don't think the above is a major problem today but if the desire is make wildcard certs ubiquitous (?), I hope people will at least think twice.

And, for the record, I don't think wildcard certs should be banned outright. I also think that Hans and his network situation represent a good example where actual need and skill of the administrator combine to give one confidence that the wildcard cert's risks are controlled, and perhaps minimized.

And, finally, I'll just say CAA sounds like a good addition to the security toolbox but does have some shortcomings. The extent to which it can limit unintended wildcard certs is probably a good thing but clearly it will rely on a sufficiently skilled administrator for any effectiveness to be realized. 


From: Ryan Sleevi
Sent: Wednesday, April 26, 2017 2:46 PM
To: Peter Kurrasch‎
Reply To: [email protected]‎
Cc: Ryan Sleevi; mozilla-dev-security-policy
Subject: Re: Removing "Wildcard DV Certs" from Potentially Problematic Practices list



On Wed, Apr 26, 2017 at 3:17 PM, Peter Kurrasch <[email protected]> wrote:
Hi Ryan--

To your first comment, I'm afraid I won't have the time to take a closer look at the discussion on 3.2.2.4. Hopefully a path from single domain to unlimited domains exists (or will). It makes sense to me (without fully considering the consequences) that a "special" validation path be used for wildcards. Perhaps it could be part of an existing validation method, I'm not sure. Bottom line though, I don't think it makes sense that anyone and everyone be allowed to obtain a wildcard cert.

Right, but there's definitely a more careful argument you'll need to make, because that is _precisely_ what is (effectively) desired, not just by CAs, but most users of more than a few TLS certificates.

I'm not even sure that I disagree with you - my history in the CA/Browser Forum hopefully demonstrates Google's deep concern with the security of the validation methods, the capabilities possible with each validation method, and the frequency of which they're performed. However, finding consensus is, at this point, difficult, despite it being plainly obvious to folks with a security background :)

For context, I've been pushing for exploring CAA methods to reduce the permissible validation methods, and CAA already has the ability to restrict wildcard issuance via issuewild to set a of CAs, for which you can then establish a defined procedure. So if your concern is helping protect a _competent_ organization from a rogue wildcard, that already exists and is (in the process of) being required.

Let's suppose we break up everyone with a server on the Internet into 3 categories based on the size of their Internet presence: substantial, intermediate, and tiny. A "substantial" presence almost certainly has a large enterprise behind it with staff and capital resources to maintain the service. If we can assume that such organizations have tighter controls over use of the domain name space, servers, certificates, and so forth, then I think some of the risks posed by wildcard certs are reduced. That being the case, I'm less concerned about this category.

Regrettably, I think this inverts the priority. "substantial" presence are often the slowest to deploy, have the most complexity in their infrastructure, and similarly, suffer the greatest risk from a rogue wildcard.
 
For an "intermediate" presence, I'm thinking of places like universities that have a large number of subdomains in use but might not have a good set of controls over use of the domain name space and certificates and so forth. In this type of setting I think the use of wildcards might be appealing because it simplifies management of the network but if the controls are not in place, there remains a certain level of risk that these certs pose. (And to be fair, this isn't to say that only academic environments face this situation or that all academic environments are not suitably managed.) This category can be of concern.

This is most organizations :)
 
The "tiny" category almost speaks for itself and probably applies to all individuals and small businesses--people who want some presence but might not be all that diligent about it. In other words, these are the systems that probably have the weakest security measures in place. These are the systems that are regularly compromised and used for spam campaigns, fake login screens, and such. This is also the category for whom wildcard certs pose the greatest risk‎ to everyone else.

I disagree that you can attribute that cost to wildcards. That's the cost of the organization itself. The wildcard doesn't really contribute or change that calculus.
 
...and all sorts of variants of the above.‎ Per the wildcard cert, all of the above domains are perfectly valid. Per the actual intent of th‎e domain holder, most of the above are surely not valid.

CAA solves that ;)
 
So, the problem becomes how a relying party may determine if the site/domain is legitimate. If I have some indicator in the browser UI that says "secure" because the FQDN matching has been satisfied, I'll probably proceed to use the page that is presented. If the indicator is missing, there is a chance I'll think twice about proceeding any further. The trouble in this situation is that if the FQDN is nefarious but satisfies the wildcard contained in the cert, I will probably make the wrong decision and open myself up to who knows what.

But that's no different than shopping Target or Home Depot and their credit card processor / POS terminal being compromised, is it? You trusted an organization with an insufficient security practice relative to the threats faced. We don't say credit cards caused Target to be hacked though! 








_______________________________________________
dev-security-policy mailing list
[email protected]
https://lists.mozilla.org/listinfo/dev-security-policy

Reply via email to