Microsoft's official enterprise support blog for AD DS and more
Hi, folks! Jonathan here again, and today I want to talk about what appears to be an increasingly common topic: migrating from a single Windows Certification Authority (CA) to a multi-tier hierarchy. I’m going to assume that you already have a basic understanding of Public Key Infrastructure (PKI) concepts, i. e., you know what a root CA is versus an issuing CA, and you understand that Microsoft CAs come in two flavors – Standalone and Enterprise. If you don’t know those things then I recommend that you take a look at this before proceeding.
It seems that many organizations had installed a single Windows CA in order to support whatever major project that may have required it. Perhaps they were rolling out System Center Configuration Manager (SCCM), or wireless, or some other certificate consuming technology and one small line item in the project’s plan was Install a CA. Over time, though, this single CA began to see a lot of use as it was leveraged more and more for purposes other than originally conceived. Suddenly, there is a need for a proper Public Key Infrastructure (PKI) and administrators are facing some thorny questions:
I’m here to tell you that you aren’t alone. There are many organizations in the same situation, and there are good answers to each of these questions. More importantly, I’m going to share those answers with you. Let’s get started, shall we?
Important Note: This blog post does not address the private key archival scenario. Stay tuned for a future blog post on migrating archived private keys from one CA to another.
(You know, I actually considered asking Ned to find some Office clip art that showed two cats fighting over a mouse, and then thought, “What if he found it?!” I decided I didn’t really want to know and bagged the idea.)
To be clear, there is absolutely no issue with installing multiple Windows root CAs in the same forest. You can deploy your new PKI and keep it from issuing certificates to your users or computers until you are good and ready for it to do so. And while you’re doing all this, the old CA will continue to chug along oblivious to the fact that it will soon be removed with extreme prejudice.
Each Windows CA you install requires some objects created for it in Active Directory. If the CA is installed on a domain member these objects are created automatically. If, on the other hand, you install the CA on a workgroup computer that is disconnected from the network, you’ll have to create these objects yourself.
Regardless, all of these objects exist under the following container in Active Directory:
CN=Public Key Services, CN=Services, CN=Configuration, DC=<forestRootPartition>
As you can see, these objects are located in the Configuration partition of Active Directory which explains why you have to be an Enterprise Admin in order to install a CA in the forest. The Public Key Services Container holds the following objects:
AIA stands for Authority Information Access, and this container is the place where each CA will publish its own certificate for applications and services to find if needed. The AIA container holds certificationAuthority objects, one for each CA. The name of the object matches the canonical name of the CA itself.
CDP stands for CRL Distribution Point (and CRL stands for Certificate Revocation List). This container is where each CA publishes its list of revoked certificates to Active Directory. In this container, you’ll find another container object whose common name matches the host name of the server on which Certificate Services is installed – one for each Windows CA in your forest. Within each server container is a cRLDistributionPoint object named for the CA itself. The actual CRL for the CA is published to this object.
The Certificate Templates container holds a list of pKICertificateTemplate objects, each one representing one of the templates you see in the Certificate Templates MMC snap-in. Certificate templates are shared objects, meaning they can be used by any Enterprise CA in the forest. There is no CA-specific information stored on these objects.
The Certification Authorities container holds a list of certificationAuthority objects representing each root CA trusted by the Enterprise. Any root CA certificate published here is distributed to each and every member of the forest as a trusted root. A Windows root CA installed on a domain server will publish its certificate here. If you install a root CA on a workgroup server you’ll have to publish the certificate here manually.
The Enrollment Services container holds a list of pKIEnrollmentService objects, each one representing an Enterprise CA installed in the forest. The pKIEnrollmentService object is used by Windows clients to locate a CA capable of issuing certificates based on a particular template. When you add a certificate template to a CA via the Certification Authority snap-in, that CA’s pKIEnrollmentService object is updated to reflect the change.
There are few other objects and containers in the Public Key Services container, but they are beyond the scope of this post. If you’re really interested in the nitty-gritty details, post a comment and I’ll address them in a future post.
To summarize, let’s look at a visual of each of these objects and containers and see how they fit together. I’ve diagrammed out an environment with three CAs. One is the Old And Busted CA, which has been tottering along for years ever since Bob the network admin put it up to issue certificates for wireless authentication.
Now that Bob has moved onto new and exciting opportunities in the field of food preparation and grease trap maintenance after that unfortunate incident with the misconfigured VLANs, his successor, Mike, has decided to deploy a new, enterprise-worthy PKI.
To that end, Mike has deployed the New Hotness Root CA, along with the More New Hotness Issuing CA. The New Hotness Root CA is an offline Standalone root, meaning it is running the Windows CA in Standalone mode on a workgroup server disconnected from the network. The New Hotness Issuing CA, however, is an online issuing CA. It’s running in Enterprise mode on a domain server.
Let’s see what the AD objects for these CAs look like:
Figure 1: Sample PKI AD objects
Figure 1: Sample PKI AD objects
We’ve come an awful long way to emphasize one simple point. As you can see, each PKI-related object in Active Directory is uniquely named, either for the CA itself or the server on which the CA is installed. Because of this, you can install a (uniquely named) CA on every server in your environment and not run into the sort of conflict that some customers fear when I talk to them about this topic. You could also press your tongue against a metal pole in the dead of winter. Of course, it would hurt, and you’d look silly, but you could do it. Same concept applies here.
So what’s the non-silly approach?
If you need to migrate your organization from the Old And Busted CA to the New Hotness PKI, then the very first thing you should do is deploy the new PKI. This requires proper planning, of course; select your platform, locate your servers, that sort of thing. I encourage you to use a Windows Server 2008 R2 platform. WS08R2 CAs are supported with a minimum schema version of 30 which means you do not need to upgrade your Windows Server 2003 domain controllers. More details are here.
Once your planning is complete, deploy your new PKI. Actual step-by-step guidance is beyond the scope of this blog post, but it is pretty well covered elsewhere. You should first take a look at the Best Practices for Implementing a Microsoft Windows Server 2003 Public Key Infrastructure. Yes, I realize this was for Windows Server 2003, but the concepts are identical for Windows Server 2008 and higher, and the scripts included in the Best Practices Guide are just as useful for the later platforms. It is also true that the Guide describes setting up a three tiered hierarchy, but again, you can easily adapt the prescriptive guidance to a two tiered hierarchy. If you want help with that then you should take a look at this post.
The major benefit to using Windows Server 2008 or higher is a neat little addition to the CAPolicy.INF file. When you install a new Enterprise CA it is preconfigured with a set of default certificate templates for which it is ready immediately to start issuing certificates. You don’t really want the CA to issue any certificates until you’re good and ready for it to do so. If the Enterprise CA wasn’t configured with any templates by default then it wouldn’t issue any certificates after the CA starts up. When you were ready to switch over to the new PKI, you’d just configure the issuing CA with the appropriate templates. It turns out that as of Windows Server 2008 you can install an Enterprise issuing CA so that the default certificate templates were not automatically configured on the CA. You accomplish this by adding a line to the CAPolicy.inf file:
Now, if at this point you’re wondering, “What is a CAPolicy.INF file, and how is it involved in setting up a CA,” then guess what? That is your clue that you need to read the Best Practices Guide linked above. It’s all in there, including samples.
“Oh…but the samples are for Windows Server 2003,” you say, accusingly. Relax; here’s a blog post I wrote earlier fully documenting the Windows Server 2008 R2 CAPolicy.INF syntax. Again, the concepts and broad strokes are all the same; just some minor details have changed. Use my earlier post to supplement the Best Practices Guide and you’ll be golden.
So you have your new PKI installed and you’re ready to migrate your organization over to it. How does one do that without impacting one’s organization too severely?
The first thing you’ll want to do is prevent the old CA from issuing any new certificates. You just uninstall it, of course, but that could cause considerable problems. What do you think would happen if that CA’s published CRL expired and it wasn’t around to publish a new one? Depending on the application using those certificates, they’d all fail to validate and become useless. Wireless clients would fail to connect, smart card users would fail to authenticate, and all sorts of other bad things would occur. The goal is to prevent any career limiting outages so you shouldn’t just uninstall that CA.
No, you should instead remove all the templates from the Certificate Templates folder using the Certification Authority MMC snap-in on the old CA. If an Enterprise CA isn’t configured with any templates it can’t issue any new certificates. On the other hand, it is still quite capable of refreshing its CRL, and this is exactly the behavior you want. Conversely, you’ll want to add those same templates you removed from the Old And Busted CA into the Certificate Templates folder on the New Hotness Issuing CA.
If you modify the contents of the Certificate Templates folder for a particular CA, that CA’s pKIEnrollmentService object must be updated in Active Directory. That means that you will have some latency as the changes replicate amongst your domain controllers. It is possible that some user in an outlying site will attempt to enroll for a certificate against the Old And Busted CA and that request will fail because the Old And Busted CA knows immediately that it should not issue any certificates. Given time, though, that error condition will fade as all domain controllers get the new changes. If you’re extremely sensitive to that kind of failure, however, then just add your templates to the New Hotness Issuing CA first, wait a day (or whatever your end-to-end replication latency is) and then remove those templates from the Old And Busted CA. In the long run, it won’t matter if the Old And Busted CA issues a few last minute certificates.
At this point all certificate requests within your organization will be processed by the New Hotness Issuing CA, but what about all those certificates issued by the Old And Busted CA that are still in use? Do you have to manually go to each user and computer and request new certificates? Well…it depends on how the certificates were originally requested.
If a certificate has been manually requested then, yes, in all likelihood you’ll need to manually update those certificates. I’m referring here to those certificates requested using the Certificates MMC snap-in, or through the Web Enrollment Pages. Unfortunately, there’s no automatic management for certificates requested manually. In reality, though, refreshing these certificates probably means changing some application or service so it knows to use the new certificate. I refer here specifically to Server Authentication certificates in IIS, OCS, SCCM, etc. Not only do you need to change the certificate, but you also need to reconfigure the application so it will use the new certificate. Given this situation, it makes sense to make your necessary changes gradually. Presumably, there is already a procedure in place for updating the certificates used by these applications I mentioned, among others I didn’t, as the current certificates expire. As time passes and each of these older, expiring certificates are replaced by new certificates issued by the new CA, you will gradually wean your organization off of the Old And Busted CA and onto the New Hotness Issuing CA. Once that is complete you can safely decommission the old CA.
And it isn’t as though you don’t have a deadline. As soon as the Old And Busted CA certificate itself has expired you’ll know that any certificate ever issued by that CA has also expired. The Microsoft CA enforces such validity period nesting of certificates. Hopefully, though, that means that all those certificates have already been replaced, and you can finally decommission the old CA.
Certificate Autoenrollment was introduced in Windows XP, and it allows the administrator to assign certificates based on a particular template to any number of forest users or computers. Triggered by the application of Group Policy, this component can enroll for certificates and renew them when they get old. Using Autoenrollment, once can easily deploy thousands of certificates very, very quickly. Surely, then, there must be an automated way to replace all those certificates issued by the previous CA?
As a matter of fact, there is.
As described above, the new PKI is up and ready to start issuing digital certificates. The old CA is still up and running, but all the templates have been removed from the Certificate Templates folder so it is no longer issuing any certificates. But you still have literally thousands of automatically enrolled certificates outstanding that need to be replaced. What do you do?
In the Certificates Templates MMC snap-in, you’ll see a list of all the templates available in your enterprise. To force all holders of a particular certificate to automatically enroll for a replacement, all you need to do is right-click on the template and select Reenroll All Certificate Holders from the context menu.
What this actually does is increment the major version number of the certificate template in question. This change is detected by the Autoenrollment component on each Windows workstation and server prompting them to enroll for the updated template, replacing any certificate they may already have. Automatically enrolled user certificates are updated in the exact same fashion.
Now, how long it takes for each certificate holder to actually finish enrolling will depend how many there are and how they connect to the network. For workstations that are connected directly to the network, user and computer certificates will be updated at the next Autoenrollment pulse.
Note: For computers, the autoenrollment pulse fires at computer startup and every eight hours thereafter. For users, the autoenrollment pulse fires at user logon and every eight hours thereafter. You can manually trigger an autoenrollment pulse by running certutil -pulse from the command line. Certutil.exe is installed with the Windows Server 2003 Administrative Tools Pack on Windows XP, but it is installed by default on the other currently supported versions of Windows.
For computers that only connect by VPN it may take longer for certificates to be updated. Unfortunately, there is no blinking light that says all the certificate holders have been reenrolled, so monitoring progress can be difficult. There are ways it could be done -- monitoring the certificates issued by the CA, using a script to check workstations and servers and verify that the certificates are issued from the new CA, etc. -- but they require some brain and brow work from the Administrator.
There is one requirement for this reenrollment strategy to work. In the group policy setting where you enable Autoenrollment, you must have the following option selected: Update certificates that use certificate templates.
If this policy option is not enabled then your autoenrolled certificates will not be automatically refreshed.
Remember, there are two autoenrollment policies -- one for the User Configuration and one for the Computer Configuration. This option must be selected in both locations in order to allow the Administrator to force both computers and users to reenroll for an updated template.
As I’ve said earlier, once you’ve configured the Old And Busted CA so that it will no longer issue certificates you shouldn’t need to touch it again until all the certificates issued by that CA have expired. As long as the CA continues to publish a revocation list, all the certificates issued by that CA will remain valid until they can be replaced. But what if you want to decommission the Old And Busted CA immediately? How could make sure that your outstanding certificates would remain viable until you can replace them with new certificates? Well, there is a way.
All X.509 digital certificates have a validity period, a defined interval time with fixed start and end dates between which the certificate is considered valid unless it has been revoked. Once the certificate is expired there is no need to check with a certificate revocation list (CRL) -- the certificate is invalid regardless of its revocation status. Revocation lists also have a validity period during which time it is considered an authoritative list of revoked certificates. Once the CRL has expired it can no longer be used to check for revocation status; a client must retrieve a new CRL.
You can use this to your advantage by extending the validity period of the Old And Busted CA’s CRL in the CA configuration to match (or exceed) the remaining lifetime of the CA certificate. For example, if the Old And Busted CA’s certificate will be valid for the next 4 years, 3 months, and 10 days, then you can set the publication interval for the CA’s CRL to 5 years and immediately publish it. The newly published CRL will remain valid for the next five years, and as long as you leave that CRL published in the defined CRL distribution points -- Active Directory and/or HTTP -- clients will continue to use it for checking revocation status. You no longer need the actual CA itself so you can uninstall it.
One drawback to this, however, is that you won’t be able to easily add any certificates to the revocation list. If you need to revoke a certificate after you’ve decommissioned the CA, then you’ll need to use the command line utility certutil.exe.
Certutil.exe -resign “Old And Busted CA.crl” +<serialNumber>
Of course, this requires that you keep the private keys associated with the CA, so you’d better back up the CA’s keys before you uninstall the role.
Wow…we’ve covered a lot of information here, so I’ll try to boil all of it down to the most important points. First, yes you can have multiple root CAs and even multiple PKIs in a single Active Directory forest. Because of the way the objects are representing those CAs are named and stored, you couldn’t possibly experience a conflict unless you tried to give more than one CA the same CA name.
Second, once the new PKI is built you’ll want to configure your old CA so that it no longer issues certificates. That job will now belong to the issuing CA in your new PKI.
Third, the ease with which you can replace all the certificates issued by the old CA with certificates issued by your new CA will depend mainly on how the certificates were first deployed. If all of your old certificates were requested manually then you will need to replace them in the same way. The easiest way to do that is replace them all gradually as they expired. On the other hand, if your old certificates were deployed via autoenrollment then you can trigger all of your autoenrollment clients to replace the old certificates with new ones from the new PKI. You can do this through the Certificate Templates MMC snap-in.
And finally, what do you do with the old CA? Well, if you don’t need the equipment you can just keep it around until it either expires or all the old certificates have been replaced. If, however, you want to get rid of it immediately you can extend the lifetime of the old CA’s CRL to match the remaining validity period of the CA certificate. Just publish a new CRL and it’ll be good until all outstanding certificates have expired. Just keep in mind that this route will limit your ability to revoke those old certificates.
If you think I missed something, or you want me to clarify a certain point, please feel free to post in the comments below.
Jonathan “Man in Black” Stephens
PS: Don’t ever challenge my Office clip art skills again, Jonathan.
Good stuff Thanks!