Choosing the right certificate revocation method in ACM Private CA

AWS Certificate Manager Private Certificate Authority (ACM PCA) is a highly available, fully managed private certificate authority (CA) service that allows you to create CA hierarchies and issue X.509 certificates from the CAs you create in ACM PCA. You can then use these certificates for scenarios such as encrypting TLS communication channels, cryptographically signing code, authenticating users, and more. But what happens if you decide to change your TLS endpoint or update your code signing entity? How do you revoke a certificate so that others no longer accept it?

In this blog post, we will cover two fully managed certificate revocation status checking mechanisms provided by ACM PCA: the Online Certificate Status Protocol (OCSP) and certificate revocation lists (CRLs). OCSP and CRLs both enable you to manage how you can notify services and clients about ACM PCA–issued certificates that you revoke. We’ll explain how these standard mechanisms work, we’ll highlight appropriate deployment use cases, and we’ll identify the advantages and downsides of each. We won’t cover configuration topics directly, but will provide you with links to that information as we go.

Certificate revocation

An X.509 certificate is a static, cryptographically signed document that represents a user, an endpoint, an IoT device, or a similar end entity. Because certificates provide a mechanism to authenticate these end entities, they are valid for a fixed period of time that you specify in the expiration date attribute when you generate a certificate. The expiration attribute is important, because it validates and regulates an end entity’s identity, and provides a means to schedule the termination of a certificate’s validity. However, there are situations where a certificate might need to be revoked before its scheduled expiration. These scenarios can include a compromised private key, the end of agreement between signed and signing organizations, user or configuration error when issuing certificates, and more. Although you can use certificates in many ways, we will refer to the predominant use case of TLS-based client-server implementations for the remainder of this blog post.

Certificate revocation can be used to identify certificates that are no longer trusted, and CRLs and OCSP are the standard mechanisms used to publish the revocation information. In addition, the special use case of OCSP stapling provides a more efficient mechanism that is supported in TLS 1.2 and later versions.

ACM PCA gives you the flexibility to use either of these mechanisms, or both. More importantly, as an ACM PCA administrator, the mechanism you choose to use is reflected in the certificate, and you must know how you want to manage revocation before you create the certificate. Therefore, you need to understand how the mechanisms work, select your strategy based on its appropriateness to your needs, and then create and deploy your certificates. Let’s look at how each mechanism works, the use cases for each, and issues to be aware of when you select a revocation strategy.

Certificate revocation using CRLs

As the name suggests, a CRL contains a list of revoked certificates. A CRL is cryptographically signed and issued by a CA, and made available for download by clients (for example, web browsers for TLS) through a CRL distribution point (CDP) such as a web server or a Lightweight Directory Access Point (LDAP) endpoint.

A CRL contains the revocation date and the serial number of revoked certificates. It also includes extensions, which specify whether the CA administrator temporarily suspended or irreversibly revoked the certificate. The CRL is signed and timestamped by the CA and can be verified by using the public key of the CA and the cryptographic algorithm included in the certificate. Clients download the CRL by using the address provided in the CDP extension and trust a certificate by verifying the signature, expiration date, and revocation status in the CRL.

CRLs provide an easy way to verify certificate validity. They can be cached and reused, which makes them resilient to network disruptions, and are an excellent choice for a server that is getting requests from many clients for the same CA. All major web browsers, OpenSSL, and other major TLS implementations support the CRL method of validating certificates.

However, the size of CRLs can lead to inefficiency for clients that are validating server identities. An example is the scenario of browsing multiple websites and downloading a CRL for each site that is visited. CRLs can also grow large over time as you revoke more certificates. Consider the World Wide Web and the number of invalidations that take place daily, which makes CRLs an inefficient choice for small-memory devices (for example, mobile, IoT, and similar devices). In addition, CRLs are not suited for real-time use cases. CRLs are downloaded periodically, a value that can be hours, days, or weeks, and cached for memory management. Many default TLS implementations, such as Mozilla, Chrome, Windows OS, and similar, cache CRLs for 24 hours, leaving a window of up to a day where an endpoint might incorrectly trust a revoked certificate. Cached CRLs also open opportunities for non-trusted sites to establish secure connections until the server refreshes the list, leading to security risks such as data breaches and identity theft.

Implementing CRLs by using ACM PCA

ACM PCA supports CRLs and stores them in an Amazon Simple Storage Service (Amazon S3) bucket for high availability and durability. You can refer to this blog post for an overview of how to securely create and store your CRLs for ACM PCA. Figure 1 shows how CRLs are implemented by using ACM PCA.

Figure 1: Certificate validation with a CRL

The workflow in Figure 1 is as follows:

1. On certificate revocation, ACM PCA updates the Amazon S3 CRL bucket with a new CRL.
   

   Note: An update to the CRL may take up to 30 minutes after a certificate is revoked.

2. The client requests a TLS connection and receives the server’s certificate.
3. The client retrieves the current CRL file from the Amazon S3 bucket and validates it.

The refresh interval is the period between when an administrator revokes a certificate and when all parties consider that certificate revoked. The length of the refresh interval can depend on how quickly new information is published and how long clients cache revocation information to improve performance.

When you revoke a certificate, ACM PCA publishes a new CRL. ACM PCA waits 5 minutes after a RevokeCertificate API call before publishing a new CRL. This process exists to accommodate multiple revocation requests in a short time frame. An update to the CRL can take up to 30 minutes to propagate. If the CRL update fails, ACM PCA makes further attempts every 15 minutes.

CRLs also have a validity period, which you define as part of the CRL configuration by using ExpirationInDays. ACM PCA uses the value in the ExpirationInDays parameter to calculate the nextUpdate field in the CRL (the day and time when ACM PCA will publish the next CRL). If there are no changes to the CRL, the CRL is refreshed at half the interval of the next update. Clients may cache CRLs while they are still valid, so not all clients will have the updated CRL with the newly revoked certificates until the previous published CRL has expired.

Certificate revocation using OCSP

OCSP removes the burden of downloading the CRL from the client. With OCSP, clients provide the serial number and obtain the certificate status for a single certificate from an OCSP Responder. The OCSP Responder can be the CA or an endpoint managed by the CA. The certificate that is returned to the client contains an authorityInfoAccess extension, which provides an accessMethod (for example, OCSP), and identifies the OCSP Responder by a URL (for example, http://example-responder:<port>) in the accessLocation. You can also specify the OCSP Responder location manually in the CA profile. The certificate status response that is returned by the OCSP Responder can be good, revoked, or unknown, and is signed by using a process similar to the CRL for protection against forgery.

OCSP status checks are conducted in real time and are a good choice for time-sensitive devices, as well as mobile and IoT devices with limited memory.

However, the certificate status needs to be checked against the OCSP Responder for every connection, therefore requiring an extra hop. This can overwhelm the responder endpoint that needs to be designed for high availability, low latency, and protection against network and system failures. We will cover how ACM PCA addresses these availability and latency concerns in the next section.

Another thing to be mindful of is that the OCSP protocol implements OCSP status checks over unencrypted HTTP that poses privacy risks. When a client requests a certificate status, the CA receives information regarding the endpoint that is being connected to (for example, domain, IP address, and related information), which can easily be intercepted by a middle party. We will address how OCSP stapling can be used to address these privacy concerns in the OCSP stapling section.

Implementing OCSP by using ACM PCA

ACM PCA provides a highly available, fully managed OCSP solution to notify endpoints that certificates have been revoked. The OCSP implementation uses AWS managed OCSP responders and a globally available Amazon CloudFront distribution that caches OCSP responses closer to you, so you don’t need to set up and operate any infrastructure by yourself. You can enable OCSP on new or existing CAs using the ACM PCA console, the API, the AWS Command Line Interface (AWS CLI), or through AWS CloudFormation. Figure 2 shows how OCSP is implemented on ACM PCA.

Note: OCSP Responders, and the CloudFront distribution that caches the OCSP response for client requests, are managed by AWS.

Figure 2: Certificate validation with OCSP

The workflow in Figure 2 is as follows:

1. On certificate revocation, the ACM PCA updates the OCSP Responder, which generates the OCSP response.
2. The client requests a TLS connection and receives the server’s certificate.
3. The client sends a query to the OCSP endpoint on CloudFront.
   

   Note: If the response is still valid in the CloudFront cache, it will be served to the client from the cache.

4. If the response is invalid or missing in the CloudFront cache, the request is forwarded to the OCSP Responder.
5. The OCSP Responder sends the OCSP response to the CloudFront cache.
6. CloudFront caches the OCSP response and returns it to the client.

The ACM PCA OCSP Responder generates an OCSP response that gets cached by CloudFront for 60 minutes. When a certificate is revoked, ACM PCA updates the OCSP Responder to generate a new OCSP response. During the caching interval, clients continue to receive responses from the CloudFront cache. As with CRLs, clients may also cache OCSP responses, which means that not all clients will have the updated OCSP response for the newly revoked certificate until the previously published (client-cached) OCSP response has expired. Another thing to be mindful of is that while the response is cached, a compromised certificate can be used to spoof a client.

Certificate revocation using OCSP stapling

With both CRLs and OCSP, the client is responsible for validating the certificate status. OCSP stapling addresses the client validation overhead and privacy concerns that we mentioned earlier by having the server obtain status checks for certificates that the server holds, directly from the CA. These status checks are periodic (based on a user-defined value), and the responses are stored on the web server. During TLS connection establishment, the server staples the certificate status in the response that is sent to the client. This improves connection establishment speed by combining requests and reduces the number of requests that are sent to the OCSP endpoint. Because clients are no longer directly connecting to OCSP Responders or the CAs, the privacy risks that we mentioned earlier are also mitigated.

Implementing OCSP stapling by using ACM PCA

OCSP stapling is supported by ACM PCA. You simply use the OCSP Certificate Status Response passthrough to add the stapling extension in the TLS response that is sent from the server to the client. Figure 3 shows how OCSP stapling works with ACM PCA.

Figure 3: Certificate validation with OCSP stapling

The workflow in Figure 3 is as follows:

1. On certificate revocation, the ACM PCA updates the OCSP Responder, which generates the OCSP response.
2. The client requests a TLS connection and receives the server’s certificate.
3. In the case of server’s cache miss, the server will query the OCSP endpoint on CloudFront.
   

   Note: If the response is still valid in the CloudFront cache, it will be returned to the server from the cache.

4. If the response is invalid or missing in the CloudFront cache, the request is forwarded to the OCSP Responder.
5. The OCSP Responder sends the OCSP response to the CloudFront cache.
6. CloudFront caches the OCSP response and returns it to the server, which also caches the response.
7. The server staples the certificate status in its TLS connection response (for TLS 1.2 and later versions).

OCSP stapling is supported with TLS 1.2 and later versions.

Selecting the correct path with OCSP and CRLs

All certificate revocation offerings from AWS run on a highly available, distributed, and performance-optimized infrastructure. We strongly recommend that you enable a certificate validation and revocation strategy in your environment that best reflects your use case. You can opt to use CRLs, OCSP, or both. Without a revocation and validation process in place, you risk unauthorized access. We recommend that you review your business requirements and evaluate the risk profile of access with an invalid certificate versus the availability requirements for your application.

In the following sections, we’ll provide some recommendations on when to select which certificate validation and revocation strategy. We’ll cover client-server TLS communication, and also provide recommendations for mutual TLS (mTLS) authentication scenarios.

Recommended scenarios for OCSP stapling and OCSP Must-Staple

If your organization requires support for TLS 1.2 and later versions, you should use OCSP stapling. If you want to reduce the application availability risk for a client that is configured to fail the TLS connection establishment when it is unable to validate the certificate, you should consider using the OCSP Must-Staple extension.

OCSP stapling

If your organization requires support for TLS 1.2 and later versions, you should use OCSP stapling. With OCSP stapling, you reduce your client’s load and connectivity requirements, which helps if your network connectivity is unpredictable. For example, if your application client is a mobile device, you should anticipate network failures, low bandwidth, limited processing capacity, and impatient users. In this scenario, you will likely benefit the most from a system that relies on OCSP stapling.

Although the majority of web browsers support OCSP stapling, not all servers support it. OCSP stapling is, therefore, typically implemented together with CRLs that provide an alternate validation mechanism or as a passthrough for when the OCSP response fails or is invalid.

OCSP Must-Staple

If you want to rely on OCSP alone and avoid implementing CRLs, you can use the OCSP Must-Staple certificate extension, which tells the connecting client to expect a stapled response. You can then use OCSP Must-Staple as a flag for your client to fail the connection if the client does not receive a valid OCSP response during connection establishment.

Recommended scenarios for CRLs, OCSP (without stapling), and combinational strategies

If your application needs to support legacy, now deprecated protocols such as TLS 1.0 or 1.1, or if your server doesn’t support OCSP stapling, you could use a CRL, OCSP, or both together. To determine which option is best, you should consider your sensitivity to CA availability, recently revoked certificates, the processing capacity of your application client, and network latency.

CRLs

If your application needs to be available independent of your CA connectivity, you should consider using a CRL. CRLs are much larger files that, from a practical standpoint, require much longer cache times to be of use, but they will be present and available for verification on your system regardless of the status of your network connection. In addition, the lookup time of a certificate within a CRL is local and therefore shorter than a network round trip to an OCSP Responder, because there are no network connection or DNS lookup times.

OCSP (without stapling)

If you are sensitive to the processing capacity of your application client, you should use OCSP. The size of an OCSP message is much smaller compared to a CRL, which allows you to configure shorter caching times that are better suited for your risk profile. To optimize your OCSP and OCSP stapling process, you should review your DNS configuration because it plays a significant role in the amount of time your application will take to receive a response.

For example, if you’re building an application that will be hosted on infrastructure that doesn’t support OCSP stapling, you will benefit from clients making an OCSP request and caching it for a short period. In this scenario, your application client will make a single OCSP request during its connection setup, cache the response, and reuse the certificate state for the duration of its application session.

Combining CRLs and OCSP

You can also choose to implement both CRLs and OCSP for your certificate revocation and validation needs. For example, if your application needs to support legacy TLS protocols while providing resiliency to network failures, you can implement both CRLs and OCSP. When you use CRLs and OCSP together, you verify certificates primarily by using OCSP; however, in case your client is unable to reach the OCSP endpoint, you can fail over to an alternative validation method (for example, CRL). This approach of combining CRLs and OCSP gives you all the benefits of OCSP mentioned earlier, while providing a backup mechanism for failure scenarios such as an unreachable OCSP Responder, invalid response from the OCSP Responder, and similar. However, while this approach adds resilience to your application, it will add management overhead because you will need to set up CRL-based and OCSP-based revocation separately. Also, remember that clients with reduced computing power or poor network connectivity might struggle as they attempt to download and process the CRL.

Recommendations for mTLS authentication scenarios

You should consider network latency and revocation propagation delays when optimizing your server infrastructure for mTLS authentication. In a typical scenario, server certificate changes are infrequent, so caching an OCSP response or CRL on your client and an OCSP-stapled response on a server will improve performance. For mTLS, you can revoke a client certificate at any time; therefore, cached responses could introduce the risk of invalid access. You should consider designing your system such that a copy of a CRL for client certificates is maintained on the server and refreshed based on your business needs. For example, you can use S3 ETags to determine whether an object has changed, and flush the server’s cache in response.

Conclusion

This blog post covered two certificate revocation methods, OCSP and CRLs, that are available on ACM PCA. Remember, when you deploy CA hierarchies for public key infrastructure (PKI), it’s important to define how to handle certificate revocation. The certificate revocation information must be included in the certificate when it is issued, so the choice to enable either CRL or OCSP, or both, has to happen before the certificate is issued. It’s also important to have highly available CRL and OCSP endpoints for certificate lifecycle management. ACM PCA provides a highly available, fully managed CA service that you can use to meet your certificate revocation and validation requirements. Get started using ACM PCA.

 
If you have feedback about this post, submit comments in the Comments section below. If you have questions about this post, contact AWS Support.

Want more AWS Security news? Follow us on Twitter.