Modern browsers support quite a few HTTP headers that provide an additional layer in any defense-in-depth strategy. If present in an HTTP response, these headers enable compatible browsers to enforce certain security properties. In early 2016, we undertook an Adobe-wide project to implement security headers.
Implementing a security control in a medium to large scale organization, consisting of many teams and projects, presents its own unique challenges. For example, should we go after select headers or all, which headers do we select, how do we encourage adoption, how do we verify, and so on. Here are a few interesting observations made in our quest to implement security headers:
Browser vs. Non-browser clients
Compatible browsers enforce security policies contained in security headers while incompatible browsers or non-browser clients simply ignore them. However, additional considerations are necessary that may be unique to your setup. For example, Adobe has a range of clients: full browser-based, headless browser-based (e.g., Chromium embedded framework), and desktop applications. Implementing the security property required by a HTTP Strict Transport Security (HSTS) header (i.e. all traffic is sent over TLS for all such clients) requires a combination of an enabling HSTS header and 302 redirects from HTTP to HTTPS. This is not as secure. The reason is that incompatible headless browser-based clients or desktop applications will ignore a HSTS header sent by the server-side. Also, if the landing page URL is provided over HTTP, the clients will continue to send requests over HTTP if the 302 redirect approach is not used. The problem of updating all such clients to have HTTPS landing URLs, which would require updating clients installed on customers’ machines, is a thorny problem. The key is to understand the unique aspects of your applications and customer base to determine how each header could be implemented.
We decided to adopt a phased implementation. For Phase 1: X-XSS-Protection, X-Content-Type-Options, X-Frame-Options. For Phase 2: HSTS, CSP (we haven’t yet considered the Public Key Pinning Extension for HTTP header). Dividing massive efforts into phases offers two main benefits:
- ability to make initial phases lightweight to build momentum with product teams – later phases can focus on more complex tasks
- the opportunity to iron out quirks in processes and automation.
Application stack vs. Operations stack
Who should we ask to enable these headers? Should it be the developers writing application code or the operations engineer managing servers? There are pros and cons of each option (e.g. specifying many security headers in server configurations such as nginx, IIS, and apache is programming language agnostic and doesn’t require changes to the code). Introducing such headers could even be made part of the hardened deployment templates (i.e. Chef recipes). However, some headers may require deeper understanding of the application landscape such as which sites, if any, should be allowed to frame it. We provided both options to teams and the table below provides some useful links that helped inform our decisions.
Affecting the change
We leveraged our existing Secure Product Life Cycle engagement process with product teams to affect this change. We use a security backlog as part of this process to capture pending security work for a team. Adding security header work to this backlog ensured that the security headers implementation work would be completed as part of our regular processes.
Having provided necessary phased guidance, the last piece of the puzzle was to develop an automated way to verify correct implementation of security headers. Manually checking status of these headers would not only be effort intensive but also repetitive and ineffective. Using publicly available scanners (such as https://securityheaders.io) is not always an option due to a stage/dev environments’ accessibility restrictions or specific phased implementation guidance. As noted here, Adobe has undertaken several company-wide security initiatives. A major one is the Security Automation Framework (SAF). SAF allows creating assertions, with various programming languages supported, and run them periodically with reporting. First, we organically compiled a list of end points for Adobe web properties that were discovered through other initiatives. Then Phase 1 and Phase 2 header checks were implemented as SAF assertions to run weekly as scans across the sites on this list. These automated scans have been instrumental in getting a bird’s eye view of adoption progress and give information needed to start a dialogue with product teams.
Security headers provide a useful layer in any defense-in-depth strategy. Most of these are relatively easy to implement. The size of an organization and number of products can introduce challenges that are not necessarily specific to implementing security headers. Within any organization, its critical to plan security initiatives so they help form a malleable ecosystem to help ease the implementation of future initiatives. It does indeed take a village to implement security controls in an organization as big as Adobe.
Senior Security Researcher