Fixing Content-Security-Policies with Cloudflare Workers

Black Hills Information Security, Inc.

Bad Habits: An ANTISOC Operation Same Problem, Different Angles: When Red Team and Blue Team Actually Talk to Each Other How to Identify and Exploit New Vulnerabilities Swapper – A Pure Regex Match/Replace Burp Extension A Practical Guide to BloodHound Data Collection Network Engineering Basics Signed, Trusted, and Abused: Proxy Execution via WebView2 Getting Started In Pentesting – Advice From The BHIS Pentest Lead Cloud Security: Tips and Resources for Securing the Cloud Lessons From A Chatbot Incident How to Lead Effective Tabletops Understanding GRC: How to Navigate Risks and Compliance Standards The “P” in PAM is for Persistence: Linux Persistence Technique Malware Analysis: How to Analyze and Understand Malware OSINT: How to Find, Use, and Control Open-Source Intelligence What to Do with Your First Home Lab When the SOC Goes to Deadwood: A Night to Remember Social Engineering and Microsoft SSPR: The Road to Pwnage is Paved with Good Intentions Common Cyber Threats Finding the Right Penetration Testing Company Deceptive-Auditing: An Active Directory Honeypots Tool The Curious Case of the Comburglar How to Set Smart Goals (That Actually Work For You) Inside the BHIS SOC: A Conversation with Hayden Covington Abusing Delegation with Impacket (Part 3): Resource-Based Constrained Delegation Why You Got Hacked – 2025 Super Edition Abusing Delegation with Impacket (Part 2): Constrained Delegation Abusing Delegation with Impacket (Part 1): Unconstrained Delegation GoSpoof – Turning Attacks into Intel Model Context Protocol (MCP) Bypassing WAFs Using Oversized Requests Getting Started with AI Hacking Part 2: Prompt Injection Wrangling Windows Event Logs with Hayabusa & SOF-ELK (Part 2) DomCat: A Domain Categorization Tool Wrangling Windows Event Logs with Hayabusa & SOF-ELK (Part 1) Microsoft Store and WinGet: Security Risks for Corporate Environments Default Web Content MailFail Commonly Abused Administrative Utilities: A Hidden Risk to Enterprise Security Stop Spoofing Yourself! Disabling M365 Direct Send Bypassing CSP with JSONP: Introducing JSONPeek and CSP B Gone Offensive Tooling Cheatsheets: An Infosec Survival Guide Resource DNS Triage Cheatsheet GraphRunner Cheatsheet Burp Suite Cheatsheet Impacket Cheatsheet Wireshark Cheatsheet Hashcat Cheatsheet EyeWitness Cheatsheet Nmap Cheatsheet Netcat (nc) Cheatsheet Hunt for Weak Spots in Your Wireless Network with Airodump-ng from the Aircrack-ng Suite Detecting ADCS Privilege Escalation Vulnerability Scanning with Nmap Getting Started with NetExec: Streamlining Network Discovery and Access How to Use Dirsearch Augmenting Penetration Testing Methodology with Artificial Intelligence – Part 3: Arcanum Cyber Security Bot How to Design and Execute Effective Social Engineering Attacks by Phone Abusing S4U2Self for Active Directory Pivoting Why Use a Macro Pad? Espanso: Text Replacement, the Easy Way Caging Copilot: Lessons Learned in LLM Security Augmenting Penetration Testing Methodology with Artificial Intelligence – Part 2: Copilot Augmenting Penetration Testing Methodology with Artificial Intelligence – Part 1: Burpference Intercepting Traffic for Mobile Applications that Bypass the System Proxy How to Root Android Phones Communicating Security to the C-Suite: A Strategic Approach Offline Memory Forensics With Volatility Getting Started with AI Hacking: Part 1 Go-Spoof: A Tool for Cyber Deception How to Test Adversary-in-the-Middle Without Hacking Tools Canary in the Code: Alert()-ing on XSS Exploits How to Hack Wi-Fi with No Wi-Fi Why Your Org Needs a Penetration Test Program Burp Suite Extension: Copy For Light at the End of the Dark Web Wi-Fi Forge: Practice Wi-Fi Security Without Hardware Avoiding Dirty RAGs: Retrieval-Augmented Generation with Ollama and LangChain Gone Phishing: Installing GoPhish and Creating a Campaign 5 Things We Are Going to Continue to Ignore in 2025 John Strand’s 5 Phase Plan For Starting in Computer Security Questions From a Beginner Threat Hunter GRC for Security Managers: From Checklists to Influence AI Large Language Models and Supervised Fine Tuning Attack Tactics 9: Shadow Creds for PrivEsc w/ Kent & Jordan One Active Directory Account Can Be Your Best Early Warning Introduction to Zeek Log Analysis Indecent Exposure: Your Secrets are Showing Creating Burp Extensions: A Beginner’s Guide Pitting AI Against AI: Using PyRIT to Assess Large Language Models (LLMs) The Top Ten List of Why You Got Hacked This Year (2023/2024) ICS Hard Knocks: Mitigations to Scenarios Found in ICS/OT Backdoors & Breaches Intro to Data Analytics Using SQL Finding Access Control Vulnerabilities with Autorize The Detection Engineering Process Cyber Risk Lessons We Can Learn From Hurricane Preparedness Intro to Desktop Application Testing Methodology What Is Penetration Testing? Adversary in the Middle (AitM): Post-Exploitation Pentesting, Threat Hunting, and SOC: An Overview

Kassie Kimball · 2021-12-04 · via Black Hills Information Security, Inc.

Kent Ickler //

Background

Over four years ago now, I wrote a blog post on fixing missing Content-Security-Policy by updating configuration on webservers: https://www.blackhillsinfosec.com/fix-missing-content-security-policy-website/. Content-Security-Policies instruct a user’s web browser how it should behave on certain security considerations.

Oh, how times have changed. Here at Black Hills Information Security (BHIS), we’ve actually migrated webservers, hosting companies, security platforms — that list goes on and on. The “best practices” for Content-Security-Policies have changed in the last four years too. On our new hosting platform, we need to set up appropriate content security headers again. Since we now use Cloudflare for our CDN and WAF provider, we have some new opportunities for fronting our Content-Security-Policies outside of the web server itself.

Initial Testing

Before you go about updating your Content-Security-Policies, it’s good to have a clear picture of how your server currently handles/sends Content-Security-Policies. A good way to test this configuration is to use a third-party tool. We can use SecurityHeaders.io to scan our website’s Content-Security-Policy configuration.

Link: https://www.securityheaders.io

In the case below, we’ve had SecurityHeaders.io scan the WildWestHackinFest.com website.

That looks bad, right? Well, maybe. It is important to note that Content-Security-Policies are used to instruct the browser how to handle security concerns within the browser. This is critical on websites where there is user interaction and sensitive information being disclosed. For example, it would be imperative that a banking website, health records portal, or other user-interaction service have appropriate Content-Security-Policy headers. In the scenario where there is no user interaction or no sensitive information disclosed, it becomes less imperative that Content-Security-Policies be configured in a very secured state.

Here’s a good example of a “not-great” configuration scenario: The US Social Security Administration has a portal where users can login and access sensitive information about their account. The portal login landing page is https://secure.ssa.gov:

Alright, so that’s a picture of what not to do.

If you’re looking to correct some of these issues, you have a couple methods afforded to you. The first is to read the blog from four years ago that demonstrates how to fix the issue by configuring your web server with the appropriate Content-Security-Headers. But there is another way.

Cloudflare Workers

Link: https://workers.cloudflare.com/

Cloudflare Workers are a serverless section of server-side-JavaScript that can perform actions or modify web traffic associated with a Cloudflare CDN/WAF protected site. In the case of our earlier example, https://wildwesthackinfest.com is a website that is served by the Cloudflare network. This allows us to use the Cloudflare Workers service to manipulate web traffic without having to update the backend (“origin”) web servers associated with the website. BHIS operates multiple websites and using Cloudflare Workers will also allow us to centralize the response-header manipulation to one programmatic location, rather than managing multiple backend server configuration.

Prerequisites

To allow a Cloudflare Worker to manipulate your web server’s traffic, you must ensure that the webservice has “Orange Cloud” configuration on Cloudflare.

You must also enable the Workers service. After logging into Cloudflare and selecting your domain, click on the “Workers” tab to get the service initiated.

Building the Worker

Once you have Cloudflare Workers enabled — on the Workers tab, click on “Manage Workers” and then “Create Service.” Give your new worker a name and select “HTTP Handler” as the starter template. Then, find the “Quick Edit” button to access the Cloudflare Worker editor/IDE.

Next, replace the existing template code with the following worker code.

Link: https://github.com/Relkci/cf-worker-header-injector/blob/main/worker.js

addEventListener('fetch', event => {
  event.respondWith(handleRequest(event.request))
})

async function handleRequest(request) {

  let originalResponse = await fetch(request)

  // pass in the original response so we can modify some of it.
  let response = new Response(originalResponse.body,originalResponse);

  response.headers.set('X-Frame-Options', 'SAMEORIGIN');
  response.headers.set('Referrer-Policy','same-origin');
  response.headers.set('Content-Security-Policy', 'default-src \'self\' \'unsafe-inline\'');
  response.headers.set('Feature-Policy','camera \'none\'; geolocation \'none\'');
  response.headers.set('X-Content-Type-Options','nosniff');
  response.headers.set('X-XSS-Protection','1; mode=block');
  response.headers.set('Permissions-Policy', 'camera=(), geolocation=(), microphone=()');
  return response
}

The above is a starter template for Content-Security-Policies. You should configure each policy as you see fit for your website. Any headers you add here will later be added to your web server’s HTTP Responses. Press “Save and Deploy.” Don’t worry, nothing is live yet.

In the next step, we will associate the new Cloudflare Worker with your website. After we do this, your worker will begin intercepting HTTP responses from your backend (origin) web server and injecting the headers specified in the worker code.

Apply the Worker

After you have your worker saved, return to the Cloudflare Workers tab associated to your domain and press “Add Route.” The Add Route functionality associates your new worker with your web server. As soon as this is done, your backend (origin) web server’s HTTP responses will be intercepted by the worker and the worker will inject the headers according to its code.

After the worker has been routed with your website, you’ll find the Workers tab now lists the active routes.

Content-Security-Policy Refinement

As soon as you associate the worker, remember that restrictive Content-Security-Policy settings may break rendering or functions of your website if you use JavaScript, CSS, or external resources. The Content-Security-Policy instructs the browser which resources are safe to load. A restrictive policy may prevent the loading of necessary resources. As you update your Content-Security-Policy for your website, you may see errors or rendering issues that are related to the new Content-Security-Policies. The easiest way to identify the Content-Security-Policy violations is to use the browser’s developer mode (typically, press F12). The Network or Console tabs will typically alert you to violations that will need to be accounted for in the Content-Security-Policy.

Here’s an example of a console error:

This error lets us know that within the script-src setting in the CSP, the stats.wp.com is not explicitly trusted. There are various ways we can go about adding trust for this source; one method is to add https://stats.wp.com to the script-src setting. After updating and re-deploying the worker, refresh and you should see that the resource was allowed to load and consequently did not present an error in the developer console.

Worker Pricing

One last note regarding the addition of the Workers service and the creation of the Worker later — Cloudflare does provide a free tier to workers, and two different price schemes: Bundled and Unbound.

If you’re worried about what that might cost you, we were too. We used the Unbound pricing scheme because it was the default. We’re probably safe to update the model to Bundled since our Workers’ execution time is less than 50ms. The Workers dashboard will show the execution time so you can make your own decision.

Anyhow, with the Unbound pricing model, we weren’t at much risk regardless. Here’s a picture of one month of Bundled worker activity on one of our busier websites:

At the end of the day, $5.38 wasn’t a bad cost to us for +1M website hits.

It’s worth noting that you can (and perhaps should) limit your worker route to typically only affect website resources that will instruct a web browser to create additional sub-queries for resources. These are typically HTML, PHP, ASP, etc. Images and videos usually do not need a CSP as they typically do not instruct the browser to create additional HTTP requests. For example, you may omit a web server directory that contains only images. The worker routes can be stacked. Setting a route to a service of “None” will prevent execution of workers.

OWASP Secure Headers Project

Link: https://owasp.org/www-project-secure-headers/

For a good resource on the types of Security Headers and their appropriate use, check out the OWASP documentation on Response Headers.

Content-Security-Policy Reference

Link: https://content-security-policy.com/

Another good resource, perhaps a bit more technical, is the Content-Security-Policy Resource. This is an update to date compilation of current CSP mechanics and support matrix by browser.

Cloudflare HSTS Settings

After you have set up your worker and configured your CSP according to your website’s needs, there are a handful of other settings in Cloudflare that will also need configuration to fully align your website with secure practices (and get that better score on SecurityHeaders.io).

Be aware however, that if your website or subdomains of your website are not also aligned according to these changes, you may experience a headache getting everything setup. Of note, once HTTP Strict Transport Security (HSTS) is enabled, there is no easy way to back out of configuration. Ensure that subdomains of your domain and your TLD website are configured with HTTPS appropriately before continuing.

On your Cloudflare Dashboard, jump to your domain and go to the SSL/TLS section and then the Edge Certificate’s tab. You will find two sections that you should configure.

Always Use HTTPS

“Always Use HTTPS” will automatically redirect port 80 from Cloudflare’s edge to HTTPS (443) for your website, regardless of if your origin server supports port 80 or serves other content from port 80.

HTTP Strict Transport Security (HSTS)

Enable HSTS and configure settings according to your need. It’s best practice to have the max-age header set to at least 6 months, but this setting carries with it some risk. Once enabled, you must always support HTTPS on your site, or a possibility exists that users will not be able to access your website.

Minimum TLS Version

Unless your users are likely to be using unsupported or obsolete web browsers or clients, it is typically safe to enable a minimum TLS version of 1.2.

Versions 1.1, 1.0, and SSL version 2 and version 3 are considered insecure by most security analysts despite a low threat probability.

Enable TLS 1.3

It’s the new rage, for this year. Enable this option to support TLS 1.3 on the Cloudflare Edge for your web server.

Post-Change Testing

After you have deployed the Cloudflare Worker and updated your Cloudflare Edge SSL/TLS options, it’s time to check out if you’ve made any impact on your “Security Headers” score. Let’s use SecurityHeaders.io again to see if there has been improvement.

More Work — CSP Refinement

We’re capped at an “A” because the website we host includes an “unsafe-inline” option in the Content-Security-Policy. Ultimately, you want to get rid of those, but it can be used as a stopgap to make your website function as you continue to build a more restrictive CSP. It’s important to note that at the end of the day, effectively every resource source used on a website should be explicitly listed within the CSP. “Unsafe” source inclusions are… “unsafe”ish.

Assisted Configuration — CSP-as-a-Service

There are some new products out that will assist you in determining your baseline Content-Security-Policy based on your existing website resources used. There is a new CSP method, “report-uri,” that instructs web browsers to report CSP violations to a webservice that then notifies web administrators of the issue. This is something akin to how DMARC reporting works, allowing a domain owner to receive reports of potentially fraudulent email, except in this case the alert is about CSP violations.

A handful of Content-Security-Policy report Service-as-a-Service products exist. They are configured to be the recipient of your users’ browser’s CSP violation reports. They aggregate the reports and then provide you with an appropriate updated baseline Content-Security-Policy.

Frankly, this is something that you can do yourself without the use of a product. Load your webpage and check the Developer-Tools console for CSP violations. Regardless, a third-party service could go a long way in automating some of the monotony, especially for very large or complex websites where manually confirming the entirety of the website rending and functioning appropriately is not possible.