CVE-2024-9308

Executive Summary

CVE-2024-9308 is an open redirect vulnerability discovered in haotian-liu/llava version v1.2.0 (LLaVA-1.6). This flaw allows a remote unauthenticated attacker to craft a specially crafted URL that redirects users to arbitrary external websites. The vulnerability is classified as CWE-601 (URL Redirection to Untrusted Site) and carries a CVSS score of 6.1 (Medium). Although the EPSS score is currently 0.0%, the public availability of proof-of-concept details and the lack of an official patch make this a significant threat for phishing, malware distribution, and credential theft campaigns. The project is an open-source multimodal AI model, and no supply chain package manager (NPM, PyPI, Go, Ruby) is involved.

CVSS Breakdown & Threat Level

• CVSS Vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:L/I:L/A:N
• Base Score: 6.1 (Medium)
• Attack Vector: Network (AV:N)
• Attack Complexity: Low (AC:L)
• Privileges Required: None (PR:N)
• User Interaction: Required (UI:R) – the victim must click the crafted link
• Scope: Changed (S:C) – the redirect can lead to a different security domain
• Confidentiality Impact: Low (C:L) – potential for credential theft
• Integrity Impact: Low (I:L) – redirection to malicious sites
• Availability Impact: None (A:N)

Threat Level: Medium. The vulnerability is easy to exploit but requires user interaction. The lack of a patch and public exploit details elevate the risk for targeted attacks.

Deep Dive & Mechanics

The open redirect vulnerability exists in the LLaVA web interface, likely in a URL parameter that handles redirection after authentication or page navigation. An attacker can craft a URL such as:

https://victim-llava-instance.com/redirect?url=http://evil.com

When a user clicks this link, the application does not validate the destination domain, allowing redirection to any external site. This can be used to:

• Phishing: Redirect users to a fake login page to steal credentials.
• Malware Distribution: Redirect to a drive-by download site.
• Credential Theft: Capture OAuth tokens or session cookies via a malicious landing page.

The vulnerability is triggered by a lack of server-side validation of the url parameter. The application likely uses a function like redirect(url) without checking if the URL belongs to an allowed domain.

Vulnerable vs. Patched Code

Vulnerable Code (v1.2.0):

# Example vulnerable redirect handler
@app.route('/redirect')
def redirect_handler():
    target = request.args.get('url')
    return redirect(target)

Patched Code (Hypothetical):

# Patched version with domain whitelist
from urllib.parse import urlparse

ALLOWED_DOMAINS = ['example.com', 'localhost']

@app.route('/redirect')
def redirect_handler():
    target = request.args.get('url')
    parsed = urlparse(target)
    if parsed.netloc in ALLOWED_DOMAINS:
        return redirect(target)
    else:
        return abort(400, "Invalid redirect target")

The patch adds a whitelist check to ensure the redirect target is within allowed domains. As of now, no official patch has been released.

Exploitation & Lab Setup

Red Team (Exploitation)

Prerequisites:

• A running instance of LLaVA v1.2.0 (LLaVA-1.6)
• A controlled external server (e.g., http://evil.com)

Exploit Steps:

1. Identify the vulnerable redirect endpoint (e.g., /redirect?url=).
2. Craft a malicious URL:

   http://victim-llava-instance.com/redirect?url=http://evil.com/phishing.html
   

3. Distribute the link via email, social media, or other channels.
4. When the victim clicks, they are redirected to the attacker-controlled site.

Proof-of-Concept (Python):

import requests

target = "http://victim-llava-instance.com/redirect?url=http://evil.com"
response = requests.get(target, allow_redirects=False)
print(f"Status: {response.status_code}")
print(f"Location header: {response.headers.get('Location')}")

Blue Team (Detection & Hardening)

Detection:

• Monitor web server logs for unusual redirect endpoint usage.
• Look for url parameters containing external domains.
• Use the Sigma rule below to detect suspicious redirects.

Hardening:

• Implement a whitelist of allowed redirect domains.
• Validate the url parameter using a server-side URL parser.
• Add a warning page before redirecting to external sites.

Indicators of Compromise (IOCs)

• URL Pattern: /redirect?url=http:// or /redirect?url=https://
• Suspicious Domains: Any external domain not in the whitelist
• Log Entries: GET /redirect?url=http://evil.com with a 302 response
• User-Agent: No specific pattern, but automated scanners may use common libraries (e.g., Python-requests)

Detection Engineering

YARA Rule

rule CVE_2024_9308_Open_Redirect {
    meta:
        description = "Detects open redirect attempts targeting LLaVA v1.2.0"
        author = "Valters IT"
        date = "2025-04-10"
        reference = "CVE-2024-9308"
    strings:
        $redirect = "/redirect?url=" nocase
        $http = "http://" nocase
        $https = "https://" nocase
    condition:
        $redirect and ($http or $https)
}

Sigma Rule

title: CVE-2024-9308 Open Redirect in LLaVA
id: 8a9b3c2d-1e5f-4a7b-8c9d-0e1f2a3b4c5d
status: experimental
description: Detects HTTP requests to the vulnerable redirect endpoint with external URLs
author: Valters IT
logsource:
    category: webserver
    product: apache
detection:
    selection:
        cs-method: 'GET'
        cs-uri-stem: '/redirect'
        cs-uri-query|contains: 'url=http'
    condition: selection
falsepositives:
    - Legitimate redirects to allowed domains (if whitelist exists)
level: high

Suricata Rule

alert http any any -> any any (msg:"CVE-2024-9308 LLaVA Open Redirect Attempt"; flow:to_server; content:"GET"; http_method; content:"/redirect?url="; http_uri; content:"http://"; distance:0; within:100; classtype:attempted-recon; sid:1000001; rev:1;)

Remediation & Mitigation

1. Immediate Action: Disable the redirect functionality or restrict it to internal domains only.
2. Code Fix: Implement a whitelist of allowed redirect targets (see patched code example).
3. Workaround: Use a reverse proxy (e.g., Nginx) to block requests to /redirect with external URLs:

   location /redirect {
       if ($arg_url ~* "^https?://") {
           return 403;
       }
   }
   

4. Monitoring: Enable logging and alerting for the /redirect endpoint.
5. User Education: Warn users not to click suspicious links related to LLaVA instances.

Valters IT
Securing Digital Frontiers