An experience with Daimler’s vulnerability reporting program

Eaton

Background

Daimler AG is a German multinational automotive corporation commonly known for being the parent company of Mercedes-Benz. Daimler runs a white hat/vulnerability reporting program.

Mercedes-Benz USA, LLC (MBUSA for short) is the distributor of Mercedes-Benz cars in the United States. MBUSA runs a variety of websites, one of those being a Dealer Help Center website built on Oracle’s Service Cloud: mercedes-benz-dealer.custhelp.com. This website is only meant to be accessed in Mercedes-Benz dealerships on computers running their “NetStar” system. It’s accessible to the outside world, but everything is supposed to be protected with a login, or limited to access only from NetStar.

This post will go on to explain how 3 vulnerabilities could have exposed a decade’s worth of sensitive content on MBUSA’s Dealer Help Center website.

The Vulnerabilities

During the course of a simple Google search, I discovered an indexed PDF from the Dealer Help Center website. By modifying my search query to exclusively find all indexed content for the Dealer Help Center website, like this, this, and this, I was able to locate a fairly significant number of confidential documents. Some documents were not confidential at all, such as vehicle instruction manuals, published spec sheets & images. However, there were some documents that were clearly marked confidential and meant for Mercedes-Benz dealership management eyes only. There are also documents containing personal information, like cell phone numbers for MBUSA management individuals. I did not find any customer data nor do I believe any customer data was at risk since this website is just a help resource for dealers.

Looking at the URL for the PDFs (example: https://mercedes-benz-dealer.custhelp.com/app/answers/detail/a_id/3553 – I call this the “answers” endpoint/page), there is an ID number at the end. I wondered what would happen if I tried different ID numbers. Some times I was asked to log in, perhaps because the page didn’t exist and it was redirecting to somewhere else. Frequently, however, it gave me a downloadable PDF. Below is the header from one:

And here’s a page from another:

This is a classic example of an enumeration attack on an unsecured endpoint. It would take very little effort to create a script to go through thousands of ID numbers and download any PDFs the request returned. Some of the documents go back more than 10 years, so there was a possibility of being able to download a decade’s worth of confidential content. I did not check for any rate limits – attempting to dump their entire network would have been abusive and likely a violation of the program rules.

Important to note is that since I did not attempt to download every single document possible, the full scale of sensitive data exposure is not known.

But wait, there’s more!

The Google searches revealed another endpoint. Example: https://mercedes-benz-dealer.custhelp.com/ci/fattach/get/10704 – I call this the “attach” endpoint.

There is another opportunity for an enumeration attack there. This appears to be where all help article attachments are stored, so it is an attractive target for data exfiltration. All I needed to do was change the number at the end to be able to download more PDFs.

Oh, and one more thing!™🍏

This one is minor compared to the last 2, but still reveals sensitive data. When visiting an answers page that is not a document, and instead an actual page with simple text content, the page’s text/body content is protected, but the HTML title leaks out! (look at the browser tab)

A wordy page title has the potential to give away some interesting information, maybe even internal announcements of upcoming car models?

Reporting Timeline

I composed an email with clear instructions on how to reproduce the problem, explained the possible impact if left unfixed, and provided a few examples of exfiltrated confidential data. I sent it to Daimler’s security email address listed on their white hat website page. The back-and-forth lasted almost 5 months. The entire email timeline is below (US Eastern time):