Behind the Scenes of Merry Maker

February 1, 2022

Eric Brandel

Director of Engineering, Cybersecurity

Caleb Walch

Distinguished Engineer, Cybersecurity

We recently introduced Target’s cybersecurity tool “Merry Maker” and open sourced the framework to help others protect against digital skimming. Here’s a deep dive into how the solution works and how we built it.
 
When beginning to design the system, we first looked for the advantages that we as defenders have when dealing with this kind of threat. First, the attacker’s code must be visible to the browser and, by extension, us. Second, to complete their attack they need to exfiltrate data. Third, the skimming code is frequently obfuscated or contains signatures unique to malicious scripts and rules can be written to detect this.
 
How Merry Maker Works
 
screenshot of code snippet describing a payload wallet rule with Eric Brandel listed as author
 
Merry Maker operates by performing realistic website activity, similar to what an actual guest would do. The Javascript and network activity generated by this is then saved for analysis. Merry Maker also looks for things including known malicious domains and new domains that it has not seen before. If either of those are discovered, the tool creates an alert. It can also run Yara rules against the contents of the requests, so defenders can look to see if specific data or types of data are being exfiltrated. For instance, it can detect a specified credit card number or the existence of base64 encoded data.
 
screenshot from Terminal with a rule "digital_skimmer_coffeemokko" running with the meta description listed as "CoffeMokko Digital Skimmer" showing Eric Brandel as the author along with strings and conditions listed below these elements.
 
Merry Maker also grabs all first and third party Javascript and can capture the present state of the Document Object Model (DOM) at any moment. All this data is returned and run through another series of rules. These rules are also written in Yara, enabling teams to quickly add detection as they see fit, and to utilize any publicly available rules.
 
Screenshot of code commands including "export class: GoogleAnalyticsRule" with a scan event called "MerryMaker" being run
 
And, finally, additional detections can be added via Typescript. This allows for more advanced detections to be created.
 
screenshot of source editor showing a TargetTechBlog log "visiting site"
 
Behind the scenes, Merry Maker utilizes Puppeteer and Headless Chrome. Puppeteer enables us to control Chrome using instructions written in Javascript, which are maintained and tested via the User Interface (UI). There is support for basic versioning of these sources and it is straightforward to switch back and forth between different copies.
 
Running that source results in thousands of events being generated. All of them are brought back and stored, and rules can be created to run against any of them. If a rule hits, an alert is fired. Merry Maker supports three different alert types:
 
  • Basic, which are displayed in the UI
  • Kafka
  • GoAlert
 
Merry Maker also supports utilizing Indicators of Compromise (IOCs) and will monitor the traffic it sees for known digital skimming domains. These are managed through the UI and enable the user to easily add domains to this “watch list.” These can be from an intel provider or from publicly available lists.
 
How We Built Merry Maker
 
diagram with two blue rectangles side by side. The left one reads "UI/API" and has "scanner" and "job handler" below, and the right side reads "JS Scope" with "headless browser" below
 
Merry Maker has four main components:
  • JS Scope (headless browser)
  • Scanner (Rules)
  • UI/API
  • Job Handler
 
horizontal arrow diagram pointing to the right with plotted points reading as follows: "Network: request, response," "Javascript: function calls, event listeners, DOM manipulation," and "Client: cookies, screenshots."
 
JS Scope
Merry Maker is built to mimic the behavior of real users as they navigate a site: searching for a product, adding it to the cart, entering payment details, and submitting the order. All client-side code that runs during those steps can only be fully executed in a real browser. JS Scope depends on Puppeteer to drive a headless browser with the addition of pre-injected scripts to hook native function calls. JS Scope gives us a better understanding into the behavior of client-side code beyond static analysis. Every hooked function call, web request, response, cookie, and other relevant calls are captured and analyzed in near real-time.
 
diagram with a blue cylinder that reads "queue" in the center with arrows pointing out to four purple rectangles that read "scanner" - the four connecting arrows read (clockwise from right to left) "unknown-domain-rule," "yara-rule," "rule-result," and "web-request"
 
Scanner
JS Scope events are sent over a message bus to be consumed by the Job Coordinator and Scanner. The Scanner will inspect the event and determine which rules need to be run. A single event may be scheduled to be inspected by more than one rule.
 
For example, a web request may be run through an IOC Rule, Unknown Domain Rule, and a Yara Rule. When a rule alerts on an event, it is immediately sent to the Job Coordinator and forwarded to the preconfigured endpoints (Kafka, Web hook, SMS). All these actions take place while the site is being scanned, instead of waiting for the scan to complete.
 
What Does Merry Maker Look Like?
 
In order to understand the functions of Merry Maker, let’s also walk through the UI in detail to share what each view represents and where a user can go to find specific information.
 
screenshot of a dashboard with alerts such as "google.analytics" and "unknown.domain" on the left side and scans such as "eb.com" and "MyBalanceNow" on the right
 
Dashboard
The Dashboard view presents the current state and health of the system. Here users can quickly see the most recent alerts, active and recently run scans, and the number of events in the queue.
 
screenshot of dashboard with "alerts view" selected, showing alerts that have come through for each active rule
 
Alerts
The Alerts view lists all recorded Rule Alerts with event source attribution.
 
screenshot of the "sites" view, showing all monitored sites and their current active status
 
Sites
The Sites view lists all monitored sites and their Active status. Users can manage existing site records, delete, and create new sites.
 
screenshot of the IOC view in the dashboard, showing active IOCs used per rule type
 
IOCs
The IOCs view allows users to manage active IOCs used in alerting for various rule types.
 
allow list screenshot showing the keys for each type of alert that is allowed and to stop rules from inspecting known events
 
Allow List
The Allow List feature prevents alerting on false positives and stops rules from inspecting known events.
 
screenshot of a HTTP GET request during a scan, showing domain, URL and headers based on the available rules
 
Request
Here is an example of an HTTP GET request captured during a scan. This data is used to alert on the domain, URI, and headers based on the available rules.
 
this example shows Timeout and ElementSource calls both hooked during a scan - these can be used to aid in detection and tracking third-party script behavior
 
Function Calls
An example set Timeout and ElementSource call hooked during a scan. These calls can be used for detection and tracking third-party JavaScript behavior.
 
Merry Maker has provided an extra layer of security and assurance for our guests as they shop online, and in open sourcing this project, it’s our hope that the rest of the security community can benefit from the framework and what we’ve learned.
 
Click here to access the Merry Maker open source project.
Click here to read an introduction to Merry Maker.

RELATED POSTS

Meet Merry Maker: How Target Protects Against Digital Skimming

By Chris Carlson, February 1, 2022
Credit card skimmers are custom-made additions to payment devices (like ATMs or gas pumps) that criminals build and use to steal a consumer’s payment information.