I stumbled upon the Apache Batik library while researching other Java-based products. It immediately caught my attention, as this library parses Scalable Vector Graphics (SVG) files and transforms them into different raster graphics formats (i.e., PNG, PDF, or JPEG). I was even more encouraged when I looked at the Batik documentation. It was obvious that such a library could be prone to Server-Side Request Forgery (SSRF) issues (e.g., loading of images from remote resources). However, the documentation shows that Batik can also:
· Load and execute remote Java classes.
Those are some neat features! On the other hand, Apache Batik protects its users from both SSRF and remote code execution (RCE) vulnerabilities through the various security modes it offers. In this blog post, I am going to show you:
· How I bypassed the default security modes: DefaultScriptSecurity (CVE-2022-40146) and DefaultExternalResourceSecurity (CVE-2022-38398).
· How to abuse the SSRF in the default Batik Transcoder to make arbitrary HTTP GET requests or to trigger an NTLM challenge.
· What configurations are vulnerable to the RCE through remote class loading.
Before we get into the details, here’s a quick video demonstrating a remote code execution vulnerability exploited through remote JAR loading.
Sample Web Application Endpoint
Apache Batik can be used in different ways. There is a pretty good chance that you have already seen it in web applications during the conversion of SVG to PDF/PNG/JPEG. Let’s define a sample endpoint that performs such a conversion:
This endpoint performs the following actions:
· Retrieves and decodes the base64-encoded SVG file.
· Creates the Apache Batik JPEG Transcoder.
· Converts the SVG to JPEG.
This is a common usage of Apache Batik. Note that there is a risk of an easy SSRF if the SVG loads an image from an external resource. Batik tries to protect against such a scenario with its external resource controls, although some of these have existing bypasses with their own CVEs assigned. Let’s quickly review those resource controls.
Apache Batik External Resource Controls
Apache Batik resource controls can be divided into two main categories:
· Script execution
· External resources controls (like images).
Let’s take a brief look at the scripting controls. They control whether script provided within an SVG will be executed. The external resource controls are similar to the scripting controls but apply to the fetching of resources such as images. For a full description of security controls, you can access the documentation here. Here’s a quick overview of the available ScriptSecurity implementations:
· NoLoadScriptSecurity – scripts are completely blocked.
· EmbededScriptSecurity [sic] – scripts embedded in the document can be executed when properly referenced.
· DefaultScriptSecurity – Embedded external scripts (as above) plus scripts coming from the same origin as the document referencing them are allowed.
· RelaxedScriptSecurity – scripts from any location can be loaded.
In my research so far, I have focused only on the default security controls.
Note that the default security control has a concept of “origin”, but what exactly does this mean? It means that the resource or script will be loaded only if it originates from the same “host” as the SVG file. For example:
· If we load a local SVG file, we can also load local scripts or resources.
· If we load a local SVG file, we cannot load scripts or resources from remote origins (e.g., through HTTP or SMB).
· If we load an SVG file through the HTTP protocol, we can load remote scripts from the same host through either HTTP or any other supported protocol, such as SMB.
In order to avoid any confusion, let’s quickly describe what “loading an SVG through the HTTP protocol” means in Batik. It specifically means that the HTTP URL is directly provided to the Batik
This does not look like a particularly common scenario, because the attacker would need to control the URL from which the SVG file is (directly) loaded into Batik.
Accordingly, the default security controls seem to be appropriate. When a local SVG file is loaded, or an SVG is provided as an InputStream, the default controls should block the loading of any remote resources.
Now let’s see how we can bypass those security checks.
DefaultScriptSecurity and DefaultExternalResourceSecurity vs URL.getHost
To start, let’s dig into the DefaultScriptSecurity constructor, which is responsible for our security check. Please note that the code of DefaultExternalResourceSecurity is almost identical, thus it will not be presented.
At , the scriptURL and docURL of ParsedURL type are provided. For the sake of simplicity, let’s say that ParsedURL wraps the Java URL class.
At  and , the respective host strings are retrieved with the getHost method. Under the hood, it retrieves the output of the Java URL.getHost.
At , hosts obtained in points  and  are compared. If they are the same, the code flow continues, and the exception will not be thrown.
At , a SecurityException is thrown if the security check is not successful.
If the document host and the script host are the same, the exception will not be thrown, and the script or resource will be loaded. Next, let’s look at how the Java URL.getHost behaves for different protocols:
It’s pretty simple. The host for a local file will be equal to null, whereas for the remote files shown here (referenced by either UNC path or HTTP) it will be equal to evil.com. If the Apache Batik TranscoderInput is created with the InputStream, the getHost method will also return null.
It seems that the default security routines work properly, and we will not be able to load remote resources during processing of SVGs coming from local files or input streams.
Unfortunately, a trivial bypass exists, and honestly, Java itself is the likely culprit. Let’s see the output of the getHost getter for the JAR protocol:
What? It seems that the host for the JAR protocol is also null. In order to properly retrieve the host from the JAR URL, the following cumbersome code can be used. We retrieve the file member from the URL and then use it to create a new URL, from which we can get the host.
In this way, the security check can be easily bypassed with the JAR protocol. Let’s have a look at a sample SVG that leads to SSRF through the image tag.
This SVG will bypass the DefaultExternalResourceSecurity control and will make an HTTP GET request to the attacker’s server. Please note that we can also use the syntax “jar:file://…”, in order to get an NTLMv2 challenge-response in a Windows environment and potentially perform an NTLM relaying attack.
Now that we have shown how to bypass the default external resources controls to get an easy SSRF through the image tag, let’s see how to get remote code execution. This has been fixed in Batik 1.15.
Apache Batik Remote Class Loading Feature
While digging through the Apache Batik documentation, I found an intriguing feature: “Referencing Java Code From a Document”.
Batik implements the Java bindings for SVG, and thus allows Java code to be referenced from script elements…
In order to use this extension, the type attribute of a script element must be set to application/java-archive. In addition, the xlink:href attribute must be the URI of a jar file that contains the code to run.
It seems that we can provide a script of type application/java-archive referencing a properly structured JAR file. We already know that we can bypass the DefaultScriptSecurity check and load files from remote locations, thus it looks like a win!
Luckily, the script execution is not enabled by default in the Apache Batik transcoder and some configuration modifications must be applied:
1) Enable script execution
Script execution can be enabled in the transcoder with the following line of code:
Please note that the Apache Batik is a “do it yourself” library and often requires a lot of customization. There is a chance that you will see such a configuration in the wild. Also, Batik defines some methods that allow dynamic verification of whether the SVG file contains scripts or not. This method can be used to automatically enable script execution if needed.
2) Fix the logical flaw to allow execution of scripts of type “application/java-archive”.
It seems that Apache Batik’s definition of default-allowed script types contains an unintended error. Let’s have a quick look:
You can clearly see that the script types are separated with a comma followed by a space. Now, let’s look at the method that retrieves the list of allowed types:
The list is created with StringTokenizer, where the delimiter is set to a comma without a space! According to that, all the allowed types other than the first one (ECMASCRIPT) will contain a leading space. For example, the allowed type will be “ application/java-archive” instead of “application/java-archive”.
The code will compare the script type declared in the SVG against the type specified in the list. You may think that this is not a problem, since we can declare a script type in the SVG that also includes a space:
<script type=” application/java-archive”>
This will not work, though. The type check for remote JAR loading is performed twice. The second check verifies if the script type is equal to “application/java-archive”, without a leading space. As a result, we will not be able to pass both checks. To enable remote class loading, we must manually modify the list of allowed script types by making an API call as follows:
Now, let’s look at how to perform remote JAR loading in Batik. To begin, we must create a class that implements the EventListenerInitializer interface. Then, we must define the initializeEventListeners method. This method will be executed after the JAR is loaded. Here’s an example of a malicious class:
The MANIFEST.MF file included in the JAR needs to specify SVG-Handler-Class. The following snippet presents an exemplary manifest:
Finally, we specify a malicious SVG that loads such a JAR and bypasses the security policies:
The following screenshot presents a proof of concept where the JAR was loaded through the HTTP protocol and the code was executed, spawning a curl process and making a request to our HTTP server.
Success! We were able to bypass the restrictions defined in DefaultScriptSecurity, load the JAR file from the remote location, and achieve remote code execution.
Scripts Enabled with Remote JAR Loading Disallowed
Due to the logical flaw described in the previous section, it’s likely that you will encounter the following configuration:
· Scripts enabled.
· The setting for allowed script types is untouched, so that, in practice, only ECMAScript is enabled.
The following snippet presents an SVG file with an ugly looking script that will execute the attacker’s command.
The following screenshot demonstrates the achieved code execution:
Execution of ECMAScript is a well-documented feature of Apache Batik and this behavior will not be fixed. The official security guide recommends securing your application with the Java SecurityManager when scripting is enabled. I’m taking a guess, but I think that we will never see a completely secure implementation of SecurityManager within Batik. It would likely break too much functionality and make the application unusable. This is still merely speculation on my part. It should also be noted that running scripts is disabled by
default and must be explicitly enabled.
UPDATE: Apache Batik 1.16 has been recently released. It introduces a hardening to Rhino script execution mechanisms. However, it’s hard to call this hardening a strong one. You are still able to call almost any class whose full name starts with “org.”. Saying that, you can still easily abuse many methods included in Apache libraries. This security check can be found here.
To wrap things up, the security checks implemented in the default external resource controllers of Apache Batik could be bypassed prior to Batik 1.15, in order to:
· Perform SSRF through the JAR protocol, producing either an HTTP GET request or NTLM relaying via an UNC path.
· Potentially achieve RCE through remote JAR loading, provided that a non-default configuration was applied.
The bugs discussed here in DefaultScriptSecurity and DefaultExternalResourceSecurity were fixed in version 1.15. Still, Apache Batik may allow the execution of arbitrary Java code when script execution is enabled, even in the latest version 1.16.
If you are using Batik or are interested in finding additional bugs, here are some things to consider:
· Developers – do not allow scripts to execute through Batik. Apply the strictest controls possible.
· Pentesters/Bug Hunters – when you test functionality that accepts SVG, try to use the SVG files presented in this blog post.
· Vulnerability researchers – when your target uses Apache Batik for SVG parsing or conversions, analyze its configuration carefully. You might have an easy win over there.
Written by admin
ZDI-24-229: Linux Kernel ksmbd Session Key Exchange Heap-based Buffer Overflow Remote Code Execution VulnerabilityMarch 2, 2024This vulnerability allows remote attackers to execute arbitrary code on affected installations of Linux Kernel. Authentication is not required to exploit this vulnerability, but only systems with ksmbd enabled are vulnerable. The ZDI has assigned a CVSS rating of 9.0....
ZDI-24-228: Linux Kernel ksmbd Negotiate Request Out-Of-Bounds Read Information Disclosure VulnerabilityMarch 2, 2024This vulnerability allows remote attackers to disclose sensitive information on affected installations of Linux Kernel. Authentication is not required to exploit this vulnerability. However, only systems with ksmbd enabled are vulnerable. The ZDI has assigned a CVSS rating of 8.6....
ZDI-24-227: Linux Kernel ksmbd Chained Request Improper Input Validation Information Disclosure VulnerabilityMarch 2, 2024This vulnerability allows remote attackers to disclose sensitive information on affected installations of Linux Kernel. Authentication is not required to exploit this vulnerability, but only systems with ksmbd enabled are vulnerable. The ZDI has assigned a CVSS rating of 9.3....