August 14, 2017 By Larry Loeb 2 min read

Researchers recently found that three of the most popular version control systems can be tricked into running arbitrary code.

Joern Schneeweisz, a security researcher for Recurity Labs, discovered vulnerabilities that could enable threat actors to steal source code from the code repository or hijack the machine that runs the repository by copying a source code project via a SSH link. This link can be hidden inside modules of the tool so that no direct user action is required to exploit the fault.

Inside the Git LFS Vulnerability

The first vulnerability, which can be configured by a .lfsconfig file, was found in Git LFS, a tool used on GitHub to clone repository files. If a URL in a submodule contains “ssh://-oProxyCommand=some-command,” instead of obtaining a clone, Git LFS would perform the command on the user’s computer. Calling this URL recursively could execute multiple commands. The payload is triggered when the module is updated.

This vulnerability also showed up in the GitLab implementation and was tracked with CVE-2017-12426. Additionally, Schneeweisz discovered that the parent Git project was open to the flaw as well.

Two Additional Vulnerabilities in Version Control Systems

After being informed of the problem, Git issued a fix. However, it seems that there were other version control systems affected by the same vulnerability, which was buried inside the Git clone operation. Bleeping Computer noted that during the attempt to fix Git LFS, Apache Subversion (SVN) and Mercurial were also found to be open to this kind of attack.

Schneeweisz noted that SVN was impacted most because it follows HTTP 301 redirects to svn+ssh:// URLs. “As a result,” he said, “an innocent-looking HTTP URL can be used to trigger a command execution with a 301 redirect.” That means that SVN doesn’t detect HTTP redirects in its repository cloning operations.

However, the issued fix addressed all the problem areas that have been found thus far in all three tools. Users of these systems should be vigilant about patching systems to avoid exploits.

More from

FYSA — VMware Critical Vulnerabilities Patched

< 1 min read - SummaryBroadcom has released a security bulletin, VMSA-2025-0004, addressing and remediating three vulnerabilities that, if exploited, could lead to system compromise. Products affected include vCenter Server, vRealize Operations Manager, and vCloud Director.Threat TopographyThreat Type: Critical VulnerabilitiesIndustry: VirtualizationGeolocation: GlobalOverviewX-Force Incident Command is monitoring activity surrounding Broadcom’s Security Bulletin (VMSA-2025-0004) for three potentially critical vulnerabilities in VMware products. These vulnerabilities, identified as CVE-2025-22224, CVE-2025-22225, and CVE-2025-22226, have reportedly been exploited in attacks. X-Force has not been able to validate those claims. The vulnerabilities…

SoaPy: Stealthy enumeration of Active Directory environments through ADWS

10 min read - Introduction Over time, both targeted and large-scale enumeration of Active Directory (AD) environments have become increasingly detected due to modern defensive solutions. During our internship at X-Force Red this past summer, we noticed FalconForce’s SOAPHound was becoming popular for enumerating Active Directory environments. This tool brought a new perspective to Active Directory enumeration by performing collection via Active Directory Web Services (ADWS) instead of directly through Lightweight Directory Access Protocol (LDAP) as other AD enumeration tools had in the past.…

Smoltalk: RCE in open source agents

26 min read - Big shoutout to Hugging Face and the smolagents team for their cooperation and quick turnaround for a fix! Introduction Recently, I have been working on a side project to automate some pentest reconnaissance with AI agents. Just after I started this project, Hugging Face announced the release of smolagents, a lightweight framework for building AI agents that implements the methodology described in the ReAct paper, emphasizing reasoning through iterative decision-making. Interestingly, smolagents enables agents to reason and act by generating…

Topic updates

Get email updates and stay ahead of the latest threats to the security landscape, thought leadership and research.
Subscribe today