In 2020, we built-in kCTF into Google’s Vulnerability Rewards Program (VRP) to assist researchers evaluating the safety of Google Kubernetes Engine (GKE) and the underlying Linux kernel. Because the Linux kernel is a key part not only for Google, however for the Web, we began closely investing on this space. We prolonged the VRP’s scope and most reward in 2021 (to $50k), then once more in February 2022 (to $91k), and at last in August 2022 (to $133k). In 2022, we additionally summarized our learnings thus far in our cookbook, and launched our experimental mitigations for the most typical exploitation strategies.

On this publish, we would wish to share our learnings and statistics in regards to the newest Linux kernel exploit submissions, how efficient our mitigations are in opposition to them, what we do to guard our customers, and, lastly, how we’re altering our program to align incentives to the areas we’re most curious about.

Learnings and Statistics

Since its inception, this system has rewarded researchers with a complete of 1.8 million USD, and previously yr, there was a transparent pattern: 60% of the submissions exploited the io_uring part of the Linux kernel (we paid out round 1 million USD for io_uring alone). Moreover, io_uring vulnerabilities have been utilized in all of the submissions which bypassed our mitigations.

Limiting io_uring

To guard our customers, we determined to restrict the utilization of io_uring in Google merchandise: 

Whereas io_uring brings efficiency advantages, and promptly reacts to safety points with complete safety fixes (like backporting the 5.15 model to the 5.10 secure tree), it’s a pretty new a part of the kernel. As such, io_uring continues to be actively developed, however it’s nonetheless affected by extreme vulnerabilities and in addition gives sturdy exploitation primitives. For these causes, we at present think about it secure just for use by trusted parts.

Transparency

At present, we make vulnerability particulars public on our spreadsheet (which now additionally contains CVE particulars), and we have now summarized completely different exploitation strategies in our cookbook. Sooner or later, to make our efforts extra clear and provides quicker suggestions to the group, we are going to ask researchers to open-source their submissions, together with the code they used.

Introducing kernelCTF

To higher align incentives with our areas of curiosity, we’re shifting our focus from GKE and kCTF to the most recent secure kernel and our mitigations. In consequence, beginning right this moment we are going to deal with kernel exploit submissions underneath a brand new identify, “kernelCTF,” with its personal reward construction and submission course of. The utmost whole payout for kernelCTF continues to be $133,337 per submission. Whereas the precise GKE kernel configuration continues to be lined by the brand new kernelCTF, exploits affecting non-kernel parts like the complete GKE stack (together with Kubernetes), the container runtime, and GKE itself, are actually individually eligible for vulnerability rewards underneath the kCTF VRP which is returning to its unique reward quantities and situations.

Conclusion

Our objective stays the identical: we’re constructing a pipeline to research, experiment, measure, and construct safety mitigations to make the Linux kernel as secure as potential, with the assistance of the safety group. We hope that over time, we will implement safety mitigations that make it tougher to take advantage of Linux kernel vulnerabilities.

With the identify change, we have now moved our communication channel to #kernelctf on Discord, with a separate #kernelctf-announcements channel. Please be a part of us there for the most recent updates relating to kernelCTF.