Cloudflare Container Storage Flaw Allowed Cross Tenant Data Exposure

Cloudflare Container Storage Flaw Allowed Cross Tenant Data Exposure

Cloudflare has fixed a security vulnerability in its Containers platform that could allow one customer’s container to access leftover data from another customer’s previously deleted container on the same server. The company and the security researchers who reported the issue confirmed that the exposed information came from unused disk space previously assigned to other containers rather than active workloads. Cloudflare stated that customers did not need to take any action as the issue had been addressed across its service. The vulnerability was reported on September 4 through Cloudflare’s bug bounty program by Oren Yomtov from security firm Accomplish.

Cloudflare Containers enables customers to run applications inside isolated containers on shared servers, with Cloudflare managing the underlying infrastructure and server allocation. Cloudflare Sandboxes, which operates on top of Containers and provides an environment for running untrusted code, including code generated by AI agents, was also affected by the issue. Researchers identified that the problem originated from the way shared disk storage was configured. Each container received storage through Linux thin provisioning technology, which manages storage allocation through fixed size blocks. When a container was removed, its previously allocated storage blocks returned to a shared pool that could later be assigned to another customer’s container. The issue occurred because the storage pool was configured to skip wiping blocks before they were reused. Normally, wiping ensures that data from a previous workload is removed before storage is assigned again. Due to the configuration, when a new container wrote only a small portion of a reused block, remaining sections could still contain data from a previous container. Researchers demonstrated the issue by writing a small four kilobyte block into unused storage space and then reading the entire block at the raw disk level. The remaining portion of the block contained data that had not been overwritten by the new container. During testing, researchers reported finding leftover material in multiple production environments, including 18 out of 24 attempts and across 20 of 22 underlying machines located across four continents.

The recovered storage fragments included directory structures, database pages, and complete SQLite database structures, according to Cloudflare. Researchers reported identifying examples such as directory listings, SQLite databases, Chromium browser profiles, environment configuration files, and credential related files in their analysis. However, they stated that their testing process focused on identifying the presence and structure of recovered data rather than reviewing customer information. Cloudflare also confirmed that the researchers maintained privacy during the investigation and securely deleted any recovered material after reporting the issue. The researchers did not demonstrate that the vulnerability could modify active customer data or disrupt running workloads. Cloudflare addressed the vulnerability through multiple remediation steps. The company first restored block wiping for newly allocated storage blocks, preventing the demonstrated attack method. Researchers confirmed on September 14 that their proof of concept no longer worked after this change. However, Cloudflare identified that already assigned container disks and cached image layers could still contain previously mapped storage information. To address this remaining risk, the company retired active container disks, cleared related caches, and restarted affected infrastructure during scheduled maintenance periods. The cleanup process was completed on September 19 before the company publicly disclosed the issue.

Cloudflare also investigated whether other parties had used the same technique. The company created detection methods based on the researchers’ proof of concept and internal testing activity, then reviewed available disk activity records. Cloudflare reported finding only authorized testing conducted by the researchers and its own security teams, with no evidence that this specific method had been used by other parties. However, the company did not disclose how long the affected configuration had existed or the exact period during which exposure was possible. Researchers separately noted that similar disk configuration concerns were observed in Cloudflare’s Browser Run product, while Cloudflare’s disclosure specifically identified Containers and Sandboxes as affected services.

Source

Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem. 

Post Comment