Cloudflare fixes cross-tenant data exposure in Containers via dm-thin patch

Cloudflare has fully remediated a cross-tenant data exposure vulnerability in Cloudflare Containers and Cloudflare Sandboxes, reported by security researcher Oren Yomtov of Accomplish on September 4, 2026. The issue stemmed from a misconfiguration in the underlying storage layer that allowed customers to recover residual data from disk blocks previously used by other tenants on the same host.

The Vulnerability Mechanism

Cloudflare Containers use Linux device mapper thin provisioning (dm-thin) to manage writable root disks within Firecracker VMs. The affected storage pools were configured with the skip_block_zeroing option, which prevents dm-thin from clearing newly allocated blocks before exposing them to a container.

When a container wrote a small amount of data (e.g., a 4 KiB block) into an unmapped region, dm-thin allocated a larger physical block (64 KiB) from a shared pool. Because zeroing was disabled, the remaining 60 KiB of the block retained data from its previous owner. A subsequent raw read of the disk could therefore reveal bytes written by other customers.

Validation and Impact

Researchers demonstrated that this technique could recover filesystem metadata, directory structures, database pages, and even structurally complete SQLite databases. Across six production placements, they identified 2,700 distinct foreign directory inodes and observed residual material on 18 of 24 placements and 20 of 22 underlying nodes across four continents.

Key limitations of the exploit included:

• Attackers could not target specific victims, workloads, or hosts.
• Exposure depended on Cloudflare’s automatic workload placement and block reassignment.
• The researchers did not demonstrate modification of active data or impact to workload availability.

Remediation and Investigation

Cloudflare applied a two-step fix across the Containers fleet:

1. Configuration Change: Removed skip_block_zeroing from dm-thin pool configurations to restore default zeroing behavior for new allocations.
2. Cache Cleanup: Retired all running container disks and removed cached image snapshots created before the mitigation to ensure existing mappings were cleared.

The company reviewed historical disk-I/O telemetry and found no evidence of malicious exploitation. All activity matching the exploit signature was attributed to the researchers and Cloudflare engineers during authorized validation. The remediation required no customer-side configuration changes.