AI-generated editorial illustration; not a documentary photograph.
Cloudflare says it has fully remediated a vulnerability in Cloudflare Containers and Cloudflare Sandboxes that could let one Workers Paid customer recover residual data from storage blocks previously used by another tenant. In a September 24, 2026 post, the company said it found no evidence that customer data had been compromised or that anyone other than the reporting researchers and Cloudflare engineers used the technique.
How the flaw exposed data
The affected Containers ran on shared multi-tenant infrastructure using Linux device mapper thin provisioning. When a container disk was deleted, its physical 64 KiB blocks returned to a pool serving multiple customer accounts. A pool setting called skip_block_zeroing meant newly allocated blocks were not cleared first.
A researcher from Accomplish showed that a new container could write a 4 KiB portion of selected blocks and then read the rest. Those remaining 60 KiB could contain bytes from a previous container, including filesystem metadata, directory structures, database pages and application data.
What researchers demonstrated
The technique did not allow an attacker to choose a particular victim, workload or host, and it could not read an actively attached disk. Exposure depended on workload placement and whether previously released blocks were reassigned. Cloudflare also said the researchers did not demonstrate modifying another customer’s active data or affecting availability.
Across six production placements, the researchers reported testing 5,614 directory blocks and identifying 2,700 foreign directory inodes through filesystem checksum analysis. They observed residual material on 18 of 24 placements and 20 of 22 underlying nodes across four continents, while reporting only aggregate results to Cloudflare and securely deleting recovered data under their control.
Fleet-wide remediation
Cloudflare removed skip_block_zeroing from the Containers storage pools, restoring allocation behavior that clears new blocks before they are exposed. The researchers independently confirmed that their proof of concept no longer worked after this change.
Because existing disks and cached OCI image layers could still contain mappings created before the fix, Cloudflare also retired running container disks, drained hosts, restarted the virtual machines and cleared cached image snapshots. The company said cleanup across the Containers fleet was completed on September 19, 2026, with no customer-side configuration changes required.
Investigation findings
Cloudflare reviewed retained historical disk-I/O telemetry using signatures derived from the proof of concept and its internal reproduction. It attributed matching activity to the researchers’ authorized testing and Cloudflare’s validation work, and said it found no additional activity consistent with exploitation of this attack path.
Source: How Cloudflare addressed a cross-tenant data exposure vulnerability in Containers
SIGNAL TITANS
Leave a Reply