GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel CephFS: folios skipped for writeback are never put, leaking page references

CVSS 7.5CVE-2026-52960Kernel, userspace & hypervisorcurated

Impact

CephFS writeback collects folios with filemap_get_folios(), which takes a reference on each, but folios removed from the batch because they are not suitable for writeback were dropped without folio_put(). Each occurrence pins a page permanently. On a node that writes heavily to CephFS - checkpoint and log traffic from training jobs - free memory erodes over time until allocations fail, and the only recovery is a reboot. The record is a one-paragraph fix tested on v6.18 and makes no claim beyond the leaked references; it is a resource-exhaustion defect, not memory corruption.

Who can reach it

Local workload on a node with CephFS mounted in-kernel; any tenant that can write files to the mount drives writeback. No special privilege is needed beyond write access to the filesystem.

What to do

Move to a stable kernel containing the folio_put() fix (three stable commits are linked; the record names no fixed version). That is a kernel change: drain the node and reboot. There is no runtime mitigation other than reducing CephFS write pressure and rebooting nodes whose free memory has already drifted.

References

Related entries

All Kernel, userspace & hypervisor entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.