GPU VulnDB

Database/Control plane, storage & DevOps

Linux CephFS client: unbounded delegated-inode intervals let a hostile MDS burn client CPU and memory

UnscoredCVE-2026-89648Control plane, storage & DevOpscurated

Impact

ceph_parse_deleg_inos() decodes attacker-controlled 64-bit interval lengths from an MDS create-with-delegation reply and loops over them inserting entries into the session xarray, with no bound on a single interval, on the number of intervals per reply, or on the accumulated count across replies. A malicious or compromised MDS can pin a client CPU in the insert loop or grow the xarray until the node runs out of memory. On a GPU node this is a resource-exhaustion denial of service driven entirely by the storage control plane: the node stops making progress on training work and typically has to be rebooted to recover. The fix caps the per-session population and the aggregate interval length accepted from one reply.

Who can reach it

A malicious or compromised Ceph MDS serving a client that performs async file creates on a CephFS mount. No local user action beyond ordinary file creation, and no client-side authentication bypass needed.

What to do

Patch the host kernel to a stable release carrying the CEPH_MAX_DELEG_INOS cap (commits below) and drain and reboot each node mounting CephFS. As an interim measure, ensure clients only mount from trusted MDS daemons. The record names no vendor fixed-version string beyond the stable commits.

References

Related entries

All Control plane, storage & DevOps 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.