GPU VulnDB

Database/Control plane, storage & DevOps

Linux nfsd: pNFS layout fence worker takes a duplicate reference and leaks the layout stateid

CVSS 9.8CVE-2026-89681Control plane, storage & DevOpscurated

Impact

The workqueue core clears WORK_STRUCT_PENDING before running the callback, so delayed_work_pending() in lm_breaker_timedout() can report no work pending while the fence worker is already running. The breaker then takes a second sc_count reference and queues a worker that coalesces with the in-flight one, and the extra reference is never dropped - the layout stateid leaks. On a pNFS-serving metadata server this is an unbounded kernel memory leak driven by clients that hold layouts and do not return them, degrading a storage node that a whole GPU cluster mounts. No memory corruption is described; the impact is resource exhaustion over time.

Who can reach it

An NFSv4.1 pNFS client holding layouts that let the server's layout breaker time out. Ordinary client access to the export; no server-side privilege.

What to do

Roll a kernel with one of the linked stable commits onto the NFS server and reboot. Only servers exporting pNFS layouts are affected; a server that does not hand out layouts has nothing to fence. As an interim measure, monitor slab growth on the server and restart it during a maintenance window before it exhausts memory.

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.