GPU VulnDB

Database/Control plane, storage & DevOps

Linux nfsd: LAYOUTGET racing a delegation revoke hits a BUG_ON and panics the NFS server

CVSS 9.8CVE-2026-89686Control plane, storage & DevOpscurated

Impact

nfsd4_alloc_layout_stateid() read fp->fi_deleg_file without fi_lock on the delegation path. The laundromat can clear fi_deleg_file concurrently when it revokes an unreturned delegation, nfsd_file_get() then returns NULL and the BUG_ON fires, panicking the server. The commit states the race is client-reachable: one client holds a delegation and ignores the recall while another opens the same file to force it, and a LAYOUTGET using the delegation stateid lands in the window. A panic on the NFS server takes down shared storage for every node that mounts it - on a GPU cluster that means every running job stalls, not one tenant.

Who can reach it

Two NFSv4.1 clients with ordinary access to the same export, one holding a delegation it does not return. No server-side privilege required.

What to do

Patch the NFS server kernel with one of the linked stable commits and reboot the server; the fix takes fi_lock around the read and replaces the BUG_ON with a clean error return. There is no configuration workaround short of disabling delegations or pNFS layouts on the export.

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.