GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel nfsd: race between cpntf publish and OFFLOAD_CANCEL oopses on an uninitialised list head

CVSS 7.5CVE-2026-89684Kernel, userspace & hypervisorcurated

Impact

The new copy-notify entry is published into the s2s_cp_stateids IDR before it is linked onto sc_cp_list, leaving a window where the entry is reachable by so_id with a zeroed list_head. Because so_id is echoed to the client as cnr_stateid, any NFSv4.2 client can drive an OFFLOAD_CANCEL into that window and make the server list_del a {NULL,NULL} node, crashing it. Server-to-server COPY is what dataset staging and checkpoint moves use, so on a busy fleet the race is hit by ordinary traffic, and the blast radius is every node with the export mounted.

Who can reach it

Any NFSv4.2 client that can perform a server-side COPY and then send OFFLOAD_CANCEL for the returned stateid. Requires whatever authentication your export demands to issue the COPY; no special privilege beyond that.

What to do

Update the host kernel to a stable release that folds the cs_type assignment and list_add into the idr_alloc_cyclic critical section, then reboot the NFS server. As an interim measure, disabling server-to-server COPY on the export removes the code path; there is no live patch described.

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.