GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel nvmet-auth: use-after-free when auth timeout work races SQ teardown

CVSS 9.8CVE-2026-89970Kernel, userspace & hypervisorcurated

Impact

nvmet_auth_sq_free() cancelled the authentication-expiry work without waiting for an already-running callback, so transport teardown could free or reuse the queue while nvmet_auth_expired_work() was still touching that struct nvmet_sq. A remote initiator that starts in-band authentication and then drops the connection at the right moment can drive the race, giving kernel memory corruption on the storage target. On a GPU fleet the exposure is the NVMe-oF target nodes serving dataset and checkpoint volumes; a crash there stalls every job with a namespace mounted from that node.

Who can reach it

Any initiator that can reach the NVMe-oF target and begin in-band authentication, then tear the queue down. The race is timing-dependent, not a deterministic exploit.

What to do

Update to a kernel containing the fix (teardown now synchronously drains the delayed work) and reboot the affected target nodes. No standalone mitigation short of disabling in-band authentication or restricting which initiators can connect. The record gives stable commits only, no distribution-fixed version.

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.