GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux nvmet-auth: racing AUTH_SEND and AUTH_RECEIVE commands cause a use-after-free write in the target kernel

UnscoredCVE-2026-99441Kernel, userspace & hypervisorcurated

Impact

On a Linux NVMe-oF target with DH-HMAC-CHAP in-band authentication enabled, a REPLY AUTH_SEND command can derive a session key while a concurrent AUTH_RECEIVE on the same queue takes the failure path and frees that key. The REPLY command then writes into freed slab memory. That gives kernel memory corruption on the storage target, and the target may crash. The race sits in the authentication exchange itself, so the host sending the commands does not have to finish authenticating. A disaggregated NVMe-oF target that serves many GPU nodes takes all of their scratch or dataset volumes down with it. The record does not say whether this goes beyond a crash.

Who can reach it

A host that can reach the NVMe-oF target's fabric port (TCP or RDMA) and open an admin or I/O queue, during DH-HMAC-CHAP authentication, so before authentication succeeds. Only targets with in-band authentication configured are exposed.

What to do

Install a kernel that carries the upstream fix (stable commits linked; the record names no distro versions) and reboot the target. That takes its namespaces offline, so schedule it with the GPU nodes that depend on them, or fail over to a multipath peer first. Until then, limit which hosts can reach the target's fabric ports.

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.