GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel qla2xxx: use-after-free when a late response interrupt reaches an already-freed request queue

CVSS 8.1CVE-2026-89848Kernel, userspace & hypervisorcurated

Impact

qla2xxx_delete_qpair() frees the request queue before releasing the response queue's MSI-X vector. In that window a response interrupt can still fire, queue qpair->q_work, and have qla24xx_process_response_queue() dereference the freed rsp->req on the LOGINOUT/CT/ELS and status paths. The cancel_work_sync() that was meant to cover this runs too late, in the response free path. The result is kernel memory corruption during Fibre Channel queue-pair teardown - which happens on adapter reset, port disruption or driver unload, not only on operator action. On a storage-attached GPU node this is an unplanned crash of a node whose in-flight training state is expensive to lose.

Who can reach it

Local and indirect - no tenant-reachable interface. It is a race triggered by FC adapter or fabric events (qpair teardown, adapter reset, driver unload) on hosts using QLogic FC HBAs; no authentication step applies.

What to do

Apply a stable kernel that releases the response-queue interrupt and flushes qpair->q_work before deleting the request queue. Kernel-image fix: drain and reboot each node with a QLogic FC HBA. Hosts without qla2xxx loaded need no action.

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.