GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel qla2xxx: double completion in async IOCB timeout frees a live stack frame

CVSS 9.8CVE-2026-89847Kernel, userspace & hypervisorcurated

Impact

When qla24xx_async_abort_cmd() fails, the timeout handler called sp->done() without checking whether it had actually removed the SRB from outstanding_cmds[]. If the response ISR completed the same handle first, the submitter in qla24xx_control_vp() has already returned and reclaimed its on-stack completion; the second sp->done() then calls complete() into that freed frame - a use-after-free in kernel stack memory. Reaching it needs Fibre Channel command timeouts racing late completions, which is a fabric-fault or flaky-target condition, not something a tenant steers. Nodes with QLogic FC HBAs attached to shared storage are the ones to look at; everyone else does not run this code.

Who can reach it

Local and hardware-dependent: requires a QLogic FC HBA and an async IOCB timing out while the response queue completes the same handle. No tenant-reachable path is described in the record.

What to do

Update to a kernel carrying the sp_found guard on this path and reboot the affected storage-attached nodes. Headless GPU nodes without qla2xxx-driven FC HBAs are unaffected and need no action. The record lists 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.