GPU VulnDB

Database/NVIDIA / GPU stack

Linux kernel drm/xe: imported dma-buf BO can outlive the exporter, giving a use-after-free on the shared resv

CVSS 7.8CVE-2026-68266NVIDIA / GPU stackcurated

Impact

An imported dma-buf buffer object borrows the exporter's reservation object but only takes a dma-buf reference after dma_buf_dynamic_attach() succeeds. If the attach fails, or ttm_bo_init_reserved() fails during creation, the BO still points at the exporter's resv and is torn down asynchronously by ttm_bo_delayed_delete(), which locks that resv possibly after the exporter has been freed - a general protection fault in mutex_can_spin_on_owner() was observed. Buffer sharing across devices and processes is normal on GPU nodes doing peer-to-peer or video/compute interop, so this is reachable by ordinary workloads, and the failure takes down the host.

Who can reach it

Local user able to import a dma-buf into the Xe driver, i.e. any tenant with a GPU pod that uses buffer sharing. Requires hitting an attach or BO-creation failure, which memory pressure makes achievable.

What to do

Apply the stable fix that holds the dma-buf reference for the whole BO lifetime and reboot the node. No configuration mitigation short of blocking dma-buf import, which breaks normal workloads.

References

Related entries

All NVIDIA / GPU stack 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.