GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel ixgbe: XPS setup on XDP Tx queues writes out of bounds on high-core-count hosts

CVSS 7.8CVE-2026-74317Kernel, userspace & hypervisorcurated

Impact

ixgbe called netif_set_xps_queue() for XDP Tx queues, which are not exposed to the netdev. On a host with 64 or more CPUs using an E610 adapter, MSI-X assignment caps the netdev at 63 queue pairs while XDP allocates 64, so configuring an XDP program drives a WARNING and then a KASAN-reported slab-out-of-bounds write in __netif_set_xps_queue(). That is kernel heap corruption triggered by a normal operational action - loading an XDP program on a dense node, which is exactly the shape of GPU and storage hosts. The practical outcome is a node that can panic or misbehave at the moment an operator or a CNI loads XDP, forcing an unplanned drain.

Who can reach it

Local and privileged: loading an XDP program requires CAP_NET_ADMIN/CAP_BPF (root, or whatever daemon manages the dataplane). No remote or unauthenticated path. Only hosts with an affected Intel adapter driven by ixgbe, 64 or more CPUs, and XDP in use are exposed.

What to do

Install a stable kernel carrying the fix that skips XPS configuration for XDP Tx queues (backported across several stable branches; check the linked commits for your branch) and reboot the node after draining it. Until then, not loading XDP programs on affected ixgbe interfaces avoids the trigger entirely, which is a realistic mitigation if XDP is not part of your dataplane.

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.