GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel AF_ALG: af_alg_pull_tsgl can reassign one page past the end of the scatterlist

CVSS 7.8CVE-2026-43078Kernel, userspace & hypervisorcurated

Impact

When page reassignment was added to af_alg_pull_tsgl(), the surrounding loop bound was not updated, so the function can try to reassign one page more than the list holds - an off-by-one over kernel scatterlist state. This sits directly behind the AF_ALG socket interface, which unprivileged processes can open by default, so on a multi-tenant GPU node any tenant pod reaches it without extra capabilities. NVD rates it 7.8 local with high impact across the board, i.e. a path from tenant code to kernel memory corruption and host root on a node whose other GPUs belong to someone else. Affected branches go back to 4.14, so long-lived nodes are in scope, not just recent kernels.

Who can reach it

Local, any authenticated user or container process that can create an AF_ALG (AF_ALG/algif_*) socket. No special device access, no GPU access, no network reachability required.

What to do

Update to a stable kernel containing the fix (backports exist on five branches per the record) and reboot each node; kernel crypto code cannot be hot-patched or reloaded here. Where tenants have no need for the kernel crypto socket interface, denying socket(AF_ALG, ...) via seccomp in the pod runtime and keeping algif_* modules unloaded removes the exposure until the reboot window.

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.