GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel mm/pagewalk: stale ACTION_AGAIN causes duplicate walk callbacks and an out-of-bounds write via mincore()

CVSS 7.8CVE-2026-89754Kernel, userspace & hypervisorcurated

Impact

When a ->pmd_entry() callback sets ACTION_AGAIN and the PMD is cleared on retry, walk->action is never reset, so the PUD-level retry added by an earlier fix re-invokes walk callbacks for a range already processed. For non-idempotent callers this is a heap out-of-bounds write; the report shows KASAN catching a slab-out-of-bounds write from __mincore_unmapped_range() reached through the ordinary mincore() syscall, triggered by an unprivileged local fuzzer. This is core memory management, not an optional subsystem, so it applies to every node in a fleet, and on a multi-tenant GPU host any tenant process with shell access can reach mincore(). Kernel heap corruption from an unprivileged syscall is a credible local privilege escalation path.

Who can reach it

Local, unprivileged. Any user process on the node that can call mincore() on a suitable mapping; no special capability or device access is required.

What to do

Move to a stable kernel carrying commit 895cd4ecbb2e (or the backport for your series) and reboot every node. There is no runtime toggle - mincore() is not something you can disable - so this is a full fleet reboot campaign, staged with drains.

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.