GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel mm/truncate: stale folio pointer during partial truncation splits the wrong folio

UnscoredCVE-2026-99462Kernel, userspace & hypervisorcurated

Impact

In truncate_inode_partial_folio(), the kernel keeps a page pointer to the folio that straddles the end of the truncation range after the first split has already unlocked it. In that window the page can be reclaimed and reused as a new large folio in the same file, so the kernel splits a folio at the wrong position, or a use-after-free window opens. The result is page-cache corruption or a host kernel crash, and on a shared GPU node that takes down every job on it. The record describes no reproducer and no privilege requirement, and the race window is narrow.

Who can reach it

Local. Needs a process that can truncate or hole-punch files on a filesystem that uses large page-cache folios, while reclaim runs at the same time. The record does not say whether an unprivileged user can trigger it reliably.

What to do

Install a kernel that contains the stable fix (commits referenced below), then drain and reboot each node. The record lists no fixed version numbers and no workaround.

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.