The Linux kernel vulnerability (CVE-2026-53154) involves the hugetlb memory management code where two code paths allocate a hugetlb folio and then call copy_user_large_folio(), which can fail due to hardware poisoning or other errors. On failure, the global hugetlb pool count is restored, but the per-VMA reservation map entry remains marked as consumed, causing a reservation leak. This affects the hugetlb_mfill_atomic_pte() resubmission path and the copy_hugetlb_page_range() fork-time copy-on-write path. The leak can cause a subsequent fault at the same address to proceed without reservation, potentially resulting in a SIGBUS signal under hugetlb pool pressure. The patch adds a missing restore_reserve_on_error() call before folio_put() on both error paths to fix the leak.

The vulnerability does not affect confidentiality or integrity but impacts availability by causing a task to receive a SIGBUS signal when accessing previously reserved huge page memory under pool pressure. This can lead to application crashes or instability in workloads using huge pages with userfaultfd or during fork-time copy-on-write operations involving pinned hugetlb anonymous pages.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. The vulnerability description indicates a fix was introduced by adding restore_reserve_on_error() calls on error paths. Users should monitor official Linux kernel advisories and apply updates when available to address this issue.