CVE-2026-74632 addresses a race condition in the Linux kernel's huge zero page management code. The vulnerability arises from a timing window where the shrinker can decrement the huge_zero_refcount to zero and overwrite huge_zero_pfn with an invalid sentinel value (~0UL) while a concurrent get_huge_zero_folio() call writes a valid huge_zero_pfn. This leads to is_huge_zero_pfn() and is_huge_zero_pmd() functions misidentifying the huge zero folio as a regular transparent huge page folio, causing incorrect behavior such as splitting the huge zero folio. The patch introduces a spinlock (huge_zero_lock) to serialize writes to huge_zero_pfn, huge_zero_folio, and huge_zero_refcount, ensuring atomicity and correct memory ordering. The fix carefully preserves the fast path atomic logic and is designed to be backportable.

The vulnerability can cause the Linux kernel to misidentify huge zero folios, leading to incorrect handling such as splitting these folios. This could affect memory management correctness and stability in systems using transparent huge pages. There is no indication of direct security impact such as privilege escalation or information disclosure from the provided data.

A patch fixing this race condition has been released and should be applied. The fix introduces a spinlock to serialize critical updates in the huge zero page code. No alternative mitigations are described. Patch status is not explicitly confirmed in the input data, so users should consult the official Linux kernel advisories or vendor patches for the current remediation status.