CVE-2026-31466

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> mm/huge_memory: fix folio isn&amp;#39;t locked in softleaf_to_folio()<br /> <br /> On arm64 server, we found folio that get from migration entry isn&amp;#39;t locked<br /> in softleaf_to_folio(). This issue triggers when mTHP splitting and<br /> zap_nonpresent_ptes() races, and the root cause is lack of memory barrier<br /> in softleaf_to_folio(). The race is as follows:<br /> <br /> CPU0 CPU1<br /> <br /> deferred_split_scan() zap_nonpresent_ptes()<br /> lock folio<br /> split_folio()<br /> unmap_folio()<br /> change ptes to migration entries<br /> __split_folio_to_order() softleaf_to_folio()<br /> set flags(including PG_locked) for tail pages folio = pfn_folio(softleaf_to_pfn(entry))<br /> smp_wmb() VM_WARN_ON_ONCE(!folio_test_locked(folio))<br /> prep_compound_page() for tail pages<br /> <br /> In __split_folio_to_order(), smp_wmb() guarantees page flags of tail pages<br /> are visible before the tail page becomes non-compound. smp_wmb() should<br /> be paired with smp_rmb() in softleaf_to_folio(), which is missed. As a<br /> result, if zap_nonpresent_ptes() accesses migration entry that stores tail<br /> pfn, softleaf_to_folio() may see the updated compound_head of tail page<br /> before page-&gt;flags.<br /> <br /> This issue will trigger VM_WARN_ON_ONCE() in pfn_swap_entry_folio()<br /> because of the race between folio split and zap_nonpresent_ptes()<br /> leading to a folio incorrectly undergoing modification without a folio<br /> lock being held.<br /> <br /> This is a BUG_ON() before commit 93976a20345b ("mm: eliminate further<br /> swapops predicates"), which in merged in v6.19-rc1.<br /> <br /> To fix it, add missing smp_rmb() if the softleaf entry is migration entry<br /> in softleaf_to_folio() and softleaf_to_page().<br /> <br /> [tujinjiang@huawei.com: update function name and comments]