{"cve":"CVE-2023-52934","nvd":{"published":"2025-03-27T17:15:43.207","modified":"2026-08-04T10:18:50.070","status":"Modified","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nmm/MADV_COLLAPSE: catch !none !huge !bad pmd lookups\n\nIn commit 34488399fa08 (\"mm/madvise: add file and shmem support to\nMADV_COLLAPSE\") we make the following change to find_pmd_or_thp_or_none():\n\n\t-       if (!pmd_present(pmde))\n\t-               return SCAN_PMD_NULL;\n\t+       if (pmd_none(pmde))\n\t+               return SCAN_PMD_NONE;\n\nThis was for-use by MADV_COLLAPSE file/shmem codepaths, where\nMADV_COLLAPSE might identify a pte-mapped hugepage, only to have\nkhugepaged race-in, free the pte table, and clear the pmd.  Such codepaths\ninclude:\n\nA) If we find a suitably-aligned compound page of order HPAGE_PMD_ORDER\n   already in the pagecache.\nB) In retract_page_tables(), if we fail to grab mmap_lock for the target\n   mm/address.\n\nIn these cases, collapse_pte_mapped_thp() really does expect a none (not\njust !present) pmd, and we want to suitably identify that case separate\nfrom the case where no pmd is found, or it's a bad-pmd (of course, many\nthings could happen once we drop mmap_lock, and the pmd could plausibly\nundergo multiple transitions due to intervening fault, split, etc). \nRegardless, the code is prepared install a huge-pmd only when the existing\npmd entry is either a genuine pte-table-mapping-pmd, or the none-pmd.\n\nHowever, the commit introduces a logical hole; namely, that we've allowed\n!none- && !huge- && !bad-pmds to be classified as genuine\npte-table-mapping-pmds.  One such example that could leak through are swap\nentries.  The pmd values aren't checked again before use in\npte_offset_map_lock(), which is expecting nothing less than a genuine\npte-table-mapping-pmd.\n\nWe want to put back the !pmd_present() check (below the pmd_none() check),\nbut need to be careful to deal with subtleties in pmd transitions and\ntreatments by various arch.\n\nThe issue is that __split_huge_pmd_locked() temporarily clears the present\nbit (or otherwise marks the entry as invalid), but pmd_present() and\npmd_trans_huge() still need to return true while the pmd is in this\ntransitory state.  For example, x86's pmd_present() also checks the\n_PAGE_PSE , riscv's version also checks the _PAGE_LEAF bit, and arm64 also\nchecks a PMD_PRESENT_INVALID bit.\n\nCovering all 4 cases for x86 (all checks done on the same pmd value):\n\n1) pmd_present() && pmd_trans_huge()\n   All we actually know here is that the PSE bit is set. Either:\n   a) We aren't racing with __split_huge_page(), and PRESENT or PROTNONE\n      is set.\n      => huge-pmd\n   b) We are currently racing with __split_huge_page().  The danger here\n      is that we proceed as-if we have a huge-pmd, but really we are\n      looking at a pte-mapping-pmd.  So, what is the risk of this\n      danger?\n\n      The only relevant path is:\n\n\tmadvise_collapse() -> collapse_pte_mapped_thp()\n\n      Where we might just incorrectly report back \"success\", when really\n      the memory isn't pmd-backed.  This is fine, since split could\n      happen immediately after (actually) successful madvise_collapse().\n      So, it should be safe to just assume huge-pmd here.\n\n2) pmd_present() && !pmd_trans_huge()\n   Either:\n   a) PSE not set and either PRESENT or PROTNONE is.\n      => pte-table-mapping pmd (or PROT_NONE)\n   b) devmap.  This routine can be called immediately after\n      unlocking/locking mmap_lock -- or called with no locks held (see\n      khugepaged_scan_mm_slot()), so previous VMA checks have since been\n      invalidated.\n\n3) !pmd_present() && pmd_trans_huge()\n  Not possible.\n\n4) !pmd_present() && !pmd_trans_huge()\n  Neither PRESENT nor PROTNONE set\n  => not present\n\nI've checked all archs that implement pmd_trans_huge() (arm64, riscv,\npowerpc, longarch, x86, mips, s390) and this logic roughly translates\n(though devmap treatment is unique to x86 and powerpc, and (3) doesn't\nnecessarily hold in general -- but that doesn't matter since\n!pmd_present() always takes failure path).\n\nAlso, add a comment above find_pmd_or_thp_or_none()\n---truncated---","cvss_v4":null,"cvss_v3":{"version":"3.1","vector":"CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:U/C:N/I:N/A:H","score":4.7},"cwes":["CWE-362"],"references":["https://git.kernel.org/stable/c/96aaaf8666010a39430cecf8a65c7ce2908a030f","https://git.kernel.org/stable/c/edb5d0cf5525357652aff6eacd9850b8ced07143"],"products":["linux:linux_kernel"],"source":"nvd","synced_at":"2026-10-19T22:00:28.034033815+00:00"},"epss":{"cve":"CVE-2023-52934","score":0.00122,"percentile":0.01974,"date":"2026-10-31"},"kev":null,"osv":[],"attribution":["This product uses the NVD API but is not endorsed or certified by the NVD.","EPSS scores: FIRST Exploit Prediction Scoring System, https://www.first.org/epss","Known exploited vulnerabilities: CISA Known Exploited Vulnerabilities Catalog, https://www.cisa.gov/known-exploited-vulnerabilities-catalog","Package advisories: OSV (https://osv.dev), from the GitHub Advisory Database, PyPA, Go, RustSec and other ecosystem databases, under CC-BY 4.0 or CC0","CVE records: the CVE Program's CVE List (https://www.cve.org), under its terms of use (https://www.cve.org/Legal/TermsOfUse). CVE is a registered trademark of The MITRE Corporation; this product is not affiliated with the CVE Program."]}