{
  "affected": [
    {
      "ranges": [
        {
          "events": [
            {
              "introduced": "1b1676180246232308885c4f37fee01cf898fdb2"
            },
            {
              "fixed": "da57b6bce646946c083a48df644d43bb9d011f5f"
            },
            {
              "fixed": "9eaedd598cd25eb0e18958e2456153a4c1f3db03"
            },
            {
              "fixed": "79cb51d538abc459f74e046d27797d36b133ce0e"
            }
          ],
          "repo": "https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git",
          "type": "GIT"
        }
      ]
    },
    {
      "package": {
        "ecosystem": "Linux",
        "name": "Kernel"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "6.10.0"
            },
            {
              "fixed": "6.18.64"
            }
          ],
          "type": "ECOSYSTEM"
        },
        {
          "events": [
            {
              "introduced": "6.19.0"
            },
            {
              "fixed": "7.3.5"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "database_specific": {
    "cna_assigner": "Linux",
    "osv_generated_from": "https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/99xxx/CVE-2026-99664.json"
  },
  "details": "In the Linux kernel, the following vulnerability has been resolved:\n\nmm/gup: recheck the PUD under PTL in follow_pud_mask()\n\nfollow_pud_mask() reads the PUD without a lock. If it is a leaf, it takes\nthe lock and calls follow_huge_pud(), which reads the entry again, but\nonly tests whether it is present.\n\nA leaf can be zapped concurrently and a page fault can fill the empty\nentry with a page table.\n\nThis currently only applies when device-dax makes use of PUD entries,\nand the fault after e.g., MADV_DONTNEED would find THPs to be disabled.\n\nThat requires a sequence like\n\n1) Fault in the PUD\n2) Disable THP\n3) Have GUP race against concurrent MADV_DONTNEED + refault\n\nWhich is rather unlikely to happen in practice, which is why we never\nreceived a report about such a problem.\n\nBefore commit 1b1676180246 (\"mm/gup: handle huge pud for\nfollow_pud_mask()\"), we performed that check in follow_devmap_pud() by\ntesting for pud_devmap().\n\nLet's fix it by just structuring follow_pud_mask() similar to\nfollow_pmd_mask(), minus protnone stuff that does not apply and minus\nFOLL_SPLIT_PMD handling. That makes it much easier to spot any such\ninconsistencies.\n\nPerform the pud_bad() check early in follow_pmd_mask() and just forward\nthe PUD value. We can drop the pud_present() check in follow_huge_pud(),\njust like we have in follow_huge_pmd().\n\nEven though we are performing more changes than strictly necessary, the\nfix should be easy to backport.",
  "id": "CVE-2026-99664",
  "modified": "2026-11-20T06:45:48.648186966Z",
  "published": "2026-11-20T05:50:14.863Z",
  "references": [
    {
      "type": "PACKAGE",
      "url": "https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git"
    },
    {
      "type": "WEB",
      "url": "https://git.kernel.org/stable/c/79cb51d538abc459f74e046d27797d36b133ce0e"
    },
    {
      "type": "WEB",
      "url": "https://git.kernel.org/stable/c/9eaedd598cd25eb0e18958e2456153a4c1f3db03"
    },
    {
      "type": "WEB",
      "url": "https://git.kernel.org/stable/c/da57b6bce646946c083a48df644d43bb9d011f5f"
    },
    {
      "type": "ADVISORY",
      "url": "https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/99xxx/CVE-2026-99664.json"
    },
    {
      "type": "ADVISORY",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-99664"
    }
  ],
  "schema_version": "1.9.2",
  "summary": "mm/gup: recheck the PUD under PTL in follow_pud_mask()"
}