{"id":"CVE-2026-99472","state":"PUBLISHED","title":"KVM: pfncache: use a dedicated invalidation sequence for cache refresh","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nKVM: pfncache: use a dedicated invalidation sequence for cache refresh\n\nThe gfn_to_pfn_cache refresh path guards against mmu notifier\ninvalidations which complete while it has dropped gpc->lock for the\nHVA->PFN lookup: hva_to_pfn_retry() samples kvm->mmu_invalidate_seq\nand retries if it changed, or if mn_active_invalidate_count is still\nelevated.\n\nThat is insufficient for HVA-based caches. mmu_invalidate_seq is only\nadvanced by kvm_mmu_invalidate_end() when the invalidated range\noverlaps a memslot, and an HVA-based cache (e.g. the Xen shared_info\npage mapped with KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA) need not be backed\nby any memslot at all. An invalidation of the cached HVA which starts\nand ends entirely within the lookup window is thus invisible to the\nretry check: mn_active_invalidate_count is back to zero and the\nsequence never moved. The refresh then publishes a mapping of a page\nwhich has already been freed, and the next reader dereferences it:\n\n  BUG: KASAN: use-after-free in kvm_xen_shared_info_init+0x3c6/0x440\n  Read of size 4 at addr ffff8880599c2900 by task syz.2.383/7257\n\nSince gfn_to_pfn_cache_invalidate_start() deliberately skips caches\nwhich are not currently valid (including one whose refresh is in\nprogress, as the refresh clears the valid flag before dropping the\nlock), the retry check is the only line of defence, and it must fire\nfor *any* invalidation, not just those hitting a memslot.\n\nAdd a dedicated kvm->gpc_invalidate_seq, incremented by every\nkvm_mmu_notifier_invalidate_range_end() under mn_invalidate_lock\nbefore mn_active_invalidate_count is decremented, and check it in\nhva_to_pfn_retry() instead of mmu_invalidate_seq. Incrementing in\nrange_end() in the same critical section as the in-progress count\nalso closes the variant where the cache is activated with the\ncontested HVA only after invalidate_range_start() has run.\n\nThe same bug is also reachable through the per-vCPU vcpu_info cache\n(KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO_HVA), where the stale mapping is\nthen dereferenced by kvm_setup_guest_pvclock() on the next KVM_RUN:\n\n  BUG: KASAN: use-after-free in kvm_setup_guest_pvclock+0x5bf/0x660\n\nThis intentionally makes refresh retry on *unrelated* mmu notifier\nevents; restoring precision (and reworking the GPC locking more\ngenerally) is left for a subsequent series.\n\nReproducers: https://david.woodhou.se/xen_shinfo_race.c\n             https://david.woodhou.se/vcpu_info_race.c","publishedAt":"2026-11-19T14:42:59.997Z","updatedAt":"2026-11-22T11:08:29.877Z","severity":{"score":7.8,"label":"High","version":"3.1","vector":"CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H","source":"Linux"},"assigner":"Linux","products":["Linux Linux"],"affected":[{"vendor":"Linux","product":"Linux","defaultStatus":"unaffected","versions":[{"version":"b9220d32799a62c634f7b0b3618da52772626952","status":"affected","lessThan":"fabc002f6396202b5ad7ad4ceec24d27aad90f3f","versionType":"git"},{"version":"b9220d32799a62c634f7b0b3618da52772626952","status":"affected","lessThan":"da260bcb56feeb6489a83f310308453a0a7de25e","versionType":"git"}],"repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["include/linux/kvm_host.h","virt/kvm/kvm_main.c","virt/kvm/pfncache.c"]},{"vendor":"Linux","product":"Linux","defaultStatus":"affected","versions":[{"version":"6.9","status":"affected"},{"version":"0","status":"unaffected","lessThan":"6.9","versionType":"semver"},{"version":"7.3.2","status":"unaffected","lessThanOrEqual":"7.3.*","versionType":"semver"},{"version":"7.4-rc1","status":"unaffected","lessThanOrEqual":"*","versionType":"original_commit_for_fix"}],"repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["include/linux/kvm_host.h","virt/kvm/kvm_main.c","virt/kvm/pfncache.c"]}],"references":["https://git.kernel.org/stable/c/fabc002f6396202b5ad7ad4ceec24d27aad90f3f","https://git.kernel.org/stable/c/da260bcb56feeb6489a83f310308453a0a7de25e"],"weaknesses":[],"ingestedAt":"2026-11-22T11:18:06.032Z","source":{"repository":"CVEProject/cvelistV5","path":"cves/2026/99xxx/CVE-2026-99472.json","commit":"247cc9beefd85903834df0592e88a44f4d599223","baselineCommit":"556c9d3a8cd17e218bb8e452ba7cd619a384ff47"},"raw":{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2026-99472","state":"PUBLISHED","datePublished":"2026-11-19T14:42:59.997Z","dateUpdated":"2026-11-22T11:08:29.877Z","assignerShortName":"Linux","assignerOrgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","dateReserved":"2026-09-25T10:25:14.417Z"},"containers":{"cna":{"title":"KVM: pfncache: use a dedicated invalidation sequence for cache refresh","descriptions":[{"lang":"en","value":"In the Linux kernel, the following vulnerability has been resolved:\n\nKVM: pfncache: use a dedicated invalidation sequence for cache refresh\n\nThe gfn_to_pfn_cache refresh path guards against mmu notifier\ninvalidations which complete while it has dropped gpc->lock for the\nHVA->PFN lookup: hva_to_pfn_retry() samples kvm->mmu_invalidate_seq\nand retries if it changed, or if mn_active_invalidate_count is still\nelevated.\n\nThat is insufficient for HVA-based caches. mmu_invalidate_seq is only\nadvanced by kvm_mmu_invalidate_end() when the invalidated range\noverlaps a memslot, and an HVA-based cache (e.g. the Xen shared_info\npage mapped with KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA) need not be backed\nby any memslot at all. An invalidation of the cached HVA which starts\nand ends entirely within the lookup window is thus invisible to the\nretry check: mn_active_invalidate_count is back to zero and the\nsequence never moved. The refresh then publishes a mapping of a page\nwhich has already been freed, and the next reader dereferences it:\n\n  BUG: KASAN: use-after-free in kvm_xen_shared_info_init+0x3c6/0x440\n  Read of size 4 at addr ffff8880599c2900 by task syz.2.383/7257\n\nSince gfn_to_pfn_cache_invalidate_start() deliberately skips caches\nwhich are not currently valid (including one whose refresh is in\nprogress, as the refresh clears the valid flag before dropping the\nlock), the retry check is the only line of defence, and it must fire\nfor *any* invalidation, not just those hitting a memslot.\n\nAdd a dedicated kvm->gpc_invalidate_seq, incremented by every\nkvm_mmu_notifier_invalidate_range_end() under mn_invalidate_lock\nbefore mn_active_invalidate_count is decremented, and check it in\nhva_to_pfn_retry() instead of mmu_invalidate_seq. Incrementing in\nrange_end() in the same critical section as the in-progress count\nalso closes the variant where the cache is activated with the\ncontested HVA only after invalidate_range_start() has run.\n\nThe same bug is also reachable through the per-vCPU vcpu_info cache\n(KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO_HVA), where the stale mapping is\nthen dereferenced by kvm_setup_guest_pvclock() on the next KVM_RUN:\n\n  BUG: KASAN: use-after-free in kvm_setup_guest_pvclock+0x5bf/0x660\n\nThis intentionally makes refresh retry on *unrelated* mmu notifier\nevents; restoring precision (and reworking the GPC locking more\ngenerally) is left for a subsequent series.\n\nReproducers: https://david.woodhou.se/xen_shinfo_race.c\n             https://david.woodhou.se/vcpu_info_race.c"}],"affected":[{"vendor":"Linux","product":"Linux","defaultStatus":"unaffected","versions":[{"version":"b9220d32799a62c634f7b0b3618da52772626952","status":"affected","lessThan":"fabc002f6396202b5ad7ad4ceec24d27aad90f3f","versionType":"git"},{"version":"b9220d32799a62c634f7b0b3618da52772626952","status":"affected","lessThan":"da260bcb56feeb6489a83f310308453a0a7de25e","versionType":"git"}],"repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["include/linux/kvm_host.h","virt/kvm/kvm_main.c","virt/kvm/pfncache.c"]},{"vendor":"Linux","product":"Linux","defaultStatus":"affected","versions":[{"version":"6.9","status":"affected"},{"version":"0","status":"unaffected","lessThan":"6.9","versionType":"semver"},{"version":"7.3.2","status":"unaffected","lessThanOrEqual":"7.3.*","versionType":"semver"},{"version":"7.4-rc1","status":"unaffected","lessThanOrEqual":"*","versionType":"original_commit_for_fix"}],"repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["include/linux/kvm_host.h","virt/kvm/kvm_main.c","virt/kvm/pfncache.c"]}],"references":[{"url":"https://git.kernel.org/stable/c/fabc002f6396202b5ad7ad4ceec24d27aad90f3f"},{"url":"https://git.kernel.org/stable/c/da260bcb56feeb6489a83f310308453a0a7de25e"}],"metrics":[{"cvssV3_1":{"version":"3.1","vectorString":"CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H","baseScore":7.8,"baseSeverity":"HIGH"},"scenarios":[{"lang":"en","value":"AV:L - The stale pfn is published by hva_to_pfn_retry() when the host process races KVM_XEN_HVM_SET_ATTR (KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA) or KVM_XEN_VCPU_SET_ATTR (KVM_XEN_VCPU_ATTR_TYPE_VCPU_INFO_HVA) against an mmu_notifier invalidation of that same HVA. The HVA comes from that local ioctl, not a remote protocol message, and kvm_vm_ioctl/kvm_vcpu_ioctl reject a caller whose mm is not kvm->mm.\nAC:L - The same process drives both sides: one thread calls kvm_gpc_activate_hva(), which drops gpc->lock inside hva_to_pfn_retry() across hva_to_pfn(), while another munmaps that HVA so kvm_mmu_notifier_invalidate_range_end() completes in the window. The attacker chooses an HVA that overlaps no memslot, so the old mmu_invalidate_seq check never moves.\nPR:L - KVM_XEN_HVM_SET_ATTR and KVM_XEN_VCPU_SET_ATTR are dispatched from kvm_vm_ioctl and kvm_vcpu_ioctl with no capability check, and kvm_dev_ioctl(KVM_CREATE_VM) has none either. A local user who can open /dev/kvm creates the VM and issues the ioctl; the only gate is kvm->mm == current->mm.\nUI:N - The attacker mmaps the HVA, issues the Xen attribute ioctl, and invalidates that mapping itself. kvm_xen_shared_info_init() then dereferences the stale khva in the same KVM_XEN_HVM_SET_ATTR call, with no other user mounting or opening anything.\nS:U - The corrupted mapping is a host-kernel kmap of the attacking process's own page, reached from that process's VM fd. A guest cannot invoke KVM_XEN_ATTR_TYPE_SHARED_INFO_HVA, and kvm_vm_ioctl returns -EIO unless current->mm == kvm->mm, so this is host user-to-kernel escalation, not a guest escape.\nC:H - After kvm_release_page_clean() drops the last reference, kvm_xen_shared_info_init() reads shared_info.wc.version through the published gpc->khva (KASAN 4-byte use-after-free), and xen_get_guest_pvclock() memcpy's pvclock_vcpu_time_info from the same stale mapping. A reallocated page makes those reads observe the new owner.\nI:H - kvm_xen_shared_info_init() writes wall-clock fields at shared_info.wc (offset 0xc00) through that khva, kvm_setup_guest_pvclock() memcpy's a pvclock struct into the vcpu_info mapping, and kvm_xen_set_evtchn_fast() test_and_set_bit()s the caller-chosen port in evtchn_pending. Those stores land in the freed page.\nA:H - The published khva is consumed after the page is freed. syzbot reports BUG: KASAN use-after-free in kvm_xen_shared_info_init() on the shared_info path and in kvm_setup_guest_pvclock() on the vcpu_info path, which oopses the host kernel."}]}],"providerMetadata":{"shortName":"Linux","orgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","dateUpdated":"2026-11-22T11:08:29.877Z"},"cpeApplicability":[{"nodes":[{"operator":"OR","negate":false,"cpeMatch":[{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.9","versionEndExcluding":"7.3.2"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"6.9","versionEndExcluding":"7.4-rc1"}]}]}],"x_generator":{"engine":"bippy-1.2.0"}}}}}