[PATCH v3 8/9] mm/migrate_device: do not migrate a uffd-protected zero page

From: Tal Zussman

Date: Tue Nov 17 2026 - 23:29:56 EST


Device migration replaces a PTE that maps the zero page with a new PTE
that is built only from the VMA flags, so its userfaultfd WP or RWP
protection is lost. The same applies to the huge zero PMD. Writes and
accesses to the page are then no longer reported to userfaultfd.

To fix it, do not migrate a uffd-protected zero page. Skip it during
collection, and check again under the page table lock before inserting,
as another thread can protect the page after it was collected.

The alternative would be to carry the uffd state over to the new entry,
but a uffd-protected zero page in a range that a driver migrates should
be rare.

While at it, read the PMD once with pmdp_get() in
migrate_vma_collect_huge_pmd().

Fixes: 292924b26024 ("userfaultfd: wp: apply _PAGE_UFFD_WP bit")
Fixes: a30b48bf1b24 ("mm/migrate_device: implement THP migration of zone device pages")
Cc: Thomas Hellström <thomas.hellstrom@xxxxxxxxxxxxxxx>
Cc: Lyude Paul <lyude@xxxxxxxxxx>
Cc: Danilo Krummrich <dakr@xxxxxxxxxx>
Cc: Felix Kuehling <Felix.Kuehling@xxxxxxx>
Cc: Jason Gunthorpe <jgg@xxxxxxxx>
Cc: Leon Romanovsky <leon@xxxxxxxxxx>
Cc: dri-devel@xxxxxxxxxxxxxxxxxxxxx
Cc: intel-xe@xxxxxxxxxxxxxxxxxxxxx
Cc: nouveau@xxxxxxxxxxxxxxxxxxxxx
Cc: amd-gfx@xxxxxxxxxxxxxxxxxxxxx
Reviewed-by: Alistair Popple <apopple@xxxxxxxxxx>
Signed-off-by: Tal Zussman <tz2294@xxxxxxxxxxxx>
---
include/linux/userfaultfd_k.h | 24 ++++++++++++++++++++++++
mm/migrate_device.c | 34 ++++++++++++++++++++++++++--------
2 files changed, 50 insertions(+), 8 deletions(-)

diff --git a/include/linux/userfaultfd_k.h b/include/linux/userfaultfd_k.h
index 5768a3902c7c..54850cdabf5e 100644
--- a/include/linux/userfaultfd_k.h
+++ b/include/linux/userfaultfd_k.h
@@ -246,6 +246,18 @@ static inline bool userfaultfd_huge_pmd_rwp(struct vm_area_struct *vma,
return userfaultfd_rwp(vma) && pmd_uffd(pmd);
}

+static inline bool userfaultfd_pte_protected(struct vm_area_struct *vma,
+ pte_t pte)
+{
+ return userfaultfd_protected(vma) && pte_uffd(pte);
+}
+
+static inline bool userfaultfd_huge_pmd_protected(struct vm_area_struct *vma,
+ pmd_t pmd)
+{
+ return userfaultfd_protected(vma) && pmd_uffd(pmd);
+}
+
static inline bool userfaultfd_armed(struct vm_area_struct *vma)
{
return vma_test_any_mask(vma, __VMA_UFFD_FLAGS);
@@ -390,6 +402,18 @@ static inline bool userfaultfd_huge_pmd_rwp(struct vm_area_struct *vma,
return false;
}

+static inline bool userfaultfd_pte_protected(struct vm_area_struct *vma,
+ pte_t pte)
+{
+ return false;
+}
+
+static inline bool userfaultfd_huge_pmd_protected(struct vm_area_struct *vma,
+ pmd_t pmd)
+{
+ return false;
+}
+
static inline bool userfaultfd_armed(struct vm_area_struct *vma)
{
return false;
diff --git a/mm/migrate_device.c b/mm/migrate_device.c
index 9e19fadfa65c..60985e9bec8b 100644
--- a/mm/migrate_device.c
+++ b/mm/migrate_device.c
@@ -142,30 +142,38 @@ static int migrate_vma_collect_huge_pmd(pmd_t *pmdp, unsigned long start,
struct folio *folio;
struct migrate_vma *migrate = walk->private;
spinlock_t *ptl;
+ pmd_t pmd;
int ret;
unsigned long write = 0;

ptl = pmd_lock(mm, pmdp);
- if (pmd_none(*pmdp)) {
+ pmd = pmdp_get(pmdp);
+ if (pmd_none(pmd)) {
spin_unlock(ptl);
return migrate_vma_collect_hole(start, end, -1, walk);
}

- if (pmd_trans_huge(*pmdp)) {
+ if (pmd_trans_huge(pmd)) {
if (!(migrate->flags & MIGRATE_VMA_SELECT_SYSTEM)) {
spin_unlock(ptl);
return migrate_vma_collect_skip(start, end, walk);
}

- folio = pmd_folio(*pmdp);
+ folio = pmd_folio(pmd);
if (is_huge_zero_folio(folio)) {
spin_unlock(ptl);
+ /*
+ * Nothing carries the uffd state of the zero page
+ * over, so skip it if its uffd bit is set.
+ */
+ if (userfaultfd_huge_pmd_protected(walk->vma, pmd))
+ return migrate_vma_collect_skip(start, end, walk);
return migrate_vma_collect_hole(start, end, -1, walk);
}
- if (pmd_write(*pmdp))
+ if (pmd_write(pmd))
write = MIGRATE_PFN_WRITE;
- } else if (!pmd_present(*pmdp)) {
- const softleaf_t entry = softleaf_from_pmd(*pmdp);
+ } else if (!pmd_present(pmd)) {
+ const softleaf_t entry = softleaf_from_pmd(pmd);

if (!softleaf_is_device_private(entry) ||
!(migrate->flags & MIGRATE_VMA_SELECT_DEVICE_PRIVATE)) {
@@ -337,6 +345,9 @@ static int migrate_vma_collect_pmd(pmd_t *pmdp,
pfn = pte_pfn(pte);
if (is_zero_pfn(pfn) &&
(migrate->flags & MIGRATE_VMA_SELECT_SYSTEM)) {
+ /* See migrate_vma_collect_huge_pmd(). */
+ if (userfaultfd_pte_protected(vma, pte))
+ goto next;
mpfn = MIGRATE_PFN_MIGRATE;
migrate->cpages++;
goto next;
@@ -871,10 +882,14 @@ static int migrate_vma_insert_huge_pmd_page(struct migrate_vma *migrate,
if (userfaultfd_missing(vma))
goto unlock_abort;

- if (is_huge_zero_pmd(*pmdp))
+ if (is_huge_zero_pmd(*pmdp)) {
+ /* The uffd bit can have been set since the collection. */
+ if (userfaultfd_huge_pmd_protected(vma, *pmdp))
+ goto unlock_abort;
flush = true;
- else if (!pmd_none(*pmdp))
+ } else if (!pmd_none(*pmdp)) {
goto unlock_abort;
+ }

add_mm_counter(vma->vm_mm, MM_ANONPAGES, HPAGE_PMD_NR);
folio_add_new_anon_rmap(folio, vma, addr, RMAP_EXCLUSIVE);
@@ -1074,6 +1089,9 @@ static void migrate_vma_insert_page(struct migrate_vma *migrate,

if (!is_zero_pfn(pfn))
goto unlock_abort;
+ /* The uffd bit can have been set since the collection. */
+ if (userfaultfd_pte_protected(vma, orig_pte))
+ goto unlock_abort;
flush = true;
} else if (!pte_none(orig_pte))
goto unlock_abort;

--
2.39.5