Skip to contentSkip to conversation

mm: userfaultfd: correct dirty flags set for both present and swap pte

RFC PATCH Superseded Barry Song 5 participants 7 messages May 8, 2025

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Barry Song

author
#0 lore raw

RFC PATCH 1/1 mm: userfaultfd: correct dirty flags set for both present and swap pte

From: Barry Song <[email protected]>

As David pointed out, what truly matters for mremap and userfaultfd move operations is the soft dirty bit. The current comment and implementation—which always sets the dirty bit for present PTEs and fails to set the soft dirty bit for swap PTEs—are incorrect. This patch updates the behavior to correctly set the soft dirty bit for both present and swap PTEs in accordance with mremap.

---
 mm/userfaultfd.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
mm/userfaultfd.c +10 -2
index e8ce92dc105f..bc473ad21202 100644
@@ -1064,8 +1064,13 @@ static int move_present_pte(struct mm_struct *mm,
 	src_folio->index = linear_page_index(dst_vma, dst_addr);
 
 	orig_dst_pte = folio_mk_pte(src_folio, dst_vma->vm_page_prot);
-	/* Follow mremap() behavior and treat the entry dirty after the move */
-	orig_dst_pte = pte_mkwrite(pte_mkdirty(orig_dst_pte), dst_vma);
+	/* Set soft dirty bit so userspace can notice the pte was moved */
+#ifdef CONFIG_MEM_SOFT_DIRTY
+	orig_dst_pte = pte_mksoft_dirty(orig_dst_pte);
+#endif
+	if (pte_dirty(orig_src_pte))
+		orig_dst_pte = pte_mkdirty(orig_dst_pte);
+	orig_dst_pte = pte_mkwrite(orig_dst_pte, dst_vma);
 
 	set_pte_at(mm, dst_addr, dst_pte, orig_dst_pte);
 out:
@@ -1100,6 +1105,9 @@ static int move_swap_pte(struct mm_struct *mm, struct vm_area_struct *dst_vma,
 	}
 
 	orig_src_pte = ptep_get_and_clear(mm, src_addr, src_pte);
+#ifdef CONFIG_MEM_SOFT_DIRTY
+	orig_src_pte = pte_swp_mksoft_dirty(orig_src_pte);
+#endif
 	set_pte_at(mm, dst_addr, dst_pte, orig_src_pte);
 	double_pt_unlock(dst_ptl, src_ptl);
 
Signature · 1 line
2.39.3 (Apple Git-146)
#1 lore raw
Quoted 40 lines · Barry Song

From: Barry Song <[email protected]>

As David pointed out, what truly matters for mremap and userfaultfd move operations is the soft dirty bit. The current comment and implementation—which always sets the dirty bit for present PTEs and fails to set the soft dirty bit for swap PTEs—are incorrect. This patch updates the behavior to correctly set the soft dirty bit for both present and swap PTEs in accordance with mremap.

---
  mm/userfaultfd.c | 12 ++++++++++--
  1 file changed, 10 insertions(+), 2 deletions(-)
mm/userfaultfd.c +7 -2
index e8ce92dc105f..bc473ad21202 100644
@@ -1064,8 +1064,13 @@ static int move_present_pte(struct mm_struct *mm,
  	src_folio->index = linear_page_index(dst_vma, dst_addr);
 
  	orig_dst_pte = folio_mk_pte(src_folio, dst_vma->vm_page_prot);
-	/* Follow mremap() behavior and treat the entry dirty after the move */
-	orig_dst_pte = pte_mkwrite(pte_mkdirty(orig_dst_pte), dst_vma);
+	/* Set soft dirty bit so userspace can notice the pte was moved */
+#ifdef CONFIG_MEM_SOFT_DIRTY
+	orig_dst_pte = pte_mksoft_dirty(orig_dst_pte);
+#endif
+	if (pte_dirty(orig_src_pte))
+		orig_dst_pte = pte_mkdirty(orig_dst_pte);
+	orig_dst_pte = pte_mkwrite(orig_dst_pte, dst_vma);
 
  	set_pte_at(mm, dst_addr, dst_pte, orig_dst_pte);
  out:

mm/userfaultfd.c static int move_swap_pte(struct mm_struct *mm, struct vm_area_struct *dst_vma, · line 1112

Quoted 8 lines
mm/userfaultfd.c +3 -0
@@ -1100,6 +1105,9 @@ static int move_swap_pte(struct mm_struct *mm, struct vm_area_struct *dst_vma,
  	}
 
  	orig_src_pte = ptep_get_and_clear(mm, src_addr, src_pte);
+#ifdef CONFIG_MEM_SOFT_DIRTY
+	orig_src_pte = pte_swp_mksoft_dirty(orig_src_pte);
+#endif
  	set_pte_at(mm, dst_addr, dst_pte, orig_src_pte);
  	double_pt_unlock(dst_ptl, src_ptl);

Yeah, I think that should be the right thing to do.

Signature · 3 lines
Cheers,

David / dhildenb
reviewer Acked-by on v1 1/1
#2 lore raw
In reply to Barry Song
Quoted 9 lines · Barry Song

From: Barry Song <[email protected]>

As David pointed out, what truly matters for mremap and userfaultfd move operations is the soft dirty bit. The current comment and implementation—which always sets the dirty bit for present PTEs and fails to set the soft dirty bit for swap PTEs—are incorrect. This patch updates the behavior to correctly set the soft dirty bit for both present and swap PTEs in accordance with mremap.

Quoted 6 lines
Signature · 1 line
Peter Xu
#3 lore raw

Thanks Barry for fixing this.

Quoted 8 lines · Peter Xu
Quoted 6 lines · Barry Song

From: Barry Song <[email protected]>

As David pointed out, what truly matters for mremap and userfaultfd move operations is the soft dirty bit. The current comment and implementation—which always sets the dirty bit for present PTEs and fails to set the soft dirty bit for swap PTEs—are incorrect.

Can you please briefly describe the consequences of not setting the soft-dirty bit? I'm wondering if it needs to be backported as a fix?

Quoted 15 lines
Quoted 9 lines

This patch updates the behavior to correctly set the soft dirty bit for both present and swap PTEs in accordance with mremap.

--
Peter Xu

author
#4 lore raw
Quoted 8 lines · Lokesh Gidra

Thanks Barry for fixing this.

Quoted 4 lines · Peter Xu
Quoted 2 lines · Barry Song

From: Barry Song <[email protected]>

Quoted 7 lines
Quoted 4 lines
Quoted 4 lines

As David pointed out, what truly matters for mremap and userfaultfd move operations is the soft dirty bit. The current comment and implementation—which always sets the dirty bit for present PTEs and fails to set the soft dirty bit for swap PTEs—are incorrect.

Can you please briefly describe the consequences of not setting the soft-dirty bit? I'm wondering if it needs to be backported as a fix?

As I understand it, this could break features like Checkpoint-Restore in Userspace (CRIU), which relies on tracking memory changes to create incremental dumps. While Android may not currently have a real-world use case for this, it would still be beneficial to backport the fix in a general way.

Quoted 11 lines
Quoted 11 lines
Quoted 10 lines

This patch updates the behavior to correctly set the soft dirty bit for both present and swap PTEs in accordance with mremap.

Quoted 4 lines
Quoted 4 lines

--
Peter Xu

Thanks
Barry

#5 lore raw
Quoted 15 lines · Barry Song
Quoted 13 lines · Lokesh Gidra

Thanks Barry for fixing this.

Quoted 8 lines · Peter Xu
Quoted 6 lines · Barry Song

From: Barry Song <[email protected]>

As David pointed out, what truly matters for mremap and userfaultfd move operations is the soft dirty bit. The current comment and implementation—which always sets the dirty bit for present PTEs and fails to set the soft dirty bit for swap PTEs—are incorrect.

Quoted 9 lines
Quoted 2 lines

Can you please briefly describe the consequences of not setting the soft-dirty bit? I'm wondering if it needs to be backported as a fix?

As I understand it, this could break features like Checkpoint-Restore in Userspace (CRIU), which relies on tracking memory changes to create incremental dumps. While Android may not currently have a real-world use case for this, it would still be beneficial to backport the fix in a general way.

Makes sense. Thanks for clarifying.

Quoted 18 lines
Quoted 11 lines
Quoted 11 lines
Quoted 10 lines

This patch updates the behavior to correctly set the soft dirty bit for both present and swap PTEs in accordance with mremap.

Quoted 4 lines
Quoted 4 lines

--
Peter Xu

Thanks
Barry

Reviewed-by
#6 lore raw
In reply to Peter Xu
Quoted 11 lines · Peter Xu
Quoted 9 lines · Barry Song

From: Barry Song <[email protected]>

As David pointed out, what truly matters for mremap and userfaultfd move operations is the soft dirty bit. The current comment and implementation—which always sets the dirty bit for present PTEs and fails to set the soft dirty bit for swap PTEs—are incorrect. This patch updates the behavior to correctly set the soft dirty bit for both present and swap PTEs in accordance with mremap.

Quoted 8 lines
Quoted 6 lines

The function was introduced in 6.8. We should probably CC stable as well.

Quoted 4 lines

--
Peter Xu