mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH] mm/userfaultfd: check exclusivity of the page being moved
@ 2026-11-17 19:18 nakixii
  2026-11-17 19:52 ` Mike Rapoport
  2026-11-17 21:15 ` David Hildenbrand (Arm)
  0 siblings, 2 replies; 4+ messages in thread
From: nakixii @ 2026-11-17 19:18 UTC (permalink / raw)
  To: Andrew Morton
  Cc: Mike Rapoport, Peter Xu, David Hildenbrand, Suren Baghdasaryan,
	Lokesh Gidra, Kairui Song, linux-mm, linux-kernel

move_pages_ptes() gets the folio of a present source PTE and then checks
PageAnonExclusive() on the first page of that folio.  PG_anon_exclusive
is per page, so for a PTE-mapped large folio this can give the wrong
answer for the page that is actually being moved.

This happens after fork().  fork() clears the flag on every page of the
folio.  Once the child exits, a write fault in the parent can reuse the
large folio, and do_wp_page() sets the flag again only on the faulting
page.  Reclaim can leave the same state behind when try_to_unmap()
replaces the earlier pages with swap entries and then fails on a later
one.

Moving that page with UFFDIO_MOVE then fails with -EBUSY, even though
the folio could be split and the page moved.  Callers that fall back to
UFFDIO_COPY on -EBUSY end up copying instead.

Check the flag on the page mapped by the source PTE and get the folio
from that page.  Locking, refcounting and the split-and-retry path stay
the same.  The batched path only handles small folios and the PMD path
already checks the page the PMD maps, so they don't need this.

Tested in an x86_64 KVM guest on mm-unstable with only 64K anonymous
mTHP enabled: map a 64K-aligned region backed by one large folio, fork,
let the child exit, write to the sixth page and move it.  Without this
patch all 20 moves failed with -EBUSY.  With it all 20 worked and the
data was correct.  Moving the first page right after fork() still fails
with -EBUSY, as it should.  The move tests in uffd-unit-tests pass with
and without the patch (the two swap tests were skipped, there is no
swap in the guest).  I didn't reproduce the reclaim case and didn't
test arm64.

I found this with Claude Code while debugging an out-of-tree swap
backend, and it wrote the first version of the fix.  This version, the
changelog and the reproducer were also done with AI help.  I've reviewed
all of it.

Fixes: adef440691ba ("userfaultfd: UFFDIO_MOVE uABI")
Link: https://lore.kernel.org/all/20261116110122.309748-1-nakixii@gmail.com/
Signed-off-by: nakixii <nakixii@gmail.com>
---

Notes:
    Split out of the RFC linked above, as Kairui suggested, and rebased
    onto mm-unstable (81188836663676f8).  The code is the same as in the
    RFC; I rewrote the comment and the changelog.

 mm/userfaultfd.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
index d65184d9a..ae9ee210b 100644
--- a/mm/userfaultfd.c
+++ b/mm/userfaultfd.c
@@ -1557,6 +1557,7 @@ static long move_pages_ptes(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd
 		 */
 		if (!src_folio) {
 			struct folio *folio;
+			struct page *page;
 			bool locked;
 
 			/*
@@ -1570,8 +1571,14 @@ static long move_pages_ptes(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd
 				goto out;
 			}
 
-			folio = vm_normal_folio(src_vma, src_addr, orig_src_pte);
-			if (!folio || !PageAnonExclusive(&folio->page)) {
+			/*
+			 * PG_anon_exclusive is per page. The first page of a
+			 * large folio can be shared while the page mapped here
+			 * is still exclusive, so check the page being moved.
+			 */
+			page = vm_normal_page(src_vma, src_addr, orig_src_pte);
+			folio = page ? page_folio(page) : NULL;
+			if (!folio || !PageAnonExclusive(page)) {
 				spin_unlock(src_ptl);
 				ret = -EBUSY;
 				goto out;

base-commit: 81188836663676f8503f3603cfe56e88264c4d50
-- 
2.43.0


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH] mm/userfaultfd: check exclusivity of the page being moved
  2026-11-17 19:18 [PATCH] mm/userfaultfd: check exclusivity of the page being moved nakixii
@ 2026-11-17 19:52 ` Mike Rapoport
  2026-11-17 21:16   ` David Hildenbrand (Arm)
  2026-11-17 21:15 ` David Hildenbrand (Arm)
  1 sibling, 1 reply; 4+ messages in thread
From: Mike Rapoport @ 2026-11-17 19:52 UTC (permalink / raw)
  To: nakixii
  Cc: Mike Rapoport, Andrew Morton, Peter Xu, David Hildenbrand,
	Suren Baghdasaryan, Lokesh Gidra, Kairui Song, linux-mm,
	linux-kernel

On Wed, Nov 18, 2026 at 03:18:53AM +0800, nakixii wrote:
> move_pages_ptes() gets the folio of a present source PTE and then checks
> PageAnonExclusive() on the first page of that folio.  PG_anon_exclusive
> is per page, so for a PTE-mapped large folio this can give the wrong
> answer for the page that is actually being moved.
> 
> This happens after fork().  fork() clears the flag on every page of the
> folio.  Once the child exits, a write fault in the parent can reuse the
> large folio, and do_wp_page() sets the flag again only on the faulting
> page.

COW reuse of PTE-mapped large folios only appeared with commit
1da190f4d0a6 ("mm: Copy-on-Write (COW) reuse support for PTE-mapped
THP"), so this scenario didn't exist when UFFDIO_MOVE was added.

> Reclaim can leave the same state behind when try_to_unmap()
> replaces the earlier pages with swap entries and then fails on a later
> one.

This one, and fork() with a single page of a PTE-mapped large folio
pinned, could produce a mix of exclusive and shared pages in a folio
before that, so the Fixes tag is right, but please make it clear in the
changelog that the fork() + reuse case is a later addition.

> Moving that page with UFFDIO_MOVE then fails with -EBUSY, even though
> the folio could be split and the page moved.  Callers that fall back to
> UFFDIO_COPY on -EBUSY end up copying instead.

It also works the other way around: if the first page is exclusive and
the page being moved is not, we needlessly split the folio before the
retry notices the page is shared and returns -EBUSY.

Worth mentioning that as well.

...

> Signed-off-by: nakixii <nakixii@gmail.com>

Please use a known identity in Signed-off-by rather than a nickname, see
"Developer's Certificate of Origin" in
Documentation/process/submitting-patches.rst.

...

> -			folio = vm_normal_folio(src_vma, src_addr, orig_src_pte);
> -			if (!folio || !PageAnonExclusive(&folio->page)) {
> +			/*
> +			 * PG_anon_exclusive is per page. The first page of a
> +			 * large folio can be shared while the page mapped here
> +			 * is still exclusive, so check the page being moved.
> +			 */
> +			page = vm_normal_page(src_vma, src_addr, orig_src_pte);
> +			folio = page ? page_folio(page) : NULL;
> +			if (!folio || !PageAnonExclusive(page)) {

This would be simpler as

			page = vm_normal_page(src_vma, src_addr, orig_src_pte);
			if (!page || !PageAnonExclusive(page)) {
				spin_unlock(src_ptl);
				ret = -EBUSY;
				goto out;
			}

			folio = page_folio(page);

>  				spin_unlock(src_ptl);
>  				ret = -EBUSY;
>  				goto out;

With that addressed

Acked-by: Mike Rapoport (Microsoft) <rppt@kernel.org>

-- 
Sincerely yours,
Mike.

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH] mm/userfaultfd: check exclusivity of the page being moved
  2026-11-17 19:18 [PATCH] mm/userfaultfd: check exclusivity of the page being moved nakixii
  2026-11-17 19:52 ` Mike Rapoport
@ 2026-11-17 21:15 ` David Hildenbrand (Arm)
  1 sibling, 0 replies; 4+ messages in thread
From: David Hildenbrand (Arm) @ 2026-11-17 21:15 UTC (permalink / raw)
  To: nakixii, Andrew Morton
  Cc: Mike Rapoport, Peter Xu, Suren Baghdasaryan, Lokesh Gidra,
	Kairui Song, linux-mm, linux-kernel

On 11/17/26 20:18, nakixii wrote:
> move_pages_ptes() gets the folio of a present source PTE and then checks
> PageAnonExclusive() on the first page of that folio.  PG_anon_exclusive
> is per page, so for a PTE-mapped large folio this can give the wrong
> answer for the page that is actually being moved.
> 
> This happens after fork().  fork() clears the flag on every page of the
> folio.  Once the child exits, a write fault in the parent can reuse the
> large folio, and do_wp_page() sets the flag again only on the faulting
> page.  Reclaim can leave the same state behind when try_to_unmap()
> replaces the earlier pages with swap entries and then fails on a later
> one.
> 
> Moving that page with UFFDIO_MOVE then fails with -EBUSY, even though
> the folio could be split and the page moved.  Callers that fall back to
> UFFDIO_COPY on -EBUSY end up copying instead.

Ah, I was about to say, nothing bad will happen because we reject to move large
folios.

So this is pretty harmless and pretty uncommon to hit.

> @@ -1570,8 +1571,14 @@ static long move_pages_ptes(struct mm_struct *mm, pmd_t *dst_pmd, pmd_t *src_pmd
>  				goto out;
>  			}
>  
> -			folio = vm_normal_folio(src_vma, src_addr, orig_src_pte);
> -			if (!folio || !PageAnonExclusive(&folio->page)) {
> +			/*
> +			 * PG_anon_exclusive is per page. The first page of a
> +			 * large folio can be shared while the page mapped here
> +			 * is still exclusive, so check the page being moved.
> +			 */
> +			page = vm_normal_page(src_vma, src_addr, orig_src_pte);
> +			folio = page ? page_folio(page) : NULL;
> +			if (!folio || !PageAnonExclusive(page)) {

Yeah, we could have also moved it after the split and test for an early
folio_likely_mapped_shared() first, but doesn't really matter.

-- 
Cheers,

David

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH] mm/userfaultfd: check exclusivity of the page being moved
  2026-11-17 19:52 ` Mike Rapoport
@ 2026-11-17 21:16   ` David Hildenbrand (Arm)
  0 siblings, 0 replies; 4+ messages in thread
From: David Hildenbrand (Arm) @ 2026-11-17 21:16 UTC (permalink / raw)
  To: Mike Rapoport, nakixii
  Cc: Andrew Morton, Peter Xu, Suren Baghdasaryan, Lokesh Gidra,
	Kairui Song, linux-mm, linux-kernel

On 11/17/26 20:52, Mike Rapoport wrote:
> On Wed, Nov 18, 2026 at 03:18:53AM +0800, nakixii wrote:
>> move_pages_ptes() gets the folio of a present source PTE and then checks
>> PageAnonExclusive() on the first page of that folio.  PG_anon_exclusive
>> is per page, so for a PTE-mapped large folio this can give the wrong
>> answer for the page that is actually being moved.
>>
>> This happens after fork().  fork() clears the flag on every page of the
>> folio.  Once the child exits, a write fault in the parent can reuse the
>> large folio, and do_wp_page() sets the flag again only on the faulting
>> page.
> 
> COW reuse of PTE-mapped large folios only appeared with commit
> 1da190f4d0a6 ("mm: Copy-on-Write (COW) reuse support for PTE-mapped
> THP"), so this scenario didn't exist when UFFDIO_MOVE was added.

You could construct something odd even without that by using MADV_DONTFORK on
parts of the folio to then fork().

-- 
Cheers,

David

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2026-11-17 21:16 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-17 19:18 [PATCH] mm/userfaultfd: check exclusivity of the page being moved nakixii
2026-11-17 19:52 ` Mike Rapoport
2026-11-17 21:16   ` David Hildenbrand (Arm)
2026-11-17 21:15 ` David Hildenbrand (Arm)

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®