[PATCH] mm/migrate_device: avoid out-of-bounds writes for compound folios

[PATCH] mm/migrate_device: avoid out-of-bounds writes for compound folios

[Hui Su]

migrate_device_range() and migrate_device_pfns() clear the entries
following a compound folio so that the PFN arrays retain their
page-granular representation.

If a compound folio extends beyond the end of the caller-provided range,
the loops clear all following folio entries without limiting them to the
number of slots remaining in the npages-sized array, causing an
out-of-bounds write.

Limit the number of entries updated and consumed to the remaining array
slots while keeping the actual number of pages in the folio unchanged.

Observed with a KASAN x86 QEMU kernel using the HMM
migrate_anon_huge_zero selftest. Closing /dev/hmm_dmirror0 after
migrating an anonymous huge page to device memory exercises:

  dmirror_fops_release()
    -> dmirror_device_evict_chunk()
      -> migrate_device_range()

Fixes: a30b48bf1b24 ("mm/migrate_device: implement THP migration of zone device pages")
Signed-off-by: Hui Su <sh_def@163.com>
---
 mm/migrate_device.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)

diff --git a/mm/migrate_device.c b/mm/migrate_device.c
index 908d2d4ec43a..67bfc1af0b29 100644
--- a/mm/migrate_device.c
+++ b/mm/migrate_device.c
@@ -1396,16 +1396,17 @@ int migrate_device_range(unsigned long *src_pfns, unsigned long start,
 	for (pfn = start, i = 0; i < npages; pfn++, i++) {
 		struct page *page = pfn_to_page(pfn);
 		struct folio *folio = page_folio(page);
-		unsigned int nr = 1;
+		unsigned long nr, slots;
 
 		src_pfns[i] = migrate_device_pfn_lock(pfn);
 		nr = folio_nr_pages(folio);
+		slots = min(nr, npages - i);
 		if (nr > 1) {
 			src_pfns[i] |= MIGRATE_PFN_COMPOUND;
-			for (j = 1; j < nr; j++)
+			for (j = 1; j < slots; j++)
 				src_pfns[i+j] = 0;
-			i += j - 1;
-			pfn += j - 1;
+			i += slots - 1;
+			pfn += slots - 1;
 		}
 	}
 
@@ -1430,15 +1431,16 @@ int migrate_device_pfns(unsigned long *src_pfns, unsigned long npages)
 	for (i = 0; i < npages; i++) {
 		struct page *page = pfn_to_page(src_pfns[i]);
 		struct folio *folio = page_folio(page);
-		unsigned int nr = 1;
+		unsigned long nr, slots;
 
 		src_pfns[i] = migrate_device_pfn_lock(src_pfns[i]);
 		nr = folio_nr_pages(folio);
+		slots = min(nr, npages - i);
 		if (nr > 1) {
 			src_pfns[i] |= MIGRATE_PFN_COMPOUND;
-			for (j = 1; j < nr; j++)
+			for (j = 1; j < slots; j++)
 				src_pfns[i+j] = 0;
-			i += j - 1;
+			i += slots - 1;
 		}
 	}
 
-- 
2.54.0

Re: [PATCH] mm/migrate_device: avoid out-of-bounds writes for compound folios

[Balbir Singh]

On Mon, Aug 17, 2026 at 03:43:51PM +0800, Hui Su wrote:
> migrate_device_range() and migrate_device_pfns() clear the entries
> following a compound folio so that the PFN arrays retain their
> page-granular representation.
> 
> If a compound folio extends beyond the end of the caller-provided range,
> the loops clear all following folio entries without limiting them to the
> number of slots remaining in the npages-sized array, causing an
> out-of-bounds write.
> 
> Limit the number of entries updated and consumed to the remaining array
> slots while keeping the actual number of pages in the folio unchanged.
> 
> Observed with a KASAN x86 QEMU kernel using the HMM
> migrate_anon_huge_zero selftest. Closing /dev/hmm_dmirror0 after
> migrating an anonymous huge page to device memory exercises:
> 
>   dmirror_fops_release()
>     -> dmirror_device_evict_chunk()
>       -> migrate_device_range()
> 
> Fixes: a30b48bf1b24 ("mm/migrate_device: implement THP migration of zone device pages")
> Signed-off-by: Hui Su <sh_def@163.com>
> ---
>  mm/migrate_device.c | 16 +++++++++-------
>  1 file changed, 9 insertions(+), 7 deletions(-)
> 
> diff --git a/mm/migrate_device.c b/mm/migrate_device.c
> index 908d2d4ec43a..67bfc1af0b29 100644
> --- a/mm/migrate_device.c
> +++ b/mm/migrate_device.c
> @@ -1396,16 +1396,17 @@ int migrate_device_range(unsigned long *src_pfns, unsigned long start,
>  	for (pfn = start, i = 0; i < npages; pfn++, i++) {
>  		struct page *page = pfn_to_page(pfn);
>  		struct folio *folio = page_folio(page);
> -		unsigned int nr = 1;
> +		unsigned long nr, slots;
>  
>  		src_pfns[i] = migrate_device_pfn_lock(pfn);
>  		nr = folio_nr_pages(folio);
> +		slots = min(nr, npages - i);
>  		if (nr > 1) {
>  			src_pfns[i] |= MIGRATE_PFN_COMPOUND;
> -			for (j = 1; j < nr; j++)
> +			for (j = 1; j < slots; j++)
>  				src_pfns[i+j] = 0;
> -			i += j - 1;
> -			pfn += j - 1;
> +			i += slots - 1;
> +			pfn += slots - 1;
>  		}
>  	}
>  
> @@ -1430,15 +1431,16 @@ int migrate_device_pfns(unsigned long *src_pfns, unsigned long npages)
>  	for (i = 0; i < npages; i++) {
>  		struct page *page = pfn_to_page(src_pfns[i]);
>  		struct folio *folio = page_folio(page);
> -		unsigned int nr = 1;
> +		unsigned long nr, slots;
>  
>  		src_pfns[i] = migrate_device_pfn_lock(src_pfns[i]);
>  		nr = folio_nr_pages(folio);
> +		slots = min(nr, npages - i);
>  		if (nr > 1) {
>  			src_pfns[i] |= MIGRATE_PFN_COMPOUND;
> -			for (j = 1; j < nr; j++)
> +			for (j = 1; j < slots; j++)
>  				src_pfns[i+j] = 0;
> -			i += j - 1;
> +			i += slots - 1;
>  		}
>  	}
>  
> -- 
> 2.54.0
> 
>

There is a similar fix sent out by Matthew, The question is, is it OK to
send in partial data upto npages (when nr > npages - i)? We should bail
early and not proceed if that is the case.

Balbir