Do not mix class->size and object size during offsets/sizes
calculation in zs_obj_write().  Size classes can merge into
clusters, based on objects-per-zspage and pages-per-zspage
characteristics, so some size classes can store objects
smaller than class->size.  This becomes problematic when
object size is much smaller than class->size - we can determine
that object spans two physical pages, because we use a larger
class->size for this, while the actual object is much smaller
and fits one physical page, so there is nothing to write to
the second page and memcpy() size calculation underflows.

We always know the exact size in bytes of the object
that we are about to write (store), so use it instead of
class->size.

Reported-by: Igor Belousov <igor.b@beldev.am>
Cc: <stable@vger.kernel.org>
Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
---
 mm/zsmalloc.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
index 70406ac94bbd..999b513c7fdf 100644
--- a/mm/zsmalloc.c
+++ b/mm/zsmalloc.c
@@ -1233,19 +1233,19 @@ void zs_obj_write(struct zs_pool *pool, unsigned long handle,
 	class = zspage_class(pool, zspage);
 	off = offset_in_page(class->size * obj_idx);
 
-	if (off + class->size <= PAGE_SIZE) {
+	if (!ZsHugePage(zspage))
+		off += ZS_HANDLE_SIZE;
+
+	if (off + mem_len <= PAGE_SIZE) {
 		/* this object is contained entirely within a page */
 		void *dst = kmap_local_zpdesc(zpdesc);
 
-		if (!ZsHugePage(zspage))
-			off += ZS_HANDLE_SIZE;
 		memcpy(dst + off, handle_mem, mem_len);
 		kunmap_local(dst);
 	} else {
 		/* this object spans two pages */
 		size_t sizes[2];
 
-		off += ZS_HANDLE_SIZE;
 		sizes[0] = PAGE_SIZE - off;
 		sizes[1] = mem_len - sizes[0];
 
-- 
2.49.0.906.g1f30a19c02-goog

On Sun, May 04, 2025 at 08:00:22PM +0900, Sergey Senozhatsky wrote:
> Do not mix class->size and object size during offsets/sizes
> calculation in zs_obj_write().  Size classes can merge into
> clusters, based on objects-per-zspage and pages-per-zspage
> characteristics, so some size classes can store objects
> smaller than class->size.  This becomes problematic when
> object size is much smaller than class->size - we can determine
> that object spans two physical pages, because we use a larger
> class->size for this, while the actual object is much smaller
> and fits one physical page, so there is nothing to write to
> the second page and memcpy() size calculation underflows.
> 
> We always know the exact size in bytes of the object
> that we are about to write (store), so use it instead of
> class->size.
> 
> Reported-by: Igor Belousov <igor.b@beldev.am>
> Cc: <stable@vger.kernel.org>
> Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>

Could you please include user-visible effects and circumstances that
Igor reported? Crash, backtrace etc, 16k pages etc. in the changelog?

This type of information helps tremendously with backports, or finding
this patch when encountering the issue in the wild.

Acked-by: Johannes Weiner <hannes@cmpxchg.org>

On (25/05/06 09:56), Johannes Weiner wrote:
> Could you please include user-visible effects and circumstances that
> Igor reported? Crash, backtrace etc, 16k pages etc. in the changelog?
> 
> This type of information helps tremendously with backports, or finding
> this patch when encountering the issue in the wild.

Fair enough.   I'll send a v2 with updated commit message.

On (25/05/04 20:00), Sergey Senozhatsky wrote:
> Do not mix class->size and object size during offsets/sizes
> calculation in zs_obj_write().  Size classes can merge into
> clusters, based on objects-per-zspage and pages-per-zspage
> characteristics, so some size classes can store objects
> smaller than class->size.  This becomes problematic when
> object size is much smaller than class->size - we can determine
> that object spans two physical pages, because we use a larger
> class->size for this, while the actual object is much smaller
> and fits one physical page, so there is nothing to write to
> the second page and memcpy() size calculation underflows.
> 
> We always know the exact size in bytes of the object
> that we are about to write (store), so use it instead of
> class->size.

I think it's

Fixes: 44f76413496e ("zsmalloc: introduce new object mapping API")

On 2025-05-04 15:00, Sergey Senozhatsky wrote:
> Do not mix class->size and object size during offsets/sizes
> calculation in zs_obj_write().  Size classes can merge into
> clusters, based on objects-per-zspage and pages-per-zspage
> characteristics, so some size classes can store objects
> smaller than class->size.  This becomes problematic when
> object size is much smaller than class->size - we can determine
> that object spans two physical pages, because we use a larger
> class->size for this, while the actual object is much smaller
> and fits one physical page, so there is nothing to write to
> the second page and memcpy() size calculation underflows.
> 
> We always know the exact size in bytes of the object
> that we are about to write (store), so use it instead of
> class->size.
> 
> Reported-by: Igor Belousov <igor.b@beldev.am>
> Cc: <stable@vger.kernel.org>
> Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>

Tested-by: Igor Belousov <igor.b@beldev.am>

> ---
>  mm/zsmalloc.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
> index 70406ac94bbd..999b513c7fdf 100644
> --- a/mm/zsmalloc.c
> +++ b/mm/zsmalloc.c
> @@ -1233,19 +1233,19 @@ void zs_obj_write(struct zs_pool *pool, 
> unsigned long handle,
>  	class = zspage_class(pool, zspage);
>  	off = offset_in_page(class->size * obj_idx);
> 
> -	if (off + class->size <= PAGE_SIZE) {
> +	if (!ZsHugePage(zspage))
> +		off += ZS_HANDLE_SIZE;
> +
> +	if (off + mem_len <= PAGE_SIZE) {
>  		/* this object is contained entirely within a page */
>  		void *dst = kmap_local_zpdesc(zpdesc);
> 
> -		if (!ZsHugePage(zspage))
> -			off += ZS_HANDLE_SIZE;
>  		memcpy(dst + off, handle_mem, mem_len);
>  		kunmap_local(dst);
>  	} else {
>  		/* this object spans two pages */
>  		size_t sizes[2];
> 
> -		off += ZS_HANDLE_SIZE;
>  		sizes[0] = PAGE_SIZE - off;
>  		sizes[1] = mem_len - sizes[0];