Re: [PATCH] s390/mm: Fix memory leak in add_marker() when kvrealloc fails

[Heiko Carstens <hca@linux.ibm.com>]

On Sun, Oct 26, 2025 at 05:13:51PM +0800, Miaoqian Lin wrote:
> When kvrealloc() fails, the original markers memory is leaked
> because the function directly assigns the NULL to the markers pointer,
> losing the reference to the original memory.
> 
> As a result, the kvfree() in pt_dump_init() ends up freeing NULL instead
> of the previously allocated memory.
> 
> Fix this by using a temporary variable to store kvrealloc()'s return
> value and only update the markers pointer on success.
> 
> Found via static anlaysis and this is similar to commit 42378a9ca553
> ("bpf, verifier: Fix memory leak in array reallocation for stack state")
> 
> Fixes: d0e7915d2ad3 ("s390/mm/ptdump: Generate address marker array dynamically")
> Cc: stable@vger.kernel.org
> Signed-off-by: Miaoqian Lin <linmq006@gmail.com>
> ---
>  arch/s390/mm/dump_pagetables.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/s390/mm/dump_pagetables.c b/arch/s390/mm/dump_pagetables.c
> index 9af2aae0a515..0f2e0c93a1e0 100644
> --- a/arch/s390/mm/dump_pagetables.c
> +++ b/arch/s390/mm/dump_pagetables.c
> @@ -291,16 +291,19 @@ static int ptdump_cmp(const void *a, const void *b)
>  
>  static int add_marker(unsigned long start, unsigned long end, const char *name)
>  {
> +	struct addr_marker *new_markers;
>  	size_t oldsize, newsize;
>  
>  	oldsize = markers_cnt * sizeof(*markers);
>  	newsize = oldsize + 2 * sizeof(*markers);
>  	if (!oldsize)
> -		markers = kvmalloc(newsize, GFP_KERNEL);
> +		new_markers = kvmalloc(newsize, GFP_KERNEL);
>  	else
> -		markers = kvrealloc(markers, newsize, GFP_KERNEL);
> -	if (!markers)
> +		new_markers = kvrealloc(markers, newsize, GFP_KERNEL);
> +	if (!new_markers)
>  		goto error;
> +
> +	markers = new_markers;

This is not better to the situation before. If the allocation fails,
markers_cnt will be set to zero, but the old valid markers pointer will stay,
which means that the next call to add_marker() will allocate a new area via
kvmalloc() instead of kvrealloc(), and thus leaking the old area too.

add_marker() needs to changes to return in a manner that both marker and
marker_cnt correlate with each other. And I guess it is also easily possible
to get rid of the two different allocation paths.

Care to send a new version?