Re: [PATCH] crypto: using size_add() for kmalloc()

[Dan Carpenter <dan.carpenter@linaro.org>]

On Mon, Apr 21, 2025 at 08:05:43PM +0800, Herbert Xu wrote:
> On Mon, Apr 21, 2025 at 01:51:06PM +0800, Su Hui wrote:
> >
> > @@ -433,7 +434,7 @@ static inline struct aead_request *aead_request_alloc(struct crypto_aead *tfm,
> >  {
> >  	struct aead_request *req;
> >  
> > -	req = kmalloc(sizeof(*req) + crypto_aead_reqsize(tfm), gfp);
> > +	req = kmalloc(size_add(sizeof(*req), crypto_aead_reqsize(tfm)), gfp);
> 
> This is just wrong.  You should fail the allocation altogether
> rather than proceeding with a length that is insufficient.

When size_add() overflows then it returns SIZE_MAX.  None of the
allocation functions can allocate SIZE_MAX bytes so kmalloc() will
fail and that's already handled correctly.  Meanwhile if
"sizeof(*req) + crypto_aead_reqsize(tfm)" overflows then the
allocation will succeed and it results in memory corruption.

This is exactly what Kees did with the mass conversion to
struct_size().  I occasionally run across places where Kees's mass
conversion patches did fix real integer overflow bugs.

regards,
dan carpenter