Re: [PATCHv2 3/7] crypto: omap-crypto: fix userspace copied buffer access

[Herbert Xu <herbert@gondor.apana.org.au>]

On Mon, May 11, 2020 at 02:19:09PM +0300, Tero Kristo wrote:
> In case buffers are copied from userspace, directly accessing the page
> will most likely fail because it hasn't been mapped into the kernel
> memory space. Fix the issue by forcing a kmap / kunmap within the
> cleanup functionality.
> 
> Signed-off-by: Tero Kristo <t-kristo@ti.com>
> ---
>  drivers/crypto/omap-crypto.c | 9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/crypto/omap-crypto.c b/drivers/crypto/omap-crypto.c
> index cc88b7362bc2..31bdb1d76d11 100644
> --- a/drivers/crypto/omap-crypto.c
> +++ b/drivers/crypto/omap-crypto.c
> @@ -178,11 +178,16 @@ static void omap_crypto_copy_data(struct scatterlist *src,
>  		amt = min(src->length - srco, dst->length - dsto);
>  		amt = min(len, amt);
>  
> -		srcb = sg_virt(src) + srco;
> -		dstb = sg_virt(dst) + dsto;
> +		srcb = kmap_atomic(sg_page(src)) + srco + src->offset;
> +		dstb = kmap_atomic(sg_page(dst)) + dsto + dst->offset;
>  
>  		memcpy(dstb, srcb, amt);
>  
> +		flush_dcache_page(sg_page(dst));

You need to check !PageSlab as it's illegal to call it on a kernel
page.  Also you should be using flush_kernel_dcache_page.  scatterwalk
uses flush_dcache_page only because when it was created the other
function didn't exist.

Would it be possible to use the sg_miter interface to do the copy?
In fact your function could possibly be added to lib/scatterlist.c
as it seems to be quite generic.

Thanks,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt