Re: OpenSSL patch to support Linux CryptoAPI.

[Evgeniy Polyakov <johnpol@2ka.mipt.ru>]

Hi.

On Fri, Aug 08, 2008 at 11:31:58AM -0700, Shasi Pulijala (spulijala@amcc.com) wrote:
>  struct crypt_op {
> -	u_int32_t	ses;
> -	u_int16_t	op;		/* i.e. COP_ENCRYPT */
>  #define COP_NONE	0
>  #define COP_ENCRYPT	1
>  #define COP_DECRYPT	2
> -	u_int16_t	flags;
> -#define	COP_F_BATCH	0x0008		/* Batch op if possible */
> -	u_int		len;
> -	caddr_t		src, dst;	/* become iov[] inside kernel */
> -	caddr_t		mac;		/* must be big enough for chosen MAC */
> -	caddr_t		iv;
> +	__u16		op;		/* i.e. COP_ENCRYPT */
> +	__u16		flags;
> +	__u16		iv_size;
> +	__u16 		assoc_size;
> +	__u32		src_size;
> +	caddr_t		src_data;
> +	caddr_t		dst_data;
> +	__u8		data[0];	/* must be big enough for chosen MAC */
>  };

If above caddr_t is what I thought (i.e. a pointer or long type),
there is no way it can be correct. It is _NOT_ allowed to put
variable sized members into structures shared between kernel
and userspace.

-- 
	Evgeniy Polyakov