[PATCH 3/4 v2] efi_runtime: Do not pass user addresses to firmware

[Borislav Petkov]

On Fri, Apr 04, 2014 at 04:26:42PM +0100, Matt Fleming wrote:
> From: Matt Fleming <matt.fleming at intel.com>
> 
> Currently there's some inconsistency with how arguments are passed to
> the firmware from the efi_runtime driver. Some values have the standard
> get_user()/put_user() calls, others do not.
> 
> Passing userspace pointers directly to the firmware is a bug because
> those addresses may fault. And if they are going to fault we'd like to
> know about it in the kernel rather than at some later time when
> executing in firmware context.
> 
> Furthermore, beginning with v3.14 of the kernel the current tests
> actually cause the kernel to crash because firmware calls are now done
> with their own, entirely separate, page tables - no user addresses are
> mapped during execution of runtime services.
> 
> This change doesn't require predication on a particular kernel version
> because the efi_runtime should really have always done this copying
> to/from userspace for every argument of the runtime services.
> 
> This patch is heavily based on one from Borislav.
> 
> Cc: Borislav Petkov <bp at alien8.de>
> Signed-off-by: Matt Fleming <matt.fleming at intel.com>
> ---
> 
> Changes in v2:
> 
>  - Delete the unused put_ucs2() function
>  - Rename get_ucs"* to copy_ucs2_* to dodge the reference counting
>    implications of the names
>  - Change function arguments to copy_ucs2_from_user_len() to mirror
>    copy_from_user() instead of put_user() to avoid confusion
>  - Expand comments for new helper functions

Acked-by: Borislav Petkov <bp at suse.de>

-- 
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--