Re: [PATCH 5/8] efi: Get the secure boot status [ver #6]

[David Howells <dhowells-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>]

Matt Fleming <matt-mF/unelCI9GS6iBeEJttW/XRex20P6io@public.gmane.org> wrote:

> On Wed, 11 Jan, at 03:27:23PM, David Howells wrote:
> > Matt Fleming <matt-mF/unelCI9GS6iBeEJttW/XRex20P6io@public.gmane.org> wrote:
> > 
> > > > +	movb	$0, BP_secure_boot(%rsi)
> > > >  #ifdef CONFIG_EFI_STUB
> > > >  	/*
> > > >  	 * The entry point for the PE/COFF executable is efi_pe_entry, so
> > > 
> > > Is clearing ::secure_boot really necessary? Any code path that goes
> > > via efi_main() will set it correctly and all other code paths should
> > > get it cleared in sanitize_boot_params(), no?
> > 
> > No.
> > 
> > The boot_params->secure_boot parameter exists whether or not efi_main() is
> > traversed (ie. if EFI isn't enabled or CONFIG_EFI_STUB=n) and, if not cleared,
> > is of uncertain value.
> >
> > Further, sanitize_boot_params() has to be modified by this patch so as not to
> > clobber the secure_boot flag.
> 
> Any new parameters that boot loaders do not know about should be
> cleared to zero by default in the boot loader because boot_params
> itself should be zero'd when allocated.

Do you mean the boot loader or the boot wrapper?  If the loader, that is
outside my control - and given the purpose of the value, I'm not sure I
want to rely on that.

> There are two cases to consider:
> 
>  1) boot_params is not zero'd
>  2) boot_params is zero'd
> 
> 1) This is a broken boot loader implementation that violates the x86
> boot specification and I would never expect ->secure_boot to have a
> valid value.

If there's a boot specification that must be complied with, why does
sanitize_boot_params() even exist?  Why does the comment on it say:

 * Deal with bootloaders which fail to initialize unknown fields in
 * boot_params to zero.  The list fields in this list are taken from
 * analysis of kexec-tools; if other broken bootloaders initialize a
 * different set of fields we will need to figure out how to disambiguate.

> It should not be special-cased in sanitize_boot_params(), it should be
> zero'd.

Sigh.  sanitize_boot_params() is part of the problem.  The startup sequence
goes something like this:

 (0) We enter the boot wrapper.

 (1) We clear the secure-boot status value [my patch adds this].

 (2) The boot wrapper *may* invoke efi_main() - which will determine the
     secure-boot status.

 (3) The boot wrapper calls extract_kernel() to decompress the kernel.

 (4) extract_kernel() calls sanitize_boot_params() which would otherwise clear
     the secure-boot flag.

 (5) The boot wrapper jumps into the main kernel image, which now does not see
     the secure boot status value we calculated.

So, no, sanitize_boot_params() must *not* zero the value unless we change the
call point for s_b_p().

> 2) In this case ->secure_boot should be zero unless modified inside of
> efi_main().

I have no idea whether this is guaranteed or not.

> Did you hit the scenario where ->secure_boot has a garbage value while
> developing these patches? I wouldn't expect to see it in practice.

I haven't actually checked what the value was before I cleared it.  But, I've
found that security people get seriously paranoid about assuming things to be
implicitly so;-).

David