Re: [PULL] modules-next

[Ben Hutchings <ben@decadent.org.uk>]

[-- Attachment #1: Type: text/plain, Size: 2460 bytes --]

On Sun, 2016-07-31 at 21:44 -0400, Linus Torvalds wrote:
> So this feels wrong to me, can you guys please explain:
> 
> On Sun, Jul 31, 2016 at 9:02 PM, Rusty Russell <rusty@rustcorp.com.au> wrote:
> > 
> > Ben Hutchings (3):
> >       module: Invalidate signatures on force-loaded modules
> >       module: Disable MODULE_FORCE_LOAD when MODULE_SIG_FORCE is enabled
> 
> forcing a load and SIG_FORCE are entirely independent issues, afaik. I
> think requiring signed modules is just a good idea. But that doesn't
> necessarily mean that you don't have a signed module that is signed
> with a key you trust, but you still want to force-load it for the
> wrong kernel version (ie maybe you have a binary-only module from your
> IT department (and your IT department is evil,but at least they sign
> it to show that the module is trust-worthy as coming from them, even
> if they have some dubious behavior), but you did some kernel updates
> that still allow the module to work but the version doesn't match any
> more).

We discussed this before and I thought you were happy with this
version.  If the use case you describe is at all common, it could
perhaps be handled by having a tool that patches the version
information and re-signs the module with a different trusted key.

> Am I missing something? What's the connection between
> MODULE_FORCE_LOAD and MODULE_SIG_FORCE? Because it smells like they
> are independent and that the above changes are very very dubious.

As I understand it:
- module signature enforcement means that root is not trusted to load
  arbitrary code into the kernel; instead the code has to be approved
  by one of the signing key holders
- force-loading a module means "I promise that this module is ABI
  compatible, even though it doesn't appear to be"

No-one signs that promise, and if it's false, the ABI differences could
mean that an otherwise benign module would compromise the kernel.  So
as I see it, the kernel should not trust a force-loaded signed module
any more than an unsigned module.

If you still think that module signature enforcement is compatible with
force-loading, I would like to know what you consider the purpose of
enforcement to be.

Ben.

> I didn't actually pull the tree, I just reacted to the pull request itself.
> 
>              Linus
-- 

Ben Hutchings
Sturgeon's Law: Ninety percent of everything is crap.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]