Re: [GIT PULL] selinux/selinux-pr-20250323

[Linus Torvalds]

On Wed, 26 Mar 2025 at 18:06, Thiébaud Weksteen <tweek@xxxxxxxxxx> wrote:
>
> Taking one example from this merge request: kexec image loading.

So this is the kind of "why" I was looking for.

> Currently, any process which has CAP_SYS_BOOT can use kexec to replace
> the existing kernel. Android has 5 processes with CAP_SYS_BOOT, only 1
> of which needs kexec functionality [1]. By using these new
> permissions, we will ensure that this process is able to call kexec,
> while prohibiting other processes. SELinux provides us strong, kernel
> enforced guarantees which can be checked at policy compile time.
> Extending on this, we will use this patchset to guarantee that kernels
> and ramdisks executed by kexec come from known, good sources.
>
> The other hooks are of similar value to Android.

Now explain to me how the firmware loading hook works, not some
hand-wavy "similar value" thing.

Because it seems entirely bogus. Exactly because the context of
firmware loading is *not* something you can depend on. There is no
"one special process" that has firmware loading capabilities.

I'm looking at selinux_kernel_load_data() in particular, where you
don't even pass it a file at all, so it's not like it could check for
"is this file integrity-protected" or anything like that. It seems to
literally say "can this process load firmware", and as I've explained,
the firmware loading is done by random processes.

               Linus