Re: [gentoo-dev] [PATCH] linux-mod.eclass: support module signing

[Mike Gilbert]

On Mon, Jun 27, 2022 at 5:11 PM Georgy Yakovlev <gyakovlev@g.o> wrote:
>
> On Mon, 2022-06-27 at 15:49 -0400, Mike Gilbert wrote:
> > On Mon, Jun 27, 2022 at 3:42 PM Georgy Yakovlev
> > <gyakovlev@g.o> wrote:
> > >
> > > On Mon, 2022-06-27 at 14:56 -0400, Mike Gilbert wrote:
> > > > On Mon, Jun 27, 2022 at 2:35 PM Kenton Groombridge
> > > > <concord@g.o> wrote:
> > > > > > so looks like we need to combine both methods and do the
> > > > > > following:
> > > > > >  - if signing requested without compression - sign in
> > > > > > pkg_preinst.
> > > > > >  - if signing requested with compression - sign in
> > > > > > src_install
> > > > > >
> > > > >
> > > > > Why can't we do both in pkg_preinst? I am thinking it would be
> > > > > best
> > > > > if
> > > > > we drop the current compression implementation and rework your
> > > > > old
> > > > > code
> > > > > to handle both compression and signing since the signing code
> > > > > is
> > > > > more or
> > > > > less already complete.
> > > >
> > > > Signing modules in pkg_preinst seems like a bad idea to me. That
> > > > means
> > > > you need to copy your private keys around to every host where the
> > > > package might be installed.
> > > >
> > > > If you sign in src_compile or src_install, you only need private
> > > > keys
> > > > on the system building your binpkg.
> > > >
> > >
> > > unfortunately portage will unconditionally strip .ko objects,
> > > rendering
> > > modules unloadable by stripping signature,  unless we do dostrip -x
> > > (requires EAPI7+, which should not be a problem nowadays, but was a
> > > problem back in 2018), which can be quite unfortunate on debug
> > > enabled
> > > kernels.
> >
> > Sounds like something to fix/change in Portage. It could probably be
> > updated to not strip the signature. However, I would guess the
> > signature needs to be updated after the binary is modified in any
> > case.
> >
> > Or as a workaround you could disable automatic striping via dostrip -
> > x
> > and run the proper commands to strip the modules in src_install as
> > well.
> >
> I think even strip itself does not have proper options not to break
> module. Several years back it was the case, basically one has to strip
> first, sign second, otherwise module will be unloadable.
>
> "Signed modules are BRITTLE as the signature is outside of the defined
> ELF container. Thus they MAY NOT be stripped once the signature is
> computed and attached. Note the entire module is the signed payload,
> including any and all debug information present at the time of
> signing."
>
> https://www.kernel.org/doc/html/v4.15/admin-guide/module-signing.html#signed-modules-and-stripping
>

In that case, I think the only viable way to make this work is to
disable automatic stripping and handle stripping via custom code in
the ebuild/eclass.