[PATCH 1/3 v2] efi_capsule: Move signature from DTB to .rodata

[Simon Glass]

Hi,

On Thu, 22 Jul 2021 at 14:54, Ilias Apalodimas
<ilias.apalodimas at linaro.org> wrote:
>
> Hi Simon,
>
> On Thu, Jul 22, 2021 at 10:46:40AM -0600, Simon Glass wrote:
> > > > > > > >> In some platforms the key is derived from the relocated DTB, which we
>
> [...]
>
> > > > > > > >> can overwrite. But I'll let Sughosh who figured it out explain the
> > > > > > > >> details.
> > > > > > > >
> > > > > > > >
> > > > > > > > On platforms where the dtb is concatenated with the u-boot image, using CONFIG_OF_SEPARATE, the fdt is also getting relocated to the main memory. We retrieve the public key from this dtb. By default, the fdtcontroladdr env variable is getting set to this relocated dtb address -- this address can also be accessed using the bdinfo command. Thus the public key can be modified before attempting the capsule update. Which is the reason why Ilias is moving the public key to the embedded rodata section.
> > > > > > >
> > > > > > > You should be clearer about what problem you are trying to solve. Are
> > > > > > > you worried about a script changing the DT? Or just it being writable
> > > > > > > in general?
> > > > > >
> > > > > > Being writable in general is my main concern. Doing fixup internally
> > > > > > from U-Boot might be something we'll always need but the ability to
> > > > > > completely change it doesn't play well security.
> > > > > >
> > > > > > >
> > > > > > > U-Boot itself is relocated also, including the rodata. So are you
> > > > > > > using the public key from the original location? What if that is not
> > > > > > > accessible after relocation?
> > > > > >
> > > > > > We are accessing he key from the relocated address.
> > > > >
> > > > > Then in what way are you protecting it? This is so confusing. Are you
> > > > > saying that you are protecting the relocated address? If so, protect
> > > > > the relocated devicetree too!
> > > > >
> > >
> > > How? DTBs if fixed up and there's a protocol proposal from Heinrich,
> > > which allows fixups from GRUB2.  So how exactly are you going to put
> > > it in r/o memory (which is what .rodata is supposed to achieve).
> >
> > Because they are different DTBs, right? Either it can be read-only or
> > can't be read-only. At first you said it could not be read-only. Now
> > you are saying it needs to be changed. Where is all this coming from?
> >
>
> Not exactly, I said it can be read only, assuming we can switch pages to
> RO, as long CONFIG_OF_EMBED is enabled.  What happens if you choose
> CONFIG_OF_PRIOR_STAGE or CONFIG_OF_SEPARATE? That means the *prior* stage
> boot loader needs to know your public key and inject in the dtb it hands
> over?
>
> > > A big portion of the DTBs we build today are horribly outdated
> > > compared to the current upstream.  Since nowdays there's a spec
> >
> > That may be true on some boards but it is not my experience, at least
> > on ARM. Anyway that is an issue for the board maintainers. I don't
> > think this has any bearing on the points we are discussing here.
> >
>
> There's a discussion for DTs and it's evolution. Some are indeed a bit
> outdated and I can find details on that.  The point I am trying to make
> here is that the closer we keep the DTBs to what linux hosts, the easier
> it's going to be to keep them up to date.
>
> > > describing what can and can't go in a DTB, I'd much rather prefer we
> > > stick to that and make a potential update easier.
> >
> > There is just so much confusion in all of this and we are going around
> > in circles. Let me try to state what I think are points of confusion.
> >
> > 1. The U-Boot DT needs to be protected against change for lots of
> > reasons (drivers misbehaving, etc.). The signature is only one of
> > them.
> >
>
> We agree on that
>
> > 2. The U-Boot DT is separate from the one passed to Linux. So
> > discussion about where U-Boot config should go in the Linux DT is not
> > germain.
>
> Not always, there's cases were you can use the same DTB. The reason we don't
> is due to the diversion we have.  Ideally you should have a single DTB,
> which can be embedded into your firmware and you can authenticate it by
> just authentication the firmware, but I've abandoned that dream long
> ago.
>
> >
> > 3. U-Boot uses DT for its configuration and that is that. It has done
> > that for about 7-8 years. U-Boot does not have a user space to provide
> > policy and configuration . It cannot do what Linux does and run
> > programs and look up filesystems to figure out how to boot. So
> > configuration / runtime info go in the DT in U-Boot. I have not seen
> > any proposal to do it any other way. I hope you can understand how
> > frustrating to have someone come from the Linux world and say, Oh it's
> > all wrong...
>
> Apologies if it sounded like that. I am not trying to point any fingers or
> judge any code that's been there for a couple of years.
> I never said it's all wrong. I can understand some hardware specific config
> going into a DT that's been used for a couple of years.  What I don't
> understand is how a signature fits that profile.
>
> > we should put it user space, etc. The alternative to DT is
> > a mishmash of random places and ideas with no schema and no
> > discoverability, etc, or a forest of CONFIG options, like it used to
> > me.
> >
> > 4. The DT is relocated anyway so is not actually read-only just
> > because you put it in the rodata area.
>
> You mean the signature here right (not the DT).  The point is if we fix the
> .rodata section and it ends up on RO memory, then you do get what we want
> and it's always available not matter who provides the DTB, or what
> expectations you have for fixing it up (hence put it in R/W).
>
> >
> > 5. You can make the DT read-only if you want to. You can make any part
> > of memory read-only. You need to create an API for that if that's what
> > you want. It would be nice to have a command to look at what is
> > protected and change it. See for example the 'mtrr' command on x86.
> >
>
> I'll repeat myself but isn't that the case for CONFIG_OF_EMBED only?
>
> > Also I would add that there has always been a DT spec. I think the
> > latest version is here:
> > https://github.com/devicetree-org/devicetree-specification/releases/tag/v0.3
> > So far as I can tell it does not talk about what can and cannot go in a DT
> >
> > Perhaps the spec you are referring to is a Linux spec. Do you have a
> > link?
>
> I don't think there's a 'linux spec'. The DTS ended up being maintained in
> the kernel. I am not gonna argue if this is a good or bad thing, but that
> codebase is trying to follow the spec you pasted
>
> > So far as I can tell, U-Boot, Zephyr, etc. have had very little
> > input into that. I know for a fact that no one has asked what I think.
> > For example, even the u-boot,xxx tags are kept in separate files in
> > U-Boot because of resistance to putting that in Linux. Zephyr
> > completely does its own thing with DT. U-Boot very much follows Binux,
> > BUT it has its own things as well, just as Linux does. I would LOVE to
> > see that change and if you would like to help with that, or have ideas
> > on how, please go ahead.
>
> I am joining the calls for that exact reason. So we can discuss that there
> I guess?

Very unfortunately this patch was silently applied to a tree and is
now in mainline, despite this discussion not being resolved.

I will send a revert until this can be resolved.

Regards,
Simon