In December 2019 we had pleasure to meet with Daniel Kiper GRUB maintainer, Software Developer, TrenchBoot technical leader at Oracle and conference speaker. Since we are working with GRUB2 across many commercial projects and our customers leverage it often we decided to take advantage of this relation and organize a meeting during which 3mdeb engineers and team leaders presented GRUB related topics. This blog post summarizes the discussion and creates reference point for interested parties:
Somehow ‘minisummit’ name survived since no one offered better title after publishing Qubes OS and 3mdeb ‘minisummit’ 2019 blog post.
What we discussed during our meeting with Daniel:
- Intro - introduction, motivation and agenda presented by me
- Redundant GRUB2 env file - Maciej Pijanowski presented redundant GRUB2 environment file, feature needed for power fail safe upgrades, similar mechanism was already implemented in U-Boot
- TPM support in GRUB2 for legacy boot mode
- update on talk presented on LPC 2019 in which Michał Żygowski highlights changes made in TrustedGRUB2 and build base for discussion about merging changes to GRUB2 mainline
- GRUB2 security features overview - general talk overviewing security features in GRUB2 presented by me
- Python 3 support in GRUB2 - talk in which Michał Żygowski presents how we used Python with GRUB2 and starts a discussion about making Python 3 first class citizen in GRUB2 bootloader
- AMD TrenchBoot support in GRUB2 - final talk in which I present status of AMD TrenchBoot support, implementation and possible paths to upstream related code
We also published all videos on our YouTube channel.
Redundant GRUB2 env file
The key problem mentioned during presentations are lack of integrity check in the current implementation of GRUB2, what means any corruption of environment file in GRUB may lead to boot failure that can be recovered only by manual intervention. Since in most cases GRUB2 is used as a bootloader in general purpose computers this was not a problem. Even if in reality it happens people know how to recover either through GRUB command line or live booting Linux distro.
Unfortunately, if trying to use GRUB2 in the embedded environment there is a risk of ending up with an unbootable product, which of course leads to all other problems on the vendor’s side.
We all agreed that this feature is needed in GRUB2 and if 3mdeb will get correct sponsoring from vendors using GRUB2 such support will be developed and contributed. On the other hand, some vendors assume that file corruption is so rare that they don’t want to spend additional engineering hours on such feature.
Before such support is available, there are still methods to mitigate the impact of the problem for which reference you can find in the presentation.
TPM support in GRUB2 for legacy boot mode
The key problem here is the lack of boot process integrity in non-UEFI systems. Since GRUB2 is in boot process chain also in legacy systems we wanted to continue discussion started at LPC 2019. General conclusion on Linux Plumbers Conference was that approach presented in “TCG PC Client Specific Implementation Specification for Conventional BIOS” is preferred. What in short means that BIOS should expose legacy interrupt INT 1Ah to legacy operating system for measurement purposes.
There are couple problems with that:
- this approach is complaint with TPM1.2, no support for TPM2.0
- GRUB2 is neither consumer or producer of INT 1Ah
- at this point GRUB2 supports TPM only through UEFI API
- GRUB2 should preserve interrupt handlers
Because the whole problem seems to not be so important (it affects the limited amount of hardware) we are thinking about the simplest way of having correct results. It seems that solution that has most sense is leveraging Rhode and Schwarz TrustedGRUB2 implementation.
The key assumption of Rhode and Schwarz project is that BIOS measures MBR and
interrupts for measuring bootloader kernel (diskboot.img
) is called from MBR.
In that way measurement chain continues. Of course something has to install
those interrupts, in this case it is SeaBIOS, which implemented previously
mentioned TCG specification.
Addressing a lack of support for TPM2.0 through INT 1Ah seems not to be a big issue since we can hide hardware version behind SeaBIOS implementation. Bootloader just need to extend PCRs and do not use any sophisticated TPM features. This approach implies that our stack needs SeaBIOS, since we were not able to find anyone who can confirm existence of legacy BIOS or CSM with INT 1Ah support. SeaBIOS of course limit us to coreboot and QEMU based platforms.
Based on previous assumption we have to admit that GRUB2 can be only consumer of INT 1Ah API. The simplest solution for that would be to port support from TrustedGRUB2.
In previously described cases, GRUB2 is the second stage bootloader. If we would like to have GRUB2 as first stage bootloader GRUB2 should be producer of INT 1Ah API. That implies targets like:
- *BSD booted from GRUB2 on top of Legacy BIOS/CSM without INT 1Ah
- any system booted from GRUB2 on top of coreboot - here GRUB2 sill can be a consumer if coreboot would install INT 1Ah API what would be a little bit bizarre in light of already having that in SeaBIOS
Moreover being a producer means providing a driver for TPM communication. According to minisummit discussion producer scenario is technically possible to implement things but may be economically not feasible.
GRUB2 security features overview
During this presentation I complained a little bit about the usability of various GRUB2 security features and suggested possible extensions that could help improve the current state of GRUB2 security. Definitely documentation could be better, but this requires time and community engagement.
Most of the talk focused on how those features help coreboot based platforms. The overall adoption of security features is slow, mostly because of a lack of integration across the system components.
Python 3 support in GRUB2
Whole topic started with our talk about “CHIPSEC as coreboot payload” from OSFC2018.
Our key concern here was lack of pre-OS firmware validation environment, there were some attempts in the past like UEFI support for MicroPython, but recently there seem to be no progress in the area.
We all agreed that support for Python in GRUB would be interesting from various points of view. One area that would be problematic is ownership and maintainership since Python is evolving. Keeping it up to date would be required and probably the cost time of developers. Other mentioned alternative was MicroPython since it would resolve the same problem but probably with less maintainer overhead.
AMD TrenchBoot support in GRUB2
From this presentation you can learn how to run most recent code and test it. Most of the presentation and discussion was about internals how things should be implemented and if what we did is acceptable.
General conclusion was that we have to implement DRTM specific relocator as it was done for other boot options.
This session was more like live code review, but we took the chance to discuss each aspect of TrenchBoot support for AMD in GRUB project. Most of the mentioned problems were already implemented and sent for review to grub-devel mailing list.
In long run we plan to provide frequent updates related to AMD support in TrenchBoot project. Next update is planned to FOSDEM 2020.
Summary
It was pleasure to whole 3mdeb to meet with Daniel and understand his perspective and vision of GRUB2 from community and commercial perspective. We definitely see possible synergies and hope to find resources to support GRUB bootloader.
Some may ask if GRUB2 existence is justifiable in light of Linux booting directly from UEFI and growth of projects like LinuxBoot. Our opinion is that we choose solutions that fits best for outcome we want to achieve. Amount of work, expertise and features set accumulated over 25 years of development cannot be ignores and largely simplify our lives in some cases. It may mean that it is faster and easier to use GRUB2 in some cases and in other leverage LinuxBoot or other solution.
If you need bootloader support or you think we can help in improving the
security of your firmware or you looking for someone who can boost your product
by leveraging advanced features of used hardware platform, feel free to
book a call with us or
drop us email to contact<at>3mdeb<dot>com
. If you are interested in similar
content feel free to sign up for our newsletter