GNU Boot - News
GNU Boot November 2024 News
Item posted by GNUtoo <gnutoo> on Sat 23 Nov 2024 07:05:25 PM UTC.
A lot has changed since the two last news from the GNU Boot project.
GNU Boot install party in Paris the 7 and 8 December 2024
People involved in the GNU Boot project will be organizing a 100% free
software install party within a bigger event that also has a regular
install party. There will also be a presentation about 100% free
software in there. The event will be mainly in French.
More details are available in French and in English in the following
link:
https://lists.gnu.org/archive/html/help-guix/2024-11/msg00112.html
GNU Boot 0.1 RC4
Many changes were made since the RC3 and since then we fixed an
important bug that prevented Trisquel from booting (If during the
Trisquel installation you chose "LVM2" and didn't encrypt the
storage, GNU Boot images with GRUB would not find the Trisquel
installation).
Because of that we decided to do a new RC4 (release candidate 4)
and to publish new GNU Boot images.
There are still some work needed before doing a 0.1 release as we want
to make it easier for less technical users to install and use GNU
Boot, but more and more of the project structure are getting in place
(website, manual, automatic tests, guix, good development procedures,
enabling build on all distributions, etc) which then makes it easier
to contribute.
We also decided to use Guix for more of the software components
we build, and since this is a big change, we will need people to
help more with testing.
Nonfree software found again, no supported device affected.
The last announcement we made was "Nonfree software found in GNU Boot
releases again, many distros affected"[1].
Some people misunderstood it (maybe we could have been more clear):
the nonfree software that we found was code that GNU Boot didn't use,
so it was easy to remove and it didn't affect the supported devices in
any way.
Finding nonfree software in 100% free distribution is also common:
this is part of the work to ensure these distribution remains 100%
free.
The first time it happened in GNU Boot we publicized it to explain why
we were re-releasing some of the GNU Boot files as it could be very
scary if this happens without any public communication.
The second time we published a news about it mainly to help propagate
the information to the affected distributions and this is probably why
it was misunderstood: it was mainly targeted at GNU Boot users and
maintainers of the affected packages. We also contacted upstream and
some affected distributions directly as well but contacting everybody
takes a lot of time so having a news about it helps. At least Debian
and Trisquel fixed the issue but we still need to contact some
distributions.
After that, and probably thanks to the previous news, Leah Rowe
contacted us on one of the GNU Boot mailing lists[2] to notify us that
she also found additional similar nonfree software in GNU Boot.
So we confirmed that and promptly removed them and re-made again the
source release. And here again even if the work was delayed a bit,
this was fast to do and it doesn't affect the supported devices in any
way.
But we also need help contacting distributions again because one of
the issue she found is very serious because it affects many
distributions and also important devices that GNU Boot doesn't
support.
The ARM trusted firmware ships a nonfree hdcp.bin binary in its source
code. ARM trusted firmware is a dependency of u-boot that is used to
support many ARM computers in other distributions (like Guix, Debian,
etc).
As contacting affected distributions is a tedious task, we also need
help to propagate the information and contact them especially because
we don't know if Leah intend or not to do that work (so far she didn't
reply when asked twice about it), so it's probably up to the GNU Boot
community as a whole (which also includes its maintainers and readers
of this news) to help here.
The details are in the commit 343515aee7ef34695ac45830fad419d9562f9c15
("coreboot: blobs.list: arm-trusted-firmware: Remove RK3399 hdcp.bin
firmware.") in the GNU Boot source code[3].
[1]https://savannah.gnu.org/news/?id=10684
[2]https://lists.gnu.org/archive/html/gnuboot-patches/2024-10/msg00028.html
[3]https://git.savannah.gnu.org/cgit/gnuboot.git/commit/?id=343515aee7ef34695ac45830fad419d9562f9c15
Website and documentation
Jordán (isf) has been contributing some Spanish translations of the
most important website pages (the landing, status and how to
contribute pages). This is important as it could help get more
contributors. These contributions also helped us improve the process
for accepting pseudonymous contributions and enabled us to fix issues.
The work on improving the website in general also continued. Many of
the website pages were reviewed and improved (there is a lot of work
there and mentioning it all would make the news way too long).
The website also now shows the git revision from which it is build and
we also helped the FSF fix some server configuration that created
issue with the deployment of the GNU Boot website (more details are in
the commit message[1]) by reporting the issue to them and testing the
fix.
Patches for making a manual are also being reviewed. While there isn't
much in the manual yet, it also enables to better organize the
documentation and it has the potential to make GNU Boot more
accessible to less technical people.
The next goals is to look how to merge part of the website inside the
manual and continue improving both the website and the manual.
[1]https://git.savannah.gnu.org/cgit/gnuboot.git/commit/?id=d1df672383f6eb8d4218fdef7fbe9ec5e41803e4
Authenticating GNU Boot source code
We now have the ability to verify the source code when downloading it
from git. This is important to avoid certain type of attacks and it
also enables to write code to automatically download, verify and build
the GNU Boot source code.
The source can be verified with the following command (it requires to
have Guix installed):
$ guix git authenticate $(git rev-parse HEAD) \
"E23C 26A5 DEEE C5FA 9CDD D57A 57BC 26A3 6871 16F6" \
-k origin/keyring
If the authentication works it will print a message like that:
guix git: successfully
authenticated commit 05c09293d9660ea1f26b5b705a089b466a0aa718
The 05c09293d9660ea1f26b5b705a089b466a0aa718 might be different in
your case.
The "E23C 26A5 DEEE C5FA 9CDD D57A 57BC 26A3 6871 16F6" part in the
command above is Adrien Bourmault (neox)'s GPG key.
How to use that will be documented more in depth in the upcoming GNU
Boot manual that is currently being reviewed. Its importance will also
be explained in more details for people not familiar with the security
issues it's meant to solve. Also note that we also welcome help for
reviewing patches.
Licensing
The GNU Boot source code has a complex history. It is based on the
last fully free software releases of Libreboot. And the Libreboot
source code history is very complex.
We found some missing authorship information in some of the files that
come from Libreboot and so we started such information from the
various git repositories that were used at some point by Libreboot or
some of the projects it was based on.
To help with this task we also added a page on the GNU Boot website
(https://www.gnu.org/software/gnuboot/web/docs/history/) to track the
status of the reconstruction of the missing authorship and to document
the GNU Boot source code history.
Upstream contributions and easier building of GNU Boot
GNU Boot is just a distribution and like most distributions, it tries
to collaborate with various upstream projects whenever possible.
Since GNU Boot relies on Guix, we improved the Guix documentation
directly to help people install Guix on Trisquel and Parabola. We also
helped Trisquel fix security issues in the Guix package by bug
reporting and testing fixes (some bugs still need to be fixed in
Parabola and Debian, and reporting issues upstream takes time).
Since we also advise to use PureOS or Trisquel to build GNU Boot we
also enabled people with Guix to produce PureOS or Trisquel chroots
with Debootstrap. This was done through contributions to Debootstrap,
and to the Guix Debootstrap package. We could then mention that in the
GNU Boot build documentation
(https://www.gnu.org/software/gnuboot/web/docs/build/) and added a
script (in contrib/start-guix-daemon.py) to support building GNU Boot
in chroots. However there are still issue with the build in chroots
that need to be fixed to producing all released files. Instructions on
how to do build in chroots is also lacking.
In addition we also added the ability to build GNU Boot with Trisquel
11 (aramo).
An apt-cacher-ng package was also contributed in Guix upstream as it
can then be used to speed-up one of the automatic tests used in GNU
Boot but the support for apt-cacher-ng was not integrated yet in GNU
Boot. Last year we also contributed a GRUB package in Guix but we
didn't have the occasion to use it yet. It will probably happen soon
though.
Building GNU Boot
How to build GNU Boot has changed a lot since GNU Boot 0.1 RC3.
Before Guix could only be optionally used to build the website.
In addition to that, Guix is now integrated in the build system so we
can now rely on Guix packages to build GNU Boot images. This also
means that you need to install Guix to build GNU Boot images.
We currently use Guix packages to build some tests. We also build some
installation utilities for the I945 ThinkPads (ThinkPad X60, X60s,
X60T and T60) but we don't have documentation for less technical
people yet on how to use them. We also would need help for testing
these computers as we have no idea if they still work fine or which
fully free distributions still work on them in practice.
We now also support the './configure' and 'make' commands to build GNU
Boot but not yet the 'make install' command as to work we would need
to adapt many of the scripts that are used during the build to be
compatible with that.
There is also less visible work that was done, like cleaning up a lot
of code, adding tests for code quality, documenting a bit the GNU Boot
source code structure, and so on.
Work on making GNU Boot reproducible also started. See
https://reproducible-builds.org/ or
https://en.wikipedia.org/wiki/Reproducible_builds for more detail on
the issue.
We took an extremely strict approach and put the checksum of some of
the things we build directly into GNU Boot and verify it the checksum
during the compilation. This enables us to automatically detect
issues without having to do anything.
We started to enable that for easy things, and we also added the
infrastructure to also use that in Guix packages as well by validating
one of the packages we use during automatic testing.
However at one point this guix package stopped being
reproducible. Since we wanted to keep that code (especially as it was
showing a good example of how to do it), we fixed the bug instead of
removing the test.
This then helped us detect a very subtle and interesting bug in one of
the components we use for automatic tests.
The bug could not be caught during testing because some time
information stored inside the FAT32 file system has a granularity of a
day, and since all the testing happened the same day, it was caught
only later on.
This bug was then fixed and the details are in the fix[1]. A bug
report was also opened upstream because bugs were found in diffoscope
along the way[2]. We still need to do some testing though to
understand if the bug is in diffoscope or one of the underlying
libraries (libguestfs) and then to report the remaining bugs to the
distributions we used during this work.
We also made it easier to update the checksum in the Guix package. If
you package software with Guix, this change is also a good example of
how not to break the '--without-tests' option when you override the
tests in the package you contribute. The commit message[3] and the
change have more details and references on all that.
[1]https://git.savannah.gnu.org/cgit/gnuboot.git/commit/?id=4c3de49fbb3b43940b43f8fdccc8e51ee7df8f46
[2]https://salsa.debian.org/reproducible-builds/diffoscope/-/issues/390
[3]https://git.savannah.gnu.org/cgit/gnuboot.git/commit/?id=40fcb94e2f7ab1df8d320f78311e623f801d8602
LVM2 bug
WodeShengli reported a very important bug[1]: GNU Boot images with
GRUB can't find LVM2 partitions if the partition itself is not
encrypted. For instance if you have LVM2 and no encryption at all or
if the disk is encrypted and that on top you have LVM2, GNU Boot will
not find the partition.
Since this is an extremely serious usability issue (because images
with GRUB are supposed to work out of the box) we spent time to fix
it.
The issue was that the GRUB configuration we ship hardcoded the name
of the LVM volumes to try to boot from. Fixing it required to be able
loop over all the partitions being found, but we found no command to
do that in GRUB (which is probably why the LVM partition names were
hardcoded in the first place).
So we started adding GRUB command options to do that but while the
code worked fine, it didn't integrate in GRUB well. So we contacted
GRUB looking for help as we would have needed to upstream our command
option in GRUB anyway.
And we were told that GRUB already had a way to do what we were
looking for so we used that to fix the issue.
We also added tests that automatically download the Trisquel installer
and installs Trisquel with LVM2 and test if GNU Boot can boot the new
Trisquel installation[2].
While this test is skipped for 32bit computers, it is still good to
have as some people will run it. The test also paves the way to add
more tests that would enable us to improve further the GRUB
configuration without breaking the boot.
[1]https://savannah.gnu.org/bugs/index.php?65663
[2]https://git.savannah.gnu.org/cgit/gnuboot.git/commit/?id=860b00bf1e798d86c8bb2a70d77633599dfa1da2
[3]https://git.savannah.gnu.org/cgit/gnuboot.git/commit/?id=9cc02ddde1e164fabfbddc8bbd3832ef9468d92d
Powered by Savane 3.14-8aba.
Corresponding source code