title: Canoeboot vs GNU Boot x-toc-enable: true ...
If you want to understand the beef Canoeboot has with GNU Boot, please read the about page, the Libreboot Binary Blob Reduction Policy and the GNU Boot article on libreboot.org - all of that energy has fuelled the creation of this page, and in fact the entire Canoeboot project. The main purpose of Canoeboot is to be technically superior to GNU Boot, while complying with all of its policies by providing ROM images and source tarballs adhering to the GNU Free System Distribution Guidelines, even though GNU FSDG is deeply flawed; this page describes precisely how that has been achieved, and it will be updated over time as Canoeboot continues to develop.
tl;dr as of 26 October 2023, Canoeboot code is about 1 year ahead of GNU Boot in terms of development, and about 2 years ahead in terms of documentation. Read on, if you want details. GNU Boot is also known as gnuboot.
Canoeboot and GNU Boot are both forks of Libreboot, designed to comply with the GNU Free System Distribution Guidelines. This page is maintained, to show the differences between these two projects.
This current version of the page pertains to Canoeboot 20231026 versus GNU Boot 0.1 RC. You can find GNU Boot ("gnuboot") on the GNU Savannah website.
You can also read the Canoeboot 20231101 changelog; that release only came out 7 days after the 20231026 release, so the rest of this page is more or less accurate, when combined with the facts on the 20231101 announcement.
This fact is very important; nonGeNUine Boot's 20230717 changelog is relative to Libreboot 20220710, and Canoeboot 20231026's changelog is relative to nonGeNUine Boot 20230717's changelog.
Therefore, this page will analyse differences in both projects, at these two
points. First, let's analyse GNU Boot, with the tag reset to 0.1-rc1
, which
corresponds to commit ID a64d284fd798d843133c9d7274bba17bd7837174
. Since GNU
Boot also contains the Libreboot history that it forked from, it contains the
Libreboot 20220710 release tag, so we can do this:
git log --graph --pretty=format:'%Cred%h%Creset %s %Creset' --abbrev-commit 20220710..0.1-rc1
Within the GNU Boot git repository, this would yield the following response:
* a64d284 .gitignore: order alphabetically
* 0df4fe5 GRUB: config from HDD/SSD: Add support for gnuboot_grub.cfg
* ce13d22 GRUB: Use GNU Boot logo
* 74b678c GRUB: Say the name GNUBoot in the grub menu
* eeddd2b build/dependencies: debian: adding python-is-python3 to build seabios properly
* 58b8e09 coreboot/fam15h: don't build ada toolchain for generic platforms
* f7c0fec coreboot/fam15h: update code base, deblob, unset CONFIG_STM (see bug #64535)
* de9297f coreboot/fam15h: fix crossgcc acpica build on newer hostcc
* c38348d coreboot/fam15h: fix for gcc/gnat building
* 0d77d99 coreboot/fam15h: fixing binutils not building properly
* b773079 coreboot/default, coreboot/fam15h: use GNU mirror for acpica
* bf17993 Continue Libreboot under the GNU project
And that's all. The fam15h-related fixed are actually merged from
the fsdg20230625
branch of Libreboot, made during July 2023. See:
https://browse.libreboot.org/lbmk.git/log/?h=fsdg20230625
The other patches are also merged (cherry-picked) from Libreboot! The above commit log is all that GNU Boot did, for their 0.1 RC1 release.
Therefore, to know the differences between Canoeboot and GNU Boot, I will copy all items first from the nonGeNUine Boot 20230717 change log, and then the Canoeboot 20231026 change log, but I will skip those entries that define features which GNU Boot already has.
On this day, GNU Boot's current commit ID (in the main
branch)
is 54c6ae497d49c233b654c171978baa77b90ffe17
from 12 October 2023. Most of
the changes since 0.1 RC1 up to that commit are just documentation changes,
and even still, only cherry-picking minor patches here and there that were
already done in Libreboot, in some cases years ago. It's worth noting that
the GNU Boot documentation is based on Libreboot documentation from late 2021
or at most, very early 2022.
I don't need to compare documentation, and it would take too long. Their documentation is 2 years out of date, what more is there to say?
Now, feature comparisons in the build systems:
Canoeboot has these boards fully supported, that GNU Boot currently lacks support for:
d862695f5f432b5c78dada5f16c293a4c3f9fce6
, 12 October 20231c13f8d85c7306213cd525308ee8973e5663a3f8
, 16 June 2021e58b870ff926415e23fc386af41ff81b2f588763
, 3 October 20231e1da7a963007d03a4e0e9a9e0ff17990bb1608d
, 24 August 20234459ed60cb1e0562bc5b40405e2b4b9bbf766d57
, 2 October 2023b2e8bd83647f664260120fdfc7d07cba694dd89e
, 17 November 20211c13f8d85c7306213cd525308ee8973e5663a3f8
, 16 June 2021f7564844f82b57078d601befadc438b5bc1fa01b
, 25 October 20211281e340ad1d90c0cc8e8d902bb34f1871eb48cf
, 24 September 2021As you can see, Canoeboot's revisions are a lot newer.
Canoeboot 20231026 contains a heavily patched version of GRUB, which contains argon2 support. This allows full decryption of LUKS2 volumes, without having to switch to different key derivation (PBKDF2) and without needing to use LUKS1. You can simply use the default LUKS2 setup provided by any distro, and it will work. More information is available about GRUB cryptomount, in the GNU+Linux guide - search on that page for LUKS2 or argon2.
GNU Boot completely lacks this feature, as of 26 October 2023. It can only support LUKS2 if the key derivation is downgraded to PBKDF2 (insecure, or to LUKS1 (also insecure).
For all intentions, the average user cannot have a fully encrypted system on
GNU Boot. They must leave /boot
unencrypted on GNU+Linux distros.
With Canoeboot, you can have encrypted /boot
very easily. This is a boon for
security, because it reduces the chance of someone tampering with your data
successfully, and combined with other steps,
can be used to reduce the risk of evil maid attacks (by making it infeasible).
Canoeboot can build firmware images for RP2040 and STM32 microcontrollers,
using pico-serprog
and stm32-vserprog
respectively. This can be used to
set up an SPI flasher of high quality, but these parts are low-cost.
GNU Boot does not support this feature, as of 26 October 2023.
There are 9 shell scripts in Canoeboot 20231026, versus about 50 in GNU Boot 0.1 RC1, because GNU Boot uses the pre-audit design; Libreboot used to have lots of very simple scripts, but ended up with a lot of code repetition.
The new lbmk design generalises all of the logic, doing away with the very hacky logic that existed in the old build system design.
The interface in Canoeboot's build system is much easier to use. For example, the commands are shorter, and easier to remember. See: cbmk maintenance manual tells you everything about the Canoeboot build system.
GNU Boot doesn't even have a maintenance manual, in their version. Their
documentation exists in the same repository as code, but their
version of docs/maintain/
does not actually contain any instructions,
at least as of commit ID a64d284fd7
on 17
August 2023.
Canoeboot has much better documentation, but this is obvious if you've been paying attention. As already stated: GNU Boot's documentation is horribly out of date, even relative to the version of Libreboot that they're using! (which itself is also horribly out of date)
Smaller doesn't mean worse; in fact, Canoeboot's build system is more efficient. It's about 1250 sloc (source lines of code), when counting shell scripts in the core of the build system. Libreboot, Canoeboot and GNU Boot build systems all use the same design, written in shell scripts.
1250 sloc in Canoeboot, versus 2111 in gnuboot; gnuboot however lacks many of the features and improvements that you're about to see below. The Canoeboot build system does several times as many things, in half the amount of code! The code is generally just more reliable, less error prone, and easier to work with, in Canoeboot. GNU Boot uses a very old version of the Libreboot build system design, from long before I started any massive audits. There have been three Libreboot build system audits in 2023, as of 26 October 2023.
Three audits, and Canoeboot has inherited the improvements of all of them. GNU Boot's design is based on the pre-audit lbmk codebase.
Build system / performance improvements in Canoeboot:
config/git/
--mtime
and option options in GNU Tar (if it is actually GNU Tar), when
creating Tar archives. This results in partially reproducible source archives,
and consistent hashes were seen in testing, but not between distros..git
within lbmk, for the build system itself, if
Git history was removed as in releases. This work around some build systems
like coreboot that use Git extensively, and are error-prone without it.-k
grub.cfg
, because it causes trouble in some
non-interactive setups where the user sees an errant message on the screen
and has to press enter. This fixes boot interruptions in some cases, allowing
normal use of the machine. The pager was initially enabled many years ago,
to make use of cat a bit easier in the GRUB shell, but the user can just
enable the pager themselves if they really want to.include/
, and main scripts in script/
,
called by the main build
scriptsetvars
function defined under include/err.sh
config/git
are now
concatenated, traversing recursively through the target directory; files first,
then directories in order, and for each directory, follow the same pattern
until all files are concatenated. This same logic is also used for patches.
This now enables use of subdirectories, in some config/patch directories.util/nvmutil
cp
to copy links.err
, or just something that belongs on the error
output (instead of standard output).-B
option in make commands./tmp
, because it might be a tmpfs with
little memory available; clone into tmp/gitclone
instead, within lbmk,
and mv
it to avoid unnecessary additional writes (mv
is much more efficient
than cp
, for this purpose).target.cfg
handling in vendor scripts, because they use
the concatenated config format instead (they always have).resources/scripts/
is now script/
,
and resources/
was renamed to config/
; ifd and gbe files were also moved
to config/ifd/
. Commands are now 1-argument instead of 2, for example
the ./build boot roms
command is now ./build roms
.read
command that is built into sh
, reading each line.
This is more efficient, and provides more robust handling on lines with
spaces in them.config/git
have
had certain depend
items removed, if a given project already defines it
under .gitmodules
(within its repository).target.cfg
files in multi-tree projects coreboot,
SeaBIOS and U-Boot. Unified to all such projects, under one script, and
with improved error handling../build dependencies x
commands still work
as root (they're the only commands available as root).list
command has been mostly unified, making it easier to tell (from
lbmk) what commands are available, without having to manually poke around
under script/
.-T0
flag is now used, universally, on xz commands. This makes xz
run
on multiple threads, greatly speeding up the creation of large tar archives.-j
in make commands, for multi-threading, but it relies
on nproc
to get thread count, so this only works if you have nproc
(you
probably don't, if you run BSD; BSD porting is still on TODO for Canoeboot).gitcheck
; now, a global git name/email config is always required.
The only behaviour (setting local config, and unsetting) was quite error-prone
under fault conditions, where cleanup may not have been provided, or when
execution was interrupted, resulting sometimes in accidentally committing
to lbmk.git
as author named lbmkplaceholder
.x_
or err
is used to provide such behaviour. This results in all
exits from lbmk being consolidated to err
, under fault conditions. - zero
exits are also consolidated, going only through the main script, which has its
own exit function called lbmk_exit
that provides TMPDIR
cleanup.err
function (and functions
that use it) inside include/err.sh
; there is also x_
which can be used
to run a command and exit automatically with non-zero status, useful because
it provides more verbose output than if you just relied on set -e
, and it
still works when a script does not use set -e
- however, it is not used
on all functions, because it works by executing $@
directly, which can break
depending on arguments. Therefore, some scripts just default to || err
for
providing breakage in scripts.elf/
vs bin/
, when it comes to
flashing coreboot ROM images; it tells them to use bin/
because those
images do contain payloads, whereas the ones under elf/
do not.elf/
, as are payloads, then they are joined separately by
the usual ROMs build script, and these cached ROMs contain many changes in
them that were previously handled by moverom
in the main ROM build script.
Under the new design, repetitive steps are avoided; payloads are inserted into
a copy of the cached ROMs under TMPDIR
, before being copied for keymaps
and small files; this eliminates delays caused by slow compression (LZMA is
always used, when inserting payloads). After crossgcc and the payloads are
compiled, the ROM with coreboot builds in under a minute, whereas it would
have previously taken several minutes on most Canoeboot-supported hardware.include/
and its functions are called by the main build script, which
provides a stub for this.script/
, inheriting the TMPDIR
variable set (and exported) by lbmk./tmp
handling; a universal TMPDIR
is set (environmental
variable) and exported to all child processes running lbmk scripts. On exit,
the main tmp directory is purged, cleaning all tmp directories under it.-e
option in sh
was
heavily relied upon to catch errors, but now errors are handled much more
verbosely. Many fault conditions previously did not make lbmk exit at all,
let alone with non-zero status, and zero status was sometimes being returned
under some edge cases that were tested. Error handling is more robust now.util/ich9utils
(containing ich9gen
) was removed, thus eliminating about
3000 source lines (of C code) from lbmk. The nvmutil
program, also provided
by and originating from the Canoeboot project, can already change GbE MAC
addresses. Coreboot's bincfg can generate ich9m descriptors, and ifdtool can
manipulate them; so the features provided by ich9utils were superfluous, since
they are available in other projects that we ship. We now ship pre-built
ifd/gbe configs on these machines, which can be modified or re-assembled
manually if you want to. This eliminates a moving part from Canoeboot, and
speeds up the build a little bit.elf
directory contains these, and
the existing bin
directory still holds the full ROM images (containing
payloads) when compiled.build/boot/roms_helper
script - the new style is much cleaner,
mandating that logic be top-down, with a main()
function defined; it's
basically inspired by the OpenBSD coding style for C programs, adapted to
shell scripts.grub.elf
is now used on all ROM
images. The grub.cfg
goes in GRUB memdisk now, but can be overridden by
inserting a grub.cfg
in CBFS; many behaviours are also controlled this way,
for example to change keymaps and other behaviours. This results in much
faster builds, because a different GRUB payload doesn't have to be added to
each new ROM image; such takes time, due to time-expensive LZMA compression.
This, plus the optimised set of GRUB modules, also makes GRUB itself load
much faster. All of the fat has been trimmed, though still quite a lot more
than a Crumb./tmp
at all, but now it's pretty reliable.cros
: Disable coreboot-related BL31 features. This fixes poweroff on gru
chromebooks. Patch courtesy of Alper Nebi Yasak.u-boot
: Increase EFI variable buffer size. This fixes an error where
Debian's signed shim allocates too many EFI variables to fit in the space
provided, breaking the boot process in Debian. Patch courtesy Alper Nebi Yasakutil/spkmodem-recv
: New utility, forked from GNU's implementation, then
re-written to use OpenBSD style(9) programming style instead of the
originally used GNU programming style, and it is uses
OpenBSD pledge()
when compiled on OpenBSD. Generally much cleaner coding
style, with better error handling than the original GNU version (it is forked
from coreboot, who forked it from GNU GRUB, with few changes made). This
is a receiving client for spkmodem, which is a method coreboot provides to
get a serial console via pulses on the PC speaker.extra.sh
directly from given coreboot tree. Unused
by any boards, but could allow expanding upon patching capabilities in lbmk
for specific mainboards, e.g. apply coreboot gerrit patches in a specific
order that is not easy to otherwise guarantee in more generalised logic of
the nonGeNUine Boot build system.util/e6400-flash-unlock
: New utility, that disables flashing protections
on Dell's own BIOS firmware, for Dell Latitude E6400. This enables nonGeNUine Boot
installation without disassembling the machine (external flashing equipment
is not required). Courtesy Nicholas Chin.cbutils
: New concept, which implements: build coreboot utilities like
cbfstool and include the binaries in a directory inside lbmk, to be re-used.
Previously, they would be compiled in-place within the coreboot build system,
often re-compiled needlessly, and the checks for whether a given util are
needed were very ad-hoc: now these checks are much more robust.
Very centralised approach, per coreboot tree, rather than selectively
compiling specific coreboot utilities, and makes the build system logic in
nonGeNUine Boot much cleaner.gru_bob
and gru_kevin
chromebooks, U-Boot is used
instead of Google's own depthcharge bootloader. It has been heavily
modified to avoid certain initialisation that is replaced by coreboot, in
such a way that U-Boot is mainly used as a bootloader providing UEFI for
compliant GNU+Linux distros and BSDs. Courtesy Alper Nebi Yasak.sh
(though, many dependencies still use GNU extensions, such as GNU
Make, so this portability is not directly useful yet, but a stepping stone.
nonGeNUine Boot eventually wants to be buildable on non-GNU, non-GNU/Linux systems,
e.g. BSD systems)nvmutil
- can randomise the MAC address on Intel GbE NICs, for
systems that use an Intel Flash Descriptorgrub.cfg
files on the HDD/SSD.So, in conclusion: this page is not trying to tell you why you should use Canoeboot; rather, it's just telling you that someone worse exists. Canoeboot and GNU Boot are both inherently flawed in their designs; both projects are completely inferior to the Libreboot project, for all the reasons laid out in the Binary Blob Reduction Policy - the Canoeboot project is provided, specifically, to prove the merits of that policy, by showing what Libreboot would have been by now if it continued adhering to GNU policy, instead of changing to its current Blob Reduction Policy.
Libreboot provides all of the same blob-free configurations as Canoeboot, when possible on any given mainboard, and that is the preference, but the FSF/GNU dogma states that the user must never run any proprietary software. This dogma is wrong; a more correct approach is to say that proprietary software is bad, because it restricts the user's freedom to study and modify the software; it removes their power to say no when the developer wants to change the program in a bad way, it leaves them at the mercy of that developer - the point is this:
Free software is good, and we should be promoting as much of it as possible. This means that a hardline no-blob approach like the policy implemented by Canoeboot (or GNU Boot for that matter), is entirely misguided, because it will only alienate those who just want some free software. Most people like the idea of software freedom but cannot go all the way yet; we should encourage people to make the right choices, but otherwise just help them in whatever way we can, meeting people where they're at right now.
And that is why Libreboot exists in the way that it does. Canoeboot serves as an example of what would happen in the best-case scenario if Libreboot never changed its policy. The best possible release of Canoeboot will still be missing a ton of boards and features from Libreboot. Indeed, the Canoeboot 20231026 and nonGeNUine Boot 20230717 both illustrate that quite nicely.