Repository navigation
v0.1.1 — install to the eMMC, and a way back
v0.1.1 — install to the eMMC, and a way back
A maintenance release. The boot chain, the kernel and the sway session are the same as
v0.1.0; what changed is the part that installs the system onto internal storage, the
SSH host keys, and two new tools that back the running machine up and put it back.
What changed since v0.1.0
- Install to internal storage works, and it is now pmbootstrap's method.
scripts/lib-cgpt.shreproduces pmbootstrap's ChromeOS install path part by part: the root
partition is sizeddev_size - root_start - 33(the secondary GPT survives, as pmbootstrap
leaves it),partproberuns after thecgpt adds, themkfslines and the/etc/fstaboptions
are the onespmb/install/*.pybuilds, and a target with a mounted partition is refused instead
of silently unmounted. - Partitions are resolved by number, not by their order in
lsblk.lsblklists a disk's
partitions in the order the kernel handed out their minors, and thepartx -d/-around
reorders them: over 16 attaches of one loop device, 5 came back as2 3 1or3 1 2, which
formatted the boot partition as the root filesystem and wrote the kpart into the wrong partition.
The layout now looks each partition up by number and asserts the threepmOS_*labels, so a
wrong resolution dies instead of destroying a card. install-emmc.shrecreates the mount pointsrsyncdrops (/dev,/proc,/sys,/run,
/mnt,/media) and refuses to write the kernel partition unless/devexists and/sbin/init
resolves. Without them the installed system mounted its root and then died inrun-initwith
No init foundand no console to say why, which read as a firmware or kernel failure.- The SSH host keys are generated on the first boot that lacks them. Debian's
sshd-keygen.serviceis gated onConditionFirstBoot=yes, and an image written to a card or the
eMMC is always a copy of a tree that already generated/etc/machine-id, so the condition was
never true, the keys were never made, andssh.servicedied on every boot with
sshd: no hostkeys available -- exiting.rootfs-setup.shships a drop-in that keys on the host
keys being absent instead, which is the thing that actually matters. scripts/backup-system.sh(new) backs the running installation up into a directory: the
kernel partition it booted from, the GPT areas,/bootand the root filesystem as gzip tars with
owners, ACLs and xattrs preserved, plus a README andSHA256SUMS.scripts/restore-system.sh
(new) is the other half: it rebuilds the three partitions from such a backup with the original
filesystem UUIDs, so the restored machine boots with its own fstab and command line.install-kpart.sh's chroot guard actually fires (stat -Land the corrected comparison, with
both sides read before refusing), andCONFIRMsurvives thesudore-exec in both
install-emmc.shandinstall-kpart.sh.build-rootfs.shships an empty/etc/machine-idand no host keys, so every card and every
install gets its own identity instead of sharing one journal and one SSH identity.scripts/00-deps.shinstalls the host build dependencies fromapkorapt, so the
toolchain is installable from the Debian system this port itself runs.
Validation
The hardware evidence this port publishes lives in reports/, taken with
scripts/check-system.sh on the reference Hayato. The kernel, the boot chain and the session are
unchanged from v0.1.0 — whose report of the booted image is clean (51 OK, 2 WARN, the
two WARNs being the upstream sbs-8-000b fuel-gauge noise and the world regulatory domain) — so
what this release changes was verified on the hardware and on loop devices instead:
- the eMMC install reaches the desktop: the running card installed onto
/dev/mmcblk0boots by
picking internal disk in the firmware menu, the same greetd prompt as the card; - the layout and its refusals were rehearsed on a loop device: 16 of 16 attaches resolve the
partitions correctly (including the 5 that came back reordered), and the geometry, partition
types, labels andcgptflags match pmbootstrap's; scripts/backup-system.shandscripts/restore-system.shwere rehearsed on a loop device over a
sparse copy of the original disk's geometry: the GPT, the kernel partition and the file count
come out identical andtar -dagrees with both archives.
Assets
| asset | what it is |
|---|---|
google-asurada.img.xz |
4 GiB disk image: the three ChromeOS GPT partitions (kernel, /boot, root), ready for dd |
google-asurada-rootfs.tar.zst |
the same root filesystem as a tarball (with /boot inside), for writing your own card with scripts/write-sd.sh |
google-asurada.vmlinuz.kpart |
the signed FIT kernel partition of that image; it embeds the image's root=UUID, so it belongs to this image and no other |
SHA256SUMS |
SHA-256 of the three assets |
sha256sum -c SHA256SUMS
xz -dc google-asurada.img.xz | sudo dd of=/dev/mmcblk1 bs=4M conv=fsyncOn the Chromebook: developer mode, crossystem dev_boot_usb=1 dev_boot_signed_only=0, insert the
card, turn it on and pick external disk in the firmware menu. It comes up silently to a greetd
prompt: user brunux, password debian, passwordless sudo — change the password before using it
for anything real. (scripts/install-emmc.sh then installs it to internal storage.)
What it does not carry
Unchanged from v0.1.0: the artifact is sanitized and asserted clean (scripts/lib-dist.sh). No
Wi-Fi profiles, no SSH host keys, no machine-id, no personal files — the machine identity and the
SSH host keys are generated on first boot.
Known limitations
- The image is 4 GiB. A larger card keeps the space after
p3unused; writing the tarball with
scripts/write-sd.shsizesp3to the whole card. - The tarball's
/etc/fstabis the unconfigured one the build leaves behind:
write-sd.shgenerates the real one. - Artifacts are not byte-reproducible yet (FIT timestamp, filesystem UUIDs).
- Battery percentage is the capacity reading only; the gauge reports
Unknown chemistry: OTC0and
spams the journal on upstream kernels — Nura's own log has the same lines.
Upstream
Vendored from pmaports commit 0e458a2417d6f5ed61c7b70621d5632e30136091 (2026-10-15).