Skip to content

tools: verify signed root images and isolate device builds - #192

Closed
Gr-linkk wants to merge 1 commit into
derekross:mainfrom
Gr-linkk:fix/release-integrity
Closed

Gr-linkk wants to merge 1 commit into
derekross:mainfrom
Gr-linkk:fix/release-integrity

Conversation

@Gr-linkk

Copy link
Copy Markdown
Contributor

What and why

A signed bundle could still accept a changed root image when its unsigned .sha256 sidecar was changed too. The installer now verifies the root image's filename and digest against the authenticated bundle manifest, and rejects an external --images directory in bundle mode.

Rootfs builds also shared output names and a work directory across devices. Builds now use device-qualified filenames, a private work directory and a lock for each device/version. Compression and the smoke check finish before publishing the output, so a failed build preserves the previous image. Bundle creation selects a matching device and checks its image digest and kernel/modules release.

Device and build

  • Phone: none: laptop build and installation tools only.
  • Kernel release: not applicable.
  • Packages changed: none.

Checks (no phone needed)

  • tools/pr-check upstream/main fix/release-integrity passes.
  • tests/all passes.
  • Package tests/version bump: not applicable; no package files changed.
  • Commit has an area prefix, rationale and an Assisted-by: trailer.

The installer tests use a real temporary signing key and signed manifests. They reject modified image bytes with an updated sidecar, a wrong image filename, an unsigned manifest and an external images directory.

The rootfs tests run the build script with a Docker stand-in: concurrent builds for different devices succeed, a second build of the same device/version is refused, failed builds leave the previous output intact, and bundles select the matching device.

On the phone

Not applicable: no phone execution or flashing was needed for these changes.

The local suite skipped checks requiring an Arch Docker image, AOSP mkbootimg, dtc or a built root image. I did not build a complete ARM rootfs or perform a real installation/flash.

  • No change to allowed partitions, storage guards or register access.

A signed bundle verified its boot images but trusted the adjacent root
image's unsigned checksum. Bind the selected root image's name and bytes
to the signed MANIFEST, and refuse an external boot-image override in
bundle mode. Preserve the explicit unsigned development flow.

Include the device in rootfs output names, use private temporary build
directories and lock each device/version. Check the staged image before
replacing a previous build. Installer and bundle selection use the root
manifest's device, including images with older filenames.

Verified with real GPG signatures and modified ext4 fixtures in
tests/install-phone, concurrent/failing container stand-ins in
tests/build-rootfs, and tests/all. No physical flash or ARM image build.

Assisted-by: Codex <noreply@openai.com>
@Gr-linkk Gr-linkk closed this Oct 30, 2026
@Gr-linkk Gr-linkk reopened this Oct 30, 2026
@derekross

Copy link
Copy Markdown
Owner

Thank you, this closes a real hole: the unsigned .sha256 could vouch for a swapped root image. The tests with a temporary signing key and the stand-in container are excellent. It conflicted with #196 on the tests/all line, so I kept both new tests, and updated two runbooks to the per-device image names. It's on main authored by you (the PR shows as closed because the commit went in directly).

@derekross derekross closed this in fedb9d7 Oct 31, 2026
h4rkl added a commit to h4rkl/omandroid that referenced this pull request Nov 1, 2026
Keep both guards in tools/make-bundle: main's root image checks (derekross#192),
then the initramfs key check. tests/make-bundle-keys gives its fixture
root image the MANIFEST that main now requires, and tests/all keeps
main's list with tests/make-bundle-keys added.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
aninmucek added a commit to aninmucek/omarchy-mobile that referenced this pull request Nov 3, 2026
From the review of derekross#414:

- The status line names derekross/linux-omarchy-mobile#192 and its
  commits. M1.3's kernel commit, ce160abd9, is derekross#192's 1253914a9 with
  the same files. M1.4's -dirty run is derekross#192's 6a2999804: the kept DTS
  is blob 381ad1e0 and the flashed DTB has sha256 9a5d92a5, as derekross#192's.
- Risks: every unmarked boot uses one of slot B's tries. If stock boots,
  decline any factory reset and take a fresh backup.
- M1.2 lists the fastboot getvar values instead of a local helper, and
  the two rollbacks pin the serial.
- NOTES.md: the COMPATIBLE and DTB rows point at derekross#192.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
derekross pushed a commit that referenced this pull request Nov 5, 2026
The Pixel 8a (akita) boots our kernel from slot B with USB networking,
SSH and the framebuffer console; the board file is
linux-omarchy-mobile#192. BRINGUP-M1.md holds the runbook in husky's
format with tests 1-5 and the image checksums; kernel.config adds the
lockup detectors the board file's bootargs rely on; device.conf gets
SIMPLEFB_TRIG_CON="0x19470030 0x3061", from TRIG_CON read once (0x3070,
husky's value) before the write went in. NOTES records the booted tree,
the command line and the display registers, and two findings: this
bootloader keeps VENDOR_CMDLINE (so `splash` takes effect), and Google's
panel driver gives a size too large for the panel, so the board sets
none yet. Only slot B's five images were written, slot B was never
marked, and the -dirty M1.4 kernel's DTS and DTB match #192's.
derekross added a commit that referenced this pull request Nov 7, 2026
The matrix re-read against what merged and ran on hardware since 2026-11-02:
- Pixel 7: Bluetooth and the torch move to testing (kernel #196, #197, #426); fingerprint moves to WIP (kernel #207; the package side, #458, is in review); AVB footers and reboot=warm noted.
- Pixel 8a: a new row, from first light and UFS behind the guard (kernel #192, #206; #414).
- Pixel 6 Pro: GPU scaling and cooling (kernel #189) and UFS suspend (kernel #188) merged.
- Pixel 9a: GPU levels merged through kernel #96; GNSS's moving-car session (#114) and the link driver (kernel #185).
- Pixel 7a: Bluetooth's device tree merged (kernel #98).
The Pixel 8 Pro's row is left to #472.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
…#192)

In a signed release bundle the root image is now bound to the signed MANIFEST (its name and SHA256 digest), and the installer checks the selected image against that record before any fastboot call, so the unsigned .sha256 can no longer vouch for a swapped image. Builds are locked per device and version and published only after they are checked; old bundles' root image lines still parse. Nothing about what is written, the allowlist or the order changes. Maintainer edits: tests/all keeps derekross#196's new test too, and two runbooks use the per-device image names.

Closes derekross#192

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
The Pixel 8a (akita) boots our kernel from slot B with USB networking,
SSH and the framebuffer console; the board file is
linux-omarchy-mobile#192. BRINGUP-M1.md holds the runbook in husky's
format with tests 1-5 and the image checksums; kernel.config adds the
lockup detectors the board file's bootargs rely on; device.conf gets
SIMPLEFB_TRIG_CON="0x19470030 0x3061", from TRIG_CON read once (0x3070,
husky's value) before the write went in. NOTES records the booted tree,
the command line and the display registers, and two findings: this
bootloader keeps VENDOR_CMDLINE (so `splash` takes effect), and Google's
panel driver gives a size too large for the panel, so the board sets
none yet. Only slot B's five images were written, slot B was never
marked, and the -dirty M1.4 kernel's DTS and DTB match derekross#192's.
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
The matrix re-read against what merged and ran on hardware since 2026-11-02:
- Pixel 7: Bluetooth and the torch move to testing (kernel derekross#196, derekross#197, derekross#426); fingerprint moves to WIP (kernel derekross#207; the package side, derekross#458, is in review); AVB footers and reboot=warm noted.
- Pixel 8a: a new row, from first light and UFS behind the guard (kernel derekross#192, derekross#206; derekross#414).
- Pixel 6 Pro: GPU scaling and cooling (kernel derekross#189) and UFS suspend (kernel derekross#188) merged.
- Pixel 9a: GPU levels merged through kernel derekross#96; GNSS's moving-car session (derekross#114) and the link driver (kernel derekross#185).
- Pixel 7a: Bluetooth's device tree merged (kernel derekross#98).
The Pixel 8 Pro's row is left to derekross#472.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants