Repository navigation
fix: verify SHA-256 on bootc-installer bundle in iso/live/src/install-flatpaks.sh, drop tuna-os fallback - #48
Conversation
…-org fallback The ISO live build downloads org.bootcinstaller.Installer.flatpak at a pinned version and imports it with --no-gpg-verify, so nothing verified the payload. It also fell back to fetching the same version tag from tuna-os/tuna-installer, a different GitHub org, letting a release there substitute the installer shipped in every Utah ISO. Pin INSTALLER_SHA256 beside INSTALLER_VERSION (same pattern as UUPD_SHA256 in the Containerfile), verify the bundle with sha256sum --check before build-import-bundle, and remove the cross-org fallback so the download only ever comes from projectbluefin/bootc-installer. Fixes #47 Signed-off-by: sec-check[bot] <sec-check[bot]@users.noreply.github.com>
hanthor
left a comment
There was a problem hiding this comment.
Digest verified correct against the real release asset
I downloaded the actual bundle and checked the pin in this PR:
$ curl -sSL -o bi.flatpak \
https://github.com/projectbluefin/bootc-installer/releases/download/v3.0.16/org.bootcinstaller.Installer.flatpak
$ ls -l bi.flatpak
-rw-r--r-- 1 root root 92530152 bi.flatpak
$ sha256sum bi.flatpak
6d68445965bf03fd628fcc9e856b162939b5f87bf4532f62725cf0e114c7eea7 bi.flatpakMatches INSTALLER_SHA256 exactly. The v3.0.16 tag exists in that repo and nowhere else relevant:
$ git ls-remote --tags https://github.com/projectbluefin/bootc-installer | grep -v '\^{}' \
| sed 's#.*refs/tags/##' | sort -V | tail -3
v3.0.13 v3.0.14 v3.0.16The security argument is sound too. flatpak build-import-bundle on a --no-gpg-verify remote means the version pin was the entire trust story, and the tuna-os/tuna-installer fallback on the same version tag in a different org was a real substitution path into every Utah live ISO. Removing it and verifying the digest before import is the right shape, and mirroring the existing UUPD_SHA256 pattern keeps it consistent with how the Containerfile already does this.
The patch applies cleanly to current main (git apply --check clean), even though the PR is based on the older e993c550.
One conflict to resolve before this merges
#65 touches the same six lines and goes the opposite direction. It repoints INSTALLER_REPO from projectbluefin/bootc-installer to tuna-os/bootc-installer and removes the fallback, but adds no digest check. That target does not publish v3.0.16:
$ curl -sSo /dev/null -w "%{http_code}\n" -L \
https://github.com/tuna-os/bootc-installer/releases/download/v3.0.16/org.bootcinstaller.Installer.flatpak
404so #65 as written breaks just iso outright. I've flagged that there. Both PRs agree the cross-org fallback must go, which is the substance of #50/#47; they disagree on which org is canonical. This PR's direction is the one supported by what is actually published.
Follow-ups
The check-download-integrity.py ${VAR} indirection gap tracked in #47 is the right follow-up and correctly scoped out. Worth noting it is load-bearing: that guard is exactly what should have caught #65's 404, and it could not, because it cannot resolve ${INSTALLER_REPO} / ${INSTALLER_VERSION}. Might be worth bumping its priority.
Also note this PR carries the hold label and is red on Build Utah — but that red is the pre-existing /usr/local/libexec breakage on main (see #55/#56), several steps before the ISO path, not anything from this change. iso/live/src/install-flatpaks.sh runs during just iso, which Build Utah does not invoke at all, so CI here cannot exercise this change either way. The digest verification above is the strongest available evidence and it checks out.
This one looks ready to me once #65's direction is settled and the hold is cleared.
Generated by Claude Code
…rojectbluefin#48 Review on projectbluefin#65 found that tuna-os/bootc-installer does not publish the pinned v3.0.16 release (it tops out at v3.0.14 on semver tags), so repointing INSTALLER_REPO there while keeping the v3.0.16 pin breaks `just iso`: the curl 404s and, with the fallback removed, nothing catches it under set -euo pipefail. Verified v3.0.16 is only served from projectbluefin/bootc-installer. projectbluefin#48 already carries the fallback removal plus a sha256sum digest pin verified against the real published asset, so this PR reduces to its docs/Justfile clarifications and keeps INSTALLER_REPO pointed at projectbluefin/bootc-installer with a comment recording why. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LNgJ8VMdoqzGequbxsP2Kc
…ifact Review on projectbluefin#65 established that repointing at tuna-os/bootc-installer while keeping the v3.0.16 pin breaks the ISO build -- that tag exists only in projectbluefin/bootc-installer, and `curl --fail` aborts the script under set -euo pipefail. f4964ec reverted the repoint in response. The comment that commit left behind is accurate but incomplete, and the missing half is the part that matters for whoever resolves projectbluefin#50: it reads as "tuna-os does not publish this", when in fact tuna-os publishes the same org.bootcinstaller.Installer.flatpak asset -- just under date-based tags rather than semver. Verified both directions: v3.0.16 projectbluefin 200, tuna-os 404 v2026.09.14-9a9a913 tuna-os 200 v2026.09.08-2684a1d tuna-os 200 v3.0.14 (tuna-os) 404 So the repoint projectbluefin#50 asks for is achievable; it is a versioning-scheme change (INSTALLER_REPO and INSTALLER_VERSION have to move together), not a blocked one. Left as a comment rather than acted on because projectbluefin#48 is open against these same lines proposing the opposite direction, and which org is canonical is the open question in projectbluefin#50. Refs projectbluefin#50 Signed-off-by: Danathar <Danathar@users.noreply.github.com>
CI note: the failing
|
hanthor
left a comment
There was a problem hiding this comment.
Verified the pin rather than trusting it: fetched https://github.com/projectbluefin/bootc-installer/releases/download/v3.0.16/org.bootcinstaller.Installer.flatpak and it hashes to 6d68445965bf03fd628fcc9e856b162939b5f87bf4532f62725cf0e114c7eea7, the value this PR commits. Dropping the tuna-os fallback is right for the reason given — a same-named tag in another org must not be able to substitute the installer. Merges clean on today's main.
Generated by Claude Code
Security Fix
Cluster:
iso/live/src/install-flatpaks.shonly (live-ISO flatpak bake-in). Disjoint from open PRs utah#40/#44 (scripts/flavors.py/ flavor matrix).install-flatpaks.shdownloadedorg.bootcinstaller.Installer.flatpakat a pinned version but verified nothing beforeflatpak build-import-bundlewith a--no-gpg-verifyremote. On download failure it fell back to the same version tag intuna-os/tuna-installer, a different GitHub org — a substitution path into every Utah live ISO.This PR:
INSTALLER_SHA256(v3.0.16 bundle, digest computed from the upstream release asset) besideINSTALLER_VERSION, mirroring theUUPD_SHA256pattern in the Containerfilesha256sum --check --strictbefore importFALLBACK_REPOfallbackFollow-up (not in this PR): extend
scripts/check-download-integrity.pyto resolve simple${VAR}indirection so extension-hidden downloads cannot evade the guard — tracked in #47.Fixes #47
Filed by sec-check agent (ACMM L4/L5 — hold-gated mode). Hold-gated: human review required.
— hive: agent=sec-check backend=copilot model=kimi-k3