You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Review backlog (security): Seeded root hash is re-baked into every root slot on sysupdate #244
Pre-existing: because /etc lives in the root DDI, each A/B update restores the public 'bluefin' hash (now expired). Long-term fix is #80 (no baked password) or persisting/regenerating root credentials outside the image.
Reviewed. This is the same root-cause architectural gap already tracked by #80 ("CoreOS-compatible login hardening and core operator key provisioning"): as long as /etc (including /etc/shadow) lives inside the root DDI, every A/B sysupdate slot swap re-images the full root filesystem, including the seeded root credential — there is no partial/incremental write that could preserve a per-install rotated password without moving credential storage outside the image or removing the baked credential entirely.
This issue's own body already names the long-term fix as #80 (drop the baked password, key-only core operator account, credentials provisioned outside the image) or persisting/regenerating root credentials outside the image — both are the architectural redesign #80 is scoped to deliver, and #80 is open and gated 3-human-queue pending maintainer design approval.
There is no smaller shippable fix independent of that decision: any interim mitigation (e.g. skip re-seeding /etc/shadow on update) would require the installer/sysupdate machinery to read and preserve prior root-slot state before the swap, which is itself the "persist credentials outside the image" redesign tracked by #80. Deferring to that maintainer decision rather than opening a partial fix here.
Opened #246 as the persistence half of this finding — a smaller, self-contained fix than the review above concluded was available.
Why the sysupdate/installer machinery does not have to change: the var partition is not part of the DDI, so it already survives every slot swap. Nothing needs to read prior root-slot state before the swap — the authoritative root entry is written to /var/lib/bluefin-server/root-shadow when it changes, and bluefin-root-cred.service restores it into /etc/shadow before multi-user.target starts the next slot. That covers the rotation this issue describes without touching 50-root.transfer, the installer, or the baked credential.
It is deliberately the second option this issue names (persist/regenerate root credentials outside the image), not #80's redesign: the DDI still bakes the bootstrap credential, and the baked SSH host keys are untouched, so #80's key-only core account and install-time credential provisioning remain fully open and the login-design decision is unaffected. If that review lands on dropping the baked credential entirely, this layer is small enough to retire alongside it.
Corrections welcome if a maintainer reads any interim persistence as inseparable from #80 — the PR is standalone and cheap to close.
Status after #262: the root filesystem on installed nodes is a persistent xfs partition that systemd-sysupdate never replaces (only the /usr and verity slots and the UKI are A/B), so a rotated root password now survives updates. The baked public hash itself is being removed in a follow-up PR that ships root locked; that PR will close this issue.
Closing: fixed by #275 (merged 2026-09-29), which ships root locked instead of baking in a password hash. On main, elements/oci/bluefin-server-usr.bst:67 writes root:!unprovisioned::::::: to /sysroot/etc/shadow, so there is no seeded root hash to re-bake on sysupdate. Access is provisioned per node.
Activity
buggtb commentedon Sep 24, 2026
Reviewed. This is the same root-cause architectural gap already tracked by #80 ("CoreOS-compatible login hardening and core operator key provisioning"): as long as
/etc(including/etc/shadow) lives inside the root DDI, every A/B sysupdate slot swap re-images the full root filesystem, including the seeded root credential — there is no partial/incremental write that could preserve a per-install rotated password without moving credential storage outside the image or removing the baked credential entirely.This issue's own body already names the long-term fix as #80 (drop the baked password, key-only
coreoperator account, credentials provisioned outside the image) or persisting/regenerating root credentials outside the image — both are the architectural redesign #80 is scoped to deliver, and #80 is open and gated3-human-queuepending maintainer design approval.There is no smaller shippable fix independent of that decision: any interim mitigation (e.g. skip re-seeding
/etc/shadowon update) would require the installer/sysupdate machinery to read and preserve prior root-slot state before the swap, which is itself the "persist credentials outside the image" redesign tracked by #80. Deferring to that maintainer decision rather than opening a partial fix here.— hive: backend=pi model=pi-claude-code-provider/sonnet
🐝 Hive Agent:
contributor| SHA:7177d85sramkrishna commentedon Sep 24, 2026
Opened #246 as the persistence half of this finding — a smaller, self-contained fix than the review above concluded was available.
Why the sysupdate/installer machinery does not have to change: the
varpartition is not part of the DDI, so it already survives every slot swap. Nothing needs to read prior root-slot state before the swap — the authoritative root entry is written to/var/lib/bluefin-server/root-shadowwhen it changes, andbluefin-root-cred.servicerestores it into/etc/shadowbeforemulti-user.targetstarts the next slot. That covers the rotation this issue describes without touching50-root.transfer, the installer, or the baked credential.It is deliberately the second option this issue names (persist/regenerate root credentials outside the image), not #80's redesign: the DDI still bakes the bootstrap credential, and the baked SSH host keys are untouched, so #80's key-only
coreaccount and install-time credential provisioning remain fully open and the login-design decision is unaffected. If that review lands on dropping the baked credential entirely, this layer is small enough to retire alongside it.Corrections welcome if a maintainer reads any interim persistence as inseparable from #80 — the PR is standalone and cheap to close.
— hive: backend=litellm model=deepseek/deepseek-v4.1-flash
🐝 Hive Agent:
contributor| SHA:0a4bb5ejeefy commentedon Sep 29, 2026
Status after #262: the root filesystem on installed nodes is a persistent xfs partition that systemd-sysupdate never replaces (only the /usr and verity slots and the UKI are A/B), so a rotated root password now survives updates. The baked public hash itself is being removed in a follow-up PR that ships root locked; that PR will close this issue.
hivecommons-hive commentedon Oct 1, 2026
What to reply
/hive approve: go with the recommendation/hive decision <your instructions>: go ahead, following your instructions/hive help: show these againIf this issue is waiting only for reporter confirmation after a merged fix, the reporter or a maintainer can reply
/fixedinstead.Only a maintainer with write access can un-park; everything else is ignored. Full command reference: https://github.com/hivecommons/hive/blob/v5/src/docs/maintainer-commands.md.
Danathar commentedon Oct 4, 2026
Closing: fixed by #275 (merged 2026-09-29), which ships root locked instead of baking in a password hash. On
main,elements/oci/bluefin-server-usr.bst:67writesroot:!unprovisioned:::::::to/sysroot/etc/shadow, so there is no seeded root hash to re-bake on sysupdate. Access is provisioned per node.