Skip to content

controlplane: no role erases a volume's signed history, and runners read who wrote it - #25

Merged
aledbf merged 1 commit into
mainfrom
f6c/writers
Oct 3, 2026
Merged

aledbf merged 1 commit into
mainfrom
f6c/writers

Conversation

@aledbf

@aledbf aledbf commented Oct 3, 2026 •

Copy link
Copy Markdown
Contributor

F6c (spin F6, spec spin#259).

What changes

  • The boundary denies s3:DeleteObjectVersion on volumes/*/HEAD and writers/* for every role, with no condition.
    • Past object_lock_days, the lock no longer keeps a version. Erasing the newest HEADs of a volume nobody wrote since would be a rollback a restore believes.
    • This corrects the spec's limit 2. The control plane's role does hold DeleteObjectVersion, because Restore lifts a delete marker, which is a version. As you pointed out, the UI's delete and restore are the database's soft delete (trashed_until), and a purge crypto-shreds the volume. So nothing in spin needs to lift a marker on a HEAD or a writer's record.
  • The runner scope reads writers/*, to hold a HEAD to its writer, and writes none of it. Only the control plane says who a writer is.

spin's per-host session policy needs the same reads (spin#264). Without both, a restore on AWS is refused.

Tests

  • New run: a_volumes_signed_history_is_erased_by_nobody. It checks:
    • the Deny on HEAD and writers versions;
    • BypassGovernanceRetention denied;
    • the runner reads writers/*;
    • no runner statement writes there.
  • I planted the HEAD resource's removal from the boundary and the run failed ("a role can erase a version of a volume's HEAD"), then reverted it.
  • task lint passes: 42 module tests.

🤖 Generated with Claude Code


View with [code]smith Autofix with [code]smith
Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.

…ead who wrote it

spin F6:
- a restore takes, of every version of a volume's HEAD, the highest-sequence one its writer signed;
- it holds that HEAD to the writer's record under writers/.

The boundary now denies s3:DeleteObjectVersion on volumes/*/HEAD and writers/* for every role.
Past the bucket's lock, erasing the newest HEADs of a volume nobody wrote since would be a rollback
a restore believes. Nothing in spin lifts a delete marker there: a workspace's delete is the
database's soft delete until the volume is crypto-shredded.

The runner scope reads writers/* and writes none of it.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
@aledbf
aledbf merged commit 230fd47 into main Oct 3, 2026
2 checks passed
@aledbf
aledbf deleted the f6c/writers branch October 3, 2026 11:40
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.

1 participant