feat(supabase): V1-2.1 M4/M5 secure sheet access and review workflow

M4 20261028100000_sheet_access_hardening + M5 20261028110000_sheet_review_workflow
(applied together; no unprotected window). Implements contract §5-§14.

- Catalogue rights table + owner RPCs; has_mfa / has_aal2_session (live
  auth.sessions check); grant_active vs caller_has split; catalogue reader
  predicate; can_edit_sheet on the sheet_editor capability (F14).
- AR-1 B: aal2 required on every platform role change, target verified factor,
  race-free last-owner guard (F18/F19); service_role loses table privileges on
  the catalogue + profiles/profile_roles and EXECUTE on the role RPCs (F12/F17).
- Reviewed lifecycle: version provenance + review columns, content/freeze/
  author triggers, rounds/comments/acknowledgments, save/submit/withdraw/
  comment/acknowledge/decide, atomic publish_sheet_version(review_id) on the
  stored verdict, deprecate/archive/restore, species RPCs, equine working copy,
  uploads with the storage check (F7) + editor INSERT policy + 2 MiB bucket.
  F1-F3,F5,F6,F8,F9,F13,F15,F16 closed. Verdict is a fixture until M6 (V1-2.3).
- O22: SQL business mirror removed, declared status check kept; pgTAP 011 and
  the Rust generator's SQL parts deleted, ledger retargeted.
- Access layer sheets.ts (closed SheetFailure kinds); old client publish gate
  deleted; types regenerated; upload tool reads via get_sheet_asset_state.

Tested on the throwaway sandbox only; DEV untouched. pgTAP 850 (real stack) /
122 (PGlite) green; real-GoTrue AR-1 B + F18 race green; fresh install and
M3->M5 upgrade green; V1-1 regression green; rust:check, typecheck, lint,
build, types drift clean. BDR-084, LRN-095.
This commit is contained in:
bchanot committed 2026-10-28 15:52:49 +01:00
1 parent 9c13bae4bc
commit 77e5fa4cf5
33 files changed
+4613 -1617

No files matched your search

+13
View File
@@ -613,3 +613,16 @@ Alt: A (owner+admin only) rejected: leaves moderator/support weak path; C (accep
Review (fresh challenger FATAL(4), all integrated, no ratified decision changed): grant needs target with verified factor; private fns: non-null actor = auth.uid() + aal2, null actor only without JWT; advisory lock both RPCs; DEV precondition before M4 = ENVIRONMENTS §3.6 hardening (public default JWT secret + demo service key + 54321 LAN → aal2 forgeable) + seed owner password changed + role holders listed; recovery reordered (ban, password set with owner, unban, enrol).
Reserves: staff read policies (is_platform_admin) stay aal1, owner decision; stolen aal2 session residual; refused attempts not in audit table.
Status: ratified; contract only, no code/migration/DEV change; V1-2.1 not authorised.
## BDR-084 — V1-2.1 M4/M5 secure SQL foundations (sandbox) (2026-10-28)
- **Status**: implemented + tested on sandbox; DEV apply is an owner action (contract 16.11)
V1-2.1 M4/M5 built + tested on sandbox (not DEV). Migrations `20261028100000_sheet_access_hardening` (M4), `20261028110000_sheet_review_workflow` (M5); applied together (M4 removes Step-4 unsafe write paths, M5 adds the reviewed workflow; no unprotected window).
Key design choices:
- AR-1 B: `private.has_aal2_session()` = aal claim == `aal2` AND the token's `session_id` names a live `auth.sessions` row of the caller at aal2 (closes the stolen-token-after-downgrade window at once, not after jwt_expiry). GoTrue v2.197 probe confirmed: verify upgrades the SAME session to aal2, token carries `aal`+`session_id`+`amr`, exp-iat=3600.
- Predicates split by caller vs stored actor: `grant_active(uid,kind)` (no assurance) for stored approvers/acknowledgers; `caller_has(kind)` = `grant_active(auth.uid())` + `has_mfa()` for the session. `can_edit_sheet` redefined on `caller_has('sheet_editor')` + caller==auth.uid() (F14).
- F12/F17: `revoke all ... from service_role` on every catalogue table + profiles/profile_roles; role RPCs `EXECUTE` revoked from service_role. service_role stays a GoTrue-admin residual (§13), not a table writer.
- F18 last-owner race: one `pg_advisory_xact_lock` in both RPCs and both private writers, guard counts active owners under the lock. Cross-revoke resolves safely (one 200, one refused; one owner remains).
- O22 mirror removed: dropped `sheet_document_issues` + the 4 business-mirror helpers; kept `sheet_document_has_shape`, `check_sheet_version_document`, `unknown_assets`; added the declared status check `sheet_all_reviewed` (absent review_status = inherited = refused). pgTAP 011 + the SQL-mirror parts of the Rust `sheet_conformance` generator deleted; ledger retargeted.
- Publish gate: `publish_sheet_version(review_id)` atomic; `guard_sheet_version_update` lets draft->published only with an approved round of same version+revision + passed verdict (no direct UPDATE publishes, even table owner). Verdict is a fixture until M6 (V1-2.3).
- Reader column grants narrowed; editor needs only `revision` on versions; authors/provenance/fingerprint reach staff via RPCs (`get_sheet_version`).
+3
View File
@@ -472,3 +472,6 @@ R5.4: `draft/{labels,changelog,review_report,drawing,overlay,app_review,svg_comp
- Owner ratified AR-1 B (BDR-083). Contract §5.7 written: rule, chain, path inventory (2 RPCs, 2 private fns, service_role table grants F17, owner-count race F18), bootstrap, factor/password/recovery, session rules, deploy order (enrol before M4 on DEV), tests §15.
- GoTrue v2.197.0 source read: enrol needs aal2 once verified factor, unenroll verified needs aal2. auth.sessions check kept as to-verify.
- Docs only. Independent security review, then commit + push feature branch. STOP at checkpoint.
## 2026-10-28
- V1-2.1 M4/M5 built+tested on sandbox (DEV untouched): M4 20261028100000 sheet_access_hardening (rights table + RPCs, has_mfa/has_aal2_session, catalogue reader predicate, can_edit_sheet on capability, AR-1 B role RPCs + F17/F18/F19, F1-F16), M5 20261028110000 sheet_review_workflow (version/sheet cols, freeze/revision/author triggers, rounds/comments/acks, lifecycle+review+publish RPCs, species, working copy, uploads+INSERT policy, mirror removed + status check). Defects F1-F19 reproduced on old schema; pgTAP 001/003/004/005/006/007/010/015/016 green (850 real-stack, 122 PGlite); sandbox real-GoTrue AR-1 B + F18 race green; fresh install + M3->M5 upgrade green; sheets.ts closed kinds, client gate deleted, types regen no drift; rust:check + typecheck + lint + build green. 011 + SQL-mirror generator deleted, ledger retargeted. BDR + LRN-095. check:all Docker image/worker stages not re-run (no app/worker/Dockerfile change). STOP at checkpoint.
+9
View File
@@ -651,3 +651,12 @@ Apply: SQL test touching a worker queue or shared bucket → neutralise or scope
Context: V1-1 DEV acceptance RESP-01. Owner saw whole-page sideways scroll everywhere at 375 px (DevTools responsive). Sandbox sweep (Chromium mobile + desktop with scrollbars shown via `ignoreDefaultArgs: ['--hide-scrollbars']`, WebKit iPhone, Firefox; 280-430 px) found none at 375; owner then measured DEV with a console snippet (cw/sw + topmost offending boxes, classes only): no overflow without zoom → browser zoom was the cause. Same sweep with long account e-mails found a real defect: dashboard `grid gap-4 md:grid-cols-2` has no explicit column below md → implicit `auto` track = cards' min-content → `truncate` e-mail never truncates, cards past screen from ~31 chars at 320 px.
Fix: `grid-cols-1` (minmax(0,1fr)) + e-mail wraps (`min-w-0 [overflow-wrap:anywhere]`). Test sweeps 320-430 + desktop with long values, proven failing on old code.
Apply: overflow report → ask how observed (zoom, preset, device type), give owner a read-only console probe for DEV, reproduce with realistic max-length values on sandbox; a grid used as a single column on phones needs `grid-cols-1`, else one unbreakable value widens every card.
## LRN-095 — pgTAP across PGlite and the real Supabase image: privileges, roles, storage triggers (2026-10-28)
Adapting pgTAP across PGlite and the real Supabase image: the fixtures must respect privileges the shim doesn't enforce.
- `auth.sessions`/`auth.mfa_factors` added to the PGlite shim (`supabase-shim.sql`) with the pinned image's enums (`aal_level`, `factor_type`, `factor_status`) and NOT-NULL columns, so aal2-session checks run Dockerless. A test that checks a live-aal2 session fabricates the session + factor rows directly.
- API roles (`authenticated`, `service_role`, `anon`) hold no privilege on `auth.*`: every test mutation of `auth.users/sessions/mfa_factors` must run under `reset role` (postgres), never while `set local role authenticated`. Same for seeding `public.*` rows a trigger guards (the content trigger needs `auth.uid()` set, so direct version seeding carries a `request.jwt.claims` sub).
- Real-stack-only divergences: `storage.objects` has a storage-API trigger forbidding direct DELETE/UPDATE (PGlite allows it) — assert "no write policy" via `pg_policies` census, not a runtime throws_ok. PostgREST maps a `raise ... errcode='42501'` to HTTP 403 with code `42501` (not P0001). An EXECUTE-revoked function raises "permission denied for function" before its own body — test that, not the body's custom error.
- A narrow operator read RPC returning JSON `null` for "no row" is cleaner than raising (PostgREST won't surface a raise as a 404 the client recognises); the tool treats `null` as absent.
- Real GoTrue proves what fabricated rows can't: `pnpm db:sandbox test` enrol+verify TOTP → aal2 token, then the role RPC works at aal2 and is refused (mfa_required, 403) at aal1; F18 cross-revoke under real concurrency leaves one owner.