mirror of
https://github.com/shankar0123/certctl.git
synced 2026-06-07 15:11:29 +00:00
5d7bc864515a900ccfcd431cba006b5813d1466e
178 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 shankar0123
|
c4ed3da30b |
fix(ci): Sprint 6 CI follow-up — staticcheck ST1021 + tenant-query baseline + skip inventory
Sprint 6 push (commits |
||
|
shankar0123
|
663b14bfd8 |
feat(retention): COMP-002-RETENTION — federated-user PII purge pipeline
Sprint 6 closure of the audit's MED-severity COMP-002-RETENTION
finding.
Pre-fix posture: the federated-user admin surface
(auth_users.go::Deactivate) sets users.deactivated_at on soft-delete,
but the PII columns (email, display_name, oidc_subject) stay
populated forever. No in-code primitive for GDPR right-to-be-
forgotten; no scheduled retention purge.
This commit ships the audit's recommended two-phase fix:
Phase 1 — operator-callable scrub primitive
internal/service/user_retention.go
UserRetentionService.DeleteUserPII(ctx, userID):
- revoke all active sessions (defense-in-depth)
- email := 'purged@redacted.local'
- display_name := '[purged]'
- oidc_subject := 'sha256:' || hex(sha256(original))
- audit_events row with action=user.purge_pii,
category=auth, actor=system
Why hash oidc_subject instead of NULL:
1. (oidc_provider_id, oidc_subject) UNIQUE constraint would
trip on multiple purged users converging to NULL
2. The hash is one-way; the original IdP-side identifier is
unrecoverable. Re-login under the same subject mints a
fresh u-id (right-to-be-forgotten semantics)
3. Forensic continuity: an operator can recompute
sha256(<known-subject>) and confirm "this user was
deactivated then purged"
users.id itself is preserved so historical
audit_events.actor = u-X rows still resolve. The forensic-
attribution chain stays intact even after the PII is gone.
Phase 2 — scheduled batch purge
internal/scheduler/scheduler.go
UserRetentionPurger interface + userRetentionLoop:
- PurgeDeactivatedUsers enumerates every user with
deactivated_at < NOW() - retention_window
- DeleteUserPII per row
- per-tick batch cap (default 200) keeps blast radius
predictable; large backlogs spread across multiple ticks
- atomic.Bool guard + 5-min per-tick context.WithTimeout
Repository contract grew a single new method:
internal/repository/user.go::ListDeactivatedBefore(ctx, t)
internal/repository/postgres/user.go: SQL-side filter
(deactivated_at IS NOT NULL AND deactivated_at < $1)
ORDER BY deactivated_at ASC, cross-tenant.
Configuration
CERTCTL_USER_RETENTION_INTERVAL default 24h
CERTCTL_USER_RETENTION_WINDOW default 30 days
CERTCTL_USER_RETENTION_BATCH_CAP default 200
Test stub additions for repository.UserRepository.ListDeactivatedBefore:
internal/auth/oidc/service_test.go::stubUsers
internal/api/handler/auth_users_test.go::stubFullUserRepo
internal/api/handler/auth_session_oidc_test.go::stubUserRepo
Documentation
docs/operator/privacy-and-retention.md
- retention pipeline diagram (day-0 deactivate → day-N purge)
- operator config table
- verification runbook (4 steps with SQL)
- what's NOT covered (deferred: DSAR export, api_keys cascade,
retroactive audit_events.details redaction)
Tests
internal/service/user_retention_test.go (NEW, 4 tests):
TestDeleteUserPII_ScrubsAndRevokes
TestDeleteUserPII_IsIdempotent
TestPurgeDeactivatedUsers_RespectsWindow
TestPurgeDeactivatedUsers_BatchCap
Verified locally:
go vet ./... (clean)
gofmt -l internal/ cmd/ (clean)
go test -short -count=1 \
./internal/service/... ./internal/scheduler/... ./internal/config/...
(all green)
Cross-sprint interaction: pairs with COMP-001-HASH (prior commit).
The user.purge_pii audit row this service emits flows through the
new hash chain, so the scrub event is itself tamper-evident.
Closes COMP-002-RETENTION. Sprint 6 is complete (2/2 findings).
|
||
|
shankar0123
|
43836aca7c |
feat(audit): COMP-001-HASH — per-row hash chain on audit_events (tamper-evidence)
Sprint 6 closure of the audit's HIGH-severity COMP-001-HASH finding.
Pre-fix posture: migration 000018 installs a WORM trigger on
audit_events that blocks UPDATE / DELETE for the application role.
But the trigger header itself documents a compliance-superuser
bypass (backup restore, retention purges, breach recovery). Without
a hash chain, that role can rewrite any row's actor / action /
details / timestamp / event_category with no on-disk trace.
HIPAA §164.312(b), FedRAMP AU-9, NIST 800-53 AU-10 want tamper-
EVIDENCE, not just tamper-prevention. This commit ships the
evidence layer.
Wire shape:
migrations/000047_audit_events_hash_chain.up.sql
+ pgcrypto extension (digest function)
+ audit_chain_head: single-row sentinel table holding the most
recent row_hash; FOR UPDATE row-lock serialises chain writes
under concurrent INSERTs so two parallel writers can't read
the same prev_hash and produce a forked chain
+ audit_events: prev_hash + row_hash columns
+ audit_events_canonical_payload(): centralised hash input
builder. UTC + microsecond ISO-8601 keeps the hash session-
timezone-independent. All columns separated by '|' so a
concatenation-ambiguity exploit can't fabricate a collision
+ audit_events_compute_hash_chain(): BEFORE-INSERT trigger
function. Reads sentinel FOR UPDATE → computes
sha256(prev_hash || id || actor || actor_type || action ||
resource_type || resource_id || details::text ||
timestamp_utc_iso || event_category) → writes both columns +
advances the sentinel
+ backfill loop walks every existing row in (timestamp ASC, id
ASC) order; WORM trigger temporarily DISABLEd inside this
migration's transaction so backfill UPDATEs land cleanly,
ENABLEd before COMMIT
+ audit_events_verify_chain(): STABLE plpgsql verifier. Walks
the chain end-to-end and returns the first break:
(first_break_id TEXT, first_break_pos INT, row_count INT)
internal/repository/postgres/audit.go
+ AuditRepository.VerifyHashChain — calls the SQL function and
maps the OUT parameters to Go return values
internal/repository/interfaces.go
+ AuditRepository.VerifyHashChain in the contract; every
in-memory mock + stub picks up the no-op implementation
internal/scheduler/scheduler.go
+ AuditChainVerifier + AuditChainBreakRecorder interfaces
+ auditChainVerifyInterval (default 6h)
+ auditChainVerifyLoop: runs once on start + every tick;
atomic.Bool guard + 5-min per-tick context timeout match every
other GC loop's pattern
internal/service/audit_chain_metric.go
+ AuditChainCounter type with atomic counters. Sticky-first-
detection on (BrokenAtID, BrokenAtPos) so the actionable
alarm doesn't drift across walks. Snapshot() returns the
full state for the metrics handler
internal/api/handler/metrics.go
+ AuditChainCounterSnapshotter interface + Prometheus
exposition for four series:
certctl_audit_chain_break_detected_total counter (the alarm)
certctl_audit_chain_verify_total counter (walks done)
certctl_audit_chain_rows gauge (last walk size)
certctl_audit_chain_last_verified_at gauge (unix seconds)
internal/config/config.go
+ AuditChainConfig{ VerifyInterval } + CERTCTL_AUDIT_CHAIN_VERIFY_INTERVAL
cmd/server/main.go
+ wires AuditChainCounter into both the scheduler (recorder) +
metrics handler (snapshotter) — single instance shared so the
writer + reader are guaranteed to converge
internal/repository/postgres/audit_chain_test.go (NEW)
+ TestAuditEventsHashChain_FreshTable: empty walk → clean
+ TestAuditEventsHashChain_AppendLinksRows: three INSERTs
produce a strictly-linked chain; prev_hash on row 0 is NULL;
verifier walks clean over the 3 rows
+ TestAuditEventsHashChain_VerifierDetectsTampering: simulate
the compliance-superuser threat model (DISABLE WORM, UPDATE
a middle row, ENABLE WORM); verifier returns the tampered
row's id at position 1
docs/operator/audit-chain.md (NEW)
+ Layered-defenses explainer (WORM + hash chain). Verifier
function reference. Recommended Prometheus alert rule.
Performance scaling table (10k to 10M rows). Step-by-step
runbook for what to do when a break is detected. Operator
configuration table.
Test-stub additions for AuditRepository.VerifyHashChain:
internal/service/testutil_test.go — mockAuditRepo
internal/service/acme_test.go — fakeAuditRepo
internal/integration/lifecycle_test.go — mockAuditRepository
internal/api/handler/scep_intune_e2e_test.go — intuneE2EAuditRepo
Verified locally:
go vet ./... (clean)
gofmt -l internal/ cmd/ (clean)
go test -short -count=1 ./internal/scheduler/... ./internal/config/...
./internal/service/... ./internal/api/handler/... ./internal/repository/...
(all green)
Verified with testcontainers + postgres:16-alpine + the migration
runner (not gated under -short — requires docker):
go test -count=1 -run TestAuditEventsHashChain ./internal/repository/postgres/...
Closes COMP-001-HASH leg of Sprint 6. COMP-002-RETENTION lands in
the next commit (separate concern: federated-user PII retention).
|
||
|
shankar0123
|
15fedbaa06 |
test(scheduler): SCALE-001 — assert claim cap via non-Pending count, not Running
Sprint 2's TestProcessPendingJobs_RespectsClaimLimit asserted
that exactly 3 jobs sat in JobStatusRunning after a 10-row
ProcessPendingJobs sweep with SetClaimLimit(3). The CI run
landed 'running-job count = 0; want 3.'
Root cause: the mock's ClaimPendingJobs flips Pending → Running
on the 3 claimed rows (atomic-claim semantics). processJob then
calls renewalService.ProcessRenewalJob, which fails on the
mock cert-repo's not-found error and calls failJob → which
transitions the row from Running → Failed. By the time the
test assertion runs, no row is still in Running.
The load-bearing SCALE-001 invariant is 'the cap STOPPED at 3.'
Whether the 3 claimed rows ended up Running, Failed, or
Completed is irrelevant to the cap — what matters is that 7
rows STAYED in Pending for the next tick.
Fix: count non-Pending (= claimed) and still-Pending (= 10
minus claimed) separately. Assert claimed=3 and stillPending=7.
LastClaimLimit=3 assertion (already passing in the failed run)
also stays as the seam-propagation pin.
This is a test-fix only — the SCALE-001 production behavior
landed correctly in
|
||
|
shankar0123
|
a485e31f63 |
fix(repo,service): SCALE-002 — push pagination into SQL for target/issuer/team/agent_group
Sprint 2 unified-master-audit closure. Pre-fix four service List
endpoints (target, issuer, team, agent_group) called repoFoo.List(ctx)
to fetch the full table then sliced in memory:
rows, _ := s.repo.List(ctx)
total := int64(len(rows))
start := (page - 1) * perPage
end := start + perPage
return rows[start:end], total, nil
This page-sliced in memory pattern marshals every row per request —
fine on small fleets but unacceptable for multi-tenant or large-fleet
deploys. The agent_group case was worse — the service explicitly
ignored page/perPage and returned the entire slice.
Fix:
- New ListPaginated(ctx, limit, offset) method on each of the four
repositories. Postgres implementations push LIMIT + OFFSET into
the SQL plus a SELECT COUNT(*) for the total. Mirrors the cursor
pattern already in internal/repository/postgres/certificate.go.
- Each ListPaginated normalises limit≤0→50 and offset<0→0,
matching the service-layer defaults that already existed.
- Repository interfaces grow the new method so adapters stay
swappable.
- Service List methods now call repoFoo.ListPaginated(ctx, perPage,
(page-1)*perPage) directly — no more memory-slice.
- AgentGroupService.ListAgentGroups closes the Bundle E / Audit
L-020 'page/perPage unused' gap.
Test changes:
- sliceWindow generic helper in testutil_test.go mirrors the SQL
LIMIT/OFFSET semantics for in-memory mocks.
- Six mock implementers (lifecycle_test, testutil_test x2,
agent_group_test, team_test) gain ListPaginated methods.
- TestTeamService_List_SCALE002_PaginationPropagatesToRepo pins
the page=2, perPage=3 → 3 rows of 10 invariant.
Closes SCALE-002.
|
||
|
shankar0123
|
037876fa0f |
fix(scheduler): SCALE-001 — cap ClaimPendingJobs per-tick (default 1000)
Sprint 2 unified-master-audit closure. Pre-fix the scheduler invoked
ClaimPendingJobs(ctx, "", 0). limit:0 loads every Pending row in a
single transaction — a 100K-job burst (cert-fleet sweep, post-outage
recovery, large agent-fleet first boot) marshalled the full queue
into process memory before boundedFanOut's semaphore could back-
pressure the upstream CAs.
Fix:
- SchedulerConfig.JobClaimLimit (env CERTCTL_SCHEDULER_JOB_CLAIM_LIMIT,
default 1000). ≤0 normalised to 1000 in SetClaimLimit — fail-safe
vs. legacy unlimited semantics.
- JobService.claimLimit threaded into the existing
ProcessPendingJobs flow; ClaimPendingJobs(ctx, "", s.claimLimit).
- cmd/server/main.go wires jobService.SetClaimLimit(cfg.Scheduler.JobClaimLimit).
- 'processing pending jobs' log line now includes claim_limit so
operators can spot the cap engaging (count == claim_limit ⇒
queue is running ahead of fan-out; bump CERTCTL_SCHEDULER_JOB_CLAIM_LIMIT
or CERTCTL_RENEWAL_CONCURRENCY).
- Test wiring keeps the legacy zero-value (unlimited) for byte-
for-byte compatibility with the existing 600+ JobService unit
tests — only production code goes through SetClaimLimit.
Regression coverage:
- mockJobRepo.LastClaimLimit records the limit passed through
ClaimPendingJobs so tests can pin the propagation.
- TestProcessPendingJobs_RespectsClaimLimit: 10 Pending rows,
SetClaimLimit(3), expect exactly 3 transition to Running plus
LastClaimLimit=3 on the mock.
- TestSetClaimLimit_NormalisesNonPositive: 0/-1/-1000 all
normalise to 1000.
Closes SCALE-001.
|
||
|
shankar0123
|
037dab7b6f |
fix(agent,service): SEC-002 — validate certificate_id shape + contain key path
Sprint 1 unified-master-audit closure. Pre-fix the agent built its
on-disk key path via:
keyPath := filepath.Join(a.config.KeyDir, job.CertificateID+".key")
migrations/000001_initial_schema.up.sql declares managed_certificates.id
as TEXT PRIMARY KEY with no shape constraint, so a compromised control
plane (or a poisoned database row) could deliver a job whose
certificate_id is '../../etc/passwd', '/absolute/path', a NUL-byte
payload, or a Windows-separator-laden string — driving arbitrary
file write or read on the agent host.
Fix (two ends; both load-bearing):
Server side:
- New internal/validation/certificate_id.go: ValidateCertificateID
pins the canonical TEXT-PK shape (^[A-Za-z0-9._-]{1,128}$, plus
explicit '.'/'..' rejection).
- CertificateService.Create now invokes ValidateCertificateID after
the existing required-fields check; malformed IDs are refused
before persistence or downstream job creation.
Agent side:
- cmd/agent/keymem.go: validateAgentCertID mirrors the server-side
shape regex. safeAgentKeyPath additionally asserts the joined
path is contained within KeyDir via filepath.Rel — even if a
future refactor bypasses the shape check, a path that escapes
KeyDir fails closed.
- poll.go + deploy.go: both filepath.Join call sites routed
through safeAgentKeyPath; rejection surfaces via reportJobStatus
so the control plane sees the failure.
Regression coverage:
- internal/validation/certificate_id_test.go: production shapes
accepted; explicit rejection table for empty, overlong, posix
traversal, absolute, Windows traversal, Windows separator, NUL
byte, newline/tab injection, drive prefix, space, unicode dots.
- cmd/agent/keymem_test.go: validateAgentCertID acceptance +
rejection tables; safeAgentKeyPath happy path + the 8 audit
vectors plus empty-keyDir refusal.
Closes SEC-002.
|
||
|
shankar0123
|
fc237de357 |
feat(audit): close P-H2 — server-side since / until time-range filters
Closes frontend-design-audit finding P-H2 (High):
AuditPage filters time-range *client-side*; comment says "server
may not support time params" — fetches the entire event window,
throws 99% away in JS
Ground-truth recon found the closure is much smaller than the
audit's "1 day backend + 2 hours frontend" estimate:
• repository AuditFilter.From / .To: ALREADY exist in
internal/repository/filters.go:57-58
• postgres.AuditRepository.List: ALREADY pushes
`timestamp >= since` + `timestamp <= until` predicates into the
SQL query (internal/repository/postgres/audit.go:107-116)
• Composite index idx_audit_events_category_timestamp on
(event_category, timestamp DESC) added in migration 000032
makes the new query hit an index scan
• MCP `certctl_audit_list_with_category` tool's docstring already
advertises `since` / `until` (internal/mcp/tools_audit_fix.go:174)
— but the server silently ignored them, making the published
contract a lie
The only missing piece was the handler exposing the params + the
frontend porting from client-side filtering. ~150 lines total.
═══════════════════════════ CHANGES ═══════════════════════════════
Service (internal/service/audit.go):
• New ListAuditEventsByFilter(ctx, since, until, category, page,
perPage) threads time bounds into the existing repository.
AuditFilter.From / .To fields.
• Existing ListAuditEvents + ListAuditEventsByCategory become
thin wrappers around the new method with zero times.
Handler (internal/api/handler/audit.go):
• Interface gains ListAuditEventsByFilter signature.
• ListAuditEvents handler parses `since` + `until` RFC3339 query
params; 400 on malformed input or `until` not after `since`.
• Single dispatch via ListAuditEventsByFilter for ALL request
shapes (with or without time bounds, with or without category).
Tests (internal/api/handler/audit_handler_test.go):
• mockAuditService gains listByFiltFunc + lastFilterSince/Until/
Category trace fields.
• 5 new subtests:
- TestListAuditEvents_WithSinceUntil — happy path, both bounds
- TestListAuditEvents_SinceOnly — one-sided open-ended
- TestListAuditEvents_InvalidSince — 400 on garbage
- TestListAuditEvents_UntilBeforeSince — 400 on reversed range
- TestListAuditEvents_TimeRangePlusCategory — composes with
auditor-role category=auth filter
Frontend (web/src/pages/AuditPage.tsx):
• TIME_RANGES dropdown now sends `since` as RFC3339 (now − N hours)
via the existing useQuery params object instead of filtering
client-side after the fact.
• Pre-P-H2 `filtered = data.data.filter(e => now-ts<N)` block
deleted (replaced by `filtered = data?.data || []`); comment
documents why for the diff reader.
OpenAPI (api/openapi.yaml):
• listAuditEvents gains `since` + `until` query-param specs
(format: date-time, description, P-H2 closure date).
• Description block explains the `since`/`until` vs `from`/`to`
naming divergence from the sibling /audit/export endpoint
(different param semantics: list = open-ended bounds, export =
required ≤ 90-day compliance window).
═══════════════════════════ VERIFICATION ═══════════════════════════
Backend (Go toolchain now wired in sandbox — go1.25.10 ARM64 from
.gomodcache, GOCACHE on /tmp partition):
• gofmt -l on all touched files: clean
• go vet ./... — exit 0
• go test -short -count=1 ./internal/api/handler/... — ok 4.195s
(existing 14 subtests + 5 new = 19/19 pass)
• go test -short -count=1 ./internal/service/... — ok 4.733s
• staticcheck ./internal/api/handler/... ./internal/service/...:
zero findings
Frontend:
• npm ci — 634 packages, exit 0 (resolves cleanly post-Hotfix #9)
• npx tsc --noEmit — exit 0
• npx vitest run src/pages/AuditPage.test.tsx — 4/4 pass
• npx vite build — built in 3.49s
Ground-truth: origin/master tip
|
||
|
shankar0123
|
b1fa4970be |
refactor(service/acme): extract orders concern to sibling file (Phase 9, 9b — deferred half of Sprint 9)
Phase 9 ARCH-M2 closure Sprint 9b — the orders cut Sprint 9
explicitly deferred. Closes the bigger half of the
internal/service/acme.go split via the Option B sibling-file pattern
(operator's post-Sprint-8 choice — package stays `service`, no
import-path churn for ~70 call sites).
Why Sprint 9b is a separate commit from Sprint 9
================================================
Sprint 9 shipped four cuts whose source ranges were each a single
contiguous region in acme.go (nonces, authz, challenges, gc — line
ranges 423-444 / 999-1018 / 1326-1561 / 1914-1965 at audit time).
Sprint 9b crosses a different shape:
1. Non-contiguous source: orders block A (lines 795-1223 pre-cut)
+ helpers block B (1237-1283 pre-cut), with
firstAvailableIssuer at 1227-1235 staying behind because it's
called from Phase 4 RevokeCert + RenewalInfo too.
2. Per-helper move-vs-stay decision: each helper in the
post-FinalizeOrder cluster needed an explicit call-graph audit
to decide whether it moves with orders or stays with the
surviving cross-concern surface in acme.go.
Same shape as the Sprint 8 / Sprint 8b split (mechanical vs harder-
shape on separate commits) — the Phase 9 prompt's "do not bundle"
rule enforcing itself.
What moved
==========
New `internal/service/acme_orders.go` (540 LOC)
-----------------------------------------------
Contains the entire Phase 2 orders concern:
- The `// --- Phase 2 — orders + authz + finalize + cert download`
banner (moves with its contents, not left as a phantom in
acme.go pointing at code that's no longer there).
- The four public order methods: CreateOrder, LookupOrder,
FinalizeOrder, LookupCertificate.
- The FinalizeOrderResult shape (consumed only by FinalizeOrder
callers).
- accountOwnsACMECert (only callsite: LookupCertificate).
- The three orders-internal ID helpers: randIDSuffix +
base32encode (random ACME entity IDs) + identifierStrings
(audit details).
Per-helper move-vs-stay analysis
================================
Grep against the post-Sprint-9 tree pinned every helper's call sites
before the cut decision:
randIDSuffix: callers in CreateOrder (4x) + FinalizeOrder
(1x) — all moving. MOVE.
base32encode: only caller is randIDSuffix. MOVE.
identifierStrings: only caller is CreateOrder. MOVE.
accountOwnsACMECert: only caller is LookupCertificate. MOVE.
firstAvailableIssuer: three call sites — FinalizeOrder (moving),
RevokeCert (staying, Phase 4), RenewalInfo
(staying, Phase 4). STAY in acme.go.
Doc-comment updated to flag cross-concern
status + explain why it's not moved.
mapACMERevocationReason: only caller is RevokeCert. STAY (already
sits in the Phase 4 region of acme.go and
belongs with its sole caller).
jwksThumbprintsEqualSvc: only caller is RotateAccountKey. STAY
(Phase 4 helper; never had an orders
relationship).
Side effect: import cleanup
===========================
With randIDSuffix moved, acme.go no longer references crypto/rand.
The `cryptorand "crypto/rand"` aliased import is removed.
Per-symbol audit confirmed every other import (context, crypto/x509,
errors, fmt, strings, sync/atomic, time, jose, internal/api/acme,
internal/config, internal/domain, internal/repository) is still
consumed by surviving code in acme.go.
Net effect
==========
acme.go: 1634 → 1158 LOC pre-doc-update; 1162 LOC post the four-line
firstAvailableIssuer doc-comment refresh (-472 net, -28.9% from the
post-Sprint-9 size). Original audit-time size was 1965 LOC; cumulative
Sprint-9 + Sprint-9b reduction: 1965 → 1162 = -803 LOC (-40.9%).
The biggest single backend hotspot from the audit is now smaller
than mcp/tools.go.
Behavior preservation contract
==============================
1. gofmt -l clean across acme.go + acme_orders.go.
2. go vet ./internal/service/... — no findings.
3. staticcheck ./internal/service/... ./cmd/server/...
./internal/api/handler/... ./internal/scheduler/...
./internal/mcp/... — no findings.
4. go test -short -count=1 ./internal/service/... — green
(including the orderTrackingRepo + TestCreateOrder_* +
TestFinalizeOrder_* + TestLookupCertificate_* surface that
pins the moved code's behavior).
5. Broader-importer suite green:
go test -short -count=1 ./cmd/server/... ./internal/api/handler/...
./internal/scheduler/...
6. Per-symbol import audit on both files (no unused imports left,
no missing imports introduced).
Same-package resolution means every call inside FinalizeOrder /
RevokeCert / RenewalInfo to firstAvailableIssuer crosses a file
boundary but stays within `package service` — zero overhead at
compile time, zero change to the public method-set on
service.ACMEService.
What remains for Phase 9
========================
Three sibling-file splits queued for Sprints 10-12:
- Sprint 10: internal/mcp/tools.go (1867 LOC) grouped by tool
domain (certificate / agent / job / discovery / admin).
- Sprint 11: internal/api/handler/auth_session_oidc.go (1577 LOC)
split per handler verb.
- Sprint 12: cmd/agent/main.go (1489 LOC) mirroring the cmd/server
pattern from Sprint 8.
Refs: ARCH-M2 (god-files), Phase 9 audit. Sprint 9b is the named
follow-on to Sprint 9; after this commit, the service-layer cut from
the audit's hotspot list is fully closed.
|
||
|
shankar0123
|
b503d27b4f |
refactor(service/acme): split into sibling files — Option B (Phase 9, 9 of N — partial)
Phase 9 ARCH-M2 closure Sprint 9. Splits internal/service/acme.go
(was 1965 LOC, the top hotspot after Sprints 1-8 finished the
config + main-binary cuts) via the Option B sibling-file pattern —
new files stay in `package service` so every external caller of
`service.ACMEService.{IssueNonce,LookupAuthz,ListAuthzsByOrder,
RespondToChallenge,GarbageCollect}` resolves the same way. Pure
mechanical relocation; no signature, no behavior, no import-graph
change.
Why Option B (not a subpackage)
================================
A subpackage (e.g. `internal/service/acme/`) would have meant
rebadging every public method receiver to its new package — that's
import-path churn for ~70 call sites across handlers, scheduler,
cmd/server wiring, MCP tools, and tests, plus the cyclic-import
risk of pulling acme back into `service` for the shared interfaces.
Option B sacrifices the encapsulation discipline a subpackage
would have given (sibling files can still reach into each other's
unexported state because Go scopes are per-package), but in
exchange the diff is restricted to file moves + four sed deletes;
zero importer touches anywhere outside this directory. The
trade-off matches every prior Sprint 1-7 config cut.
What moved
==========
New `internal/service/acme_nonces.go` (46 LOC)
----------------------------------------------
The IssueNonce method (RFC 8555 §6.5 Replay-Nonce issuance). The
nonceAdapter type — which wraps ACMERepo.ConsumeNonce for the JWS
verifier — stays in acme.go alongside VerifyJWS because it's
verification-infrastructure plumbing, not a server-issues-nonce
concern.
New `internal/service/acme_authz.go` (45 LOC)
---------------------------------------------
LookupAuthz + ListAuthzsByOrder (the authz read-side). Authz write-
side (status cascade after challenge validation) lives in
acme_challenges.go alongside recordChallengeOutcome where it
belongs operationally; the authz creation path stays inside
CreateOrder in acme.go (orders own per-order authz row creation).
New `internal/service/acme_challenges.go` (267 LOC)
---------------------------------------------------
The whole Phase 3 challenge dispatch + validator callback concern:
the `// --- Phase 3 — challenge dispatch + validator callback ---`
banner, the ChallengeResponseShape struct, the HTTP-facing
RespondToChallenge method (which transitions challenge → processing
and submits to the validator pool), and the asynchronous
recordChallengeOutcome callback (which persists final challenge
status and cascades the parent authz + order status). Largest
single extract this sprint by line count.
New `internal/service/acme_gc.go` (74 LOC)
------------------------------------------
The Phase 5 ACME GC sweep: scheduler-invoked GarbageCollect entry
point (3 sweeps: nonces, expired authzs, expired orders) and the
atomicAddUint64 counter helper (only consumed by the sweep body
for the rows-affected-N case the default `bump` doesn't cover).
What deferred
=============
Sprint 9 was originally scoped to ship 5 sub-files (nonces / authz /
challenges / orders / gc). The orders cut — CreateOrder +
LookupOrder + FinalizeOrder + LookupCertificate + the orders
helpers (randIDSuffix / base32encode / identifierStrings /
firstAvailableIssuer / accountOwnsACMECert / mapACMERevocationReason) +
FinalizeOrderResult — is ~700 LOC spread across multiple non-
contiguous regions in acme.go, with the orders helpers also feeding
into RevokeCert / RenewalInfo on the Phase 4 side. Disentangling
which helpers move with orders vs which stay with Phase 4 needs a
focused sprint of its own to avoid leaving a half-cut helper
declared in one file but called from a sibling — which works
(same package) but defeats the point of organising by concern.
Deferred to a potential Sprint 9b.
Net effect
==========
acme.go: 1965 → 1634 LOC (-331). Four new sibling files at 432 LOC
total. The headline 1965-LOC hotspot drops below the next-tier
candidates (mcp/tools.go, auth_session_oidc.go, cmd/agent/main.go).
Behavior preservation contract
==============================
1. gofmt -l clean across all 5 affected files.
2. go vet ./internal/service/... — no findings.
3. staticcheck ./internal/service/... — no findings.
4. go test -short -count=1 ./internal/service/... — green.
5. Broader-importer build green:
go build ./cmd/server/... ./internal/api/handler/...
./internal/scheduler/... ./internal/mcp/...
6. Broader-importer tests green:
go test -short -count=1 ./cmd/server/... ./internal/api/handler/...
./internal/scheduler/...
7. Per-import-symbol audit: all 8 imports remaining in acme.go
(context, cryptorand, x509, errors, fmt, strings, sync/atomic,
time, jose, internal/api/acme, internal/config, internal/domain,
internal/repository) verified used by surviving code. New
sibling files carry only the imports their extracted code needs.
The Option B sibling-file shape means same-package resolution
preserves access to ACMEService's unexported state from every
extracted method without any visibility tweaks. Worth noting for
the future: this also means a careless future caller could reach
through file boundaries and re-tangle concerns; the file headers
document the intended boundary but Go's tooling won't enforce it.
Why this is a partial sprint
============================
Splitting into 4 of 5 named sub-files now (vs blocking until orders
is also clean) keeps the hotspot count down with this commit and
lets a follow-up Sprint 9b focus exclusively on the orders cut
without re-touching the four files this sprint ships. Same
"smallest useful slice, document the rest" cadence as Sprint 8
splitting into 8a (mechanical) + 8b (behavior-aware).
Refs: ARCH-M2 (god-files), Phase 9 audit. Last in the config /
service hotspot chain before the agent + mcp + auth-session cuts
land in Sprints 10-12.
|
||
|
shankar0123
|
21aeed4f4e |
legal: addlicense headers + normalize legacy variants (Phase 0 RED-4)
Phase 0 closure (Path B2, post-rewrite):
addlicense sweep — adds the canonical certctl LLC copyright + BUSL-1.1
SPDX header to every production Go file. Template:
// Copyright 2026 certctl LLC. All rights reserved.
// SPDX-License-Identifier: BUSL-1.1
Coverage: 338 / 338 production Go files (cmd/ + internal/, excluding
*_test.go and **/testdata/**). Pre-sweep coverage was 22 / 338 (6.5%);
post-sweep is 338 / 338 (100%).
Normalized 22 pre-existing legacy headers (`// Copyright (c) certctl`
+ `// SPDX-License-Identifier: BSL-1.1`) and 1 file using a
`Certctl Contributors` attribution. The legacy SPDX ID `BSL-1.1`
is non-standard; the official SPDX identifier for Business Source
License 1.1 is `BUSL-1.1` (capital U). All 338 files now share the
canonical form.
Generated via:
addlicense -c "certctl LLC" -y 2026 \
-f cowork/legal/copyright-header.tpl \
-ignore '**/testdata/**' -ignore '**/*_test.go' \
cmd/ internal/
Verification:
find cmd internal -name '*.go' -not -name '*_test.go' \
-not -path '*/testdata/*' \
-exec grep -L '^// Copyright 2026 certctl LLC' {} \; | wc -l
Returns: 0
gofmt clean. Header additions are comments only, no compile impact.
Closes: cowork/certctl-architecture-diligence-audit.html#fix-RED-4
|
||
|
shankar0123
|
02438ad9e1 |
ci: floor raise + doc drift (Phase 3 closure — TEST-H1/H2/M1/M2/M3/M4/L1, ARCH-H3/L1/L2/L3/L4)
Twelve findings from the architecture diligence audit's Phase 3 bundle
closed in one PR. All touch the CI workflows + small doc-drift fixes
across the production Go tree + migration headers.
CI workflow changes
====================
TEST-H1 — Race detection on ./... -short
.github/workflows/ci.yml:106 was a 9-package explicit list. Audit
finding TEST-H1 flagged that 25+ packages (internal/auth/*,
internal/repository/*, internal/mcp, internal/scep, internal/pkcs7,
internal/api/router, internal/api/acme, internal/cli, internal/cms,
internal/config, internal/deploy, internal/integration,
internal/ratelimit, internal/secret, internal/trustanchor, all of
cmd/) silently dropped off race coverage.
Post-fix: 'go test -race -short ./... -count=1 -timeout 600s'.
76 testing.Short() guards already cover testcontainers + live-DB
integration suites, so -short keeps the long-running tests out.
TEST-H2 — Cross-platform build matrix
New 'cross-platform-build' job in ci.yml. Matrix:
ubuntu-latest + windows-latest + macos-latest, fail-fast: false.
Builds cmd/server + cmd/agent + cmd/cli + cmd/mcp-server on each.
Catches Windows-specific regressions (path separators, file
permissions, exec.Command semantics) the pre-Phase-3 Ubuntu-only
CI missed.
TEST-L1 — actions/setup-go cache: true (explicit)
setup-go v5 defaults cache: true; making it explicit so a future
setup-go upgrade can't silently flip it. Re-runs hit the Go module
+ build cache instead of recompiling cold.
TEST-M1 — Mutation-testing floor at 55%
security-deep-scan.yml::go-mutesting step rewritten. Removed
continue-on-error + per-package '|| true'. New post-loop check
extracts every 'The mutation score is X.YZ' line and fails the
step if any package drops below 0.55. Floor rationale: starter
ratio catches major regressions without rejecting the audit's
'this is OK' steady state; raise quarterly.
TEST-M2 — 3 advisory deep-scan gates promoted to blocking
Removed continue-on-error: true from:
- gosec (filtered to G201/G202/G304/G108 high-signal rules:
SQL-injection + path-traversal + pprof-exposed)
- osv-scanner (multi-ecosystem CVE; complements govulncheck
which is already blocking in ci.yml)
- trivy image scan (--severity HIGH,CRITICAL --exit-code 1)
continue-on-error count: 15 → 11.
ZAP / schemathesis / nuclei / testssl stay advisory because their
false-positive rates on https://localhost:8443-targeted DAST runs
are high.
TEST-M3 — Playwright harness stub
web/package.json adds '@playwright/test' devDep + 'e2e' / 'e2e:install'
npm scripts. web/playwright.config.ts ships single chromium project
with webServer block pointing at 'npm run dev'. web/src/__tests__/
e2e/smoke.spec.ts proves the harness wires through. The full 15-flow
suite ships in frontend-design-audit Phase 8 (TEST-H1 in THAT audit);
this is the wiring + a single smoke test as the regression floor.
New Makefile target: 'make e2e-test'.
Doc/code drift fixes
====================
TEST-M4 + ARCH-L2 — Skip inventory artifact + CI guard
scripts/skip-inventory.sh walks every t.Skip site under cmd/ +
internal/ + deploy/test/ and emits docs/testing/skip-inventory.md
grouped by package with file:line:expression triples. Current
inventory: 142 t.Skip sites, 76 testing.Short() guards.
scripts/ci-guards/skip-inventory-drift.sh regenerates and fails on
diff (excluding the 'Last reviewed' timestamp line which drifts
daily). The Markdown is the canonical acquisition-diligence artifact
for 'what tests are being skipped and why.'
ARCH-H3 — MCP catalogue floor reconciliation
Audit framing was '121 vs floor 150 — doc/code drift.' Live count
via the test's actual regex over all 5 tool files (tools.go +
tools_audit_fix.go + tools_auth.go + tools_auth_bundle2.go +
tools_est.go): 155 unique 'Name: "certctl_*"' declarations.
Pre-Phase-3 audit measured tools.go in isolation (121) and missed
the other 4 files (+34 unique names). The test at
internal/ciparity/surface_parity_test.go::TestSurfaceParity_MCP
passes today (155 ≥ 150). Added a clarifying comment near
mcpBaselineFloor explaining the measurement scope so future
reviewers don't repeat the audit's framing error.
STATUS: stale — no code drift, just a measurement scoping error in
the audit.
ARCH-L1 — panic() rationale comments
5 panic sites in production Go (excluding _test.go):
- internal/repository/postgres/tx.go:84
- internal/service/issuer.go:861 (mustJSON)
- internal/service/est.go:728 (mustParseTime)
- internal/service/acme.go:1288 (rand source failure — already documented)
- internal/pkcs7/certrep.go:270 (OID marshal — already documented)
Added ARCH-L1 rationale comments to the 3 sites that didn't have
them. All 5 are defensible impossible-path / rethrow / hardcoded-
constant guards.
ARCH-L3 — Migration IF-NOT-EXISTS carve-outs
4 migrations skip the literal 'IF NOT EXISTS' token but ARE
idempotent via different Postgres patterns:
- 000014_policy_violation_severity_check.up.sql: ALTER TABLE
ADD CONSTRAINT CHECK doesn't accept IF NOT EXISTS; idempotency
via DROP CONSTRAINT IF EXISTS preamble.
- 000018_audit_events_worm.up.sql: CREATE OR REPLACE FUNCTION
+ DROP TRIGGER IF EXISTS + CREATE TRIGGER + DO $$ pg_roles
existence check. CREATE TRIGGER doesn't take IF NOT EXISTS.
- 000030_rbac_admin_perms.up.sql: INSERT ... ON CONFLICT DO NOTHING.
- 000039_audit_crit1_perms.up.sql: same INSERT + ON CONFLICT pattern.
Added ARCH-L3 header comments to each explaining the carve-out so
reviewers don't flag the missing literal token.
STATUS: largely stale — migrations are already idempotent.
ARCH-L4 — TODO/FIXME → see #<descriptor>
5 TODOs rewritten to the allowed 'see #<descriptor>' pattern:
- internal/repository/postgres/auth.go:220 → see #bundle-2-scope-fk
- internal/connector/discovery/gcpsm/gcpsm.go:547 → see #gcpsm-pagination
- internal/service/audit.go:244 → see #audit-pagination-count
- internal/service/job.go:295, 299 → see #validation-job-impl
New CI guard scripts/ci-guards/no-todo-in-prod.sh grep-fails any
new TODO/FIXME in cmd/ + internal/ (excluding _test.go); allows
'see #N' / 'see #<descriptor>' patterns.
Sandbox limitation
==================
The 6.1 GB certctl working tree fills the sandbox volume; go1.25.10
toolchain download fails with 'no space left on device' (sandbox has
1.25.9; go.mod requires 1.25.10). Local 'go test' / 'go build' NOT
run in this commit. Operator must run 'make verify' on their
workstation before push per CLAUDE.md operating rules.
The smoke.spec.ts NOT executed in the sandbox (no chromium installed).
Operator runs 'cd web && npm install && npx playwright install
--with-deps chromium && npm run e2e' on first wire-up.
All CI guards (no-todo-in-prod, skip-inventory-drift, G-3
env-docs-drift, doc-rot-detector, and every existing guard) verified
clean by running each individually.
Closes: cowork/certctl-architecture-diligence-audit.html#fix-TEST-H1,
cowork/certctl-architecture-diligence-audit.html#fix-TEST-H2,
cowork/certctl-architecture-diligence-audit.html#fix-TEST-M1,
cowork/certctl-architecture-diligence-audit.html#fix-TEST-M2,
cowork/certctl-architecture-diligence-audit.html#fix-TEST-M3,
cowork/certctl-architecture-diligence-audit.html#fix-TEST-M4,
cowork/certctl-architecture-diligence-audit.html#fix-TEST-L1,
cowork/certctl-architecture-diligence-audit.html#fix-ARCH-H3,
cowork/certctl-architecture-diligence-audit.html#fix-ARCH-L1,
cowork/certctl-architecture-diligence-audit.html#fix-ARCH-L2,
cowork/certctl-architecture-diligence-audit.html#fix-ARCH-L3,
cowork/certctl-architecture-diligence-audit.html#fix-ARCH-L4
|
||
|
shankar0123
|
d60a0ac297 |
fix(security): close BUNDLE 1 — server+agent connector config validation chain
Bundle 1 closure (2026-05-12 acquisition diligence audit). Closes the
acquisition-blocker chain: target.edit (default r-operator grant per
migrations/000029_rbac.up.sql:196) → arbitrary reload_command stored
without validation → agent createTargetConnector json.Unmarshal-only
→ sh -c on agent host. README's 'shell injection prevention on all
connector scripts' claim is now true at the chain level.
Server-side: new internal/connector/target/configcheck package + a
configcheck.Validate call in target.go::Create + ::Update +
::CreateTarget + ::UpdateTarget (all 4 entry points). Rejects shell
metacharacters in reload_command / validate_command / restart_command
for nginx, apache, haproxy, postfix/dovecot, javakeystore, ssh. Sentinel
errors.Is(err, service.ErrInvalidConnectorConfig) available for handler
400 mapping. Non-shell connector types (F5, IIS, Caddy, Traefik, Envoy,
cloud targets, K8s) are no-ops by design.
Agent-side: defense-in-depth connector.ValidateConfig(ctx, configJSON)
call in cmd/agent/main.go inserted between createTargetConnector and
DeployCertificate. This catches (a) configs pre-dating the server gate,
(b) encrypted-blob tampering, (c) per-connector filesystem invariants
that the server can't check.
F5 (S2 finding): proven docs-vs-code drift, not a security bug. The
applyDefaults function never set Insecure=true; runtime default has
always been Go zero-value (false → TLS verified). Three lying 'default
true' comments in f5/f5.go (lines 30, 45-47, 126) rewritten to match
actual code behavior.
Docs (C4 + C9): README L12 + L68 narrowed — 'any CA / any server' →
'Twelve native CA connectors plus an OpenSSL adapter; fifteen native
deployment-target connectors plus a proxy-agent pattern.' 'Every deploy
goes through atomic-write + ...' narrowed to file-based connectors with
inline link to per-target guarantee matrix. New deployment-model.md §1.6
ships a 15-target × 8-property guarantee table covering atomic write /
owner-perms / SHA-256 idempotency / pre-deploy snapshot / on-failure
rollback / post-deploy TLS verify / Prometheus counters / shell-injection
validation — including the K8s preview honesty marker (CLAIM-H4).
Tests: internal/connector/target/configcheck/configcheck_test.go covers
14 shell-injection payloads (semicolon, pipe, backtick, dollar-paren,
redirect, and-chain, newline, double-quote, escape, dollar-var) × 7
shell-using connectors + benign-command acceptance + non-shell no-op
behavior + empty config + malformed JSON. All pass.
Verification (run from /sessions/gifted-blissful-pasteur/mnt/cowork/certctl):
go fmt ./... # clean (no diffs)
go vet ./... # clean (no findings)
go test -short -count=1 ./internal/... ./cmd/...
# 60+ packages all ok, zero FAIL
Audit-Closes: BUNDLE-1 RT-C1 SEC-M4 CLAIM-M2 CLAIM-L3
Audit-Verifies-False: S2 (F5 'default insecure' was a comment lie, code was always secure)
|
||
|
shankar0123
|
0152bdf567 |
fix(auth/rbac): scope-aware ActorRole revoke (A-4)
HIGH-10's UNIQUE (actor, role, scope_type, scope_id, tenant) uniqueness
extension lets an operator grant the same role to the same actor at
multiple scopes (e.g. r-operator on profile=p-acme AND profile=p-globex).
But ActorRoleRepository.Revoke's WHERE clause omitted (scope_type,
scope_id) — a single call deleted every variant. Selective revoke was
unrepresentable; operators had to drop all and re-grant N-1, opening
a race window where the actor's access was briefly different.
Closure across all layers (handler → service → repo → MCP → GUI client),
preserving the legacy "revoke all variants" contract for unmodified
callers:
internal/repository/auth.go
- New ActorRoleRevokeOptions struct. Zero value = legacy semantic;
non-empty ScopeType narrows to one variant.
- New ErrActorRoleNotFound sentinel for scoped no-match (HTTP 404).
internal/repository/postgres/auth.go
- Revoke signature extended with opts. Empty opts.ScopeType uses
the legacy SQL (no scope WHERE), zero-row delete = no error.
- Non-empty narrows with `scope_type = $5 AND scope_id IS NOT
DISTINCT FROM $6` — the IS-NOT-DISTINCT-FROM is load-bearing,
vanilla `=` would silently miss the (global, NULL) case because
NULL ≠ NULL in standard SQL.
- Selective revoke with zero matching rows returns
ErrActorRoleNotFound; operators get feedback on typos.
internal/service/auth/actor_role_service.go
- Revoke takes opts. Audit row's details map records the scope so
SIEMs can distinguish wide-vs-selective revokes:
`scope: "all_variants"` for the legacy path, or
`scope_type` + `scope_id` for selective. Privilege check
(auth.role.assign) and reserved-actor guard unchanged.
internal/api/handler/auth.go
- RevokeRoleFromKey parses optional `?scope_type=` / `?scope_id=`
query params via new parseRevokeScope helper.
- Validation mirrors AssignRoleToKey: scope_id forbidden with
scope_type=global, required with profile/issuer, invalid
scope_type → 400. scope_id without scope_type also → 400.
- writeAuthError maps ErrActorRoleNotFound to 404.
internal/mcp/tools_auth.go + types.go
- AuthRevokeKeyRoleInput gains optional ScopeType + ScopeID with
jsonschema descriptions explaining the dual-mode contract.
- Tool call site appends URL-encoded query params when ScopeType
is set; legacy callers (no scope_type) emit the bare DELETE
path unchanged.
web/src/api/client.ts
- authRevokeKeyRole signature: optional 3rd argument
`{ scope_type?, scope_id? }`. Pre-A-4 call sites (no opts arg)
keep firing the bare DELETE — fully backward compatible. The
GUI KeysPage's per-row revoke button (still one row per role,
pre-Fix-12) continues to use the legacy shape; future GUI work
can pass scope params for per-variant rows.
docs/operator/rbac.md
- New "Revoke: legacy 'all variants' vs scope-selective" subsection
under "From the HTTP API" with curl examples for both modes plus
the audit-row payload shape that lets SOC/SIEM tell them apart.
Regression coverage:
Repository (testcontainers, skipped under -short — 6 tests in
internal/repository/postgres/auth_revoke_scope_test.go):
TestRevokeActorRole_NoOpts_RemovesAllVariants
TestRevokeActorRole_WithScope_RemovesOnlyMatching
TestRevokeActorRole_WithGlobalScope_RemovesOnlyGlobal — pins the
IS-NOT-DISTINCT-FROM branch (global, NULL)
TestRevokeActorRole_NoMatch_ReturnsNotFound — pins the new sentinel
TestRevokeActorRole_NoOpts_NoMatch_IsNoOp — pins the legacy
idempotence contract
TestRevokeActorRole_IssuerScope_RemovesOnlyMatching — pin the
issuer-scope half (profile + issuer are symmetric scope types)
Handler (7 new tests in auth_test.go):
TestAuthHandler_RevokeRoleFromKey — extended to assert no scope
filter is forwarded when query string is empty (legacy behaviour)
TestAuthHandler_RevokeRoleFromKey_A4_ScopedProfile
TestAuthHandler_RevokeRoleFromKey_A4_ScopedGlobal
TestAuthHandler_RevokeRoleFromKey_A4_RejectsScopeIDWithGlobal
TestAuthHandler_RevokeRoleFromKey_A4_RejectsMissingScopeID
TestAuthHandler_RevokeRoleFromKey_A4_RejectsScopeIDWithoutScopeType
TestAuthHandler_RevokeRoleFromKey_A4_RejectsInvalidScopeType
TestAuthHandler_RevokeRoleFromKey_A4_ScopedNotFoundReturns404
MCP (2 new table rows in tools_per_tool_test.go):
Scoped revoke with scope_type=profile + scope_id=p-acme →
`?scope_type=profile&scope_id=p-acme`
Scoped revoke with scope_type=global (no scope_id) →
`?scope_type=global`
Service-layer test plumbing (service_test.go) updated for new opts
arg: 4 existing call sites pass repository.ActorRoleRevokeOptions{}
to keep their pre-A-4 semantics; the fakeActorRoleRepo.Revoke
implementation now mirrors the postgres scope-aware behaviour
(legacy zero-value vs scoped narrowing + ErrActorRoleNotFound on
no-match).
Verify gate green: gofmt clean, go vet clean, go test -short across
repository/postgres, service/auth, api/handler, and mcp. The
pre-existing KeysPage.test.tsx failure observed on the baseline
commit (reproduced via `git stash` earlier in Fix 03) is unrelated;
my client.ts change adds an optional third argument and is fully
backward-compatible.
Spec at cowork/auth-bundles-fixes-2026-05-11/04-high-actor-role-revoke-scope.md.
Audit doc updated: new row A-4 (2026-05-11) CLOSED appended to the
status table at the bottom of cowork/auth-bundles-audit-2026-05-10.md.
Operator-visible advisory in CHANGELOG.md v2.1.0 release notes under
Security (non-BREAKING — legacy callers are unchanged).
Depends on Fix 01 (the scope-aware EffectivePermissions read path on
branch fix/audit-2026-05-11/crit-actor-role-scope-reads). This fix
makes the inverse op selectively reversible; without Fix 01 the read
side would mis-evaluate scoped grants anyway, making selective revoke
moot at runtime.
|
||
|
shankar0123
|
a123263498 |
fix(auth/rbac): close HIGH-10 lying field — EffectivePermissions reads actor-role scope (A-1)
Audit 2026-05-11 A-1 closure. Spec at
cowork/auth-bundles-fixes-2026-05-11/01-crit-actor-role-scope-reads.md.
WHAT.
The HIGH-10 closure (commit
|
||
|
shankar0123
|
912ec3f547 |
fix(audit): ship streaming NDJSON audit export endpoint (HIGH-9 / HIGH-11)
Audit 2026-05-10 HIGH-9 + HIGH-11 closure. HIGH-10 deferred to v3.
HIGH-9 (verification only): Fix 01's CRIT-1 router-gate sweep already
wraps every role-mgmt route with rbacGate. Verified via grep:
- GET /api/v1/auth/roles → auth.role.list
- POST /api/v1/auth/roles → auth.role.create
- GET /api/v1/auth/roles/{id} → auth.role.list
- PUT /api/v1/auth/roles/{id} → auth.role.edit
- DELETE /api/v1/auth/roles/{id} → auth.role.delete
- POST /api/v1/auth/roles/{id}/permissions → auth.role.edit
- DELETE /api/v1/auth/roles/{id}/permissions/{perm} → auth.role.edit
- POST /api/v1/auth/keys/{id}/roles → auth.role.assign
- DELETE /api/v1/auth/keys/{id}/roles/{role_id} → auth.role.revoke
Defense-in-depth invariant restored: privilege check fires at BOTH
router and service layers; AST-level coverage is pinned by
TestRouterRBACGateCoverage (Fix 01's CI guard).
HIGH-11: ship GET /api/v1/audit/export — streaming NDJSON audit export
gated by audit.export. Pre-fix, the permission was seeded into r-admin
and r-auditor (migration 000031) but no endpoint enforced it; r-auditor's
claim was misleading capability advertisement. Post-fix:
- internal/api/handler/audit.go::ExportAudit emits one JSON event per
line as application/x-ndjson — the de-facto compliance-archive
format consumed by SIEMs (Splunk universal forwarder, Elastic
Filebeat, Vector).
- Required from/to (RFC3339) bounded to a 90-day max window;
optional category filter (cert_lifecycle/auth/config); optional
limit capped at 100k rows.
- Content-Disposition: attachment; filename="certctl-audit-<from>_to_<to>.ndjson"
so curl + browser downloads land with a sensible filename.
- Recursively self-audits: every successful export emits an
audit.export row capturing actor + range + category + row count
so compliance reviewers can see who pulled which evidence and when.
- Service layer: AuditService.ExportEventsByFilter reuses the
existing repository.AuditFilter (From/To/EventCategory already
supported); no SQL duplication.
- OpenAPI parity exception added for the streaming-shape route
(matches the ACME/SCEP/EST precedent at
internal/api/router/openapi_parity_test.go::SpecParityExceptions).
Regression matrix in audit_export_test.go (7 cases):
- TestExportAudit_StreamsNDJSONLines (happy path; pins content-type +
content-disposition + JSON-per-line shape + recursive self-audit)
- TestExportAudit_RejectsRangeBeyond90Days (100-day window → 400)
- TestExportAudit_RejectsMissingFromOrTo (3 cases)
- TestExportAudit_RejectsInvalidCategory (unknown enum → 400)
- TestExportAudit_AcceptsValidCategoryFilter (auth filter passes through)
- TestExportAudit_RejectsNonGET (POST → 405)
- TestExportAudit_RejectsToBeforeFrom (inverted range → 400)
The auditor role's surface is now complete (read + export). The
handler interface is extended with ExportEventsByFilter +
RecordEventWithCategory; mockAuditService satisfies both with a
self-audit trace (lastAuditAction / lastAuditCategory / lastAuditActor).
HIGH-10 (scope + expiry on assignRoleRequest): DEFERRED to v3.
Schema column already exists (ActorRole.ExpiresAt); load-bearing wire
remains v3 work. Documented carve-out at HIGH-10's annotation.
Refs: cowork/auth-bundles-audit-2026-05-10.md HIGH-9 HIGH-11
Spec: cowork/auth-bundles-fixes-2026-05-10/12-high-9-10-11-role-mgmt-cleanup.md
|
||
|
shankar0123
|
f5ba17114d |
fix(audit): close silence-leg of HIGH-6; emit WARN on audit-write failure
Audit 2026-05-10 HIGH-6 partial closure (silence leg). The audit
identified two distinct gaps in the auth surface's audit-emit pattern:
(1) silence — `_ = audit.RecordEventWithCategory(...)` discards the
error, so a DB hiccup or connection reset between action and
audit-row INSERT goes completely unnoticed. CWE-778; SOC 2 / NIST
AU-9 compliance requires every authorization event to be durably
logged, and 'we have an audit log' is a weaker claim than 'every
authorization event is durably logged.'
(2) non-transactional — the audit row uses a separate connection
from the action's tx, so partial failure leaves an orphan action
row that committed with no audit trail. Decision 8 of the
auth-bundles-index requires action + audit row atomic.
This commit closes leg (1) fully across all six audit-emit call sites
in the auth surface:
- internal/service/auth/actor_role_service.go::recordAudit
- internal/service/auth/role_service.go::recordAudit
- internal/auth/bootstrap/service.go::ValidateAndMint
- internal/auth/breakglass/service.go::recordAudit
- internal/auth/session/service.go::recordAudit
- internal/api/handler/auth_session_oidc.go::recordAudit
- internal/service/profile.go::Update (Phase 9 approval-bypass)
Each `_ = ...` swallow is replaced with:
if err := audit.RecordEventWithCategory(...); err != nil {
slog.WarnContext(ctx, '<surface> audit write failed (action
committed; audit row may be missing)',
'action', action, 'actor_id', actor, 'resource_id', resource,
'err', err)
}
Operators monitoring audit-write failures now see structured WARN
logs with action + actor + resource attribution; missing audit rows
can be cross-referenced against monitoring without manual SELECT-from-
audit-table.
Infrastructure for leg (2) (transactional commit) is also landed in
this commit:
- service.AuditService.RecordEventWithCategoryWithTx (new method;
accepts repository.Querier from postgres.WithinTx — the existing
helper used by the issuer-coverage audit closure)
- service/auth.AuditService interface declares the new method
- test stub fakeAudit.RecordEventWithCategoryWithTx satisfies the
extended interface
The eight per-path WithinTx-refactors documented in
cowork/auth-bundles-fixes-2026-05-10/10-high-6-atomic-audit-commit.md
(role grant/revoke, session revoke, breakglass set/remove, approval
submit/approve/reject, OIDC provider CRUD, bootstrap consume) are
deferred to a v3 follow-on bundle. Each requires reshaping the
corresponding repository methods to accept *Tx variants; collectively
that's ~2 days of refactor work that warrants its own bundle. The
silence-leg closure is the high-impact, low-risk subset that catches
the common-failure case (DB connection drops, audit-table outage).
Refs: cowork/auth-bundles-audit-2026-05-10.md HIGH-6
Spec: cowork/auth-bundles-fixes-2026-05-10/10-high-6-atomic-audit-commit.md
|
||
|
shankar0123
|
5d79e53ad0 |
auth-bundle-1 follow-on: close coverage gaps to clear Phase 12 floors
CI run #486 (post-Bundle-1 merge + Go 1.25.10 bump) failed three coverage-threshold gates: internal/api/handler 74.7% < floor 75 (-0.3pp) internal/auth 66.3% < floor 85 (-18.7pp) internal/service/auth 51.1% < floor 85 (-33.9pp) The Phase 12 gate file's "85% with negative-test coverage" claim turned out to be aspirational — the read-side and Update-path methods on RoleService / PermissionService / ActorRoleService had zero unit-test coverage, and internal/auth's keystore + HasPermission helper had zero tests. This commit closes the gap without lowering the gate. Per-package CI-style averages after this commit (per scripts/check-coverage-thresholds.sh's per-function-mean): internal/api/handler 76.1% (+1.4pp, margin +1.1pp) internal/auth 90.5% (+24.2pp, margin +5.5pp) internal/service/auth 93.7% (+42.6pp, margin +8.7pp) Tests added: internal/service/auth/service_test.go (+18 tests, +518 LOC): PermissionService.List, PermissionService.GetByName, RoleService.Get (4 paths), RoleService.List (system caller), RoleService.Update (4 paths), RoleService.ListPermissions (3 paths), RoleService.AddPermission/RemovePermission round-trip + gate paths, RoleService.Delete (success + nil-caller + no-perm + audit), RoleService.Create (nil-caller), ActorRoleService.ListForActor (self-bypass + cross-actor + nil-caller + system + with-perm), ActorRoleService.Effective- Permissions (same shape), ActorRoleService.ListKeys (3 paths + system bypass), ActorRoleService.Revoke (4 paths), Authorizer edge cases (empty actorID short-circuit, empty tenantID default, scoped-grant-without-scope-id no-match invariant, repo-error wrap-and-return, HoldsAnyOf early-exit), recordAudit nil-arm short-circuits. internal/auth/keystore_test.go (NEW, +175 LOC): StaticKeyStore.Len, StaticKeyStore.LookupByHash hit + miss, MutableKeyStore seeded lookup + Len, Add registers new key, AddHashed registers from precomputed hash, AddHashed replaces on duplicate hash (idempotent boot-loader contract), HasPermission no-actor / default-actor-type / checker-error / scoped-check threading. internal/auth/bootstrap/service_test.go (+36 LOC): Service.Available nil-receiver/nil-strategy short-circuit, Service.Available delegates to Strategy when configured. internal/api/handler/auth_test.go (+208 LOC): GetRole returns role + permissions, GetRole 404 + 401, UpdateRole 200 + invalid-JSON-400 + 401, ListKeys returns actor list + 401, RemoveRolePermission 204 (global + scoped) + 401, rolePermToResponse scope encoding pin via GetRole. Verified: gofmt -l . clean (touched files only). go vet ./internal/auth/... ./internal/service/auth/... ./internal/api/handler/ rc=0. go test -count=1 -short on the four packages green. CI-style per-function averages computed via the live scripts/check-coverage-thresholds.sh arithmetic — all three gated packages clear their floors with margin. Per CLAUDE.md "complete path" + "do not lower the gate to make CI green": gate file unchanged. The 85/85/75 floors stand. |
||
|
shankar0123
|
cbb47aaf5d |
auth-bundle-1 Phase 11 + 12: RBAC MCP tools + negative-test coverage gate
# Phase 11 — RBAC MCP tools
12 new tools in internal/mcp/tools_auth.go mirroring the Phase-4
+ Phase-7 HTTP surface so operators driving certctl from Claude
/ VS Code / any MCP client get the same management capability
the GUI + CLI already expose:
certctl_auth_me GET /v1/auth/me
certctl_auth_list_roles GET /v1/auth/roles
certctl_auth_get_role GET /v1/auth/roles/{id}
certctl_auth_create_role POST /v1/auth/roles
certctl_auth_update_role PUT /v1/auth/roles/{id}
certctl_auth_delete_role DELETE /v1/auth/roles/{id}
certctl_auth_list_permissions GET /v1/auth/permissions
certctl_auth_add_permission_to_role POST /v1/auth/roles/{id}/permissions
certctl_auth_remove_permission_from_role DELETE /v1/auth/roles/{id}/permissions/{perm}
certctl_auth_list_keys GET /v1/auth/keys
certctl_auth_assign_role_to_key POST /v1/auth/keys/{id}/roles
certctl_auth_revoke_role_from_key DELETE /v1/auth/keys/{id}/roles/{role_id}
Each tool routes through the existing HTTP client (no parallel
business logic), so permission gates fire server-side: a
non-admin caller's MCP tool invocation returns whatever 403 the
underlying HTTP handler emits, fenced via errorResult for LLM-
prompt-injection defense.
Input types in internal/mcp/types.go (AuthRoleIDInput,
AuthCreateRoleInput, AuthUpdateRoleInput,
AuthRolePermissionGrantInput, AuthRolePermissionRevokeInput,
AuthAssignKeyRoleInput, AuthRevokeKeyRoleInput) carry
jsonschema descriptions so the MCP consumer's tool catalogue
shows operator-friendly hints.
internal/mcp/tools_auth_test.go ships 14 tests:
- TestAuthMCP_AllToolsRegister (registration must not panic)
- TestAuthMCP_PathsAndMethods (table-driven, 12 rows pinning
each tool's HTTP method + URL)
- TestAuthMCP_ForbiddenSurfacesFencedError (12 tools × 403
mock → error surface)
internal/mcp/tools_per_tool_test.go's allHappyPathCases extended
with the 12 new rows so the in-memory dispatch coverage gate
(TestMCP_RegisterTools_DispatchableToolCount) stays green at the
new total of 139 registered tools.
Re-derived total via 'grep -cE "gomcp\.AddTool\(" internal/mcp/tools*.go':
133 (121 in tools.go + 12 in tools_auth.go).
# Phase 12 — negative-test coverage gate
Audit of the prompt's 12 negative-test paths against existing
coverage:
1. Missing actor → 401 ✓ TestRequirePermission_NoActorReturns401, TestRBACGate_NoActorReturns401
2. No roles → 403 ✓ TestRequirePermission_DeniedActorReturns403, TestRBACGate_AuditorRole_403sOnAdminRoutes
3. Role lacks specific perm → 403 ✓ same suite
4. Wrong scope → 403 ✓ TestAuthorizer_SpecificScopeMatchesExactID (wrongID arm)
5. Self-grant w/o auth.role.assign → 403 ✓ TestActorRoleService_GrantRequiresAuthRoleAssign
6. Bootstrap token wrong → 401 ✓ TestEnvTokenStrategy_WrongTokenReturnsInvalidToken, TestBootstrapHandler_Mint_WrongToken_401
7. Bootstrap used twice → 410 ✓ TestEnvTokenStrategy_OneShotConsumption, TestBootstrapHandler_Mint_TwiceReturns410
8. Bootstrap when admin exists → 410 ✓ TestEnvTokenStrategy_AdminExistsClosesPath, TestBootstrapHandler_Mint_AdminExists410
9. Role delete with assignees → 409 NEW: TestRoleService_DeleteWithActorsAssignedReturns409
10. Profile-edit loophole → gated ✓ TestProfileEdit_RequiresApprovalLoopholeClosed
11. Permission not in catalog → 400 ✓ TestRoleService_AddPermissionRejectsNonCanonical
12. Scope ID for nonexistent resource → 404 (validation deferred — no FK constraint between role_permissions.scope_id and the resource tables; documented for a future bundle)
Filled the gap at #9 with TestRoleService_DeleteWithActorsAssignedReturns409
which pins the repository sentinel pass-through (postgres FK
ON DELETE RESTRICT → repository.ErrAuthRoleInUse → service
returns the sentinel verbatim → handler maps to HTTP 409).
# Coverage gates
.github/coverage-thresholds.yml gains 2 entries:
- internal/auth: floor 85
- internal/service/auth: floor 85
.github/workflows/ci.yml's coverage test command extended with
./internal/auth/... and ./internal/api/router/... so the
threshold check has data to evaluate.
# Protocol-endpoint not-gated test (Category F)
internal/api/router/phase12_protocol_allowlist_test.go (new)
adds 3 router-level invariant tests:
- TestPhase12_ProtocolEndpointsNotGated: AST-walks router.go,
asserts no rbacGate(...) call references a path under any
protocol-endpoint prefix (/acme, /scep, /.well-known/est,
/.well-known/pki/ocsp, /.well-known/pki/crl).
- TestPhase12_IsProtocolEndpoint_CoversCanonicalPrefixes:
pins auth.IsProtocolEndpoint against the canonical prefix
set; if a future protocol lands without lockstep allowlist
update, this fails.
- TestPhase12_RBACGateRoutesAreUnderAPIv1: belt-and-braces —
every rbacGate-wrapped route MUST start with /api/v1/.
Catches accidental cross-prefix wraps.
Complements the existing TestRequirePermission_ProtocolEndpointBypassesGate
(middleware-level) + TestRouter_AuthExemptAllowlist_PinsActualRegistrations
(allowlist drift) so the Category F invariant is pinned at all
three layers (middleware + router + dispatch).
# Verifications
* gofmt clean repo-wide.
* go vet ./... clean.
* staticcheck across internal/auth + handler + router + cli +
service + repository + cmd + domain + mcp: clean.
* go test -short -count=1 green across internal/auth (incl.
bootstrap), internal/api/handler, internal/api/router,
internal/cli, internal/service (incl. auth),
internal/domain/auth, internal/mcp, cmd/server, cmd/cli.
|
||
|
shankar0123
|
69a508dfcf |
auth-bundle-1 Phase 9 + 10: approval-bypass closure + RBAC GUI
# Phase 9 — approval-bypass closure (Decision 9, option a)
* Migration 000033_approval_kinds.up.sql: ALTER TABLE
issuance_approval_requests ADD COLUMN approval_kind +
payload JSONB; relax certificate_id + job_id to nullable;
CHECK (approval_kind IN ('cert_issuance','profile_edit'))
+ CHECK (per-kind nullability invariant) + index on
approval_kind. Idempotent throughout via DO blocks.
* domain.ApprovalKind enum (cert_issuance / profile_edit) +
IsValidApprovalKind. ApprovalRequest gains Kind +
Payload []byte for the pending profile diff.
* postgres.ApprovalRepository.Create + scanApprovalRow extended
to round-trip the new columns; certificate_id + job_id
switched to sql.NullString so profile_edit rows persist
cleanly. Default Kind=cert_issuance preserves back-compat
for every Phase-7-2026-05-03 caller.
* ApprovalService.RequestProfileEditApproval: new entry point
that creates a pending profile-edit row carrying the
serialized profile diff. Bypass mode (CERTCTL_APPROVAL_BYPASS)
short-circuits the same way it does for cert_issuance.
* ApprovalService.SetProfileEditApply hook: cmd/server/main.go
registers a closure that deserializes req.Payload + persists
via profileRepo.Update + emits a profile.edit_applied audit
row with category=auth. The hook avoids the Approval ↔
Profile import cycle.
* ProfileService.UpdateProfile: gates when (a) the live
profile carries RequiresApproval=true, OR (b) the proposed
edit would set it true. Returns ErrProfileEditPendingApproval
with the new approval ID; ProfileHandler maps to HTTP 202
Accepted + {pending_approval_id}. Both arms close the
flip-flop loophole because every transition through an
approval-tier profile fires the gate.
* TestProfileEdit_RequiresApprovalLoopholeClosed pins all 3
bypass attempts (flip-off / kept-on / flip-on) gated; nil-
approval-service preserves pre-Phase-9 direct-apply for
test fixtures.
* Approval service tests gain 4 profile_edit rows: pending row
shape; same-actor self-approve rejected with
ErrApproveBySameActor (load-bearing two-person integrity);
approve fails-closed when apply callback unwired;
apply callback invoked on approve.
* docs/reference/profiles.md (new) explains the gate +
edit response shape (202) + same-actor invariant + bypass
+ audit hooks.
# Phase 10 — RBAC management GUI
* useAuthMe hook (web/src/hooks/useAuthMe.ts): TanStack Query
fetches /api/v1/auth/me on app boot, caches for 60s, exposes
hasPerm(p) + hasAnyPerm + isAdmin predicates. Every Phase-10
page consumes this on mount + gates affordances against the
cached effective_permissions slice. Server-side enforcement
is the load-bearing gate; client-side hide/disable is UX.
* New routes:
- /auth/roles — list (auth.role.list); create-role modal
(auth.role.create) hidden when missing.
- /auth/roles/:id — detail + permissions; edit
(auth.role.edit), delete (auth.role.delete), add/remove
permission affordances each gated.
- /auth/keys — list of every actor with role grants; assign
+ revoke modals (auth.role.assign). actor-demo-anon
flagged system-managed; mutation buttons hidden for it.
- /auth/settings — stub showing /v1/auth/me identity +
bootstrap-endpoint availability via /v1/auth/bootstrap.
* AuditPage extended with category filter ('All categories'
+ the 3 enum values from migration 000032). Selection flows
to the API call params + the URL-driven query state.
* Layout: 3 new nav entries (Roles / API Keys / Auth Settings).
* api/client.ts: 12 new exported functions for the RBAC
surface (authMe, list/get/create/update/delete role,
list/add/remove role permissions, list keys, assign/revoke
key role, bootstrap-availability probe).
* data-testid attributes on every interactive element so a
future Playwright suite can assert behavior without brittle
CSS selectors.
* Empty state, error state, and unsaved-changes warnings on
every form per the prompt's implementation rules.
# Frontend tests
* RolesPage.test.tsx (6 tests): list render, empty state,
error state, hide-create-button-without-perm,
show-create-button-with-perm, submit-create-modal.
* KeysPage.test.tsx (3 tests): demo-anon flagged
system-managed (no buttons), permission-gated affordance
hide for auditor caller, assign-modal-POST contract.
* AuthSettingsPage.test.tsx (2 tests): identity surface,
bootstrap-OPEN-status surface.
* AuditPage.test.tsx (+1): category-filter select renders
with the 4 documented options.
15 frontend tests total in src/pages/auth/ + the audit
category-filter test; all pass via npx vitest run.
# Verifications
* go vet ./... clean.
* staticcheck across internal/auth + handler + router + cli +
service + repository + cmd + domain: clean.
* gofmt -l clean repo-wide.
* go test -short -count=1 green across internal/service,
internal/api/handler, internal/api/router, internal/auth,
internal/auth/bootstrap, internal/service/auth,
internal/domain/auth, cmd/server, cmd/cli, internal/cli.
* npx tsc --noEmit clean.
* npm run build green (vite build produces dist/index.html
+ 946KB JS bundle; chunk-size warning is pre-existing).
* npx vitest run src/pages/auth/ src/pages/AuditPage.test.tsx
green (15 tests, 4 files).
|
||
|
shankar0123
|
af4fa12724 |
auth-bundle-1 Phase 8 follow-up: classify issuer/target audit rows + auditor end-to-end tests + gofmt drift
Self-audit caught five real gaps in 3ef45e2; this commit closes them. # Phase 8 — issuer/target audit rows now classified as 'config' The Phase 8 prompt explicitly required existing config-mutation calls (issuer config, target config, etc.) to write event_category=config. The |
||
|
shankar0123
|
3ef45e2ad4 |
auth-bundle-1 Phase 6-7-8: bootstrap path + scope-down CLI + auditor-role split
# Phase 6 — day-0 admin bootstrap * internal/auth/bootstrap/ (new package): Strategy interface + EnvTokenStrategy with constant-time compare, one-shot consumption via sync.Mutex, optional admin-existence probe. Bundle 2's OIDC- first-admin will plug in alongside as an alternate Strategy. * BootstrapService.ValidateAndMint: validates the operator's CERTCTL_BOOTSTRAP_TOKEN, mints a 32-byte (64-hex-char) random API key value, persists the SHA-256 hash to api_keys, grants r-admin via actor_roles, AddHashed's the runtime keystore so the just- minted key authenticates the next request without restart, and records bootstrap.consume to the audit trail with category=auth. * internal/auth/keystore.go (new): KeyStore interface + StaticKeyStore (immutable env-var-only path) + MutableKeyStore (env-var keys + DB-loaded api_keys + runtime AddHashed). The auth middleware now consumes a KeyStore so the bootstrap path can extend the lookup table at runtime. * migrations/000031_api_keys.up/down.sql: api_keys table with (id, name UNIQUE, key_hash UNIQUE, tenant_id, admin, created_by, created_at, expires_at, last_used_at). Idempotent. * /v1/auth/bootstrap GET (probe) + POST (mint) — auth-exempt. Both routes documented in api/openapi.yaml + AuthExemptRouterRoutes allowlist updated. The token never leaves internal/auth/bootstrap; the minted plaintext key flows only into the HTTP response body. * Startup warning emitted when CERTCTL_BOOTSTRAP_TOKEN is set AND admin actors already exist (config drift signal). * Tests: 4 strategy invariants (empty token born disabled, wrong token=ErrInvalidToken without consumption, one-shot consumption, admin-exists closes path), 5 service tests (happy path + actor- name validation + propagation of strategy errors + nil-deps guard + 32-byte entropy budget), 8 HTTP-handler tests (status 201/410/401/400 mapping + token-leak hygiene scan of slog + audit details + Location header). Token-leak test redirects slog.Default to a buffer for the test scope. # Phase 7 — API-key migration + scope-down CLI * GET /v1/auth/keys handler + service method ListKeys backed by ActorRoleRepository.ListDistinctActors. Returns one row per (actor_id, actor_type) pair with the slice of role IDs they hold. Permission: auth.role.list. * internal/cli/auth_scope_down.go: AuthListKeys, AuthScopeDown (interactive), AuthScopeDownNonInteractive (JSON config), AuthScopeDownSuggest (--suggest with optional --apply). The synthetic actor-demo-anon is filtered out of every interactive / bulk path; non-interactive flow logs and skips it explicitly. * SuggestRoleFromAuditEvents (pure function): walks 30 days of audit events per actor and returns the narrowest matching role (admin / mcp / viewer / agent / operator) plus a one-line reason. Classification: any admin-shaped action wins; otherwise all-MCP → mcp; all-read-only → viewer; all-agent-shaped → agent; otherwise operator. Test table pins all six classifications. * CLI subcommand tree extended: 'auth keys list' + 'auth keys scope-down [--non-interactive <cfg>] [--suggest [--apply]]'. * CHANGELOG.md leads v2.1.0 with the SECURITY: AUDIT YOUR API KEYS call-out + four flow examples. # Phase 8 — auditor role + event_category column * migrations/000032_audit_category.up/down.sql: ALTER TABLE audit_events ADD COLUMN event_category TEXT NOT NULL DEFAULT 'cert_lifecycle' + CHECK constraint (cert_lifecycle/auth/config) + (event_category) and (event_category, timestamp DESC) indexes for the auditor-filter query path. WORM trigger from migration 000018 continues to enforce append-only at the DB layer (DDL is not blocked). * domain.AuditEvent gains EventCategory string (omitempty); domain.EventCategoryCertLifecycle / Auth / Config constants. * AuditService.RecordEventWithCategory sibling of RecordEvent; legacy callers stay on RecordEvent (defaults to cert_lifecycle). Auth callers (RoleService, ActorRoleService, BootstrapService) switched to RecordEventWithCategory(..., 'auth', ...). * GET /v1/audit?category=<cat>: handler accepts the optional query param, validates against the enum (400 on invalid value), dispatches through ListAuditEventsByCategory. OpenAPI updated with the new query param + AuditEvent.event_category schema. * Postgres AuditRepository.Create now writes event_category; AuditRepository.List filters on it; AuditFilter.EventCategory gates the WHERE clause. * Tests: 5 audit-category-filter HTTP tests (dispatch routing, back-compat fallback, 400 for invalid values, all 3 enum values accepted, page+category combine, JSON output surfaces the field). 3 auditor-role invariants (auditor holds exactly audit.read+audit.export, no mutating perms, disjoint from viewer except audit.read). # Cross-phase wiring * HandlerRegistry.Bootstrap field added; cmd/server/main.go wires the bootstrap service ahead of RegisterHandlers (extracted assembleNamedAPIKeys helper into auth_backfill.go, moved the keystore + bootstrap construction up alongside the auth repos). * AuthCheckResolver / AuthActorRoleService extended with ListKeys to satisfy the Phase 7 surface; existing fakes updated. * fakeAudit + mockAuditService stubs in tests gain RecordEventWithCategory + ListAuditEventsByCategory; existing tests untouched. # Verifications * gofmt -l: clean across every modified file. * go vet ./...: clean. * staticcheck across internal/auth + handler + router + cli + service + repository + cmd + domain: clean. * go test -short -count=1: green across every Bundle-1-touched package — internal/auth (incl. bootstrap), internal/api/handler, internal/api/router, internal/cli, internal/service/auth, internal/service, internal/domain/auth, internal/repository/postgres, cmd/server, cmd/cli, plus internal/scheduler, internal/api/middleware, cmd/agent, internal/mcp. |
||
|
shankar0123
|
bd54d5f7fa |
auth-bundle-1 Phase 2: RBAC service layer + Authorizer primitive
Bundle 1 / Phase 2: ships PermissionService, RoleService, ActorRoleService, and the Authorizer primitive that Phase 3 RequirePermission middleware calls on every gated request.
Authorizer.CheckPermission semantics: a grant matches when (a) the permission name equals the requested permission AND (b) the grant is global-scoped OR the grant scope_type+scope_id exactly match the request. Global beats specific; per-resource grants widen the effective set rather than shadowing global. Hot-path query is one ActorRoleRepository.EffectivePermissions JOIN call (already shipped in Phase 1) plus an in-memory walk; Phase 12 will add benchmarks + caching if the JOIN cost shows up at scale.
Privilege-escalation guard: ActorRoleService.Grant and Revoke require the caller to hold auth.role.assign globally. Without it, ErrSelfRoleAssignment. System callers (AsSystemCaller()) bypass the check; bootstrap, migrations, scheduler-initiated grants use this path. Reserved actor actor-demo-anon is rejected on Grant + Revoke so the demo path stays alive even after a misclick (ErrAuthReservedActor).
Caller abstraction: every service entry point takes *Caller (ActorID, ActorType, TenantID, IsSystem). CallerFromContext is a stub returning ErrUnauthenticated; Phase 3 wires the middleware-context bridge that fills the Caller from request context. The contract is pinned by TestCallerFromContext_Phase2ReturnsUnauthenticated so the Phase 3 upgrade is observable.
Audit recording: every mutating service operation calls AuditService.RecordEvent. Bundle 1 Phase 8 adds the event_category column + parameter and back-fills 'auth' for these calls; until then the rows go in with the default category.
Test coverage: in-memory fakeRoleRepo / fakePermissionRepo / fakeActorRoleRepo / fakeAudit pin the privilege-escalation invariants (ErrUnauthenticated for nil caller, ErrForbidden for missing perm, ErrInvalidPermission for non-canonical permission name, ErrSelfRoleAssignment for Grant without auth.role.assign, ErrAuthReservedActor for actor-demo-anon mutations, system-caller bypass) without requiring testcontainers. Phase 12 will add live-Postgres integration coverage.
Branch: dev/auth-bundle-1. Phase 1 was
|
||
|
shankar0123
|
0e06f6c4fc |
cli: promote --force on renew + require --reason on revoke (closes P3-1, P3-2)
Closes findings P3-1 and P3-2 from the 2026-05-05 CLI/API/MCP↔GUI parity
audit (cowork/cli-gui-parity-audit-2026-05-05/RESULTS.md). Both findings
flagged hidden defaults that the CLI was sending without exposing them
to operators: `force=false` baked into every renew payload, and a silent
fallback to `reason="unspecified"` whenever --reason was omitted.
P3-1 — promote --force on `certs renew` (full end-to-end plumbing)
The pre-2026-05-05 CLI sent `{"force": false}` in the renew body. The
API handler never decoded it — a textbook "lying field" per the
operator's CLAUDE.md "complete path, not the easy path" rule: the body
field stored a value, claimed to do something, and silently did nothing
because the wire never reached the consumer. Adding a --force flag that
also went unread would have created another lying field.
This commit takes the complete path:
service.CertificateService.TriggerRenewal grew a `force bool` parameter
(internal/service/certificate.go). When force=true, the
RenewalInProgress block is overridden so operators can recover stuck
in-flight renewals where a previous job hung without releasing the
status flag. Archived and Expired remain terminal blockers regardless
of force — those are semantic dead-ends that --force should not paper
over (archived = decommissioned, expired = issue a new cert instead of
renewing a dead one).
handler.CertificateHandler.TriggerRenewal parses force from
?force=true (or ?force=1) query param, OR {"force": true} JSON body,
whichever the client picks. Defaults to false. Passes through to the
service.
internal/cli/client.go::RenewCertificate(id, force bool) sends
?force=true on the URL when --force is set. The historical hardcoded
`{"force": false}` body is gone — no more lying field.
cmd/cli/main.go dispatches `certs renew <id> [--force]` (ID-first
flag-second convention matches the existing `agents retire <id>
[--force]`).
P3-2 — require --reason on `certs revoke` (Option A: strict refusal)
The pre-2026-05-05 CLI dropped to `--reason unspecified` whenever the
operator omitted the flag. Compliance reporting (RFC 5280 §5.3.1, PCI-
DSS §3.6, HIPAA §164.312) relies on the reason code being meaningful;
silent fallback defeats the audit trail because every revocation looks
identical.
cmd/cli/main.go dispatch refuses to send when --reason is empty,
prints the canonical RFC 5280 §5.3.1 reason-code menu, and exits
non-zero.
internal/cli/client.go exposes ValidRevokeReasons() returning the
canonical camelCase list (unspecified, keyCompromise, caCompromise,
affiliationChanged, superseded, cessationOfOperation, certificateHold,
removeFromCRL, privilegeWithdrawn, aaCompromise) and
NormalizeRevokeReason() that accepts both camelCase and snake_case
inputs and normalises to the canonical wire form. Off-list reasons
are rejected at dispatch with the menu re-printed.
Test pins:
internal/cli/client_test.go::TestClient_RenewCertificate_ForceFlag —
--force=true sends ?force=true with empty body; --force=false sends
no query and no body.
internal/cli/client_test.go::TestNormalizeRevokeReason +
TestValidRevokeReasons — canonical-camelCase + snake_case + reject-
off-enum behaviour.
cmd/cli/dispatch_test.go::TestHandleCerts_Revoke_RequiresReason +
TestHandleCerts_Revoke_RejectsUnknownReason +
TestHandleCerts_Renew_ForceFlag — dispatch-layer pins for the same
contracts.
internal/api/handler/certificate_handler_test.go::TestTriggerRenewal_
ForceQueryParam — query-param passthrough (no-flag, force=true,
force=1, force=false) flows through to the service-layer parameter.
internal/service/certificate_test.go::TestTriggerRenewal_
ForceOverridesInProgress — force=false preserves the
RenewalInProgress block; force=true clears it.
Existing TestTriggerRenewal_Archived extended to assert force=true
still blocks Archived (terminal-state guarantee).
Docs: docs/reference/cli.md updated with the --force example for renew
and the strict --reason semantics for revoke (including snake_case
input acceptance).
Acceptance gate (verified):
- go build ./cmd/server/... ./cmd/agent/... ./cmd/cli/...
./cmd/mcp-server/... clean.
- go vet ./... clean.
- go test -short -count=1 ./... pass repo-wide.
- bash scripts/ci-guards/openapi-handler-parity.sh clean
(router 178, OpenAPI 144, exceptions 36 — unchanged; we add
parameter parsing, not routes).
- gofmt -l clean.
|
||
|
shankar0123
|
75097909e9 | |||
|
shankar0123
|
b0fc067317 |
security: close CodeQL #17 (log injection) + #23 (SSRF false-positive reopen)
Two CodeQL alerts in one sweep — both medium-impact follow-ups
on already-merged guards.
Alert #17 — go/log-injection (CWE-117) at
internal/api/middleware/middleware.go:58:
log.Printf("[%s] %s %s %d %v", requestID, r.Method, r.URL.Path, ...)
r.Method and r.URL.Path are attacker-controllable (Go's net/http
percent-decodes path segments before they reach handlers, so
r.URL.Path can contain CR/LF in the decoded form even though raw
HTTP request lines cannot). An attacker who controls a URL can
forge new log entries by embedding %0A%0Afake-log-line.
Fix: introduce scrubLogValue helper that replaces CR/LF/NUL with
spaces. Apply to both r.Method and r.URL.Path. Replacement is
structural (collapse to space) not destructive (drop) so an
operator scanning the log still sees the field was present, just
neutralized. Cheap fast path when the value contains no control
chars (the common case).
The deprecation comment on this function recommends NewLogging
(slog with structured fields) where the logger escapes per-field
natively. The Logging function is preserved for back-compat
callers; the scrubber is the load-bearing CWE-117 defense for the
legacy path.
Alert #23 — go/request-forgery (CWE-918) at scep_probe.go:271:
CodeQL reopened the alert after commit
|
||
|
shankar0123
|
e6919cdaba |
security(scep_probe): re-validate URL inside scepHTTPGet to close CodeQL #23 (CWE-918)
CodeQL alert #23 (go/request-forgery, CWE-918 SSRF) flagged the client.Do(req) sink at internal/service/scep_probe.go:232 because the URL parameter to scepHTTPGet is taint-traced from the user- supplied input to ProbeSCEP without the analyzer recognizing the upstream sanitizer. The defense-in-depth was already in place: 1. validation.ValidateSafeURL at ProbeSCEP entry (line 75) — rejects obvious SSRF targets (loopback / link-local / cloud metadata literals) before any network call. 2. validation.SafeHTTPDialContext on the http.Transport — re-resolves the host at dial time and rejects connections to reserved IP ranges. This is the authoritative SSRF + DNS- rebinding guard. Even if step 1 was bypassed, the dial would still fail. But CodeQL's taint tracker doesn't follow the validator across function boundaries, so the alert stays open even though the code is safe. This commit re-runs validation.ValidateSafeURL inside scepHTTPGet immediately before http.NewRequestWithContext — sanitizer in the same function as the sink, which CodeQL recognizes as a guard. Bonus defense-in-depth: any future call site that wires a URL into scepHTTPGet without going through ProbeSCEP (e.g. a new code path that directly probes a discovered URL) inherits the same SSRF guard automatically. Fail-closed by default. The validator dispatch matches ProbeSCEP's pattern — tests override via s.scepValidateURL to hit httptest loopback servers; production callers use validation.ValidateSafeURL. The probe's existing httptest-based tests continue to work unchanged. Verified locally: gofmt: clean. go vet ./...: exit 0. go test -short ./internal/service/...: ok 4.029s (every existing scep_probe test still green — the new revalidation is a no-op for tests that go through ProbeSCEP because the same validator already passed once at entry). Reference: https://github.com/certctl-io/certctl/security/code-scanning/23 Closes CodeQL alert #23 (go/request-forgery). |
||
|
shankar0123
|
62523fb845 |
service: 10 IntermediateCAService tests + in-memory fake repo (Rank 8 commit 2.5)
Service-layer pin for Rank 8. The fake IntermediateCARepository's
WalkAncestry mirrors the postgres recursive-CTE semantics
(leaf-first ordering, terminate at parent_ca_id IS NULL) so the
AssembleChain pin carries the same weight the production repo would.
Tests:
TestIntermediateCA_CreateRoot_RegistersOperatorSuppliedSelfSigned
Happy path. RFC 5280 §3.2 self-signed root + matching key gets
persisted with parent_ca_id=NULL, state=active, KeyDriverID=...
TestIntermediateCA_CreateRoot_RejectsNonSelfSigned
RFC 5280 §3.2 enforcement. Cert whose embedded public key
doesn't match the actual signer fails CheckSignatureFrom →
ErrCANotSelfSigned.
TestIntermediateCA_CreateRoot_RejectsKeyMismatch
Operator-boundary defense in depth. Cert is well-formed
self-signed but the supplied keyDriverID resolves to a
different key → ErrCAKeyMismatch.
TestIntermediateCA_CreateChild_PathLenTighteningEnforced
RFC 5280 §4.2.1.9 enforcement. Child whose path-len equals or
exceeds parent's → ErrPathLenExceeded. Strictly-tighter child
succeeds.
TestIntermediateCA_CreateChild_NameConstraintsSubset
RFC 5280 §4.2.1.10 enforcement. Widening rejected
("evil.com" outside parent's "example.com"); subdomain
narrowing succeeds ("internal.example.com").
TestIntermediateCA_AssembleChain_4DeepHierarchy ← LOAD-BEARING
The pin the local connector tree-mode delegates to. Builds
root → policy → issuing-A → issuing-B and asserts AssembleChain
returns 4 CERTIFICATE blocks in leaf-to-root order with
matching subject CommonNames at each depth.
TestIntermediateCA_Retire_RefusesIfActiveChildren
Drain-first semantics. retiring → retired with active children
refuses with ErrCAStillHasActiveChildren.
TestIntermediateCA_Retire_TwoPhaseConfirm
First call: active → retiring (no confirm). Second call without
confirm: surfaces "pass confirm=true". Second call with
confirm: retiring → retired.
TestIntermediateCA_MetricsRecordedPerOutcome
Snapshot pin. CreateRoot bumps create_root, CreateChild bumps
create_child, Retire(active) bumps retire_retiring, all
dimensioned by issuer_id.
TestIntermediateCA_LoadHierarchy_FlatList
Returns every CA for an issuer ordered by created_at; caller
renders the tree from parent_ca_id.
Test infrastructure:
fakeIntermediateCARepo — sync.Mutex-guarded map.
WalkAncestry walks
parent_ca_id from leafID
to root (or terminates on
cycle, defense-in-depth).
Compile-time interface
guard.
testCAFixture — mints a self-signed root
cert+key in process,
Adopt()s the key under
a stable ref so CreateRoot
can resolve it.
newTestService — wires IntermediateCAService
with fake repo +
signer.MemoryDriver +
mockAuditRepo (already
lives in testutil_test.go)
+ IntermediateCAMetrics.
Verified locally:
gofmt: clean.
go vet ./...: exit 0.
go test -short -count=1 -run TestIntermediateCA ./internal/service/...
PASS (10/10)
go test -short -count=1 ./internal/service/...: ok 3.844s
Reference: cowork/rank-8-intermediate-ca-hierarchy-prompt.md, commit 2.5.
|
||
|
shankar0123
|
fb54ebcb62 |
service: IntermediateCAService + IntermediateCAMetrics + RFC 5280 enforcement
Rank 8 of the 2026-05-03 deep-research deliverable, commit 2 of 5.
Service-layer wiring for first-class N-level CA hierarchy management.
The connector rewrite that activates this surface lands in commit 3.
Files added:
internal/service/intermediate_ca.go — IntermediateCAService
with 6 methods:
CreateRoot:
registers operator-
supplied root cert+key
reference. Validates
RFC 5280 §3.2 self-
signed (subject ==
issuer + signature
verifies). Cross-
checks the supplied
keyDriverID resolves
to a signer whose
public key matches
the cert (rejects
mismatched bundles
at registration
time, not at first
CreateChild — the
ErrCAKeyMismatch
sentinel).
CreateChild:
generates child key
via signer.Driver,
signs the cert via
the parent's signer.
Enforces RFC 5280
§4.2.1.9 (path-len
tightening) +
§4.2.1.10
(NameConstraints
subset semantics) at
service layer fail-
closed. Defaults
child path-len to
parent-1 when
unset; caps child
validity at parent's
not_after (RFC 5280
§4.1.2.5).
Retire: two-phase
drain — first call
active → retiring,
second call (with
confirm=true)
retiring → retired.
Refuses retired
transition if active
children still exist
(the
ErrCAStillHasActiveChildren
sentinel — drain-
first semantics).
Get / LoadHierarchy:
thin repo wrappers.
AssembleChain: walks
WalkAncestry (the
recursive CTE
shipped in commit 1)
and returns the
leaf-to-root PEM
bundle for the
local connector to
attach to
IssuanceResult.
internal/service/intermediate_ca_metrics.go — IntermediateCAMetrics:
per-(issuer_id, kind)
counter, mirrors the
ApprovalMetrics +
ExpiryAlertMetrics
pattern. RecordCreate
(root/child) +
RecordRetire
(retiring/retired).
SnapshotIntermediateCA
for the Prometheus
exposer.
Defense in depth retained:
- NEVER persist CA private key bytes in the row. KeyDriverID is the
only key reference; signer.Driver.Load resolves it at signing time.
- The Driver interface has 3 methods (Load/Generate/Name) — no
Import surface. CreateRoot accepts a pre-positioned KeyDriverID
rather than raw key bytes; the operator owns where the root key
physically lives. Future PKCS11Driver / CloudKMSDriver close the
file-on-disk leg without touching this service.
Verified locally:
gofmt: clean.
go vet ./internal/service/...: exit 0.
go build ./internal/service/...: exit 0.
Deferred to commit 2.5 (or fold into commit 3, operator's call):
- 9 service-level tests including:
* TestIntermediateCA_CreateRoot_RegistersOperatorSuppliedSelfSigned
* TestIntermediateCA_CreateRoot_RejectsNonSelfSigned
* TestIntermediateCA_CreateRoot_RejectsKeyMismatch
* TestIntermediateCA_CreateChild_PathLenTighteningEnforced
* TestIntermediateCA_CreateChild_NameConstraintsSubset
* TestIntermediateCA_AssembleChain_4DeepHierarchy ← LOAD-BEARING
* TestIntermediateCA_Retire_RefusesIfActiveChildren
* TestIntermediateCA_Retire_TwoPhaseConfirm
* TestIntermediateCA_MetricsRecordedPerOutcome
Test setup needs: in-memory IntermediateCARepository fake +
signer.MemoryDriver (already exists) + helper to generate test root
cert+key. Fake repo's WalkAncestry implementation needs to mirror
the recursive-CTE semantics for the AssembleChain pin to be
meaningful. Total ~500 lines of test code; non-trivial setup.
Out of scope of THIS commit (commits 3-5):
- Local connector rewrite + byte-equivalence pin
(TestLocal_HierarchyMode_SingleVsTree_ByteIdentical).
- 4 admin-gated handler endpoints + OpenAPI extension.
- web/src/pages/IssuerHierarchyPage.tsx.
- docs/intermediate-ca-hierarchy.md sysadmin runbook.
- cmd/server/main.go wiring.
Reference: cowork/rank-8-intermediate-ca-hierarchy-prompt.md.
|
||
|
shankar0123
|
31e50d987f |
ci: fix Rank 7 lint + openapi-handler-parity drift on master
Two CI failures from the Rank 7 chain push (#438): Go Build & Test — staticcheck ST1021: internal/service/approval_metrics.go:97 comment for ApprovalDecisionEntry doesn't start with the type name internal/service/approval_metrics.go:130 comment for ApprovalPendingAgeSnapshot doesn't start with the type name Frontend Build — scripts/ci-guards/openapi-handler-parity.sh: 4 router routes have no OpenAPI operationId: GET /api/v1/approvals GET /api/v1/approvals/{id} POST /api/v1/approvals/{id}/approve POST /api/v1/approvals/{id}/reject The Rank 7 commit-3 spec deferred OpenAPI extension to commit 4 with a 'batched alongside the integration changes' note; commit 4 didn't actually add them. This commit closes that gap. Fixes: approval_metrics.go — split the doc comment that was attached to SnapshotApprovalDecisions (the function) but visually preceded ApprovalDecisionEntry (the type), so the type appeared to staticcheck as having a comment that named the function instead of the type. Same fix on ApprovalPendingAgeSnapshot. Now each exported type has its own type-name-leading comment per Go convention. api/openapi.yaml — added 4 new operationIds (listApprovalRequests, getApprovalRequest, approveApprovalRequest, rejectApprovalRequest) + new ApprovalRequest schema component under components/schemas. Inline 401 response (the Unauthorized component does not exist in this spec; the canonical pattern in the rest of the file is inline 'description: Authentication required'). The two-person integrity contract surface is documented in the description of the approve / reject endpoints so external readers see the RBAC contract from the spec alone. Verified locally: go vet ./internal/service/...: exit 0. scripts/ci-guards/openapi-handler-parity.sh: clean (140 ops vs 174 routes, 36 documented exceptions). Third CI failure (image-and-supply-chain) was a transient apt-fetch 'Connection reset by peer' from deb.debian.org while pulling libasan6_10.2.1-6_amd64.deb. Not a code issue; just re-run the workflow. No code change needed. |
||
|
shankar0123
|
aebfd8bd7c |
Revert "chore: drop 'Infisical' label from internal references"
This reverts commit
|
||
|
shankar0123
|
19706e56b3 |
chore: drop 'Infisical' label from internal references
Strategic naming cleanup. Earlier doc-comments + commit messages framed Rank 4 / Rank 5 / Rank 7 work as 'Rank N of the 2026-05-03 Infisical deep-research deliverable' — the 'Infisical' qualifier was a holdover from the original deep-research framing where Infisical (a competing secrets-management platform) was the comparator. Keeping the comparator's name in our source adds noise without value; an external reader sees 'Infisical' and assumes a dependency or shared lineage rather than reading it as the competitive context it was. Mechanical sed across 34 files (32 source / docs + 2 follow-up Python passes to collapse 'deep-research deep-research' duplicates that emerged where the original phrase wrapped across lines): s|Infisical deep-research|deep-research|g s|infisical-deep-research-results|deep-research-results-2026-05-03|g s|infisical-deep-research-prompt|deep-research-prompt-2026-05-03|g s|infisical-deep-research|deep-research|g s|Infisical|deep-research|g s|deep-research deep-research|deep-research|g # collapse-pass Net diff: 63 insertions / 64 deletions across cmd/, docs/, internal/, migrations/. Pure text substitution; zero behavior change. Code path unchanged — go vet clean, tests for TestApproval pass on both internal/service and internal/api/handler packages. Workspace docs (cowork/) carry the same references and will be swept separately — they're not under certctl/ git control. The two filename references (cowork/infisical-deep-research-results.md + cowork/infisical-deep-research-prompt.md) get renamed alongside that sweep to deep-research-results-2026-05-03.md / deep-research-prompt-2026-05-03.md so cross-references in the certctl repo doc-comments resolve cleanly. |
||
|
shankar0123
|
03c61f4c20 |
scheduler, certificate, renewal: gate issuance on profile-driven approval
Closes Rank 7 of the 2026-05-03 Infisical deep-research deliverable
(cowork/infisical-deep-research-results.md Part 5). Pre-fix, certctl
issued certificates unattended — every renewal-loop tick that crossed
a renewal threshold created a Job at Status=Pending which the
scheduler dispatched directly to the issuer connector. PCI-DSS Level
1, FedRAMP Moderate / High, SOC 2 Type II, and HIPAA-regulated PHI
customers all ask the same procurement question: "How do you enforce
two-person integrity on cert issuance?" Today's answer: "We don't."
After this commit chain: "Per-profile RequiresApproval=true creates a
parallel ApprovalRequest row; the renewal-loop creates the Job at
Status=AwaitingApproval; an authorized approver (different from the
requester per the same-actor RBAC check) calls
POST /api/v1/approvals/{id}/approve, transitioning the Job to
Pending; the scheduler picks it up."
This commit (4 of 4) wires the gate into the manual TriggerRenewal
entry point + main.go service construction + Config.Approval +
docs + WORKSPACE-ROADMAP follow-up entries. The previous commits
in the chain shipped:
- 1 (
|
||
|
shankar0123
|
8043e2bbac |
service: ApprovalService + ApprovalMetrics + 8 table-driven tests
Rank 7 of the 2026-05-03 Infisical deep-research deliverable, commit 2 of 4
(cowork/rank-7-approval-workflow-primitive-prompt.md). Builds on the
foundation in commit
|
||
|
shankar0123
|
8b75e0311b |
chore: rename Go module path to github.com/certctl-io/certctl
Mechanical sed across the main go.mod's module declaration, the f5-mock-icontrol
sub-module's go.mod, every Go file's import path (361 files), and a rebuild of
the checked-in f5-mock-icontrol binary so its embedded build-info reflects the
new module path. No behavior change.
Choice B from cowork/transfer-certctl-to-org.md, executed 2026-05-04. Choice A
(keep module path declared as github.com/shankar0123/certctl regardless of
repo URL) shipped on the day of the org transfer (2026-05-03) since we had no
external Go consumers; this commit closes that deferral.
Backward-compat: GitHub HTTP redirects continue to forward
github.com/shankar0123/certctl → github.com/certctl-io/certctl at the URL
level, but Go's module proxy uses the path declared in go.mod as the
canonical name. Pre-fix, anyone trying `go get github.com/certctl-io/certctl/...`
hit a "module path mismatch" error because go.mod said
github.com/shankar0123/certctl and the URL they fetched it from said
certctl-io/certctl. Post-fix, the canonical name and the URL agree, so
go get / go install / external Go consumers / Go-tooling integrations
work cleanly via either the new path (preferred) or the old path (which
redirects and Go follows the redirect for source fetch).
Anyone still importing the old path inside their own code keeps working
provided they update their go.mod's `require` line to match — the module
path declared in their consumer's go.sum / go.mod is the authoritative
import name, so a mass sed across their import statements is the migration
on the consumer side. No external consumers exist today.
Diff shape:
361 *.go files — import path replacement only
2 go.mod — module declaration replacement only
1 binary — deploy/test/f5-mock-icontrol/f5-mock-icontrol rebuilt
so embedded build-info reflects the new path (8618965 vs
8618933 bytes; 32-byte diff is the build-info change)
Total: 364 files, 730 insertions / 730 deletions, net-zero size, pure
mechanical substitution.
Verification:
gofmt: 17 files needed re-alignment after sed (the new path is one char
shorter than the old, so column-aligned import groups drifted). Applied
`gofmt -w` to fix.
go mod tidy: clean exit on both modules.
go vet ./...: clean exit.
go build ./...: clean exit.
go test -short -count=1 on representative packages: all green
(internal/domain, internal/validation, internal/crypto, internal/crypto/signer,
cmd/agent). Test output now reads `ok github.com/certctl-io/certctl/...`
confirming the module path resolves correctly.
binary: f5-mock-icontrol rebuilt; `strings | grep shankar0123` returns
nothing; `strings | grep certctl-io/certctl` shows the new module path
embedded in build-info.
Files intentionally NOT touched in this commit:
README.md / CHANGELOG.md / docs/ / etc. — already swept to certctl-io
URLs in commit
|
||
|
shankar0123
|
8a56a78282 |
target(azurekv): SDK-driven Azure Key Vault target connector
Closes Rank 5 (Azure half) of the 2026-05-03 Infisical deep-research deliverable (cowork/infisical-deep-research-results.md Part 5). Pre-fix, certctl had no path to deploy certs to Azure-managed TLS- termination endpoints (Application Gateway / Front Door / App Service / Container Apps) — operators terminating TLS at Azure had to use manual `az keyvault certificate import` invocations or external automation. This commit lands the SDK-driven Azure Key Vault target connector that closes the gap, mirroring the AWS ACM target shape shipped in commit |
||
|
shankar0123
|
edf6bee7f8 |
target(awsacm): SDK-driven AWS Certificate Manager target connector
Closes Rank 5 (AWS half) of the 2026-05-03 Infisical deep-research
deliverable (cowork/infisical-deep-research-results.md Part 5).
Pre-fix, certctl had no path to deploy certs to AWS-managed TLS-
termination endpoints (ALB / CloudFront / API Gateway / App Runner)
— operators terminating TLS at AWS had to use Infisical secret-sync,
manual aws-cli imports, or external automation. This commit lands
the SDK-driven AWS Certificate Manager target connector that closes
the gap end-to-end.
Architecture:
- internal/connector/target/awsacm/awsacm.go — Connector wraps
*acm.Client behind the ACMClient interface seam (mirrors
awsacmpca's ACMPCAClient pattern from the issuer side).
LoadDefaultConfig handles the standard AWS credential chain
(IRSA / EC2 instance profile / SSO / env vars); no embedded
creds in connector Config.
- Pre-deploy snapshot via DescribeCertificate + GetCertificate so
on-import-failure rollback restores the previous cert. Mirrors
the Bundle 5 IIS pattern + the Bundle 7/8 WinCertStore /
JavaKeystore patterns. Surfaces rollback success/failure via
the existing certctl_deploy_rollback_total Prometheus counter
label set.
- Provenance tags: certctl-managed-by=certctl + certctl-
certificate-id=<mc-id> set automatically on every import. ACM
strips tags on re-import, so the connector calls
AddTagsToCertificate post-import to keep the provenance pair
fresh. Operators looking up a cert ARN by managed-cert ID
(Terraform data source, CloudFormation output) match against
these tags.
- DeploymentRequest.KeyPEM held in agent memory only — never
written to disk. Aligns with the pull-only deployment model
documented in CLAUDE.md.
Tests:
- awsacm_test.go: 15-subtest happy-path + validation matrix
covering ValidateConfig (success / missing-region / malformed-
region / malformed-ARN / reserved-tag rejection),
DeployCertificate (fresh import / rotate-in-place / rollback-
on-serial-mismatch / rollback-also-fails / empty-key-rejected /
no-client-rejected), ValidateOnly (returns sentinel),
ValidateDeployment (serial match / mismatch / no-ARN-yet).
- awsacm_failure_test.go: 5 per-error-class contract tests
mirroring the awsacmpca_failure_test.go shape (commit
|
||
|
shankar0123
|
109f32ff41 |
notifications: per-policy multi-channel expiry-alert routing
Closes Rank 4 of the 2026-05-03 Infisical deep-research deliverable
(see cowork/infisical-deep-research-results.md Part 5). Pre-fix,
RenewalService.CheckExpiringCertificates already ran daily,
RenewalPolicy.AlertThresholdsDays drove per-cert thresholds, and
NotificationService.SendThresholdAlert deduped per (cert, threshold)
— but the channel was hardcoded to Email
(internal/service/notification.go:118 pre-fix). Operators who
configured PagerDuty / Slack / Teams / OpsGenie via
CERTCTL_PAGERDUTY_ROUTING_KEY etc. got nothing at any threshold
unless SMTP was also wired. Their first signal of an expired cert
was a 3 AM outage.
This commit lands the routing matrix on top of the existing
infrastructure:
1. RenewalPolicy gains AlertChannels (per-tier channel list) +
AlertSeverityMap (per-threshold tier assignment) +
EffectiveAlertChannels / EffectiveAlertSeverity accessors.
Default*() helpers preserve the back-compat Email-only
behaviour for operators who haven't touched their policies
post-upgrade. Migration 000026 adds the JSONB columns
idempotently.
2. NotificationService.SendThresholdAlertOnChannel — the new
per-channel dispatch helper. Old SendThresholdAlert stays as
an Email-only alias so non-policy callers (admin "send test
alert" surfaces) keep working byte-for-byte.
3. NotificationService.HasThresholdNotificationOnChannel — per-
(cert, threshold, channel) deduplication so a transient
PagerDuty 5xx today does NOT suppress today's Slack alert and
tomorrow's PagerDuty retry will still fire.
4. RenewalService.sendThresholdAlerts walks the resolved channel
set per threshold tier, fans out to every configured channel,
handles per-channel failures independently, defensively drops
off-enum channels with an audit row trail, and records a per-
channel audit event with metadata.channel + metadata.severity_tier.
5. service.ExpiryAlertMetrics — atomic counter table mirrored on
the VaultRenewalMetrics shape from the 2026-05-03 audit fix #5
(commit
|
||
|
shankar0123
|
0792271dc6 |
vault: add automatic token renewal at TTL/2 + Prometheus metric
Closes Top-10 fix #5 of the 2026-05-03 issuer-coverage audit (see cowork/issuer-coverage-audit-2026-05-03/RESULTS.md). Pre-fix, the VaultPKI adapter authenticated with a static token and never called renew-self. Long-lived deploys hit token expiry; the first operator-visible signal was failed cert renewals on production targets. This commit: 1. Connector.Start(ctx) spawns a goroutine that calls POST /v1/auth/token/renew-self at TTL/2 cadence (computed from a one-shot lookup-self at startup). Honours ctx.Done() for graceful shutdown via a per-loop done channel + Stop(). 2. On `renewable: false` response (initial lookup OR any subsequent renewal), the loop emits a WARN, increments the not_renewable counter, and exits. The operator must rotate the token before Vault's Max TTL elapses. 3. New Prometheus counter certctl_vault_token_renewals_total with labels result={success,failure,not_renewable}. Registered alongside existing certctl_issuance_* counters in internal/api/handler/metrics.go. 4. ERROR-level logging on renewal failure with operator-actionable substring ("vault token renewal failed; rotate the token before TTL expires") so journalctl + grep find it. Loop keeps ticking after a failure — transient blips don't kill it. New optional issuer.Lifecycle interface: type Lifecycle interface { Start(ctx context.Context) error Stop() } Connectors that hold no background goroutines (almost all of them) do not implement this — IssuerRegistry.StartLifecycles / StopLifecycles feature-detect via type assertion. New lifecycle-bearing connectors plug in by implementing the interface; no further registry plumbing required. Wiring (cmd/server/main.go): - service.NewVaultRenewalMetrics() instance is shared between issuerRegistry.SetVaultRenewalMetrics (so Vault connectors built by Rebuild get a recorder) and metricsHandler.SetVaultRenewals (so the Prometheus exposer emits the new series). - issuerRegistry.StartLifecycles(ctx) is called after issuerService.BuildRegistry; defer issuerRegistry.StopLifecycles is paired so goroutines exit cleanly on signal. - IssuerConnectorAdapter.Underlying() exposes the wrapped issuer.Connector so registry-level machinery can reach the concrete connector behind the adapter without duplicating the wiring at every call site. Tests (internal/connector/issuer/vault/vault_renew_test.go): - TestVault_RenewLoop_TickAtHalfTTL — three ticks → three renewals, all "success". - TestVault_RenewLoop_StopsOnNotRenewable — second renewal returns renewable=false, loop exits, third tick fires no HTTP call. - TestVault_RenewLoop_FailureSurfacesViaMetric — first renewal 403 bumps "failure", second renewal succeeds → loop kept ticking. - TestVault_RenewLoop_CtxCancellation_StopsCleanly — Stop returns within 200ms after ctx cancel. - TestVault_RenewLoop_StartsNothingWhenNotRenewable — token already non-renewable at boot ⇒ no goroutine, "not_renewable" metric increments at startup so operators see it in Grafana. - TestVault_ComputeInterval — 4 cases pinning TTL/2 + minRenewInterval floor. - TestVault_RenewSelf_ParseFailure_NamesActionableInError — surfaced error contains "vault token renewal failed" + "rotate the token". Cadence is dynamic — every successful renewal re-derives TTL/2 from the renewed lease's lease_duration, so a short bootstrap token that gets renewed up to a longer Max TTL shifts to the longer cadence automatically (defends against degenerate fast ticking on a token whose Max TTL is far longer than its initial TTL). Documentation: - docs/connectors.md Vault PKI section gains "Token TTL + automatic renewal" subsection (operator-facing: cadence, metric, renewable=false rotation playbook). Out of scope (intentional, flagged in the audit follow-up): - AppRole / Kubernetes / AWS IAM auth methods (different renewal semantics). - Hot-reload of rotated token from disk (operator restarts today; future: GUI/MCP issuer-update path triggers Rebuild which Stops the old connector and Starts the new one). - Auto-re-auth after token death (operator playbook owns it). CHANGELOG.md is intentionally not hand-edited (per CHANGELOG.md itself: "no longer maintains a hand-edited per-version changelog; per-release notes are auto-generated from commit messages between consecutive tags"). Verified locally: - gofmt clean. - go vet ./internal/service/... ./internal/api/handler/... ./internal/connector/issuer/vault/... ./cmd/server/... clean. - go test -short -count=1 ./internal/connector/issuer/vault/... ./internal/service/... ./internal/api/handler/... green. - go test -race -count=10 -run 'TestVault_RenewLoop|TestVault_ComputeInterval' ./internal/connector/issuer/vault/... green. Audit reference: cowork/issuer-coverage-audit-2026-05-03/RESULTS.md Top-10 fix #5. |
||
|
shankar0123
|
bee47f0318 |
acme-server: cert-manager integration test + production hardening (Phase 5/7)
Closes the production-readiness loop on the ACME surface. After this
commit, certctl ships per-account rate limits + a GC sweeper for
expired ACME state + a kind-driven cert-manager 1.15 integration test
+ a lego-driven RFC conformance harness + a k6 loadtest scenario for
the unauthenticated ACME path.
Architecture:
- Rate limits live in-memory + per-replica. Restart wipes the
counters; orders/hour caps are eventual-consistency anyway. A
3-replica certctl-server fleet behind an LB effectively has 3x
the configured throughput per account; persistent rate limiting
is a follow-up if production telemetry shows abuse patterns we
can't catch in a single restart cycle. Per-key + per-action
isolation: ActionNewOrder/acc-1, ActionKeyChange/acc-1, and
ActionChallengeRespond/<challenge-id> are independent buckets.
- GC loop follows the existing scheduler-loop pattern (atomic.Bool
+ sync.WaitGroup; see crlGenerationLoop for shape). Three
independent SQL sweeps per tick (DELETE expired nonces; UPDATE
pending authzs whose expires_at < now() to expired; UPDATE
pending/ready/processing orders whose expires_at < now() to
invalid). Each sweep is a single statement; failures are logged-
and-continued so a failing nonces sweep doesn't block authzs.
Per-sweep 1m timeout bounds a stuck Postgres.
- cert-manager integration test is gated on KIND_AVAILABLE so CI
skips it cleanly (kind is too heavy for per-PR). Operators run
locally via 'make acme-cert-manager-test'; the harness brings up
a fresh cluster each run + tears it down on Cleanup.
- lego conformance harness drives a real ACME client through
register → run → cert-PEM-landed against a hermetic certctl
stack. Catches RFC-shape regressions third-party clients would
hit before they ship.
- k6 ACME-flow scenario hammers the unauthenticated surface
(directory + new-nonce + ARI synthetic-id) at 100 VUs × 5m. JWS-
signed flows are out of scope for k6 (no JWS support); they're
covered by the lego harness above.
What ships:
- internal/api/acme/ratelimit.go (+ ratelimit_test.go: 7 cases —
disable-when-perHour-zero, capacity, per-key isolation, per-
action isolation, refill-over-time, RetryAfter, concurrent-access
with -race + 200 goroutines × 200 calls).
- internal/repository/postgres/acme.go: 4 new methods —
CountActiveOrdersByAccount + GCExpiredNonces + GCExpireAuthorizations
+ GCInvalidateExpiredOrders. Each a single SQL statement.
- internal/service/acme.go: SetRateLimiter + GarbageCollect +
rate-limit gates at 3 entry points (CreateOrder + RotateAccountKey
+ RespondToChallenge) + concurrent-orders gate at CreateOrder.
2 new sentinels (ErrACMERateLimited, ErrACMEConcurrentOrdersExceeded);
5 new GC metrics (gc_runs / gc_run_failures / gc_nonces_reaped /
gc_authzs_expired / gc_orders_invalidated).
- internal/scheduler/scheduler.go: ACMEGarbageCollector interface +
acmeGCRunning atomic.Bool + acmeGCInterval + 2 setters (SetACME-
GarbageCollector + SetACMEGCInterval) + acmeGCLoop following the
crlGenerationLoop shape.
- internal/api/handler/acme.go: writeServiceError gains rateLimited
(429 + RFC 8555 §6.7) + concurrent-orders-exceeded mappings.
- internal/config/config.go: 5 new env vars
(CERTCTL_ACME_SERVER_RATE_LIMIT_ORDERS_PER_HOUR=100,
CERTCTL_ACME_SERVER_RATE_LIMIT_CONCURRENT_ORDERS=5,
CERTCTL_ACME_SERVER_RATE_LIMIT_KEY_CHANGE_PER_HOUR=5,
CERTCTL_ACME_SERVER_RATE_LIMIT_CHALLENGE_RESPONDS_PER_HOUR=60,
CERTCTL_ACME_SERVER_GC_INTERVAL=1m).
- cmd/server/main.go: NewRateLimiter() + SetRateLimiter() at
startup; conditional SetACMEGarbageCollector(acmeService) +
SetACMEGCInterval(cfg.ACMEServer.GCInterval) when Enabled+
GCInterval > 0.
- deploy/test/acme-integration/: kind-config.yaml + cert-manager-
install.sh + clusterissuer-trust-authenticated.yaml +
clusterissuer-challenge.yaml + certificate-test.yaml + conformance-
lego.sh + certmanager_test.go (//go:build integration + KIND_AVAILABLE
gate).
- deploy/test/loadtest/k6/acme_flow.js + README ACME-flows section.
- Makefile: 2 new PHONY targets (acme-cert-manager-test +
acme-rfc-conformance-test).
- docs/acme-server.md: status flipped to Phase 5; Configuration
table grows 5 rows; new 'Phase 5 — operational guidance' section
explaining rate-limit math + GC sweeper semantics + cert-manager
integration + lego conformance + k6 baseline.
Tests:
- 'go vet ./...' clean across the repo.
- 'go test -short -count=1 ./internal/...' green across every
affected package (service / acme / handler / scheduler / repo /
config).
- 'go vet -tags=integration ./deploy/test/acme-integration/' clean
(the integration test compiles cleanly with the build tag).
- The kind/cert-manager harness is gated behind KIND_AVAILABLE so
CI skips by default; operators run locally via 'make acme-cert-
manager-test'.
Engineering history: cowork/WORKSPACE-CHANGELOG.md 'ACME-Server-5'.
|
||
|
shankar0123
|
1e1bc9b3b4 |
ci: fix Phase 4 post-push unused-symbol failures
CI on commit
|
||
|
shankar0123
|
4dc8d3fa5b |
acme-server: key rollover + revocation + ARI (Phase 4/7)
Closes the RFC 8555 + RFC 9773 surface beyond the issuance happy-path:
- POST /acme/profile/<id>/key-change (RFC 8555 §7.3.5)
- POST /acme/profile/<id>/revoke-cert (RFC 8555 §7.6)
- GET /acme/profile/<id>/renewal-info/<cert-id> (RFC 9773 ARI)
After this commit, ACME clients can rotate account keys, revoke certs
through the ACME surface (rather than only via the certctl GUI/API),
and fetch ARI for proactive renewal scheduling.
Architecture:
- Key rollover: outer JWS verified against the registered account key
(existing kid path); the inner JWS — embedded as the outer's payload
— verified against the embedded NEW jwk in a new dedicated routine
(ParseAndVerifyKeyChangeInner) that enforces RFC 8555 §7.3.5
inner-only invariants: MUST use jwk + MUST NOT use kid, payload
.account == outer.kid, payload.oldKey thumbprint-equals registered.
A single WithinTx swaps the stored thumbprint+pem and writes the
audit row. Concurrent-rollover safety via SELECT…FOR UPDATE on the
conflicting account row in UpdateAccountJWKWithTx; the loser
observes the winner's new thumbprint and is told to retry (409).
- Revocation: two auth paths. kid → AccountOwnsCertificate single-
indexed COUNT lookup over acme_orders. jwk → constant-time RFC 7638
thumbprint compare against the cert's pubkey. Both paths route
through service.RevocationSvc.RevokeCertificateWithActor so the
existing CRL/OCSP refresh + audit + metrics pipeline applies. RFC
5280 §5.3.1 numeric reason codes clamp to certctl's
domain.ValidRevocationReasons; codes 8 (removeFromCRL) + 10
(aACompromise) clamp to 'unspecified' since they aren't in the set.
- ARI is GET-only and unauth per RFC 9773 §4. Cert-id wire shape is
base64url(AKI).base64url(serial); ParseARICertID strict-decodes,
SerialHex emits the canonical certctl-shape lowercase-no-leading-
zeros hex used in certificate_versions.serial_number.
ComputeRenewalWindow has 3 branches: bound RenewalPolicy →
[notAfter - days, notAfter - days/2]; no policy → last 33% of
validity; past expiry → [now, now + 1d] (renew immediately).
Retry-After honors CERTCTL_ACME_SERVER_ARI_POLL_INTERVAL.
What ships:
- internal/api/acme/{keychange,ari}.go (+ phase4_test.go: 15 tests).
- internal/api/acme/order.go: RevokeCertRequest wire shape.
- internal/api/handler/acme.go: KeyChange, RevokeCert, RenewalInfo
+ 11 new writeServiceError mappings.
- internal/repository/postgres/acme.go: UpdateAccountJWKWithTx (FOR
UPDATE + expectedOldThumbprint precondition; ErrACMEAccountKey-
ConcurrentUpdate sentinel) + AccountOwnsCertificate.
- internal/service/acme.go: RotateAccountKey + RevokeCert +
RenewalInfo; CertificateRevoker + RenewalPolicyLookup interfaces;
SetRevocationDelegate + SetRenewalPolicyLookup wiring; 11 new
sentinels; 6 new metrics.
- internal/service/acme_phase4_test.go: service-layer tests for
RotateAccountKey (happy + duplicate-key) + RevokeCert (kid mismatch
+ jwk mismatch + jwk happy + already-revoked + reason-clamping) +
RenewalInfo (disabled + bad cert-id).
- internal/api/router/router.go: 6 new register calls (3 per-profile
+ 3 shorthand). Router parity exceptions extended in lockstep
(in-tree SpecParityExceptions + CI-only openapi-handler-exceptions
.yaml).
- cmd/server/main.go: SetRevocationDelegate(revocationSvc) +
SetRenewalPolicyLookup(renewalPolicyRepo) at startup.
- internal/config/config.go: CERTCTL_ACME_SERVER_ARI_ENABLED (default
true) + CERTCTL_ACME_SERVER_ARI_POLL_INTERVAL (default 6h);
BuildDirectory's ariEnabled flag now flips on under
cfg.ARIEnabled.
- docs/acme-server.md: phase status flipped to Phase 4; endpoints
table grows 6 rows (3 per-profile + 3 shorthand); FAQ section
appended explaining how to rotate keys, revoke certs, and consume
ARI.
Tests:
- 'go vet ./...' clean across the repo.
- 'go test -short -count=1 ./...' green across every package.
- phase4_test.go covers: keychange happy-path + 5 negatives +
MapKeyChangeErrorToProblem coverage; ARI cert-id round-trip + 6
malformed cases + BuildARICertID from a generated cert; window-
math 3 branches.
- service-layer tests confirm: RotateAccountKey atomically swaps the
thumbprint (verifies persisted state) and rejects duplicate keys;
RevokeCert routes through the stub RevocationSvc with the right
actor string + reason on the jwk path, rejects mismatched keys,
rejects already-revoked certs, clamps reason codes correctly;
RenewalInfo respects ARIEnabled + cert-id format.
Engineering history: cowork/WORKSPACE-CHANGELOG.md 'ACME-Server-4'.
|
||
|
shankar0123
|
62513ad12f |
ci: fix Phase 3 post-push CI failures (contextcheck + ST1021)
CI on commit
|
||
|
shankar0123
|
9bc845304e |
acme-server: HTTP-01 + DNS-01 + TLS-ALPN-01 challenge validation (Phase 3/7)
Wires up the actual challenge-validation machinery so profiles in
acme_auth_mode='challenge' resolve end-to-end. After this commit,
cert-manager 1.15+ with `solver: http01: ingress` against a
challenge-mode profile completes a real HTTP-01 flow and gets a cert.
DNS-01 + TLS-ALPN-01 share the same code path with the appropriate
validator selection.
Architecture (the load-bearing parts):
- 3 separate semaphore-bounded worker pools (one per challenge type),
so HTTP-01 and DNS-01 can't starve each other under load. Default
weight 10 per type; tunable via CERTCTL_ACME_SERVER_HTTP01_CONCURRENCY,
DNS01_CONCURRENCY, TLSALPN01_CONCURRENCY.
- 30s per-challenge timeout (configurable via PoolConfig.PerChallengeTimeout).
- HTTP-01 validator runs validation.IsReservedIPForDial (newly
exported wrapper preserving the existing private impl byte-for-byte
for the network scanner + ValidateSafeURL paths) on the resolved
IP — both at the initial dial and every redirect hop. SSRF probes
into private IP space are refused before the connect.
- DNS-01 validator uses a dedicated resolver pointed at
CERTCTL_ACME_SERVER_DNS01_RESOLVER (default 8.8.8.8:53) — does
NOT use the system resolver to keep behavior deterministic across
deployments. Wildcard handling: `*.example.com` queries
_acme-challenge.example.com.
- TLS-ALPN-01 validator (RFC 8737) connects with ALPN `acme-tls/1`,
inspects the id-pe-acmeIdentifier extension (OID 1.3.6.1.5.5.7.1.31),
asserts the ASN.1 OCTET STRING value equals SHA-256 of the key
authorization. Cert chain is intentionally NOT validated
(InsecureSkipVerify=true is correct per RFC 8737 — the proof is
in the extension, not the chain). Documented in docs/tls.md L-001
table + the //nolint:gosec comment carries the justification.
SSRF guard: same posture as HTTP-01.
- Validation is asynchronous: handler accepts the POST and returns
200 immediately with status=processing; the worker-pool fires a
callback that updates challenge → authz → order in a fresh
background-context WithinTx. The order auto-promotes to `ready`
when ALL authzs become valid; auto-fails to `invalid` when ANY
authz becomes invalid.
What ships:
- internal/api/acme/challenge.go: KeyAuthorization (RFC 8555 §8.1) +
DNS01TXTRecordValue (§8.4) + TLSALPN01ExtensionValue (RFC 8737 §3)
helpers; IDPEAcmeIdentifierOID; ChallengeProblemFromError mapper
(4-way: connection / dns / tls / incorrectResponse); 9 sentinel
errors covering every named failure mode.
- internal/api/acme/validators.go: ChallengeValidator interface;
Pool dispatcher with 3 semaphores + per-type in-flight + peak
gauges; HTTP01Validator + DNS01Validator + TLSALPN01Validator
implementations; Drain method called from cmd/server/main.go's
shutdown sequence.
- internal/api/acme/validators_test.go: KeyAuthorization round-trip,
DNS01 / TLS-ALPN-01 helper tests, SSRF rejection, bounded-
concurrency saturation test (peak-in-flight ≤ cap), type-isolation
test (HTTP-01 saturation doesn't block DNS-01), UnknownType test,
7-case ChallengeProblemFromError mapping.
- internal/repository/postgres/acme.go: GetChallengeByID +
UpdateChallengeWithTx + UpdateAuthzStatusWithTx.
- internal/service/acme.go: SetValidatorPool wires the *acme.Pool;
RespondToChallenge dispatches with account-ownership assertion +
KeyAuthorization computation + processing-status transition (atomic
+ audit); recordChallengeOutcome callback persists the final
challenge + cascading authz + order-promote/-fail in one WithinTx +
audit row. 4 new metrics.
- internal/api/handler/acme.go: Challenge handler; round-trips
account.JWKPEM through ParseJWKFromPEM to recover the *jose.JSONWebKey
the validator pool needs.
- internal/api/router/router.go + openapi_parity_test.go +
api/openapi-handler-exceptions.yaml: 2 new routes (per-profile +
shorthand for challenge/{chall_id}) with parity exceptions.
- cmd/server/main.go: constructs the Pool at startup with the
per-type concurrency caps from cfg.ACMEServer; ACMEService.ValidatorPool()
accessor exposed for the shutdown drain sequence.
- internal/validation/ssrf.go: exported IsReservedIPForDial wrapper
(private impl unchanged; network scanner + ValidateSafeURL paths
byte-identical with prior behavior).
- docs/tls.md: L-001 InsecureSkipVerify table extended with the
TLS-ALPN-01 validator justification (RFC 8737 §3).
- docs/acme-server.md: phase status updated; endpoints table grows
the challenge row; phases-cross-reference flips Phase 3 → live.
Tests:
- 80%+ coverage on the new files.
- BoundedConcurrency test: 10 challenges submitted against an
HTTP-01 pool of weight 3; observed peak-in-flight ≤ 3, all 10
eventually complete, post-Drain in-flight returns to 0.
- TypeIsolation test: HTTP-01 saturation does NOT block a DNS-01
submission; DNS-01 callback fires within 2s.
- SSRF rejection test: a Validate against `localhost` is refused
before the dial (ErrChallengeReservedIP or ErrChallengeConnection).
Engineering history: cowork/WORKSPACE-CHANGELOG.md "ACME-Server-3".
|
||
|
shankar0123
|
c351bba41a |
acme-server: orders + authorizations + finalize + cert download (Phase 2/7)
Closes the issuance loop in trust_authenticated mode (commits |
||
|
shankar0123
|
44a85d6f85 |
acme-server: account resource + JWS verifier (Phase 1b/7)
Layers JWS-authenticated POST machinery onto the Phase 1a foundation
(commit
|
||
|
shankar0123
|
ec88a61274 |
acme-server: foundation — directory + new-nonce + per-profile routing (Phase 1a/7)
First slice of the RFC 8555 ACME server endpoint (master plan at cowork/acme-server-endpoint-prompt.md, per-phase prompts at cowork/acme-server-prompts/). This commit lands the smallest viable end-to-end deployable slice: an ACME client running curl -sk https://certctl/acme/profile/<id>/directory curl -sk -I https://certctl/acme/profile/<id>/new-nonce successfully fetches the directory document and a Replay-Nonce. Account creation, JWS verification, orders, challenges, and revocation are all out of scope for this phase and arrive in Phases 1b–4. Closes the Rank 1 LHF from the 2026-05-03 Infisical deep-research (cowork/infisical-deep-research-results.md). Pre-fix, certctl was an ACME consumer only — no /acme/directory endpoint, no JWS verifier, no challenge validators. K8s customers running cert-manager could not point at certctl as an ACME issuer; they had to deploy a certctl agent on every node. What ships: - internal/api/acme/{directory,nonce,errors}.go (+ tests). - internal/api/handler/acme.go + acme_handler_test.go. - internal/repository/postgres/acme.go (nonce ops only — Phase 1b extends with account CRUD; Phases 2-4 extend with order / authz / challenge CRUD). - internal/service/acme.go (BuildDirectory + IssueNonce stubs; Phase 1b adds VerifyJWS / NewAccount / etc.). - migrations/000025_acme_server.{up,down}.sql ships the full 5-table ACME schema (acme_accounts / acme_orders / acme_authorizations / acme_challenges / acme_nonces) PLUS the per-profile certificate_profiles.acme_auth_mode column. Phase 1a actively uses only acme_nonces; remaining tables are empty until Phases 1b-4 plug in. - internal/config/config.go: ACMEServerConfig struct + ACMEServer field on Config. Env vars use CERTCTL_ACME_SERVER_* prefix to avoid colliding with the existing consumer-side ACMEConfig at config.go:1746 (CERTCTL_ACME_DIRECTORY_URL / PROFILE / CHALLENGE_TYPE etc.). Phase 1a wires Enabled + DefaultAuthMode + DefaultProfileID + NonceTTL + DirectoryMeta; Order/Authz TTLs + per-challenge-type concurrency caps + DNS01 resolver are reserved fields parsed in 1a so operators can set them ahead of Phases 2/3. - cmd/server/main.go: wire ACMEHandler into the HandlerRegistry literal alongside the existing certificate / EST / SCEP / etc. handlers. - internal/api/router/router.go: HandlerRegistry.ACME field + 6 Register calls (3 per-profile + 3 shorthand). - internal/api/router/openapi_parity_test.go: 6 new entries in SpecParityExceptions. ACME is a wire-protocol surface (JWS-signed JSON over HTTPS per RFC 7515) whose semantics are dictated by RFC 8555 + RFC 9773 rather than by an OpenAPI document, same precedent as SCEP/EST. The canonical reference is docs/acme-server.md. - docs/acme-server.md: Phase-1a-shaped reference. Configuration table for every CERTCTL_ACME_SERVER_* env var. Per-profile auth-mode decision tree skeleton. TLS trust bootstrap section flagging cert-manager's ClusterIssuer.spec.acme.caBundle requirement (the single biggest first-time-deploy footgun; the full cert-manager walkthrough lands in Phase 6 but the requirement is documented up front). Architecture decisions baked in: - URL family is /acme/profile/<id>/* (per-profile, canonical) with /acme/* shorthand active when CERTCTL_ACME_SERVER_DEFAULT_PROFILE_ID is set. Path matches existing per-profile precedent in EST + SCEP. - Auth mode is per-profile (acme_auth_mode column on certificate_profiles), NOT server-wide. One certctl-server can serve trust_authenticated for an internal-PKI profile and challenge for a public-trust-style profile simultaneously. The column is read at request time, not cached at server start — operators flipping a profile's mode via SQL take effect on the next order without restart. - Nonces are DB-backed (acme_nonces table). Survive server restart. The RFC 8555 §6.5 replay defense requires the store to outlast the client's nonce caching window; an in-memory-only nonce store would lose every in-flight order on restart. - Per-op atomic counters on service.ACMEService.Metrics() — certctl_acme_directory_total, certctl_acme_directory_failures_total, certctl_acme_new_nonce_total, certctl_acme_new_nonce_failures_total. Naming follows certctl frozen decision 0.10 cardinality discipline. Phase 1b will extend with new_account counters; Phase 2 with order / finalize / cert; Phase 3 with per-challenge-type counters. Audit fixes #11 + #12 (cowork/acme-server-prompts/audit-additions.md) applied: - #11: CERTCTL_ACME_SERVER_* prefix avoids the consumer-side CERTCTL_ACME_* namespace collision. - #12: prior-attempt WIP from two failed Phase-1 dispatches was discarded at phase start; this commit starts from a clean tree. Tests: - 14 unit tests in internal/api/acme/ (directory, nonce, errors). - 7 handler-level tests via httptest.NewServer + mockACMEService (mirrors the mockSCEPService pattern at scep_handler_test.go). - 7 service-layer tests with mocked repo + injected profileLookup. - All pass under -race -count=1 -short. Deferred to Phase 1b: - JWS verification (go-jose v4 — see master-prompt §8a for the API surface and audit doc for the speculation pitfalls). - new-account / account/<id> endpoints + AccountService. - Nonce *consumption* path (issue path is in this commit; consume is only invoked by JWS-verified POSTs which Phase 1b adds). Engineering history: cowork/WORKSPACE-CHANGELOG.md "ACME-Server-1a". Per-phase implementation plan: cowork/acme-server-prompts/. Master plan + audit fixes: cowork/acme-server-endpoint-prompt.md + cowork/acme-server-prompt-audit.md + cowork/acme-server-prompts/audit-additions.md. |
||
|
shankar0123
|
475421457f |
fix(test): TestBoundedFanOut_SkipsAgentRoutedDeployments race on seenIDs slice
CI race detector flagged TestBoundedFanOut_SkipsAgentRoutedDeployments
on commit
|
||
|
shankar0123
|
35e18bfc56 |
scheduler: bound renewal concurrency via CERTCTL_RENEWAL_CONCURRENCY
Closes the #9 acquisition-readiness blocker from the 2026-05-01 issuer coverage audit. Pre-fix, JobService.ProcessPendingJobs ran every claimed job sequentially in a single goroutine: safe but slow, and operators with large fleets had no lever to dial throughput up. Switching to fire-and-forget per-job goroutines would have unbounded the upstream-CA call rate and tripped DigiCert / Entrust / Sectigo rate limits — certctl's response to 429 was to retry on the next tick, re-fanning out the same calls and digging deeper into the limit. Operators need a knob. This commit: - Adds CERTCTL_RENEWAL_CONCURRENCY env var (default 25) loaded via the existing getEnvInt pattern in internal/config/config.go. Documented inline as the cap for the per-tick renewal/issuance/ deployment goroutine fan-out, with operator-tuning guidance: permissive upstream limits + large fleets (>10k certs) → 100; strict limits or async-CA-heavy fleets → 25 or lower. - Wires golang.org/x/sync/semaphore.Weighted around the per-job goroutine launch in JobService.ProcessPendingJobs. Acquire(ctx, 1) is the load-bearing piece — it BLOCKS the loop when at the cap, providing real backpressure rather than fire-and-forget. The fan-out is split into processPendingJobsSequential (legacy, preserved for unit-test wiring that doesn't call SetRenewalConcurrency) and processPendingJobsConcurrent (production, delegates to a generic boundedFanOut helper). - boundedFanOut takes the per-job work as a closure so the cap can be tested directly without standing up the renewal/deployment service graph. processed/failed counters use atomic.Int64 to avoid mutex overhead on every job completion; final log line reads both AFTER wg.Wait so the counts reflect every dispatched job. ctx-aware Acquire ensures a shutdown ctx cancel interrupts the dispatch loop promptly; in-flight goroutines drain via Wait before the function returns so no goroutine outlives the scheduler tick. - shouldSkipJob extracted as a package-private helper so the agent-routed-deployment skip logic is shared between the sequential and concurrent paths byte-for-byte (the audit prompt's "channel-based semaphore without ctx-aware acquire" anti-pattern is explicitly avoided — semaphore.Weighted.Acquire returns on ctx done; channel <- struct{}{} would block forever). - SetRenewalConcurrency setter on JobService normalises ≤0 to 1. semaphore.NewWeighted(0) constructs a semaphore that blocks every Acquire forever; the normalisation prevents a misconfigured env var from wedging the scheduler. - cmd/server/main.go wires SetRenewalConcurrency(cfg.Scheduler. RenewalConcurrency) on the freshly-built jobService, immediately after SetAuditService. Production deployments always take the bounded path; tests that build JobService directly via NewJobService keep their strict-sequential behaviour because renewalConcurrency is the zero value. - Tests in internal/service/job_concurrency_test.go: * TestBoundedFanOut_CapHolds — primary regression guard. 50 jobs × 50ms work × cap=5 → asserts peak in-flight never exceeds 5 AND reaches 5 at least once (catches both upper-bound regressions and gates that incorrectly cap below the configured value). Lock-free max via CompareAndSwap so the measurement instrument doesn't itself constrain concurrency. * TestBoundedFanOut_AllJobsRun — lower-bound: every non-skipped job is dispatched. * TestBoundedFanOut_SkipsAgentRoutedDeployments — pins the shouldSkipJob contract. * TestBoundedFanOut_CtxCancelInterrupts — ctx cancellation interrupts a stuck fan-out within the timeout budget. * TestBoundedFanOut_FailedJobsCounted — per-job errors don't abort the fan-out. * TestSetRenewalConcurrency_NormalizesNonPositive — ≤0 → 1 fail-safe pinned across negative/zero/positive inputs. - docs/features.md: scheduler-loop table augmented with the concurrency-cap env-var pointer alongside the job-processor row. - docs/architecture.md: Concurrency Safety section gains a paragraph explaining the cap, the operator-tuning guidance, the ctx-aware Acquire semantics, and the audit reference. Operator-facing impact: the first big renewal sweep no longer takes down the upstream CA's rate-limit budget. Existing deployments get the bounded path automatically (default 25); operators can override via env var without code changes. Verified locally: - gofmt -l . clean - go vet ./... clean - staticcheck ./... clean - go test -short -count=1 across service / scheduler / config / integration: green - Six new tests under TestBoundedFanOut* + TestSetRenewalConcurrency*: green Audit reference: cowork/issuer-coverage-audit-2026-05-01/RESULTS.md Top-10 fix #9. |
||
|
shankar0123
|
fefa5a5fd7 |
acme: support serial-only revocation via local cert-version lookup
Closes the #7 acquisition-readiness blocker from the 2026-05-01 issuer coverage audit. Pre-fix, ACME RevokeCertificate at acme.go:L519-L529 returned the literal error "ACME revocation by serial not supported in V1; provide certificate DER". RFC 8555 §7.6 genuinely requires the cert DER bytes (not just the serial), but a CLM platform's job is to abstract over that limitation. Operators routinely have only the serial in hand: lost PEM, rotated key, GUI revoke action driven by a row in the certs list. This commit: - Adds CertificateLookupRepo interface at the ACME connector boundary (connector boundary, NOT a service/repository import — the connector accepts whatever satisfies the shape). Production wiring in cmd/server/main.go injects the postgres CertificateRepository; tests inject a fake. - Adds CertificateRepository.GetVersionBySerial(ctx, issuerID, serial) + interface declaration in repository/interfaces.go, returning the certificate_versions row whose SerialNumber matches, scoped to the issuer via JOIN on managed_certificates. Mirrors the existing GetByIssuerAndSerial shape but returns the version (where PEMChain lives). Per RFC 5280 §5.2.3 the issuer scope is required for determinism. - Adds SetCertificateLookup + SetIssuerID setters on *acme.Connector. Mirror the pattern local.Connector already uses for OCSP responder wiring. Both must be wired before serial-only revoke works; unwired state falls back to a more actionable error pointing at the wiring requirement (the historical "not supported" wording is retired). - Rewrites RevokeCertificate end-to-end: lookup → empty-PEM check → pem.Decode → block.Type == "CERTIFICATE" check → ensureClient → golang.org/x/crypto/acme.Client.RevokeCert(ctx, accountKey, der, reasonCode). RFC 8555 §7.6 case 1 (revocation request signed with account key) — the same account key issued the cert, so authority is intrinsic. The not-found path returns an actionable operator- facing error pointing at the local-store requirement. - Adds mapRevocationReason translating RFC 5280 §5.3.1 reason strings (unspecified, keyCompromise, cACompromise, affiliationChanged, superseded, cessationOfOperation, certificateHold, removeFromCRL, privilegeWithdrawn, aACompromise) into golang.org/x/crypto/acme. CRLReasonCode. Accepts canonical camelCase + underscore_lower + ALL_CAPS_UNDERSCORE. Nil reason → 0 (unspecified). Unknown reason errors rather than silently demoting (operators rely on the reason for compliance reporting). - Wiring update in service/issuer_registry.go: SetACMECertLookup setter on the registry; Rebuild type-asserts *acme.Connector and calls SetCertificateLookup + SetIssuerID, mirroring the existing *local.Connector branch. cmd/server/main.go calls issuerRegistry.SetACMECertLookup(certificateRepo) immediately after SetIssuanceMetrics — the postgres repo satisfies the interface via GetVersionBySerial. - Tests: * acme_revoke_test.go (new): TestRevokeCertificate_NoCertLookupWired, TestRevokeCertificate_NoIssuerIDWired, TestRevokeCertificate_LookupReturnsNotFound (operator-facing "may not have been issued through certctl" hint pinned), TestRevokeCertificate_LookupArbitraryError, TestRevokeCertificate_VersionPEMEmpty (corrupt-row guard), TestRevokeCertificate_PEMMalformed_NoBlock, TestRevokeCertificate_PEMMalformed_WrongType (PRIVATE KEY block rejected as not a CERTIFICATE). * TestMapRevocationReason_TableDriven: full RFC 5280 reason set plus camelCase / underscore / ALL-CAPS variants plus nil-reason and unknown-reason cases. * acme_failure_test.go: renamed TestRevokeCertificate_AlwaysError → TestRevokeCertificate_UnwiredCertLookupFallback; the test still exercises the same backward-compat branch but now asserts the new "CertificateLookup wiring" error wording. - Mock-repo updates (3 sites): mockCertificateRepository in internal/integration/lifecycle_test.go, mockCertRepo in internal/service/testutil_test.go, mockCertRepoWithGetError in internal/service/shortlived_test.go each gain a GetVersionBySerial implementation that mirrors the GetByIssuerAndSerial logic but returns the version row. - docs/connectors.md ACME section: new "Revocation by serial number" subsection covering the workflow, the local-store requirement (cert was issued through certctl, not imported), the reason-code mapping with the three accepted spelling variants, and a pointer to the audit reference. Out of scope (intentional, per spec): - Recovering the DER from outside the local cert store (CT logs, CSR + signature reconstruction). If the cert wasn't issued through certctl, revoke-by-serial via certctl isn't possible. - Revocation via the cert's private key (RFC 8555 §7.6 case 2). The account-key path covers all certctl-issued certs because the same account key issued them. - Pebble-backed integration test for the happy path. Pebble integration is the right home for that — the unit tests in this commit pin all failure-mode branches before the network call, and the wiring branch in Rebuild is exercised by the existing TestIssuerRegistryRebuild paths. Verified locally: - gofmt -l . clean - go vet ./... clean - staticcheck ./... clean - go test -short -count=1 across connector, service, repository, integration, api/middleware, api/handler: green Audit reference: cowork/issuer-coverage-audit-2026-05-01/RESULTS.md Top-10 fix #7. |