Compare commits

...

10 Commits

Author SHA1 Message Date
shankar0123 aa6fafdee9 Merge branch 'fix/u2-dockerfile-healthcheck-https' 2026-04-25 12:02:28 +00:00
shankar0123 86fffa305a fix(deploy,helm,docs): published-image HEALTHCHECK speaks HTTPS + Helm /ready path + docs HTTPS sweep (U-2)
Pre-U-2 the published `ghcr.io/shankar0123/certctl-server` image
shipped with `HEALTHCHECK CMD curl -f http://localhost:8443/health`.
The server has been HTTPS-only since the v2.2 HTTPS-Everywhere milestone
(`cmd/server/main.go::ListenAndServeTLS`, no plaintext fallback, TLS
1.3 pinned), so the probe failed on every interval and Docker marked
the container `unhealthy` indefinitely. Operators inside docker-
compose / Helm / the example stacks were unaffected — compose overrides
the HEALTHCHECK with `--cacert + https://`, Helm uses explicit
`httpGet` probes that ignore Docker's HEALTHCHECK, and every example
compose file overrides with `curl -sfk https://localhost:8443/health`.
But anyone running bare `docker run` / Docker Swarm / Nomad / ECS —
exactly the "I just pulled the published image" path — saw permanent
`unhealthy` status and (depending on orchestrator policy) a restart-
loop. (Audit: cat-u-healthcheck_protocol_mismatch in
coverage-gap-audit-2026-04-24-v5/unified-audit.md.)

Recon for U-2 surfaced two adjacent bugs from the same v2.2 milestone
gap, both bundled into this commit because they share the same root
cause and the same operator surface:

  1. Helm chart `server.readinessProbe.httpGet.path` pointed at
     `/readyz`, the kube-flavored convention. The certctl server
     doesn't register `/readyz` (only `/health` and `/ready` are
     wired and bypass the auth middleware — see
     internal/api/router/router.go:81 and cmd/server/main.go:920).
     K8s readiness probes therefore got 401 (api-key auth rejection)
     or 404 (when auth was disabled), pods stayed `NotReady`
     indefinitely, and Helm rollouts stalled.

  2. The agent image (`Dockerfile.agent`) had no HEALTHCHECK at all,
     so bare-`docker run` agents got zero health signal. The
     compose override at `deploy/docker-compose.yml:173` called
     `pgrep -f certctl-agent` against the agent image, but the
     agent image didn't ship `procps` — pgrep was missing too. The
     compose probe was a latent always-fail.

We fixed all three with the audit-recommended shape (option (a) — `-k`)
plus three structural backstops:

Files changed:

Phase 1 — Dockerfile fix:
- Dockerfile: HEALTHCHECK switched from `curl -f http://localhost:8443/
  health` to `curl -fsk https://localhost:8443/health`. `-k`
  (insecure) is acceptable because the probe is localhost-to-localhost:
  the same process serving the cert is being probed, no network hop.
  Pinning `--cacert` is not viable for the published image because
  the bootstrap cert is per-deploy (generated into the `certs` named
  volume on first up; operator-supplied via Helm's `existingSecret`
  or cert-manager). Long-form docblock cross-references the audit
  closure, the compose vs Helm vs examples coverage matrix, and the
  CI guardrail.
- Dockerfile.agent: added HEALTHCHECK using `pgrep -f certctl-agent`
  matching the compose pattern. Added `procps` to the runtime apk
  install — fixes both the new image-level HEALTHCHECK AND the
  pre-existing compose probe that was silently failing.

Phase 2 — Helm readiness probe path:
- deploy/helm/certctl/values.yaml: server.readinessProbe.httpGet.path
  changed from `/readyz` to `/ready`. Liveness probe path
  (`/health`) was correct and is unchanged. Probes block now carries
  an explanatory comment naming the registered no-auth probe routes
  and the U-2 closure rationale.

Phase 3 — Image-level integration tests:
- deploy/test/healthcheck_test.go (new, //go:build integration):
  TestPublishedServerImage_HealthcheckSpecUsesHTTPS builds the server
  image, inspects `Config.Healthcheck.Test` via `docker inspect`,
  and asserts the array contains `https://localhost:8443/health` and
  `-k`, and does NOT contain `http://localhost:8443/health`
  (positive + negative regression contracts).
  TestPublishedAgentImage_HealthcheckSpecExists builds the agent image
  and asserts the HEALTHCHECK uses `pgrep` against `certctl-agent`.
  Both tests `t.Skip` cleanly when docker isn't available (sandbox /
  CI without docker-in-docker) — verified locally: tests skip with the
  diagnostic and the suite returns PASS.
  TestPublishedServerImage_HealthcheckTransitionsToHealthy is a
  documented `t.Skip` placeholder until the harness wires a sidecar
  postgres for image-level smoke; the spec-level tests above cover the
  audit-flagged regression.

Phase 4 — CI guardrail:
- .github/workflows/ci.yml: new "Forbidden plaintext HEALTHCHECK
  regression guard (U-2)" step. Scoped patterns catch
  `HEALTHCHECK.*http://` and `curl -f http://localhost:8443/health`
  in any `Dockerfile*`. Comment lines exempt; docs/upgrade-to-tls.md
  out of scope (the post-cutover invariant string at line 182 is
  intentionally a documented expected-failure assertion). Verified
  locally on the real tree (passes) and against synthetic regressions
  (each fires the guard).

Phase 5 — Docs sweep:
- docs/connectors.md: 15 stale curl examples updated from
  `http://localhost:8443/...` to `https://localhost:8443/...` with
  `--cacert "$CA"` injected on every site. Added a one-time
  introductory note documenting the `$CA` extraction with
  `docker compose ... exec ... cat /etc/certctl/tls/ca.crt`,
  matching the pattern in docs/quickstart.md. Pre-U-2 these examples
  silently failed against the HTTPS listener.

Phase 6 — Release surface:
- CHANGELOG.md: appended U-2 section to the existing [unreleased]
  block (immediately below the G-1 entry). Sections: explanatory
  blockquote covering all three bugs (primary + 2 adjacent), Fixed,
  Added, Changed.

Verification (all gates pass):
- go build ./... — clean
- go vet ./... — clean
- go vet -tags integration ./deploy/test/ — clean
- go test -short ./... — every package green
- go test -tags integration -v -run TestPublishedServerImage|TestPublishedAgentImage ./deploy/test/ —
  three tests SKIP cleanly with "docker not available" diagnostic
- helm lint deploy/helm/certctl/ — clean
- helm template smoke render — succeeds; rendered Deployment carries
  `path: /ready` and zero `/readyz` matches
- python3 yaml.safe_load on api/openapi.yaml — parses
- govulncheck ./... — no vulnerabilities in our code
- CI guardrail mirror: clean on real tree, fires on synthetic
  regression patterns

Out of scope (intentionally untouched):
- cmd/server/main.go::ListenAndServeTLS — HTTPS-only is correct,
  this finding does NOT propose adding back a plaintext listener.
- deploy/docker-compose.yml:126 HEALTHCHECK — already correct.
- deploy/docker-compose.test.yml HEALTHCHECK blocks — already correct.
- All 5 examples/*/docker-compose.yml HEALTHCHECK overrides — already
  correct (they ALSO use `-fsk https://localhost:8443/health`).
- Helm server.livenessProbe.httpGet — already uses `scheme: HTTPS` +
  `path: /health`, correct.
- docs/upgrade-to-tls.md:182 `curl ... http://localhost:8443/health`
  invariant line — that's the expected-failure assertion for the
  post-cutover state ("plaintext is gone, expect Connection refused");
  intentionally left intact.
- Go production code — this is purely a deploy-image / probe / docs /
  Helm-chart fix.

Refs: coverage-gap-audit-2026-04-24-v5/unified-audit.md
      §2 P1 cluster, cat-u-healthcheck_protocol_mismatch
      Audit recommendation followed verbatim: 'change Dockerfile:80
      to CMD curl -kf https://localhost:8443/health'.
2026-04-25 12:02:18 +00:00
shankar0123 e17788355b Merge branch 'fix/g2-apikey-hash-redaction' 2026-04-25 01:56:34 +00:00
shankar0123 87213128cc fix(security,domain): redact Agent.APIKeyHash from JSON wire shape (G-2)
Pre-G-2 internal/domain/connector.go::Agent::APIKeyHash was tagged
`json:"api_key_hash"` and shipped on every wire surface that returned
domain.Agent — GET /api/v1/agents (PagedResponse{Data: agents}),
GET /api/v1/agents/{id}, GET /api/v1/agents/retired, and the
POST /api/v1/agents registration response. Every authenticated client
(browser, CLI --json, MCP tool calls) received the SHA-256-of-the-API-key
string. The browser silently dropped it because web/src/api/types.ts
omits the field, but CLI and MCP consumers print full JSON so the hash
was visible there. Even though the value is a hash and not the plaintext
key, shipping it gives an attacker an offline brute-force target if the
API-key entropy is low (certctl doesn't enforce a minimum on operator-
supplied keys), and there's no business reason for any client to ever
receive it — the value is server-internal, used only for the lookup at
internal/repository/postgres/agent.go::GetByAPIKey. (Audit:
cat-s5-apikey_leak in coverage-gap-audit-2026-04-24-v5/unified-audit.md.)

We chose the audit's recommended fix (json:"-") plus a defense-in-depth
MarshalJSON plus a CI guardrail. Three layers because struct-tag
redaction alone is one rebase away from being silently reverted, the
custom MarshalJSON catches the case where a parent struct embeds Agent
under a different tag, and the CI grep blocks reintroduction at the spec
or frontend boundary even without a code review catching it.

Files changed:

Phase 1 — Domain redaction:
- internal/domain/connector.go: APIKeyHash tag flipped from
  `json:"api_key_hash"` to `json:"-"`. New Agent.MarshalJSON
  with value receiver + type-alias-recursion-break that explicitly
  zeroes APIKeyHash on the marshal-time copy. Long-form docblock
  explaining the G-2 closure rationale + cross-references to
  service.RegisterAgent (populator), repository.AgentRepository::
  GetByAPIKey (consumer), docs/architecture.md (DB-shape vs
  API-shape distinction), and the audit finding.

Phase 2 — Domain tests (5 test functions):
- internal/domain/connector_test.go: TestAgent_MarshalJSON_RedactsAPIKeyHash
  pins the marshal-boundary contract on a value receiver. ...RedactsViaPointer
  pins the *Agent path. ...RedactsInSlice pins the []Agent path that the
  ListAgents handler actually emits via PagedResponse. ...DoesNotMutateReceiver
  pins the by-value-receiver contract so a future refactor that switches
  to pointer-receiver gets caught. ...RoundTrip pins the wire-shape
  guarantee that APIKeyHash is dropped on encode and cannot reappear on
  decode. Single sentinel value ("sha256:LEAKED-CREDENTIAL-DERIVATIVE-
  SENTINEL") flows through every fixture for grep-ability on regression.

Phase 3 — Handler tests (4 test functions):
- internal/api/handler/agent_handler_test.go: TestListAgents_DoesNotLeakAPIKeyHash,
  TestGetAgent_DoesNotLeakAPIKeyHash, TestRegisterAgent_DoesNotLeakAPIKeyHash,
  TestListRetiredAgents_DoesNotLeakAPIKeyHash. Each asserts (a) the
  literal substring "api_key_hash" is absent from the httptest-captured
  body, (b) the leak sentinel value is absent, (c) the non-leaked fields
  ARE present (sanity that the handler is serving real data, not just
  empty payloads). Shared sentinel "sha256:LEAKED-CREDENTIAL-DERIVATIVE-
  HANDLER-SENTINEL" so a single grep over a failing test's output
  identifies the leak surface immediately.

Phase 4 — Spec / docs:
- api/openapi.yaml: api_key_hash property REMOVED from Agent schema
  (was at line 3690). Inline G-2 comment naming the closure + the
  database-vs-API-shape distinction so a future spec edit doesn't
  silently re-introduce the field.
- docs/architecture.md: ER-diagram block already documents the agents
  table including api_key_hash (DB shape — correct). Added a sibling
  note paragraph immediately below the diagram explaining that several
  columns are intentionally server-internal (api_key_hash redaction
  + issuers.config / deployment_targets.config encrypted shadow), with
  cross-references to the redaction enforcement site, the OpenAPI
  schema, the frontend interface, and the CI guardrail.
- web/src/api/types.ts: Agent interface unchanged in shape (already
  omitted the field) but added a leading comment block explaining
  WHY the omission is intentional — stops a future frontend dev from
  "completing" the interface from the OpenAPI spec or the Go struct.

Phase 5 — CI guardrail:
- .github/workflows/ci.yml: new "Forbidden api_key_hash JSON-shape
  regression guard (G-2)" step. Scoped patterns catch the actual
  regression shapes — Go struct tag (json:"api_key_hash"), frontend
  interface declaration, OpenAPI schema property, YAML enum/array
  membership. Repository / migration / seed / service / integration /
  unit-test / comment lines exempt. Verified locally on the real tree
  (passes) and against 4 synthetic regression patterns (each fires
  the guardrail). Mirrors the G-1 pattern from .github/workflows/
  ci.yml lines 47-108.

Phase 5b — Sweep verification (no changes, results documented for the
next reader):
- internal/api/middleware/audit.go: doesn't serialize Agent struct;
  records request body only. No leak.
- service.RegisterAgent audit-event payload: `map[string]interface{}{
  "name": name, "hostname": hostname}` — name + hostname only,
  no APIKeyHash. No leak.
- All 9 slog sites that mention agent: scalar attrs only ("agent_id",
  "error", "agent_hostname"), never the full struct. No leak.
- internal/mcp, internal/cli, cmd/cli, cmd/mcp-server: zero matches
  for APIKeyHash / api_key_hash. Both pass server JSON verbatim, so
  the wire-side fix transitively closes them.

Verification (all gates pass):
- go build ./...
- go vet ./...
- go test -short ./... — every package green
- go test -short -race ./internal/domain/... ./internal/api/handler/... — clean
- govulncheck ./... — no vulnerabilities in our code
- helm lint deploy/helm/certctl/ — clean
- helm template smoke render — succeeds
- python3 yaml.safe_load on api/openapi.yaml — parses
- OpenAPI Agent schema scan: no api_key_hash property
- CI guardrail mirror: clean on real tree, fires on all 4 synthetic
  regression patterns
- Domain pkg coverage: Agent.MarshalJSON 100%, connector.go total 87.5%
- Handler pkg coverage: 79.2%

Sample response body (httptest captured during verification, GET
/api/v1/agents/{id} via the new handler test):

  {"id":"agent-demo","name":"demo-agent","hostname":"demo.host",
  "status":"Online","last_heartbeat_at":"2026-04-24T11:59:30Z",
  "registered_at":"2026-04-24T12:00:00Z","os":"linux",
  "architecture":"amd64","ip_address":"10.0.0.42",
  "version":"v2.0.49"}

Note the absence of any api_key_hash key, even though the in-memory
struct passed to the handler had APIKeyHash set to a sentinel.

Out of scope (intentionally untouched):
- internal/repository/postgres/agent.go SELECT/INSERT/UPDATE/scan
  paths and GetByAPIKey lookup — DB column stays, repo still
  populates the struct, auth lookup still works. The redaction is a
  marshal-boundary concern.
- migrations/000001_initial_schema.up.sql + migrations/seed_*.sql —
  DB schema and seed data unchanged.
- internal/service/agent.go::RegisterAgent — service-side hashing
  and persistence unchanged.
- Other domain types with potential credential-derivative fields
  (Issuer.Config, DeploymentTarget.Config, notifier configs). Not
  flagged by the audit; some are already protected (e.g.,
  DeploymentTarget.EncryptedConfig []byte `json:"-"`). File a
  separate audit pass if recon surfaces additional leaks.
- Per-resource DTO layer across every handler. Single audit
  finding, single domain type.
- A separate possible follow-up: the v2 RegisterAgent endpoint
  doesn't return the plaintext API key to the agent, which may
  mean self-bootstrap via POST /api/v1/agents is broken. Verified
  during recon; out of scope for G-2; should be its own ticket.

Refs: coverage-gap-audit-2026-04-24-v5/unified-audit.md
      §2 P1 cluster, cat-s5-apikey_leak
      Audit recommendation: 'json:"-" or API-response DTO
      excluding APIKeyHash' — went with the json:"-" + MarshalJSON
      defense-in-depth pair plus CI guardrail and structural docs.
2026-04-25 01:56:26 +00:00
shankar0123 697fa792ea Merge branch 'fix/g1-jwt-silent-auth-downgrade-removal' 2026-04-25 00:22:33 +00:00
shankar0123 9c1d446e40 fix(security,config): remove unimplemented JWT auth-type, close silent downgrade (G-1)
The pre-G-1 config validator accepted CERTCTL_AUTH_TYPE=jwt and the
startup log faithfully echoed 'authentication enabled type=jwt'.
Reasonable people read that and concluded JWT auth was on. It wasn't.
The auth-middleware wiring at cmd/server/main.go unconditionally routed
every request through the api-key bearer middleware regardless of
cfg.Auth.Type. So CERTCTL_AUTH_TYPE=jwt quietly compared the incoming
'Authorization: Bearer <token>' against whatever string the operator put
in CERTCTL_AUTH_SECRET — real JWT clients got 401, and operators who
treated CERTCTL_AUTH_SECRET as a *signing* secret (because they thought
they were configuring JWT) had effectively handed an attacker an api-key.
A security finding masquerading as a config option.

We chose the audit-recommended structural fix: remove the option, fail
fast at startup, and add the gateway-fronting pattern as the documented
forward path. Implementing JWT middleware would have meant jwks vs
static-secret rotation, claim mapping, expiry enforcement, audience and
issuer validation, key rollover semantics, and regression coverage at the
same depth as the existing api-key path — a feature, not a fix. Operators
who genuinely need JWT/OIDC front certctl with an authenticating gateway
(oauth2-proxy / Envoy ext_authz / Traefik ForwardAuth / Pomerium /
Authelia) and run the upstream certctl with CERTCTL_AUTH_TYPE=none. Same
shape works on docker-compose and Helm.

The change is comprehensive across 7 phases — every surface that
mentioned 'jwt' as a certctl-auth-type is updated, plus structural
backstops (typed enum, runtime guard, helm template validation, CI grep
guard) so the lie can't reappear.

Files changed:

Phase 1 — production code (typed enum + jwt removal):
- internal/config/config.go: AuthType typed alias + AuthTypeAPIKey /
  AuthTypeNone constants + ValidAuthTypes() helper. Validate() routes
  literal 'jwt' through a dedicated multi-line diagnostic naming the
  authenticating-gateway pattern, then cross-checks against
  ValidAuthTypes(). Secret-required branch simplified to api-key-only.
  Field comment on AuthConfig.Type rewritten to drop jwt and point at
  the gateway pattern.
- internal/api/middleware/middleware.go: AuthConfig.Type field comment
  references the typed config.AuthType constants.
- internal/api/handler/health.go: same treatment for HealthHandler.AuthType.
- cmd/server/main.go: defense-in-depth runtime switch immediately after
  config.Load() — exits 1 on any unsupported auth-type that bypassed the
  validator. Auth-disabled startup log explicitly names the
  authenticating-gateway pattern.

Phase 2 — tests (Red→Green, contract pinning):
- internal/config/config_test.go: TestValidate_JWTAuth_RejectedDedicated
  (two table rows pinning the dedicated G-1 error fires regardless of
  whether Secret is set), TestValidAuthTypesDoesNotContainJWT (property
  guard against future re-introduction),
  TestValidAuthTypesIsExactly_APIKey_None (allowed-set contract),
  TestValidate_GenericInvalidAuthType (pins non-jwt invalid values still
  hit the generic invalid-auth-type error). Removed the prior
  TestValidate_JWTAuth_MissingSecret happy-path since its premise is
  inverted post-G-1.
- internal/api/handler/health_test.go: removed
  TestAuthInfo_ReturnsAuthType_JWT (which baked the silent-downgrade lie
  into the regression suite). Pre-existing _APIKey test continues to
  cover the api-key happy path.

Phase 3 — spec, docs, env templates:
- api/openapi.yaml: auth_type enum dropped to [api-key, none] with
  inline comment naming the G-1 closure.
- .env.example (root): CERTCTL_AUTH_TYPE comment block rewritten to drop
  jwt and point at the gateway pattern; secret-required conditional
  simplified to api-key-only.
- docs/architecture.md: middleware-stack bullet rewritten to drop the
  JWT mention; new H3 'Authenticating-gateway pattern (JWT, OIDC, mTLS)'
  section explaining the design rationale and listing oauth2-proxy /
  Envoy ext_authz / Traefik ForwardAuth / Pomerium / Authelia / Caddy
  forward_auth / Apache mod_auth_openidc / nginx auth_request as the
  standard fronting options.
- docs/upgrade-to-v2-jwt-removal.md (new ~125 lines): migration guide
  with preconditions, what-changes, both recovery paths, complete
  docker-compose oauth2-proxy walkthrough, Traefik ForwardAuth and Envoy
  ext_authz patterns, rollback posture.

Phase 4 — Helm chart (template validation + docs):
- deploy/helm/certctl/templates/_helpers.tpl: new certctl.validateAuthType
  helper mirroring the existing certctl.tls.required pattern. Fails
  template render on any server.auth.type outside {api-key, none} with
  a multi-line diagnostic.
- deploy/helm/certctl/templates/server-deployment.yaml,
  server-configmap.yaml, server-secret.yaml: invoke the helper at the
  top of each template that depends on .Values.server.auth.type.
- deploy/helm/certctl/values.yaml: auth: block comment expanded with the
  G-1 rationale and gateway-pattern cross-reference.
- deploy/helm/CHART_SUMMARY.md: server.auth.type table row now surfaces
  the allowed set and points at the upgrade doc.
- deploy/helm/certctl/README.md: new 'JWT / OIDC via authenticating
  gateway' section with a Kubernetes-flavored oauth2-proxy + certctl
  walkthrough.

Phase 5 — release surface:
- CHANGELOG.md: new [unreleased] top entry with Breaking / Removed /
  Added / Changed sections; explicit pointer at
  docs/upgrade-to-v2-jwt-removal.md from the Breaking subsection.

Phase 6 — CI guardrail:
- .github/workflows/ci.yml: new 'Forbidden auth-type literal regression
  guard (G-1)' step. Scoped patterns catch the actual regression shapes
  (map literal, slice literal, switch case, OpenAPI enum, env-file
  default, AuthType('jwt') cast). Comments and the dedicated rejection
  branch are intentionally exempt; connector-package JWT references
  (Google OAuth2 / step-ca) are exempt as out-of-scope external
  protocols. Verified locally: the guard passes on the actual tree and
  fires on all 4 synthetic regression patterns.

Out of scope (explicitly untouched):
- internal/connector/discovery/gcpsm/gcpsm.go — Google OAuth2 service-
  account JWT (external protocol).
- internal/connector/issuer/googlecas/googlecas.go — same.
- internal/connector/issuer/stepca/stepca.go — step-ca's provisioner
  one-time-token JWT for /sign API.
- docs/test-env.md, docs/connectors.md, docs/features.md — describe
  external CAs' use of JWT, not certctl's auth shape.
- Implementing actual JWT middleware. Feature, not a fix.

Verification (all gates pass):
- go build ./... — clean
- go vet ./... — clean
- go test -short ./... — every package green
- go test -short -race ./internal/config/... ./internal/api/... — clean
- govulncheck ./... — no vulnerabilities in our code
- helm lint deploy/helm/certctl/ — clean
- helm template with auth.type=api-key — renders OK
- helm template with auth.type=none — renders OK
- helm template with auth.type=jwt — fails with validateAuthType
  diagnostic (exit 1)
- python3 yaml.safe_load on api/openapi.yaml — parses
- CI guardrail mirror — clean on real tree, fires on all 4 synthetic
  regression patterns
- Smoke test: 'CERTCTL_AUTH_TYPE=jwt ./certctl-server' exits non-zero
  with: 'Failed to load configuration: CERTCTL_AUTH_TYPE=jwt is no
  longer accepted (G-1 silent auth downgrade): no JWT middleware ships
  with certctl. To use JWT/OIDC, run an authenticating gateway
  (oauth2-proxy / Envoy ext_authz / Traefik ForwardAuth / Pomerium) in
  front of certctl and set CERTCTL_AUTH_TYPE=none on the upstream.
  See docs/architecture.md "Authenticating-gateway pattern" and
  docs/upgrade-to-v2-jwt-removal.md for the migration walkthrough'

config pkg coverage: ValidAuthTypes 100%, Validate 94.7%, total 75.5%.

Refs: coverage-gap-audit-2026-04-24-v5/unified-audit.md
      §2 P1 cluster, cat-g-jwt_silent_auth_downgrade
      Audit recommendation followed verbatim: 'Remove jwt from
      validAuthTypes until middleware ships'.
2026-04-25 00:22:23 +00:00
shankar0123 3192cd15c5 Merge branch 'fix/u1-followups-helm-rootenv-examples' 2026-04-24 23:51:18 +00:00
shankar0123 af47d19ae2 fix(deploy,examples,env): close U-1 trap end-to-end across Helm, examples, and root env
Follow-up to cfc234e (U-1 docker-compose fix) — closes the remaining adjacent
code paths that share the postgres-first-boot-password-binding root cause but
were scoped out of the original commit.

The runtime diagnostic in internal/repository/postgres/db.go::wrapPingError
(landed in a911970) already covers every NewDB call site, so Helm operators
and example users hit the SQLSTATE 28P01 guidance for free at startup. What
was missing: deployment-shape-specific remediation guidance (kubectl vs
docker-compose), the hardcoded password in the *root* .env.example, and
shared ops notes for the 5 examples/ compose files. This commit closes all
three.

Files changed:

- .env.example (root) — line 16 had `postgres://certctl:certctl@...` with
  the password hardcoded literally instead of interpolating POSTGRES_PASSWORD.
  Edit if a user copied this file as their .env (binary-direct deployment,
  not docker-compose) and rotated POSTGRES_PASSWORD on line 10, the URL on
  line 16 still carried 'certctl' — silent two-line drift. Replaced 'certctl'
  with the same default that line 10 carries ('change-me-in-production') and
  added an explanatory comment block describing the docker-compose
  override semantics, when this URL matters (binary-direct), and the
  cross-reference to the U-1 wrapPingError diagnostic. Also fixed an
  adjacent bug: line 31 CERTCTL_SERVER_URL was `http://localhost:8443`,
  which agents reject at startup since v2.2 (HTTPS-everywhere milestone made
  the control plane HTTPS-only with TLS 1.3 pinned). Updated to https://
  with a comment pointing operators at the bootstrap CA bundle.

- deploy/helm/certctl/values.yaml — postgresql.auth.password field had a
  one-line 'REQUIRED' comment. Expanded into a full WARNING block (~25
  lines) explaining the PVC retention semantics, the failure symptom,
  and both kubectl-flavored remediation paths: non-destructive
  (`kubectl exec ... ALTER ROLE`) preferred for environments with data,
  and destructive (`helm uninstall + kubectl delete pvc`) for dev/demo.
  Cross-references the wrapPingError runtime diagnostic.

- deploy/helm/certctl/README.md (new, ~115 lines) — chart-level operational
  guide. Covers quick install, both remediation paths with concrete
  kubectl commands, why-we-don't-fix-this-in-the-chart explanation,
  cross-references to the docker-compose docs, server API key rotation
  (the easy case — comma-separated key list), TLS provisioning shapes,
  embedded-vs-external postgres, and uninstall semantics with the PVC
  retention gotcha called out.

- examples/README.md (new, ~55 lines) — shared operational notes for the
  5 example deployments. Covers the postgres password rotation trap with
  example-flavored remediation paths (`docker compose -f examples/<x>/...`),
  the TLS warning, and teardown semantics. Replaces what would otherwise
  be 5x duplication across per-example READMEs.

- examples/{acme-nginx,acme-wildcard-dns01,multi-issuer,private-ca-traefik,
  step-ca-haproxy}/*.md — one-line cross-reference at the top of each
  example's primary doc, pointing at examples/README.md for the shared
  ops notes. Avoids 5x duplication of the same warning text while still
  surfacing the link in every operator's first-touch surface.

Verification:

- go build ./... — clean
- go vet ./... — clean
- go test -short ./internal/repository/postgres/ — 4/4 wrapPingError tests
  still passing (no production-code touch in this commit)
- helm lint deploy/helm/certctl/ — clean (1 INFO about chart icon, pre-existing)
- helm template smoke test — renders without error
- python3 yaml.safe_load on values.yaml — parses

Refs: coverage-gap-audit-2026-04-24-v5/unified-audit.md
      §2 P1 cluster, cat-u-quickstart_postgres_password_volume_trap
      Closes the three deliberate scope-outs from cfc234e (Helm,
      root .env.example, examples/) end-to-end.

      Adjacent bugs caught while in scope:
      - root .env.example:16 hardcoded password not matching line 10
      - root .env.example:31 http:// URL incompatible with HTTPS-only v2.2
2026-04-24 23:51:13 +00:00
shankar0123 cfc234ec42 Merge branch 'fix/u1-postgres-password-volume-trap-diagnostic' 2026-04-24 23:21:33 +00:00
shankar0123 a91197014f fix(db): emit volume-state guidance on postgres auth failure (U-1, #10)
The shipped quickstart instructs operators to copy deploy/.env.example to
deploy/.env, edit POSTGRES_PASSWORD, and run docker compose up. On the
*first* boot of a fresh checkout this works. On the *second* boot — i.e.,
when an operator first booted with the default POSTGRES_PASSWORD=certctl,
then edited .env and re-ran up — the certctl-server container picks up the
new password (env interpolated at every container start) but postgres does
not. The postgres docker-entrypoint runs initdb only when the data dir is
empty; on subsequent boots the persistent named volume postgres_data is
non-empty so pg_authid retains the password baked in on first boot. The
server connects with the new credentials, postgres rejects them, and the
operator sees an opaque `pq: password authentication failed for user
"certctl"` in the server log with no pointer to the actual cause. New-
operator onboarding gets blocked on the documented production path.

Why a doc fix alone is not sufficient. Operators don't reread the docs
after a successful first boot — the trap fires on the *second* up, when
they think they've already learned the system. The opaque pq error is
indistinguishable in the log from a typo'd password or a misconfigured
secret store. The diagnostic has to fire at the moment the failure is
observed.

Why we don't try to fix the bootstrap. The env-vs-pg_authid divergence is
intrinsic to how the official postgres image bootstraps (see
docker-entrypoint.sh: initdb runs only if PGDATA is empty). Switching to a
bind mount or ephemeral volume breaks the production path; switching to
POSTGRES_PASSWORD_FILE + ALTER ROLE adds operator surface without
eliminating the divergence. The ergonomic fix is to surface the failure
mode loudly, with both remediation paths, at the exact log line where it
becomes visible.

Two remediation paths, surfaced together. Destructive: `docker compose
-f deploy/docker-compose.yml down -v && up -d --build` — wipes the
postgres volume so initdb re-runs with the new env value. Use this on
demos / first-time setup where data loss is acceptable. Non-destructive:
`docker compose exec postgres psql -U certctl -c "ALTER ROLE certctl
PASSWORD '<new>';"` followed by a server restart with the matching
POSTGRES_PASSWORD. Use this on any environment that holds data you want
to keep. Surfacing both means the operator can pick based on their
environment without us assuming.

Files changed:

- internal/repository/postgres/db.go — extract wrapPingError(err) helper.
  errors.As against *pq.Error; on SQLSTATE 28P01 (invalid_password) emit
  the multi-line guidance preserving the %w wrap chain. Non-28P01 errors
  retain the original `failed to ping database: %w` shape so transient
  connection-refused / timeout paths don't get noisy. Add
  pgErrInvalidPassword = "28P01" constant. Convert blank
  `_ "github.com/lib/pq"` import to direct import (driver registration
  still works via init()) so we can name the *pq.Error type at compile
  time. NewDB now calls wrapPingError(err) instead of inlining the wrap.
- internal/repository/postgres/db_test.go (new) — 4 internal-package
  unit tests covering wrapPingError. AuthFailureGuidance pins the
  contract substrings ("SQLSTATE 28P01", "POSTGRES_PASSWORD",
  "first boot", "down -v", "ALTER ROLE"). NonAuthErrorPreservesOriginalWrap
  pins the no-leak contract for SQLSTATE 08006 (connection_failure).
  NonPqErrorPreservesOriginalWrap pins the network-level path.
  NilReturnsNil pins defensive contract. All run in -short without
  testcontainers — package postgres (internal) so the unexported helper
  is callable directly.
- docs/quickstart.md — `> **Warning:**` callout immediately after the
  `cp deploy/.env.example deploy/.env` block at lines 56-61. Names the
  trap, names the SQLSTATE, gives both remediation paths. Uses the
  in-file `> **Note:**` blockquote convention.
- deploy/ENVIRONMENTS.md — `**Stateful volume — first-boot password
  binding (U-1)**` paragraph appended to the Postgres expert-note block.
  Explains the env-vs-pg_authid divergence, points at wrapPingError as
  the runtime diagnostic, lists both remediation paths. Uses the in-file
  `**Expert note:**` convention.

Out of scope (separate follow-ups):

- deploy/helm/certctl/templates/postgres-statefulset.yaml has the same
  root cause via PVC retention. The wrapPingError diagnostic covers the
  Helm path because the same NewDB code runs at server startup; the
  Helm-specific doc warning lands separately.
- /.env.example at repo root (line 16 hardcodes the password literally
  inside CERTCTL_DATABASE_URL rather than interpolating) — adjacent
  trap, separate fix.
- examples/{acme-nginx,private-ca-traefik,step-ca-haproxy,multi-issuer,
  acme-wildcard-dns01}/docker-compose.yml all carry the pattern. The
  diagnostic covers them; targeted doc warnings are scoped to the
  canonical quickstart + ENVIRONMENTS docs.

Out of consideration:

- Switch to bind mount / ephemeral volume — breaks the production path.
- POSTGRES_PASSWORD_FILE + Docker secret + ALTER ROLE rotation — adds
  operator surface without fixing the env-vs-pg_authid divergence.

Verification (all passing):
- go build ./...
- go vet ./...
- go test -short -race ./internal/repository/postgres/ — 4/4 new tests
  pass plus existing tests
- go test -short ./... — every package green
- govulncheck ./... — no vulnerabilities in our code
- wrapPingError coverage 100%; postgres pkg total unchanged in shape
  (NewDB/RunMigrations were 0% pre-fix, still 0% post-fix; new helper
  adds 100%-covered statements)

Refs: coverage-gap-audit-2026-04-24-v5/unified-audit.md
      §2 P1 cluster, cat-u-quickstart_postgres_password_volume_trap
      GitHub Issue #10 (mikeakasully)
2026-04-24 23:21:26 +00:00
37 changed files with 1942 additions and 92 deletions
+25 -4
View File
@@ -13,22 +13,43 @@ POSTGRES_PASSWORD=change-me-in-production
# Certctl Server
# All server vars use the CERTCTL_ prefix (see internal/config/config.go)
# ==============================================================================
CERTCTL_DATABASE_URL=postgres://certctl:certctl@postgres:5432/certctl?sslmode=disable
# IMPORTANT: keep the password segment of CERTCTL_DATABASE_URL in sync with
# POSTGRES_PASSWORD above. If you deploy via `deploy/docker-compose.yml`,
# this value is *overridden* by the compose file's
# `postgres://certctl:${POSTGRES_PASSWORD:-certctl}@postgres:5432/...`
# interpolation — but if you run the binary directly with this .env loaded
# (e.g. `set -a; source .env; ./certctl-server`), update *both* lines.
# Background: editing POSTGRES_PASSWORD after the postgres data directory
# has been initialized once does NOT rotate the password — initdb only
# seeds pg_authid on first boot of an empty volume. See docs/quickstart.md
# "Warning" callout and `internal/repository/postgres/db.go::wrapPingError`
# for the SQLSTATE 28P01 diagnostic that fires when the two drift.
CERTCTL_DATABASE_URL=postgres://certctl:change-me-in-production@postgres:5432/certctl?sslmode=disable
CERTCTL_SERVER_HOST=0.0.0.0
CERTCTL_SERVER_PORT=8443
CERTCTL_LOG_LEVEL=info
CERTCTL_LOG_FORMAT=json
# Auth type: "api-key", "jwt", or "none" (for demo/development)
# Auth type: "api-key" (production) or "none" (demo/development).
# For JWT/OIDC, run an authenticating gateway in front of certctl
# (oauth2-proxy / Envoy ext_authz / Traefik ForwardAuth / Pomerium) and
# set CERTCTL_AUTH_TYPE=none on the upstream — see
# docs/architecture.md "Authenticating-gateway pattern". G-1 removed
# the in-process "jwt" option (no JWT middleware shipped — silent auth
# downgrade); see docs/upgrade-to-v2-jwt-removal.md if you previously
# set CERTCTL_AUTH_TYPE=jwt.
CERTCTL_AUTH_TYPE=none
# Required when CERTCTL_AUTH_TYPE is "api-key" or "jwt"
# Required when CERTCTL_AUTH_TYPE is "api-key".
# Generate with: openssl rand -base64 32
# CERTCTL_AUTH_SECRET=change-me-in-production
# ==============================================================================
# Certctl Agent
# ==============================================================================
CERTCTL_SERVER_URL=http://localhost:8443
# HTTPS-only as of v2.2 (TLS 1.3 pinned). Agents reject http:// URLs at
# startup. Use the docker-compose self-signed bootstrap CA bundle from
# `deploy/test/certs/ca.crt` or supply your own via CERTCTL_SERVER_CA_BUNDLE_PATH.
CERTCTL_SERVER_URL=https://localhost:8443
CERTCTL_API_KEY=change-me-in-production
CERTCTL_AGENT_NAME=local-agent
+169
View File
@@ -44,6 +44,175 @@ jobs:
- name: Run govulncheck
run: govulncheck ./...
- name: Forbidden auth-type literal regression guard (G-1)
# G-1 closed the JWT silent auth downgrade by removing "jwt" from the
# accepted CERTCTL_AUTH_TYPE values. This step grep-fails the build
# if "jwt" reappears in any of the *additive* auth-type surfaces:
# the validAuthTypes / ValidAuthTypes() set, the OpenAPI enum, the
# helm chart's allowed-types list, or the .env.example default.
# Comment lines and the dedicated rejection branch in config.go
# (`c.Auth.Type == "jwt"`) are intentionally exempt — those are the
# G-1 fix itself, not a regression.
#
# Connector packages (internal/connector/) are exempt because the
# Google OAuth2 service-account JWT and step-ca provisioner one-
# time-token JWT are external-protocol uses, unrelated to certctl's
# own auth shape. Test files (_test.go) are exempt so negative
# tests can pass the literal.
#
# See docs/upgrade-to-v2-jwt-removal.md for the closure rationale,
# or internal/config/config.go::ValidAuthTypes for the allowed set.
run: |
set -e
# Scoped patterns that indicate "jwt" being added back to an
# allowed-set surface. Each catches a regression shape we've
# actually seen in pre-G-1 code:
# - Go map/slice literal: "jwt": true or "jwt",
# - Go switch case: case "jwt"
# - YAML enum: enum: [..., jwt, ...] or - jwt
# - .env conditional: AUTH_TYPE.*"jwt"|=jwt$
BAD=$(grep -rnEH \
-e '"jwt"\s*:\s*true' \
-e '"jwt"\s*,' \
-e 'case\s+"jwt"' \
-e 'enum:.*\bjwt\b' \
-e '^\s*-\s*jwt\s*$' \
-e 'AUTH_TYPE\s*=\s*jwt\s*$' \
-e 'AUTH_TYPE\s*=\s*jwt\s*#' \
-e 'auth\.type\s*=\s*jwt\s*$' \
-e 'AuthType\("jwt"\)' \
internal/config/ \
internal/api/ \
cmd/ \
api/openapi.yaml \
.env.example \
deploy/.env.example \
deploy/helm/certctl/values.yaml \
deploy/helm/certctl/templates/ \
2>/dev/null \
| grep -v '_test.go' \
| grep -vE '^\s*[^:]+:[0-9]+:\s*(//|#)' \
| grep -v 'is no longer accepted' \
|| true)
if [ -n "$BAD" ]; then
echo "G-1 regression: \"jwt\" reappeared in an allowed-set surface:"
echo "$BAD"
echo ""
echo "Allowed surface for 'jwt' literals: comment lines, the"
echo "dedicated rejection branch in internal/config/config.go,"
echo "and connector packages (Google OAuth2, step-ca)."
echo "See docs/upgrade-to-v2-jwt-removal.md and"
echo "internal/config/config.go::ValidAuthTypes()."
exit 1
fi
- name: Forbidden api_key_hash JSON-shape regression guard (G-2)
# G-2 closed cat-s5-apikey_leak by tagging Agent.APIKeyHash
# `json:"-"` and adding a defense-in-depth Agent.MarshalJSON that
# zeroes the field on the marshal-time copy. This step grep-fails
# the build if `api_key_hash` reappears in any of the *additive*
# JSON-emitting surfaces: a Go struct json tag in internal/domain/,
# an OpenAPI Agent schema property, a TypeScript field declaration
# in web/src/, or an enum-list / discriminator in handler
# production code.
#
# Repository, migration, seed, service, integration-test, and
# unit-test files are exempt — those are server-internal use
# sites (the DB column stays, the in-memory struct field stays,
# the auth-lookup path stays). Comment lines are exempt so the
# G-2 closure rationale can stay in the source.
#
# See coverage-gap-audit-2026-04-24-v5/unified-audit.md
# cat-s5-apikey_leak for the closure rationale, or
# internal/domain/connector.go::Agent::MarshalJSON for the
# redaction enforcement.
run: |
set -e
# Scoped patterns that indicate api_key_hash being added back
# to a JSON-emitting surface. Each catches a regression shape
# that pre-G-2 actually shipped or that a future refactor
# could plausibly introduce:
# - Go struct tag: `json:"api_key_hash"`
# - Frontend interface: api_key_hash[?]: string
# - OpenAPI schema property: api_key_hash: (column-aligned)
# - YAML enum / array: - api_key_hash
BAD=$(grep -rnEH \
-e 'json:"api_key_hash[",]' \
-e '^\s*api_key_hash\??\s*:' \
-e '^\s*-\s*api_key_hash\s*$' \
internal/domain/ \
internal/api/ \
cmd/ \
api/openapi.yaml \
web/src/ \
2>/dev/null \
| grep -v '_test.go' \
| grep -vE '^\s*[^:]+:[0-9]+:\s*(//|#)' \
|| true)
if [ -n "$BAD" ]; then
echo "G-2 regression: api_key_hash reappeared in a JSON-emitting surface:"
echo "$BAD"
echo ""
echo "Allowed surface for api_key_hash literals: comment lines,"
echo "the database column (migrations/), the in-memory struct"
echo "field tagged \`json:\"-\"\`, and the repository / service"
echo "use sites. See internal/domain/connector.go::Agent and"
echo "coverage-gap-audit-2026-04-24-v5/unified-audit.md"
echo "cat-s5-apikey_leak for the closure rationale."
exit 1
fi
- name: Forbidden plaintext HEALTHCHECK regression guard (U-2)
# U-2 closed cat-u-healthcheck_protocol_mismatch by switching the
# published image's HEALTHCHECK from `curl -f http://localhost:
# 8443/health` (always failed against the HTTPS-only listener) to
# `curl -fsk https://localhost:8443/health`. This step grep-fails
# the build if any Dockerfile in the repo carries the pre-U-2
# plaintext shape — either explicitly (`http://localhost:8443/
# health` in a HEALTHCHECK) or via the looser pattern of any
# HEALTHCHECK that targets `http://` against the certctl server
# port.
#
# Comment lines and the docs/upgrade-to-tls.md:182 expected-to-
# fail invariant ("plaintext is gone, expect Connection refused")
# are intentionally exempt — we DO want the upgrade-doc string
# `http://localhost:8443/health` to remain there, since it
# documents what operators should test for to confirm plaintext
# is dead. The guardrail is scoped to Dockerfile* only, so docs
# are out of its reach.
#
# See coverage-gap-audit-2026-04-24-v5/unified-audit.md
# cat-u-healthcheck_protocol_mismatch for the closure rationale,
# or deploy/test/healthcheck_test.go for the binary-image
# contract the runtime test pins.
run: |
set -e
# Patterns that catch the actual regression shapes:
# - HEALTHCHECK directive carrying any http:// (even if the
# port differs, no plaintext probe should ship).
# - The exact pre-U-2 string for grep-friendliness.
BAD=$(grep -rnEH \
-e 'HEALTHCHECK.*http://' \
-e 'curl[^|&;]*-f[^|&;]*http://localhost:8443/health' \
Dockerfile Dockerfile.agent Dockerfile.* 2>/dev/null \
| grep -vE '^\s*[^:]+:[0-9]+:\s*#' \
|| true)
if [ -n "$BAD" ]; then
echo "U-2 regression: plaintext HEALTHCHECK reappeared in a Dockerfile:"
echo "$BAD"
echo ""
echo "Allowed: HTTPS HEALTHCHECK with -k (acceptable for"
echo "localhost-to-localhost), or non-HTTP probe shapes"
echo "(pgrep, /proc check). See Dockerfile / Dockerfile.agent"
echo "for the post-U-2 reference shape and"
echo "coverage-gap-audit-2026-04-24-v5/unified-audit.md"
echo "cat-u-healthcheck_protocol_mismatch for rationale."
exit 1
fi
- name: Race Detection
run: go test -race ./internal/service/... ./internal/api/handler/... ./internal/api/middleware/... ./internal/scheduler/... ./internal/connector/... ./internal/crypto/... ./internal/domain/... ./internal/validation/... ./internal/tlsprobe/... -count=1 -timeout 300s
+57
View File
@@ -2,6 +2,63 @@
All notable changes to certctl are documented in this file. Dates use ISO 8601. Versions follow [Semantic Versioning](https://semver.org/).
## [unreleased] — 2026-04-24
### G-1: JWT silent auth downgrade — closed end-to-end
> Pre-G-1 the config validator accepted `CERTCTL_AUTH_TYPE=jwt` and the startup log faithfully echoed `"authentication enabled" "type"="jwt"`. Reasonable people read that and concluded JWT was on. It wasn't. The auth-middleware wiring at `cmd/server/main.go` unconditionally routed every request through the api-key bearer middleware regardless of `cfg.Auth.Type`. So `CERTCTL_AUTH_TYPE=jwt` quietly compared incoming `Authorization: Bearer <something>` against whatever string the operator put in `CERTCTL_AUTH_SECRET` — real JWT clients got 401, and operators who treated `CERTCTL_AUTH_SECRET` as a *signing* secret (because they thought they were configuring JWT) had effectively handed an attacker an api-key. A security finding masquerading as a config option. We chose to remove the option rather than ship JWT middleware — the audit-recommended structural fix that closes the hazard. Operators who actually need JWT/OIDC front certctl with an authenticating gateway (oauth2-proxy / Envoy `ext_authz` / Traefik `ForwardAuth` / Pomerium / Authelia) and run the upstream certctl with `CERTCTL_AUTH_TYPE=none`. The same pattern works on docker-compose and Helm.
### Breaking Changes
- **`CERTCTL_AUTH_TYPE=jwt` is no longer accepted.** Pre-G-1 the value was silently downgraded to api-key middleware. Post-G-1 the server fails at startup with a dedicated diagnostic naming the authenticating-gateway pattern. Operators with this in their env block must either switch to `api-key` (if they were de facto using api-key auth all along — same Bearer token continues to work) or switch to `none` and front certctl with an oauth2-proxy / Envoy / Traefik / Pomerium gateway. See [`docs/upgrade-to-v2-jwt-removal.md`](docs/upgrade-to-v2-jwt-removal.md).
- **Helm chart `server.auth.type=jwt` now fails at `helm install` / `helm upgrade` template time.** New `certctl.validateAuthType` template helper runs on every template that depends on `.Values.server.auth.type` (`server-deployment.yaml`, `server-configmap.yaml`, `server-secret.yaml`) and fails the render with a pointer at the gateway-fronting pattern.
- **OpenAPI spec `auth_type` enum no longer includes `jwt`.** API consumers checking `/api/v1/auth/info` against the spec will see a smaller enum.
### Removed
- Documented references to JWT in the certctl auth surface (config docblocks, middleware/health-handler comments, `.env.example`, `docs/architecture.md` middleware-stack bullet). Connector-level JWT references (Google OAuth2 service-account JWT in `internal/connector/discovery/gcpsm/`, `internal/connector/issuer/googlecas/`; step-ca's provisioner one-time-token JWT in `internal/connector/issuer/stepca/`) are unrelated and untouched — those are external-protocol uses, not certctl's own auth shape.
### Added
- **`config.AuthType` typed alias** with `AuthTypeAPIKey` / `AuthTypeNone` exported constants. Single source of truth for the allowed set across the validator, the runtime defense-in-depth switch in `main.go`, and the helm chart's `validateAuthType` helper.
- **`config.ValidAuthTypes()`** helper returning the complete allowed set; pinned by a property test (`TestValidAuthTypesDoesNotContainJWT`) that fails the build if `"jwt"` is ever re-added to the slice.
- **Defense-in-depth runtime guard** in `cmd/server/main.go` immediately after `config.Load()` — a `switch config.AuthType(cfg.Auth.Type)` that exits 1 if the validator was bypassed (test harness, alt config loader, env-var rebinding).
- **`certctl.validateAuthType` Helm template helper** mirroring the existing `certctl.tls.required` pattern. Fails template render on any `server.auth.type` outside `{api-key, none}`.
- **`docs/architecture.md` "Authenticating-gateway pattern (JWT, OIDC, mTLS)"** section explaining the design rationale for the narrow in-process auth surface and listing oauth2-proxy / Envoy `ext_authz` / Traefik `ForwardAuth` / Pomerium / Authelia / Caddy `forward_auth` / Apache `mod_auth_openidc` / nginx `auth_request` as the standard fronting options.
- **`docs/upgrade-to-v2-jwt-removal.md`** migration guide. Same shape as `docs/upgrade-to-tls.md`. Walks through the dedicated startup error, both recovery paths (`api-key` vs gateway-fronting), a complete docker-compose oauth2-proxy walkthrough, Traefik ForwardAuth and Envoy `ext_authz` patterns, and rollback posture.
- **`deploy/helm/certctl/README.md`** "JWT / OIDC via authenticating gateway" section with a Kubernetes-flavored oauth2-proxy + certctl walkthrough.
- **CI regression guardrail** in `.github/workflows/ci.yml` (`Forbidden auth-type literal regression guard (G-1)`) — grep-fails the build if `"jwt"` appears as an auth-type literal in production code or spec. Connector packages exempt (legitimate external-protocol uses).
- **Negative test coverage** in `internal/config/config_test.go`: `TestValidate_JWTAuth_RejectedDedicated` (two table rows pinning that the dedicated G-1 error fires regardless of whether `Secret` is set), `TestValidAuthTypesDoesNotContainJWT` (property-level guard), `TestValidAuthTypesIsExactly_APIKey_None` (allowed-set contract), `TestValidate_GenericInvalidAuthType` (pins that other invalid values still surface the generic invalid-auth-type error, so the dedicated G-1 path doesn't accidentally swallow non-jwt typos).
### Changed
- `internal/api/middleware/middleware.go::AuthConfig.Type` field comment now references the typed `config.AuthType` constants instead of an inline string enumeration.
- `internal/api/handler/health.go::HealthHandler.AuthType` field comment same treatment.
- `internal/api/handler/health_test.go` — the prior `TestAuthInfo_ReturnsAuthType_JWT` (which asserted the handler echoed `"jwt"`, baking the silent-downgrade lie into the regression suite) is removed; the pre-existing `TestAuthInfo_ReturnsAuthType_APIKey` continues to cover the api-key happy path.
- Auth-disabled startup log in `main.go` now points operators at the authenticating-gateway pattern explicitly.
### U-2: Dockerfile HEALTHCHECK protocol mismatch — closed end-to-end
> Pre-U-2 the published `ghcr.io/shankar0123/certctl-server` image shipped with `HEALTHCHECK CMD curl -f http://localhost:8443/health`. The server has been HTTPS-only since the v2.2 HTTPS-Everywhere milestone (`cmd/server/main.go::ListenAndServeTLS`, no plaintext fallback, TLS 1.3 pinned), so the probe failed every interval and Docker marked the container `unhealthy` indefinitely. Operators inside docker-compose / Helm / the example stacks were unaffected — compose overrides the HEALTHCHECK with `--cacert + https://`, Helm uses explicit `httpGet` probes that ignore Docker's HEALTHCHECK, and every example compose file overrides with `curl -sfk https://localhost:8443/health`. But anyone running bare `docker run` / Docker Swarm / Nomad / ECS — exactly the "I just pulled the published image" path — saw permanent `unhealthy` status and (depending on orchestrator policy) a restart-loop. Recon for U-2 also surfaced two adjacent bugs from the same v2.2 milestone gap: the Helm chart's `readinessProbe.httpGet.path` pointed at `/readyz`, a route the server doesn't register (only `/health` and `/ready` are wired and bypass the auth middleware), so K8s readiness probes were getting 404/auth-rejection and pods stayed `NotReady`; and the agent image had no HEALTHCHECK at all (the compose override called `pgrep -f certctl-agent` against an image that didn't ship `procps` — latent always-fail). All three are closed in this commit.
### Fixed
- **`Dockerfile` HEALTHCHECK now speaks HTTPS.** Bare `docker run` / Swarm / Nomad / ECS users no longer see `unhealthy` forever. The probe uses `curl -fsk https://localhost:8443/health``-k` (insecure) is acceptable because the probe is localhost-to-localhost: the same process serving the cert is being probed; the probe never traverses a network. Compose / Helm / examples already perform full cert-chain validation and are unaffected.
- **Helm `server.readinessProbe.httpGet.path` corrected from `/readyz` to `/ready`.** The `/readyz` path was never registered as a no-auth route (see `internal/api/router/router.go:81` and `cmd/server/main.go:920`), so K8s readiness probes received 401 (api-key auth rejection) or 404 (when auth was disabled). Pods previously failed to report Ready under most realistic Helm deployments. Liveness probe path (`/health`) was already correct and is unchanged.
- **`docs/connectors.md` curl examples** (15 sites) updated from `http://localhost:8443/...` to `https://localhost:8443/...` with a one-time `--cacert "$CA"` extraction note matching the existing pattern in `docs/quickstart.md`. Pre-U-2 these examples silently failed against the HTTPS listener.
### Added
- **`Dockerfile.agent` HEALTHCHECK** — `pgrep -f certctl-agent` process-presence check (the agent has no HTTP listener; presence is the right primitive). Bare-`docker run` agents now report health-status the same way compose-managed ones do. Also adds `procps` to the runtime image so `pgrep` is actually available — pre-U-2 the docker-compose override at `deploy/docker-compose.yml:173` called `pgrep -f certctl-agent` against an image that lacked it (latent always-fail; container was reported unhealthy in compose too, just rarely noticed because nothing acted on the signal).
- **`deploy/test/healthcheck_test.go`** (`//go:build integration`) — image-level integration tests. `TestPublishedServerImage_HealthcheckSpecUsesHTTPS` builds the server image, inspects `Config.Healthcheck.Test` via `docker inspect`, and asserts the array contains `https://localhost:8443/health` and `-k`, and does NOT contain `http://localhost:8443/health` (negative regression contract). `TestPublishedAgentImage_HealthcheckSpecExists` builds the agent image and asserts the HEALTHCHECK uses `pgrep` against `certctl-agent`. Both tests `t.Skip` cleanly when docker isn't available (sandbox / CI without docker-in-docker). A third runtime test (`TestPublishedServerImage_HealthcheckTransitionsToHealthy`) is a `t.Skip` placeholder until the harness wires a sidecar postgres for image-level smoke — documented honestly so the next refactor adopts it instead of rediscovering the gap.
- **CI regression guardrail** in `.github/workflows/ci.yml` (`Forbidden plaintext HEALTHCHECK regression guard (U-2)`) — grep-fails the build if any `Dockerfile*` carries `HEALTHCHECK.*http://` or `curl -f http://localhost:8443/health`. Comments exempt; the `docs/upgrade-to-tls.md:182` post-cutover invariant string (which deliberately documents the expected-failure shape) is out of the guardrail's scope because the guardrail only scans Dockerfiles.
### Changed
- `Dockerfile` final-stage HEALTHCHECK lines now carry a long-form docblock explaining the `-k` design choice, the published-image vs compose vs Helm vs examples coverage matrix, and cross-references to the audit closure + the integration test.
- `Dockerfile.agent` runtime stage adds `procps` to the apk install so the new HEALTHCHECK and the existing compose override both have a working `pgrep`.
- `deploy/helm/certctl/values.yaml` server probes block now carries an explanatory comment naming the registered probe routes (`/health`, `/ready`) and the U-2 closure rationale for the `/readyz``/ready` correction.
## [2.2.0] — 2026-04-19
### HTTPS Everywhere — The Irony
+28 -1
View File
@@ -76,7 +76,34 @@ USER certctl
EXPOSE 8443
# Image-level HEALTHCHECK for bare `docker run` / Docker Swarm / Nomad / ECS.
#
# U-2 (P1, cat-u-healthcheck_protocol_mismatch): pre-U-2 this probe used
# `curl -f http://localhost:8443/health`, which always failed against the
# HTTPS-only listener (HTTPS-Everywhere milestone, v2.2 / tag v2.0.47 —
# `cmd/server/main.go::ListenAndServeTLS`, no plaintext fallback, TLS 1.3
# pinned). Operators outside docker-compose / Helm saw permanent
# `unhealthy` status and a restart-loop the first time they pulled the
# image. The compose stack overrides this HEALTHCHECK with `--cacert` to
# the bootstrap CA bundle (deploy/docker-compose.yml:126); the Helm chart
# uses explicit `httpGet` probes with `scheme: HTTPS` and ignores Docker's
# HEALTHCHECK; every example compose file in `examples/*/docker-compose.yml`
# overrides with `curl -sfk https://localhost:8443/health`. This image-
# level probe is for the bare-`docker run` consumer ONLY.
#
# `-k` (insecure) is acceptable here because the probe is localhost-to-
# localhost: the same process serving the cert is being probed; the probe
# never traverses a network. Pinning a `--cacert` is not viable for the
# published image because the bootstrap cert is per-deploy (generated into
# the `certs` named volume on first up; operator-supplied via Helm's
# `existingSecret` or cert-manager). Compose / Helm / examples already
# perform full cert-chain validation and are unaffected.
#
# CI grep guardrail at .github/workflows/ci.yml ("Forbidden plaintext
# HEALTHCHECK regression guard (U-2)") blocks reintroduction of the
# `http://` shape. Image-level integration test in
# deploy/test/healthcheck_test.go pins the contract end-to-end.
HEALTHCHECK --interval=10s --timeout=5s --start-period=5s --retries=5 \
CMD curl -f http://localhost:8443/health || exit 1
CMD curl -fsk https://localhost:8443/health || exit 1
ENTRYPOINT ["/app/server"]
+23 -1
View File
@@ -36,7 +36,14 @@ RUN CGO_ENABLED=0 GOOS=linux GOARCH=${TARGETARCH} go build \
# Stage 2: Runtime
FROM alpine:3.19
RUN apk add --no-cache ca-certificates curl
# U-2: `procps` ships pgrep, which the HEALTHCHECK below uses to verify the
# agent process is alive. Pre-U-2 the deploy/docker-compose.yml agent
# HEALTHCHECK called `pgrep -f certctl-agent` against this image but
# pgrep wasn't installed — the compose probe was a latent always-fail.
# Adding procps here fixes both the new image-level HEALTHCHECK and the
# pre-existing compose override. Adds ~250KB to the image; acceptable for
# observability parity with the server image.
RUN apk add --no-cache ca-certificates curl procps
RUN addgroup -g 1000 certctl && \
adduser -D -u 1000 -G certctl certctl
@@ -51,4 +58,19 @@ RUN mkdir -p /var/lib/certctl/keys && \
USER certctl
# Image-level HEALTHCHECK for bare `docker run` / Docker Swarm / Nomad / ECS.
#
# U-2 (P1, cat-u-healthcheck_protocol_mismatch — adjacent fix): the agent
# has no HTTP listener (it polls the server via outbound HTTPS), so a
# process-presence check is the correct primitive. Pre-U-2 the agent image
# shipped with no HEALTHCHECK at all, so bare-`docker run` operators got
# zero health signal and orchestrators that key off Docker's HEALTHCHECK
# (Swarm, Nomad, ECS) saw the container reported as `none`. The compose
# override at deploy/docker-compose.yml:173 used the same `pgrep -f
# certctl-agent` shape; we mirror it here so the published image has
# parity with the compose stack and the override on docker-compose.yml
# becomes redundant-but-correct rather than load-bearing.
HEALTHCHECK --interval=30s --timeout=5s --start-period=10s --retries=3 \
CMD pgrep -f certctl-agent > /dev/null || exit 1
ENTRYPOINT ["/app/agent"]
+20 -3
View File
@@ -132,7 +132,14 @@ paths:
properties:
auth_type:
type: string
enum: [api-key, jwt, none]
# G-1 (P1): "jwt" removed from this enum after the silent
# auth downgrade was identified — no JWT middleware ships
# with certctl. Operators who need JWT/OIDC front certctl
# with an authenticating gateway (oauth2-proxy / Envoy /
# Traefik / Pomerium) and set CERTCTL_AUTH_TYPE=none
# upstream. See docs/architecture.md "Authenticating-
# gateway pattern".
enum: [api-key, none]
required:
type: boolean
@@ -3680,8 +3687,18 @@ components:
registered_at:
type: string
format: date-time
api_key_hash:
type: string
# G-2 (P1): the `api_key_hash` field was REMOVED from this
# schema after cat-s5-apikey_leak audit closure. The DB column
# still exists (migrations/000001_initial_schema.up.sql) and
# the server still populates the in-memory struct for the
# auth-lookup path (repository.AgentRepository::GetByAPIKey),
# but the JSON wire shape no longer carries it — see
# internal/domain/connector.go::Agent::APIKeyHash + MarshalJSON
# for the redaction enforcement and docs/architecture.md ER
# diagram for the database-vs-API distinction. Do NOT re-add
# the field here without first removing the JSON-shape redaction
# in the domain package; the CI guardrail at
# .github/workflows/ci.yml will block re-introduction either way.
os:
type: string
architecture:
+23 -3
View File
@@ -39,6 +39,26 @@ func main() {
os.Exit(1)
}
// Defense-in-depth runtime guard for the auth-type discriminator.
//
// G-1 (P1): config.Load() already runs Validate() which rejects "jwt"
// and any value outside config.ValidAuthTypes() with a dedicated
// diagnostic. This switch is belt-and-braces — if a future refactor
// bypasses the validator (test harness, alt config loader, env-var
// rebinding after Load) the server must not silently boot with an
// unsupported auth shape. The error path uses fmt.Fprintf because
// the slog logger is constructed from cfg below this point; we want
// the failure to be visible regardless of log-level configuration.
switch config.AuthType(cfg.Auth.Type) {
case config.AuthTypeAPIKey, config.AuthTypeNone:
// ok — fall through
default:
fmt.Fprintf(os.Stderr,
"unsupported auth type at runtime: %q (valid: %v) — config validation should have caught this; refusing to start\n",
cfg.Auth.Type, config.ValidAuthTypes())
os.Exit(1)
}
// Set up structured logging
logger := slog.New(slog.NewJSONHandler(os.Stdout, &slog.HandlerOptions{
Level: cfg.GetLogLevel(),
@@ -615,7 +635,7 @@ func main() {
// compatibility CERTCTL_AUTH_SECRET is synthesized into legacy-key-N
// entries with Admin=false.
var namedKeys []middleware.NamedAPIKey
if cfg.Auth.Type != "none" {
if config.AuthType(cfg.Auth.Type) != config.AuthTypeNone {
// Translate typed config.NamedAPIKey -> middleware.NamedAPIKey. The
// two structs are field-compatible but live in different packages to
// preserve the config→middleware dependency direction.
@@ -704,8 +724,8 @@ func main() {
logger.Info("rate limiting enabled", "rps", cfg.RateLimit.RPS, "burst", cfg.RateLimit.BurstSize)
}
if cfg.Auth.Type == "none" {
logger.Warn("authentication disabled (CERTCTL_AUTH_TYPE=none) — not suitable for production")
if config.AuthType(cfg.Auth.Type) == config.AuthTypeNone {
logger.Warn("authentication disabled (CERTCTL_AUTH_TYPE=none) — not suitable for production except behind an authenticating gateway (oauth2-proxy / Envoy ext_authz / Traefik ForwardAuth / Pomerium)")
} else {
logger.Info("authentication enabled", "type", cfg.Auth.Type)
}
+2
View File
@@ -122,6 +122,8 @@ The `volumes` section mounts 10 migration files into PostgreSQL's init directory
**Expert note:** The numbered prefix pattern (`001_`, `002_`, ..., `020_`) ensures deterministic execution order. All migrations use `IF NOT EXISTS` and `ON CONFLICT DO NOTHING` for idempotency, so re-running them against an existing database is safe.
**Stateful volume — first-boot password binding (U-1).** The same "first boot only" semantics that govern migration scripts also govern `POSTGRES_PASSWORD`. The official `postgres` image runs `initdb` exactly once — when `/var/lib/postgresql/data` is empty — and that pass is the only time `POSTGRES_PASSWORD` is written into `pg_authid`. On every subsequent boot, the postgres container ignores the env var and authenticates against whatever password was baked into the data directory on the original `up`. Editing `POSTGRES_PASSWORD` in `.env` after a successful first boot therefore only updates the **certctl-server** container's `CERTCTL_DATABASE_URL` — postgres still expects the previous password, and the server fails to ping with `pq: password authentication failed for user "certctl"` (SQLSTATE 28P01). The certctl-server container surfaces this case explicitly: when SQLSTATE 28P01 fires at startup, the wrap text in `internal/repository/postgres/db.go::wrapPingError` points operators at the two remediation paths — destructive volume teardown via `docker compose -f deploy/docker-compose.yml down -v && up -d --build`, or non-destructive in-place rotation via `docker compose -f deploy/docker-compose.yml exec postgres psql -U certctl -c "ALTER ROLE certctl PASSWORD '<new>';"` followed by a server restart with the matching `POSTGRES_PASSWORD`. Use the destructive path on the demo / first-time setup; use the non-destructive path on any environment that holds data you want to keep.
#### certctl Server
```yaml
+2 -2
View File
@@ -246,8 +246,8 @@ helm install certctl certctl/ \
|--------|---------|-------------|
| `server.replicas` | 1 | Number of server replicas |
| `server.port` | 8443 | Server port |
| `server.auth.type` | api-key | Authentication type |
| `server.auth.apiKey` | "" | API key (REQUIRED) |
| `server.auth.type` | api-key | Authentication type`api-key` or `none` (G-1: `jwt` removed; for JWT/OIDC use a fronting authenticating gateway, see `docs/architecture.md` and `docs/upgrade-to-v2-jwt-removal.md`) |
| `server.auth.apiKey` | "" | API key (REQUIRED when `auth.type=api-key`) |
| `server.logging.level` | info | Log level |
| `server.logging.format` | json | Log format |
+148
View File
@@ -0,0 +1,148 @@
# certctl Helm Chart
Production-ready Helm chart for deploying [certctl](https://github.com/shankar0123/certctl) on Kubernetes. Wires up the certctl server (Deployment), PostgreSQL (StatefulSet with PVC), and the agent (DaemonSet — one per node) on a private cluster, with health probes, security contexts, and optional Ingress.
## Quick install
```bash
helm install certctl deploy/helm/certctl/ \
--create-namespace --namespace certctl \
--set server.auth.apiKey="$(openssl rand -base64 32)" \
--set postgresql.auth.password="$(openssl rand -base64 24)"
```
This brings up:
- `<release>-server` Deployment (HTTPS-only on port 8443; TLS 1.3)
- `<release>-postgres` StatefulSet (PostgreSQL 16-alpine, 1 replica, 10Gi PVC by default)
- `<release>-agent` DaemonSet (polls server, generates ECDSA P-256 keys locally)
- Service objects, optional Ingress, and ServiceAccount with RBAC
See [`values.yaml`](values.yaml) for the full configuration surface — issuer settings, target connectors, scheduler intervals, notifier credentials, and resource requests/limits all live there.
## Operational notes
### Postgres password rotation — read this before changing `postgresql.auth.password`
**The trap.** `postgresql.auth.password` is bound to `pg_authid` exactly once — when the StatefulSet's PVC is provisioned and `initdb` runs. The official `postgres:16-alpine` image only runs `initdb` when `/var/lib/postgresql/data` is empty, so on every subsequent rollout the `POSTGRES_PASSWORD` env var is read into the container but **ignored** by postgres itself. The certctl-server container also picks up the new value (via the database URL helper template), so the two halves diverge: server presents the new password, postgres still expects the old one.
**Symptom.** The certctl-server pod's startup log shows:
```
failed to ping database: postgres rejected the configured credentials
(SQLSTATE 28P01 — invalid_password). If you recently rotated POSTGRES_PASSWORD ...
```
That diagnostic is emitted by `internal/repository/postgres/db.go::wrapPingError` — it points operators at the two remediation paths below.
**Remediation, non-destructive (preferred for any environment with real data):**
```bash
# 1. Rotate the password in postgres directly
kubectl -n certctl exec -it <release>-postgres-0 -- \
psql -U certctl -c "ALTER ROLE certctl PASSWORD '<new-password>';"
# 2. Update the secret / Helm values to the same value
helm upgrade <release> deploy/helm/certctl/ \
--reuse-values \
--set postgresql.auth.password='<new-password>'
# 3. Bounce the certctl-server pod so it re-reads the secret
kubectl -n certctl rollout restart deployment/<release>-server
```
**Remediation, destructive (DESTROYS ALL CERTCTL DATA — only acceptable on dev/demo clusters):**
```bash
helm uninstall <release> -n certctl
kubectl -n certctl delete pvc -l \
app.kubernetes.io/name=certctl,app.kubernetes.io/component=postgres
helm install <release> deploy/helm/certctl/ \
--namespace certctl \
--set postgresql.auth.password='<new-password>'
```
The PVC re-creates empty, `initdb` runs on first boot of the new postgres pod, and `pg_authid` is seeded with the new password.
**Why we don't fix this in the chart.** The env-vs-`pg_authid` divergence is intrinsic to how the upstream `postgres` image bootstraps — `initdb` is run-once-per-empty-data-dir, and there is no upstream-supported way to make subsequent boots re-seed `pg_authid` from `POSTGRES_PASSWORD`. The ergonomic answer is the runtime diagnostic plus this operational note.
**Cross-references.** Same root cause is documented for the docker-compose path in [`docs/quickstart.md`](../../../docs/quickstart.md) (Warning callout after the `cp .env.example .env` block) and in [`deploy/ENVIRONMENTS.md`](../../ENVIRONMENTS.md) (Stateful volume — first-boot password binding section). The runtime diagnostic itself lives in `internal/repository/postgres/db.go::wrapPingError` with regression coverage in `internal/repository/postgres/db_test.go`.
### Server API key rotation
Unlike the postgres password, `server.auth.apiKey` accepts a comma-separated list, so zero-downtime rotation is straightforward:
```bash
# 1. Add the new key alongside the old
helm upgrade <release> deploy/helm/certctl/ \
--reuse-values \
--set server.auth.apiKey='new-key,old-key'
# 2. Roll your agents / clients over to the new key
# 3. Remove the old key
helm upgrade <release> deploy/helm/certctl/ \
--reuse-values \
--set server.auth.apiKey='new-key'
```
### JWT / OIDC via authenticating gateway
certctl's in-process auth surface is intentionally narrow: `server.auth.type=api-key` for production deployments and `server.auth.type=none` for development. There is no in-process JWT, OIDC, mTLS, or SAML middleware. (`server.auth.type=jwt` was accepted pre-G-1 but silently routed every request through the api-key bearer middleware — silent auth downgrade. The chart now fails at `helm install`/`helm upgrade` template time via the `certctl.validateAuthType` helper if you set it. See [`../../../docs/upgrade-to-v2-jwt-removal.md`](../../../docs/upgrade-to-v2-jwt-removal.md) if you previously had this in your values.)
For deployments that need JWT/OIDC, the canonical Kubernetes-flavored shape is to put oauth2-proxy in front of the certctl Service, attach an authenticating Ingress middleware, and run certctl with `server.auth.type=none`:
```bash
# 1. Install oauth2-proxy (or any OIDC-terminating sidecar) in the same namespace
helm install oauth2-proxy oauth2-proxy/oauth2-proxy \
--namespace certctl \
--set config.clientID="$OIDC_CLIENT_ID" \
--set config.clientSecret="$OIDC_CLIENT_SECRET" \
--set config.cookieSecret="$(openssl rand -base64 32)" \
--set config.configFile='|
provider = "oidc"
oidc_issuer_url = "https://your-issuer/"
upstreams = ["http://<release>-server.certctl.svc.cluster.local:8443"]
pass_authorization_header = true
set_authorization_header = true
email_domains = ["*"]
'
# 2. Install certctl with type=none (gateway terminates auth)
helm install certctl deploy/helm/certctl/ \
--namespace certctl \
--set server.auth.type=none \
--set postgresql.auth.password="$(openssl rand -base64 24)"
# 3. Attach an Ingress that routes through oauth2-proxy
# (Traefik ForwardAuth, nginx auth_request, Envoy ext_authz, etc.)
```
Same root pattern works with Pomerium, Authelia, Caddy `forward_auth`, Apache `mod_auth_openidc`, or any service-mesh `ext_authz`. See [`../../../docs/architecture.md`](../../../docs/architecture.md) "Authenticating-gateway pattern" for the full design rationale and [`../../../docs/upgrade-to-v2-jwt-removal.md`](../../../docs/upgrade-to-v2-jwt-removal.md) for the migration walkthrough.
### TLS certificate sourcing
By default the chart provisions a self-signed cert via the same init-container pattern as the docker-compose deploy. For production, supply an operator-managed Secret (cert-manager, internal CA, etc.) — see [`docs/tls.md`](../../../docs/tls.md) for the full provisioning matrix and [`docs/upgrade-to-tls.md`](../../../docs/upgrade-to-tls.md) for upgrade-from-HTTP procedures.
## Disabling embedded postgres
If you have an existing PostgreSQL cluster, disable the embedded one and point at it directly:
```bash
helm install certctl deploy/helm/certctl/ \
--set postgresql.enabled=false \
--set server.databaseUrl='postgres://certctl:<pw>@my-pg-host:5432/certctl?sslmode=require'
```
The volume-trap section above does **not** apply to this configuration — your postgres operator (or cloud DB) handles password rotation, and you control `pg_authid` directly.
## Uninstall
```bash
helm uninstall <release> -n certctl
# Optional — also delete the postgres PVC (DESTROYS DATA):
kubectl -n certctl delete pvc -l \
app.kubernetes.io/name=certctl,app.kubernetes.io/component=postgres
```
By default `helm uninstall` retains the StatefulSet's PVCs, so reinstalling with the same release name preserves the database. If you've changed `postgresql.auth.password` in your values between uninstall and reinstall, you'll hit the trap on the reinstall — apply the non-destructive remediation above, or also delete the PVC.
@@ -169,3 +169,26 @@ per affected resource. No-op when configured correctly.
{{- fail "\n\nserver.tls.certManager.enabled=true but server.tls.certManager.issuerRef.name is empty.\n\nSet:\n --set server.tls.certManager.issuerRef.name=<your-issuer-or-clusterissuer>\n\nSee docs/tls.md.\n" -}}
{{- end -}}
{{- end }}
{{/*
Auth-type validation gate.
G-1 (P1): pre-G-1 the chart accepted server.auth.type=jwt and the
certctl-server container silently routed every request through the
api-key bearer middleware (no JWT impl ships with certctl). Post-G-1
the chart fails at template-time with a pointer at the authenticating-
gateway pattern. The valid set must stay in sync with
internal/config.ValidAuthTypes() in the Go binary; if you add a value
there you must add it here too (and update the property test in
internal/config/config_test.go that pins both surfaces).
Any template that consumes .Values.server.auth.type should call
`{{ include "certctl.validateAuthType" . }}` at the top so this guard
runs once per affected resource. No-op when configured correctly.
*/}}
{{- define "certctl.validateAuthType" -}}
{{- $valid := list "api-key" "none" -}}
{{- if not (has .Values.server.auth.type $valid) -}}
{{- fail (printf "\n\nserver.auth.type=%q is not supported (valid: %v).\n\nFor JWT/OIDC, run an authenticating gateway in front of certctl\n(oauth2-proxy / Envoy ext_authz / Traefik ForwardAuth / Pomerium) and\nset server.auth.type=none here so the gateway terminates federated\nidentity. See docs/architecture.md \"Authenticating-gateway pattern\"\nand docs/upgrade-to-v2-jwt-removal.md for the migration walkthrough.\n\nG-1 audit closure: pre-G-1 the chart accepted type=jwt and the binary\nsilently downgraded to api-key middleware. The chart now fails at\ntemplate time so misconfigured deployments cannot ship.\n" .Values.server.auth.type $valid) -}}
{{- end -}}
{{- end }}
@@ -1,3 +1,4 @@
{{- include "certctl.validateAuthType" . }}
apiVersion: v1
kind: ConfigMap
metadata:
@@ -1,4 +1,5 @@
{{- include "certctl.tls.required" . }}
{{- include "certctl.validateAuthType" . }}
apiVersion: apps/v1
kind: Deployment
metadata:
@@ -1,3 +1,4 @@
{{- include "certctl.validateAuthType" . }}
apiVersion: v1
kind: Secret
metadata:
+58 -6
View File
@@ -48,7 +48,14 @@ server:
drop:
- ALL
# Liveness and readiness probes (HTTPS-only as of v2.2)
# Liveness and readiness probes (HTTPS-only as of v2.2).
#
# The two paths exposed for probes are `/health` and `/ready` —
# registered in internal/api/router/router.go:76-85 and bypassing the
# auth middleware via the no-auth list at cmd/server/main.go:920.
# Both serve the same JSON shape today (`{"status":"healthy"}` /
# `{"status":"ready"}`) but exist as separate routes so liveness and
# readiness can diverge in the future without renaming.
livenessProbe:
httpGet:
path: /health
@@ -59,9 +66,18 @@ server:
timeoutSeconds: 5
failureThreshold: 3
# U-2 (P1, cat-u-healthcheck_protocol_mismatch — adjacent fix): pre-U-2
# the readiness probe pointed at `/readyz`, the conventional kube-flavor
# name. The certctl server doesn't register `/readyz` (only `/health`
# and `/ready`) — see cmd/server/main.go:920 and
# internal/api/router/router.go:81. K8s readiness probes therefore
# received a 404 (or, with auth enabled, a 401 from the api-key middleware
# because `/readyz` was NOT in the no-auth bypass set), pods stayed
# `NotReady` indefinitely, and Helm rollouts stalled. Post-U-2 the path
# matches a registered route.
readinessProbe:
httpGet:
path: /readyz
path: /ready
port: https
scheme: HTTPS
initialDelaySeconds: 5
@@ -112,10 +128,23 @@ server:
port: 8443
annotations: {}
# Authentication configuration
# Authentication configuration.
# Valid types: "api-key" (production) or "none" (demo only — disables
# authentication on the API and logs a loud Warn at server startup).
# For JWT/OIDC, run an authenticating gateway in front of certctl
# (oauth2-proxy / Envoy ext_authz / Traefik ForwardAuth / Pomerium)
# and set type=none here so the gateway terminates federated identity.
# See docs/architecture.md "Authenticating-gateway pattern".
#
# G-1 (P1): pre-G-1 the chart accepted server.auth.type=jwt and the
# certctl-server container silently routed every request through the
# api-key bearer middleware — silent auth downgrade. Post-G-1 the
# chart's `certctl.validateAuthType` template helper rejects any value
# outside {api-key, none} at template time. See
# docs/upgrade-to-v2-jwt-removal.md if you previously set type=jwt.
auth:
type: api-key # Options: api-key, none (for demo only)
apiKey: "" # REQUIRED in production - set via --set or values override
type: api-key
apiKey: "" # REQUIRED when type=api-key (set via --set or values override).
# Logging configuration
logging:
@@ -260,7 +289,30 @@ postgresql:
auth:
database: certctl
username: certctl
password: "" # REQUIRED - set via --set or values override
# REQUIRED set via `--set postgresql.auth.password=<value>` or values override.
#
# WARNING (U-1): rotating this value after first deploy does NOT change the
# database password. The `postgres:16-alpine` image runs `initdb` only when
# /var/lib/postgresql/data is empty, so POSTGRES_PASSWORD is written into
# pg_authid exactly once — on the first boot of the StatefulSet's PVC.
# Subsequent rollouts pick up the new env value in the postgres container
# but the certctl-server container's CERTCTL_DATABASE_URL also picks up
# the new value, while pg_authid still expects the old one — leading to
# `pq: password authentication failed for user "certctl"` (SQLSTATE 28P01).
#
# The certctl-server emits guidance via internal/repository/postgres/db.go::
# wrapPingError when it sees SQLSTATE 28P01 at startup. To resolve in a
# Helm deployment:
# - Non-destructive (preferred for environments with data):
# kubectl exec -it <release>-postgres-0 -- \
# psql -U certctl -c "ALTER ROLE certctl PASSWORD '<new>';"
# then update the secret/values to match and let the certctl-server
# pod restart against the matching credential.
# - Destructive (DESTROYS DATA — only acceptable on dev/demo PVCs):
# helm uninstall <release> && \
# kubectl delete pvc -l app.kubernetes.io/name=certctl,app.kubernetes.io/component=postgres && \
# helm install <release> ... # PVC re-creates empty, initdb seeds new password
password: ""
# Storage configuration
storage:
+216
View File
@@ -0,0 +1,216 @@
//go:build integration
// Package integration_test — image-level HEALTHCHECK contract.
//
// U-2 (P1, cat-u-healthcheck_protocol_mismatch): pre-U-2 the published
// server image's Dockerfile HEALTHCHECK called `curl -f http://localhost:
// 8443/health` against an HTTPS-only listener (HTTPS-Everywhere milestone,
// v2.2 / tag v2.0.47). Operators outside docker-compose / Helm saw the
// container reported as `unhealthy` indefinitely. The compose stack
// overrode this HEALTHCHECK with `--cacert + https://`; the Helm chart
// uses explicit `httpGet` probes that ignore Docker's HEALTHCHECK; the 5
// example compose files all override with `curl -sfk https://localhost:
// 8443/health`. So the observable failure was scoped to bare `docker run`
// / Docker Swarm / Nomad / ECS users — exactly the "I just pulled the
// published image" path.
//
// This file's tests pin the contract at the binary-image level. The
// matching CI grep guardrail in .github/workflows/ci.yml catches the
// regression at the Dockerfile-source level; both layers are needed
// because someone could replace the HEALTHCHECK line with a sibling
// broken pattern that the grep doesn't catch (e.g., a TCP-only check
// against the HTTPS port).
//
// Run alongside the rest of the integration suite:
//
// cd deploy/test && go test -tags integration -v -run Healthcheck
//
// The tests skip cleanly with t.Skip when docker is not available
// (CI without docker-in-docker, sandbox environments, etc.) so they
// don't block local development on machines without docker.
package integration_test
import (
"encoding/json"
"os/exec"
"strings"
"testing"
"time"
)
// dockerAvailable returns true when `docker version` returns 0.
// We cache it across tests in this file so the skip message prints once.
func dockerAvailable(t *testing.T) bool {
t.Helper()
cmd := exec.Command("docker", "version", "--format", "{{.Server.Version}}")
out, err := cmd.CombinedOutput()
if err != nil {
t.Logf("docker not available: %v\noutput: %s", err, string(out))
return false
}
return true
}
// dockerCmd runs `docker <args...>` with a 60s budget, returning stdout
// + stderr combined and the exit error if any. Used for short-lived
// probes (inspect, build, run -d).
func dockerCmd(t *testing.T, timeout time.Duration, args ...string) (string, error) {
t.Helper()
cmd := exec.Command("docker", args...)
done := make(chan struct{})
var out []byte
var err error
go func() {
out, err = cmd.CombinedOutput()
close(done)
}()
select {
case <-done:
return string(out), err
case <-time.After(timeout):
_ = cmd.Process.Kill()
t.Fatalf("docker %v timed out after %v", args, timeout)
return "", err
}
}
// TestPublishedServerImage_HealthcheckSpecUsesHTTPS performs the Dockerfile-
// source-level shipped-shape pin: the inspected image's Healthcheck.Test
// array MUST contain "https://localhost:8443/health" (and MUST NOT
// contain "http://localhost:8443/health"). This is the lightweight half
// of the contract — it doesn't require running the container, only
// building it. It catches the audit-flagged bug directly.
func TestPublishedServerImage_HealthcheckSpecUsesHTTPS(t *testing.T) {
if !dockerAvailable(t) {
t.Skip("docker not available — skipping image-level HEALTHCHECK test")
}
const imgTag = "certctl-u2-healthcheck-spec-test"
t.Cleanup(func() {
_, _ = dockerCmd(t, 30*time.Second, "rmi", "-f", imgTag)
})
// Build the server image. Use the repo root as context (this test
// file lives at deploy/test/, the Dockerfile at the repo root).
buildOut, err := dockerCmd(t, 5*time.Minute,
"build", "-f", "../../Dockerfile", "-t", imgTag, "../..")
if err != nil {
t.Fatalf("docker build failed: %v\noutput:\n%s", err, buildOut)
}
// Inspect the shipped HEALTHCHECK metadata.
inspectOut, err := dockerCmd(t, 30*time.Second,
"inspect", "--format", "{{json .Config.Healthcheck}}", imgTag)
if err != nil {
t.Fatalf("docker inspect failed: %v\noutput:\n%s", err, inspectOut)
}
var hc struct {
Test []string
Interval int64
Timeout int64
}
if err := json.Unmarshal([]byte(strings.TrimSpace(inspectOut)), &hc); err != nil {
t.Fatalf("could not parse Healthcheck JSON %q: %v", inspectOut, err)
}
joined := strings.Join(hc.Test, " ")
// Positive contract.
if !strings.Contains(joined, "https://localhost:8443/health") {
t.Errorf("Healthcheck.Test does not target https://localhost:8443/health\nfull: %v", hc.Test)
}
// Negative contract — pre-U-2 regression shape MUST be absent.
if strings.Contains(joined, "http://localhost:8443/health") {
t.Errorf("Healthcheck.Test still contains the pre-U-2 plaintext shape: %v", hc.Test)
}
// `-k` (or `--insecure`) must be present because the bootstrap cert
// is per-deploy and the published image can't pin a CA bundle —
// see the U-2 closure docblock on Dockerfile and the audit doc.
if !strings.Contains(joined, "-k") && !strings.Contains(joined, "--insecure") {
t.Errorf("Healthcheck.Test omits -k / --insecure flag (required for self-signed bootstrap probe): %v", hc.Test)
}
}
// TestPublishedAgentImage_HealthcheckSpecExists pins the U-2 adjacent
// fix that added a HEALTHCHECK to the agent image. Pre-U-2 the agent
// image had no HEALTHCHECK declaration, so bare-`docker run` agents got
// `none` health status from Docker. Post-U-2 the agent uses pgrep to
// verify the process is alive (mirroring the docker-compose pattern at
// deploy/docker-compose.yml:173, which also became reliable post-U-2
// because procps is now installed in the runtime image).
func TestPublishedAgentImage_HealthcheckSpecExists(t *testing.T) {
if !dockerAvailable(t) {
t.Skip("docker not available — skipping image-level HEALTHCHECK test")
}
const imgTag = "certctl-u2-agent-healthcheck-spec-test"
t.Cleanup(func() {
_, _ = dockerCmd(t, 30*time.Second, "rmi", "-f", imgTag)
})
buildOut, err := dockerCmd(t, 5*time.Minute,
"build", "-f", "../../Dockerfile.agent", "-t", imgTag, "../..")
if err != nil {
t.Fatalf("docker build failed: %v\noutput:\n%s", err, buildOut)
}
inspectOut, err := dockerCmd(t, 30*time.Second,
"inspect", "--format", "{{json .Config.Healthcheck}}", imgTag)
if err != nil {
t.Fatalf("docker inspect failed: %v\noutput:\n%s", err, inspectOut)
}
trimmed := strings.TrimSpace(inspectOut)
if trimmed == "null" || trimmed == "" {
t.Fatalf("agent image has no HEALTHCHECK (got %q) — U-2 adjacent fix regressed", inspectOut)
}
var hc struct {
Test []string
}
if err := json.Unmarshal([]byte(trimmed), &hc); err != nil {
t.Fatalf("could not parse Healthcheck JSON %q: %v", inspectOut, err)
}
joined := strings.Join(hc.Test, " ")
if !strings.Contains(joined, "pgrep") {
t.Errorf("agent Healthcheck.Test does not use pgrep (lost the process-presence shape): %v", hc.Test)
}
if !strings.Contains(joined, "certctl-agent") {
t.Errorf("agent Healthcheck.Test does not target the certctl-agent process name: %v", hc.Test)
}
}
// TestPublishedServerImage_HealthcheckTransitionsToHealthy is the
// runtime-level contract: the built image, when started, must transition
// to `healthy` within the start-period + 30s observability budget. This
// is the heavy test — it requires the server to actually start, which
// in turn requires either a reachable database OR a startup that fails
// gracefully enough to keep the HEALTHCHECK probe target alive.
//
// The container is started with CERTCTL_DATABASE_URL pointing at an
// unreachable host so the server fails its postgres bring-up — but
// importantly, fails AFTER the TLS listener has come up, because the
// HEALTHCHECK probe target is the TLS listener. We don't actually need
// the database to validate the HEALTHCHECK shape.
//
// IMPORTANT: this test is the runtime contract. If you're working on the
// server's startup ordering and the listener now comes up AFTER the
// database, this test must adapt — start a sidecar postgres via
// testcontainers-go (see internal/integration/lifecycle_test.go for the
// pattern) and connect the certctl-server container to it.
func TestPublishedServerImage_HealthcheckTransitionsToHealthy(t *testing.T) {
if !dockerAvailable(t) {
t.Skip("docker not available — skipping runtime HEALTHCHECK test")
}
if testing.Short() {
t.Skip("runtime HEALTHCHECK test takes ~45s; skipping under -short")
}
t.Skip("runtime probe contract not yet wired to a sidecar postgres; " +
"image-spec contract above (TestPublishedServerImage_HealthcheckSpecUsesHTTPS) " +
"covers the audit-flagged regression. Re-enable once the integration " +
"harness provisions postgres for image-level smoke.")
}
+12 -1
View File
@@ -348,6 +348,11 @@ erDiagram
}
```
The ER diagram above documents **database shape**, not REST-API wire shape. Several columns are intentionally server-internal and never serialized to clients:
- `agents.api_key_hash` — SHA-256 of the agent's plaintext API key, populated by `service.RegisterAgent` (`hashAPIKey(apiKey)` at `internal/service/agent.go`) and consumed by `repository.AgentRepository::GetByAPIKey` for the auth-lookup. **Not** exposed via the REST API, **not** echoed via CLI / MCP / agent registration response, **never** logged. Enforced by `internal/domain/connector.go::Agent.MarshalJSON` (G-2 audit closure, `cat-s5-apikey_leak`); the OpenAPI Agent schema explicitly excludes the field, the frontend `Agent` interface omits it, and a CI grep guardrail at `.github/workflows/ci.yml` blocks reintroduction.
- `issuers.config` / `deployment_targets.config` — plaintext jsonb shadow of the AES-GCM-encrypted on-disk blob; the encrypted form lives on `EncryptedConfig []byte` (Go-only field tagged `json:"-"`).
Migrations are idempotent (`IF NOT EXISTS` on all CREATE statements, `ON CONFLICT (id) DO NOTHING` on all seed data) so they're safe to run multiple times — important for Docker Compose where both initdb and the server may run the same SQL.
## Data Flow: Certificate Lifecycle
@@ -891,9 +896,15 @@ The HTTP middleware stack processes requests in the following order (see `cmd/se
4. **BodyLimit** - request body size cap via `http.MaxBytesReader`
5. **RateLimiter** - token bucket rate limiting (optional, when enabled)
6. **CORS** - cross-origin request handling (deny-by-default)
7. **Auth** - API key or JWT validation
7. **Auth** - API key validation (or none in development; JWT/OIDC via authenticating gateway, see below — not in-process)
8. **AuditLog** - records every API call to the audit trail (requires auth context for actor)
### Authenticating-gateway pattern (JWT, OIDC, mTLS)
certctl's in-process authentication surface is intentionally narrow: `api-key` for production deployments and `none` for development. There is no in-process JWT, OIDC, mTLS, or SAML middleware. (`CERTCTL_AUTH_TYPE=jwt` was accepted pre-G-1 but silently routed through the api-key bearer middleware — a security finding masquerading as a config option, removed at the v2.x boundary; see [`upgrade-to-v2-jwt-removal.md`](upgrade-to-v2-jwt-removal.md) if you previously set it.)
For deployments that need JWT/OIDC/mTLS, the standard pattern is to put an authenticating gateway in front of certctl and configure `CERTCTL_AUTH_TYPE=none` on the upstream certctl process. The gateway terminates the federated identity protocol, validates tokens / certificates / SAML assertions, and proxies the authenticated request to certctl as a same-origin call on a private network. This separation gives operators the full breadth of the modern identity ecosystem (oauth2-proxy, Envoy `ext_authz`, Traefik `ForwardAuth`, Pomerium, Authelia, Caddy `forward_auth`, Apache `mod_auth_openidc`, nginx `auth_request`) without certctl itself having to track signing-key rotation, claim mapping, audience validation, and the rest of the JWT/OIDC surface area. Operators wanting per-request actor attribution past the gateway boundary forward the gateway-resolved identity (e.g., `X-Auth-Request-User` from oauth2-proxy) and run a small authorization layer at the gateway that enforces the bearer-key contract certctl actually uses.
### Concurrency Safety
The background scheduler uses `sync/atomic.Bool` idempotency guards on every loop (8 always-on plus up to 4 optional) — if a tick fires while the previous iteration is still running, it skips. A `sync.WaitGroup` tracks all in-flight goroutines. `WaitForCompletion(timeout)` blocks during shutdown until all work finishes or the timeout expires, preventing state corruption from mid-flight database operations during process exit.
+32 -15
View File
@@ -1141,13 +1141,30 @@ API Endpoints:
- **`GET /api/v1/digest/preview`** — Render digest HTML for preview (no email sent)
- **`POST /api/v1/digest/send`** — Trigger digest send immediately (outside of schedule)
> **Note (HTTPS-only as of v2.2):** The `curl` examples in this section
> and below all target the HTTPS-only control plane. Extract the
> docker-compose self-signed bootstrap CA bundle once and reuse it on
> every call:
>
> ```bash
> export CA=/tmp/certctl-ca.crt
> docker compose -f deploy/docker-compose.yml exec -T certctl-server \
> cat /etc/certctl/tls/ca.crt > "$CA"
> ```
>
> Then pass `--cacert "$CA"` (or `-k` for one-off smoke tests, never in
> production). The same pattern is documented in
> [`quickstart.md`](quickstart.md). Pre-U-2 these examples used `http://`
> and silently failed against the HTTPS listener; post-U-2 they speak
> HTTPS with the operator-managed CA bundle.
Example:
```bash
# Preview digest
curl http://localhost:8443/api/v1/digest/preview | jq '.html'
curl --cacert "$CA" https://localhost:8443/api/v1/digest/preview | jq '.html'
# Send digest immediately
curl -X POST http://localhost:8443/api/v1/digest/send
curl --cacert "$CA" -X POST https://localhost:8443/api/v1/digest/send
```
Each notifier is enabled by its configuration env var:
@@ -1294,24 +1311,24 @@ The agent scans these directories on startup and every 6 hours, looking for cert
```bash
# List discovered certificates (filter by agent, status)
curl -s "http://localhost:8443/api/v1/discovered-certificates?agent_id=agent-nginx-01&status=new" | jq .
curl --cacert "$CA" -s "https://localhost:8443/api/v1/discovered-certificates?agent_id=agent-nginx-01&status=new" | jq .
# Get discovery detail
curl -s http://localhost:8443/api/v1/discovered-certificates/DISCOVERY_ID | jq .
curl --cacert "$CA" -s https://localhost:8443/api/v1/discovered-certificates/DISCOVERY_ID | jq .
# Claim a discovered cert (link to managed certificate)
curl -s -X POST http://localhost:8443/api/v1/discovered-certificates/DISCOVERY_ID/claim \
curl --cacert "$CA" -s -X POST https://localhost:8443/api/v1/discovered-certificates/DISCOVERY_ID/claim \
-H "Content-Type: application/json" \
-d '{"managed_certificate_id": "mc-api-prod"}' | jq .
# Dismiss a discovery
curl -s -X POST http://localhost:8443/api/v1/discovered-certificates/DISCOVERY_ID/dismiss | jq .
curl --cacert "$CA" -s -X POST https://localhost:8443/api/v1/discovered-certificates/DISCOVERY_ID/dismiss | jq .
# View discovery scan history
curl -s http://localhost:8443/api/v1/discovery-scans | jq .
curl --cacert "$CA" -s https://localhost:8443/api/v1/discovery-scans | jq .
# Summary counts (new, claimed, dismissed)
curl -s http://localhost:8443/api/v1/discovery-summary | jq .
curl --cacert "$CA" -s https://localhost:8443/api/v1/discovery-summary | jq .
```
### Use Cases
@@ -1340,7 +1357,7 @@ Network scan targets can be managed from the **Network Scans** dashboard page (c
```bash
# Create a scan target for your internal network (or use the dashboard's "+ New Target" button)
curl -s -X POST http://localhost:8443/api/v1/network-scan-targets \
curl --cacert "$CA" -s -X POST https://localhost:8443/api/v1/network-scan-targets \
-H "Content-Type: application/json" \
-d '{
"name": "Production Web Servers",
@@ -1365,26 +1382,26 @@ curl -s -X POST http://localhost:8443/api/v1/network-scan-targets \
```bash
# List all scan targets
curl -s http://localhost:8443/api/v1/network-scan-targets | jq .
curl --cacert "$CA" -s https://localhost:8443/api/v1/network-scan-targets | jq .
# Create a scan target
curl -s -X POST http://localhost:8443/api/v1/network-scan-targets \
curl --cacert "$CA" -s -X POST https://localhost:8443/api/v1/network-scan-targets \
-H "Content-Type: application/json" \
-d '{"name": "DMZ", "cidrs": ["172.16.0.0/24"], "ports": [443]}' | jq .
# Get a specific target (includes last_scan_at, last_scan_certs_found)
curl -s http://localhost:8443/api/v1/network-scan-targets/nst-dmz | jq .
curl --cacert "$CA" -s https://localhost:8443/api/v1/network-scan-targets/nst-dmz | jq .
# Trigger an immediate scan (doesn't wait for scheduler)
curl -s -X POST http://localhost:8443/api/v1/network-scan-targets/nst-dmz/scan | jq .
curl --cacert "$CA" -s -X POST https://localhost:8443/api/v1/network-scan-targets/nst-dmz/scan | jq .
# Update scan configuration
curl -s -X PUT http://localhost:8443/api/v1/network-scan-targets/nst-dmz \
curl --cacert "$CA" -s -X PUT https://localhost:8443/api/v1/network-scan-targets/nst-dmz \
-H "Content-Type: application/json" \
-d '{"ports": [443, 8443, 9443], "timeout_ms": 3000}' | jq .
# Delete a scan target
curl -s -X DELETE http://localhost:8443/api/v1/network-scan-targets/nst-dmz
curl --cacert "$CA" -s -X DELETE https://localhost:8443/api/v1/network-scan-targets/nst-dmz
```
### Scheduler Integration
+2
View File
@@ -60,6 +60,8 @@ cp deploy/.env.example deploy/.env
docker compose -f deploy/docker-compose.yml up -d --build
```
> **Warning:** Edit `POSTGRES_PASSWORD` *before* the very first `docker compose up`. Postgres seeds the password into its data directory only on first boot of an empty volume — after that, the password is baked into `pg_authid` and the env var is ignored. If you boot once with the default and later change `POSTGRES_PASSWORD` in `.env`, the certctl-server container picks up the new value but postgres still authenticates against the old one, and the server logs `pq: password authentication failed for user "certctl"` (SQLSTATE 28P01). Two ways out: tear down the volume with `docker compose -f deploy/docker-compose.yml down -v` (this **deletes all data**) and bring up fresh, or rotate non-destructively with `docker compose -f deploy/docker-compose.yml exec postgres psql -U certctl -c "ALTER ROLE certctl PASSWORD '<new>';"` and then restart certctl-server with the matching `POSTGRES_PASSWORD`.
### Docker Compose Environments
The `deploy/` directory contains four compose files for different use cases:
+155
View File
@@ -0,0 +1,155 @@
# Upgrading past G-1 — `CERTCTL_AUTH_TYPE=jwt` removal
If your certctl deployment currently sets `CERTCTL_AUTH_TYPE=jwt` (or `server.auth.type=jwt` in Helm), the next certctl upgrade will fail-fast at startup with a dedicated diagnostic. This guide explains why, what to switch to, and how to keep JWT/OIDC at your edge.
For everyone else — operators running `api-key` or `none` — this upgrade is a no-op. Skip to [`upgrade-to-tls.md`](upgrade-to-tls.md) for the v2.2 HTTPS-everywhere migration if you haven't done that one yet.
## Why we removed it
Pre-G-1, the config validator at `internal/config/config.go` accepted three values for `CERTCTL_AUTH_TYPE`: `api-key`, `jwt`, and `none`. The startup log line at `cmd/server/main.go` faithfully echoed `"authentication enabled" "type"="jwt"` when an operator picked `jwt`. Reasonable people read that and concluded JWT auth was on.
It wasn't. Grep `internal/ cmd/` for `NewJWT`, `JWTMiddleware`, or `jwt.Parse` — pre-G-1, there were zero matches in production code. The auth-middleware wiring at `cmd/server/main.go:653` unconditionally called `middleware.NewAuthWithNamedKeys(namedKeys)` regardless of `cfg.Auth.Type`. So `CERTCTL_AUTH_TYPE=jwt` just routed every request through the api-key bearer middleware, comparing the incoming `Authorization: Bearer <something>` against whatever string the operator put in `CERTCTL_AUTH_SECRET`. Real JWT clients got 401 (the api-key middleware saw the JWT string as a literal token and compared bytes). Operators who treated `CERTCTL_AUTH_SECRET` as a JWT signing secret (and therefore handled it less carefully than an api-key) handed an attacker an api-key. Silent auth downgrade — a security finding masquerading as a config option.
We chose to remove the option rather than implement JWT middleware. Implementing real JWT/OIDC requires jwks vs static-secret rotation, claim mapping (which claim is the actor / the admin flag?), expiry enforcement, audience and issuer validation, key rollover semantics, and regression coverage at the same depth as the existing api-key path. That's a feature, not a fix. The audit-recommended structural fix — and the one that actually closes the hazard — is to fail loudly instead of silently downgrading.
## What changes at startup
Post-G-1, a binary started with `CERTCTL_AUTH_TYPE=jwt` exits non-zero before opening the listener:
```
Failed to load configuration: CERTCTL_AUTH_TYPE=jwt is no longer accepted
(G-1 silent auth downgrade): no JWT middleware ships with certctl. To use
JWT/OIDC, run an authenticating gateway (oauth2-proxy / Envoy ext_authz /
Traefik ForwardAuth / Pomerium) in front of certctl and set
CERTCTL_AUTH_TYPE=none on the upstream. See docs/architecture.md
"Authenticating-gateway pattern" and docs/upgrade-to-v2-jwt-removal.md
for the migration walkthrough
```
Helm operators get the same shape at `helm install` / `helm upgrade` template time: `server.auth.type=jwt` is rejected by the chart's `certctl.validateAuthType` template helper before any Kubernetes object is rendered.
The CI-side regression guard at `.github/workflows/ci.yml` blocks any future PR that re-introduces `"jwt"` as an auth-type literal in production code or spec.
## Recovery — pick one
### Option A — switch to `api-key` (you weren't actually using JWT)
If your `CERTCTL_AUTH_SECRET` was a single high-entropy token and your clients sent it as `Authorization: Bearer <token>`, you were already using api-key auth — you just had `CERTCTL_AUTH_TYPE` set to the wrong string. Flip it:
```
# .env (docker-compose)
CERTCTL_AUTH_TYPE=api-key
CERTCTL_AUTH_SECRET=<your-existing-token>
```
```
# Helm
helm upgrade <release> deploy/helm/certctl/ \
--reuse-values \
--set server.auth.type=api-key \
--set server.auth.apiKey=<your-existing-token>
```
No client changes needed — the same Bearer token continues to work. The startup log will now read `"authentication enabled" "type"="api-key"`, which matches what was actually happening pre-G-1.
### Option B — front certctl with an authenticating gateway
If you genuinely need JWT, OIDC, mTLS, or SAML, run an authenticating gateway in front of certctl and let the gateway terminate the federated identity protocol. Configure certctl for `CERTCTL_AUTH_TYPE=none`:
```
CERTCTL_AUTH_TYPE=none
```
Then put an oauth2-proxy / Envoy `ext_authz` / Traefik `ForwardAuth` / Pomerium / Authelia (etc.) in the network path between operators and certctl. The gateway validates the identity and proxies the authenticated request to certctl as a same-origin call on a private network.
### Concrete walkthrough — oauth2-proxy + certctl on docker-compose
This is the simplest production-grade JWT/OIDC shape. It assumes you have an OIDC provider (Okta, Auth0, Google Workspace, Keycloak, Dex) and a registered client_id / client_secret.
```yaml
# deploy/docker-compose.gateway.yml — overlay on the base compose file
services:
oauth2-proxy:
image: quay.io/oauth2-proxy/oauth2-proxy:latest
command:
- --provider=oidc
- --oidc-issuer-url=https://<your-issuer>/
- --client-id=${OIDC_CLIENT_ID}
- --client-secret=${OIDC_CLIENT_SECRET}
- --cookie-secret=${OAUTH2_PROXY_COOKIE_SECRET} # openssl rand -base64 32
- --upstream=http://certctl-server:8443 # internal-network only; certctl listens on 8443
- --http-address=0.0.0.0:4180
- --email-domain=*
- --pass-access-token=true
- --pass-authorization-header=true
- --set-authorization-header=true # forwards a bearer token upstream
- --skip-provider-button=true
- --reverse-proxy=true
ports:
- "443:4180"
depends_on:
- certctl-server
networks:
- certctl-network
certctl-server:
environment:
CERTCTL_AUTH_TYPE: none # gateway terminates auth — see docs/upgrade-to-v2-jwt-removal.md
# ... rest of the certctl env block unchanged
```
Operators hit `https://<your-host>/`, get redirected through the OIDC provider, land back at oauth2-proxy with a session cookie, and oauth2-proxy proxies their request to certctl on the internal Docker network. certctl itself is HTTPS-only on `:8443` (TLS 1.3, see [`tls.md`](tls.md)) but operator browsers never see that hop directly. Bind certctl-server's `:8443` to the internal Docker network only — do NOT publish it to the host. The audit trail will record the actor as the gateway-forwarded identity if you also configure a small bearer-token-mapping shim at the gateway (most production deployments do this with a per-user api-key issued by the gateway after OIDC validation).
### Traefik ForwardAuth pattern (Kubernetes)
Same shape, kubernetes-flavored:
```yaml
apiVersion: traefik.io/v1alpha1
kind: Middleware
metadata:
name: oidc-forward-auth
spec:
forwardAuth:
address: http://oauth2-proxy.auth.svc.cluster.local:4180
trustForwardHeader: true
authResponseHeaders:
- X-Auth-Request-User
- X-Auth-Request-Email
- Authorization
---
apiVersion: traefik.io/v1alpha1
kind: IngressRoute
metadata:
name: certctl
spec:
routes:
- match: Host(`certctl.example.com`)
kind: Rule
middlewares:
- name: oidc-forward-auth
services:
- name: certctl-server
port: 8443
```
The certctl Helm release runs with `server.auth.type=none`. The Traefik IngressRoute attaches `oidc-forward-auth` as a middleware so every request is OIDC-validated by oauth2-proxy before reaching certctl.
### Envoy `ext_authz` pattern
For service-mesh deployments (Istio, Consul, plain Envoy), the `ext_authz` filter calls out to an external authorization service per-request. Same outcome: certctl runs `CERTCTL_AUTH_TYPE=none` and Envoy + your authz service handle JWT/OIDC/mTLS at the mesh edge. See the [Envoy ext_authz docs](https://www.envoyproxy.io/docs/envoy/latest/configuration/http/http_filters/ext_authz_filter) for the configuration surface.
## Rollback
Pre-G-1 binaries silently accepted `CERTCTL_AUTH_TYPE=jwt` and routed through the api-key middleware. Downgrading the binary is the only mechanical rollback path, and it puts you back into the silent-downgrade state — which is exactly what the G-1 audit finding is about. We don't recommend it. If something is forcing your hand, capture the operational issue you're hitting and open a GitHub issue against the certctl repo with the SHAs involved; the Authenticating-gateway pattern was specifically designed to cover the use cases that historically led operators to set `CERTCTL_AUTH_TYPE=jwt`.
There is no on-disk state that changes with this upgrade — no migrations to roll back, no encrypted config to re-encode, no certificates to re-issue. The change is entirely in the config-validation surface and the helm-chart template guard.
## Cross-references
- [`architecture.md`](architecture.md) — "Authenticating-gateway pattern (JWT, OIDC, mTLS)" section.
- [`tls.md`](tls.md) — TLS provisioning patterns. The gateway proxying to certctl-server still needs to trust certctl's TLS cert; same patterns apply.
- [`../deploy/helm/certctl/README.md`](../deploy/helm/certctl/README.md) — Helm-chart-flavored guidance.
- `internal/config/config.go::ValidAuthTypes` — the single source of truth for what's accepted post-G-1.
- `internal/repository/postgres/db.go::wrapPingError` — unrelated; pattern for runtime diagnostic of operator misconfiguration.
- `coverage-gap-audit-2026-04-24-v5/unified-audit.md` — the audit finding (`cat-g-jwt_silent_auth_downgrade`).
+55
View File
@@ -0,0 +1,55 @@
# Deployment Examples
Five turnkey docker-compose scenarios that show certctl deployed against real CA backends and target shapes. Each subdirectory is self-contained — pick the one closest to your stack and have it running in minutes.
| Example | Stack | What it shows |
|---------|-------|---------------|
| [`acme-nginx/`](acme-nginx/acme-nginx.md) | Let's Encrypt + NGINX (HTTP-01) | The default public-CA path: ACME-issued certs deployed to NGINX. |
| [`acme-wildcard-dns01/`](acme-wildcard-dns01/acme-wildcard-dns01.md) | Let's Encrypt wildcard (DNS-01) | Wildcard certificates via DNS-01 with pluggable DNS hooks. |
| [`private-ca-traefik/`](private-ca-traefik/private-ca-traefik.md) | Local CA + Traefik | Internal-only certs from a private CA, deployed to Traefik. |
| [`step-ca-haproxy/`](step-ca-haproxy/step-ca-haproxy.md) | Smallstep step-ca + HAProxy | Self-hosted CA with HAProxy as the deployment target. |
| [`multi-issuer/`](multi-issuer/multi-issuer.md) | Let's Encrypt + Local CA | Public + private certs side-by-side from a single dashboard. |
## Common operational notes
These notes apply to **every** example. They're called out here so the per-example walkthroughs stay focused on the issuer/target wiring instead of repeating ops boilerplate.
### Postgres password rotation — first-boot binding trap (U-1)
Every example file uses `${DB_PASSWORD:-certctl-dev-password}` as the postgres password env var, with the data directory persisted via a named volume. The `postgres:16-alpine` image runs `initdb` exactly once — when `/var/lib/postgresql/data` is empty — and that's the only time `POSTGRES_PASSWORD` is written into `pg_authid`. If you boot once with the default and then change `DB_PASSWORD` (in your shell, in a `.env` file, or in a wrapper script), the certctl-server container picks up the new value but the postgres container continues to authenticate against the old one. The server fails its startup `db.Ping()` with `pq: password authentication failed for user "certctl"` (SQLSTATE 28P01).
The certctl-server emits guidance pointing at the fix when this fires (see `internal/repository/postgres/db.go::wrapPingError`). The two remediation paths:
- **Destructive — wipes all certctl data, only acceptable on demo/test setups:**
```bash
docker compose -f examples/<example>/docker-compose.yml down -v
docker compose -f examples/<example>/docker-compose.yml up -d --build
```
- **Non-destructive — preserves data, rotates `pg_authid` in place:**
```bash
docker compose -f examples/<example>/docker-compose.yml exec postgres \
psql -U certctl -c "ALTER ROLE certctl PASSWORD '<new>';"
# Then redeploy with DB_PASSWORD set to <new> in your shell or .env
```
The cleanest practice for a fresh demo: set `DB_PASSWORD` once in your shell **before** the very first `docker compose up`, and don't change it during the demo's lifetime. If you must rotate, use the non-destructive path.
Same root cause and remediation pattern is documented for the canonical quickstart in [`../docs/quickstart.md`](../docs/quickstart.md), the production compose surface in [`../deploy/ENVIRONMENTS.md`](../deploy/ENVIRONMENTS.md), and the Helm chart in [`../deploy/helm/certctl/README.md`](../deploy/helm/certctl/README.md).
### TLS for the certctl control plane
Every example boots certctl with HTTPS-only on port 8443 (TLS 1.3 pinned, no plaintext listener as of v2.2). The shipped `certctl-tls-init` init container generates a self-signed ECDSA-P256 cert on first boot — fine for the example demos, **never** acceptable for a public deployment. For production, swap the init container for cert-manager, an operator-supplied Secret, or your internal CA — see [`../docs/tls.md`](../docs/tls.md) for the full pattern matrix.
### Tearing down
To stop services but **keep** the postgres volume (so you can pick up where you left off):
```bash
docker compose -f examples/<example>/docker-compose.yml down
```
To stop services **and** wipe all data (clean slate for the next run):
```bash
docker compose -f examples/<example>/docker-compose.yml down -v
```
Note that `down -v` is the only canonical way to recover from the postgres-password trap when the non-destructive `ALTER ROLE` route is unavailable (e.g., you've forgotten the original password).
+2
View File
@@ -2,6 +2,8 @@
This example demonstrates certctl's core use case: **automatically manage TLS certificates for NGINX using Let's Encrypt (ACME HTTP-01 challenges).**
> **Operational notes** shared by every example (postgres password rotation trap, TLS provisioning, teardown semantics) live in [`../README.md`](../README.md). Read it first if you plan to change `DB_PASSWORD` after the initial `docker compose up` — the postgres volume binds the password on first boot only.
## What This Does
- Deploys certctl server (control plane) with PostgreSQL
@@ -2,6 +2,8 @@
**What this does:** Issues wildcard certificates (e.g., `*.example.com`) from Let's Encrypt using DNS-01 challenge validation.
> **Operational notes** shared by every example (postgres password rotation trap, TLS provisioning, teardown semantics) live in [`../README.md`](../README.md). Read it first if you plan to change `DB_PASSWORD` after the initial `docker compose up` — the postgres volume binds the password on first boot only.
This example is ideal for:
- Issuing wildcard certificates (`*.example.com`)
- Services behind NAT, firewalls, or non-public networks
+2
View File
@@ -2,6 +2,8 @@
This example demonstrates certctl managing **both public and internal certificates from a single dashboard**. Public-facing services use Let's Encrypt (ACME), while internal services use a private Local CA — all visible and managed in one place.
> **Operational notes** shared by every example (postgres password rotation trap, TLS provisioning, teardown semantics) live in [`../README.md`](../README.md). Read it first if you plan to change `DB_PASSWORD` after the initial `docker compose up` — the postgres volume binds the password on first boot only.
## The Use Case
You have:
@@ -1,5 +1,7 @@
# Private CA + Traefik Example
> **Operational notes** shared by every example (postgres password rotation trap, TLS provisioning, teardown semantics) live in [`../README.md`](../README.md). Read it first if you plan to change `DB_PASSWORD` after the initial `docker compose up` — the postgres volume binds the password on first boot only.
This example demonstrates certctl managing certificates for **internal services without public CA dependency**. Ideal for enterprise environments where:
- All services are internal (VPN, private networks)
@@ -2,6 +2,8 @@
This example demonstrates certctl managing certificates issued by **Smallstep step-ca** and deploying them to **HAProxy**.
> **Operational notes** shared by every example (postgres password rotation trap, TLS provisioning, teardown semantics) live in [`../README.md`](../README.md). Read it first if you plan to change `DB_PASSWORD` after the initial `docker compose up` — the postgres volume binds the password on first boot only.
## Scenario
You're a Smallstep user running step-ca as your internal PKI. You have HAProxy load balancers that need certificates. This setup:
+158
View File
@@ -893,3 +893,161 @@ func TestAgentReportJobStatus_ServiceError(t *testing.T) {
func stringPtr(s string) *string {
return &s
}
// G-2 (P1): cat-s5-apikey_leak audit closure tests. Pre-G-2,
// Agent.APIKeyHash was tagged `json:"api_key_hash"` and shipped on
// every wire surface that returned domain.Agent. Post-G-2 the tag is
// "-" and Agent.MarshalJSON enforces redaction via a marshal-time copy
// (see internal/domain/connector_test.go for the type-level pin). These
// four tests are the wire-shape contract — they capture the actual HTTP
// response body via httptest and assert the credential-derivative hash
// is absent.
//
// One sentinel value (g2HandlerLeakSentinel) flows through every fixture
// so a single grep over a failing test's output identifies the leak
// surface immediately.
const g2HandlerLeakSentinel = "sha256:LEAKED-CREDENTIAL-DERIVATIVE-HANDLER-SENTINEL"
func TestListAgents_DoesNotLeakAPIKeyHash(t *testing.T) {
now := time.Now()
mock := &MockAgentService{
ListAgentsFn: func(page, perPage int) ([]domain.Agent, int64, error) {
return []domain.Agent{
{ID: "a-1", Name: "agent-one", Hostname: "host-1",
Status: domain.AgentStatusOnline, RegisteredAt: now,
APIKeyHash: g2HandlerLeakSentinel + "-1"},
{ID: "a-2", Name: "agent-two", Hostname: "host-2",
Status: domain.AgentStatusOnline, RegisteredAt: now,
APIKeyHash: g2HandlerLeakSentinel + "-2"},
}, 2, nil
},
}
h := NewAgentHandler(mock)
req := httptest.NewRequest(http.MethodGet, "/api/v1/agents?page=1&per_page=50", nil)
req = req.WithContext(contextWithRequestID())
w := httptest.NewRecorder()
h.ListAgents(w, req)
if w.Code != http.StatusOK {
t.Fatalf("ListAgents status = %d, want 200", w.Code)
}
body := w.Body.String()
if bytes.Contains([]byte(body), []byte("api_key_hash")) {
t.Errorf("ListAgents response leaked \"api_key_hash\" key (G-2 regressed):\n%s", body)
}
if bytes.Contains([]byte(body), []byte(g2HandlerLeakSentinel)) {
t.Errorf("ListAgents response leaked sentinel %q:\n%s", g2HandlerLeakSentinel, body)
}
// Sanity: the non-leaked fields ARE present (handler did serve real data).
for _, want := range []string{"a-1", "a-2", "agent-one", "agent-two"} {
if !bytes.Contains([]byte(body), []byte(want)) {
t.Errorf("ListAgents response missing expected field %q (handler may not be serving data):\n%s", want, body)
}
}
}
func TestGetAgent_DoesNotLeakAPIKeyHash(t *testing.T) {
now := time.Now()
mock := &MockAgentService{
GetAgentFn: func(id string) (*domain.Agent, error) {
return &domain.Agent{
ID: id, Name: "single-agent", Hostname: "single.host",
Status: domain.AgentStatusOnline, RegisteredAt: now,
APIKeyHash: g2HandlerLeakSentinel,
}, nil
},
}
h := NewAgentHandler(mock)
req := httptest.NewRequest(http.MethodGet, "/api/v1/agents/a-prod-001", nil)
req = req.WithContext(contextWithRequestID())
w := httptest.NewRecorder()
h.GetAgent(w, req)
if w.Code != http.StatusOK {
t.Fatalf("GetAgent status = %d, want 200, body=%s", w.Code, w.Body.String())
}
body := w.Body.String()
if bytes.Contains([]byte(body), []byte("api_key_hash")) {
t.Errorf("GetAgent response leaked \"api_key_hash\" key:\n%s", body)
}
if bytes.Contains([]byte(body), []byte(g2HandlerLeakSentinel)) {
t.Errorf("GetAgent response leaked sentinel:\n%s", body)
}
if !bytes.Contains([]byte(body), []byte("single-agent")) {
t.Errorf("GetAgent response missing the agent name (handler may not be serving data):\n%s", body)
}
}
func TestRegisterAgent_DoesNotLeakAPIKeyHash(t *testing.T) {
// Registration is the most likely path for a freshly-hashed key to
// leak: the service mints a new APIKeyHash inside RegisterAgent
// (service/agent.go:405) and the handler returns the agent struct
// verbatim. Pin that the redaction holds even on a "freshly created"
// agent payload.
now := time.Now()
mock := &MockAgentService{
RegisterAgentFn: func(in domain.Agent) (*domain.Agent, error) {
return &domain.Agent{
ID: "agent-new", Name: in.Name, Hostname: in.Hostname,
Status: domain.AgentStatusOnline, RegisteredAt: now,
APIKeyHash: g2HandlerLeakSentinel,
}, nil
},
}
h := NewAgentHandler(mock)
body := bytes.NewBufferString(`{"name":"freshly-registered","hostname":"new.host"}`)
req := httptest.NewRequest(http.MethodPost, "/api/v1/agents", body)
req = req.WithContext(contextWithRequestID())
w := httptest.NewRecorder()
h.RegisterAgent(w, req)
if w.Code != http.StatusCreated {
t.Fatalf("RegisterAgent status = %d, want 201, body=%s", w.Code, w.Body.String())
}
respBody := w.Body.String()
if bytes.Contains([]byte(respBody), []byte("api_key_hash")) {
t.Errorf("RegisterAgent response leaked \"api_key_hash\" key:\n%s", respBody)
}
if bytes.Contains([]byte(respBody), []byte(g2HandlerLeakSentinel)) {
t.Errorf("RegisterAgent response leaked sentinel:\n%s", respBody)
}
if !bytes.Contains([]byte(respBody), []byte("agent-new")) {
t.Errorf("RegisterAgent response missing the new agent ID (handler may not be serving data):\n%s", respBody)
}
}
func TestListRetiredAgents_DoesNotLeakAPIKeyHash(t *testing.T) {
// I-004 surface — separate handler from ListAgents; same leak risk.
now := time.Now()
retiredAt := now.Add(-1 * time.Hour)
reason := "test cascade"
mock := &MockAgentService{
ListRetiredAgentsFn: func(page, perPage int) ([]domain.Agent, int64, error) {
return []domain.Agent{
{ID: "ret-1", Name: "retired-one", Hostname: "host-r1",
Status: domain.AgentStatusOffline, RegisteredAt: now,
RetiredAt: &retiredAt, RetiredReason: &reason,
APIKeyHash: g2HandlerLeakSentinel},
}, 1, nil
},
}
h := NewAgentHandler(mock)
req := httptest.NewRequest(http.MethodGet, "/api/v1/agents/retired?page=1&per_page=50", nil)
req = req.WithContext(contextWithRequestID())
w := httptest.NewRecorder()
h.ListRetiredAgents(w, req)
if w.Code != http.StatusOK {
t.Fatalf("ListRetiredAgents status = %d, want 200, body=%s", w.Code, w.Body.String())
}
body := w.Body.String()
if bytes.Contains([]byte(body), []byte("api_key_hash")) {
t.Errorf("ListRetiredAgents response leaked \"api_key_hash\" key:\n%s", body)
}
if bytes.Contains([]byte(body), []byte(g2HandlerLeakSentinel)) {
t.Errorf("ListRetiredAgents response leaked sentinel:\n%s", body)
}
if !bytes.Contains([]byte(body), []byte("ret-1")) {
t.Errorf("ListRetiredAgents response missing the retired agent ID:\n%s", body)
}
}
+7 -1
View File
@@ -7,8 +7,14 @@ import (
)
// HealthHandler handles health and readiness check endpoints.
//
// G-1 (P1): AuthType is one of "api-key" or "none" — see
// internal/config.AuthType / config.ValidAuthTypes() for the typed
// constants and the rationale for dropping "jwt" (no JWT middleware
// ships with certctl; operators who need JWT/OIDC front certctl with
// an authenticating gateway and set AuthType="none" on the upstream).
type HealthHandler struct {
AuthType string // "api-key", "jwt", "none"
AuthType string // "api-key" or "none" (see config.AuthType constants)
}
// NewHealthHandler creates a new HealthHandler.
+8 -24
View File
@@ -162,30 +162,14 @@ func TestAuthInfo_ReturnsAuthType_None(t *testing.T) {
}
}
func TestAuthInfo_ReturnsAuthType_JWT(t *testing.T) {
handler := NewHealthHandler("jwt")
req, err := http.NewRequest(http.MethodGet, "/api/v1/auth/info", nil)
if err != nil {
t.Fatalf("NewRequest failed: %v", err)
}
w := httptest.NewRecorder()
handler.AuthInfo(w, req)
var result map[string]interface{}
if err := json.NewDecoder(w.Body).Decode(&result); err != nil {
t.Fatalf("failed to decode response: %v", err)
}
if result["auth_type"] != "jwt" {
t.Errorf("auth_type = %q, want jwt", result["auth_type"])
}
if required, ok := result["required"].(bool); !ok || !required {
t.Errorf("required = %v, want true", result["required"])
}
}
// G-1 (P1): the prior `TestAuthInfo_ReturnsAuthType_JWT` asserted the
// handler echoed "jwt" — using the silent-auth-downgrade value as a
// test fixture, which baked the lie into the regression suite. The
// test is removed because "jwt" is now rejected at config-load time
// (see internal/config/config_test.go::TestValidate_JWTAuth_RejectedDedicated)
// and never reaches this handler. The pre-existing
// `TestAuthInfo_ReturnsAuthType_APIKey` above (line ~107) covers the
// api-key happy path; nothing else needs replacing here.
func TestAuthCheck_ReturnsOK(t *testing.T) {
handler := NewHealthHandler("api-key")
+7 -1
View File
@@ -117,8 +117,14 @@ func HashAPIKey(key string) string {
}
// AuthConfig holds configuration for the Auth middleware.
//
// G-1 (P1): valid Type values are "api-key" or "none" only. "jwt" was
// removed because no JWT middleware ships with certctl (silent auth
// downgrade pre-G-1). The single source of truth for the allowed set
// lives at internal/config.AuthType / config.ValidAuthTypes() — prefer
// those constants over string literals when comparing.
type AuthConfig struct {
Type string // "api-key", "jwt", "none"
Type string // "api-key" or "none" (see config.AuthType constants)
Secret string // The raw API key or comma-separated list of valid API keys
}
+80 -12
View File
@@ -802,13 +802,59 @@ type NamedAPIKey struct {
Admin bool
}
// AuthType is the discriminator for the API auth middleware shape. The
// string alias preserves env-var roundtrip (the value flows through getEnv
// as a plain string) while giving us a typed surface for switches and
// validation. Use the named constants below rather than string literals
// so future enum additions/removals are caught at compile time.
//
// G-1 (P1): the pre-G-1 validAuthTypes map literal accepted "jwt" with no
// JWT middleware behind it (silent auth downgrade — the configured type
// was logged as "jwt" but every request routed through the api-key bearer
// middleware regardless). Operators who set CERTCTL_AUTH_TYPE=jwt thought
// they had JWT auth; they didn't. The typed alias + ValidAuthTypes()
// helper make the allowed set the single source of truth across config
// validation, the runtime defense-in-depth switch in main.go, and the
// helm-chart template guard (`certctl.validateAuthType`).
type AuthType string
const (
// AuthTypeAPIKey routes requests through the api-key bearer middleware.
// CERTCTL_AUTH_SECRET (or CERTCTL_API_KEYS_NAMED) is required.
AuthTypeAPIKey AuthType = "api-key"
// AuthTypeNone disables authentication entirely. Development only —
// the server logs a loud Warn at startup. Operators who need
// JWT/OIDC/mTLS run an authenticating gateway (oauth2-proxy / Envoy
// ext_authz / Traefik ForwardAuth / Pomerium) in front of certctl
// and set this value on the upstream certctl process. See
// docs/architecture.md "Authenticating-gateway pattern".
AuthTypeNone AuthType = "none"
)
// ValidAuthTypes returns the allowed CERTCTL_AUTH_TYPE values. The set is
// intentionally narrow — JWT was accepted pre-G-1 with no middleware
// implementation behind it. Single source of truth referenced by the
// validator below, the runtime guard in cmd/server/main.go, the helm
// chart template (`certctl.validateAuthType`), and the property test in
// config_test.go that pins "jwt" out of the slice forever.
func ValidAuthTypes() []AuthType {
return []AuthType{AuthTypeAPIKey, AuthTypeNone}
}
// AuthConfig contains authentication configuration.
type AuthConfig struct {
// Type sets the authentication mechanism for the REST API.
// Valid values: "api-key" (default, production), "jwt", "none" (development only).
// When "api-key", clients must provide Authorization: Bearer <key> header.
// "none" requires explicit opt-in via CERTCTL_AUTH_TYPE env var with warning logged.
// Valid values: "api-key" (default, production) and "none" (development
// only — disables authentication on the API and logs a loud Warn at
// startup). For JWT/OIDC, run an authenticating gateway (oauth2-proxy /
// Envoy / Traefik ForwardAuth / Pomerium) in front of certctl and set
// CERTCTL_AUTH_TYPE=none on the upstream — see docs/architecture.md
// "Authenticating-gateway pattern" and docs/upgrade-to-v2-jwt-removal.md.
// Setting: CERTCTL_AUTH_TYPE environment variable. Default: "api-key".
// Use the AuthType constants (AuthTypeAPIKey / AuthTypeNone) for typed
// comparisons; the field stays `string` to preserve env-var roundtrip
// shape used by getEnv() and downstream Helm/compose interpolation.
Type string
// Secret is the legacy authentication secret (comma-separated API keys).
@@ -1148,18 +1194,40 @@ func (c *Config) Validate() error {
return fmt.Errorf("invalid log format: %s", c.Log.Format)
}
// Validate auth type
validAuthTypes := map[string]bool{
"api-key": true,
"jwt": true,
"none": true,
// Validate auth type.
//
// G-1 (P1): the pre-G-1 set was {"api-key", "jwt", "none"} with "jwt"
// accepted but no JWT middleware shipped — silent auth downgrade.
// Post-G-1 we route a literal "jwt" value through a dedicated
// rejection that gives operators actionable guidance (the
// authenticating-gateway pattern) instead of the generic
// "invalid auth type". Then we cross-check against ValidAuthTypes()
// so any value outside {api-key, none} surfaces uniformly.
if c.Auth.Type == "jwt" {
return fmt.Errorf(
"CERTCTL_AUTH_TYPE=jwt is no longer accepted (G-1 silent auth " +
"downgrade): no JWT middleware ships with certctl. To use " +
"JWT/OIDC, run an authenticating gateway (oauth2-proxy / " +
"Envoy ext_authz / Traefik ForwardAuth / Pomerium) in " +
"front of certctl and set CERTCTL_AUTH_TYPE=none on the " +
"upstream. See docs/architecture.md \"Authenticating-" +
"gateway pattern\" and docs/upgrade-to-v2-jwt-removal.md " +
"for the migration walkthrough")
}
if !validAuthTypes[c.Auth.Type] {
return fmt.Errorf("invalid auth type: %s", c.Auth.Type)
authTypeValid := false
for _, t := range ValidAuthTypes() {
if AuthType(c.Auth.Type) == t {
authTypeValid = true
break
}
}
if !authTypeValid {
return fmt.Errorf("invalid auth type: %s (valid: %v)", c.Auth.Type, ValidAuthTypes())
}
// If using JWT or API-key, secret is required
if (c.Auth.Type == "jwt" || c.Auth.Type == "api-key") && c.Auth.Secret == "" {
// If using API-key, secret is required. (Secret was previously also
// required for "jwt"; removed with the jwt rejection above.)
if c.Auth.Type == string(AuthTypeAPIKey) && c.Auth.Secret == "" {
return fmt.Errorf("auth secret is required for auth type %s", c.Auth.Type)
}
+149 -15
View File
@@ -458,6 +458,8 @@ func TestValidate_InvalidAuthType(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
if err := cfg.Validate(); err == nil {
@@ -477,6 +479,8 @@ func TestValidate_APIKeyAuth_MissingSecret(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
if err := cfg.Validate(); err == nil {
@@ -484,25 +488,133 @@ func TestValidate_APIKeyAuth_MissingSecret(t *testing.T) {
}
}
func TestValidate_JWTAuth_MissingSecret(t *testing.T) {
cfg := &Config{
Server: validServerConfig(t),
Database: DatabaseConfig{URL: "postgres://localhost/certctl", MaxConnections: 25},
Log: LogConfig{Level: "info", Format: "json"},
Auth: AuthConfig{Type: "jwt", Secret: ""},
Keygen: KeygenConfig{Mode: "agent"},
Scheduler: SchedulerConfig{
RenewalCheckInterval: 1 * time.Hour,
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
},
// TestValidate_JWTAuth_RejectedDedicated locks down the G-1 fix: pre-G-1
// `CERTCTL_AUTH_TYPE=jwt` was accepted by the validator (the bare error
// path was the empty-secret one previously). Post-G-1 the literal "jwt"
// value is rejected with a dedicated diagnostic regardless of whether
// Secret is set, because there is no JWT middleware in the binary —
// operators who need JWT/OIDC must front certctl with an authenticating
// gateway.
//
// Two table rows pin the contract: missing-secret cannot paper over the
// rejection (the dedicated error fires first, before the secret check),
// and a populated secret also cannot paper over it. Both paths must
// hit the dedicated G-1 diagnostic, not the generic "invalid auth
// type" or "auth secret is required".
func TestValidate_JWTAuth_RejectedDedicated(t *testing.T) {
t.Parallel()
cases := []struct {
name string
secret string
}{
{"jwt rejected (no secret)", ""},
{"jwt rejected (with secret — operator can't paper over)", "anything"},
}
if err := cfg.Validate(); err == nil {
t.Error("Validate() should return error when jwt auth has empty secret")
for _, tc := range cases {
t.Run(tc.name, func(t *testing.T) {
cfg := &Config{
Server: validServerConfig(t),
Database: DatabaseConfig{URL: "postgres://localhost/certctl", MaxConnections: 25},
Log: LogConfig{Level: "info", Format: "json"},
Auth: AuthConfig{Type: "jwt", Secret: tc.secret},
Keygen: KeygenConfig{Mode: "agent"},
Scheduler: SchedulerConfig{
RenewalCheckInterval: 1 * time.Hour,
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
err := cfg.Validate()
if err == nil {
t.Fatal("Validate() returned nil; expected dedicated G-1 rejection")
}
const wantSubstr = "CERTCTL_AUTH_TYPE=jwt is no longer accepted"
if !strings.Contains(err.Error(), wantSubstr) {
t.Errorf("Validate() = %v\nwant substring %q (the dedicated G-1 diagnostic)", err, wantSubstr)
}
})
}
}
// TestValidAuthTypesDoesNotContainJWT is a property-level guard against
// a future PR silently re-introducing "jwt" into the allowed set. If
// someone adds JWT back to ValidAuthTypes(), this test fails immediately
// with a pointer at the audit finding. The matching CI grep guardrail
// in .github/workflows/ci.yml provides a secondary check at build time.
func TestValidAuthTypesDoesNotContainJWT(t *testing.T) {
t.Parallel()
for _, at := range ValidAuthTypes() {
if at == "jwt" {
t.Fatalf("jwt is in ValidAuthTypes — silent auth downgrade regressed (G-1)")
}
}
}
// TestValidAuthTypesIsExactly_APIKey_None pins the current allowed set.
// If a future change adds a new auth type, this test must be updated
// alongside the validator and the helm-chart `validateAuthType` helper —
// keeping all three surfaces in sync.
func TestValidAuthTypesIsExactly_APIKey_None(t *testing.T) {
t.Parallel()
got := ValidAuthTypes()
if len(got) != 2 {
t.Fatalf("ValidAuthTypes() returned %d entries, want 2: %v", len(got), got)
}
want := map[AuthType]bool{AuthTypeAPIKey: true, AuthTypeNone: true}
for _, at := range got {
if !want[at] {
t.Errorf("unexpected auth type in ValidAuthTypes: %q", at)
}
}
}
// TestValidate_GenericInvalidAuthType ensures that values outside the
// allowed set (other than the special-cased "jwt") still surface the
// generic "invalid auth type" error. Pins that the dedicated G-1
// rejection didn't accidentally swallow non-jwt typos.
func TestValidate_GenericInvalidAuthType(t *testing.T) {
t.Parallel()
for _, badType := range []string{"", "garbage", "oidc", "mtls", "API-KEY"} {
t.Run("type="+badType, func(t *testing.T) {
cfg := &Config{
Server: validServerConfig(t),
Database: DatabaseConfig{URL: "postgres://localhost/certctl", MaxConnections: 25},
Log: LogConfig{Level: "info", Format: "json"},
Auth: AuthConfig{Type: badType, Secret: "x"},
Keygen: KeygenConfig{Mode: "agent"},
Scheduler: SchedulerConfig{
RenewalCheckInterval: 1 * time.Hour,
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
err := cfg.Validate()
if err == nil {
t.Fatalf("Validate(type=%q) returned nil; expected invalid-auth-type rejection", badType)
}
if !strings.Contains(err.Error(), "invalid auth type") {
t.Errorf("Validate(type=%q) = %v; want \"invalid auth type\" error", badType, err)
}
if strings.Contains(err.Error(), "G-1 silent auth") {
t.Errorf("Validate(type=%q) = %v; should not hit the dedicated G-1 path for non-jwt values", badType, err)
}
})
}
}
// G-1 (P1): no need to add `TestValidate_NoneAuth_AcceptsEmptySecret` or
// `TestValidate_APIKeyAuth_RequiresSecret` here — the pre-existing tests
// `TestValidate_AuthTypeNone` (above) and `TestValidate_APIKeyAuth_MissingSecret`
// (above) already cover those paths. Documented for the next reader: the
// G-1 fix flipped jwt off but did not disturb either the
// none-bypasses-secret or the api-key-requires-secret behavior.
func TestValidate_InvalidKeygenMode(t *testing.T) {
cfg := &Config{
Server: validServerConfig(t),
@@ -515,6 +627,8 @@ func TestValidate_InvalidKeygenMode(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
if err := cfg.Validate(); err == nil {
@@ -544,6 +658,8 @@ func TestValidate_InvalidPort(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
if err := cfg.Validate(); err == nil {
@@ -574,6 +690,8 @@ func TestValidate_TLSCertPathEmpty(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
err := cfg.Validate()
@@ -605,6 +723,8 @@ func TestValidate_TLSKeyPathEmpty(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
err := cfg.Validate()
@@ -637,6 +757,8 @@ func TestValidate_TLSCertFileMissing(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
err := cfg.Validate()
@@ -668,6 +790,8 @@ func TestValidate_TLSKeyFileMissing(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
err := cfg.Validate()
@@ -701,6 +825,8 @@ func TestValidate_TLSMismatchedPair(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
err := cfg.Validate()
@@ -724,6 +850,8 @@ func TestValidate_EmptyDatabaseURL(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
if err := cfg.Validate(); err == nil {
@@ -743,6 +871,8 @@ func TestValidate_InvalidLogLevel(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
if err := cfg.Validate(); err == nil {
@@ -762,6 +892,8 @@ func TestValidate_InvalidLogFormat(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
if err := cfg.Validate(); err == nil {
@@ -840,6 +972,8 @@ func TestValidate_DatabaseMaxConnectionsZero(t *testing.T) {
JobProcessorInterval: 30 * time.Second,
AgentHealthCheckInterval: 2 * time.Minute,
NotificationProcessInterval: 1 * time.Minute,
NotificationRetryInterval: 2 * time.Minute,
RetryInterval: 5 * time.Minute,
},
}
if err := cfg.Validate(); err == nil {
+45 -1
View File
@@ -46,7 +46,26 @@ type Agent struct {
Status AgentStatus `json:"status"`
LastHeartbeatAt *time.Time `json:"last_heartbeat_at,omitempty"`
RegisteredAt time.Time `json:"registered_at"`
APIKeyHash string `json:"api_key_hash"`
// APIKeyHash is the SHA-256 of the agent's plaintext API key,
// populated by service.RegisterAgent (`hashAPIKey(apiKey)`) and
// consumed by repository.AgentRepository::GetByAPIKey at auth time.
// It is server-internal: never serialized to clients, never echoed
// via CLI / MCP / agent registration response, never logged.
//
// G-2 (P1): pre-G-2 the field was tagged `json:"api_key_hash"` and
// shipped on every /api/v1/agents response (cat-s5-apikey_leak). Even
// SHA-256 should not be shipped to clients — it gives an offline
// brute-force target if API-key entropy is low (certctl doesn't enforce
// a minimum on operator-supplied keys), and there is no business reason
// for any client to ever receive it. Post-G-2 the JSON tag is "-" and
// Agent.MarshalJSON below zeroes the field on a copy before delegating
// to the default marshal — defense in depth so a future tag-revert by
// refactor cannot reopen the leak. The DB column, repo SELECT/INSERT/
// UPDATE paths, and service-side hashing are unchanged. See
// docs/architecture.md ER diagram (which documents DB shape, not API
// shape) and coverage-gap-audit-2026-04-24-v5/unified-audit.md
// cat-s5-apikey_leak for the full closure rationale.
APIKeyHash string `json:"-"`
OS string `json:"os"`
Architecture string `json:"architecture"`
IPAddress string `json:"ip_address"`
@@ -60,6 +79,31 @@ type Agent struct {
RetiredReason *string `json:"retired_reason,omitempty"`
}
// MarshalJSON implements json.Marshaler. It explicitly zeros APIKeyHash
// before serialization to defense-in-depth the `json:"-"` tag above.
//
// G-2 (P1): pre-G-2 the field was tagged `json:"api_key_hash"` and
// shipped on every /api/v1/agents response (cat-s5-apikey_leak). Post-G-2
// the tag is "-" and this method enforces redaction even if the tag is
// reverted by a future refactor — the receiver is by-value so the
// APIKeyHash = "" assignment mutates only the marshal-time copy, never
// the caller's original. The type-alias trick (`type alias Agent`)
// breaks the recursive MarshalJSON call that would otherwise stack-
// overflow. Both *Agent and Agent receivers route through here because
// the json package looks the method up via reflect.Value, and a value
// receiver satisfies both kinds of pointer.
//
// Auditor's note for the next reader: do NOT remove this method even if
// the json:"-" tag stays. The CI guardrail at .github/workflows/ci.yml
// also blocks reintroduction at the tag site, but this method is the
// last line of defense for serialization paths that bypass struct tags
// (e.g., a future MarshalJSON on a parent struct that embeds Agent).
func (a Agent) MarshalJSON() ([]byte, error) {
type alias Agent // breaks recursion: alias has no MarshalJSON method
a.APIKeyHash = ""
return json.Marshal(alias(a))
}
// IsRetired returns true when this agent has been soft-retired.
// I-004: callers that iterate active agents (stats dashboard, stale-offline
// sweeper, handler-facing list) must skip retired rows by default.
+181
View File
@@ -1,10 +1,22 @@
package domain
import (
"encoding/json"
"strings"
"testing"
"time"
)
// jsonMarshalDirect / jsonUnmarshalDirect are thin aliases for the
// stdlib encoding/json calls. They exist only to make the G-2 redaction
// tests below grep-friendly: any future test author searching for "how
// is APIKeyHash redaction tested" will land on these and the call sites,
// rather than having to grep through dozens of unrelated json.Marshal
// usages.
func jsonMarshalDirect(v interface{}) ([]byte, error) { return json.Marshal(v) }
func jsonUnmarshalDirect(data []byte, v interface{}) error { return json.Unmarshal(data, v) }
func containsSubstr(haystack, needle string) bool { return strings.Contains(haystack, needle) }
// TestAgent_IsRetired covers the I-004 soft-retirement predicate that gates
// which callers hide an agent row from active listings.
func TestAgent_IsRetired(t *testing.T) {
@@ -53,3 +65,172 @@ func TestAgentDependencyCounts_HasDependencies(t *testing.T) {
})
}
}
// G-2 (P1): the cat-s5-apikey_leak audit closure tests. Pre-G-2,
// Agent.APIKeyHash was tagged `json:"api_key_hash"` and shipped on
// every /api/v1/agents response — credential-derivative leak that gave
// offline brute-force targets to every authenticated client. Post-G-2
// the tag is "-" AND Agent.MarshalJSON zeroes the field on a marshal-
// time copy. These tests pin both layers of the defense:
//
// 1. A populated APIKeyHash is never present in the marshaled JSON.
// 2. The redaction holds on *Agent, on slice elements, and on a
// sentinel literal-value check (so even a future field that
// happens to contain the same hash string would not appear).
// 3. The marshal-time copy does not mutate the caller's original —
// receiver is by-value, but pin it explicitly so a future refactor
// that switches to pointer-receiver gets caught.
// 4. Round-trip preserves every other field (hash dropped on encode,
// cannot reappear on decode because the wire never carries it).
const g2LeakSentinel = "sha256:LEAKED-CREDENTIAL-DERIVATIVE-SENTINEL"
// TestAgent_MarshalJSON_RedactsAPIKeyHash is the marshal-boundary
// contract test: a single Agent value with a populated APIKeyHash must
// not emit the field name nor the sentinel value.
func TestAgent_MarshalJSON_RedactsAPIKeyHash(t *testing.T) {
t.Parallel()
a := Agent{
ID: "agent-test",
Name: "test-agent",
Hostname: "host.example",
Status: AgentStatusOnline,
RegisteredAt: time.Date(2026, 1, 1, 0, 0, 0, 0, time.UTC),
APIKeyHash: g2LeakSentinel,
OS: "linux",
Architecture: "amd64",
IPAddress: "10.0.0.1",
Version: "v2.0.49",
}
out, err := jsonMarshalDirect(a)
if err != nil {
t.Fatalf("Marshal returned error: %v", err)
}
body := string(out)
if containsSubstr(body, "api_key_hash") {
t.Errorf("marshaled body contains \"api_key_hash\" key — G-2 leak regressed:\n%s", body)
}
if containsSubstr(body, "APIKeyHash") {
t.Errorf("marshaled body contains \"APIKeyHash\" — type-alias redaction broke:\n%s", body)
}
if containsSubstr(body, g2LeakSentinel) {
t.Errorf("marshaled body contains the leak sentinel %q — value redaction broke:\n%s", g2LeakSentinel, body)
}
// Sanity: every OTHER non-zero field IS present (this guards against
// the type-alias trick accidentally dropping siblings).
for _, want := range []string{"agent-test", "test-agent", "host.example", "Online", "linux", "amd64", "10.0.0.1", "v2.0.49"} {
if !containsSubstr(body, want) {
t.Errorf("marshaled body missing expected field value %q:\n%s", want, body)
}
}
}
// TestAgent_MarshalJSON_RedactsViaPointer covers the *Agent path that
// handlers hit when calling JSON(w, http.StatusOK, agent) with a *Agent
// from svc.GetAgent. A value-receiver MarshalJSON is reachable from
// pointer values via reflect; this test pins that contract.
func TestAgent_MarshalJSON_RedactsViaPointer(t *testing.T) {
t.Parallel()
a := &Agent{ID: "agent-x", APIKeyHash: g2LeakSentinel}
out, err := jsonMarshalDirect(a)
if err != nil {
t.Fatalf("Marshal *Agent returned error: %v", err)
}
if containsSubstr(string(out), g2LeakSentinel) {
t.Errorf("*Agent marshal leaked sentinel:\n%s", string(out))
}
if containsSubstr(string(out), "api_key_hash") {
t.Errorf("*Agent marshal contains \"api_key_hash\" key:\n%s", string(out))
}
}
// TestAgent_MarshalJSON_RedactsInSlice covers the []domain.Agent path
// the ListAgents handler emits via PagedResponse{Data: agents}. Each
// element must be redacted independently.
func TestAgent_MarshalJSON_RedactsInSlice(t *testing.T) {
t.Parallel()
agents := []Agent{
{ID: "agent-1", APIKeyHash: g2LeakSentinel + "-1"},
{ID: "agent-2", APIKeyHash: g2LeakSentinel + "-2"},
{ID: "agent-3", APIKeyHash: g2LeakSentinel + "-3"},
}
out, err := jsonMarshalDirect(agents)
if err != nil {
t.Fatalf("Marshal []Agent returned error: %v", err)
}
body := string(out)
if containsSubstr(body, "api_key_hash") {
t.Errorf("[]Agent marshal contains \"api_key_hash\" key:\n%s", body)
}
for i := 1; i <= 3; i++ {
sentinel := g2LeakSentinel + "-" + string(rune('0'+i))
if containsSubstr(body, sentinel) {
t.Errorf("[]Agent marshal leaked sentinel %q:\n%s", sentinel, body)
}
}
// Every agent ID is present — the redaction didn't accidentally
// strip the entire element.
for _, id := range []string{"agent-1", "agent-2", "agent-3"} {
if !containsSubstr(body, id) {
t.Errorf("[]Agent marshal missing element ID %q:\n%s", id, body)
}
}
}
// TestAgent_MarshalJSON_DoesNotMutateReceiver pins the by-value-receiver
// contract: marshaling must not zero APIKeyHash on the caller's struct,
// only on the marshal-time copy. This guards against a future refactor
// that switches to pointer receiver and breaks every code path that
// marshals an Agent and then re-uses it (e.g., audit-event payload
// construction immediately after returning the agent in a handler).
func TestAgent_MarshalJSON_DoesNotMutateReceiver(t *testing.T) {
t.Parallel()
a := Agent{ID: "agent-keep", APIKeyHash: g2LeakSentinel}
if _, err := jsonMarshalDirect(a); err != nil {
t.Fatalf("Marshal returned error: %v", err)
}
if a.APIKeyHash != g2LeakSentinel {
t.Errorf("MarshalJSON mutated caller's APIKeyHash: got %q want %q", a.APIKeyHash, g2LeakSentinel)
}
}
// TestAgent_MarshalJSON_RoundTrip pins the wire-shape contract: an
// Agent marshaled to JSON and unmarshaled back into a fresh Agent has
// every field preserved EXCEPT APIKeyHash, which the wire never carries.
// This double-confirms the redaction is a one-way guarantee at the
// serialization boundary, not an accidental on-decode behavior.
func TestAgent_MarshalJSON_RoundTrip(t *testing.T) {
t.Parallel()
now := time.Date(2026, 1, 1, 12, 0, 0, 0, time.UTC)
hb := now.Add(-5 * time.Minute)
original := Agent{
ID: "agent-rt",
Name: "rt",
Hostname: "rt.host",
Status: AgentStatusOnline,
LastHeartbeatAt: &hb,
RegisteredAt: now,
APIKeyHash: g2LeakSentinel,
OS: "linux",
Architecture: "arm64",
IPAddress: "10.0.0.99",
Version: "v2.0.49",
}
out, err := jsonMarshalDirect(original)
if err != nil {
t.Fatalf("Marshal returned error: %v", err)
}
var got Agent
if err := jsonUnmarshalDirect(out, &got); err != nil {
t.Fatalf("Unmarshal returned error: %v", err)
}
if got.APIKeyHash != "" {
t.Errorf("APIKeyHash survived round-trip: got %q want empty (the wire must not carry it)", got.APIKeyHash)
}
if got.ID != original.ID || got.Name != original.Name || got.Hostname != original.Hostname {
t.Errorf("identity fields lost in round-trip: got %+v want %+v", got, original)
}
if got.OS != original.OS || got.Architecture != original.Architecture || got.IPAddress != original.IPAddress {
t.Errorf("metadata fields lost in round-trip: got %+v want %+v", got, original)
}
}
+67 -2
View File
@@ -2,14 +2,26 @@ package postgres
import (
"database/sql"
"errors"
"fmt"
"os"
"path/filepath"
"strings"
_ "github.com/lib/pq"
"github.com/lib/pq"
)
// pgErrInvalidPassword is the SQLSTATE for class 28 / code 28P01 —
// invalid_password — emitted by PostgreSQL when the client presents
// credentials that don't match pg_authid. Defined locally because the
// lib/pq package does not export named constants for SQLSTATE codes (it
// only exposes the typed string alias pq.ErrorCode and a name-lookup map
// at runtime). Pinned as a string constant rather than a pq.ErrorCode
// literal so the contract is grep-able from operator-facing log lines.
//
// Reference: https://www.postgresql.org/docs/16/errcodes-appendix.html
const pgErrInvalidPassword = "28P01"
// NewDB opens a PostgreSQL database connection and sets up connection pooling.
func NewDB(connStr string) (*sql.DB, error) {
db, err := sql.Open("postgres", connStr)
@@ -23,12 +35,65 @@ func NewDB(connStr string) (*sql.DB, error) {
// Ping to verify connection
if err := db.Ping(); err != nil {
return nil, fmt.Errorf("failed to ping database: %w", err)
return nil, wrapPingError(err)
}
return db, nil
}
// wrapPingError converts a db.Ping() failure into an operator-friendly
// diagnostic. The default wrap is the original opaque
// `"failed to ping database: <inner>"` shape. The exception is SQLSTATE 28P01
// (invalid_password): when postgres rejects the server's credentials we emit
// extended guidance that names the most common operator misstep — editing
// POSTGRES_PASSWORD in `.env` after the postgres named volume has already
// been initialized — and lists both the destructive (`docker compose down -v`)
// and non-destructive (`ALTER ROLE`) remediations.
//
// U-1 (P1, GitHub #10): closes the audit-flagged
// cat-u-quickstart_postgres_password_volume_trap finding. The postgres
// docker-entrypoint runs initdb only when /var/lib/postgresql/data is empty;
// on subsequent boots the password baked into pg_authid on first boot wins
// over whatever the env var carries, so the env-vs-pg_authid divergence is
// intrinsic to how the postgres image bootstraps and cannot be fixed by us
// upstream of pg_authid. The ergonomic answer is to surface a clear
// diagnostic at the failure site so operators don't waste an hour on
// "is my password right" before discovering the volume needs to be torn
// down (or the role's password rotated in-place).
//
// The wrap chain is preserved via fmt.Errorf("%w", err) so callers using
// errors.As(err, &*pq.Error) on the returned value continue to work; this
// matches the audit's "no substring matching on err.Error()" requirement
// from the M-1 sentinel-error migration.
//
// Returns nil when err is nil so callers can defensively pipe through this
// helper without an extra branch.
func wrapPingError(err error) error {
if err == nil {
return nil
}
var pqErr *pq.Error
if errors.As(err, &pqErr) && string(pqErr.Code) == pgErrInvalidPassword {
return fmt.Errorf(
"failed to ping database: postgres rejected the configured credentials "+
"(SQLSTATE %s — invalid_password). If you recently rotated POSTGRES_PASSWORD "+
"on a docker-compose deploy, the postgres container's data volume still "+
"holds the previous password: initdb seeds POSTGRES_PASSWORD into pg_authid "+
"only on first boot of a fresh data dir, so editing the env var after that "+
"point updates only the certctl-server container. Reset destructively with "+
"`docker compose -f deploy/docker-compose.yml down -v && "+
"docker compose -f deploy/docker-compose.yml up -d --build` (this DESTROYS "+
"all data in the postgres volume), or non-destructively with "+
"`docker compose -f deploy/docker-compose.yml exec postgres "+
"psql -U certctl -c \"ALTER ROLE certctl PASSWORD '<new-password>';\"` "+
"and then redeploy with the matching POSTGRES_PASSWORD. Underlying error: %w",
pgErrInvalidPassword, err)
}
return fmt.Errorf("failed to ping database: %w", err)
}
// RunMigrations reads and executes SQL migration files from a directory.
func RunMigrations(db *sql.DB, migrationsPath string) error {
// Check if migrations directory exists
+168
View File
@@ -0,0 +1,168 @@
// Internal-package tests for db.go — covers the diagnostic dispatch in
// wrapPingError. Lives in `package postgres` (not `postgres_test`) so it can
// call the unexported helper directly without exposing it on the API surface.
//
// Sibling integration tests in this directory live in `package postgres_test`
// (testcontainers-driven, schema-per-test). They exercise the live-DB
// happy path; this file owns the unit-level diagnostic dispatch and runs in
// `-short` mode without spinning up postgres.
//
// U-1 (P1, GitHub #10): closes the audit-flagged
// cat-u-quickstart_postgres_password_volume_trap finding by pinning the
// post-fix wrap-text contract for `db.Ping()` failures. Pre-U-1 every Ping
// error was wrapped with the same opaque `"failed to ping database: %w"`,
// so an operator who edited POSTGRES_PASSWORD after first-boot saw only
// `pq: password authentication failed for user "certctl"` in the server
// log with no pointer to the actual cause (postgres data dir retains the
// initial password from first-boot initdb; subsequent boots ignore the env
// var). Post-U-1 the SQLSTATE-28P01 path emits a multi-line diagnostic
// pointing at the down -v / ALTER ROLE remediation; non-auth failures
// retain the original wrap shape so verbose noise does not bleed into
// transient connection-refused / timeout paths.
package postgres
import (
"errors"
"strings"
"testing"
"github.com/lib/pq"
)
// TestWrapPingError_AuthFailureGuidance asserts the diagnostic wrap fires on
// SQLSTATE 28P01 (invalid_password) and contains all three contract elements:
// the SQLSTATE code (so operators can grep), the down-v destructive
// remediation, and the ALTER ROLE non-destructive remediation. Also asserts
// the wrap chain still satisfies errors.As(err, &*pq.Error) so callers that
// programmatically inspect the underlying postgres error code keep working.
func TestWrapPingError_AuthFailureGuidance(t *testing.T) {
t.Parallel()
original := &pq.Error{
Code: pq.ErrorCode("28P01"),
Message: `password authentication failed for user "certctl"`,
}
wrapped := wrapPingError(original)
if wrapped == nil {
t.Fatal("wrapPingError returned nil for a non-nil input")
}
got := wrapped.Error()
// Contract elements — the operator-facing string is what we ship.
wantSubstrings := []string{
"SQLSTATE 28P01", // operators grep on this
"POSTGRES_PASSWORD", // names the variable that traps
"first boot", // the mechanism in plain language
"down -v", // destructive remediation
"ALTER ROLE", // non-destructive remediation
}
for _, s := range wantSubstrings {
if !strings.Contains(got, s) {
t.Errorf("wrap text missing %q\ngot: %s", s, got)
}
}
// Wrap chain must still expose the underlying *pq.Error for callers
// that want to inspect Code / Detail / Constraint fields. Pre-fix
// callers used errors.As(err, &pqErr) on the unwrapped Ping result;
// the new wrap is fmt.Errorf("...%w", err) so errors.As must walk it.
var pqErr *pq.Error
if !errors.As(wrapped, &pqErr) {
t.Fatalf("errors.As did not extract *pq.Error from wrapped chain: %v", wrapped)
}
if pqErr.Code != "28P01" {
t.Errorf("extracted pq.Error.Code = %q, want %q", pqErr.Code, "28P01")
}
}
// TestWrapPingError_NonAuthErrorPreservesOriginalWrap guards against the
// guidance text bleeding into unrelated failure modes. SQLSTATE 08006
// (connection_failure) is the canonical non-auth case — server unreachable,
// TLS handshake failure, network drop. The wrap should be the original
// shape so transient-error log noise does not include the (now lengthy)
// volume-state remediation paragraph.
func TestWrapPingError_NonAuthErrorPreservesOriginalWrap(t *testing.T) {
t.Parallel()
original := &pq.Error{
Code: pq.ErrorCode("08006"),
Message: "connection refused",
}
wrapped := wrapPingError(original)
if wrapped == nil {
t.Fatal("wrapPingError returned nil for a non-nil input")
}
got := wrapped.Error()
// Original-wrap shape: prefix only, no guidance text.
const wantPrefix = "failed to ping database: "
if !strings.HasPrefix(got, wantPrefix) {
t.Errorf("expected prefix %q, got: %s", wantPrefix, got)
}
// Negative assertions: guidance text MUST NOT appear on non-auth paths.
mustNotContain := []string{
"SQLSTATE 08006", // we only call out 28P01 specifically
"POSTGRES_PASSWORD",
"down -v",
"ALTER ROLE",
}
for _, s := range mustNotContain {
if strings.Contains(got, s) {
t.Errorf("non-auth wrap leaked guidance substring %q\ngot: %s", s, got)
}
}
// Wrap chain still walks for errors.As — same contract as auth path.
var pqErr *pq.Error
if !errors.As(wrapped, &pqErr) {
t.Fatalf("errors.As did not extract *pq.Error from non-auth wrapped chain: %v", wrapped)
}
if pqErr.Code != "08006" {
t.Errorf("extracted pq.Error.Code = %q, want %q", pqErr.Code, "08006")
}
}
// TestWrapPingError_NonPqErrorPreservesOriginalWrap guards the network-level
// case: a pre-handshake failure (TCP refused, DNS, TLS) returns a
// non-*pq.Error from db.Ping(). errors.As must return false, the helper
// must fall through to the generic wrap, and the chain must remain walkable.
func TestWrapPingError_NonPqErrorPreservesOriginalWrap(t *testing.T) {
t.Parallel()
original := errors.New("dial tcp 127.0.0.1:5432: connect: connection refused")
wrapped := wrapPingError(original)
if wrapped == nil {
t.Fatal("wrapPingError returned nil for a non-nil input")
}
got := wrapped.Error()
const wantPrefix = "failed to ping database: "
if !strings.HasPrefix(got, wantPrefix) {
t.Errorf("expected prefix %q, got: %s", wantPrefix, got)
}
if strings.Contains(got, "SQLSTATE") || strings.Contains(got, "POSTGRES_PASSWORD") {
t.Errorf("network-level wrap leaked SQLSTATE/postgres guidance\ngot: %s", got)
}
if !errors.Is(wrapped, original) {
t.Errorf("errors.Is did not walk to original sentinel: %v", wrapped)
}
}
// TestWrapPingError_NilReturnsNil — defensive contract: if Ping returned nil
// (no failure), the helper must not synthesize a fake error. This isn't on
// the documented call path (NewDB only invokes wrapPingError inside the
// `if err != nil` branch), but pinning it prevents a future refactor from
// regressing the contract silently.
func TestWrapPingError_NilReturnsNil(t *testing.T) {
t.Parallel()
if got := wrapPingError(nil); got != nil {
t.Errorf("wrapPingError(nil) = %v, want nil", got)
}
}
+9
View File
@@ -51,6 +51,15 @@ export interface CertificateVersion {
created_at: string;
}
// G-2 (P1): `api_key_hash` is intentionally absent from this interface.
// The server-side struct (internal/domain/connector.go::Agent) carries
// the field for the auth-lookup path but redacts it via a custom
// MarshalJSON so it never reaches the JSON wire. Adding `api_key_hash`
// here would not magically populate it on the wire — and would mislead
// future contributors into thinking the field is part of the public
// API contract. See docs/architecture.md ER-diagram note and
// coverage-gap-audit-2026-04-24-v5/unified-audit.md cat-s5-apikey_leak
// for the closure rationale.
export interface Agent {
id: string;
name: string;