mirror of
https://github.com/shankar0123/certctl.git
synced 2026-06-09 18:38:53 +00:00
2e97cc10b84ee1d0f297bc3a2b2af9e6c08321ed
3 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 shankar0123
|
9b6294e83d |
auth-bundle-2 Phase 14: session + OIDC validation benchmarks (steady-state + cold paths) + auth-benchmarks.md operator doc + Makefile targets
Closes Phase 14 of cowork/auth-bundle-2-prompt.md. Ships four
benchmarks producing four numbers + the operator-doc table; three
default-tag benchmarks runnable on every CI runner, the fourth
(cold-cache OIDC) runnable on operator-side Docker hosts via the
new make target.
Files
=====
internal/auth/session/bench_test.go (NEW):
* BenchmarkSession_SteadyState (target p99 < 1ms; measured 5µs).
Warm in-memory repo + warm session row. Pure CPU: parseCookie +
HMAC verify + map lookup + sentinel checks.
* BenchmarkSession_ColdProcess (target p99 < 10ms; measured 7.1ms).
Same pipeline but with a configurable per-call delay simulating
a 1ms Postgres RTT on each repo call. Two repo calls per
Validate (signing-key fetch + session-row fetch) = 2ms minimum;
Go time.Sleep granularity adds ~1-2ms jitter. Documented why
testcontainers Postgres isn't viable inside b.N: 30+ second
container boot incompatible with per-iteration timing.
* slowSessionRepo + slowKeyRepo wrappers add the per-call delay
via time.Sleep; they delegate to the existing in-memory stubs.
* reportPercentiles helper sorts + reports p50/p95/p99/max via
b.ReportMetric (Go testing.B doesn't surface percentiles
natively).
internal/auth/oidc/bench_test.go (NEW):
* BenchmarkOIDC_SteadyState (target p99 < 5ms; measured 1.5ms).
Drives full HandleCallback against an in-process mockIdP
(httptest.Server localhost loopback). Pre-warmed JWKS cache via
RefreshKeys at setup. Pipeline: pre-login consume + state
compare + token exchange (localhost ~50-200µs) + go-oidc
Verify (RSA-2048 sig verify + alg pin) + service-layer iss/
aud/azp/at_hash/exp/iat/nonce re-checks + group-claim
resolution + group→role mapping + user upsert + session mint.
* The localhost-loopback /token call adds ~100-500µs of TCP
overhead vs pure crypto; the prompt's "no network calls"
steady-state framing accommodates this since the localhost
loopback is the closest practical proxy for a same-region
IdP /token call (which adds 5-15ms in production).
internal/auth/oidc/bench_keycloak_test.go (NEW, //go:build integration):
* BenchmarkOIDC_ColdCache (target p99 < 200ms; operator-runs).
Drives RefreshKeys against a live Keycloak container from the
Phase 10 testfixtures harness. Each iteration evicts the
in-process cache + re-fetches discovery + re-fetches JWKS over
real HTTP + re-runs the IdP-downgrade-attack defense.
* Network-bounded: the cold path is dominated by HTTPS RTT to
the IdP discovery endpoint, NOT crypto. The 200ms cap
accommodates a geographically-distant IdP (~150ms RTT) plus
the in-process JWKS fetch + downgrade-defense logic (~5ms
locally).
* Reuses the sharedKeycloak fixture from
integration_keycloak_test.go (Phase 10) so the benchmark
doesn't pay the 60-90s container boot cost separately. Skips
with a clear message if invoked without the integration test
setup.
* Reports p50/p95/p99/max in MILLISECONDS (vs the
microsecond-granularity steady-state benchmarks) since the
cold path is two orders of magnitude slower.
internal/auth/oidc/service_test.go (MODIFIED):
* Refactored newMockIdP(t *testing.T) to delegate to a new
newMockIdPWithTB(t testing.TB) sibling. Standard Go pattern
for sharing test fixtures between *testing.T and *testing.B.
No behavior change for existing service_test.go tests; the
benchmark file in bench_test.go calls newMockIdPWithTB(b)
to get the same fixture.
docs/operator/auth-benchmarks.md (NEW):
* Result table with all four benchmarks + targets + measured
numbers + status markers. Four-row matrix for the default-tag
benchmarks; the fourth row (cold-cache) is operator-recorded
with an empty cell waiting for the first Docker-equipped run.
* Hardware floor section pinning the 4 vCPU / 8 GiB RAM /
Postgres 16 / Go 1.25 baseline. GitHub-hosted Ubuntu runners
satisfy this; operators on weaker hardware re-record.
* "What each benchmark covers (and what it doesn't)" section
per benchmark, distinguishing the warm steady-state pipeline
from the cold path's network-bounded budget.
* "Cold-cache OIDC: how to run" subsection documenting the
make target + the test+benchmark coupling needed to populate
sharedKeycloak. Operator-recorded baseline table seeded
empty for first runs.
* "Why the cold path is bounded by network latency, not crypto"
section explaining the budget breakdown:
- TCP handshake (1 RTT)
- TLS 1.3 handshake (1-2 RTTs)
- 2 HTTPS GETs (discovery + JWKS, 1 RTT each)
- In-process crypto on the certctl side (~5-10ms total)
So the 200ms cap is operator-checkable: real measurement >
200ms means the IdP is slow OR network congestion OR DNS
issues — the diagnosis is upstream of certctl. Real
measurement < 200ms means the IdP is on a fast same-region
link.
* Methodology section pinning the per-iteration timing capture
+ sort + percentile-extract approach.
* Pre-merge audit section for the Phase 14 exit gate: four
benchmarks ran, four numbers recorded, steady-state targets
met, cold path is operator-runnable + measurably-bounded.
Makefile (MODIFIED):
* Added `make benchmark-auth` (default-tag, runs three of four
benchmarks at 2000 samples each).
* Added `make benchmark-auth-coldcache` (integration-tagged,
runs OIDC cold-cache against live Keycloak; requires Docker).
* Both targets carry explanatory comment blocks.
docs/README.md (MODIFIED):
* Added the auth-benchmarks.md doc to the Operator nav table
alongside performance-baselines.md.
Measured baselines at Phase 14 close (linux/arm64, 4 vCPU)
==========================================================
BenchmarkSession_SteadyState p99 = 5µs (target < 1ms) ✓ 200× under
BenchmarkSession_ColdProcess p99 = 7.1ms (target < 10ms) ✓
BenchmarkOIDC_SteadyState p99 = 1.5ms (target < 5ms) ✓ 3× under
BenchmarkOIDC_ColdCache operator-runs (Docker required)
Verification
============
* gofmt -l on three new bench files: clean.
* go vet ./internal/auth/session/... ./internal/auth/oidc/...: clean
(default tag).
* go vet -tags integration ./internal/auth/oidc/...: clean (integration
tag covers the bench_keycloak_test.go file).
* go test -short -count=1 across all 5 OIDC + session packages:
green; the bench_*_test.go files compile but don't run under
-short (testing.Short() guards + benchmarks are not selected
by -run pattern).
* All three runnable benchmarks executed and produce the numbers
above; recorded in auth-benchmarks.md.
|
||
|
shankar0123
|
1d01c87663 |
auth-bundle-2 Phase 7 + Phase 7.5: OIDC first-admin bootstrap +
break-glass admin (Argon2id, lockout, default-OFF, surface-invisibility)
Phase 7 — OIDC first-admin bootstrap (Decision 3):
- Optional AdminBootstrapHook closure on *oidc.Service. When wired,
HandleCallback consults the hook AFTER group resolution + user
upsert and BEFORE the empty-mapping fail-closed check. Hook
receives (providerID, groups, userID); returns grantAdmin=true
when the user matches CERTCTL_BOOTSTRAP_ADMIN_GROUPS AND no
admin exists yet in the tenant.
- cmd/server/main.go wires the hook as a closure that:
* Filters by CERTCTL_BOOTSTRAP_OIDC_PROVIDER_ID (if configured).
* Probes AdminExists via authActorRoleRepo (admin-already-exists
silently returns false; bootstrap mode is one-shot per tenant).
* Walks group intersection.
* On match: grants r-admin via authActorRoleRepo.Grant + emits
the bootstrap.oidc_first_admin audit row with
event_category=auth + INFO log.
- Coexists with the Bundle 1 env-var-token bootstrap. Both paths
can be configured; first match wins (admin-existence probe
short-circuits the second).
- HandleCallback's empty-mapping fail-closed check moved AFTER the
hook so a fresh deployment with zero group_role_mappings can
still mint the first admin.
- 5 tests in service_test.go: hook grants admin on match, hook
returns false preserves empty-mapping fail-closed, admin-already-
exists silently falls through to normal mapping, hook-error wraps
+ bubbles, idempotent when admin is already in the mapped role set.
Phase 7.5 — Break-glass admin (Decision 4, default-OFF):
Migration 000038 ships:
- breakglass_credentials table — at-most-one-credential-per-actor
(UNIQUE(actor_id)), Argon2id PHC-format password_hash, lockout
state machine (failure_count, locked_until, last_failure_at).
FK CASCADE on users(id) so deleting a user atomically removes
their credential.
- Two new permissions seeded into r-admin only:
auth.breakglass.admin — set/rotate/unlock/remove credentials.
auth.breakglass.login — actor uses break-glass to log in.
CanonicalPermissions extended in lockstep.
internal/auth/breakglass/service.go (~580 LOC):
- Service.Enabled() reflects CERTCTL_BREAKGLASS_ENABLED.
- SetPassword: Argon2id with OWASP 2024 params (m=64MiB, t=3, p=4,
salt=16 random bytes, output=32 bytes); per-password random salt;
PHC-format hash output. Min 12 / max 256 byte input.
- Authenticate: constant-time-compare via subtle.ConstantTimeCompare
on every code path. Identical 401 + identical timing across the
wrong-password / locked-account / non-existent-actor paths so an
attacker cannot probe whether a given actor has break-glass
configured. Non-existent-actor + locked-account paths run a
verifyDummy() Argon2id pass for timing parity. Lockout state
machine: failure_count++ on every wrong attempt; threshold (default
5) trips locked_until = NOW() + duration (default 15m). Successful
Authenticate resets the counter. Reset-window: failures aged out
after CERTCTL_BREAKGLASS_LOCKOUT_RESET_INTERVAL (default 1h)
auto-reset on next attempt.
- Unlock + RemoveCredential: admin-only (auth.breakglass.admin
gated at the router via rbacGate). Audit rows on every operation.
- All public methods refuse to act when Enabled()==false (returns
ErrDisabled; the handler maps to HTTP 404 — surface invisibility).
internal/repository/postgres/breakglass.go ships the 5-method
postgres impl with atomic single-statement IncrementFailure (so
concurrent racing wrong-password attempts can't observe an
intermediate state and slip past the threshold) and idempotent
ResetFailureCount.
internal/api/handler/auth_breakglass.go ships the 4-endpoint HTTP
surface:
- POST /auth/breakglass/login (auth-exempt; 5/min rate-limited per
source IP via the existing rate limiter; returns 404 when
disabled). On success sets the post-login session cookie + CSRF
cookie via SessionService.Create + 204. On any failure:
uniform 401 + identical timing (the service has already audited
the specific failure category).
- POST /api/v1/auth/breakglass/credentials (auth.breakglass.admin)
- POST /api/v1/auth/breakglass/credentials/{actor_id}/unlock
(auth.breakglass.admin)
- DELETE /api/v1/auth/breakglass/credentials/{actor_id}
(auth.breakglass.admin)
Admin endpoints share the surface-invisibility property: when
CERTCTL_BREAKGLASS_ENABLED=false, every admin endpoint also returns
404 (not 403) so probing via the admin surface gets the same signal
as probing the login endpoint.
Tests (internal/auth/breakglass/service_test.go):
All 8 Phase 7.5 spec-mandated negative cases:
1. Service.Enabled()==false → all ops return ErrDisabled.
2. Wrong password → ErrInvalidCredentials, failure_count++,
audit row with event_category=auth.
3. Failure_count exceeds threshold → locked, subsequent attempts
(including with the CORRECT password) return identical-shape
401 while the lockout window holds.
4. Lockout window expires → next attempt with correct password
succeeds + resets the counter.
5. Password < 12 bytes (or > 256 bytes) → ErrWeakPassword.
6. Password leak hygiene — the service has zero slog calls; the
audit-row map literal never includes the password plaintext.
7. Argon2id hash never appears in logs OR API responses — pinned
by `json:"-"` tag on BreakglassCredential.PasswordHash + a
belt-and-braces json.Marshal probe asserting the hash bytes
never appear in the marshaled output.
8. Constant-time-compare verified via timing-statistical test —
wrong-password vs no-credential paths take statistically
indistinguishable time (within 5x ratio). The verifyDummy()
hash compute on the no-credential + locked paths is what
keeps timing parity; absent that, an attacker could side-
channel "actor doesn't have a credential" via timing.
Plus coverage-lift batch covering: SetPassword first-time vs rotate,
no-caller-id rejection, no-target-id rejection, RNG failure surface,
Authenticate happy-path mints session, no-credential audit row,
session-mint-failure surface, FailureResetInterval recycle, Unlock
+ RemoveCredential happy paths, hash-format unit tests (round-trip,
mismatch, malformed/wrong-version/bad-base64 formats), nil-audit +
nil-session pass-through.
Coverage on internal/auth/breakglass/ at 91.5% per-statement (above
the Phase 7.5 spec ≥ 90% floor).
cmd/server/main.go wiring:
- Constructs breakglassRepo + breakglassService + breakglassHandler
after the OIDC service block.
- breakglassSessionMinterAdapter shim bridges *session.Service.Create
to the breakglass.SessionMinter port.
- Logs WARN at boot when CERTCTL_BREAKGLASS_ENABLED=true (operator
visibility for the deliberate SSO-bypass).
internal/config/config.go gains:
- AuthConfig.BootstrapAdminGroups + BootstrapOIDCProviderID for
Phase 7 (CERTCTL_BOOTSTRAP_ADMIN_GROUPS comma-list +
CERTCTL_BOOTSTRAP_OIDC_PROVIDER_ID).
- AuthConfig.Breakglass nested struct with 4 env vars
(CERTCTL_BREAKGLASS_ENABLED + LOCKOUT_THRESHOLD + LOCKOUT_DURATION
+ LOCKOUT_RESET_INTERVAL).
Router wiring:
- 4 new breakglass routes registered when reg.AuthBreakglass != nil;
public login route via direct r.mux.Handle (auth-exempt), 3 admin
routes via r.Register + rbacGate(auth.breakglass.admin).
- POST /auth/breakglass/login pinned in AuthExemptRouterRoutes
allowlist with Phase 7.5 justification.
- SpecParityExceptions extended with 4 new entries documenting
the Phase 7.5 deferral of full per-endpoint OpenAPI rows
(handler doc-block at the top of auth_breakglass.go is the
operator-facing reference).
Threat model (encoded in service.go + auth_breakglass.go doc-blocks
+ migration 000038 docstrings, to be promoted to docs/operator/auth-
threat-model.md in Phase 12):
- Break-glass is a deliberate bypass of the SSO security boundary.
An attacker who phishes the password OR finds it in a compromised
password manager bypasses MFA, OIDC, and every group-claim gate.
- Recommendation: keep CERTCTL_BREAKGLASS_ENABLED=false in steady-
state. Enable only during SSO-broken incidents. Disable after
recovery.
- WebAuthn pairing (v3 per Decision 12) is the load-bearing second
factor. Without it, break-glass is best treated as an emergency-
only path.
- Audit trail surfaces every break-glass action under
event_category=auth; the auditor role can monitor for unexpected
break-glass logins.
Verifications: gofmt clean, go vet clean across all touched packages,
go test -short -count=1 green across internal/auth/oidc (3.0s; new
Phase 7 hook tests integrated alongside the 21+ Phase 3 negatives),
internal/auth/breakglass (3.6s; 8 spec-mandated negatives + coverage
batch passing), internal/config + internal/domain/auth + internal/api/
router + internal/api/handler all green, no regressions in Bundle 1
packages.
|
||
|
shankar0123
|
854135dfb7 |
auth-bundle-2 Phase 3: OIDC service (HandleAuthRequest, HandleCallback,
RefreshKeys), hand-rolled group-claim resolver, 21+ negative-test
matrix, token-leak hygiene, IdP downgrade-attack defense
Phase 3 of the bundle ships the business logic that turns the Phase 2
storage primitives into a working OpenID Connect 1.0 + RFC 7636 PKCE
authorization-code flow against any enterprise IdP (Okta / Azure AD /
Google Workspace / Keycloak / Authentik / Auth0).
Service surface:
- Service.HandleAuthRequest(providerID) -> authURL, cookie, preLoginID
Builds the IdP redirect with PKCE-S256 (mandatory; RFC 9700 §2.1.1),
server-generated 32-byte state + nonce, persisted to the pre-login
row keyed by the cookie value.
- Service.HandleCallback(cookie, code, state, ip, ua) -> *CallbackResult
11-step validation: pre-login lookup-and-consume (single-use),
constant-time state compare, code-for-token exchange with PKCE
verifier, ID-token verify (alg pin via go-oidc/v3), service-layer
re-checks of iss / aud / azp (multi-aud requires it; mismatch
rejected) / at_hash (REQUIRED when access_token returned —
Phase 3 lifts the OIDC core "MAY" to a service-level "MUST") /
exp / iat-window / nonce, group-claim resolution with userinfo
fallback, group->role mapping (fail-closed on no match),
user upsert, session mint via SessionMinter port.
- Service.RefreshKeys(providerID) — explicit cache eviction +
re-load. Re-runs the IdP downgrade-attack defense so a provider
that later rotates to advertising HS* / none is caught BEFORE the
next user login attempt.
Security posture (every fail-closed branch is a sentinel error +
test):
- Algorithm pinning: allow-list {RS256, RS512, ES256, ES384, EdDSA};
deny-list {HS256, HS384, HS512, none}. Belt-and-braces re-check
via isDisallowedAlg after go-oidc.Verify.
- PKCE-S256 mandatory (oauth2.GenerateVerifier + S256ChallengeOption);
`plain` rejection sentinel exists for defense-in-depth.
- State + nonce: 32-byte crypto/rand, base64url-no-pad,
constant-time compare, single-use.
- IdP downgrade-attack defense: at provider creation / RefreshKeys,
reject any IdP whose discovery doc advertises HS* / none in
id_token_signing_alg_values_supported.
- JWKS fail-closed: in-flight login fails 503; existing sessions
untouched. isJWKSFetchError detects the gooidc verify-error
shape; ErrJWKSUnreachable is the wire mapping.
- Token-leak hygiene: ID tokens, access tokens, refresh tokens,
authorization codes, PKCE verifiers, state, nonce, signing key
bytes — NEVER logged at any level. logging_test.go pins the
invariant via a slog buffer + grep-assert across HandleAuthRequest,
HandleCallback, alg rejection, and provider-load paths.
Group-claim resolver (internal/auth/oidc/groupclaim/):
- Hand-rolled per Decision 10 (no JSON-path lib; ~150 LOC).
- URL-shape paths (https:// / http://) treated as a single
literal key — Auth0 namespaced claims like
https://your-namespace/groups work without splitting on the
dots in the URL.
- Dot-separated paths walked through nested map[string]interface{}.
- []interface{} / []string / single-string normalized to []string;
bool / number / object / nil → fail closed.
- 18 unit tests + sentinels (ErrPathEmpty, ErrSegmentMissing,
ErrSegmentNotObject, ErrInvalidValueType).
Test surface:
- service_test.go: 57 test functions including all 21 prompt-mandated
negative cases (wrong aud / wrong iss / expired / unknown alg /
alg=none / HMAC alg / azp missing on multi-aud / azp mismatched /
at_hash missing / at_hash mismatched / iat in future / iat too old /
nonce mismatched / state mismatched / state replayed / PKCE plain
sentinel / pre-login replay / forged cookie / IdP downgrade /
group-claim missing / group-claim unmapped) plus the userinfo
fallback matrix (happy path + endpoint-missing + endpoint-failing +
userinfo-also-empty), HandleAuthRequest entry point + RNG-failure
paths, upsertUser update + create + display-name fallback +
Validate-error paths, decryptClientSecret real-encrypt round-trip
+ bad-passphrase, alg-parser malformed-header matrix.
- logging_test.go: 4 hygiene tests pinning no token / code / verifier /
state / cookie / client_secret / alg name appears in any captured
log line.
- groupclaim/resolver_test.go: 18 cases covering Okta string-array,
Keycloak realm_access.roles, Auth0 namespaced URL claim,
single-string normalization, deeply-nested 3-segment walks, and
every fail-closed branch.
Coverage:
internal/auth/oidc 92.2% (floor: 90)
internal/auth/oidc/groupclaim 100.0% (floor: 95)
internal/auth/oidc/domain 96.2% (floor: 90)
Coverage gates added at .github/coverage-thresholds.yml so a future
regression in any fail-closed branch fails CI before the commit lands.
Phase 3 of cowork/auth-bundle-2-prompt.md is closed. Next up: Phase 4
(Session service: cookies, revocation, sliding-vs-absolute expiry).
|