assay-auth 0.6.0

Authentication, OIDC (client + provider), passkey, Argon2, JWT, Biscuit capability tokens, session management, and Zanzibar-style authorization for assay-engine.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
//! OIDC client — discovery, PKCE, callback, userinfo.
//!
//! Plan 12c task 5.1 reference. We wrap the [`openidconnect`] 4 typed
//! `CoreClient` per upstream so callers don't have to thread its
//! type-state generics through every handler. Each provider is
//! discovered once at registration time (`<issuer>/.well-known/openid-configuration`)
//! and the resulting client is cached behind a slug key.
//!
//! The phase-5 surface is intentionally library-only:
//!
//! - [`OidcRegistry`] — slug-keyed registry of discovered providers
//! - [`OidcClient`] — wraps one upstream's discovered metadata + RP creds
//! - [`UpstreamProvider`] — POD record (slug + issuer + client id/secret +
//!   scopes); matches the `auth.upstream_providers` row shape that admin
//!   CRUD will land in a later plan
//! - [`UpstreamUserInfo`] — verified result of one login round-trip
//!
//! Engine boot constructs an empty registry; populated providers come
//! from a future admin API or seed config (out of phase 5 scope).

use std::collections::{BTreeMap, HashMap};
use std::sync::Arc;
use std::time::{Duration, Instant};

use openidconnect::core::{
    CoreAuthenticationFlow, CoreClient, CoreIdToken, CoreIdTokenClaims, CoreIdTokenVerifier,
    CoreJsonWebKey, CoreJsonWebKeySet, CoreJwsSigningAlgorithm, CoreProviderMetadata,
    CoreUserInfoClaims,
};
use openidconnect::reqwest as oidc_reqwest;
use openidconnect::{
    AuthorizationCode, ClaimsVerificationError, ClientId, ClientSecret, CsrfToken,
    EndpointMaybeSet, EndpointNotSet, EndpointSet, IssuerUrl, JsonWebKeySetUrl, Nonce,
    OAuth2TokenResponse, PkceCodeChallenge, PkceCodeVerifier, RedirectUrl, Scope,
    SignatureVerificationError, SubjectIdentifier, TokenResponse,
};
use parking_lot::RwLock;
use url::Url;

use crate::error::{Error, Result};

/// Fallback scopes requested during upstream federation login when a
/// row's `scopes` column is empty. The admin API normally fills the
/// column from the request body (or its server-side default of
/// `'["openid","email","profile"]'`); this constant only fires for
/// rows that pre-date the V5 migration filling in the default.
pub const DEFAULT_UPSTREAM_SCOPES: &[&str] = &["openid", "email", "profile"];

/// Lower bound on how long a fetched upstream JWKS is trusted before a
/// proactive re-fetch — even when the certs endpoint advertises a
/// shorter `max-age`. Keeps us from re-fetching on essentially every
/// login.
const JWKS_MIN_TTL: Duration = Duration::from_secs(300);

/// Upper bound on the JWKS cache TTL. Google advertises multi-hour
/// `max-age`s on its certs endpoint; we cap the trusted window so a
/// cache that never gets a rotation-triggered refetch still self-heals
/// within a bounded interval.
const JWKS_MAX_TTL: Duration = Duration::from_secs(6 * 60 * 60);

/// TTL used when the certs response carries no usable `Cache-Control:
/// max-age` directive.
const JWKS_DEFAULT_TTL: Duration = Duration::from_secs(60 * 60);

/// Minimum gap between *reactive* (unknown-`kid`) JWKS re-fetches. A
/// burst of tokens carrying a `kid` we'll never hold — bogus, or from a
/// misconfigured upstream — can't stampede the certs endpoint faster
/// than this.
const JWKS_MIN_REFETCH_INTERVAL: Duration = Duration::from_secs(60);

/// In-memory, refreshable cache of one upstream's signing keys.
///
/// Seeded from the discovery metadata at registration, then re-fetched
/// from `jwks_uri` when the keys go stale — either the TTL lapsed
/// (honoring the certs endpoint's `Cache-Control: max-age`) or an
/// incoming id_token carried a `kid` we don't yet hold (upstream key
/// rotation). Before this cache existed the verifier pinned the
/// boot-time keys forever, so every login signed by a rotated Google
/// key failed with `Signature verification failed`.
struct JwksCache {
    keys: Vec<CoreJsonWebKey>,
    /// Deadline for a proactive refresh, set from the last fetch's TTL.
    expires_at: Instant,
    /// When we last hit `jwks_uri` — rate-limits reactive refetches.
    last_fetch: Instant,
}

/// POD record describing one upstream identity provider. Mirrors the
/// planned `auth.upstream_providers` table shape (see plan 12d) so the
/// admin API can `INSERT … RETURNING *` and feed the row directly into
/// [`OidcRegistry::add`] without a translation step.
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct UpstreamProvider {
    /// Stable slug used in routes (`/login/{slug}`) and as the
    /// `auth.user_upstream.provider` column value. Lower-snake-case
    /// matches the rest of the codebase's naming.
    pub slug: String,
    /// Issuer URL — the value the discovery doc lives under
    /// (`<issuer>/.well-known/openid-configuration`).
    pub issuer: String,
    /// RP client id registered with the upstream.
    pub client_id: String,
    /// RP client secret registered with the upstream. Stored as
    /// plaintext here because phase 5 has no secret-at-rest envelope yet
    /// — admin CRUD lands with the encryption story.
    pub client_secret: String,
    /// Scopes requested at authorize time. Common set:
    /// `["openid", "email", "profile"]`. `openid` is added implicitly
    /// by [`openidconnect`]; we forward the rest unchanged.
    pub scopes: Vec<String>,
    /// Per-IdP authorize-URL parameters (`prompt`, `hd`, `domain_hint`,
    /// `idp_*`, …). Validated against the
    /// [`crate::oidc_provider::auth_params`] whitelist before the row
    /// reaches this struct. Empty by default.
    pub auth_params: BTreeMap<String, String>,
}

/// Verified userinfo returned by [`OidcClient::complete_login`]. Carries
/// the canonical fields the rest of the auth stack needs to upsert into
/// `auth.users` + `auth.user_upstream`. `raw_claims` carries the
/// id_token's full claim set so callers can pluck custom claims (e.g.
/// `groups`, `roles`) without a second parse.
#[derive(Clone, Debug)]
pub struct UpstreamUserInfo {
    pub provider: String,
    pub subject: String,
    pub email: Option<String>,
    pub email_verified: bool,
    pub name: Option<String>,
    pub picture: Option<String>,
    pub raw_claims: serde_json::Value,
}

/// A single discovered upstream — wraps the [`openidconnect`] typed
/// client and the PoD metadata used to construct it.
///
/// The CoreClient generic state after `from_provider_metadata +
/// set_redirect_uri` is `<EndpointSet, EndpointNotSet, EndpointNotSet,
/// EndpointNotSet, EndpointMaybeSet, EndpointMaybeSet>` — auth URL set
/// (so `authorize_url` works), token + userinfo MaybeSet (we error at
/// runtime if the upstream's discovery doc is missing one).
pub struct OidcClient {
    inner: CoreClient<
        EndpointSet,
        EndpointNotSet,
        EndpointNotSet,
        EndpointNotSet,
        EndpointMaybeSet,
        EndpointMaybeSet,
    >,
    /// Original PoD record for round-trip / introspection
    /// (e.g. admin "what's configured" pages).
    provider: UpstreamProvider,
    /// Owned redirect URL — `set_redirect_uri` consumed it on the
    /// builder, but operators sometimes want it back without re-parsing.
    redirect_uri: RedirectUrl,
    /// `jwks_uri` from the upstream's discovery doc — where we re-fetch
    /// signing keys when the cache goes stale.
    jwks_uri: JsonWebKeySetUrl,
    /// Algorithms the upstream advertised for id_token signing
    /// (`id_token_signing_alg_values_supported`), captured at discovery
    /// so a rebuilt verifier restricts algorithms exactly as the
    /// metadata-derived one did.
    signing_algs: Vec<CoreJwsSigningAlgorithm>,
    /// Refreshable signing-key cache. See [`JwksCache`].
    jwks: Arc<RwLock<JwksCache>>,
    /// Dedicated HTTP client for JWKS re-fetches. Kept separate from the
    /// `openidconnect`/`oauth2` reqwest used for token exchange: that
    /// crate pins a different `reqwest` major, so we fetch keys with the
    /// auth crate's own `reqwest` (matching [`crate::external_jwt`]).
    jwks_http: reqwest::Client,
}

impl OidcClient {
    /// Borrow the original PoD record.
    pub fn provider(&self) -> &UpstreamProvider {
        &self.provider
    }

    /// Borrow the configured redirect URI.
    pub fn redirect_uri(&self) -> &RedirectUrl {
        &self.redirect_uri
    }

    /// Step 1 of the authorization-code-+-PKCE flow. Generates a PKCE
    /// pair, asks the [`openidconnect`] client for the redirect URL,
    /// returns the URL alongside the verifier + nonce for round-trip
    /// (callers persist them, typically in the session).
    ///
    /// `state` lets callers pin a known CSRF value (e.g. the session id)
    /// rather than the library-generated random one — useful when the
    /// callback handler uses `state` to look the in-progress login up.
    /// Pass [`CsrfToken::new_random`] via `CsrfToken::new(...)` if you
    /// don't have one already.
    pub fn start_login(&self, state: CsrfToken) -> StartedLogin {
        let (pkce_challenge, pkce_verifier) = PkceCodeChallenge::new_random_sha256();
        let mut request = self.inner.authorize_url(
            CoreAuthenticationFlow::AuthorizationCode,
            move || state,
            Nonce::new_random,
        );
        for scope in &self.provider.scopes {
            // `openid` scope is added by openidconnect when
            // `use_openid_scope` is true (default after
            // `from_provider_metadata`); skip a duplicate so the URL
            // stays clean.
            if scope == "openid" {
                continue;
            }
            request = request.add_scope(Scope::new(scope.clone()));
        }
        for (k, v) in &self.provider.auth_params {
            request = request.add_extra_param(k.clone(), v.clone());
        }
        let (url, csrf_token, nonce) = request.set_pkce_challenge(pkce_challenge).url();
        StartedLogin {
            url,
            csrf_token,
            nonce,
            pkce_verifier,
        }
    }

    /// Step 2 — exchange the upstream's `code` for tokens, validate the
    /// id_token against the cached JWKS + nonce, and (when the upstream
    /// publishes a userinfo endpoint) supplement the claims with a
    /// userinfo call.
    ///
    /// `pkce_verifier` and `nonce` must be the values returned from
    /// [`OidcClient::start_login`] for the same login — callers persist
    /// them server-side keyed by `state`.
    pub async fn complete_login(
        &self,
        code: String,
        pkce_verifier: PkceCodeVerifier,
        nonce: Nonce,
    ) -> Result<UpstreamUserInfo> {
        let http = build_oidc_http_client(HttpClientOptions::default())?;
        let token_response = self
            .inner
            .exchange_code(AuthorizationCode::new(code))
            .map_err(|e| Error::Oidc(format!("exchange_code config: {e}")))?
            .set_pkce_verifier(pkce_verifier)
            .request_async(&http)
            .await
            .map_err(|e| Error::Oidc(format!("token exchange: {e}")))?;

        let id_token = token_response
            .id_token()
            .ok_or_else(|| Error::Oidc("upstream returned no id_token".to_string()))?;
        let claims = self.verify_id_token_claims(id_token, &nonce).await?;

        let subject = claims.subject().to_string();
        let mut email = claims.email().map(|e| e.to_string());
        let mut email_verified = claims.email_verified().unwrap_or(false);
        let mut name = claims
            .name()
            .and_then(|map| map.get(None))
            .map(|n| n.to_string());
        let mut picture = claims
            .picture()
            .and_then(|map| map.get(None))
            .map(|u| u.to_string());

        // Best-effort userinfo fetch. Some upstreams omit email/name from
        // the id_token and only expose them via /userinfo. If the
        // upstream doesn't publish a userinfo endpoint or the call fails,
        // we keep what the id_token gave us — login still works, the
        // missing fields just show up as None.
        let mut raw_claims =
            serde_json::to_value(claims).unwrap_or_else(|_| serde_json::json!({"sub": subject}));
        if let Ok(req) = self.inner.user_info(
            token_response.access_token().clone(),
            Some(SubjectIdentifier::new(subject.clone())),
        ) && let Ok(userinfo) = req.request_async(&http).await
        {
            let user_claims: CoreUserInfoClaims = userinfo;
            if email.is_none() {
                email = user_claims.email().map(|e| e.to_string());
            }
            if !email_verified {
                email_verified = user_claims.email_verified().unwrap_or(email_verified);
            }
            if name.is_none() {
                name = user_claims
                    .name()
                    .and_then(|map| map.get(None))
                    .map(|n| n.to_string());
            }
            if picture.is_none() {
                picture = user_claims
                    .picture()
                    .and_then(|map| map.get(None))
                    .map(|u| u.to_string());
            }
            // Merge userinfo into raw_claims so downstream code that
            // wants e.g. `groups` from userinfo can pluck it out.
            if let Ok(userinfo_value) = serde_json::to_value(&user_claims) {
                merge_json(&mut raw_claims, userinfo_value);
            }
        }

        Ok(UpstreamUserInfo {
            provider: self.provider.slug.clone(),
            subject,
            email,
            email_verified,
            name,
            picture,
            raw_claims,
        })
    }

    /// Verify the upstream id_token against the cached signing keys,
    /// transparently refreshing the JWKS when it has gone stale.
    ///
    /// Two refresh triggers, matching the OIDC key-rotation guidance:
    ///  * **Proactive** — the cache TTL (from the certs endpoint's
    ///    `Cache-Control: max-age`) has lapsed, so we re-fetch before
    ///    even trying.
    ///  * **Reactive** — verification fails with `NoMatchingKey`: the
    ///    token's `kid` isn't one we hold, i.e. the upstream just
    ///    rotated. We re-fetch once (rate-limited) and retry.
    ///
    /// A proactive-refresh network failure is non-fatal — we fall back
    /// to the cached keys and let verification decide. Only a missing
    /// key triggers the reactive retry; other failures (bad audience,
    /// expired, genuinely bad signature) can't be fixed by re-fetching
    /// keys, so we surface them directly.
    async fn verify_id_token_claims<'t>(
        &self,
        id_token: &'t CoreIdToken,
        nonce: &Nonce,
    ) -> Result<&'t CoreIdTokenClaims> {
        let mut refreshed = false;
        let keys = if self.jwks.read().expires_at <= Instant::now() {
            match self.refresh_jwks().await {
                Ok(fresh) => {
                    refreshed = true;
                    fresh
                }
                Err(e) => {
                    tracing::warn!(
                        slug = %self.provider.slug,
                        error = %e,
                        "proactive jwks refresh failed; verifying against cached keys"
                    );
                    self.jwks.read().keys.clone()
                }
            }
        } else {
            self.jwks.read().keys.clone()
        };

        let verifier = self.id_token_verifier(keys)?;
        match id_token.claims(&verifier, nonce) {
            Ok(claims) => Ok(claims),
            Err(e)
                if is_unknown_signing_key(&e) && !refreshed && self.reactive_refetch_allowed() =>
            {
                tracing::info!(
                    slug = %self.provider.slug,
                    "id_token kid not in cached jwks; refetching upstream keys (likely rotation)"
                );
                let fresh = self.refresh_jwks().await?;
                let verifier = self.id_token_verifier(fresh)?;
                id_token
                    .claims(&verifier, nonce)
                    .map_err(|e| Error::Oidc(format!("id_token verify: {e}")))
            }
            Err(e) => Err(Error::Oidc(format!("id_token verify: {e}"))),
        }
    }

    /// Build an id_token verifier from `keys`, mirroring the
    /// public/confidential-client choice and allowed-algorithms set that
    /// [`CoreClient::id_token_verifier`] would have produced from the
    /// discovery metadata.
    fn id_token_verifier(&self, keys: Vec<CoreJsonWebKey>) -> Result<CoreIdTokenVerifier<'static>> {
        let issuer = IssuerUrl::new(self.provider.issuer.clone())
            .map_err(|e| Error::Oidc(format!("issuer url {}: {e}", self.provider.issuer)))?;
        let client_id = ClientId::new(self.provider.client_id.clone());
        let jwks = CoreJsonWebKeySet::new(keys);
        let verifier = if self.provider.client_secret.is_empty() {
            CoreIdTokenVerifier::new_public_client(client_id, issuer, jwks)
        } else {
            CoreIdTokenVerifier::new_confidential_client(
                client_id,
                ClientSecret::new(self.provider.client_secret.clone()),
                issuer,
                jwks,
            )
        };
        Ok(verifier.set_allowed_algs(self.signing_algs.clone()))
    }

    /// Whether enough time has elapsed since the last fetch to permit a
    /// reactive (unknown-`kid`) refetch.
    fn reactive_refetch_allowed(&self) -> bool {
        self.jwks.read().last_fetch.elapsed() >= JWKS_MIN_REFETCH_INTERVAL
    }

    /// Re-fetch the upstream JWKS, store it (with a TTL derived from the
    /// certs endpoint's `Cache-Control: max-age`), and return the fresh
    /// keys.
    async fn refresh_jwks(&self) -> Result<Vec<CoreJsonWebKey>> {
        let uri = self.jwks_uri.url().to_string();
        let resp = self
            .jwks_http
            .get(&uri)
            .send()
            .await
            .map_err(|e| Error::Oidc(format!("fetch jwks {uri}: {e}")))?
            .error_for_status()
            .map_err(|e| Error::Oidc(format!("fetch jwks {uri}: {e}")))?;
        let ttl = jwks_cache_ttl(resp.headers());
        let fetched: CoreJsonWebKeySet = resp
            .json()
            .await
            .map_err(|e| Error::Oidc(format!("parse jwks {uri}: {e}")))?;
        let keys = fetched.keys().clone();

        let now = Instant::now();
        {
            let mut cache = self.jwks.write();
            cache.keys = keys.clone();
            cache.expires_at = now + ttl;
            cache.last_fetch = now;
        }
        tracing::debug!(
            slug = %self.provider.slug,
            ttl_secs = ttl.as_secs(),
            keys = keys.len(),
            "refreshed upstream jwks"
        );
        Ok(keys)
    }
}

/// Result of [`OidcClient::start_login`]. The HTTP layer redirects the
/// user to `url` and persists the rest server-side (typically in the
/// session payload, keyed by `csrf_token` so the callback can look the
/// in-progress login up via the `state` query param).
pub struct StartedLogin {
    pub url: Url,
    pub csrf_token: CsrfToken,
    pub nonce: Nonce,
    pub pkce_verifier: PkceCodeVerifier,
}

/// Slug-keyed registry of discovered upstreams.
///
/// Cheap to clone — interior is `Arc<RwLock<…>>` so HTTP handlers can
/// share a single registry while admin endpoints add / remove providers
/// at runtime.
#[derive(Clone, Default)]
pub struct OidcRegistry {
    inner: Arc<RwLock<HashMap<String, Arc<OidcClient>>>>,
}

impl OidcRegistry {
    /// Empty registry — engine boot creates one of these and feeds it to
    /// [`crate::ctx::AuthCtx::with_oidc`]. Providers are added later via
    /// admin CRUD or seed config.
    pub fn new() -> Self {
        Self::default()
    }

    /// Discover and cache one upstream. Performs a network round-trip to
    /// `<issuer>/.well-known/openid-configuration` plus the JWKS fetch,
    /// so call this from boot or from an admin endpoint, not from a
    /// per-request handler.
    ///
    /// `redirect_uri` is the absolute URL the upstream redirects back
    /// to after login (typically `<public_url>/login/<slug>/callback`).
    pub async fn add(&self, provider: UpstreamProvider, redirect_uri: Url) -> Result<()> {
        let issuer = IssuerUrl::new(provider.issuer.clone())
            .map_err(|e| Error::Oidc(format!("issuer url {}: {e}", provider.issuer)))?;
        let http = build_oidc_http_client(HttpClientOptions::default())?;
        let metadata = CoreProviderMetadata::discover_async(issuer, &http)
            .await
            .map_err(|e| Error::Oidc(format!("discover {}: {e}", provider.slug)))?;
        // Capture the bits we need to rebuild an id_token verifier later
        // (on JWKS refresh) before `from_provider_metadata` consumes the
        // metadata. `discover_async` already fetched the JWKS into the
        // metadata, so `initial_keys` is a valid seed for the cache.
        let jwks_uri = metadata.jwks_uri().clone();
        let signing_algs = metadata.id_token_signing_alg_values_supported().clone();
        let initial_keys = metadata.jwks().keys().clone();
        let redirect = RedirectUrl::new(redirect_uri.to_string())
            .map_err(|e| Error::Oidc(format!("redirect_uri {redirect_uri}: {e}")))?;
        let client_secret = if provider.client_secret.is_empty() {
            None
        } else {
            Some(ClientSecret::new(provider.client_secret.clone()))
        };
        let inner = CoreClient::from_provider_metadata(
            metadata,
            ClientId::new(provider.client_id.clone()),
            client_secret,
        )
        .set_redirect_uri(redirect.clone());
        let jwks_http = reqwest::Client::builder()
            .timeout(Duration::from_secs(10))
            .build()
            .map_err(|e| Error::Oidc(format!("build jwks http client: {e}")))?;
        let now = Instant::now();
        let client = OidcClient {
            inner,
            provider: provider.clone(),
            redirect_uri: redirect,
            jwks_uri,
            signing_algs,
            jwks: Arc::new(RwLock::new(JwksCache {
                keys: initial_keys,
                expires_at: now + JWKS_DEFAULT_TTL,
                last_fetch: now,
            })),
            jwks_http,
        };
        self.inner
            .write()
            .insert(provider.slug.clone(), Arc::new(client));
        Ok(())
    }

    /// Look up a discovered upstream by slug. Returns the same `Arc`
    /// stored at registration time so callers can hold the client for
    /// the duration of a long-running flow.
    pub fn client(&self, slug: &str) -> Option<Arc<OidcClient>> {
        self.inner.read().get(slug).cloned()
    }

    /// List the slugs of every registered provider (for admin /
    /// debugging UIs).
    pub fn slugs(&self) -> Vec<String> {
        self.inner.read().keys().cloned().collect()
    }

    /// Remove a provider from the registry. Returns `true` if a row was
    /// dropped. Pending in-flight logins keep working because they hold
    /// an `Arc<OidcClient>` from before the removal.
    pub fn remove(&self, slug: &str) -> bool {
        self.inner.write().remove(slug).is_some()
    }

    /// Number of registered providers — handy for tests + metrics.
    pub fn len(&self) -> usize {
        self.inner.read().len()
    }

    /// Whether the registry is empty.
    pub fn is_empty(&self) -> bool {
        self.inner.read().is_empty()
    }
}

/// Tunables for the discovery / token / userinfo HTTP client.
///
/// Defaults: 5s connect timeout, 10s overall request timeout.
#[derive(Clone, Debug)]
pub struct HttpClientOptions {
    pub connect_timeout_secs: u64,
    pub request_timeout_secs: u64,
}

impl Default for HttpClientOptions {
    fn default() -> Self {
        // TODO: plumb via [auth.oidc] config (discovery_connect_timeout_secs,
        //       discovery_request_timeout_secs).
        Self {
            connect_timeout_secs: 5,
            request_timeout_secs: 10,
        }
    }
}

/// Build the reqwest client `openidconnect` uses for discovery, token
/// exchange, JWKS fetches, and userinfo. We disable redirects on the
/// security advice in the [`openidconnect`] crate docs (SSRF mitigation)
/// and use rustls — matches the rest of assay's HTTP stack.
///
/// Plumbs explicit connect/read timeouts so a stored `issuer` pointing
/// at a hung endpoint can't pin the engine on boot or admin upsert.
/// SSRF protection is handled at the literal-host layer in
/// [`crate::oidc_provider::issuer_validation`].
pub fn build_oidc_http_client(opts: HttpClientOptions) -> Result<oidc_reqwest::Client> {
    oidc_reqwest::ClientBuilder::new()
        .redirect(oidc_reqwest::redirect::Policy::none())
        .connect_timeout(Duration::from_secs(opts.connect_timeout_secs))
        .timeout(Duration::from_secs(opts.request_timeout_secs))
        .build()
        .map_err(|e| Error::Oidc(format!("build oidc http client: {e}")))
}

/// TTL for a freshly-fetched JWKS, parsed from `Cache-Control: max-age`
/// and clamped to `[JWKS_MIN_TTL, JWKS_MAX_TTL]`. Falls back to
/// [`JWKS_DEFAULT_TTL`] when no usable directive is present.
fn jwks_cache_ttl(headers: &reqwest::header::HeaderMap) -> Duration {
    headers
        .get(reqwest::header::CACHE_CONTROL)
        .and_then(|v| v.to_str().ok())
        .and_then(parse_max_age_secs)
        .map(|secs| Duration::from_secs(secs).clamp(JWKS_MIN_TTL, JWKS_MAX_TTL))
        .unwrap_or(JWKS_DEFAULT_TTL)
}

/// Pull the `max-age` value (in seconds) out of a `Cache-Control`
/// header. Returns `None` if there's no `max-age` directive (e.g.
/// `s-maxage` only, or `no-cache`).
fn parse_max_age_secs(cache_control: &str) -> Option<u64> {
    cache_control.split(',').find_map(|directive| {
        directive
            .trim()
            .strip_prefix("max-age")?
            .trim_start()
            .strip_prefix('=')?
            .trim()
            .parse::<u64>()
            .ok()
    })
}

/// Whether an id_token verification error means the signing key wasn't
/// found — i.e. the token's `kid` isn't in our cached JWKS, the
/// fingerprint of upstream key rotation. Other verification failures
/// (bad audience, expired, genuinely bad signature) are *not* fixable by
/// re-fetching keys, so we don't retry on them.
fn is_unknown_signing_key(e: &ClaimsVerificationError) -> bool {
    matches!(
        e,
        ClaimsVerificationError::SignatureVerification(SignatureVerificationError::NoMatchingKey)
    )
}

/// Recursive merge of two JSON values — used so userinfo claims top up
/// the id_token claims without overwriting them. Object fields merge
/// recursively; everything else is replaced.
fn merge_json(target: &mut serde_json::Value, src: serde_json::Value) {
    match (target, src) {
        (serde_json::Value::Object(a), serde_json::Value::Object(b)) => {
            for (k, v) in b {
                merge_json(a.entry(k).or_insert(serde_json::Value::Null), v);
            }
        }
        (slot, src) => {
            // Don't clobber a non-null target with a null source — the
            // id_token's value wins when userinfo doesn't add anything.
            if !src.is_null() {
                *slot = src;
            }
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn registry_starts_empty() {
        let reg = OidcRegistry::new();
        assert!(reg.is_empty());
        assert_eq!(reg.len(), 0);
        assert!(reg.client("google").is_none());
        assert!(reg.slugs().is_empty());
    }

    #[test]
    fn merge_json_merges_objects_and_keeps_existing_on_null() {
        let mut a = serde_json::json!({"email": "a@x", "groups": ["a"]});
        let b = serde_json::json!({"email": serde_json::Value::Null, "name": "Alice"});
        merge_json(&mut a, b);
        assert_eq!(a["email"], "a@x");
        assert_eq!(a["name"], "Alice");
        assert_eq!(a["groups"], serde_json::json!(["a"]));
    }

    #[test]
    fn upstream_provider_record_is_clonable() {
        let p = UpstreamProvider {
            slug: "google".to_string(),
            issuer: "https://accounts.google.com".to_string(),
            client_id: "client".to_string(),
            client_secret: "secret".to_string(),
            scopes: vec!["openid".to_string(), "email".to_string()],
            auth_params: BTreeMap::new(),
        };
        let dup = p.clone();
        assert_eq!(p, dup);
    }

    #[test]
    fn parses_max_age_from_cache_control() {
        assert_eq!(parse_max_age_secs("max-age=3600"), Some(3600));
        assert_eq!(
            parse_max_age_secs("public, max-age=600, must-revalidate"),
            Some(600)
        );
        assert_eq!(parse_max_age_secs("max-age = 42"), Some(42));
        assert_eq!(parse_max_age_secs("private, no-cache"), None);
        // `s-maxage` is a distinct directive — must not be mistaken for
        // `max-age`.
        assert_eq!(parse_max_age_secs("s-maxage=120"), None);
    }

    #[test]
    fn jwks_ttl_clamps_and_defaults() {
        use reqwest::header::{CACHE_CONTROL, HeaderMap, HeaderValue};

        // No header → default.
        assert_eq!(jwks_cache_ttl(&HeaderMap::new()), JWKS_DEFAULT_TTL);

        let ttl = |v: &'static str| {
            let mut h = HeaderMap::new();
            h.insert(CACHE_CONTROL, HeaderValue::from_static(v));
            jwks_cache_ttl(&h)
        };
        // Below floor → clamped up; above ceiling → clamped down.
        assert_eq!(ttl("max-age=10"), JWKS_MIN_TTL);
        assert_eq!(ttl("public, max-age=999999"), JWKS_MAX_TTL);
        // In range → passthrough.
        assert_eq!(ttl("max-age=1800"), Duration::from_secs(1800));
    }

    #[test]
    fn only_missing_key_triggers_refetch() {
        // A missing `kid` (rotation) is retryable...
        assert!(is_unknown_signing_key(
            &ClaimsVerificationError::SignatureVerification(
                SignatureVerificationError::NoMatchingKey
            )
        ));
        // ...but a genuine bad signature or a non-signature failure is
        // not — re-fetching keys wouldn't help.
        assert!(!is_unknown_signing_key(
            &ClaimsVerificationError::SignatureVerification(
                SignatureVerificationError::CryptoError("bad sig".into())
            )
        ));
        assert!(!is_unknown_signing_key(&ClaimsVerificationError::Expired(
            "stale".into()
        )));
    }

    /// Discovery against an unreachable URL should fail with `Error::Oidc`,
    /// not panic. We don't network out from unit tests; this just exercises
    /// the error path.
    #[tokio::test]
    async fn discover_unreachable_issuer_returns_oidc_error() {
        let reg = OidcRegistry::new();
        let provider = UpstreamProvider {
            slug: "ghost".to_string(),
            issuer: "http://127.0.0.1:1/oidc".to_string(),
            client_id: "client".to_string(),
            client_secret: "secret".to_string(),
            scopes: vec!["openid".to_string()],
            auth_params: BTreeMap::new(),
        };
        let redirect = Url::parse("https://example.com/login/ghost/callback").unwrap();
        let result = reg.add(provider, redirect).await;
        assert!(matches!(result, Err(Error::Oidc(_))));
    }
}