Files
immich-folio/lib/admin/auth.ts

112 lines
3.8 KiB
TypeScript
Raw Normal View History

/**
* Admin authentication utilities.
* Uses HMAC-signed session tokens stored in HttpOnly cookies.
*/
import crypto from 'crypto';
import { env } from '../env';
security: fix pre-auth bypasses and restore Immich 3.x compatibility (#378) Seven security issues from an audit, plus album pages and the map coming up empty against Immich 3.x. Verified against a real Immich 3.0.3 instance in a production build. Security: - Admin session forgery via a hardcoded 'dev-fallback-secret' signing key that bypassed the production AUTH_SECRET guard - /api/map ignored album passwords and returned working asset tokens; rewriting ?size=thumbnail to ?size=original downloaded full-resolution originals from protected albums - Admin login had no rate limit, and the client IP came from a spoofable header, so rotating it bypassed every limit in the app - Raw Immich asset UUIDs were serialized into the RSC payload - Album tokens never expired server-side; a captured token replayed forever - Path traversal into the Immich API via an unvalidated album id - /admin ran with no enforced CSP (only Report-Only reached it) TRUSTED_PROXIES is replaced by TRUSTED_PROXY_HOPS. The old variable matched proxy IPs against the socket peer address, which self-hosted Next.js never exposes, so the check could never pass: unset, the limiter read a spoofable header; set, every request shared one bucket and a single client could rate-limit the whole site. Migrates automatically. Immich 3.x: GET /albums/:id no longer embeds assets. Assets now come from POST /search/metadata with withExif: true — without that flag EXIF is omitted and grid aspect ratios, the lightbox panel and map markers stay empty. Drops support for Immich below 3.0, which now logs a warning. Upgrade: requires Immich 3.0+; set TRUSTED_PROXY_HOPS (nginx = 1); visitors of protected areas and admin sessions must log in once more. 91 unit tests (25 new); the new admin-auth and rate-limit cases were confirmed to fail against the pre-fix code. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 18:57:57 +02:00
import { resolveAuthSecret } from '../secret';
import { cookies } from 'next/headers';
const COOKIE_NAME = 'folio_admin_session';
const SESSION_DURATION_MS = 24 * 60 * 60 * 1000; // 24 hours
security: stop leaking the admin password length through timing (#395) verifyAdminPassword compared the raw strings, so timingSafeEqual could only run once the lengths already matched. The early return on a length mismatch is observable: an attacker who can time login attempts learns how long ADMIN_PASSWORD is, which is exactly the information that makes a brute-force search tractable. The dummy comparison that guarded the mismatch path did not help — it ran on Buffer.alloc(attempt.length), which is attacker-sized and leaves the branch timing intact anyway. HMAC both sides to a fixed 32 bytes first, then compare. Every attempt now takes the same path regardless of length. Keyed with AUTH_SECRET rather than a bare SHA-256 so the intermediate values are useless without the secret, matching how lib/auth.ts already compares subpage passwords. MAX_PASSWORD_LENGTH caps an attempt at 1024 characters, rejected in the route before the body reaches the hashing path and again in the library so no caller can bypass it. 1024 is far above any real passphrase — a cap a user could actually hit would be a lockout, not a fix. Supersedes #390 and #388, which fixed the same comparison with an unkeyed digest. The .jules/sentinel.md note the original branch carried is left out: .jules/ is gitignored since cf2e8e6. Adds 3 tests: a wrong password of exactly the right length, an oversized attempt, and that a long configured password still logs in. Verified: 210 unit tests passing (207 before), vitest exit 0, tsc 0 errors, production build exit 0, prettier clean. Co-authored-by: ralksta <ralksta@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-05 12:13:32 +02:00
/**
* Longest password attempt the login endpoint will look at.
*
* Generous enough that no real passphrase hits it — the point is only to keep
* an unbounded request body out of the hashing path, not to constrain what
* users may choose.
*/
export const MAX_PASSWORD_LENGTH = 1024;
function getSigningKey(): Buffer {
security: fix pre-auth bypasses and restore Immich 3.x compatibility (#378) Seven security issues from an audit, plus album pages and the map coming up empty against Immich 3.x. Verified against a real Immich 3.0.3 instance in a production build. Security: - Admin session forgery via a hardcoded 'dev-fallback-secret' signing key that bypassed the production AUTH_SECRET guard - /api/map ignored album passwords and returned working asset tokens; rewriting ?size=thumbnail to ?size=original downloaded full-resolution originals from protected albums - Admin login had no rate limit, and the client IP came from a spoofable header, so rotating it bypassed every limit in the app - Raw Immich asset UUIDs were serialized into the RSC payload - Album tokens never expired server-side; a captured token replayed forever - Path traversal into the Immich API via an unvalidated album id - /admin ran with no enforced CSP (only Report-Only reached it) TRUSTED_PROXIES is replaced by TRUSTED_PROXY_HOPS. The old variable matched proxy IPs against the socket peer address, which self-hosted Next.js never exposes, so the check could never pass: unset, the limiter read a spoofable header; set, every request shared one bucket and a single client could rate-limit the whole site. Migrates automatically. Immich 3.x: GET /albums/:id no longer embeds assets. Assets now come from POST /search/metadata with withExif: true — without that flag EXIF is omitted and grid aspect ratios, the lightbox panel and map markers stay empty. Drops support for Immich below 3.0, which now logs a warning. Upgrade: requires Immich 3.0+; set TRUSTED_PROXY_HOPS (nginx = 1); visitors of protected areas and admin sessions must log in once more. 91 unit tests (25 new); the new admin-auth and rate-limit cases were confirmed to fail against the pre-fix code. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 18:57:57 +02:00
// Bind the key to ADMIN_PASSWORD as well, so rotating the password
// immediately invalidates every outstanding session token.
const secret = resolveAuthSecret();
return crypto
.createHash('sha256')
.update(`admin:${secret}:${env.ADMIN_PASSWORD ?? ''}`)
.digest();
}
/** Create a signed session token. */
export function createAdminToken(): string {
const payload = {
role: 'admin',
iat: Date.now(),
exp: Date.now() + SESSION_DURATION_MS,
};
const data = Buffer.from(JSON.stringify(payload)).toString('base64url');
const key = getSigningKey();
const sig = crypto.createHmac('sha256', key).update(data).digest('base64url');
return `${data}.${sig}`;
}
/** Verify a session token. Returns true if valid. */
export function verifyAdminToken(token: string): boolean {
const parts = token.split('.');
if (parts.length !== 2) return false;
const [data, sig] = parts;
const key = getSigningKey();
const expectedSig = crypto.createHmac('sha256', key).update(data).digest('base64url');
security: fix pre-auth bypasses and restore Immich 3.x compatibility (#378) Seven security issues from an audit, plus album pages and the map coming up empty against Immich 3.x. Verified against a real Immich 3.0.3 instance in a production build. Security: - Admin session forgery via a hardcoded 'dev-fallback-secret' signing key that bypassed the production AUTH_SECRET guard - /api/map ignored album passwords and returned working asset tokens; rewriting ?size=thumbnail to ?size=original downloaded full-resolution originals from protected albums - Admin login had no rate limit, and the client IP came from a spoofable header, so rotating it bypassed every limit in the app - Raw Immich asset UUIDs were serialized into the RSC payload - Album tokens never expired server-side; a captured token replayed forever - Path traversal into the Immich API via an unvalidated album id - /admin ran with no enforced CSP (only Report-Only reached it) TRUSTED_PROXIES is replaced by TRUSTED_PROXY_HOPS. The old variable matched proxy IPs against the socket peer address, which self-hosted Next.js never exposes, so the check could never pass: unset, the limiter read a spoofable header; set, every request shared one bucket and a single client could rate-limit the whole site. Migrates automatically. Immich 3.x: GET /albums/:id no longer embeds assets. Assets now come from POST /search/metadata with withExif: true — without that flag EXIF is omitted and grid aspect ratios, the lightbox panel and map markers stay empty. Drops support for Immich below 3.0, which now logs a warning. Upgrade: requires Immich 3.0+; set TRUSTED_PROXY_HOPS (nginx = 1); visitors of protected areas and admin sessions must log in once more. 91 unit tests (25 new); the new admin-auth and rate-limit cases were confirmed to fail against the pre-fix code. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-20 18:57:57 +02:00
// timingSafeEqual throws on length mismatch — guard first so a malformed
// cookie yields a clean 401 instead of an unhandled RangeError (500).
const sigBuf = Buffer.from(sig);
const expectedBuf = Buffer.from(expectedSig);
if (sigBuf.length !== expectedBuf.length) return false;
if (!crypto.timingSafeEqual(sigBuf, expectedBuf)) {
return false;
}
try {
const payload = JSON.parse(Buffer.from(data, 'base64url').toString());
if (payload.role !== 'admin') return false;
if (typeof payload.exp !== 'number' || payload.exp < Date.now()) return false;
return true;
} catch {
return false;
}
}
security: stop leaking the admin password length through timing (#395) verifyAdminPassword compared the raw strings, so timingSafeEqual could only run once the lengths already matched. The early return on a length mismatch is observable: an attacker who can time login attempts learns how long ADMIN_PASSWORD is, which is exactly the information that makes a brute-force search tractable. The dummy comparison that guarded the mismatch path did not help — it ran on Buffer.alloc(attempt.length), which is attacker-sized and leaves the branch timing intact anyway. HMAC both sides to a fixed 32 bytes first, then compare. Every attempt now takes the same path regardless of length. Keyed with AUTH_SECRET rather than a bare SHA-256 so the intermediate values are useless without the secret, matching how lib/auth.ts already compares subpage passwords. MAX_PASSWORD_LENGTH caps an attempt at 1024 characters, rejected in the route before the body reaches the hashing path and again in the library so no caller can bypass it. 1024 is far above any real passphrase — a cap a user could actually hit would be a lockout, not a fix. Supersedes #390 and #388, which fixed the same comparison with an unkeyed digest. The .jules/sentinel.md note the original branch carried is left out: .jules/ is gitignored since cf2e8e6. Adds 3 tests: a wrong password of exactly the right length, an oversized attempt, and that a long configured password still logs in. Verified: 210 unit tests passing (207 before), vitest exit 0, tsc 0 errors, production build exit 0, prettier clean. Co-authored-by: ralksta <ralksta@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-05 12:13:32 +02:00
/**
* Verify admin password.
*
* Both sides are HMAC'd to a fixed 32 bytes before the constant-time compare.
* Comparing the raw strings meant `timingSafeEqual` could only run when the
* lengths already matched, so the early return on a length mismatch leaked how
* long ADMIN_PASSWORD is. Hashing first makes every attempt take the same path.
*
* Keyed with AUTH_SECRET rather than a bare digest, matching how `lib/auth.ts`
* compares subpage passwords: the intermediate values are then useless to
* anyone who cannot also read the secret.
*/
export function verifyAdminPassword(password: string): boolean {
const adminPw = env.ADMIN_PASSWORD;
if (!adminPw) return false;
security: stop leaking the admin password length through timing (#395) verifyAdminPassword compared the raw strings, so timingSafeEqual could only run once the lengths already matched. The early return on a length mismatch is observable: an attacker who can time login attempts learns how long ADMIN_PASSWORD is, which is exactly the information that makes a brute-force search tractable. The dummy comparison that guarded the mismatch path did not help — it ran on Buffer.alloc(attempt.length), which is attacker-sized and leaves the branch timing intact anyway. HMAC both sides to a fixed 32 bytes first, then compare. Every attempt now takes the same path regardless of length. Keyed with AUTH_SECRET rather than a bare SHA-256 so the intermediate values are useless without the secret, matching how lib/auth.ts already compares subpage passwords. MAX_PASSWORD_LENGTH caps an attempt at 1024 characters, rejected in the route before the body reaches the hashing path and again in the library so no caller can bypass it. 1024 is far above any real passphrase — a cap a user could actually hit would be a lockout, not a fix. Supersedes #390 and #388, which fixed the same comparison with an unkeyed digest. The .jules/sentinel.md note the original branch carried is left out: .jules/ is gitignored since cf2e8e6. Adds 3 tests: a wrong password of exactly the right length, an oversized attempt, and that a long configured password still logs in. Verified: 210 unit tests passing (207 before), vitest exit 0, tsc 0 errors, production build exit 0, prettier clean. Co-authored-by: ralksta <ralksta@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-05 12:13:32 +02:00
if (password.length > MAX_PASSWORD_LENGTH) return false;
security: stop leaking the admin password length through timing (#395) verifyAdminPassword compared the raw strings, so timingSafeEqual could only run once the lengths already matched. The early return on a length mismatch is observable: an attacker who can time login attempts learns how long ADMIN_PASSWORD is, which is exactly the information that makes a brute-force search tractable. The dummy comparison that guarded the mismatch path did not help — it ran on Buffer.alloc(attempt.length), which is attacker-sized and leaves the branch timing intact anyway. HMAC both sides to a fixed 32 bytes first, then compare. Every attempt now takes the same path regardless of length. Keyed with AUTH_SECRET rather than a bare SHA-256 so the intermediate values are useless without the secret, matching how lib/auth.ts already compares subpage passwords. MAX_PASSWORD_LENGTH caps an attempt at 1024 characters, rejected in the route before the body reaches the hashing path and again in the library so no caller can bypass it. 1024 is far above any real passphrase — a cap a user could actually hit would be a lockout, not a fix. Supersedes #390 and #388, which fixed the same comparison with an unkeyed digest. The .jules/sentinel.md note the original branch carried is left out: .jules/ is gitignored since cf2e8e6. Adds 3 tests: a wrong password of exactly the right length, an oversized attempt, and that a long configured password still logs in. Verified: 210 unit tests passing (207 before), vitest exit 0, tsc 0 errors, production build exit 0, prettier clean. Co-authored-by: ralksta <ralksta@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-05 12:13:32 +02:00
const secret = resolveAuthSecret();
const attemptHash = crypto.createHmac('sha256', secret).update(password).digest();
const expectedHash = crypto.createHmac('sha256', secret).update(adminPw).digest();
return crypto.timingSafeEqual(attemptHash, expectedHash);
}
/** Check if the current request has a valid admin session. */
export async function isAdminAuthenticated(): Promise<boolean> {
const cookieStore = await cookies();
const token = cookieStore.get(COOKIE_NAME)?.value;
if (!token) return false;
return verifyAdminToken(token);
}
/** Check if admin panel is enabled (password is set). */
export function isAdminEnabled(): boolean {
return !!env.ADMIN_PASSWORD;
}
export { COOKIE_NAME, SESSION_DURATION_MS };