forked from baron/baron-sso
레포 업데이트
This commit is contained in:
146
docs/superpowers/specs/2026-04-01-headless-login-debug-design.md
Normal file
146
docs/superpowers/specs/2026-04-01-headless-login-debug-design.md
Normal file
@@ -0,0 +1,146 @@
|
||||
# Headless Login Debug and Error Classification Design
|
||||
|
||||
## Scope
|
||||
|
||||
This design covers issue `#502` only.
|
||||
|
||||
- Target endpoint: `POST /api/v1/auth/headless/password/login`
|
||||
- Goal: classify 401 failure reasons for headless password login, return a safe response to RP clients, and emit richer diagnostics only when debug logging is enabled
|
||||
- Non-goal: introduce a shared auth-wide error/observability layer across all handlers. That follow-up work is tracked separately in issue `#503`.
|
||||
|
||||
## Problem
|
||||
|
||||
Current behavior makes production debugging slow.
|
||||
|
||||
- Multiple assertion failures collapse into `invalid_client_assertion`
|
||||
- The handler often returns early without emitting a reason-specific log line
|
||||
- RP clients cannot distinguish safe failure categories such as audience mismatch vs signature mismatch
|
||||
- Server operators cannot quickly tell whether the failure came from client assertion validation or user credential validation
|
||||
|
||||
## Constraints
|
||||
|
||||
- Response policy must follow the agreed level `2`: return `code + short safe message`
|
||||
- Debug logs may include level `3`-style diagnostics, but only when the logger is configured for debug level
|
||||
- Sensitive values must never be logged: raw `client_assertion`, password, session token, cookies
|
||||
- Existing successful headless password login flow must remain unchanged
|
||||
- Existing policy requires failing tests first
|
||||
|
||||
## Recommended Approach
|
||||
|
||||
Introduce a small headless-password-specific failure classification layer inside `auth_handler.go`.
|
||||
|
||||
The classification layer should map each failure into:
|
||||
|
||||
- `status`
|
||||
- `code`
|
||||
- `safeMessage`
|
||||
- `logMessage`
|
||||
- `debugFields`
|
||||
|
||||
This stays local to the current handler so the change remains small, but the structure should be reusable enough to extract later in issue `#503`.
|
||||
|
||||
## Failure Categories
|
||||
|
||||
### Client assertion failures
|
||||
|
||||
- `invalid_client_assertion_parse`
|
||||
- `invalid_client_assertion_signature`
|
||||
- `invalid_client_assertion_iss_sub`
|
||||
- `invalid_client_assertion_expired`
|
||||
- `invalid_client_assertion_not_before`
|
||||
- `invalid_client_assertion_iat_future`
|
||||
- `invalid_client_assertion_audience`
|
||||
- `invalid_client_assertion_jwks_load`
|
||||
|
||||
### Credential failure
|
||||
|
||||
- `password_or_email_mismatch`
|
||||
|
||||
## Response Contract
|
||||
|
||||
Responses should expose only safe details.
|
||||
|
||||
Examples:
|
||||
|
||||
- `{"code":"invalid_client_assertion_audience","error":"Client assertion audience mismatch"}`
|
||||
- `{"code":"invalid_client_assertion_signature","error":"Client assertion signature verification failed"}`
|
||||
- `{"code":"password_or_email_mismatch","error":"Invalid credentials"}`
|
||||
|
||||
The response must not echo:
|
||||
|
||||
- expected vs received audience values
|
||||
- full claim payload
|
||||
- kid mismatch internals beyond the safe message
|
||||
- raw token values
|
||||
|
||||
## Logging Strategy
|
||||
|
||||
### Normal log level
|
||||
|
||||
Emit one structured warning/error log per classified failure with:
|
||||
|
||||
- `reason_code`
|
||||
- `client_id`
|
||||
- `path`
|
||||
- `req_id` if present in context/log pipeline
|
||||
|
||||
Message examples:
|
||||
|
||||
- `headless password login client assertion failed`
|
||||
- `headless password login credential authentication failed`
|
||||
|
||||
### Debug log level
|
||||
|
||||
Add diagnostic fields that are useful for root cause analysis:
|
||||
|
||||
- `expected_audiences`
|
||||
- `received_audiences`
|
||||
- `received_kid`
|
||||
- `jwks_refreshed`
|
||||
- `claim_issuer`
|
||||
- `claim_subject`
|
||||
- `claim_expires_at`
|
||||
- `claim_not_before`
|
||||
- `claim_issued_at`
|
||||
- `login_challenge_prefix`
|
||||
|
||||
The `login_challenge` should be truncated before logging.
|
||||
|
||||
## Implementation Shape
|
||||
|
||||
Inside `backend/internal/handler/auth_handler.go`:
|
||||
|
||||
1. Add a small internal type for classified headless assertion errors.
|
||||
2. Make `validateHeadlessClientAssertionClaims` return structured reasons instead of generic strings.
|
||||
3. Make `verifyHeadlessClientAssertion` return a classified failure that the handler can both:
|
||||
- convert to a safe HTTP response
|
||||
- log with optional debug fields
|
||||
4. Add a small helper that checks whether debug logging is enabled for the current default `slog` logger.
|
||||
5. Log credential authentication failures with their reason code as well.
|
||||
|
||||
## Testing
|
||||
|
||||
Add failing tests first in `backend/internal/handler/auth_handler_login_test.go`.
|
||||
|
||||
Required tests:
|
||||
|
||||
- audience mismatch returns `401 + invalid_client_assertion_audience`
|
||||
- signature mismatch returns `401 + invalid_client_assertion_signature`
|
||||
- iss/sub mismatch returns `401 + invalid_client_assertion_iss_sub`
|
||||
- expired assertion returns `401 + invalid_client_assertion_expired`
|
||||
- invalid credentials still returns `401 + password_or_email_mismatch`
|
||||
- debug logger captures diagnostic fields for assertion audience mismatch
|
||||
- non-debug logger does not emit debug-only fields
|
||||
|
||||
## Risks
|
||||
|
||||
- Over-classifying too much detail into RP responses could create unnecessary information disclosure
|
||||
- Logging helper implementation must not assume a specific handler type beyond standard `slog`
|
||||
- Existing tests may assume the older generic `invalid_client_assertion` code and will need intentional updates
|
||||
|
||||
## Success Criteria
|
||||
|
||||
- Operators can distinguish signature, claims, and credential failures from logs
|
||||
- RP clients get safe but actionable reason codes
|
||||
- Debug mode includes additional diagnostics without leaking secrets
|
||||
- Existing success path and previously covered headless login behavior remain green
|
||||
Reference in New Issue
Block a user