Lukas
f62785aaf2
This commit implements fixes for three independent bugs in the fail2ban configuration and integration layer: 1. Task 1: Detect UnknownJailException and prevent silent failures - Added JailNotFoundError detection in jail_service.reload_all() - Enhanced error handling in config_file_service to catch JailNotFoundError - Added specific error message with logpath validation hints - Added rollback test for this scenario 2. Task 2: Fix iptables-allports exit code 4 (xtables lock contention) - Added global banaction setting in jail.conf with -w 5 lockingopt - Removed redundant per-jail banaction overrides from bangui-sim and blocklist-import - Added production compose documentation note 3. Task 3: Suppress log noise from unsupported backend/idle commands - Implemented capability detection to cache command support status - Double-check locking to minimize lock contention - Avoids sending unsupported get <jail> backend/idle commands - Returns default values without socket calls when unsupported All changes include comprehensive tests and maintain backward compatibility.
184 lines
13 KiB
Markdown
184 lines
13 KiB
Markdown
# BanGUI — Task List
|
|
|
|
This document breaks the entire BanGUI project into development stages, ordered so that each stage builds on the previous one. Every task is described in prose with enough detail for a developer to begin work. References point to the relevant documentation.
|
|
|
|
---
|
|
|
|
## Agent Operating Instructions
|
|
|
|
These instructions apply to every AI agent working in this repository. Read them fully before touching any file.
|
|
|
|
### Repository Layout
|
|
|
|
```
|
|
backend/app/ FastAPI application (Python 3.12+, async)
|
|
models/ Pydantic v2 models
|
|
repositories/ Database access (aiosqlite)
|
|
routers/ FastAPI routers — thin, delegate to services
|
|
services/ Business logic; all fail2ban interaction lives here
|
|
tasks/ APScheduler background jobs
|
|
utils/ Shared helpers (fail2ban_client.py, ip_utils.py, …)
|
|
backend/tests/ pytest-asyncio test suite mirroring app/ structure
|
|
Docker/ Compose files, Dockerfiles, dev config for fail2ban
|
|
fail2ban-dev-config/ Bind-mounted into the fail2ban container in debug mode
|
|
frontend/src/ React + TypeScript SPA (Vite, Fluent UI)
|
|
fail2ban-master/ Vendored fail2ban source — DO NOT EDIT
|
|
Docs/ Architecture, design notes, this file
|
|
```
|
|
|
|
### Coding Conventions
|
|
|
|
- **Python**: async/await throughout. Use `structlog` for logging (`log.info/warning/error` with keyword args). Pydantic v2 models. Type-annotated functions.
|
|
- **Error handling**: Raise domain-specific exceptions (`JailNotFoundError`, `JailOperationError`, `Fail2BanConnectionError`) from service functions; let routers map them to HTTP responses. Never swallow exceptions silently in routers.
|
|
- **fail2ban socket**: All communication goes through `app.utils.fail2ban_client.Fail2BanClient`. Use `_safe_get` when a missing command is non-fatal; use `_ok()` when the response is required.
|
|
- **Tests**: Mirror the `app/` tree under `tests/`. Use `pytest-asyncio` + `unittest.mock.patch` or `AsyncMock`. Keep fixtures in `conftest.py`. Run with `pytest backend/tests` from the repo root (venv at `.venv/`).
|
|
- **Docker dev config**: `Docker/fail2ban-dev-config/` is bind-mounted read-write into the fail2ban container. Changes here take effect after `fail2ban-client reload`.
|
|
- **Compose**: Use `Docker/compose.debug.yml` for local development; `Docker/compose.prod.yml` for production. Never commit secrets.
|
|
|
|
### How to Verify Changes
|
|
|
|
1. Run the backend test suite: `cd /home/lukas/Volume/repo/BanGUI && .venv/bin/pytest backend/tests -x -q`
|
|
2. Check for Python type errors: `.venv/bin/mypy backend/app` (if mypy is installed).
|
|
3. For runtime verification, start the debug stack: `docker compose -f Docker/compose.debug.yml up`.
|
|
4. Inspect fail2ban logs inside the container: `docker exec bangui-fail2ban-dev fail2ban-client status` and `docker logs bangui-fail2ban-dev`.
|
|
|
|
### Agent Workflow
|
|
|
|
1. Read the task description fully. Read every source file mentioned before editing anything.
|
|
2. Make the minimal change that solves the stated problem — do not refactor surrounding code.
|
|
3. After every code change run the test suite to confirm no regressions.
|
|
4. If a fix requires a config file change in `Docker/fail2ban-dev-config/`, also update `Docker/compose.prod.yml` or document the required change if it affects the production config volume.
|
|
5. Mark the task complete only after tests pass and the root-cause error no longer appears.
|
|
|
|
---
|
|
|
|
## Bug Fixes — fail2ban Runtime Errors (2026-03-14)
|
|
|
|
The following three independent bugs were identified from fail2ban logs. They are ordered by severity. Resolve them in order; each is self-contained.
|
|
|
|
---
|
|
|
|
### Task 1 — `UnknownJailException('airsonic-auth')` on reload
|
|
|
|
**Priority**: High
|
|
**Files**: `Docker/fail2ban-dev-config/fail2ban/jail.d/airsonic-auth.conf`, `backend/app/services/config_file_service.py`
|
|
|
|
#### Observed error
|
|
|
|
```
|
|
fail2ban.transmitter ERROR Command ['reload', '--all', [], [['start', 'airsonic-auth'],
|
|
['start', 'bangui-sim'], ['start', 'blocklist-import']]] has failed.
|
|
Received UnknownJailException('airsonic-auth')
|
|
```
|
|
|
|
#### Root cause
|
|
|
|
When a user activates the `airsonic-auth` jail through BanGUI, `config_file_service.py` writes a local override and then calls `jail_service.reload_all(socket_path, include_jails=['airsonic-auth'])`. The reload stream therefore contains `['start', 'airsonic-auth']`. fail2ban cannot create the jail object because its `logpath` (`/remotelogs/airsonic/airsonic.log`) does not exist in the dev environment — the airsonic service is not running and no log volume is mounted. fail2ban registers a config-level parse failure for the jail and the server falls back to `UnknownJailException` when the reload asks it to start that name.
|
|
|
|
The current error handling in `config_file_service.py` catches the reload exception and rolls back the config file, but it does so with a generic `except Exception` and logs only a warning. The real error is never surfaced to the user in a way that explains why activation failed.
|
|
|
|
#### What to implement
|
|
|
|
1. **Dev-config fix** (Docker/fail2ban-dev-config): The `airsonic-auth.conf` file is a sample jail showing how a remote-log jail is configured. Add a prominent comment block at the top explaining that this jail is intentionally `enabled = false` and that it will fail to start unless the `/remotelogs/airsonic/` log directory is mounted into the fail2ban container. This makes the intent explicit and prevents confusion.
|
|
|
|
2. **Logpath pre-validation in the activation service** (`config_file_service.py`): Before calling `reload_all` during jail activation, read the `logpath` values from the jail's config (parse the `.conf` and any `.local` override using the existing `conffile_parser`). For each logpath that is not `/dev/null` and does not contain a glob wildcard, check whether the path exists on the filesystem (the backend container shares the fail2ban config volume with read-write access). If any required logpath is missing, abort the activation immediately — do not write the local override, do not call reload — and return a `JailActivationResponse` with `active=False` and a clear `message` explaining which logpath is missing. Add a `validation_warnings` entry listing the missing paths.
|
|
|
|
3. **Specific exception detection** (`jail_service.py`): In `reload_all`, when `_ok()` raises `ValueError` and the message matches `unknownjail` / `unknown jail` (use the existing `_is_not_found_error` helper), re-raise a `JailNotFoundError` instead of the generic `JailOperationError`. Update callers in `config_file_service.py` to catch `JailNotFoundError` separately and include the jail name in the activation failure message.
|
|
|
|
#### Acceptance criteria
|
|
|
|
- Attempting to activate `airsonic-auth` (without mounting the log volume) returns a 422-class response with a message mentioning the missing logpath — no `UnknownJailException` appears in the fail2ban log.
|
|
- Activating `bangui-sim` (whose logpath exists in dev) continues to work correctly.
|
|
- All existing tests in `backend/tests/` pass without modification.
|
|
|
|
---
|
|
|
|
### Task 2 — `iptables-allports` action fails with exit code 4 (`Script error`) on `blocklist-import` jail
|
|
|
|
**Priority**: High
|
|
**Files**: `Docker/fail2ban-dev-config/fail2ban/jail.d/blocklist-import.conf`, `Docker/fail2ban-dev-config/fail2ban/jail.d/bangui-sim.conf`, `Docker/fail2ban-dev-config/fail2ban/jail.conf`
|
|
|
|
#### Observed error
|
|
|
|
```
|
|
fail2ban.utils ERROR 753c588a7860 -- returned 4
|
|
fail2ban.actions ERROR Failed to execute ban jail 'blocklist-import' action
|
|
'iptables-allports' … Error starting action
|
|
Jail('blocklist-import')/iptables-allports: 'Script error'
|
|
```
|
|
|
|
#### Root cause
|
|
|
|
Exit code 4 from an iptables invocation means the xtables advisory lock could not be obtained (another iptables call was in progress simultaneously). The `iptables-allports` `actionstart` script does not pass the `-w` (wait-for-lock) flag by default. In the Docker dev environment, fail2ban starts multiple jails in parallel during `reload --all`. Each jail's `actionstart` sub-process calls iptables concurrently. The xtables lock contention causes some of them to exit with code 4. fail2ban interprets any non-zero exit from an action script as `Script error` and aborts the action.
|
|
|
|
A secondary risk: if the host kernel uses `nf_tables` (iptables-nft) but the container runs `iptables-legacy` binaries, the same exit code 4 can also appear due to backend mismatch.
|
|
|
|
#### What to implement
|
|
|
|
1. **Add the xtables lock-wait flag globally** in `Docker/fail2ban-dev-config/fail2ban/jail.conf`, under the `[DEFAULT]` section. Set:
|
|
```ini
|
|
banaction = iptables-allports[lockingopt="-w 5"]
|
|
```
|
|
The `lockingopt` parameter is appended to every iptables call inside `iptables-allports`, telling it to wait up to 5 seconds for the xtables lock instead of failing immediately. This is the canonical fix recommended in the fail2ban documentation for containerised deployments.
|
|
|
|
2. **Remove the redundant `banaction` overrides** from `blocklist-import.conf` and `bangui-sim.conf` (both already specify `banaction = iptables-allports` which now comes from the global default). Removing the per-jail override keeps configuration DRY and ensures the lock-wait flag applies consistently.
|
|
|
|
3. **Production compose note**: Add a comment in `Docker/compose.prod.yml` near the fail2ban service block stating that the `fail2ban-config` volume must contain a `jail.d/` with `banaction = iptables-allports[lockingopt="-w 5"]` or equivalent, because the production volume is not pre-seeded by the repository. This is a documentation action only — do not auto-seed the production volume.
|
|
|
|
#### Acceptance criteria
|
|
|
|
- After `docker compose -f Docker/compose.debug.yml restart fail2ban`, the fail2ban log shows no `returned 4` or `Script error` lines during startup.
|
|
- `docker exec bangui-fail2ban-dev fail2ban-client status blocklist-import` reports the jail as running.
|
|
- `docker exec bangui-fail2ban-dev fail2ban-client status bangui-sim` reports the jail as running.
|
|
|
|
---
|
|
|
|
### Task 3 — Suppress log noise from unsupported `get <jail> idle/backend` commands
|
|
|
|
**Priority**: Low
|
|
**Files**: `backend/app/services/jail_service.py`
|
|
|
|
#### Observed error
|
|
|
|
```
|
|
fail2ban.transmitter ERROR Command ['get', 'bangui-sim', 'idle'] has failed.
|
|
Received Exception('Invalid command (no get action or not yet implemented)')
|
|
fail2ban.transmitter ERROR Command ['get', 'bangui-sim', 'backend'] has failed.
|
|
Received Exception('Invalid command (no get action or not yet implemented)')
|
|
… (repeated ~15 times per polling cycle)
|
|
```
|
|
|
|
#### Root cause
|
|
|
|
`_fetch_jail_summary()` in `jail_service.py` sends `["get", name, "backend"]` and `["get", name, "idle"]` to the fail2ban daemon via the socket. The running fail2ban version (LinuxServer.io container image) does not implement these two sub-commands in its transmitter. fail2ban logs every unrecognised command as an `ERROR` on its side. BanGUI already handles this gracefully — `asyncio.gather(return_exceptions=True)` captures the exceptions and `_safe_bool` / `_safe_str` fall back to `False` / `"polling"` respectively. The BanGUI application does not malfunction, but the fail2ban log is flooded with spurious `ERROR` lines on every jail-list refresh (~15 errors per request across all jails, repeated on every dashboard poll).
|
|
|
|
#### What to implement
|
|
|
|
The fix is a capability-detection probe executed once at startup (or at most once per polling cycle, re-used for all jails). The probe sends `["get", "<first_jail>", "backend"]` and caches the result in a module-level boolean flag.
|
|
|
|
1. In `jail_service.py`, add a module-level `asyncio.Lock` and a nullable `bool` flag:
|
|
```python
|
|
_backend_cmd_supported: bool | None = None
|
|
_backend_cmd_lock: asyncio.Lock = asyncio.Lock()
|
|
```
|
|
|
|
2. Add an async helper `_check_backend_cmd_supported(client, jail_name) -> bool` that:
|
|
- Returns the cached value immediately if already determined.
|
|
- Acquires `_backend_cmd_lock`, checks again (double-check idiom), then sends `["get", jail_name, "backend"]`.
|
|
- Sets `_backend_cmd_supported = True` if the command returns without exception, `False` if it raises any `Exception`.
|
|
- Returns the flag value.
|
|
|
|
3. In `_fetch_jail_summary()`, call this helper (using the first available jail name) before the main `asyncio.gather`. If the flag is `False`, replace the `["get", name, "backend"]` and `["get", name, "idle"]` gather entries with `asyncio.coroutine` constants that immediately return the default values (`"polling"` and `False`) without sending any socket command. This eliminates the fail2ban-side log noise entirely without changing the public API or the fallback values.
|
|
|
|
4. If `_backend_cmd_supported` becomes `True` (future fail2ban version), the commands are sent as before and real values are returned.
|
|
|
|
#### Acceptance criteria
|
|
|
|
- After one polling cycle where `backend`/`idle` are detected as unsupported, the fail2ban log shows zero `Invalid command` lines for those commands on subsequent refreshes.
|
|
- The `GET /api/jails` response still returns `backend: "polling"` and `idle: false` for each jail (unchanged defaults).
|
|
- All existing tests in `backend/tests/` pass.
|
|
- The detection result is reset if `_backend_cmd_supported` is `None` (i.e., on application restart), so a fail2ban upgrade that adds support is detected automatically within one polling cycle.
|
|
|
|
---
|
|
|