monoadmin/the-other-dude
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(ci): use TRUNCATE CASCADE for test cleanup, remove superpowers docs

Browse Source 
- TRUNCATE CASCADE reliably cleans all test data regardless of FK order
- Remove docs/superpowers/ from git tracking (already in .gitignore)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Jason Staack
2026-03-15 06:27:34 -05:00
parent 93138f0483
commit 9085d90b93
5 changed files with 4 additions and 2077 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1196
docs/superpowers/plans/2026-03-14-setup-script.md
View File

File diff suppressed because it is too large Load Diff

346
docs/superpowers/specs/2026-03-14-saas-tiers-design.md
View File

@@ -1,346 +0,0 @@
# SaaS Tiers, Invite System & Plan Limits — Design Spec
## Overview
Add a tier/quota system to TOD that limits tenants, devices, and users per account. Includes a one-time invite system for onboarding new users with configurable plan limits, and a disabled-by-default public signup page for future homelab tier.
**Branch:** `saas-tiers` (diverges from open-source `main`)
**Target audience:** Super_admin (you) managing a SaaS offering, and invited users who manage their own tenants.
## Design Decisions
- **No new "Account" model** — the existing user model is extended. The "account owner" is just the user who accepted an invite and creates tenants.
- **Per-user plan limits** — a `plan_limits` table keyed by user ID stores max_tenants, max_devices_per_tenant, max_users_per_tenant.
- **No limits row = no limits** — super_admin users never have a `plan_limits` row, so they're unlimited.
- **Invite-based onboarding** — super_admin generates one-time invite links (32 bytes / 256 bits entropy). No self-registration yet (homelab signup page exists but is disabled by default).
- **Existing RBAC preserved** — invited users become `tenant_admin` of tenants they create. No new roles.
- **`user_tenants` join table** — tracks which tenants a user belongs to and their role in each. The user's `tenant_id` column becomes the "currently active" tenant. Tenant switching updates this field and re-issues the JWT.
- **Enforcement at creation time** — limits are checked when creating devices, users, or tenants. Not on every request.
- **Invited users start as bcrypt** — with `must_upgrade_auth=True`. SRP upgrade happens on first login, consistent with existing user creation flow.
## Data Model
### New Tables
#### `plan_limits`
Per-user quota configuration. If no row exists for a user, no limits are enforced (super_admin behavior).
| Column | Type | Default | Description |
|--------|------|---------|-------------|
| `id` | UUID PK | gen_random_uuid() | Primary key |
| `user_id` | UUID FK (users.id), unique | — | The account owner |
| `max_tenants` | integer | 2 | Max tenants this user can own |
| `max_devices_per_tenant` | integer | 10 | Max devices per tenant |
| `max_users_per_tenant` | integer | 10 | Max users per tenant (0 = owner only) |
| `plan_name` | varchar(50) | "invite" | Plan identifier: "invite", "homelab", "custom" |
| `created_at` | timestamptz | now() | |
| `updated_at` | timestamptz | now() | |
RLS policy: super_admin can read/write all rows. Users can read their own row.
#### `user_tenants`
Join table tracking which tenants a user belongs to and their role in each.
| Column | Type | Default | Description |
|--------|------|---------|-------------|
| `id` | UUID PK | gen_random_uuid() | Primary key |
| `user_id` | UUID FK (users.id) | — | The user |
| `tenant_id` | UUID FK (tenants.id) | — | The tenant |
| `role` | varchar(50) | "tenant_admin" | User's role in this tenant |
| `created_at` | timestamptz | now() | |
Unique constraint on `(user_id, tenant_id)`. This table allows a single user (single email, single password) to be a member of multiple tenants without duplicating user rows.
The existing `users.tenant_id` column is retained as the "currently active" tenant. The `switch-tenant` endpoint updates this field and re-issues the JWT.
#### `invites`
One-time invite tokens generated by super_admin.
| Column | Type | Default | Description |
|--------|------|---------|-------------|
| `id` | UUID PK | gen_random_uuid() | Primary key |
| `token` | varchar(64), unique, indexed | — | URL-safe random token (32 bytes / 256 bits entropy) |
| `plan_name` | varchar(50) | "invite" | Plan to assign when claimed |
| `created_by` | UUID FK (users.id) | — | Super_admin who created it |
| `claimed_by` | UUID FK (users.id), nullable | — | User who claimed it |
| `claimed_at` | timestamptz, nullable | — | When it was claimed |
| `expires_at` | timestamptz | — | 7 days from creation |
| `created_at` | timestamptz | now() | |
No RLS — only accessible via super_admin endpoints and the public claim endpoint (which validates the token directly).
### Modified Tables
#### `tenants`
Add column:
| Column | Type | Default | Description |
|--------|------|---------|-------------|
| `owner_id` | UUID FK (users.id), nullable | — | User who created/owns this tenant. Null for bootstrap/super_admin-created tenants (always unlimited). |
#### `system_settings`
New key-value entry:
| Key | Default Value | Description |
|-----|---------------|-------------|
| `homelab_signup_enabled` | `"false"` | Controls public signup page visibility |
### Default Plan Values
| Plan | max_tenants | max_devices_per_tenant | max_users_per_tenant |
|------|------------|----------------------|---------------------|
| invite | 2 | 10 | 10 |
| homelab | 1 | 5 | 0 (owner only) |
| custom | (set by super_admin) | (set by super_admin) | (set by super_admin) |
### Migration Notes
- Existing tenants get `owner_id = NULL` (treated as unlimited / super_admin-owned).
- Existing users get a corresponding `user_tenants` row for their current `tenant_id` with their current `role`.
- No `plan_limits` rows are created for existing users (unlimited by default).
## Enforcement Logic
Limits are checked at creation time only — not on every request.
### Device Creation
**Endpoints:** `POST /api/tenants/{id}/devices`, VPN onboard endpoint
1. Look up tenant's `owner_id`
2. If `owner_id` is NULL → no limit (super_admin-owned tenant)
3. Look up `plan_limits` for owner. If no row → no limit.
4. Count devices in tenant (within the same transaction for onboard)
5. If count >= `max_devices_per_tenant` → return 422: `"Device limit reached (5/5)"`
### User Creation
**Endpoint:** `POST /api/tenants/{id}/users`
1. Look up tenant's `owner_id`
2. If `owner_id` is NULL → no limit.
3. Look up `plan_limits` for owner. If no row → no limit.
4. Count active users in tenant
5. If count >= `max_users_per_tenant` → return 422: `"User limit reached (10/10)"`
6. Homelab plan (`max_users_per_tenant = 0`) means only the owner exists — no additional users.
### Tenant Creation
**Endpoint:** `POST /api/tenants`
Currently super_admin only. Change: allow users with a `plan_limits` row to create tenants within their limit.
1. Look up `plan_limits` for current user. If no row → no limit (super_admin).
2. Count tenants where `owner_id = current_user.id`
3. If count >= `max_tenants` → return 422: `"Tenant limit reached (2/2)"`
4. Create tenant with `owner_id = current_user.id`
5. Add `user_tenants` row: `(current_user.id, new_tenant.id, "tenant_admin")`
6. Update `users.tenant_id = new_tenant.id` (switch to the new tenant)
## Invite System
### Creating Invites (Super_admin)
**Endpoint:** `POST /api/invites`
Rate limit: 20/minute
Request body:
```json
{
"plan_name": "invite" // optional, defaults to "invite"
}
```
Response:
```json
{
"id": "uuid",
"token": "abc123...",
"url": "https://app.theotherdude.net/invite/abc123...",
"plan_name": "invite",
"expires_at": "2026-03-21T16:00:00Z",
"created_at": "2026-03-14T16:00:00Z"
}
```
- Generates 32-byte URL-safe random token (256 bits entropy — brute force infeasible)
- Sets `expires_at` to 7 days from now
### Managing Invites (Super_admin)
- `GET /api/invites` — list all invites with status (pending/claimed/expired)
- `DELETE /api/invites/{id}` — revoke an unclaimed invite
### Validating an Invite (Public)
**Endpoint:** `GET /api/invites/{token}/validate`
Rate limit: 5/minute per IP
No auth required. Returns:
```json
{ "valid": true }
```
Or `{ "valid": false }` — no reason disclosed to prevent information leakage about token states.
### Claiming an Invite (Public)
**Endpoint:** `POST /api/invites/{token}/claim`
Rate limit: 5/minute per IP
No auth required. Request body:
```json
{
"name": "Jane Doe",
"email": "jane@example.com",
"password": "securepassword123"
}
```
Flow:
1. Validate token (exists, not claimed, not expired). Return generic 400 "Invalid or expired invite" for any failure (no distinction between expired/claimed/not-found).
2. Check email uniqueness globally
3. Create user with `role = "tenant_admin"`, `tenant_id = NULL`, `must_upgrade_auth = True` (bcrypt, upgrades to SRP on first login)
4. Create `plan_limits` row with plan defaults based on `invite.plan_name`
5. Mark invite as claimed (`claimed_by`, `claimed_at`)
6. Issue JWT with special `onboarding = true` claim (see Onboarding State below)
7. Frontend redirects to tenant creation page
### Onboarding State
After claiming an invite, the user has `tenant_id = NULL` and `role = "tenant_admin"`. The existing RLS middleware blocks non-super_admin users with no tenant. To handle this:
- The JWT issued during claim includes an `onboarding: true` claim
- The tenant context middleware is modified: if `onboarding = true`, allow access to a whitelist of endpoints only:
- `POST /api/tenants` (create first tenant)
- `GET /api/plan/usage` (see their limits)
- `POST /api/auth/logout`
- All other endpoints return 403: "Please create a tenant first"
- After creating their first tenant, the user gets a normal JWT with `tenant_id` set
## Tenant Switching
Users who belong to multiple tenants can switch between them.
**Endpoint:** `POST /api/auth/switch-tenant`
Request body:
```json
{
"tenant_id": "uuid"
}
```
Flow:
1. Look up `user_tenants` for `(current_user.id, target_tenant_id)`. If no row → 403 "You do not have access to this tenant".
2. Update `users.tenant_id = target_tenant_id`
3. Issue new JWT with the target `tenant_id` and the role from `user_tenants.role`
4. Return new access token + refresh token
**Listing available tenants:**
`GET /api/auth/tenants` — returns all tenants the current user belongs to (from `user_tenants`), including the currently active one.
## API Summary
### New Endpoints
| Method | Path | Auth | Rate Limit | Description |
|--------|------|------|------------|-------------|
| `POST` | `/api/invites` | super_admin | 20/min | Create invite |
| `GET` | `/api/invites` | super_admin | — | List all invites |
| `DELETE` | `/api/invites/{id}` | super_admin | 5/min | Revoke invite |
| `GET` | `/api/invites/{token}/validate` | public | 5/min/IP | Check if invite is valid |
| `POST` | `/api/invites/{token}/claim` | public | 5/min/IP | Register via invite |
| `POST` | `/api/auth/switch-tenant` | authenticated | 20/min | Switch active tenant |
| `GET` | `/api/auth/tenants` | authenticated | — | List user's tenants |
| `GET` | `/api/settings/signup-status` | public | — | Check if homelab signup is enabled |
| `GET` | `/api/plan/usage` | authenticated | — | Get current plan limits and usage |
| `PUT` | `/api/admin/users/{user_id}/plan` | super_admin | 20/min | Update a user's plan limits |
### Modified Endpoints
| Method | Path | Change |
|--------|------|--------|
| `POST` | `/api/tenants` | Allow users with plan_limits to create; set `owner_id`; add `user_tenants` row |
| `POST` | `/api/tenants/{id}/devices` | Add device limit enforcement |
| `POST` | `/api/tenants/{id}/vpn/peers/onboard` | Add device limit enforcement (before device creation in transaction) |
| `POST` | `/api/tenants/{id}/users` | Add user limit enforcement |
### Usage Response Schema
`GET /api/plan/usage` returns:
```json
{
"plan_name": "invite",
"tenants": { "current": 1, "max": 2 },
"active_tenant": {
"tenant_id": "uuid",
"devices": { "current": 3, "max": 10 },
"users": { "current": 2, "max": 10 }
}
}
```
Returns device/user counts for the currently active tenant.
## Frontend Changes
### New Pages
- **`/invite/{token}`** — public invite claim page. Standalone (not behind auth). Shows registration form or "Invalid or expired invite" error.
- **`/signup`** — public homelab signup page. Disabled by default. Shows "Not accepting signups" when `homelab_signup_enabled` is false.
- **`/settings/invites`** — super_admin invite management. Create, list, copy link, revoke.
### Modified Components
- **Top nav / sidebar** — tenant switcher dropdown for users who belong to multiple tenants. Shows current tenant name, lists available tenants from `GET /api/auth/tenants`, "Create Tenant" option if under limit.
- **Tenant list** — "Create Tenant" button visible to users with a plan_limits row (not just super_admin). Disabled with tooltip if at limit.
- **Tenant detail (super_admin view)** — shows plan limits and current usage. Editable by super_admin.
- **Device list** — subtle usage indicator: "3/10 devices" near the header. Only shown when limits exist.
- **User list** — subtle usage indicator: "2/10 users" near the header. Only shown when limits exist.
- **System settings (super_admin)** — "Enable homelab signups" toggle.
## Audit Logging
The following operations produce audit log entries:
- Invite created (by super_admin)
- Invite claimed (by new user)
- Invite revoked (by super_admin)
- Tenant created by non-super_admin user
- Tenant switched
- Plan limits updated by super_admin
## Error Handling
| Scenario | HTTP Status | Message |
|----------|-------------|---------|
| Device limit reached | 422 | "Device limit reached ({count}/{max})" |
| User limit reached | 422 | "User limit reached ({count}/{max})" |
| Tenant limit reached | 422 | "Tenant limit reached ({count}/{max})" |
| Invalid/expired/claimed invite | 400 | "Invalid or expired invite" |
| Email already registered | 409 | "Email already in use" |
| Signup disabled | 403 | "Not accepting signups at this time" |
| Switch to unjoined tenant | 403 | "You do not have access to this tenant" |
| Onboarding user hits non-whitelisted endpoint | 403 | "Please create a tenant first" |
## Out of Scope
- Billing / Paddle integration
- Homelab self-registration activation (page exists but disabled)
- VPN per-tenant network isolation (separate spec)
- Email notifications for invites (super_admin copies the link)
- Usage metering / analytics dashboard
- Plan upgrade/downgrade flows
- Tenant deletion by non-super_admin users (remains super_admin only)

249
docs/superpowers/specs/2026-03-14-setup-script-design.md
View File

@@ -1,249 +0,0 @@
# TOD Production Setup Script — Design Spec
## Overview
An interactive Python setup wizard (`setup.py`) that walks a sysadmin through configuring and deploying TOD (The Other Dude) for production. The script minimizes manual configuration by auto-generating secrets, capturing OpenBao credentials automatically, building images sequentially, and verifying service health.
**Target audience:** Technical sysadmins unfamiliar with this specific project.
## Design Decisions
- **Python 3.10+** — already required by the stack, enables rich input handling and colored output.
- **Linear wizard with opinionated defaults** — grouped sections, auto-generate everything possible, only prompt for genuine human decisions.
- **Integrated OpenBao bootstrap** — script starts the OpenBao container, captures unseal key and root token, updates `.env.prod` automatically (no manual copy-paste).
- **Sequential image builds** — builds api, poller, frontend, winbox-worker one at a time to avoid OOM on low-RAM machines.
- **Re-runnable** — safe to run again; detects existing `.env.prod` and offers to overwrite, back up (`.env.prod.backup.<ISO timestamp>`), or abort.
## Prerequisite: Database Rename
The codebase currently uses `mikrotik` as the database name. Before the setup script can use `tod`, these files must be updated:
- `docker-compose.yml` — default `POSTGRES_DB` and healthcheck (`pg_isready -d`)
- `docker-compose.prod.yml` — hardcoded poller `DATABASE_URL` (change to `${POLLER_DATABASE_URL}`)
- `docker-compose.staging.yml` — if applicable
- `scripts/init-postgres.sql` — `GRANT CONNECT ON DATABASE` statements
- `.env.example` — all URL references
The setup script will use `POSTGRES_DB=tod`. These file changes are part of the implementation, not runtime.
Additionally, `docker-compose.prod.yml` hardcodes the poller's `DATABASE_URL`. This must be changed to `DATABASE_URL: ${POLLER_DATABASE_URL}` so the setup script's generated value is used.
## Script Flow
### Phase 1: Pre-flight Checks
- Verify Python 3.10+
- Verify Docker Engine and Docker Compose v2 are installed and the daemon is running
- Check for existing `.env.prod` — if found, offer: overwrite / back up and create new / abort
- Warn if less than 4GB RAM available
- Check if key ports are in use (5432, 6379, 4222, 8001, 3000, 51820) and warn
### Phase 2: Interactive Configuration (Linear Wizard)
Six sections, presented in order:
#### 2.1 Database
| Prompt | Default | Notes |
|--------|---------|-------|
| PostgreSQL superuser password | (required, no default) | Validated non-empty, min 12 chars |
Auto-generated:
- `POSTGRES_DB=tod`
- `app_user` password via `secrets.token_urlsafe(24)` (yields ~32 base64 chars)
- `poller_user` password via `secrets.token_urlsafe(24)` (yields ~32 base64 chars)
- `DATABASE_URL=postgresql+asyncpg://postgres:<pw>@postgres:5432/tod`
- `SYNC_DATABASE_URL=postgresql+psycopg2://postgres:<pw>@postgres:5432/tod`
- `APP_USER_DATABASE_URL=postgresql+asyncpg://app_user:<app_pw>@postgres:5432/tod`
- `POLLER_DATABASE_URL=postgres://poller_user:<poller_pw>@postgres:5432/tod`
#### 2.2 Security
No prompts. Auto-generated:
- `JWT_SECRET_KEY` via `secrets.token_urlsafe(64)` (yields ~86 base64 chars)
- `CREDENTIAL_ENCRYPTION_KEY` via `base64(secrets.token_bytes(32))` (yields 44 base64 chars)
- Display both values to the user with a "save these somewhere safe" note
#### 2.3 Admin Account
| Prompt | Default | Notes |
|--------|---------|-------|
| Admin email | `admin@the-other-dude.dev` | Validated as email-like |
| Admin password | (enter or press Enter to generate) | Min 12 chars if manual; generated passwords are 24 chars |
#### 2.4 Email (Optional)
| Prompt | Default | Notes |
|--------|---------|-------|
| Configure SMTP now? | No | If no, skip with reminder |
| SMTP host | (required if yes) | |
| SMTP port | 587 | |
| SMTP username | (optional) | |
| SMTP password | (optional) | |
| From address | (required if yes) | |
| Use TLS? | Yes | |
#### 2.5 Web / Domain
| Prompt | Default | Notes |
|--------|---------|-------|
| Production domain | (required) | e.g. `tod.staack.com` |
Auto-derived:
- `APP_BASE_URL=https://<domain>`
- `CORS_ORIGINS=https://<domain>`
#### 2.6 Summary & Confirmation
Display all settings grouped by section. Secrets are partially masked (first 8 chars + `...`). Ask for confirmation before writing.
### Phase 3: Write `.env.prod`
Write the file with section comments and timestamp header. Also generate `scripts/init-postgres-prod.sql` with the generated `app_user` and `poller_user` passwords baked in (PostgreSQL init scripts don't support env var substitution).
Format:
```bash
# ============================================================
# TOD Production Environment — generated by setup.py
# Generated: <ISO timestamp>
# ============================================================
# --- Database ---
POSTGRES_DB=tod
POSTGRES_USER=postgres
POSTGRES_PASSWORD=<input>
DATABASE_URL=postgresql+asyncpg://postgres:<pw>@postgres:5432/tod
SYNC_DATABASE_URL=postgresql+psycopg2://postgres:<pw>@postgres:5432/tod
APP_USER_DATABASE_URL=postgresql+asyncpg://app_user:<app_pw>@postgres:5432/tod
POLLER_DATABASE_URL=postgres://poller_user:<poller_pw>@postgres:5432/tod
# --- Security ---
JWT_SECRET_KEY=<generated>
CREDENTIAL_ENCRYPTION_KEY=<generated>
# --- OpenBao (KMS) ---
OPENBAO_ADDR=http://openbao:8200
OPENBAO_TOKEN=PLACEHOLDER_RUN_SETUP
BAO_UNSEAL_KEY=PLACEHOLDER_RUN_SETUP
# --- Admin Bootstrap ---
FIRST_ADMIN_EMAIL=<input>
FIRST_ADMIN_PASSWORD=<input-or-generated>
# --- Email ---
# <configured block or "unconfigured" note>
SMTP_HOST=
SMTP_PORT=587
SMTP_USER=
SMTP_PASSWORD=
SMTP_USE_TLS=true
SMTP_FROM_ADDRESS=noreply@example.com
# --- Web ---
APP_BASE_URL=https://<domain>
CORS_ORIGINS=https://<domain>
# --- Application ---
ENVIRONMENT=production
LOG_LEVEL=info
DEBUG=false
APP_NAME=TOD - The Other Dude
# --- Storage ---
GIT_STORE_PATH=/data/git-store
FIRMWARE_CACHE_DIR=/data/firmware-cache
WIREGUARD_CONFIG_PATH=/data/wireguard
WIREGUARD_GATEWAY=wireguard
CONFIG_RETENTION_DAYS=90
# --- Redis & NATS ---
REDIS_URL=redis://redis:6379/0
NATS_URL=nats://nats:4222
# --- Poller ---
POLL_INTERVAL_SECONDS=60
CONNECTION_TIMEOUT_SECONDS=10
COMMAND_TIMEOUT_SECONDS=30
# --- Remote Access ---
TUNNEL_PORT_MIN=49000
TUNNEL_PORT_MAX=49100
TUNNEL_IDLE_TIMEOUT=300
SSH_RELAY_PORT=8080
SSH_IDLE_TIMEOUT=900
# --- Config Backup ---
CONFIG_BACKUP_INTERVAL=21600
CONFIG_BACKUP_MAX_CONCURRENT=10
```
### Phase 4: OpenBao Bootstrap
1. Start postgres and openbao containers only: `docker compose -f docker-compose.yml -f docker-compose.prod.yml --env-file .env.prod up -d postgres openbao`
2. Wait for openbao container to be healthy (timeout 60s)
3. Run `docker compose logs openbao 2>&1` and parse the `OPENBAO_TOKEN=` and `BAO_UNSEAL_KEY=` lines using regex (init.sh prints these to stdout during container startup, which is captured in Docker logs)
4. Update `.env.prod` by replacing the `PLACEHOLDER_RUN_SETUP` values with the captured credentials
5. On failure: `.env.prod` retains placeholders, print instructions for manual capture via `docker compose logs openbao`
### Phase 5: Build Images
Build sequentially to avoid OOM:
```
docker compose -f docker-compose.yml -f docker-compose.prod.yml build api
docker compose -f docker-compose.yml -f docker-compose.prod.yml build poller
docker compose -f docker-compose.yml -f docker-compose.prod.yml build frontend
docker compose -f docker-compose.yml -f docker-compose.prod.yml build winbox-worker
```
Show progress for each. On failure: stop, report which image failed, suggest rerunning.
### Phase 6: Start Stack
```
docker compose -f docker-compose.yml -f docker-compose.prod.yml --env-file .env.prod up -d
```
### Phase 7: Health Check
- Poll service health for up to 60 seconds
- Report status of: postgres, redis, nats, openbao, api, poller, frontend, winbox-worker
- On success: print access URL (`https://<domain>`) and admin credentials
- On timeout: report which services are unhealthy, suggest `docker compose logs <service>`
## Database Init Script
`scripts/init-postgres.sql` hardcodes `app_password` and `poller_password`. Since PostgreSQL's `docker-entrypoint-initdb.d` scripts don't support environment variable substitution, the setup script generates `scripts/init-postgres-prod.sql` with the actual generated passwords baked in. The docker-compose.prod.yml volume mount will be updated to use this file instead.
## Login Page Fix
`frontend/src/routes/login.tsx` lines 235-241 contain a "First time?" hint showing `.env` credential names. This will be wrapped in `{import.meta.env.DEV && (...)}` so it only appears in development builds. Vite's production build strips DEV-gated code entirely.
## Error Handling
| Scenario | Behavior |
|----------|----------|
| Docker not installed/running | Fail early with clear message |
| Existing `.env.prod` | Offer: overwrite / back up / abort |
| Port already in use | Warn (non-blocking) with which port and likely culprit |
| OpenBao init fails | `.env.prod` retains placeholders, print manual capture steps |
| Image build fails | Stop, show failed image, suggest retry command |
| Health check timeout (60s) | Report unhealthy services, suggest log commands |
| Ctrl+C before Phase 3 | Graceful exit, no files written |
| Ctrl+C during/after Phase 3 | `.env.prod` exists (possibly with placeholders), noted on exit |
## Re-runnability
- Detects existing `.env.prod` and offers choices
- Won't regenerate secrets if valid ones exist (offers to keep or regenerate)
- OpenBao re-init is idempotent (init.sh handles already-initialized state)
- Image rebuilds are safe (Docker layer caching)
- Backup naming: `.env.prod.backup.<ISO timestamp>`
## Dependencies
- Python 3.10+ (stdlib only — no pip packages required)
- Docker Engine 24+
- Docker Compose v2
- Stdlib modules: `secrets`, `subprocess`, `shutil`, `json`, `re`, `datetime`, `pathlib`, `getpass`, `socket` (for port checks)

274
docs/superpowers/specs/2026-03-14-vpn-isolation-design.md
View File

@@ -1,274 +0,0 @@
# Per-Tenant VPN Network Isolation — Design Spec
## Overview
Isolate WireGuard VPN networks per tenant so that devices in one tenant's VPN cannot reach devices in another tenant's VPN. Each tenant gets a unique `/24` subnet auto-allocated from `10.10.0.0/16`, with iptables rules blocking cross-subnet traffic.
**Branch:** `main` (this is a security fix, not SaaS-specific)
## Design Decisions
- **Single `wg0` interface** — WireGuard handles thousands of peers on one interface with negligible performance impact. No need for per-tenant interfaces.
- **Per-tenant `/24` subnets** — allocated from `10.10.0.0/16`, giving 255 tenants (index 1–255). Index 0 is reserved. Expandable to `10.0.0.0/8` if needed (note: `_next_available_ip()` materializes all hosts in the subnet, so subnets larger than `/24` require refactoring that function).
- **Auto-allocation only** — `setup_vpn()` picks the next available subnet. No manual override.
- **Global config sync** — one `wg0.conf` with all tenants' peers. Rebuilt on any VPN change. Protected by a PostgreSQL advisory lock to prevent concurrent writes.
- **Global server keypair** — a single WireGuard server keypair stored in `system_settings`, replacing per-tenant server keys. Generated on first `setup_vpn()` call or during migration.
- **iptables isolation** — cross-subnet traffic blocked at the WireGuard container's firewall. IPv6 blocked too.
- **Device-side config is untrusted** — isolation relies entirely on server-side enforcement (AllowedIPs `/32` + iptables DROP). A malicious device operator changing their `allowed-address` to `10.10.0.0/16` on their router gains nothing — the server only routes their assigned `/32`.
## Data Model Changes
### Modified: `vpn_config`
| Column | Change | Description |
|--------|--------|-------------|
| `subnet_index` | **New column**, integer, unique, not null | Maps to third octet: index 1 = `10.10.1.0/24` |
| `subnet` | Default changes | No longer `10.10.0.0/24`; derived from `subnet_index` |
| `server_address` | Default changes | No longer `10.10.0.1/24`; derived as `10.10.{index}.1/24` |
| `server_private_key` | **Deprecated** | Kept in table for rollback safety but no longer used. Global key in `system_settings` is authoritative. |
| `server_public_key` | **Deprecated** | Same — kept but unused. All peers use the global public key. |
### New: `system_settings` entries
| Key | Description |
|-----|-------------|
| `vpn_server_private_key` | Global WireGuard server private key (encrypted with CREDENTIAL_ENCRYPTION_KEY) |
| `vpn_server_public_key` | Global WireGuard server public key (plaintext) |
### Allocation Logic
```
subnet_index = first available integer in range [1, 255] not already in vpn_config
subnet = 10.10.{subnet_index}.0/24
server_address = 10.10.{subnet_index}.1/24
```
Allocation query (atomic, gap-filling):
```sql
SELECT MIN(x) FROM generate_series(1, 255) AS x
WHERE x NOT IN (SELECT subnet_index FROM vpn_config)
```
If no index available → 422 "VPN subnet pool exhausted".
Unique constraint on `subnet_index` provides safety against race conditions. On conflict, retry once.
## VPN Service Changes
### `setup_vpn(db, tenant_id, endpoint)`
Current behavior: creates VpnConfig with hardcoded `10.10.0.0/24` and generates a per-tenant server keypair.
New behavior:
1. **Get or create global server keypair:** check `system_settings` for `vpn_server_private_key`. If not found, generate a new keypair and store both the private key (encrypted) and public key. This happens on the first `setup_vpn()` call on a fresh install.
2. Allocate next `subnet_index` using the gap-filling query
3. Set `subnet = 10.10.{index}.0/24`
4. Set `server_address = 10.10.{index}.1/24`
5. Store the global public key in `server_public_key` (for backward compat / display)
6. Call `sync_wireguard_config(db)` (global, not per-tenant)
### `sync_wireguard_config(db)`
Current signature: `sync_wireguard_config(db, tenant_id)` — builds config for one tenant.
New signature: `sync_wireguard_config(db)` — builds config for ALL tenants.
**Concurrency protection:** acquire a PostgreSQL advisory lock (`pg_advisory_xact_lock(hash)`) before writing. This prevents two simultaneous peer additions from producing a corrupt `wg0.conf`.
**Atomic write:** write to a temp file, then `os.rename()` to `wg0.conf`. This prevents the WireGuard container from reading a partially-written file.
New behavior:
1. Acquire advisory lock
2. Read global server private key from `system_settings` (decrypt it)
3. Query ALL enabled `VpnConfig` rows (across all tenants, using admin engine to bypass RLS)
4. For each, query enabled `VpnPeer` rows
5. Build single `wg0.conf`:
```ini
[Interface]
Address = 10.10.0.1/16
ListenPort = 51820
PrivateKey = {global_server_private_key}
# --- Tenant: {tenant_name} (10.10.1.0/24) ---
[Peer]
PublicKey = {peer_public_key}
PresharedKey = {preshared_key}
AllowedIPs = 10.10.1.2/32
# --- Tenant: {tenant_name_2} (10.10.2.0/24) ---
[Peer]
PublicKey = {peer_public_key}
PresharedKey = {preshared_key}
AllowedIPs = 10.10.2.2/32
```
6. Write to temp file, `os.rename()` to `wg0.conf`
7. Touch `.reload` flag
8. Release advisory lock
### `_next_available_ip(db, tenant_id, config)`
No changes needed — already scoped to `tenant_id` and uses the config's subnet. With unique subnets per tenant, IPs are naturally isolated. Note: this function materializes all `/24` hosts into a list, which is fine for `/24` (253 entries) but must be refactored if subnets larger than `/24` are ever used.
### `add_peer(db, tenant_id, device_id, ...)`
Changes:
- Calls `sync_wireguard_config(db)` instead of `sync_wireguard_config(db, tenant_id)`
- **Validate `additional_allowed_ips`:** if provided, reject any subnet that overlaps with `10.10.0.0/16` (the VPN address space). Only non-VPN subnets are allowed (e.g., `192.168.1.0/24` for site-to-site routing). This prevents a tenant from claiming another tenant's VPN subnet in their AllowedIPs.
### `remove_peer(db, tenant_id, peer_id)`
Minor change: calls `sync_wireguard_config(db)` instead of `sync_wireguard_config(db, tenant_id)`.
### Tenant deletion hook
When a tenant is deleted (CASCADE deletes vpn_config and vpn_peers), call `sync_wireguard_config(db)` to regenerate `wg0.conf` without the deleted tenant's peers. Add this to the tenant deletion endpoint.
### `read_wg_status()`
No changes — status is keyed by peer public key, which is unique globally. The existing `get_peer_handshake()` lookup continues to work.
## WireGuard Container Changes
### iptables Isolation Rules
Update `docker-data/wireguard/custom-cont-init.d/10-forwarding.sh`:
```bash
#!/bin/sh
# Enable forwarding between Docker network and WireGuard tunnel
# Idempotent: check before adding to prevent duplicates on restart
iptables -C FORWARD -i eth0 -o wg0 -j ACCEPT 2>/dev/null || iptables -A FORWARD -i eth0 -o wg0 -j ACCEPT
iptables -C FORWARD -i wg0 -o eth0 -j ACCEPT 2>/dev/null || iptables -A FORWARD -i wg0 -o eth0 -j ACCEPT
# Block cross-subnet traffic on wg0 (tenant isolation)
# Peers in 10.10.1.0/24 cannot reach peers in 10.10.2.0/24
iptables -C FORWARD -i wg0 -o wg0 -j DROP 2>/dev/null || iptables -A FORWARD -i wg0 -o wg0 -j DROP
# Block IPv6 forwarding on wg0 (prevent link-local bypass)
ip6tables -C FORWARD -i wg0 -j DROP 2>/dev/null || ip6tables -A FORWARD -i wg0 -j DROP
# NAT for return traffic
iptables -C POSTROUTING -t nat -o wg0 -j MASQUERADE 2>/dev/null || iptables -t nat -A POSTROUTING -o wg0 -j MASQUERADE
echo "WireGuard forwarding and tenant isolation rules applied"
```
Rules use `iptables -C` (check) before `-A` (append) to be idempotent across container restarts.
The key isolation layers:
1. **WireGuard AllowedIPs** — each peer can only send to its own `/32` IP (cryptographic enforcement)
2. **iptables `wg0 → wg0` DROP** — blocks any traffic that enters and exits the tunnel interface (peer-to-peer)
3. **iptables IPv6 DROP** — prevents link-local IPv6 bypass
4. **Separate subnets** — no IP collisions between tenants
5. **`additional_allowed_ips` validation** — blocks tenants from claiming VPN address space
### Server Address
The `[Interface] Address` changes from `10.10.0.1/24` to `10.10.0.1/16` so the server can route to all tenant subnets.
## Routing Changes
### Poller & API
No changes needed. Both already route `10.10.0.0/16` via the WireGuard container.
### setup.py
Update `prepare_data_dirs()` to write the updated forwarding script with idempotent rules and IPv6 blocking.
## RouterOS Command Generation
### `onboard_device()` and `get_peer_config()`
These generate RouterOS commands for device setup. Changes:
- `allowed-address` changes from `10.10.0.0/24` to `10.10.{index}.0/24` (tenant's specific subnet)
- `endpoint-address` and `endpoint-port` unchanged
- Server public key changes to the global server public key (read from `system_settings`)
## Migration
### Database Migration
1. Generate global server keypair:
- Create keypair using `generate_wireguard_keypair()`
- Store in `system_settings`: `vpn_server_private_key` (encrypted), `vpn_server_public_key` (plaintext)
2. Add `subnet_index` column to `vpn_config` (integer, unique, not null)
3. For existing VpnConfig rows (may be multiple if multiple tenants have VPN):
- Assign sequential `subnet_index` values starting from 1
- Update `subnet` to `10.10.{index}.0/24`
- Update `server_address` to `10.10.{index}.1/24`
4. For existing VpnPeer rows:
- Remap IPs: `10.10.0.X` → `10.10.{tenant's index}.X` (preserve the host octet)
- Example: Tenant A (index 1) peer at `10.10.0.2` → `10.10.1.2`. Tenant B (index 2) peer at `10.10.0.2` → `10.10.2.2`. No collision.
5. Regenerate `wg0.conf` using the new global sync function
### Device-Side Update Required
This is a **breaking change** for existing VPN peers. After migration:
- Devices need updated RouterOS commands:
- New server public key (global key replaces per-tenant key)
- New VPN IP address (`10.10.0.X` → `10.10.{index}.X`)
- New allowed-address (`10.10.{index}.0/24`)
- The API should expose a "regenerate commands" endpoint or show a banner in the UI indicating that VPN reconfiguration is needed.
### Migration Communication
After the migration runs:
- Log a warning with the list of affected devices
- Show a banner in the VPN UI: "VPN network updated — devices need reconfiguration. Click here for updated commands."
- The existing "View Setup Commands" button in the UI will show the correct updated commands.
## API Changes
### Modified Endpoints
| Method | Path | Change |
|--------|------|--------|
| `POST` | `/api/tenants/{id}/vpn` | `setup_vpn` allocates subnet_index, uses global server key |
| `GET` | `/api/tenants/{id}/vpn` | Returns tenant's specific subnet info |
| `GET` | `/api/tenants/{id}/vpn/peers/{id}/config` | Returns commands with tenant-specific subnet and global server key |
| `POST` | `/api/tenants/{id}/vpn/peers` | Validates `additional_allowed_ips` doesn't overlap `10.10.0.0/16` |
| `DELETE` | `/api/tenants/{id}` | Calls `sync_wireguard_config(db)` after cascade delete |
### No New Endpoints
The isolation is transparent — tenants don't need to know about it.
## Error Handling
| Scenario | HTTP Status | Message |
|----------|-------------|---------|
| No available subnet index (255 tenants with VPN) | 422 | "VPN subnet pool exhausted" |
| Subnet index conflict (race condition) | — | Retry allocation once |
| `additional_allowed_ips` overlaps VPN space | 422 | "Additional allowed IPs must not overlap the VPN address space (10.10.0.0/16)" |
## Testing
- Create two tenants with VPN enabled → verify they get different subnets (`10.10.1.0/24`, `10.10.2.0/24`)
- Add peers in both → verify IPs don't collide
- From tenant A's device, attempt to ping tenant B's device → verify it's blocked
- Verify `wg0.conf` contains peers from both tenants with correct subnets
- Verify iptables rules are in place after container restart (idempotent)
- Verify `additional_allowed_ips` with `10.10.x.x` subnet is rejected
- Delete a tenant → verify `wg0.conf` is regenerated without its peers
- Disable a tenant's VPN → verify peers excluded from `wg0.conf`
- Empty state (no enabled tenants) → verify `wg0.conf` has only `[Interface]` section
- Migration: multiple tenants sharing `10.10.0.0/24` → verify correct remapping to unique subnets
## Audit Logging
- Subnet allocated (tenant_id, subnet_index, subnet)
- Global server keypair generated (first-run event)
- VPN config regenerated (triggered by which operation)
## Out of Scope
- Multiple WireGuard interfaces (not needed at current scale)
- Manual subnet assignment
- IPv6 VPN support (IPv6 is blocked as a security measure)
- Per-tenant WireGuard listen ports
- VPN-level rate limiting or bandwidth quotas