Goal: Demonstrate VoIP stream capture by hooking Android's audioserver (specifically libaudioflinger.so) and recording uplink (mic) and downlink (speaker) PCM buffers when the system is in communication mode.

Project Design & Methodology

Architecture:

• Controller App (Android)
  UI with 9 buttons to manage SELINUX policy setup, payload injection, data movement, and playback.

• Payload (libaudiohook.so)
  C++ shared library (built as part of the app project) that hooks internal AudioFlinger methods using Dobby and writes raw PCM buffers to a session directory.

• Post-Processing
  Additional tool, tools/audioparser.py converts raw .ac(Downlink)/.bc(Uplink) files to WAV. The app’s file manager can also convert & play in-app.

• Research Utilities
  Frida scripts for initial function tracing & offset discovery; analysis files for symbol/offset and other snippets.

Controller App

This Android application provides a graphical interface for managing VoIP recording through the audioserver. It integrates with a native payload that hooks into the Android audio stack to capture microphone and speaker audio from VoIP applications.

The app serves as a control center, allowing you to inject required SELinux policies, start/stop monitoring, manage recorded data, and play captured VoIP audio files.

Features

The main activity contains 9 control buttons:

1. Inject Policies
   Injects the required SELinux policies for audioserver and this app to work together.

2. Start Monitoring
   Injects the shared library payload into audioserver, enabling VoIP audio capture by hooking RecordTrack and PlaybackTrack. Uses AndKittyInjector for Process Injection.

3. Stop Monitoring
   Restarts the audioserver process, stopping monitoring and removing injected hooks.

4. Copy Data
   Copies captured raw PCM audio from temp directory /data/local/tmp/voip to /sdcard/voip.

   • Cleans up the temporary directory afterward.

5. View Logs
   Displays logs related to injection, hooking, and recording status.

6. Enable SELinux
   Restores SELinux enforcing mode.

7. Disable SELinux
   Sets SELinux to permissive mode, Use when injected policies fail.

8. Play Audios
   Opens the custom VoIP File Manager (/sdcard/voip/):

   • Converts raw PCM files into WAV format in-memory.

+----------------------+    
|                      |    
|   App MainActivity   |    
|                      |    
|                      |    
|(9 Control Buttons UI)|    
|                      |    
|                      |    
+----------+-----------+    
           |                
           |                
           |                
+----------v-----------+    
|                      |    
|       JNI Layer      |    
|                      |    
| Selinux, Process Inje|    
| ction,               |    
|                      |    
|                      |    
+----------------------+    

Hooking Payload

This shared library is injected into audioserver to detect AUDIO_MODE_IN_COMMUNICATION and capture VoIP audio streams directly from Android’s audio stack for predefined set of package names.

It uses the Dobby inline hooking framework to intercept critical methods inside libaudioflinger.so.
Both uplink (microphone) and downlink (speaker) audio streams are captured and written to session files for later processing.

Hooking Points

The library installs hooks on the following functions inside AudioFlinger and related classes:

Target Applications

The library filters streams based on UID → Package mapping.

Currently monitored apps:

• com.whatsapp
• com.whatsapp.w4b (WhatsApp Business)
• org.thoughtcrime.securesms (Signal)
• org.telegram.messenger
• org.telegram.messenger.web

You can add other applications also iniside audioserver_hook.cpp:

static const char* kTargetPackages[] = {  
    "com.whatsapp",  
    "com.whatsapp.w4b",  
    "org.thoughtcrime.securesms",  
    "org.telegram.messenger",  
    "org.telegram.messenger.web"  
};

File Output

• A session directory is created at injection into audioserver under: /data/local/tmp/voip/audioserver_session_<timestamp>/

• Files are first created with a .tmp extension, then renamed on track stop:

  • Uplink (mic) → <packageName>_<sampleRate>_<timestamp>.bc
  • Downlink (speaker) → <packageName>_<sampleRate>_<timestamp>.ac

• Example session:

	/data/local/tmp/voip/audioserver_session_1725389200/
	├── com.whatsapp_48000_1725389201_123456.ac
	├── com.whatsapp_48000_1725389201_123789.bc
	

Offsets

Since target library is stripped and methods are not exported in libaudioflinger.so, we need exact offsets of methods and other related components for successful hooking.

We can extract those offsets from static analysis using IDA-PRO or from dynamic analysis using frida. Once offsets are known we can replace them inside offsets.h header file.

Required Offsets:

Methods:

1. AUDIOFLINGER_SETMODE_OFFSET
   • Method: android::AudioFlinger::setMode(audio_mode_t)
   • Mangled name: _ZN7android12AudioFlinger7setModeE11audio_mode_t
2. RECORDTRACK_GETNEXTBUFFER_OFFSET
   • Method: android::AudioFlinger::RecordThread::RecordTrack::getNextBuffer(android::AudioBufferProvider::Buffer *)
   • Mangled name: _ZN7android12AudioFlinger12RecordThread11RecordTrack13getNextBufferEPNS_19AudioBufferProvider6BufferE
3. TRACK_GETNEXTBUFFER_OFFSET
   • Method: android::AudioFlinger::PlaybackThread::Track::getNextBuffer(android::AudioBufferProvider::Buffer *)
   • Mangled name: _ZN7android12AudioFlinger14PlaybackThread5Track13getNextBufferEPNS_19AudioBufferProvider6BufferE
4. TRACK_STOP_OFFSET
   • Method: android::AudioFlinger::PlaybackThread::Track::stop(void)
   • Mangled name: _ZN7android12AudioFlinger14PlaybackThread5Track4stopEv

Other Offsets:

1. SAMPLERATE_OFFSET
2. UID_OFFSET
3. ISOUT_OFFSET
4. FORMAT_OFFSET
5. CHANNELMASK_OFFSET
6. ATTR_OFFSET

These can be obtained from static analysis.

• Library - libaudioflinger.so
• Method - android::AudioFlinger::ThreadBase::TrackBase::TrackBase

TrackBase::TrackBase(
        IAfThreadBase *thread,
            const sp<Client>& client,
            const audio_attributes_t& attr,
            uint32_t sampleRate,
            audio_format_t format,
            audio_channel_mask_t channelMask,
            size_t frameCount,
            void *buffer,
            size_t bufferSize,
            audio_session_t sessionId,
            pid_t creatorPid,
            uid_t clientUid,
            bool isOut,
            const alloc_type alloc,
            track_type type,
            audio_port_handle_t portId,
            std::string metricsId)

Notes

• Only IN_COMMUNICATION mode is captured. Normal audio (music, ringtone, etc.) is ignored.
• Special handling for Telegram and Signal enforces correct sample rates (Telegram: 48000 Hz, Signal: 8000 Hz) to ignore other unnecessary audios like phone ring and phone disconnect sounds.
• PCM is stored raw; the controller app later converts PCM → WAV for playback.

 ┌─────────────────────────────┐
 │        VoIP Apps            │
 │  (WhatsApp / Signal / TG)   │
 └───────────────┬─────────────┘
                 │ Audio in/out
                 ▼
 ┌─────────────────────────────┐
 │       audioserver           │
 │  (libaudioflinger.so)       │
 └───────┬─────────┬───────────┘
         │         │
   uplink│         │downlink
 (mic)   │         │(speaker)
         ▼         ▼
 ┌───────────-──┐   ┌────────────-─┐
 │ RecordTrack  │   │  Track       │
 │ getNextBuffer│   │ getNextBuffer│
 └───────┬────-─┘   └───────┬────-─┘
         │                 │
   ┌─────▼─────┐     ┌─────▼─────┐
   │ hook_...  │     │ hook_...  │
   │ (uplink)  │     │ (downlink)│
   └─────┬─────┘     └─────┬─────┘
         │                 │
         └──────┬───┬──────┘
                │   │
                ▼   ▼
     ┌───────────────────────────┐
     │  /data/local/tmp/voip/    │
     │  audioserver_session_*    │
     │   - *.bc  (uplink)        │
     │   - *.ac  (downlink)      │
     └───────────────────────────┘

Target Function Identification

Process used:

1. Dynamic tracing - Frida:
   • Used Frida to trace all function calls during a VOIP Call
   • Filtered all functions that are responsible for audio buffers handling.
   • Then created Frida scripts to capture and handle audio buffers.
2. AOSP review & class mapping
   • AudioFlinger manages playback/capture via PlaybackThread & RecordThread.
   • Per-client audio is represented by PlaybackThread::Track (downlink) and RecordThread::RecordTrack (uplink).
3. Filtering to App specific streams
   • Used UID→package mapping to restrict to target apps.
4. Stability and offsets
   • Final native payload resolves addresses via base + hardcoded offsets (kept in app/src/main/cpp/include/offsets.h) to survive stripped symbols.

Detecting MODE_IN_COMMUNICATION

We hook AudioFlinger::setMode(int mode) and maintain a global state:

• On each call, we log the mode change and update g_currentAudioMode.
• Hooks (in both mic and speaker hooks) check g_currentAudioMode == AUDIO_MODE_IN_COMMUNICATION.
• This ensures we only dump buffers while a VoIP session is active, avoiding music/ringtones/device playback/mic recordings etc.

Building & Running the PoC

Project Layout

.
├── Android_App/ 
|   ├── app/                           # Android Studio project (App + Payload)
|	│   ├── src/
|	│   │   ├── main/
|	│   │   │   ├── java/              # Controller App Java/Kotlin sources
|	│   │   │   ├── cpp/               # Payload source (built with the app)
|	│   │   │   │   └── include/
|	│   │   │   │       └── offsets.h  # Function offsets for tested builds
├── tools/
│   ├── audioparser.py             # Converts .ac/.bc raw files → WAV
│   ├── frida/                     # Frida scripts for initial hook discovery
│   │   ├── host_audioserver.py
│   │   ├── agent_audioserver.js
├── app-release.apk                 # Signed release apk
└── README.md                       # Root README (design, build, run instructions)

Prerequisites

• Android Studio + NDK
• Rooted emulator (e.g., rootAVD) or test device you own (e.g., Magisk)
• Test VoIP apps installed (same ones you own/administer)
• Python for audioparser.py

Build the App (which also builds the payload)

1. Open app/ in Android Studio.
2. Ensure NDK is installed (SDK Manager → SDK Tools → NDK).
3. Build Debug variant. The Gradle/CMake config will compile the native payload (libaudiohook.so) as part of the project.

Install on Test Target

adb install -r app/build/outputs/apk/debug/app-debug.apk or Install app-release.apk provided: adb install -r app-release.apk

If you separate artifacts, you can also push the shared library for testing: adb push app/src/main/jniLibs/arm64-v8a/libaudiohook.so /data/local/tmp/

Run the PoC (Emulator/Owned Device)

1. Launch the Controller App -> grant root when prompted.
2. Tap "Inject Policies"
   Prepares SELinux rules required for the app and audioserver interaction in your test environment.
3. Tap “Start Monitoring”
   The app uses AndKittyInjector to inject the payload into audioserver.
4. Place a test VoIP call (e.g., your own WhatsApp/Signal/Telegram account).
5. Tap “Stop Monitoring”
   Restarts audioserver.
6. Tap “Copy Data”
   Moves session files from /data/local/tmp/voip/… -> /sdcard/voip/ (and clears the temp directory).
7. Playback
   • In-app: “Play Audios” converts PCM -> WAV and plays.
   • Or offline:
     root@kitploit:~

      adb pull /sdcard/voip/
     python3 tools/audioparser.py voip/
     

Additional buttons:

• “View Logs”: See injector/hook status.
• “Enable SELinux” / “Disable SELinux”: Toggle enforcing/permissive in your lab if policies are too strict.
• “Delete Data”: Clear both /data/local/tmp/voip/ and /sdcard/voip/.

Tested Versions & Devices

• Emulator: AVD API 34 (x86_64) - Android 14 - rooted with rootAVD
• Device: Samsung Galaxy S22 (ARM64) - Android 14 - Build S901EXXSCEYB1 - rooted with Magisk

Tested Apps

• Whatsapp
• Whatsapp Business
• Signal
• Telegram

Notable Challenges & Solutions

1. Stripped symbols
   • Issue: Symbols cannot be located directly via dlsym.
   • Fix: Use Offsets from Static analysis.
2. Different Symbol/Offset Across Builds
   • Issue: Stripped symbols vary by build; direct dlsym may fail.
   • Fix: Use base + hardcoded offsets for tested builds (offsets.h), validated with Frida symbol dumps and IDA-PRO.
3. Mode Detection Accuracy
   • Issue: Recording outside VoIP sessions yields noisy data.
   • Fix: Hooks captures on AudioFlinger::setMode(IN_COMMUNICATION); log transitions; store a global mode flag.
4. SELinux
   • Issue: Writing to temp directories & IPC with audioserver can be blocked.
   • Fix: Provide a Inject Policies step to add necessary rules; expose Enable/Disable SELinux toggles for debugging in test environments.
5. File Lifecycle & Renaming
   • Issue: Avoiding partial/corrupt outputs on thread stop.
   • Fix: Write to .tmp first; on stop() rename to .ac (downlink) or (uplink).

References (Open-Source & Docs)

• Dobby (Inline Hooking Framework) - https://www.github.com/jmpews/Dobby
• AndKittyInjector (Android process injector) - https://github.com/MJx0/AndKittyInjector
• AOSP
  • frameworks/av/services/audioflinger (AudioFlinger, PlaybackThread, RecordThread)
  • system/media/audio (audio formats, attributes, channel masks)