clsdumper

       _         _
   ___| |___  __| |_   _ _ __ ___  _ __   ___ _ __
  / __| / __|/ _` | | | | '_ ` _ \| '_ \ / _ \ '__|
 | (__| \__ \ (_| | |_| | | | | | | |_) |  __/ |
  \___|_|___/\__,_|\__,_|_| |_| |_| .__/ \___|_|
                                   |_|

Android Dynamic Class Dumper — dump all DEX files from running Android apps using Frida.

clsdumper hooks into a running (or freshly spawned) Android application and extracts every DEX file it can find using 9 different strategies. It works against apps with anti-Frida protections, packed/encrypted DEX files, and dynamically loaded code.

Features

• 9 extraction strategies — ART runtime walk, function hooking, memory scanning, OAT/VDEX parsing, and more
• Anti-Frida bypass — blocks signal-handler registration, filters /proc/self/maps, neutralizes watchdog threads
• Spawn & attach — start the app fresh or hook into an already running process
• Automatic deduplication — agent-side fast hash + host-side SHA-256
• Class extraction — optionally decompile DEX files into individual .smali classes via androguard
• Works without Java bridge — native strategies run outside Java.perform(), so they work even when the Java bridge crashes

Quick Start

bashCopy
# Install from PyPI
pip install clsdumper

# Dump DEX from a running app (USB-connected device)
clsdumper com.example.app

# Spawn the app and dump
clsdumper com.example.app --spawn

# Attach by PID
clsdumper 12345

Requirements

• Python 3.10+
• Rooted Android device with frida-server running
• USB debugging enabled (or use --host for TCP)

Usage

clsdumper [target] [options]

Option	Description
target	Package name or PID of the target app
-o, --output DIR	Output directory (default: ./dump_<target>)
--spawn	Spawn the app instead of attaching
--host HOST	Frida server host for TCP connection (default: USB)
--strategies LIST	Comma-separated list of strategies to use
--no-scan	Disable memory scan strategy
--deep-scan	Enable deep scan (CDEX files, slower)
--extract-classes	Extract individual classes from dumped DEX files
--no-anti-frida	Disable anti-Frida bypass
-d, --debug	Enable debug output
--list	List running processes on the device
--list-apps	List installed applications
-v, --version	Show version

Examples

bashCopy
# Use specific strategies only
clsdumper com.example.app --strategies fart_dump,oat_extract

# Spawn with class extraction
clsdumper com.example.app --spawn --extract-classes

# Connect to remote frida-server
clsdumper com.example.app --host 192.168.1.100

# List all installed apps
clsdumper --list-apps

Strategies

#	Strategy	Phase	Description
1	art_walk	Native	Walks Runtime -> ClassLinker -> DexFile structs in ART
2	open_common_hook	Native	Hooks DexFile::OpenCommon in libdexfile.so
3	memory_scan	Native	Scans readable memory regions for DEX magic bytes
4	cookie	Java	Reads mCookie field from ClassLoaders via reflection
5	classloader_hook	Java	Monitors loadClass / DexClassLoader / InMemoryDexClassLoader
6	mmap_hook	Native	Intercepts mmap/mmap64 calls (not in default set)
7	oat_extract	Native	Parses mapped .vdex / .oat files for embedded DEX
8	fart_dump	Native	Hooks DefineClass + walks class_table_ (best coverage)
9	dexfile_constructor	Native	Hooks OatDexFile C++ constructors

Default set: all except mmap_hook (causes performance issues when combined with other hooks).

Building from Source

The Frida agent is pre-built and included as clsdumper/frida/scripts/agent.js. To modify the agent:

bashCopy
# Install agent dependencies
cd agent && npm install

# Build the TypeScript agent
npm run build

# Install the Python package
cd .. && pip install -e .

How It Works

clsdumper runs in three phases:

1. Phase 0 — Anti-Frida bypass: Hooks sigaction/signal to block anti-debugging signal handlers, replaces /proc/self/maps reads via memfd_create, and monitors pthread_create to neutralize watchdog threads.

2. Phase 1 — Native strategies: Runs outside Java.perform() using direct memory access and native function hooks. This works even when the Java bridge fails (common with heavily protected apps).

3. Phase 2 — Java strategies: Uses the Java bridge to inspect ClassLoaders and hook class loading. Falls back gracefully if the bridge is unavailable.

All found DEX files are deduplicated (fast djb2 hash on the agent, SHA-256 on the host) and saved with metadata including the strategy that found them.

Output

dump_com.example.app/
  dex/
    classes_001.dex    # Dumped DEX files
    classes_002.dex
    ...
  classes/             # Only with --extract-classes
    com/example/...
  metadata.json        # Dump metadata and per-file info

License

MIT