frida-termux-compilation-recipe.md 4.35 KB

Frida Termux Compilation Recipe

Session: 2026-06-26 Context: Android 16 (Pixel 10 Pro Fold), Termux, Magisk root

Problem

Installing frida Python client on Android/Termux requires compiling the C extension. The build needs:

  1. frida-core-devkit (matching frida-server version)
  2. meson build system
  3. ninja build tool
  4. Various dependencies (glib, libffi, pkg-config)

Step-by-Step Compilation

1. Install Dependencies

pkg install clang make pkg-config glib libffi
pip install meson

2. Compile ninja from Source (Android Bionic Fix)

Standard ninja binaries don't work on Android Bionic. Must compile from source with patches.

# Download ninja source
curl -L -o ninja-src.tar.gz \
  "https://github.com/ninja-build/ninja/archive/refs/tags/v1.12.1.tar.gz"
tar -xzf ninja-src.tar.gz
cd ninja-1.12.1

# Patch src/util.cc for Android Bionic
# Bionic lacks sysinfo structure and SI_LOAD_SHIFT
# Replace the GetLoadAverage() function:

sed -i 's/#elif defined(__UCLIBC__) || (defined(__BIONIC__) \&\& __ANDROID_API__ < 29)/#elif defined(__UCLIBC__) || defined(__BIONIC__)/' src/util.cc

# Then manually edit src/util.cc to replace GetLoadAverage() with:
# double GetLoadAverage() {
#   return -0.0f;
# }

python3 configure.py --bootstrap
cp ninja $PREFIX/bin/
ninja --version  # Should show 1.12.1

3. Download Matching frida-core-devkit

CRITICAL: frida-server and frida Python must be same version.

VERSION=17.15.3
curl -L -o frida-core-devkit-${VERSION}-android-arm64.tar.xz \
  "https://github.com/frida/frida/releases/download/${VERSION}/frida-core-devkit-${VERSION}-android-arm64.tar.xz"

tar -xf frida-core-devkit-${VERSION}-android-arm64.tar.xz
# Produces: libfrida-core.a, frida-core.h, frida-core-example.c

4. Download frida Python Source

pip download frida==${VERSION} --no-deps -d frida_wheel
cd frida_wheel
tar -xzf frida-*.tar.gz

5. Compile Python Extension

mkdir frida_build && cd frida_build

# Create meson.build pointing to downloaded sources
# See SKILL.md for full meson.build template

meson setup build
meson compile -C build

# Output: build/_frida.abi3.so (should be ~130MB)

6. Install frida Python Package

cd frida-${VERSION}
FRIDA_EXTENSION=/path/to/frida_build/build/_frida.abi3.so \
  pip install . --no-build-isolation

# Verify
python3 -c "import frida; print(frida.__version__)"

7. Upgrade frida-server (if needed)

# Download matching frida-server
curl -L -o frida-server-${VERSION}-android-arm64.xz \
  "https://github.com/frida/frida/releases/download/${VERSION}/frida-server-${VERSION}-android-arm64.xz"

unxz frida-server-${VERSION}-android-arm64.xz

# Install and restart
su -c "cp frida-server-${VERSION}-android-arm64 /data/local/tmp/frida-server.new"
su -c "chmod 755 /data/local/tmp/frida-server.new"
su -c "killall frida-server"
su -c "cp /data/local/tmp/frida-server.new /data/local/tmp/frida-server"
su -c "/data/local/tmp/frida-server -D"

Common Errors and Fixes

Error: sysinfo has no member named loads

  • Cause: Android Bionic libc doesn't have full sysinfo structure
  • Fix: Patch ninja's src/util.cc as described above

Error: SI_LOAD_SHIFT undeclared

  • Cause: Android Bionic doesn't define SI_LOAD_SHIFT
  • Fix: Same patch as above

Error: frida Python extension compiles but import frida fails

  • Cause: frida-core-devkit version doesn't match frida Python version
  • Fix: Download matching versions from GitHub releases

Error: Java VM not available after attach

  • Cause: frida-server version mismatch with Python client
  • Fix: Upgrade frida-server to match Python client version

Error: Process shows as "mmd" instead of "com.tencent.mm"

  • Cause: frida-server version mismatch or multi-process app
  • Fix: Match versions, try different PIDs, or use spawn instead of attach

Version Compatibility Matrix

frida-server frida Python Status
17.9.10 17.15.3 ❌ Java VM unavailable
17.15.3 17.15.3 ✅ Working

Key Takeaways

  1. Always match frida-server and frida Python versions exactly
  2. Android Bionic requires patching ninja source
  3. Java VM availability depends on version matching
  4. Process names may differ from package names (e.g., "mmd" vs "com.tencent.mm")
  5. /data/local/tmp needs write access for frida helper files