● CVE-2024-31317 / Zero-Click RCE

CVE-2024-31317: Android Stagefright MediaCodec Out-of-Bounds Write Teardown

How an unsigned integer wrap during H.264 NAL unit length decoding in libstagefright enabled zero-click remote code execution via video files.

💡 비유를 통한 쉬운 설명 (ELI5)

Imagine a video projector that reads the size of each video slide from the film itself. A film has a label: 'This slide is 4 billion bytes long'. The projector calculates the size using a small pocket calculator that wraps back around to 12. It creates a tiny frame for 12 bytes, and then the film feeds 4 billion bytes into it, jamming the projector and taking over the movie theater.

핵심 개념 및 서브시스템 용어 해설

libstagefright: The Android C++ multimedia parsing engine responsible for demuxing and decoding MP4, MKV, and streaming video.
NAL Unit (Network Abstraction Layer): The packet format used to deliver H.264/AVC and H.265/HEVC video frames.
media.codec Sandbox: The isolated Linux process where video decoding hardware acceleration runs on Android.
Zero-Click Exploit: An attack that achieves remote code execution without requiring the victim to click a link or open an application.

단계별 공격 실행 메커니즘

Step 1

1. Deliver Malicious Video

The attacker sends a crafted MP4 video via chat or MMS.

Step 2

2. Background Thumbnail Parsing

Android's media scanner automatically demuxes the video without user interaction.

Step 3

3. Integer Wrap in NALU Length

libstagefright adds the NAL unit size to an offset, wrapping past 32 bits.

Step 4

4. Heap Buffer Overwrite & RCE

The decoder writes video data into an undersized heap buffer, achieving code execution.

소스 코드 비교: 치명적 취약점 vs 보안 강화 패치

일반 개발자가 쉽게 이해할 수 있는 고수준 소스 코드로 제공 (어셈블리/바이너리 제외).

패치되지 않은 취약점

// VULNERABLE: frameworks/av/media/libstagefright/MPEG4Extractor.cpp
status_t MPEG4Extractor::parseChunk(off64_t *offset) {
    uint32_t chunk_size = readU32();
    
    // ROOT CAUSE:
    // Raw unsigned 32-bit addition wraps around!
    // If chunk_size is 0xFFFFFFF0, adding header (0x20) results in 0x10 (16 bytes!)
    uint32_t alloc_size = chunk_size + sizeof(ChunkHeader);
    
    uint8_t *buffer = (uint8_t *)malloc(alloc_size);
    // Copies full chunk_size, overflowing buffer!
    mDataSource->readAt(*offset, buffer, chunk_size);
    return OK;
}

보안 강화 패치

// SECURE: frameworks/av/media/libstagefright/MPEG4Extractor.cpp patch
status_t MPEG4Extractor::parseChunk(off64_t *offset) {
    uint32_t chunk_size = readU32();
    
    // 1. Enforce safe integer arithmetic check using __builtin_add_overflow
    uint32_t alloc_size;
    if (__builtin_add_overflow(chunk_size, sizeof(ChunkHeader), &alloc_size)) {
        return ERROR_MALFORMED; // Abort on integer overflow
    }
    
    // 2. Impose strict maximum upper bound ceiling on media chunks
    if (alloc_size > MAX_MEDIA_CHUNK_SIZE) {
        return ERROR_MALFORMED;
    }
    
    uint8_t *buffer = (uint8_t *)malloc(alloc_size);
    mDataSource->readAt(*offset, buffer, chunk_size);
    return OK;
}

엔지니어링 보안 강화 체크리스트

Use compiler built-in overflow arithmetic (`__builtin_add_overflow`) when calculating media buffer allocations.
Impose explicit upper bound ceilings on parsed chunk and NAL unit sizes before invoking `malloc()`.
Ensure media parsing processes operate under strict seccomp-bpf system call filters.
Fuzz media parsers continuously with AFL++ and LibFuzzer using diverse MP4/MKV corpora.

← 전체 보안 디렉터리 보기 모든 플랫폼 보안 업데이트 →