MKV与MatroskaExtractor

mkv封装格式相关内容可以参考：MKV 文件格式解析_Martyn哥的博客-CSDN博客_mkv 格式解析

本篇主要是用来记录MatroskaExtractor是如何解析EBML的，如何解析EBML header以及SEGMENT的

构造函数

先看构造函数，android MKV文件的解析需要依赖外部库mkvparser，首先会去解析EBML header（这个部分在sniff中会先调用一次判断是否为mkv文件）

MatroskaExtractor::MatroskaExtractor(DataSourceHelper *source)
    : mDataSource(source),
      mReader(new DataSourceBaseReader(mDataSource)),
      mSegment(NULL),
      mExtractedThumbnails(false),
      mIsWebm(false),
      mSeekPreRollNs(0) {
    off64_t size;
    mIsLiveStreaming =
        (mDataSource->flags()
            & (DataSourceBase::kWantsPrefetching
                | DataSourceBase::kIsCachingDataSource))
        && mDataSource->getSize(&size) != OK;

    // 1、解析EBML header
    mkvparser::EBMLHeader ebmlHeader;
    long long pos;
    if (ebmlHeader.Parse(mReader, pos) < 0) {
        return;
    }

    if (ebmlHeader.m_docType && !strcmp("webm", ebmlHeader.m_docType)) {
        mIsWebm = true;
    }

    // 2、创建Segment
    long long ret =
        mkvparser::Segment::CreateInstance(mReader, pos, mSegment);

    if (ret) {
        CHECK(mSegment == NULL);
        return;
    }

    if (mIsLiveStreaming) {
        // from mkvparser::Segment::Load(), but stop at first cluster
        ret = mSegment->ParseHeaders();
        if (ret == 0) {
            long len;
            ret = mSegment->LoadCluster(pos, len);
            if (ret >= 1) {
                // no more clusters
                ret = 0;
            }
        } else if (ret > 0) {
            ret = mkvparser::E_BUFFER_NOT_FULL;
        }
    } else {
        // 3、parse Segment header
        ret = mSegment->ParseHeaders();
        if (ret < 0) {
            ALOGE("Segment parse header return fail %lld", ret);
            delete mSegment;
            mSegment = NULL;
            return;
        } else if (ret == 0) {
            // 找到Cues
            const mkvparser::Cues* mCues = mSegment->GetCues();
            // 找到SeekHead
            const mkvparser::SeekHead* mSH = mSegment->GetSeekHead();
            if ((mCues == NULL) && (mSH != NULL)) {
                size_t count = mSH->GetCount();
                const mkvparser::SeekHead::Entry* mEntry;
                for (size_t index = 0; index < count; index++) {
                    mEntry = mSH->GetEntry(index);
                    if (mEntry->id == libwebm::kMkvCues) { // Cues ID
                        long len;
                        long long pos;
                        mSegment->ParseCues(mEntry->pos, pos, len);
                        mCues = mSegment->GetCues();
                        ALOGV("find cue data by seekhead");
                        break;
                    }
                }
            }

            if (mCues) {
                long len;
                ret = mSegment->LoadCluster(pos, len);
                ALOGV("has Cue data, Cluster num=%ld", mSegment->GetCount());
            } else  {
                long status_Load = mSegment->Load();
                ALOGW("no Cue data,Segment Load status:%ld",status_Load);
            }
        } else if (ret > 0) {
            ret = mkvparser::E_BUFFER_NOT_FULL;
        }
    }

    if (ret < 0) {
        char uri[1024];
        if(!mDataSource->getUri(uri, sizeof(uri))) {
            uri[0] = '\0';
        }
        ALOGW("Corrupt %s source: %s", mIsWebm ? "webm" : "matroska",
                uriDebugString(uri).c_str());
        delete mSegment;
        mSegment = NULL;
        return;
    }

    // 获取tracks
    addTracks();
}

EBMLHeader::Parse

下图是EMBL Tree解析的mkv文件，EBML_ReadVersion是element id（0x47），对应红色部分；后面的s(1):0x1，括号里的1表示数据长度，冒号后面的0x01表示数据内容的长度为1，对应绿色部分（这里的计算比较复杂一些，使用的是ReadUInt）；最后面的0x01表示数据内容；

long long EBMLHeader::Parse(IMkvReader* pReader, long long& pos) {
  if (!pReader)
    return E_FILE_FORMAT_INVALID;

  long long total, available;

  long status = pReader->Length(&total, &available);

  if (status < 0)  // error
    return status;

  pos = 0;

  // 检查0x1A的位置
  while (pos < kMaxScanBytes) {
    status = pReader->Read(pos, 1, &scan_byte);

    if (status < 0)  // error
      return status;
    else if (status > 0)
      return E_BUFFER_NOT_FULL;

    if (scan_byte == kEbmlByte0)
      break;

    ++pos;
  }

  long len = 0;
  // 读取ID
  const long long ebml_id = ReadID(pReader, pos, len);

  if (ebml_id == E_BUFFER_NOT_FULL)
    return E_BUFFER_NOT_FULL;

  // 检查ID是否为0x1A45DFA3，mkv文件会以这四个字节为开头
  if (len != 4 || ebml_id != libwebm::kMkvEBML)
    return E_FILE_FORMAT_INVALID;

  // Move read pos forward to the EBML header size field.
  // 修改文件读取偏移量
  pos += 4;

  // Read length of size field.
  // 接下来的一个字节表示数据长度，下面这部分到ReadUInt之前都是在检查数据长度是否正常
  long long result = GetUIntLength(pReader, pos, len);

  if (result < 0)  // error
    return E_FILE_FORMAT_INVALID;
  else if (result > 0)  // need more data
    return E_BUFFER_NOT_FULL;

  if (len < 1 || len > 8)
    return E_FILE_FORMAT_INVALID;

  if ((total >= 0) && ((total - pos) < len))
    return E_FILE_FORMAT_INVALID;

  if ((available - pos) < len)
    return pos + len;  // try again later

  // 真正读取数据长度
  // Read the EBML header size.
  result = ReadUInt(pReader, pos, len);

  if (result < 0)  // error
    return result;

  pos += len;  // consume size field

  // pos now designates start of payload

  if ((total >= 0) && ((total - pos) < result))
    return E_FILE_FORMAT_INVALID;

  if ((available - pos) < result)
    return pos + result;

  const long long end = pos + result;

  Init();

  while (pos < end) {
    long long id, size;

    status = ParseElementHeader(pReader, pos, end, id, size);

    if (status < 0)  // error
      return status;

    if (size == 0)
      return E_FILE_FORMAT_INVALID;

    if (id == libwebm::kMkvEBMLVersion) {
      m_version = UnserializeUInt(pReader, pos, size);

      if (m_version <= 0)
        return E_FILE_FORMAT_INVALID;
    } else if (id == libwebm::kMkvEBMLReadVersion) {
      m_readVersion = UnserializeUInt(pReader, pos, size);

      if (m_readVersion <= 0)
        return E_FILE_FORMAT_INVALID;
    } else if (id == libwebm::kMkvEBMLMaxIDLength) {
      m_maxIdLength = UnserializeUInt(pReader, pos, size);

      if (m_maxIdLength <= 0)
        return E_FILE_FORMAT_INVALID;
    } else if (id == libwebm::kMkvEBMLMaxSizeLength) {
      m_maxSizeLength = UnserializeUInt(pReader, pos, size);

      if (m_maxSizeLength <= 0)
        return E_FILE_FORMAT_INVALID;
    } else if (id == libwebm::kMkvDocType) {
      if (m_docType)
        return E_FILE_FORMAT_INVALID;

      status = UnserializeString(pReader, pos, size, m_docType);

      if (status)  // error
        return status;
    } else if (id == libwebm::kMkvDocTypeVersion) {
      m_docTypeVersion = UnserializeUInt(pReader, pos, size);

      if (m_docTypeVersion <= 0)
        return E_FILE_FORMAT_INVALID;
    } else if (id == libwebm::kMkvDocTypeReadVersion) {
      m_docTypeReadVersion = UnserializeUInt(pReader, pos, size);

      if (m_docTypeReadVersion <= 0)
        return E_FILE_FORMAT_INVALID;
    }

    pos += size;
  }

  if (pos != end)
    return E_FILE_FORMAT_INVALID;

  // Make sure DocType, DocTypeReadVersion, and DocTypeVersion are valid.
  if (m_docType == NULL || m_docTypeReadVersion <= 0 || m_docTypeVersion <= 0)
    return E_FILE_FORMAT_INVALID;

  // Make sure EBMLMaxIDLength and EBMLMaxSizeLength are valid.
  if (m_maxIdLength <= 0 || m_maxIdLength > 4 || m_maxSizeLength <= 0 ||
      m_maxSizeLength > 8)
    return E_FILE_FORMAT_INVALID;

  return 0;
}

long long ReadID(IMkvReader* pReader, long long pos, long& len) {
  if (pReader == NULL || pos < 0)
    return E_FILE_FORMAT_INVALID;

  // Read the first byte. The length in bytes of the ID is determined by
  // finding the first set bit in the first byte of the ID.
  unsigned char temp_byte = 0;
  // 先读取一个字节，读到得应该是0x1A
  int read_status = pReader->Read(pos, 1, &temp_byte);

  if (read_status < 0)
    return E_FILE_FORMAT_INVALID;
  else if (read_status > 0)  // No data to read.
    return E_BUFFER_NOT_FULL;

  if (temp_byte == 0)  // ID length > 8 bytes; invalid file.
    return E_FILE_FORMAT_INVALID;

  int bit_pos = 0;
  const int kMaxIdLengthInBytes = 4;
  const int kCheckByte = 0x80;

  // Find the first bit that's set.
  // 这里会对应到博文里面的 ID数据长度等于起始0的个数加1
  bool found_bit = false;
  for (; bit_pos < kMaxIdLengthInBytes; ++bit_pos) {
    if ((kCheckByte >> bit_pos) & temp_byte) {   // 用0x80不断向右移位，并且和读到数据与运算，计算0的个数
      found_bit = true;
      break;
    }
  }

  if (!found_bit) {
    // The value is too large to be a valid ID.
    return E_FILE_FORMAT_INVALID;
  }

  // Read the remaining bytes of the ID (if any).
  const int id_length = bit_pos + 1;      // ID长度等于起始0的个数加1
  long long ebml_id = temp_byte;
  for (int i = 1; i < id_length; ++i) {
    ebml_id <<= 8;
    read_status = pReader->Read(pos + i, 1, &temp_byte);

    if (read_status < 0)
      return E_FILE_FORMAT_INVALID;
    else if (read_status > 0)
      return E_BUFFER_NOT_FULL;

    ebml_id |= temp_byte;   // 继续往后读取ID，并做与运算，返回真正的ID
  }

  len = id_length;
  return ebml_id;
}

long long GetUIntLength(IMkvReader* pReader, long long pos, long& len) {
  if (!pReader || pos < 0)
    return E_FILE_FORMAT_INVALID;

  long long total, available;

  int status = pReader->Length(&total, &available);
  if (status < 0 || (total >= 0 && available > total))
    return E_FILE_FORMAT_INVALID;

  len = 1;

  if (pos >= available)
    return pos;  // too few bytes available

  unsigned char b;

  // 读取下一个字节 0xA3
  status = pReader->Read(pos, 1, &b);

  if (status != 0)
    return status;

  if (b == 0)  // we can't handle u-int values larger than 8 bytes
    return E_FILE_FORMAT_INVALID;

  unsigned char m = 0x80;

  // 数据长度为起始0的个数 + 1
  while (!(b & m)) {
    m >>= 1;
    ++len;
  }

  return 0;  // success
}

long long ReadUInt(IMkvReader* pReader, long long pos, long& len) {
  if (!pReader || pos < 0)
    return E_FILE_FORMAT_INVALID;

  len = 1;
  unsigned char b;
  int status = pReader->Read(pos, 1, &b);

  if (status < 0)  // error or underflow
    return status;

  if (status > 0)  // interpreted as "underflow"
    return E_BUFFER_NOT_FULL;

  if (b == 0)  // we can't handle u-int values larger than 8 bytes
    return E_FILE_FORMAT_INVALID;

  unsigned char m = 0x80;

  // 数据长度为起始0的个数 + 1（这个长度为数据长度的长度）
  while (!(b & m)) {
    m >>= 1;
    ++len;
  }

  // result为数据内容的长度，第一个字节需要用m取反并做与运算
  long long result = b & (~m);
  ++pos;

  for (int i = 1; i < len; ++i) {
    status = pReader->Read(pos, 1, &b);

    if (status < 0) {
      len = 1;
      return status;
    }

    if (status > 0) {
      len = 1;
      return E_BUFFER_NOT_FULL;
    }

    result <<= 8;
    result |= b;

    ++pos;
  }

  return result;
}

Segment::CreateInstance

看完上面的内容，Segment是如何创建的就很简单了

long long Segment::CreateInstance(IMkvReader* pReader, long long pos,
                                  Segment*& pSegment) {
  if (pReader == NULL || pos < 0)
    return E_PARSE_FAILED;

  pSegment = NULL;

  long long total, available;

  const long status = pReader->Length(&total, &available);

  if (status < 0)  // error
    return status;

  if (available < 0)
    return -1;

  if ((total >= 0) && (available > total))
    return -1;

  for (;;) {
    if ((total >= 0) && (pos >= total))
      return E_FILE_FORMAT_INVALID;

    // Read ID
    long len;
    long long result = GetUIntLength(pReader, pos, len);

    if (result)  // error, or too few available bytes
      return result;

    if ((total >= 0) && ((pos + len) > total))
      return E_FILE_FORMAT_INVALID;

    if ((pos + len) > available)
      return pos + len;

    const long long idpos = pos;
    // 读取element id
    const long long id = ReadID(pReader, pos, len);

    if (id < 0)
      return E_FILE_FORMAT_INVALID;

    pos += len;  // consume ID

    // Read Size
    // 下面是老套的数据长度检查
    result = GetUIntLength(pReader, pos, len);

    if (result)  // error, or too few available bytes
      return result;

    if ((total >= 0) && ((pos + len) > total))
      return E_FILE_FORMAT_INVALID;

    if ((pos + len) > available)
      return pos + len;
    // 读取数据内容长度
    long long size = ReadUInt(pReader, pos, len);

    if (size < 0)  // error
      return size;

    pos += len;  // consume length of size of element

    // Pos now points to start of payload

    // Handle "unknown size" for live streaming of webm files.
    const long long unknown_size = (1LL << (7 * len)) - 1;

    // Segment的id为0x18 53 80 67
    if (id == libwebm::kMkvSegment) {
      if (size == unknown_size)
        size = -1;

      else if (total < 0)
        size = -1;

      else if ((pos + size) > total)
        size = -1;
      // 创建Segment，传入参数为Segment element起始位置、数据内容开始位置、数据长度
      pSegment = new (std::nothrow) Segment(pReader, idpos, pos, size);
      if (pSegment == NULL)
        return E_PARSE_FAILED;

      return 0;  // success
    }

    if (size == unknown_size)
      return E_FILE_FORMAT_INVALID;

    if ((total >= 0) && ((pos + size) > total))
      return E_FILE_FORMAT_INVALID;

    if ((pos + size) > available)
      return pos + size;

    pos += size;  // consume payload
  }
}

Segment::ParseHeaders

这个方法用于parse Segment中的track、seekhead等信息，比较重要，它会找到对应的element id以及数据，如何解析数据的这里就不贴代码了，如果碰到问题可以去查看相关的parse方法

long long Segment::ParseHeaders() {
  // Outermost (level 0) segment object has been constructed,
  // and pos designates start of payload.  We need to find the
  // inner (level 1) elements.
  long long total, available;

  const int status = m_pReader->Length(&total, &available);

  if (status < 0)  // error
    return status;

  if (total > 0 && available > total)
    return E_FILE_FORMAT_INVALID;

  const long long segment_stop = (m_size < 0) ? -1 : m_start + m_size;

  if ((segment_stop >= 0 && total >= 0 && segment_stop > total) ||
      (segment_stop >= 0 && m_pos > segment_stop)) {
    return E_FILE_FORMAT_INVALID;
  }

  for (;;) {
    if ((total >= 0) && (m_pos >= total))
      break;

    if ((segment_stop >= 0) && (m_pos >= segment_stop))
      break;

    long long pos = m_pos;
    const long long element_start = pos;

    // Avoid rolling over pos when very close to LLONG_MAX.
    unsigned long long rollover_check = pos + 1ULL;
    if (rollover_check > LLONG_MAX)
      return E_FILE_FORMAT_INVALID;

    if ((pos + 1) > available)
      return (pos + 1);

    // 很熟悉，检查数据长度
    long len;
    long long result = GetUIntLength(m_pReader, pos, len);

    if (result < 0)  // error
      return result;

    if (result > 0) {
      // MkvReader doesn't have enough data to satisfy this read attempt.
      return (pos + 1);
    }

    if ((segment_stop >= 0) && ((pos + len) > segment_stop))
      return E_FILE_FORMAT_INVALID;

    if ((pos + len) > available)
      return pos + len;

    const long long idpos = pos;
    // 读取ID
    const long long id = ReadID(m_pReader, idpos, len);

    if (id < 0)
      return E_FILE_FORMAT_INVALID;

    // 如果ID 为Cluster则结束本次循环
    if (id == libwebm::kMkvCluster)
      break;

    pos += len;  // consume ID

    if ((pos + 1) > available)
      return (pos + 1);

    // Read Size
    // 检查数据长度
    result = GetUIntLength(m_pReader, pos, len);

    if (result < 0)  // error
      return result;

    if (result > 0) {
      // MkvReader doesn't have enough data to satisfy this read attempt.
      return (pos + 1);
    }

    if ((segment_stop >= 0) && ((pos + len) > segment_stop))
      return E_FILE_FORMAT_INVALID;

    if ((pos + len) > available)
      return pos + len;

    // 读取数据内容长度
    const long long size = ReadUInt(m_pReader, pos, len);

    if (size < 0 || len < 1 || len > 8) {
      // TODO(tomfinegan): ReadUInt should return an error when len is < 1 or
      // len > 8 is true instead of checking this _everywhere_.
      return size;
    }

    pos += len;  // consume length of size of element

    // Avoid rolling over pos when very close to LLONG_MAX.
    rollover_check = static_cast<unsigned long long>(pos) + size;
    if (rollover_check > LLONG_MAX)
      return E_FILE_FORMAT_INVALID;

    const long long element_size = size + pos - element_start;

    // Pos now points to start of payload

    if ((segment_stop >= 0) && ((pos + size) > segment_stop))
      return E_FILE_FORMAT_INVALID;

    // We read EBML elements either in total or nothing at all.

    if ((pos + size) > available)
      return pos + size;

    // 如果ID为Info，则去创建SegmentInfo
    if (id == libwebm::kMkvInfo) {
      if (m_pInfo)
        return E_FILE_FORMAT_INVALID;

      m_pInfo = new (std::nothrow)
          SegmentInfo(this, pos, size, element_start, element_size);

      if (m_pInfo == NULL)
        return -1;

      const long status = m_pInfo->Parse();

      if (status)
        return status;
    // 如果ID为Track，则去创建Track
    } else if (id == libwebm::kMkvTracks) {
      if (m_pTracks)
        return E_FILE_FORMAT_INVALID;

      m_pTracks = new (std::nothrow)
          Tracks(this, pos, size, element_start, element_size);

      if (m_pTracks == NULL)
        return -1;
      // 调用Tracks.parse方法来解析Track信息
      const long status = m_pTracks->Parse();

      if (status)
        return status;
    // 如果ID为Cues，则创建Cues对象
    } else if (id == libwebm::kMkvCues) {
      if (m_pCues == NULL) {
        m_pCues = new (std::nothrow)
            Cues(this, pos, size, element_start, element_size);

        if (m_pCues == NULL)
          return -1;
      }
    // 如果id为SeekHead，那么创建SeekHead对象
    } else if (id == libwebm::kMkvSeekHead) {
      if (m_pSeekHead == NULL) {
        m_pSeekHead = new (std::nothrow)
            SeekHead(this, pos, size, element_start, element_size);

        if (m_pSeekHead == NULL)
          return -1;

        const long status = m_pSeekHead->Parse();

        if (status)
          return status;
      }
    } else if (id == libwebm::kMkvChapters) {
      if (m_pChapters == NULL) {
        m_pChapters = new (std::nothrow)
            Chapters(this, pos, size, element_start, element_size);

        if (m_pChapters == NULL)
          return -1;

        const long status = m_pChapters->Parse();

        if (status)
          return status;
      }
    } else if (id == libwebm::kMkvTags) {
      if (m_pTags == NULL) {
        m_pTags = new (std::nothrow)
            Tags(this, pos, size, element_start, element_size);

        if (m_pTags == NULL)
          return -1;

        const long status = m_pTags->Parse();

        if (status)
          return status;
      }
    }

    m_pos = pos + size;  // consume payload
  }

  if (segment_stop >= 0 && m_pos > segment_stop)
    return E_FILE_FORMAT_INVALID;

  if (m_pInfo == NULL)  // TODO: liberalize this behavior
    return E_FILE_FORMAT_INVALID;

  if (m_pTracks == NULL)
    return E_FILE_FORMAT_INVALID;

  return 0;  // success
}

本篇大致就先到这，了解了MKV是如何解析EBML的，后续如果有问题就很容易去定位了。