[ARROW-11518] [C++] [Parquet] Parquet reader crashes when reading boolean columns

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.0.0
• Fix Version/s: 6.0.0
• Component/s: C++
• Labels: pull-request-available
• External issue URL: https://github.com/apache/arrow/issues/27394
• Assignee: Andrey Klochkov
• Reporter: Andrey Klochkov
• Created: 06/Feb/21 07:28
• Updated: 11/Jan/23 08:20
• Resolved: 27/Jul/21 14:36

Description

Parquet file reader crashes while reading boolean columns in TypedColumnReaderImpl<DType>::Skip.

The calculation of the buffer size in the code below is not correct as value_byte_size is 1 for booleans, and the same buffer is used for definition and repetition levels data which requires 2 bytes per value.

     // This will be enough scratch space to accommodate 16-bit levels or any
      // value type
      std::shared_ptr<ResizableBuffer> scratch = AllocateBuffer(
          this->pool_, batch_size * type_traits<DType::type_num>::value_byte_size);

      do {
        batch_size = std::min(batch_size, rows_to_skip);
        values_read =
            ReadBatch(static_cast<int>(batch_size),
                      reinterpret_cast<int16_t*>(scratch->mutable_data()),
                      reinterpret_cast<int16_t*>(scratch->mutable_data()),
                      reinterpret_cast<T*>(scratch->mutable_data()), &values_read);
        rows_to_skip -= values_read;
      } while (values_read > 0 && rows_to_skip > 0);

Issue Links

• links to: GitHub Pull Request #9432