[ORC-414] [C++] ORC files with malformed protobuf objects can crash a release build

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.5.3
• Fix Version/s: 1.5.7, 1.6.1, 1.7.0
• Component/s: C++
• Assignee: Quanlong Huang
• Reporter: Quanlong Huang
• Created: 09/Oct/18 12:39
• Updated: 27/Oct/19 06:32
• Resolved: 22/Oct/19 23:19

Description

ORC files can be produced by any external tools. Some corrupt files may contain malformed protobuf objects which can crash the process. The attachment is an example.

In a debug build, protobuf will throw exceptions for this file:

$ build/tools/src/orc-scan maleformed_protobuf.orc 
[libprotobuf FATAL /mnt/volume1/impala-orc/orc/build/c++/libs/thirdparty/protobuf_ep-install/include/google/protobuf/repeated_field.h:1522] CHECK failed: (index) < (current_size_): 
Caught exception in maleformed_protobuf.orc: CHECK failed: (index) < (current_size_):

It hits a DCHECK which is eliminated in a release build.

1518 template <typename TypeHandler>
1519 inline const typename TypeHandler::Type&
1520 RepeatedPtrFieldBase::Get(int index) const {
1521   GOOGLE_DCHECK_GE(index, 0);
1522   GOOGLE_DCHECK_LT(index, current_size_);
1523   return *cast<TypeHandler>(rep_->elements[index]);
1524 }

In a release build, the process crash immediately, which means any system integrated with the orc-lib will crash when processing such kind of files.

$ build/tools/src/orc-scan maleformed_protobuf.orc 
Segmentation fault (core dumped)

The stacktrace for this crash:

#0  0x0000000000588c1e in orc::ReaderImpl::ReaderImpl(std::shared_ptr<orc::FileContents>, orc::ReaderOptions const&, unsigned long, unsigned long) ()
#1  0x000000000058b1ee in orc::createReader(std::unique_ptr<orc::InputStream, std::default_delete<orc::InputStream> >, orc::ReaderOptions const&) ()
#2  0x00000000005847c0 in scanFile (out=..., filename=0x7ffcf03a173d "maleformed_protobuf.orc", batchSize=batchSize@entry=1024) at /mnt/volume1/impala-orc/orc/tools/src/FileScan.cc:32
#3  0x0000000000584150 in main (argc=<optimized out>, argv=<optimized out>) at /mnt/volume1/impala-orc/orc/tools/src/FileScan.cc:84

We may need to introduce checksums to avoid this.

Issue Links

• blocks: IMPALA-6772 Enable test_scanners_fuzz for ORC format
• links to: GitHub Pull Request #438