SeekToFirst() on non-exsiting collection cause segmentation fault

[Paweł Karczewski]

In the code below I create iterator to the collection before adding any element to it by SSet() method.

TEST_F(EngineBasicTest, TestSeekToFirst) {

  const std::string empty_collection = "Empty";
  ASSERT_EQ(Engine::Open(db_path.c_str(), &engine, configs, stdout),
            Status::Ok);

  auto iter = engine->NewSortedIterator(empty_collection);
  iter->SeekToFirst();
  ASSERT_FALSE(iter->Valid());

}

As Collection would be implicitly created on first SSet() it do not exists during SeekToFirst() call. In that case iter->Valid() should gracefully return false, as such iterator obviously is invalid. Instead of that it cause segmentation fault.

Test output:

localhost/kvdk:build# PMEM_IS_PMEM_FORCE=1 ./dbtest --gtest_filter="*SeekToFirst*"
Note: Google Test filter = *SeekToFirst*
[==========] Running 1 test from 1 test suite.
[----------] Global test environment set-up.
[----------] 1 test from EngineBasicTest
[ RUN      ] EngineBasicTest.TestSeekToFirst
[LOG] time 0 ms: Initializing PMEM size 17179869184 in file /mnt/pmem0/data/data
[LOG] time 2087 ms: Map pmem space done
[LOG] time 2091 ms: In restoring: iterated 0 records
Segmentation fault (core dumped)

[peifeng si]

Hi karczex, Thank you for reporting this issue.

You need to check if the iter is not null before using it:

auto iter = engine->NewSortedIterator(empty_collection);
if (iter) {
  iter->SeekToFirst();
  ASSERT_FALSE(iter->Valid());
}

And your finding reminds us to add checking for "iter" in the tests.cpp, tutorial.cpp and bench.cpp. 👍

Thanks,
Peifeng

[Paweł Karczewski]

Thanks for quick response :)