K-ANOY opened a new issue, #1205:
URL: https://github.com/apache/arrow-java/issues/1205
### Describe the bug, including details regarding any error messages,
version, and platform.
I found a possible JNI array leak in `FromSchemaByteArray()` when the
serialized schema cannot be parsed.
File: `dataset/src/main/cpp/jni_util.cc`
Function: `FromSchemaByteArray`
Relevant code:
```cpp
jbyte* schemaBytes_data =
env->GetByteArrayElements(schemaBytes, nullptr);
auto serialized_schema = std::make_shared<arrow::Buffer>(
reinterpret_cast<uint8_t*>(schemaBytes_data),
schemaBytes_len);
arrow::io::BufferReader buf_reader(serialized_schema);
ARROW_ASSIGN_OR_RAISE(
std::shared_ptr<arrow::Schema> schema,
arrow::ipc::ReadSchema(&buf_reader, &in_memo))
env->ReleaseByteArrayElements(
schemaBytes, schemaBytes_data, JNI_ABORT);
return schema;
```
`GetByteArrayElements()` returns a pointer that must be paired with
`ReleaseByteArrayElements()`.
`ARROW_ASSIGN_OR_RAISE` returns immediately when `ReadSchema()` returns an
error. On that path, the release below the macro is skipped, so the acquired
array elements remain unreleased:
```text
GetByteArrayElements succeeds
-> ReadSchema returns an error
-> ARROW_ASSIGN_OR_RAISE returns
-> ReleaseByteArrayElements is skipped
```
The function is used by the public native `createDataset()` method:
```cpp
schema = JniGetOrThrow(
FromSchemaByteArray(env, schema_bytes));
```
Malformed, corrupted, or incompatible serialized schema bytes can therefore
reach this path. Repeated failed calls can retain copied array buffers or
keep
Java arrays pinned, depending on the JVM implementation.
Suggested fix: release the elements before propagating the parse result, for
example:
```cpp
auto schema_result =
arrow::ipc::ReadSchema(&buf_reader, &in_memo);
env->ReleaseByteArrayElements(
schemaBytes, schemaBytes_data, JNI_ABORT);
return schema_result;
```
An RAII guard for `schemaBytes_data` would also ensure release if additional
early returns are introduced later.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
