On Thu, 29 Oct 2026 00:58:32 GMT, David Holmes <[email protected]> wrote:
>> Fixes leaks I could identify on `Agent_OnAttach` failure paths. >> >> In `libinstrument`: >> - the changes to `Agent_OnAttach` are the important ones, while >> `DEF_Agent_OnLoad` and `loadAgent` are updated for hygiene and uniformity >> (when they fail the process gets killed anyway) >> - besides implementing a proper cleanup (the "Improve failure cleanup in >> libinstrument" commit), I also reordered a few agent initialization steps to >> not create the JVM TI environment and change the classpath before the >> provided JAR is validated (the "Reorder blocks to minimize unrecoverable >> changes" commit) >> >> The HotSpot change is adding a destructor to `JvmtiAgent` for the `delete` >> in `JvmtiAgentList::load_agent` executed when `Agent_OnAttach` fails. >> >> Tested by running `jdk_instrument`, `svc_tools`, `vmTestbase_nsk_aod`, >> `hotspot/vmTestbase/nsk/jvmti/AttachOnDemand`, >> `hotspot/serviceability/jvmti/RedefineClasse`. >> >> --------- >> - [x] I confirm that I make this contribution in accordance with the >> [OpenJDK Interim AI Policy](https://openjdk.org/legal/ai). > > src/java.instrument/share/native/libinstrument/InvocationAdapter.c line 259: > >> 257: releaseAndReturn: >> 258: if (result != JNI_OK) { >> 259: destroyJPLISAgent(agent); > > Is this a safe/valid call if the error came from `createNewJPLISAgent`? This is safe because `createNewJPLISAgent` returns either a successfully created agent or `NULL` and `destroyJPLISAgent` early-returns when it receives a `NULL` ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/33608#discussion_r4409615284
