Re: RFR: JDK-8147388: Add diagnostic commands to attach JVMTI agent.

[Yasumasa Suenaga]

I uploaded webrev:
  http://cr.openjdk.java.net/~ysuenaga/JDK-8147388/webrev.02/

This webrev can load java agent and native agent through JVMTI.agent_load dcmd.
Could you review again?


Thanks,

Yasumasa


On 2016/01/18 22:20, Dmitry Samersoff wrote:

Yasumasa,

Can we use VLA (Variable Length Arrays) ?

Apple LLVM version 6.1.0 (clang-602.0.53) (based on LLVM 3.6.0svn)
Target: x86_64-apple-darwin14.5.0
Thread model: posix

Compiles it just fine.

-Dmitry

On 2016-01-18 16:09, Yasumasa Suenaga wrote:

Hi Dmitry,

1. It might be better to have one jcmd to run both java and native java
agent. If agent library name ends with ".jar" we can assume it's java
agent.

Okay, I'll try it.

if (_libpath.value() == NULL) {
     error ...
}

I will add it.
However, I note you that _libpath is given mandatory flag.
http://mail.openjdk.java.net/pipermail/serviceability-dev/2016-January/018661.html

char options[option_len];

Can we use VLA (Variable Length Arrays) ?
I'm worried that several C++ compiler cannot compile this code.
http://clang.llvm.org/compatibility.html#vla


Thanks,

Yasumasa


On 2016/01/18 19:38, Dmitry Samersoff wrote:

Yasumasa,

1. It might be better to have one jcmd to run both java and native java
agent. If agent library name ends with ".jar" we can assume it's java
agent.

2. Please get rid of malloc/free and check _libpath.value() for NULL at
ll. 295 and below.


if (_libpath.value() == NULL) {
     error ...
}

if (_option.value() == NULL) {
     JvmtiExport::load_agent_library("instrument",
                  "false", _libpath.value(), output());
    return;
}

size_t option_len = \
     strlen(_libpath.value()) + strlen(_option.value()) + 1;

char options[option_len];

....

-Dmitry


On 2016-01-15 16:33, Yasumasa Suenaga wrote:

Hi,

I added permissions and tests in new webrev:
    http://cr.openjdk.java.net/~ysuenaga/JDK-8147388/webrev.01/

Two tests (LoadAgentDcmdTest, LoadJavaAgentDcmdTest) work fine.


Thanks,

Yasumasa


On 2016/01/15 17:20, Staffan Larsen wrote:

This is a good improvement overall.

The new diagnostic commands need to have proper permissions set:

      static const JavaPermission permission() {
         JavaPermission p =
{"java.lang.management.ManagementPermission",
                            “control", NULL};
        return p;
     }

And as David said: tests! See hotspot/test/serviceability/dcmd/jvmti.

Thanks,
/Staffan

On 14 jan. 2016, at 15:00, Yasumasa Suenaga <yasue...@gmail.com>
wrote:

Hi all,

We can use Attach API to attach JVMTI agent to live process.
However, we have to write Java code for it.

If we can attach JVMTI agents through jcmd, it is very useful.
So I want to add two new diagnostic commands:

* JVMTI.agent_load: Load JVMTI native agent.
* JVMTI.javaagent_load: Load JVMTI java agent.

I maintain two JVMTI agents - HeapStats [1] and JLivePatcher [2].
[1] is native agent, [2] is java agent.
They provide a program for attaching to live process.

I guess that various JVMTI agents provide own attach mechanism like
them.
I think that we should provide general way to attach.

I've uploaded webrev. Could you review it?
    http://cr.openjdk.java.net/~ysuenaga/JDK-8147388/webrev.00/

I'm jdk9 committer, however I cannot access JPRT.
So I need a sponsor.


Thanks,

Yasumasa


[1] http://icedtea.classpath.org/wiki/HeapStats
[2] https://github.com/YaSuenag/jlivepatcher  (in Japanese)