Proposal: Post debug Java code to target JVM and exec it

[David Holmes]

On 17/07/2023 12:21 pm, Yi Yang wrote:
> Actually, this is not a very new and strange idea. I often encounter 
> similar needs in actual debugging and troubleshooting. In addition, this 
> feature has also been implemented in some external tools 
> (https://arthas.aliyun.com/en/doc/ognl.html 
> <https://arthas.aliyun.com/en/doc/ognl.html>). It would be very 
> convenient for problem diagnosis if JDK natively had this capability

Okay please take this discussion to serviceability-dev - cc'd.

Thanks,
David

>     ------------------------------------------------------------------
>     From:David Holmes <david.holmes at oracle.com>
>     Send Time:2023 Jul. 13 (Thu.) 20:34
>     To:"YANG, Yi" <qingfeng.yy at alibaba-inc.com>; jdk-dev
>     <jdk-dev at openjdk.org>
>     Subject:Re: Proposal: Post debug Java code to target JVM and exec it
> 
>     On 13/07/2023 8:57 pm, Yi Yang wrote:
>      > Proposal: Post debug Java code to target JVM and exec it
>      >
>      > Hello, I have an early idea to post a piece of Java code to a
>     local JVM
>      > process via jcmd or jexec, like this:
>      >
>      > jcmd VM.exec Debug.java
>      > jexec Debug.java
>      >
>      > class Debug {
>      > public static void main(){
>      > ApplicationProperties c = loadClass("ApplicationProperties");
>      > // print static field of running Java program
>      > System.out.println(ApplicationProperties.field1);
>      > // call static method of running Java program
>      > ApplicationProperties.printAll();
>      > // ....
>      > }
>      > }
>      >
>      > With this approach, we can do some useful things, such as:
>      > - Get some runtime data of the program
>      > - Troubleshoot Java problems
>      > - In rare cases, it can also assist in JVM problem diagnosis
>      >
>      > To protect the attach listener, we may need to create a new
>     ExecThread
>      > at JVM startup and execute Debug::main in this thread. Do you
>     have any
>      > comments on this idea?
> 
>     So basically you want the ability to ask a running JVM to execute an
>     arbitrary piece of java code. That may have some uses but in general
>     seems incredibly dangerous - how would you safeguard such a facility?
> 
>     Cheers,
>     David
> 
>      > Thank you.