Tom Elrod wrote:
Didn't realize this was going to cause a discussion, so we might need to
move this to dev list. That aside, jboss remoting currently handles remote classloading in that
when a invocation request is deserialized on the server (happens same
way on client as well), if it can not find a class within the request
(or result in case of client), it will go back across the wire and
request the class (note that supports bi-directional classloading). This happens until all the classes are loaded on the far end. The
classloader(s) used for the deserialization is controlled within the
AbstractInvoker, where the different classloaders to be used can be
programmatically set. Yes, this does work and has been tested.
This obviously brings up security issues when allowing remote clients to ship bytecode across the wire and is something that needs to be addressed. I do believe that dynamic classloading is needed within remoting for certain cases (know we need it at Vocalocity), so don't want to get rid of it. Would prefer to add capability to ship principal across wire to validate requester if security is turned on (maybe by default). There is already a similar approach for this used within jsr 160 (jmx remoting). This part (security) probably warrants group discussion on list.
The oneway question about performance that started this thread, however,
should be straight forward. I personally feel the way I presented is
acceptable in terms for performance.
I need to reread your question on oneway, but I thought it would be implemented outside of JBoss Remoting and within AOP interceptors.
Bill
-----Original Message-----
From: Scott M Stark [mailto:[EMAIL PROTECTED] Sent: Tuesday, November 25, 2003 10:24 AM
To: Bill Burke
Cc: [EMAIL PROTECTED]
Subject: Re: one way remoting
RMI can do bidirectional class loading if the objects being loaded are associated with a RMI stub so that there is an annotated codebase in the marshalling stream.
You can prevent ClassLoaders from being created and classes from being loaded in specific package namespaces. You can certainly add new security checks to the UCL layer to extend this. Remote classes are not that big an issue when running with a security manager because the permissions are based on where code was loaded from. Priviledged ops are only going to be granted to known server codebases, not arbitrary clients who may be uploading code.
Bill Burke wrote:
Yes, that's what I meant. Nobody currently uses JBoss in this way anyways as we don't support it right now. (Or can RMI do this?).
This
may be pretty dangerous anyways. Can Java security prevent new
classes
from being created?
Bill
-- ================ Bill Burke Chief Architect JBoss Group LLC. ================
------------------------------------------------------- This SF.net email is sponsored by: SF.net Giveback Program. Does SourceForge.net help you be more productive? Does it help you create better code? SHARE THE LOVE, and help us help YOU! Click Here: http://sourceforge.net/donate/ _______________________________________________ JBoss-Development mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/jboss-development
