Hi Dinis,
On 29 Mar 2006, at 05:52, Dinis Cruz wrote:
Thanks for confirming this (I wonder how many other other Java
developers are aware of this (especially the ones not focused on
security)).
Most I've worked with aren't really aware of the security manager,
never mind bytecode verification.
It is an issue, but the security risk in the real world may be a bit
overstated. If I were a maliciously minded attacker that wanted
users to execute my evil Java program, I wouldn't need to mess about
with the lack of verification, I could just write evil code in
perfectly verifiable format and rely on users to execute it.
Can anyone come up with attack vectors that exploit lack of
verification on downloaded code that couldn't be exploited by other
(easier) means?
Stephen, do you have any idea of what is the current percentage of
'real
world' Java applications are executed:
a) with verification
b) on a secure sandbox
Very few. As Jeff mentioned some Java Application servers ship with
a security policy enabled, but the policy doesn't restrict anything
(e.g. JBoss), other's show you how to run with a sec policy, but
don't apply it by default (e.g. Tomcat). In some cases, with the
more complex app servers a sec policy would be of little security
benefit because the server needs so much access in order to function
properly that the policy could be considered completely open.
In some ways I think we're applying double standards here. Just
because a virtual machine offers the facility for defining a security
policy and verification doesn't mean that it _has_ to use it. There
are native executable programs that I trust, so why should a program
that runs in a VM be subject to more stringent security controls just
because they're available? IMO whether code needs to be sandboxed
and controlled by a policy should be decided on a case by case basis
rather than a blanket rule.
Note that for example I have seen several Java Based Financial
Applications which are executed on the client which either require
local
installation (via setup.exe / App.msi) or require that the user grants
that Java application more permissions that the ones allocated to a
normal Sandboxed browser based Java App.
This is quite common for an app, and granting more permissions is
fine as long as those are tightly controlled by the java security
policy.
Humm, this is indeed interesting. Ironically, the 1.1 and 2.0 versions
of the CLR will thrown an exception in this case (even in Full Trust).
Since verification is not performed on that .Net Assembly, the CLR
might
pick up this information when it is resolving the method's relative
address into the real physical addresses (i.e. during JIT).
Using the same code with an Applet loaded from the filesystem throws
an IllegalAccessError exception as it should.
What do you mean by 'Applet loaded from the filesystem'?
Where? In a Browser?
If you load an applet in a browser using a url such as: file:///data/
stuff/launch.html then no verification is performed.
But if you access the applet using http/s then it will be verified.
cheers,
--
Stephen de Vries
Corsaire Ltd
E-mail: stephen@corsaire.com
Tel: +44 1483 226014
Fax: +44 1483 226068
Web: http://www.corsaire.com
-------------------------------------------------------------------------
This List Sponsored by: SpiDynamics
ALERT: "How A Hacker Launches A Web Application Attack!"
Step-by-Step - SPI Dynamics White Paper
Learn how to defend against Web Application Attacks with real-world
examples of recent hacking methods such as: SQL Injection, Cross Site
Scripting and Parameter Manipulation
https://download.spidynamics.com/1/ad/web.asp?Campaign_ID=701300000003gRl
--------------------------------------------------------------------------
