river-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Peter <j...@zeus.net.au>
Subject Re: [VOTE] Release Apache River 3.0.0
Date Fri, 07 Oct 2016 02:25:01 GMT
The question is of course where to next?

As people are aware I've been working on addressing security issues and 
how to make River better and more secure.  I've been working on this 
outside the project because my attempts to discuss it caused heated 
arguments.  I figured that if I could demonstrate a working example that 
people could try out, it could allevieate any misunderstandings that may 
have developed due to language or culture differences.

River's approach to security (a result of the Jini Davis project) is 
quite complex and contains a flaw borne out of two limitations around 
the time it was developed:

   1. The assumption that the Java sandbox and java serialization was
      secure (we know today this isn't the case).
   2. Interfaces cannot be changed (no longer true with java 8), in this
      case ServiceRegistrar was designed prior to the later added on
      security features.

What's wrong with River's approach?

Answer:  It validates and authenticates after downloading code and 
deserializing untrusted data, using the proxy trust framework, by asking 
an already downloaded and deserialized service proxy for a bootstrap 
proxy, the client code then uses the boostrap proxy to determine if the 
service proxy can be trusted.  Too little too late.  Why not instead 
recieve a bootstrap proxy, deserialized using input validation, without 
code download, authenticate the remote endpoint, then ask the bootstrap 
proxy for the service proxy?

The trouble with the existing approach today is an attacker has 
opportunity to take control of a computer using deserialization alone.  
For those who think a network firewall is sufficient protection and the 
flawed security design isn't an issue on local networks, even in air 
gapped networks, an attacker can leave a USB key in a car park 
containing malware that looks for network transmissions that contain 
java serialized data, hoping that someone will pick it up and plug it 
into their pc, the malware will send serial data containing a gadget 
attack to a listening network port that accepts java serial data and 
take over the infected computer.

All network communications using standard java serialization must be 
both authenticated and encrypted, prior to reading in any data to ensure 
that the data is trusted.

I think we can all accept that these vulnerabilities exist and googling 
java serialization gadget attacks should convince even the most doubtful 

Relevant links:

I would like the opportunity to explain more, and go through working 
examples of solutions before we start arguing about whether we should 
solve these problems.  Anyone reading the Apache Way will realise that 
security is a mandatory feature of Apache Software and therefore we 
should consider how we should fix existing security issues in River and 
while doing so, take the opportunity to make security simpler to 
implement.  Arguments should not be about the relevance of security 
issues, but rather the suitablility of solutions.



On 6/10/2016 2:04 PM, Bryan Thompson wrote:
> Excellent.  A great step.
> Bryan
> On Wednesday, October 5, 2016, Peter Firmstone<peter.firmstone@zeus.net.au>
> wrote:
>> Results:
>> 3 binding votes
>> 1 non binding
>> None against.
>> The artifacts have been published, we need to wait 24 hours before
>> announcing.
>> Regards,
>> Peter.
>> Sent from my Samsung device.

View raw message