Re: Iptables discussion

[Joshua Brindle <jbrindle@tresys.com>]

Karl MacMillan wrote:

>>-----Original Message-----
>>From: Ivan Gyurdiev [mailto:gyurdiev@redhat.com]
>>Sent: Friday, July 22, 2005 9:45 AM
>>To: Karl MacMillan
>>Cc: 'Daniel J Walsh'; 'Joshua Brindle'; selinux@tycho.nsa.gov
>>Subject: RE: Iptables discussion
>>
>>
>>    
>>
>>>Solving the user-interface issues can be done more effectively, in my
>>>      
>>>
>>opinion,
>>    
>>
>>>by hiding both iptables and SElinux. My preference would be to extend the
>>>SELinux policy language to be able to express the kind of controls that you
>>>      
>>>
>>are
>>    
>>
>>>interested in expressing and create a configuration tool (gui or text based)
>>>that generates the policy. That would leave a policy.conf or equivalent that
>>>could be analyzed for correctness.
>>>      
>>>
>>I think whatever rules are automatically generated will be at a low
>>level of complexity, because anything else would be better handled by
>>writing policy. Given this, I think it will be trivial to generate
>>additions to policy.conf to address your analysis concern. Where
>>exactly this is done is an implementation detail :)
>>
>>    
>>
>
>Good - after re-reading you and Dan's messages I realized that you might be
>simply suggesting that the iptables program generate SELinux policy. That makes
>sense to me and is consistent with what I am saying about a user-interface. The
>only other comment is that I would like this to be optional. We have a need to
>create systems that have a fixed policy.
>
>  
>
>>What's missing here, is a good API to work with policy, so that you
>>can manipulate policy internals with anything other than checkpolicy.
>>That's why I'm asking for an intermediate representation. My particular
>>implementation of it may not be very good (suggestions welcome),
>>but it's certainly better than what's out there - I don't think
>>passing in policy-dependent integer id's, and exposing internal data
>>structures will make a successful api.
>>    
>>
>
>Why not continue down the path that we have started with libsemanage? I think it
>is better to have a well-defined set of local machine customizations that are
>available via libsemanage than forcing applications to do direct policy
>manipulation. This will, I think, result in an smaller, easier to use API that
>will allow a transition path to the policy server (which will hopefully allow us
>to do these kinds of customizations across multiple machines). It also helps
>with policy updates - by clearly defining local customizations and upstream
>policy it is easier for package managers and admins to handle policy updates.
>  
>
This may cause consistency problems. Adding a iptable rule would add it 
to the current running firewall config (in memory) and may or may not 
store it in a state file to be rerun on boot. If adding that rule 
triggers a policy change that presumably becomes part of the binary 
policy on disk you will end up with alot of extra rules. Also, removing 
rules is much harder than adding them, and it's probably not uncommon to 
delete firewall rules. with respect to analysis, perhaps the analysis 
tools should be able to analyze a linked module format policy and we 
should use modules to add these rules (on the backend behind libsemanage 
ofcourse) that way you can still analyze the policy after iptables has 
added rules :) .

--
This message was distributed to subscribers of the selinux mailing list.
If you no longer wish to subscribe, send mail to majordomo@tycho.nsa.gov with
the words "unsubscribe selinux" without quotes as the message.