On 12/03/13 03:38, Tom Eastep wrote:
> On 3/11/13 5:05 PM, "[email protected]" <[email protected]> wrote:
>
>> Hi
>>
>> On Mon, Mar 11, 2013, at 06:32 PM, Tom Eastep wrote:
>>> I would also like to recommend separate LANs for the servers and other
>>> client systems. I dislike not having a firewall between the two, because
>>> your internet-facing servers are the most likely targets of hackers.
>> Logically separate LANs are certainly an option.  As for physically
>> separated, for now, we're interface limited.
>>
>> So, iiuc, something like:
>>
>> --------------------------------
>> Firewall Box:
>>      ext intfc: eth0, x.x.x.17/24, Gateway x.x.x.1
>>              DNS daemon, listen @ 192.168.0.100/24
>>      int intfc: eth1, 172.16.1.100/24
>>
>> Mail Server Box:
>>      intfc: eth0, 192.168.1.25/24
>>      
>> Xen Server Box:
>>      Dom0
>>              intfc: eth0, 172.16.1.200/24
>>              br0 -> 
>>                      DomU1
>>                              intfc: eth0, 10.0.1.200/24
>>                      DomU2
>>                              intfc: eth0, 10.0.2.200/24
>>
>> Desktops
>>      intfc: eth0, 172.16.1.XX
>> --------------------------------
>>
>> would provide (some of) that separation ?
> Some (but not much). Broadcast traffic on the LAN is visible to all hosts,
> so each of the subnets is quickly exposed to the other.
>
> -Tom
> You do not need a parachute to skydive. You only need a parachute to
> skydive twice.
>
>
>
>
>
> ------------------------------------------------------------------------------
> Symantec Endpoint Protection 12 positioned as A LEADER in The Forrester  
> Wave(TM): Endpoint Security, Q1 2013 and "remains a good choice" in the  
> endpoint security space. For insight on selecting the right partner to 
> tackle endpoint security challenges, access the full report. 
> http://p.sf.net/sfu/symantec-dev2dev
> _______________________________________________
> Shorewall-users mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/shorewall-users
Does that apply equally if using properly configured vlan tagging etc? 
Of course I'm aware that even with placing some local rules to enforce
it a fully root compromised box would allow them to enable, disable,
change or otherwise play games with your boxes vlan setup as much as
they wanted to.  I'm also tend to think when it comes down to that,
firstly it would act to limit the potential attack vectors from the
simply logically separated idea where probably the most basic forms of
attack on a vulnerable PHP/Rails/CGI web application could I suspect be
sufficient to enable manipulation of services accepting broadcast
traffic even while the http daemon and the rest of the system remain
secure and with no more privileges than they were supposed to have.  If
manipulating such as vlan tags and/or disabling iptables/selinux or
similar policy enforcement regulating outgoing traffic is required I'd
have thought some form of system wide compromise with privilege
escalation would be a minimum (For the sake of argument assuming that
all internet facing servers have fully dropped root and not merely
switched euid or similar uid=0).

If I'm right on that part I would personally be inclined to consider
that to be reasonably acceptable especially on a temporary basis, mainly
because to my mind once a hostile agent has successfully managed to gain
root on a local system you have bigger problems than broadcast
disruption to worry about, especially if the compromised machine is one
that is trusted by internal clients to serve content or handle sensitive
data etc which usually be pretty much all of them in my experience often
to the point of seeming irrational.  Actually it's a little bit off on a
tangent but saying that reminds me of setup I came across earlier this
year which just seems like a fitting example how never to set up a
firewall.  It's interesting in a way because it's mostly well set up
everything on it's own interfaces and independent only for a badly
thought out policy rule to turn it into a disaster waiting to happen.

It does go off on a tangent so if it doesn't interest you it's safe to
just skip reading from this paragraph onwards.

Basic topology:
Internet connectivity through a pair of 155Mbps fibre connections to two
different providers (a stones through from LINX is a lucky place to have
an office) each terminating on one of two physical gateways with a
gigabit ethernet crosslink between them for load balancing and
fallover.  Filtered into an internal network of three basic zones a
group of internal servers predominantly an oracle DB cluster but also
the usual like In/Outbound Mail, Directory services.  Second subnet of
servers consists of a group of HTTP servers being used to serve both the
the external website and the intranet site, 2 MX relays which handle
mail actually entering and leaving and redundant set of proxy servers
that handle outbound traffic.  The final zone has ~650 employee
workstations sub-netted to departments but the interdepartmental routers
don't do any filtering two or three iptables rules that look like
someone's band-aid job that nobody remembers why they were even added.

Topology all looks great until you look at the actual filtering rules it
is set up such that only the second group of servers communicates with
external hosts, filtering is well set up between the two sets of servers
also, internal servers can send/receive from the proxy/MX/DNS servers
only what they actually need to perform their function.  Then you have
the policy from the main LAN to the border server group and I can only
presume whoever set this up fell asleep before they got to this stage
apart from the rules to prevent direct Internet-LAN and LAN-Internet
traffic it's an ACCEPT policy both ways, going to all that effort to
keep it secured from direct contact with the Internet only to present 38
single points of failure for the entire house of cards directly to the
Internet with a vast array of dameons to play with (Apache2 w PHP and
Rails and 6 different web applications publicly visible ones anyway,
Bind, Squid, Socks proxy Dante I think, Postfix, Several stupidly placed
back-end MySQL servers, Courier IMAP providing email access for mobile
employees, Rsyncd, some others but forget which packages an ftpd).

And not sure how I ended up off on that tangent just came to me there
and seemed like a really great example for anyone how *NOT* to setup a
firewall.  Takes far too much work getting them all set up right to
offer every script kiddie who gets bored a menu to pick from to walk
right on around it, once I saw the setup and figured out which server
got compromised only surprise to me was how on earth it hadn't happened
earlier.

Attachment: signature.asc
Description: OpenPGP digital signature

------------------------------------------------------------------------------
Symantec Endpoint Protection 12 positioned as A LEADER in The Forrester  
Wave(TM): Endpoint Security, Q1 2013 and "remains a good choice" in the  
endpoint security space. For insight on selecting the right partner to 
tackle endpoint security challenges, access the full report. 
http://p.sf.net/sfu/symantec-dev2dev
_______________________________________________
Shorewall-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/shorewall-users

Reply via email to