Aaron's Worthless Words

A few years ago, I developed a Perl-based application that take a template file and pukes out standardized access rules for new hosts as they’re added to the network.  This works great for making sure that each host is able to be managed properly.  This solution, however, is not very flexible.  If I need to remove a host’s access, I may have to take out 20 rules individually.  That’s not really cool, so, at the suggestion of a coworker, I’m working on a solution that uses objects, object-groups, and nested object-groups.  This should minimize the configured rules and allow new host rules to be added and removed by simply adding hosts to object-groups.

Example time.  Let’s say you have a bunch of RFC1918 addresses behind your firewall that all need HTTP access to one network on the InterTubes.  First thing to do is to create the objects that will be involved; in this case, that’s all the networks and/or ranges.  To be more specific, 192.0.2.0/24 is the public IP to which the hosts need access.   The internal hosts are 192.168.0.0/24 and the IP range 10.0.0.1-25.  Yes, I know the names are terrible.

object network NET1
 subnet 192.0.2.0 255.255.255.0
object network NET2
 subnet 192.168.0.0 255.255.255.0
object network NET3
 range 10.0.0.1 10.0.0.25

Now, we can use some Snort-like configuration to create object-groups that include the objects we just created.  In this case, we’re creating an InterWebs-based object-group and another for local addresses.

object-group network REMOTE-NETS
 network-object object NET1
object-group network LOCAL-NETS
 network-object object NET2
 network-object object NET3

Now we can use these object-groups to create ACLs.   You’ve done this before, right?

access-list TEST-ACL extended permit tcp object-
   group LOCAL-NETS object-group REMOTE-NETS eq www

To be sure it worked as expected, let’s take a look at the ACLs.  The format sucks because the lines are so long; sorry about that.

firewall# show access-list TEST-ACL
access-list TEST-ACL; 7 elements; name hash: 0x5329ed72
access-list TEST-ACL line 1 extended permit tcp object-group LOCAL-NETS object-group REMOTE-NETS eq www 0x1abfa4a0
  access-list TEST-ACL line 1 extended permit tcp 192.168.0.0 255.255.255.0 192.0.2.0 255.255.255.0 eq www (hitcnt=0) 0x50797e0c
  access-list TEST-ACL line 1 extended permit tcp host 10.0.0.1 192.0.2.0 255.255.255.0 eq www (hitcnt=0) 0xa2159c9d
  access-list TEST-ACL line 1 extended permit tcp 10.0.0.2 255.255.255.254 192.0.2.0 255.255.255.0 eq www (hitcnt=0) 0x93f1c362
  access-list TEST-ACL line 1 extended permit tcp 10.0.0.4 255.255.255.252 192.0.2.0 255.255.255.0 eq www (hitcnt=0) 0x512fc827
  access-list TEST-ACL line 1 extended permit tcp 10.0.0.8 255.255.255.248 192.0.2.0 255.255.255.0 eq www (hitcnt=0) 0x7b11e96f
  access-list TEST-ACL line 1 extended permit tcp 10.0.0.16 255.255.255.248 192.0.2.0 255.255.255.0 eq www (hitcnt=0) 0xc302aa0e
  access-list TEST-ACL line 1 extended permit tcp 10.0.0.24 255.255.255.254 192.0.2.0 255.255.255.0 eq www (hitcnt=0) 0x2ea75962

Cool.  Everything looks great, and everyone should have the access they need.  If a new host with the IP of 172.16.0.28 comes online inside the network, you add a new nested object-group that includes that host.  Access is automagically updated, so there’s no need for more ACL lines.  Another method is to add the new host directly to the LOCAL-NETS object-group, but that’s going to limit the ways to address that box and related hosts in an ACL.  I suggest you just add the new object to the object-group.

As a bonus, you can also nest object-groups into each other.  For example, we can create an object-group that includes our the LOCAL-NETS and REMOTE-NETS object-groups.

object-group network ALL-NETS
 group-object LOCAL-NETS
 group-object REMOTE-NETS

I don’t know where you’d ever use that specific object-group, but you could use this technique in other ways.  I’m looking to create object-groups for each interface of the firewall and creating a super-object (my term) to allow the standard access stuff.  You could do the same for office networks; each office has it’s own object-group for access that is also nested in an object that provides basic access to the TubeWebs or something.  Use your imagination.  :)

Send any questions my way.

This article is based on an ASA 5505 running 8.3.1.  Most of the config above should be portable to any 8.x except for declaring the objects. In other versions of 8.x, you may have to add host directly to the object-group.  Running on 7.x and below may be a different story.

Other reading:  http://aconaway.com/2009/10/01/object-groups-in-the-asafwsmpix/

Director’s Commentary:

Audio clip: Adobe Flash Player (version 9 or above) is required to play this audio clip. Download the latest version here. You also need to have JavaScript enabled in your browser.

I’ll start off with a warning.  I’ve been running 8.3.1 on my home 5505 for a few hours now.  Not only is this not really enough time for a thorough review, it’s also not the environment to test enterprise-level configurations.  There are also a lot of details missing that I just don’t know about yet, so please do some research on your own to figure out what’s going to break if you upgrade your ASA.

If you haven’t heard, Cisco has released version 8.3.1 of their ASA operating system.  I’m excited about this for only one reason – Smart Tunnels with tunnel policies.

If you’ve never heard of Smart Tunnels, you’re probably not alone.  I don’t know why they’re not more popular than they are, but I dig them.  A user connects to a URL, logs in, and a little applet loads on the machine that is used to proxy traffic through the ASA.  It doesn’t proxy all your traffic, though; only traffic from applications that you define are sent through the tunnel.  There is a huge problem that I can’t stand, though.  What if you need to SSH through the firewall and to your local LAN at the same time?  The smart tunnel applet doesn’t care or even know what you want to do; it tunnels all the traffic from the application.  Not good, eh?

The big change to this in 8.3.1 is the addition of tunnel policies to the smart tunnels.  According to the release notes, you can now dictate which connections do and don’t go through the smart tunnel.  Now, I can configure the tunnel so that some traffic goes through the ASA to get to the production gear, but other traffic pukes out the NIC normally.  I know a lot of users who are going to like not having to log in and out all day.

Note: I may do an article on smart tunnels once everything slows down a bit.  It’s a solid way to implement a clientless VPN that doesn’t require administrative access on the machine to run.

The big feature that everyone is talking about, though, is the change to the way NAT is done.  Back in the day (that means earlier this morning), if I wanted to configure a static NAT, I’d do something like this to create a static and a service NAT to two different boxes.

firewall(config)#static (inside,outside) 192.0.2.1 192.168.1.100
firewall(config)#static (inside,outside) tcp interface ssh 192.168.1.101 ssh

Now, you create an object and give that object all the attributes.  I think Cisco calls this auto-NAT.  I have no idea what the auto part means.  In our example, we would do something like this.

firewall(config)#object network TESTHOST1
firewall(config-network-object)#host 192.168.1.100
firewall(config-network-object)#nat (inside,outside) static 192.0.2.1
firewall(config)#object network TEST2
firewall(config-network-object)#host 192.168.1.101
firewall(config-network-object)#nat (inside,outside) static interface service tcp ssh ssh

I would say that the configuration is easier to parse with your eyes if the ASA didn’t break up the configuration into two parts.  If you were to do a show run and look for our configuration, you would have to look in two places.  The first part declares the object name and the host/subnet/IP range for which it’s associated.  The next part, which comes after the ACLs, declares the NAT stuff.

object network TESTHOST1
 host 192.168.1.100
object network TESTHOST2
 host 192.168.1.101

[SNIP a billion lines of ACL]

pager lines 24
logging enable
logging timestamp
logging buffer-size 8192
logging buffered informational
logging asdm informational
logging host inside x.x.x.x
flow-export destination inside x.x.x.x 12345
mtu outside 1500
mtu guests 1500
mtu inside 1500
icmp unreachable rate-limit 1 burst-size 1
icmp permit x.x.x.0 255.255.255.0 inside
asdm image disk0:/asdm-631.bin
asdm history enable
arp timeout 14400

object network TESTHOST1
 nat (inside,outside) static 192.0.2.1

object network TESTHOST2
 nat (inside,outside) static interface service tcp ssh ssh

It may be simpler to configure, but it’s not simpler to figure out later.  I’d rather have single lines of static statements; at least I can use regex on those efficiently.

There is a bright side to the new NAT thing, though.  Because the NAT statements are configured in the object, you can now reference the real IP of the host in ACLs instead of the NATted IP.  This will help those of us who use firewalls with 488249284 interfaces and that many NATs for each host.  If we wanted to allow access to the SSH host in the example, we would write an ACL that allows access to 192.168.1.101 instead of finding the NATted address on that interface and building the rules to that address.

Speaking of ACLs, you can actually create a global access-group.  Instead of creating an ACL with rules and an access-group to bind to an interface, you can build one single ACL and configure an access-group with the global directive to basically apply that ACL to all interfaces.  A few quick tests show that you can have both interface and global access-group configured simultaneously and that interface ACLs will be executed first.  I need to do some more testing to figure out exactly how these work together.

Everyone should upgrade, right?  Nope.  I don’t ever upgrade to something cool just because it’s cool.  I also don’t like to have to buy more hardware to go up a minor revision.  Take a look at the the memory requirements for 8.3.1; every model up to the 5510 requires more than the base amount to upgrade.  I got lucky since my 5505 has 512MB in it already, but I would hate to have to justify quadrupling (!) the RAM in a 5540 just for some cool features.

Send any rotten tomatoes questions to me.

Sources:

• 8.3.1 Release Notes
• Smart Tunnels on 8.2
• NAT on 8.2