Jail question

[Ivan Voras]

On 10/14/10 18:20, Matthew Law wrote:
> I have a single box on which I would like to run openvpn, smtp (postfix,
> dspam, greylist, clamav), imap (dovecot) apache22 and bind.  This box also
> acts as a network gateway so it would give an attacker carte blanche to
> the internal nets if it was compromised, which makes me nervous.  The plan
> is to run openvpn as the only unjailed service and the rest of the
> services in a single jail or their own jails.
>
> I have never touched jails before and I'm a bit unsure of the best way to
> go.  I realise that I can jail a service or a copy of the whole system
> (service would be preferable for space efficiency) but I am unclear on how
> to deal with IP addresses in jailed environments and if I should create
> individual jails or a single jail for all services.  At the moment I am
> leaning toward a single system jail for everything so I can keep the space
> in which openvpn runs as uncluttered as possible and also have a single
> postgres instance shared by the other services.  Basically, if any of the
> public services in the jail are compromised I would like to make it very
> hard for the attacker to see the internal network.

Since jails can do many things there are many "helper" utilities that 
can do much to simplify the process. If you can hack python, you can, 
for example, modify my script at 
http://ivoras.sharanet.org/stuff/mkjails.py which I've used to create a 
thousand very light-weight jails which are started and managed using 
only standard FreeBSD tools.

In any case, read rc.conf(5) man page for the jail_* settings.

> If I use this scheme must I use separate public IPs for openvpn and the
> services jail or is it possible to use a single IP or some NAT/PAT scheme?
> -this box currently has 4 x NICs split into 2x lagg interfaces in failover
> mode (one public, one private), if that makes any difference....

This is the more complex question; I think that everything which needs 
direct access to the NIC (i.e. BPF, DHCP, IPFW, etc.) will need to be 
run on the host system. TCP services will work inside jails without 
problems, but with jails it's almost the same as if they were on another 
system. If you do use NAT you will have to configure it on the host. 
Instead, you can also use TCP proxies (like bsdproxy). It's up to you 
how much complexity do you want in your system, but for simplicity I 
would set up a single outward-facing IP address and then proxy TCP 
services where I need them.