On 9/21/07, Scott Wells wrote:
(I'd hoped you would have prefaced that with a statement like "these
are my stock options talking, but...")
This is the kind of bad advice that virtualization companies (and
naive users of those technologies) need to stop spreading. This
security model is flawed, and people should not rely on these virtual
machine environments to provide firewall services.
Here's an entirely realistic scenario at this point:
- Administrator pays loads of money for VMware ESX; for better ROI, he
intends to replace several systems on the network with one big system
running a number of VMs. Maybe there is a full DMZ (say, 10 hosts) on
this box. One virtual machine is configured as a firewall, intended to
provide packet filtering and other network security services for the
other DMZ VMs.
- A vulnerability is discovered that allows an attacker who has
presence in one VM to execute arbitrary code on the host OS, or
transfer files between guest and host. (Both of these have happened
already. In fact, VMware Tools seems to be the perfect bit of flawed
gateway software to make this even easier.) Virtualized segmentation
is compromised at this point.
- Attacker now has presence on host OS and can fully control all 10 of
the VMs running on the host. VM segmentation was supposed to prevent
this, remember? This includes the firewall which he can now play fun
games with such as overwriting the ruleset. He can sniff network
traffic for all the VM hosts since he has direct access to the host
interface. In one short subversion, 10 (11) systems have been
compromised through one flawed security model. A weakness in one VM
becomes the thing that makes compromising all the others dramatically
easier. Why subject your firewall to that?
At least in a traditional non-virtualized firewall model, the attacker
would have to pull out real exploits and attack real (secured)
services to compromise the firewall, and it wouldn't fall at the same
time as the other hosts.
Yes, these kinds of of flaws have (so far) been able to be patched, but
a. They're becoming more frequent as more research goes into breaking out of VMs
b. The impact of these flaws can be so high it doesn't justify risking
the integrity of an entire network of machines at the same time when
you get bit by it.
Feel free to lump all of your IIS webservers onto a VM environment and
let that get owned up and down. At least have the good sense to
physically seperate your firewall (and other network security devices)
out of that.
DS
| Greg KH | [GIT PATCH] driver core patches against 2.6.24 |
| debian developer | Re: Dual-Licensing Linux Kernel with GPL V2 and GPL V3 |
| Vu Pham | Re: [Scst-devel] Integration of SCST in the mainstream Linux kernel |
| Adrian Bunk | Re: Linux 2.6.21 |
git: | |
| Gerrit Renker | [PATCH 15/37] dccp: Set per-connection CCIDs via socket options |
| Jarek Poplawski | [PATCH] pkt_sched: Destroy gen estimators under rtnl_lock(). |
| Radu Rendec | Endianness problem with u32 classifier hash masks |
| Benjamin Herrenschmidt | [PATCH 0/11] ibm_newemac: Candidate patches for 2.6.25 |
