RE: More on VMWare poor guest isolation design
On Mon, 27 Aug 2007, M. Burnett wrote:
I should probably have already ended this discussion, but it reminds me of a
discussion I had on this same list almost ten years ago trying to explain to
Microsoft why a vulnerability that discloses physical paths is a big enough
deal to bother patching. Their argument was that they couldn't see the risk
of disclosing a physical path, and if someone could do something with that
path then they could probably discover the path in the first place. My
argument was that it really doesn't matter what the current risks might be,
that's really not the point, let's just fix it anyway. It turns out later
there were a number of IIS issues where people could execute or access
files, but they needed to know the physical path first.
Dude, you're talking about apples and oranges. Path disclosure in a web app
is bad, period, and should be considered a security risk. But the API
you're complaining about is a *legitimate* feature with legitimate uses.
Yes, it's a feature that can be very badly abused, so enabling it needs some
forethought and intelligence.
I've said this once already, but it bears repeating: your concerns deserve
discussion in context of vmware best practices. But I personally don't
believe it merits discussion as a vulnerability. It's no more a
vulberability than, say, not setting a password on your Windows
administrator account. It's obviously idiotic, but not a flaw in the
software stack.
I think some of you are overanalyzing this issue. I am well aware that there
are other ways to accomplish the same thing in many instances, I am not
saying I have introduced a spectacular new attack vector. I would categorize
this threat standing on its own as medium to low, depending on your
environment. But the fact is that this thing bypasses normal OS security
mechanisms and we simply cannot imagine how that might be used by an
attacker in the future. Some of you keep trying to point out that owning the
host always means owning the guests, but that isn't always the case,
especially if you are not a full administrator on the host machine.
*If* you can use the API to spawn a process in a vm owned and operated by
another user *then*, and only then, do you have a legitimate vulnerability.
But you're basically complaining about being able to shoot yourself in the
foot. It is still incumbent on the host admin to prevent unauthorized
access, and *you* to prevent unauthorized use of your account. If those two
imperatives are competently met, then vmware's functionality is of little
concern.
I know that for a lot of years people have been saying that once someone can
access the physical box, there's nothing more you can do. Well, that's just
not true anymore. You very well can protect a physical machine and you
should be able to protect a virtual guest from its host. There's no way a
non-admin user is going to be able to modify the RAM of a vm. And in Windows
Vista, if not already blocked, even as an administrator I would have to
explicitly allow a worm to access the RAM or disk of a virtual machine. No
worm is going to access a vm's resources without a UAC prompt coming up.
You've got a lot more confidence in Vista then I do. Regardless, here's the
practical reality: you have a unprivileged process which can send commands
to control a vm running with privileged resources, right? As someone else
pointed out: why not just pause the VM (which writes the vm address space
to a *user*-owned file), edit it, and restart it? I'd be very surprised if
there wasn't more that could be done to a live vm as well.
Anyway you cut it, UAC is worthless in this circumstance.
The argument that owning a physical machine automatically means game over
just isn't true. We should be able to say the same thing about a VM.
I'm sorry, but your expectations for the use and value of virtual machines
is very much out of step with reality.
--Arthur Corliss
Live Free or Die