Re: [PATCH] System Wide Capability Bounding Set

From: Serge E. Hallyn
Date: Fri Jan 28 2011 - 13:52:14 EST

Quoting Steve Grubb (sgrubb@xxxxxxxxxx):
> Hi Serge,
> On Thursday, January 27, 2011 09:02:55 am Serge E. Hallyn wrote:
> > What is the attack vector you're actually envisioning? Does some
> > trojan come in and overwrite a program which which it hopes the
> > kernel will execute? Or is there just an existing vuln in such
> > a program? Are there other ways we can address these? Can we find
> > a way to classify the kernel-spawned userspace programs? Perhaps
> > based on the selinux context assigned to the program, we can assign
> > some level of trust that noone could have modified the source?
> I think that what is causing the confusion is that we are considering a different
> threat model than the normal, historic view. The way its normally viewed, if you have
> root, you can do anything you want to a machine. The threat model revolves around
> becoming root on a machine and defense rests on splitting root so a complete system
> compromise might not occur.

I wasn't looking at it as root being all-powerful, but rather as kernel
being all-powerful.

> Today, people want to have multi-tenant hosting using virtual machines whereby they
> give away root control of the guest VM. If you were renting system space, you would
> expect root access. That would make a nice juicy hacking target because you don't know
> who else is sharing the physical machine with you and they might have something in
> their VM worth stealing.
> So, the threat model becomes how do we prevent one guest from attacking another? We
> have sVirt which prevents resource based attacks from occurring. Its pretty effective
> for that. However, what if the bad guy wants to start attacking the hypervisor
> directly in effort to start attacking the host OS?

I always assume they are :)

> They need to be able to run arbitrary code in ring 0 of the VM. That means the hosting
> provider might want to eliminate some capabilities from the whole kernel so that they
> have some assurance that a root user cannot get arbitrary code running in ring 0
> without knowing a kernel level exploit. Also assume that the root user has no control
> over the kernel or modules or initrd which are kept on a read only partition enforced
> by the hypervisor. And the hosting provider will make kernel updates as kernel
> security releases are made.
> This kind of turns around some of the threat modeling that people have always made.
> There are not a whole lot of changes that need to be made. I think there was one other
> patch that we needed to prevent arbitrary code injection. Eric's initial patch was
> overly generous in my opinion. It allowed further modification of the global bounding
> set after boot had finished and could probably be used for mischief as pointed out.
> Perhaps the setting should be immutable after any change to it - which is really how
> its intended to be used. Or maybe even only a subset of the bounding set is modifiable.
> Using a wrapper program is a NOGO because the admin renting the machine would be able
> to overwrite the wrapper and then they have arbitrary code running with full privs and

Not sure I've got this. Wrapper program in the VM he can over-write,
but then he can overwrite the kernel too. But what we are worried about
is the host, so you must mean that. But if the wrapper program is of
type noone_may_write_this_t, then wouldn't finding a way to replace that
be as hard as overwriting the host kernel? Which, of course, still
remains as a viable attack vector for the guest admin, whether you have
this bounding set or not.

In other words, we have to accept that the TCB is always not just the
kernel, but some user-space too. And yes, the wrapper program here
would be part of the TCB.

Again, what I like about the wrapper idea is that it plugs in at one
place in the kernel tree (kernel/ktrhead.c), and can do other setup
if the host admin likes.

> we trust it will do the right thing. We need all modification to the running kernel out
> of reach from root in that VM.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at