Re: Physical access vulnerabilities and auto-mounting

[Nelson Elhage <nelhage () ksplice com>]

I don't have any definite opinions here about where to draw which
lines, but I want to point out that in addition to physical attack
vectors, virtualization tools are also potentially affected by these
kinds of bugs. If you try to mount an untrusted VM's virtual disk
image from somewhere, you're also vulnerable to that VM triggering
bugs in the filesystem or other layers.

- Nelson

On Tue, Feb 22, 2011 at 11:17 PM, Dan Rosenberg
<dan.j.rosenberg () gmail com> wrote:
> I originally started writing this as a response to the recent CVE
> requests for issues in partition handling, but thought it might be a
> useful discussion on its own.  I was wondering if there are any
> clear-cut policies on issues involving physical access, since these
> can be very difficult in terms of assigning blame.
>
> For example, many Linux distributions will auto-mount filesystems on
> removable storage, often going so far as to load corresponding kernel
> modules for filesystems that aren't compiled in or don't already have
> an LKM loaded.  Sometimes, this will happen even if the screen is
> locked.
>
> Incidentally, many Linux filesystem implementations don't have
> especially robust error handling for failures during attempts to mount
> corrupt filesystems.  As an example, I have a deliberately corrupted
> btrfs filesystem that triggers a BUG() if you attempt to mount it.  I
> formatted a USB stick with this filesystem, so now I have a USB stick
> that will panic the kernels of distributions that support
> auto-mounting, in some cases even when the screen is locked.
>
> Should this be considered a vulnerability?  Probably.  But what should
> be fixed?  Should auto-mounting be disabled entirely?  Is it no longer
> a vulnerability if auto-mounting is disabled only when the screen is
> locked?  Should all filesystems have graceful error handling for every
> possible edge case that can occur when dealing with corruption?
>
> I'd be interested to hear opinions on this.  And depending on how the
> discussion goes, I'd be happy to provide more details on specific
> cases, such as the btrfs example.
>
> -Dan