Re: UFS Bug: FreeBSD 6.1/6.2/7.0: MOKB-08-11-2006, CVE-2006-5824, MOKB-03-11-2006, CVE-2006-5679

On Fri, Nov 24, 2006 at 03:15:43PM -0500, Bill Moran wrote:
On Fri, 24 Nov 2006 21:04:30 +0100
Lutz Boehne <lboehne@xxxxxxxxxxx> wrote:

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Out of the box you need to be root to mount things. Once you have
root access to a box you don't need silly things like this to crash
it.

If you've gone out of your way to configure your box in such a way
that a non-root user can mount arbitrary UFS filesystems then they
certainly don't need to waste their time with buffer-overflows and
the like. They can simply mount a filesystem with any number of SUID
root binaries on it and have their way with the box.

Either way, while it's senseless to argue that the buffer overflows
don't exist, anyone in a positiion to actually exploit them doesn't
need them to be malicious.

I do quite not agree with your analysis.

Firstly, if you set the vfs.usermount sysctl to 1, users can mount any
filesystem from a device they have read access to to any directory they
own, _but_ if the user does so, FreeBSD will automatically mount that
filesystem nosuid. So the intent is to give a local user the possibilty
to mount a filesystem without gaining full control over the machine.

Secondly, why would people go out of their way to set that sysctl to 1?
I can see this happen in environments where users are not supposed to
have full control over their desktop machines, but where they need to
transfer data to/from USB flash drives.

Thirdly, while I'm talking about desktop machines, many desktop Linux
distributions are configured such they will _automatically_ mount USB
media once those are plugged in (and pop up an icon on the KDE or GNOME
desktop). It's only a matter of time until such functionality will be
available on FreeBSD (maybe it already is?) and widely used on desktop
machines (e.g. on Laptops, in Internet Cafes), as it seems to be quite
user friendly. On such machines an attacker would not even need a local
user account.

While one might say that these attack scenarios all require physical
access (and we all know that physical access is game over, right;)),
simply plugging in a USB memory device is much more inconspicious than
other "physical" attacks, like rebooting a box into single user mode
(which one could additionally secure with a password prompt).

I don't think anyone is arguing whether or not this is a bug. It is.

I will argue, however, that it does not constitute a security flaw, which
is what the MOKB folks claim. If a user has the ability to graft untrusted
filesystems onto the filesystem tree, that user is in one of a few scenerios:
1) They are root or equivalent.
2) They have physical access to the machine.
3) They are working on a machine that is secured incorrectly.

If #1, then it's a mute point, as root can DOS a machine without any kernel
bugs. If #2, it's a mute point, as physical access bypasses any software
security anyway.

That is not really true.

*Unlimited* physical access can get you around any
software security, but an attacker often does not have unlimited physical
access.

Take for example public computers in a library or an internet café
or similar (or just a computer room in a school.) A user there can probably
try to mount a CD-ROM or a floppy or a USB-stick without anybody noticing or
caring much if they do notice. If that same user were instead to remove the
case to put in his own harddisk to use as a boot device, then it is very
likely that somebody will investigate what said user is doing.

There is also the fact to consider that the more powerful attack vectors
that physical access opens up tend to take a bit of time to use.
If it takes 10 minutes to open a case and modify the innards then it is not
possible to get access undetected while the rightful user goes to another
room to fetch something (for example.) If it takes 10 seconds to mount an
USB stick and get root through some filesystem bug then you can do it while
somebodys back is turned.
The time it takes to mount an attack is very important in many cases to
determine if an attack is actually feasible or not.


And #3 is a mute point, since any system can be configured
to be insecure by a properly skilled idiot, and the kernel hackers can't be
expected to program around idiotic sysadmins.

So, yes, it is a bug that needs to be fixed. But I don't see it as a security
issue.

It is a security issue, but perhaps not one of the most critical ones (in
particular it does not allow remote breakins which are generally the most
worrisome kind.)



--
<Insert your favourite quote here.>
Erik Trulsson
ertr1013@xxxxxxxxxxxxx
_______________________________________________
freebsd-security@xxxxxxxxxxx mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-security
To unsubscribe, send any mail to "freebsd-security-unsubscribe@xxxxxxxxxxx"