Re: taskstats root only breaking iotop

[Vasiliy Kulikov <segoon@openwall.com>]

On Mon, Oct 03, 2011 at 15:31 +0300, Adrian Bunk wrote:
> On Sun, Oct 02, 2011 at 02:54:57PM +0400, Vasiliy Kulikov wrote:
> > (cc'ed kernel-hardening)
> > 
> > On Sun, Oct 02, 2011 at 12:22 +0200, Guillaume Chazarain wrote:
> > > On Sun, Oct 2, 2011 at 2:20 AM, Linus Torvalds
> > > <torvalds@linux-foundation.org> wrote:
> > > > So I don't see why you ask for it. What could possibly be a valid use-case?
> > > 
> > > Right, kbyte granularity is enough.
> > 
> > It is not enough.  In some border cases an attacker may still learn
> > private information given the counters with _arbitrary_ granularity:
> > 
> > http://www.openwall.com/lists/oss-security/2011/06/29/9
> 
> If you request a CVE for that, shouldn't there also be a CVE for
> /proc/<pid>/cmdline being readable by all users?
> 
> I'd expect "ps -ef" to be more likely to give private information to an 
> attacker than counters with kbyte granularity, or am I wrong on that?

I agree that world-readable cmdline can be a privacy issue in some
cases.  I tried to push a patch introducing procfs mount option to
restrict /proc/PID/ to PID owner to address the issue (as world-readable
cmdline and other files are already used by plenty of programs and
unconditionally breaking backward compatibility is not good, a
configuration mechanism is needed), but it didn't receive positive
feedback.  A more detailed explanation from Solar Designer:

http://www.openwall.com/lists/kernel-hardening/2011/06/20/5

Andrew Morton complained that it is too specific to our needs and one
might want to define more fine granted procfs security model:

http://www.openwall.com/lists/kernel-hardening/2011/06/21/3

I've tried to address it and defined per-file policy:

http://www.openwall.com/lists/kernel-hardening/2011/08/10/12

No comments so far :(

-- 
Vasiliy Kulikov
http://www.openwall.com - bringing security into open computing environments