Message-ID: <20121231151356.GA11118@moon>
Date:	Mon, 31 Dec 2012 19:13:57 +0400
From:	Cyrill Gorcunov <gorcunov@...nvz.org>
To:	Eric Paris <eparis@...hat.com>
Cc:	Andrew Vagin <avagin@...allels.com>,
	Andrey Vagin <avagin@...nvz.org>, linux-kernel@...r.kernel.org,
	Andrew Morton <akpm@...ux-foundation.org>,
	Kees Cook <keescook@...omium.org>,
	Serge Hallyn <serge.hallyn@...onical.com>,
	"Eric W. Biederman" <ebiederm@...ssion.com>,
	James Morris <jmorris@...ei.org>
Subject: Re: [PATCH] prctl: fix validation of an address

On Mon, Dec 31, 2012 at 09:27:14AM -0500, Eric Paris wrote:
> On Mon, 2012-12-31 at 14:14 +0400, Andrew Vagin wrote:
> > On Sun, Dec 30, 2012 at 05:03:07PM -0500, Eric Paris wrote:
> > > On Sat, 2012-12-29 at 15:00 +0400, Andrey Vagin wrote:
> > > > The address should be bigger than dac_mmap_min_addr, because
> > > > a process with CAP_RAWIO can map a vma bellow mmap_min_addr.
> > > 
> > > NAK
> > 
> > Currently prctl(PR_SET_MM_*, addr, ) returns EINVAL for valid addresses.
> > I think it's a bug. Are you agree?
> 
> Can you help me understand how prctl(PR_SET_MM_*, relates to
> checkpoint/restore?  My worry here is that somehow this interface could

Here how we use it (from userspace code)

	ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_START_CODE, (long)args->mm.mm_start_code, 0);
	ret |= sys_prctl_safe(PR_SET_MM, PR_SET_MM_END_CODE, (long)args->mm.mm_end_code, 0);
	...

the values of mm.mm_start_code and such are saved in image file and obtained
during checkpoint stage. Note the prctl_set_mm requires the caller to have
CAP_SYS_RESOURCE privilege granted.

> be used to bypass the security properties of mmap_min_addr.    I have no
> idea how the interface is used, so I don't know if my fears are founded.
> When I hear 'restore' I think of a privileged application setting up
> some unprivileged application based on untrusted data.  My fear is that
> some unpriv application, that doesn't have permission to map below
> mmap_min_addr, may be able to trick the privileged application, which
> would have this permission, into doing it on its behalf.  Does that make
> sense?  Is that a realistic scenario with how this interface is used?

> 
> > CONFIG_LSM_MMAP_MIN_ADDR could not be got from user space.
> > 
> > This application can use a real value of mmap_min_addr, but it is not
> > provided into userspace.
> 
> Unrelated to this patch issue, but I guess either could be exposed if
> there is a need.
> 
> > Currently a task can have user memory area bellow dac_mmap_min_addr,
> > but prctl returns -EINVAL for such addresses.
> > How can I understand the reason, if I know that the address is valid?
> 
> Talking about dac_mmap_min_addr is wrong.  The capabilities security
> module uses dac_mmap_min_addr but other LSMs can (and obviously do) use
> other things.  mmap_min_addr is just the shorthand to make sure you
> clear all hurdles.  Breaking those hurdles up outside of the security
> subsystem is wrong.
> 
> The kernel makes the decision on what is valid via security_mmap_addr().
> Assuming there are no security fears of an untrusted application
> tricking some priviledged application to set up these maps the answer is
> just calling security_mmap_addr() instead of doing if(addr <
> mmap_min_addr) return -EINVAL;

If only I've not missed something obvious, the check for security_mmap_addr() here
instead of poking the mmap_min_addr looks more correct for me. Andrew?

> I don't know if it is a good idea to allow this interface to ever go
> below mmap_min_addr, but I do know that using (or even thinking about)
> dac_mmap_min_addr is wrong and you should be looking at
> security_mmap_addr() if you look at anything...

	Cyrill
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists