Re: [PATCH 2/3] binfmt_elf: Verify signature of signed elf binary

[Vivek Goyal <vgoyal@redhat.com>]

On Thu, Jan 17, 2013 at 04:58:02PM +0200, Kasatkin, Dmitry wrote:
> On Wed, Jan 16, 2013 at 11:53 PM, Vivek Goyal <vgoyal@redhat.com> wrote:
> > On Wed, Jan 16, 2013 at 02:24:50PM -0500, Mimi Zohar wrote:
> > [..]
> >> > > Sorry, this is out of scope for IMA.  Dmitry has looked into this, but
> >> > > I'm not sure where it stands at the moment.
> >> >
> >> > Ok, so that's one reason that why I wrote these patcehs. IMA currently
> >> > is not doing following things to make sure address space of signed images
> >> > is not modified by others.
> >> >
> >> > - Protecting against modifications to pages on swap.
> >> > - Protecting against modifications by ptrace.
> >> > - Protecting against modifications which bypassed filesystem and directly
> >> >   wrote to the block.
> >> >
> >> > Locking down all the pages of signed binaries in memory hopefully should
> >> > solve above problems.
> >>
> >> Signing and verifying ELF executables goes back a long time ~2003/4,
> >> from a number of esteemed kernel developers, including Greg-KH and Serge
> >> Hallyn.
> >>
> >> IMA-appraisal isn't limited to appraising a single type of file, but is
> >> a generic mechanism for appraising all files.  If there are issues that
> >> aren't being addressed, then by all means, please help by addressing
> >> them.  Duplicating a large portion of the code is not productive.
> >
> > So do you have ideas on how to address above mentioned issues. Do they
> > fit into the realm of IMA/EVM or I just need to write separate code (which
> > I have already done).
> >
> > With above issues, IMA stuff for executable files sounds incomplete.
> >
> 
> swap is a last resort. healthy system uses swap minimally.
> It is very easy to add /etc/crypttab which allows to have encrypted swap
> 
> # <target name> <source device>         <key file>      <options>
> swap   /dev/sda6       /dev/urandom    swap
> 
> And it will eliminate plenty of possible attacks.
> Processes have also RW data, modification of those will create huge
> risk for the system...

Hm..., encrypted swap is interesting. It might have overhead too. It is
one way of doing thing, but I think it would be better if we don't rely
on user has setup encrypted swap and provide a way to lock down memory of
signed executables.

> 
> But certain locking extensions like you implemented can be added to IMA as well.

Cool. That would help. I am very new to IMA. If you have some ideas on
how to go about it, I can start looking into implementing it.

> 
> It was said about ptrace already.

Sorry, did not get what was said about ptrace already? We need to disable
ptrace for signed executables. So can IMA set a process flag upon
signature verification and exec() code can give additional capability. And
as Eric mentioned that should automatically take care of ptrace issue.

> 
> > - Protecting against modifications which bypassed filesystem and directly  wrote to the block.
> 
> Can you please tell a bit more how this patch protect against direct
> writing to the blocks?

If you have loaded all the pages from disk and locked them in memory and
verified the signature, then even if somebody modifies a block on disk
it does not matter. We will not read pages from disk anymore for this
exec(). We verified the signature of executable loaded in memory and
in-memory copy is intact.

So next time somebody tries to execute same binary it will fail (because
of modified block).

Thanks
Vivek