Re: [PATCH 08/28] SECURITY: Allow kernel services to override LSM settings for task actions [try #2]

[Stephen Smalley <sds@tycho.nsa.gov>]

On Tue, 2007-12-11 at 15:04 -0800, Casey Schaufler wrote:
> --- David Howells <dhowells@redhat.com> wrote:
> 
> > Stephen Smalley <sds@tycho.nsa.gov> wrote:
> > 
> > > All your code has to do is invoke a function provided by libselinux.
> > 
> > Calling libselinux means it's a special case for a specific LSM.
> > 
> > I think the best way to do this, then, has to be to dlopen the appropriate
> > LSM
> > library.  That way I don't need to do any conditional compilation or linking,
> > but can build all the bits in to cachefilesd and have the appropriate one
> > selected by the /etc/cachefilesd.conf.
> > 
> > So, what do I invoke in libselinux, how do I configure it, and how do I
> > integrate the config into my RPM and install it?
> > 
> > And then what does it give me that I can hand to the kernel (a context string
> > for SELinux, I presume), how do I get the kernel to make a check on it, how
> > do
> > I configure the check and how do I install that config from my RPM (I presume
> > I just need to modify the .fc, .if and .te files that I have already)?
> 
> That seems like an awful lot of work. I suggest that what you
> put in /etc/cachefilesd.conf is a line like:
> 
>    security_context:"<whatever>"
> 
> and have your daemon pass "<whatever>" into the kernel using
> a cachefile mechanism. The kernel code can call
> security_secctx_to_secid("<whatever>") to determine if it's valid.
> No need to invoke LSM specific code in your daemon. You may need
> to have an application, say cachefileselinuxcontext, that will
> read the current policy and spit out an appropriate value of
> "<whatever>", but that can be separate and LSM specific without
> mucking up your basic infrastructure applications. 

That sounds workable, although I think he will want a more specific hook
than security_secctx_to_secid(), or possibly a second hook call, that
would not only validate the context but authorize the use of it by the
cachefilesd process.  And then the security_task_kernel_act_as() hook
just takes the secid as input rather than the task struct of the daemon,
and applies it.  At that point, nfsd can use the same mechanism for
setting the acting SID based on the client process after doing its own
authorization.

> > > That mostly works, but it means that an update to policy may require an
> > > update to /etc/cachefilesd.conf, or that switching from one policy to
> > > another might likewise require changing that file.  Versus using a
> > > separate policy-provided config file for the label.
> > 
> > Whilst that's a fair point, if it's in a config file somewhere, then someone
> > may want to change it or someone may want to provide a second file for a
> > second cache with a different security label.
> >
> > > BTW, as should be obvious, some LSMs aren't label-based at all, so it
> > > would need to be optional.
> > 
> > Aargh.  In which case it might not be possible to make the SELinux context
> > passing from userspace -> kernel generic for all LSMs:-(
> 
> For LSM's that don't use labels what you will have to pass in
> won't be a label, it will be something else. But since any LSM
> that wants to do networking or audit will have to deal with
> secid's and secctx's the method outlined above ought to fit the
> bill.

-- 
Stephen Smalley
National Security Agency

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