Re: [PATCH 3/5] v2 seccomp_filters: Enable ftrace-based system callfiltering

[david]

On Thu, 26 May 2011, Ingo Molnar wrote:

> * david@xxxxxxx <david@xxxxxxx> wrote:
>
> > On Thu, 26 May 2011, Ingo Molnar wrote:
> >
> > > * Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > > It also gets rid of all configuration - one of the things that
> > > > makes most security frameworks (look at selinux, but also just
> > > > ACL's etc) such a crazy rats nest is the whole "set up for other
> > > > processes". If it's designed very much to be about just the "self"
> > > > process (after initialization etc), then I think that avoids pretty
> > > > much all the serious issues.
> > >
> > > That's how the event filters work currently: even when inherited they
> > > get removed when exec-ing a setuid task, so they cannot leak into
> > > privileged context and cannot modify execution there.
> > >
> > > Inheritance works when requested, covering only same-credential child
> > > tasks, not privileged successors.
> >
> > this is a very reasonable default, but there should be some way of
> > saying that you want the restrictions to carry over to the suid
> > task (I really know what I'm doing switch)
>
> Unless you mean that root should be able to do it it's a security
> hole both for events and for filters:
>
> - for example we dont want really finegrained events to be used to
>   BTS hw-trace sshd and thus enable it to discover cryptographic
>   properties of the private key sshd is using.
>
> - we do not want to *modify* the execution flow of a setuid program,
>   that can lead to exploits: by pushing the privileged codepath into
>   a condition that can never occur on a normal system - and thus can
>   push it into doing something it was not intended to do.
>
>   data damage could be done as well: for example if the privileged
>   code is logging into a system file then modifying execution can
>   damage the log file.
>
> So it's not a good idea in general to allow unprivileged code to
> modify the execution of privileged code. In fact it's not a good idea
> to allow it to simply *observe* privileged code. It must remain a
> black box with very few information leaking outwards.

I was thinking of the use case of the real sysadmin (i.e. root) wanting to 
be able to constrain things. I can see why you would not want to allow 
normal users to do this.

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