Re: [PATCH] seccomp.2: Explain arch checking, value (non-)truncation, expand example

[Jann Horn]

On Mon, Mar 16, 2015 at 03:25:56PM -0700, Kees Cook wrote:
> On Mon, Mar 16, 2015 at 11:01 AM, Jann Horn <jann@xxxxxxxxx> wrote:
> > Document some more-or-less surprising things about seccomp.
> > I'm not sure whether changing the example code like that is a
> > good idea - maybe that part of the patch should be left out?
> >
> > Demo code for the X32 issue:
> > https://gist.github.com/thejh/c5b670a816bbb9791a6d
> >
> > Demo code for full 64bit registers being visible in seccomp
> > if the i386 ABI is used on a 64bit system:
> > https://gist.github.com/thejh/c37b27aefc44ab775db5
> 
> So, it is probably worth noting the x32 ABI somewhere, and seccomp.2
> is probably reasonable, though maybe it should be explicitly detailed
> in syscall.2?

I guess that would be sensible. However, I still think that the seccomp
manpage should mention it, too, or advise the reader to also carefully
read the syscall.2 manpage.


> In the seccomp.2 manpage, though, I think we should discourage
> blacklisting, since whitelisting is a much more effective way to do
> attack surface reduction on syscalls. (And, as such, x32 would be
> already eliminated from x86-64 filters.)

I agree, whitelisting should be encouraged. However, as far as I can
tell, people use seccomp not only for proper, strict sandboxing, but
also to e.g. fix small security problems in containers or to provide
additional precautions for them. In that case, I think that the use
of a blacklist is more understandable, and various project use
seccomp that way or at least support the use of blacklists: The
default policy of LXC is a blacklist, sandstorm.io uses a seccomp
blacklist and blacklists specific ptrace calls, systemd-nspawn uses a
blacklist (although the manpage says that that's meant as a precaution
against accidental damage, not as a security measure), systemd
supports both whitelists and blacklists in the SystemCallFilter
directive.


> It is, however, reasonable to update the example just so it can be
> super-explicit, though I'd change the comments to say something more
> direct about the whitelisting vs blacklisting, like "While this
> example uses whitelisting,

You mean you would want to change the example to use whitelisting?
That sounds like a good idea.


> this is how an overlapping syscall ABI
> could be tested." or something. Additionally, I think it would be
> better to test for >= instead of & to avoid having to reload the
> syscall nr.

Yes, sounds good.
--
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/