Re: ANN: libseccomp

[Paul Moore <paul@paul-moore.com>]

On Monday, April 09, 2012 06:46:02 PM Paul Moore wrote:
> On Monday, April 09, 2012 04:51:30 PM Will Drewry wrote:
> > On Mon, Apr 9, 2012 at 4:32 PM, Paul Moore <paul@paul-moore.com> wrote:
> > > On Monday, April 09, 2012 12:16:30 PM Kees Cook wrote:
> > >> On Mon, Apr 9, 2012 at 11:58 AM, Paul Moore wrote:
> > >> I see that the arch check happens during _gen_bpf_build_bpf(), which
> > >> is excellent. Do you have any thoughts about including a call to
> > >> prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) by default as well?
> > > 
> > > That is a good question, and I guess it comes down to another question
> > > of if anyone would want to use seccomp without NO_NEW_PRIVS.  If the
> > > answer is no then I'm comfortable adding it into the seccomp_load()
> > > function; however, if the answer is yes we might want to do something
> > > different.
> > > 
> > > I haven't given much thought to this yet, so if you or anyone else feels
> > > strongly about the issue - either pro or con - I'd appreciate hearing
> > > the argument.
> > 
> > I guess the question is if there is an expectation that this library
> > be used with something like lxc, where a whole, functional system with
> > suid/fcaps binaries is contained.  In that world, it may not be
> > desirable to set the nnp bit.   The same is true if, for some reason,
> > the init task was to set a system-wide filter.
> > 
> > Most likely, default use of nnp is probably "the right thing", but
> > it'd be nice to be able to annotate when you really want to allow
> > privileged contexts to set filters without nnp.
> 
> Okay, that seems reasonable: default to NO_NEW_PRIVS, but provide an
> override mechanism.
> 
> I've been wanting a mechanism/API for tweaking some of the default library
> parameters for the past few weeks, this is likely the last bit of motivation
> I need to start working on this.  I'll look into it once the license issue
> is sorted.

A quick update - I just added support for setting NO_NEW_PRIVS when loading 
the seccomp filter into the kernel.  By default libseccomp attempts to set 
NO_NEW_PRIVS but does not fail if prctl(NO_NEW_PRIVS) returns with an error; 
however, both the attempt to set NO_NEW_PRIVS and the fact that libseccomp 
does not fail on error are configurable via the application if you don't like 
the defaults for your particular use case.

-- 
paul moore
www.paul-moore.com