Re: [PATCH v5 next 1/5] modules:capabilities: add request_module_cap()

[Luis R. Rodriguez]

On Tue, Nov 28, 2017 at 02:18:18PM -0800, Kees Cook wrote:
> On Tue, Nov 28, 2017 at 2:12 PM, Luis R. Rodriguez <mcgrof@xxxxxxxxxx> wrote:
> > On Tue, Nov 28, 2017 at 01:39:58PM -0800, Kees Cook wrote:
> >> On Tue, Nov 28, 2017 at 1:16 PM, Luis R. Rodriguez <mcgrof@xxxxxxxxxx> wrote:
> >> > And *all* auto-loading uses aliases? What's the difference between auto-loading
> >> > and direct-loading?
> >>
> >> The difference is the process privileges. Unprivilged autoloading
> >> (e.g. int n_hdlc = N_HDLC; ioctl(fd,
> >> TIOCSETD, &n_hdlc)), triggers a privileged call to finit_module()
> >> under CAP_SYS_MODULE.
> >
> > Ah, so system call implicated request_module() calls.
> 
> Yup. Unprivileged user does something that ultimately hits a
> request_module() in the kernel. Then the kernel calls out with the
> usermode helper (which has CAP_SYS_MODULE) and calls finit_module().

Thanks, using this terminology is much better to understand than auto-loading,
given it does make it clear an unprivileged call was one that initiated the
request_module() call, there are many uses of request_module() which *are*
privileged.

> > OK and since CAP_SYS_MODULE is much more restrictive one could argue, what's the
> > point here?
> 
> The goal is to block an unprivileged user from being able to trigger a
> module load without blocking root from loading modules directly.

I see now. Do we have an audit of all system calls which implicate a
request_module() call? Networking is a good example for sure to start
off with but I was curious if we have a grasp of how wide spread this
could be.

I'll go review the patches again now with all this in mind.

  Luis