Re: [PATCH bpf-next v5 4/7] bpf: lsm: Implement attach, detach and execution

[KP Singh <kpsingh@chromium.org>]

On 24-Mär 14:21, Stephen Smalley wrote:
> On Tue, Mar 24, 2020 at 2:06 PM KP Singh <kpsingh@chromium.org> wrote:
> >
> > On 24-Mär 11:01, Kees Cook wrote:
> > > On Tue, Mar 24, 2020 at 01:49:34PM -0400, Stephen Smalley wrote:
> > > > On Tue, Mar 24, 2020 at 12:25 PM Casey Schaufler <casey@schaufler-ca.com> wrote:
> > > > >
> > > > > On 3/24/2020 7:58 AM, Stephen Smalley wrote:
> > > > > > On Tue, Mar 24, 2020 at 10:50 AM KP Singh <kpsingh@chromium.org> wrote:
> > > > > >> On 24-Mär 10:35, Stephen Smalley wrote:
> > > > > >>> On Mon, Mar 23, 2020 at 12:46 PM KP Singh <kpsingh@chromium.org> wrote:
> > > > > >>>> From: KP Singh <kpsingh@google.com>
> > > > > >>>> diff --git a/kernel/bpf/bpf_lsm.c b/kernel/bpf/bpf_lsm.c
> > > > > >>>> index 530d137f7a84..2a8131b640b8 100644
> > > > > >>>> --- a/kernel/bpf/bpf_lsm.c
> > > > > >>>> +++ b/kernel/bpf/bpf_lsm.c
> > > > > >>>> @@ -9,6 +9,9 @@
> > > > > >>>>  #include <linux/btf.h>
> > > > > >>>>  #include <linux/lsm_hooks.h>
> > > > > >>>>  #include <linux/bpf_lsm.h>
> > > > > >>>> +#include <linux/jump_label.h>
> > > > > >>>> +#include <linux/kallsyms.h>
> > > > > >>>> +#include <linux/bpf_verifier.h>
> > > > > >>>>
> > > > > >>>>  /* For every LSM hook  that allows attachment of BPF programs, declare a NOP
> > > > > >>>>   * function where a BPF program can be attached as an fexit trampoline.
> > > > > >>>> @@ -27,6 +30,32 @@ noinline __weak void bpf_lsm_##NAME(__VA_ARGS__) {}
> > > > > >>>>  #include <linux/lsm_hook_names.h>
> > > > > >>>>  #undef LSM_HOOK
> > > > > >>>>
> > > > > >>>> +#define BPF_LSM_SYM_PREFX  "bpf_lsm_"
> > > > > >>>> +
> > > > > >>>> +int bpf_lsm_verify_prog(struct bpf_verifier_log *vlog,
> > > > > >>>> +                       const struct bpf_prog *prog)
> > > > > >>>> +{
> > > > > >>>> +       /* Only CAP_MAC_ADMIN users are allowed to make changes to LSM hooks
> > > > > >>>> +        */
> > > > > >>>> +       if (!capable(CAP_MAC_ADMIN))
> > > > > >>>> +               return -EPERM;
> > > > > >>> I had asked before, and will ask again: please provide an explicit LSM
> > > > > >>> hook for mediating whether one can make changes to the LSM hooks.
> > > > > >>> Neither CAP_MAC_ADMIN nor CAP_SYS_ADMIN suffices to check this for SELinux.
> > > > > >> What do you think about:
> > > > > >>
> > > > > >>   int security_check_mutable_hooks(void)
> > > > > >>
> > > > > >> Do you have any suggestions on the signature of this hook? Does this
> > > > > >> hook need to be BPF specific?
> > > > > > I'd do something like int security_bpf_prog_attach_security(const
> > > > > > struct bpf_prog *prog) or similar.
> > > > > > Then the security module can do a check based on the current task
> > > > > > and/or the prog.  We already have some bpf-specific hooks.
> > > > >
> > > > > I *strongly* disagree with Stephen on this. KRSI and SELinux are peers.
> > > > > Just as Yama policy is independent of SELinux policy so KRSI policy should
> > > > > be independent of SELinux policy. I understand the argument that BDF programs
> > > > > ought to be constrained by SELinux, but I don't think it's right. Further,
> > > > > we've got unholy layering when security modules call security_ functions.
> > > > > I'm not saying there is no case where it would be appropriate, but this is not
> > > > > one of them.
> > > >
> > > > I explained this previously.  The difference is that the BPF programs
> > > > are loaded from a userspace
> > > > process, not a kernel-resident module.  They already recognize there
> > > > is a difference here or
> > > > they wouldn't have the CAP_MAC_ADMIN check above in their patch.  The
> > > > problem with that
> > > > check is just that CAP_MAC_ADMIN doesn't necessarily mean fully
> > > > privileged with respect to
> > > > SELinux, which is why I want an explicit hook.  This gets a NAK from
> > > > me until there is such a hook.
> > >
> > > Doesn't the existing int (*bpf_prog)(struct bpf_prog *prog); cover
> > > SELinux's need here? I.e. it can already examine that a hook is being
> > > created for the LSM (since it has a distinct type, etc)?
> >
> > I was about to say the same, specifically for the BPF use-case, we do
> > have the "bpf_prog" i.e. :
> >
> > "Do a check when the kernel generate and return a file descriptor for
> > eBPF programs."
> >
> > SELinux can implement its policy logic for BPF_PROG_TYPE_LSM by
> > providing a callback for this hook.
> 
> Ok.  In that case do we really need the capable() check here at all?

We do not have a specific capable check for BPF_PROG_TYPE_LSM programs
now. There is a general check which requires CAP_SYS_ADMIN when
unprivileged BPF is disabled:

in kernel/bpf/sycall.c:

        if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
	        return -EPERM;

AFAIK, Most distros disable unprivileged eBPF.

Now that I look at this, I think we might need a CAP_MAC_ADMIN check
though as unprivileged BPF being enabled will result in an
unprivileged user being able to load MAC policies.

- KP