Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220808094623.387348-1-asavkov@redhat.com> <20220808094623.387348-2-asavkov@redhat.com>
 <CAP01T76dELOx8p_iky_Py_VcqDbQtaL-4d=zrFiCbFhMdVEmNA@mail.gmail.com>
 <YvEEXsdo/fCnoEFY@samus.usersys.redhat.com> <CAP01T74kqdAeZbmnVA2uDRiB-8tjuWtdw-q_2V5fL6wQ==rTEA@mail.gmail.com>
In-Reply-To: <CAP01T74kqdAeZbmnVA2uDRiB-8tjuWtdw-q_2V5fL6wQ==rTEA@mail.gmail.com>
From:   Andrii Nakryiko <andrii.nakryiko@gmail.com>
Date:   Mon, 8 Aug 2022 17:37:34 -0700
Message-ID: <CAEf4BzaGmBZ7aXuX2ty1eB2jddurHyranxPs2cfKkkPB_hoX9A@mail.gmail.com>
Subject: Re: [PATCH bpf-next v3 1/3] bpf: add destructive kfunc flag
To:     Kumar Kartikeya Dwivedi <memxor@gmail.com>
Cc:     Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Andrii Nakryiko <andrii@kernel.org>, bpf@vger.kernel.org,
        netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
        Andrea Arcangeli <aarcange@redhat.com>,
        Daniel Vacek <dvacek@redhat.com>,
        Jiri Olsa <olsajiri@gmail.com>, Song Liu <song@kernel.org>,
        Daniel Xu <dxu@dxuuu.xyz>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Mon, Aug 8, 2022 at 6:33 AM Kumar Kartikeya Dwivedi <memxor@gmail.com> wrote:
>
> On Mon, 8 Aug 2022 at 14:41, Artem Savkov <asavkov@redhat.com> wrote:
> >
> > On Mon, Aug 08, 2022 at 02:14:33PM +0200, Kumar Kartikeya Dwivedi wrote:
> > > On Mon, 8 Aug 2022 at 11:48, Artem Savkov <asavkov@redhat.com> wrote:
> > > >
> > > > Add KF_DESTRUCTIVE flag for destructive functions. Functions with this
> > > > flag set will require CAP_SYS_BOOT capabilities.
> > > >
> > > > Signed-off-by: Artem Savkov <asavkov@redhat.com>
> > > > ---
> > > >  include/linux/btf.h   | 1 +
> > > >  kernel/bpf/verifier.c | 5 +++++
> > > >  2 files changed, 6 insertions(+)
> > > >
> > > > diff --git a/include/linux/btf.h b/include/linux/btf.h
> > > > index cdb376d53238..51a0961c84e3 100644
> > > > --- a/include/linux/btf.h
> > > > +++ b/include/linux/btf.h
> > > > @@ -49,6 +49,7 @@
> > > >   * for this case.
> > > >   */
> > > >  #define KF_TRUSTED_ARGS (1 << 4) /* kfunc only takes trusted pointer arguments */
> > > > +#define KF_DESTRUCTIVE  (1 << 5) /* kfunc performs destructive actions */
> > > >
> > >
> > > Please also document this flag in Documentation/bpf/kfuncs.rst.
> >
> > Ok, will do.
> >
> > > And maybe instead of KF_DESTRUCTIVE, it might be more apt to call this
> > > KF_CAP_SYS_BOOT. While it is true you had a destructive flag for
> > > programs being loaded earlier, so there was a mapping between the two
> > > UAPI and kfunc flags, what it has boiled down to is that this flag
> > > just requires CAP_SYS_BOOT (in addition to other capabilities) during
> > > load. So that name might express the intent a bit better. We might
> > > soon have similar flags encoding requirements of other capabilities on
> > > load.
> > >
> > > The flag rename is just a suggestion, up to you.
> >
> > This makes sense right now, but if going forward we'll add stricter
> > signing requirements or other prerequisites we'll either have to rename
> > the flag back, or add those as separate flags. I guess the decision here
>
> IMO we should do that when the time comes, for now it should reflect
> the current state.

But names should be also semantically meaningful, so KF_DESTRUCTIVE
explains that kfunc can do destructive operations, which is better
than just declaring that kfunc needs CAP_SYS_BOOT, as the latter is
current implementation detail which has no bearing on kfunc definition
itself.

Unless we anticipate we'll have another "destructive" kfunc not using
KF_DESTRUCTIVE and instead we'll add some other
KF_CAP_SYS_WHATEVERELSE?

> To me this helper requiring cap_sys_boot is just like how some
> existing stable helpers are gated behind bpf_capable or
> perfmon_capable.
> When it requires that the program calling it be signed, we can revisit this.
>
> > depends on whether some of non-destructive bpf programs might ever require
> > CAP_SYS_BOOT capabilities or not.
>
> These are just internal kernel flags, so refactoring/renaming is not a
> big deal when it is needed. E.g. we've changed just how kfuncs are
> registered twice since the support was added not long ago :).
>
> >
> > --
> >  Artem
> >