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;
Feedback-ID: i6a694271:Fastmail
Date:   Mon, 25 Jul 2022 14:23:45 -0500
From:   Daniel Xu <dxu@dxuuu.xyz>
To:     Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc:     Artem Savkov <asavkov@redhat.com>,
        Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Andrii Nakryiko <andrii@kernel.org>, bpf <bpf@vger.kernel.org>,
        Network Development <netdev@vger.kernel.org>,
        LKML <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>
Subject: Re: [PATCH bpf-next 1/4] bpf: add BPF_F_DESTRUCTIVE flag for
 BPF_PROG_LOAD
Message-ID: <20220725192345.jlrqyfktpmttiypp@fedora>
References: <20220720114652.3020467-1-asavkov@redhat.com>
 <20220720114652.3020467-2-asavkov@redhat.com>
 <CAADnVQ+mt1iEsXUGBeL-dgXRoRwPxoz+G=aRcZTkhx2AA10R-A@mail.gmail.com>
 <YtolJfvSGjSSwbc3@sparkplug.usersys.redhat.com>
 <CAADnVQLyCc7reM1By+TYBaNGh1SBpVqyNyT+WJXOooCqX_w2GA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAADnVQLyCc7reM1By+TYBaNGh1SBpVqyNyT+WJXOooCqX_w2GA@mail.gmail.com>
Precedence: bulk

On Thu, Jul 21, 2022 at 09:32:51PM -0700, Alexei Starovoitov wrote:
> On Thu, Jul 21, 2022 at 9:18 PM Artem Savkov <asavkov@redhat.com> wrote:
> >
> > On Thu, Jul 21, 2022 at 07:02:07AM -0700, Alexei Starovoitov wrote:
> > > On Wed, Jul 20, 2022 at 4:47 AM Artem Savkov <asavkov@redhat.com> wrote:
> > > >
> > > > +/* If BPF_F_DESTRUCTIVE is used in BPF_PROG_LOAD command, the loaded program
> > > > + * will be able to perform destructive operations such as calling bpf_panic()
> > > > + * helper.
> > > > + */
> > > > +#define BPF_F_DESTRUCTIVE      (1U << 6)
> > >
> > > I don't understand what value this flag provides.
> > >
> > > bpf prog won't be using kexec accidentally.
> > > Requiring user space to also pass this flag seems pointless.
> >
> > bpf program likely won't. But I think it is not uncommon for people to
> > run bpftrace scripts they fetched off the internet to run them without
> > fully reading the code. So the idea was to provide intermediate tools
> > like that with a common way to confirm user's intent without
> > implementing their own guards around dangerous calls.
> > If that is not a good enough of a reason to add the flag I can drop it.
>
> The intent makes sense, but bpftrace will set the flag silently.
> Since bpftrace compiles the prog it knows what helpers are being
> called, so it will have to pass that extra flag automatically anyway.
> You can argue that bpftrace needs to require a mandatory cmdline flag
> from users to run such scripts, but even if you convince the bpftrace
> community to do that everybody else might just ignore that request.
> Any tool (even libbpf) can scan the insns and provide flags.

FWIW I added --unsafe flag to bpftrace a while ago for
situations/helpers such as these. So this load flag would work OK for
bpftrace.

[...]
> Do you have other ideas to achieve the goal:
> 'cannot run destructive prog by accident' ?
>
> If we had an UI it would be a question 'are you sure? please type: yes'.
>
> I hate to propose the following, since it will delay your patch
> for a long time, but maybe we should only allow signed bpf programs
> to be destructive?

I don't have any opinion on the signing part but I do think it'd be nice
if there was some sort of opt-in mechanism. It wouldn't be very nice if
some arbitrary tracing tool panicked my machine. But I suppose tracing
programs could already do some significant damage by bpf_send_signal()ing
random processes.

Thanks,
Daniel