netdev - Re: [PATCH bpf-next 04/16] net: Add BPF_XDP_EGRESS as a bpf_attach

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200424005308.kguqn53qti26uvp6@ast-mbp.dhcp.thefacebook.com>
Date:   Thu, 23 Apr 2020 17:53:08 -0700
From:   Alexei Starovoitov <alexei.starovoitov@...il.com>
To:     Toke Høiland-Jørgensen <toke@...hat.com>
Cc:     David Ahern <dsahern@...il.com>, David Ahern <dsahern@...nel.org>,
        Network Development <netdev@...r.kernel.org>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Prashant Bhole <prashantbhole.linux@...il.com>,
        Jason Wang <jasowang@...hat.com>,
        Jesper Dangaard Brouer <brouer@...hat.com>,
        Toshiaki Makita <toshiaki.makita1@...il.com>,
        Daniel Borkmann <daniel@...earbox.net>,
        John Fastabend <john.fastabend@...il.com>,
        Alexei Starovoitov <ast@...nel.org>,
        Martin KaFai Lau <kafai@...com>,
        Song Liu <songliubraving@...com>, Yonghong Song <yhs@...com>,
        Andrii Nakryiko <andriin@...com>,
        David Ahern <dahern@...italocean.com>
Subject: Re: [PATCH bpf-next 04/16] net: Add BPF_XDP_EGRESS as a
 bpf_attach_type

On Fri, Apr 24, 2020 at 01:49:03AM +0200, Toke Høiland-Jørgensen wrote:
> Alexei Starovoitov <alexei.starovoitov@...il.com> writes:
> 
> > On Thu, Apr 23, 2020 at 07:05:42PM +0200, Toke Høiland-Jørgensen wrote:
> >> >> >
> >> >> > Looks like there is indeed a bug in prog_type_ext handling code that
> >> >> > is doing
> >> >> > env->ops = bpf_verifier_ops[tgt_prog->type];
> >> >> > I'm not sure whether the verifier can simply add:
> >> >> > prog->expected_attach_type = tgt_prog->expected_attach_type;
> >> >> > and be done with it.
> >> >> > Likely yes, since expected_attach_type must be zero at that point
> >> >> > that is enforced by bpf_prog_load_check_attach().
> >> >> > So I suspect it's a single line fix.
> >> >>
> >> >> Not quite: the check in bpf_tracing_prog_attach() that enforces
> >> >> prog->expected_attach_type==0 also needs to go. So 5 lines :)
> >> >
> >> > prog_ext's expected_attach_type needs to stay zero.
> >> > It needs to be inherited from tgt prog. Hence one line:
> >> > prog->expected_attach_type = tgt_prog->expected_attach_type;
> >> 
> >> Not sure I follow you here? I ended up with the patch below - without
> >> the first hunk I can't attach freplace funcs to an xdp egress prog
> >> (since the expected_attach_type will have been propagated from
> >> verification time), and so that check will fail. Or am I missing
> >> something?
> >> 
> >> -Toke
> >> 
> >> 
> >> 
> >> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> >> index d85f37239540..40c3103c7233 100644
> >> --- a/kernel/bpf/syscall.c
> >> +++ b/kernel/bpf/syscall.c
> >> @@ -2381,10 +2381,6 @@ static int bpf_tracing_prog_attach(struct bpf_prog *prog)
> >>                 }
> >>                 break;
> >>         case BPF_PROG_TYPE_EXT:
> >> -               if (prog->expected_attach_type != 0) {
> >> -                       err = -EINVAL;
> >> -                       goto out_put_prog;
> >> -               }
> >>                 break;
> >
> > ahh. that extra check.
> > I think it's better to keep it for extra safety.
> > Here all expected_attach_type have clear checks depending on prog_type.
> > There is no other place where it's that obvious.
> > The verifier does similar thing earlier, but it's not that clear.
> > I think the better fix would to set expected_attach_type = 0 for PROG_TYPE_EXT
> > at the end of do_check, since we're overriding this field temporarily
> > during verification.
> 
> OK, sure, can do. I do agree it's better to keep the check. I'll send a
> proper patch tomorrow, then.
> 
> As far as a adding a selftest for this, I think the most natural thing
> would be to add it on top of David's tests for xdp_egress, since that's
> what hit this - would you be OK with that? And if so, should I send the
> main patch straight away and hold off on the selftest, or should I split
> them, or hold off on the whole thing?

I think the issue is not related to xdp egress.
Hence I'd like to push the fix along with selftest into bpf tree.
The selftest can be:
void noinline do_bind((struct bpf_sock_addr *ctx)
{
  struct sockaddr_in sa = {};

  bpf_bind(ctx, (struct sockaddr *)&sa, sizeof(sa));
  return 0;
}
SEC("cgroup/connect4")
int connect_v4_prog(struct bpf_sock_addr *ctx)
{
  return do_bind(ctx);
}

and freplace would replace do_bind() with do_new_bind()
that also calls bpf_bind().
I think without the fix freplace will fail to load, because
availability of bpf_bind() depends on correct prog->expected_attach_type.

I haven't looked at the crash you mentioned in the other email related
to xdp egress set. That could be different issue. I hope it's the same thing :)