From mboxrd@z Thu Jan  1 00:00:00 1970
From: Tycho Andersen <tycho.andersen@canonical.com>
Subject: Re: [PATCH 6/6] ebpf: allow BPF_REG_X in src_reg conditional jumps
Date: Fri, 4 Sep 2015 16:43:50 -0600
Message-ID: <20150904224350.GU26679@smitten>
References: <1441382664-17437-1-git-send-email-tycho.andersen@canonical.com>
 <1441382664-17437-7-git-send-email-tycho.andersen@canonical.com>
 <20150904210619.GF1842@Alexeis-MacBook-Pro-2.local>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: Kees Cook <keescook@chromium.org>,
	Alexei Starovoitov <ast@kernel.org>,
	Will Drewry <wad@chromium.org>,
	Oleg Nesterov <oleg@redhat.com>,
	Andy Lutomirski <luto@amacapital.net>,
	Pavel Emelyanov <xemul@parallels.com>,
	"Serge E. Hallyn" <serge.hallyn@ubuntu.com>,
	Daniel Borkmann <daniel@iogearbox.net>,
	linux-kernel@vger.kernel.org, netdev@vger.kernel.org
To: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Return-path: <linux-kernel-owner@vger.kernel.org>
Content-Disposition: inline
In-Reply-To: <20150904210619.GF1842@Alexeis-MacBook-Pro-2.local>
Sender: linux-kernel-owner@vger.kernel.org
List-Id: netdev.vger.kernel.org

Hi Alexei,

On Fri, Sep 04, 2015 at 02:06:19PM -0700, Alexei Starovoitov wrote:
> On Fri, Sep 04, 2015 at 10:04:24AM -0600, Tycho Andersen wrote:
> > The classic converter generates conditional jumps with:
> > 
> > if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
> >         ...
> > } else {
> >         insn->dst_reg = BPF_REG_A;
> >         insn->src_reg = BPF_REG_X;
> >         insn->imm = fp->k;
> >         bpf_src = BPF_SRC(fp->code);
> > }
> > 
> > but here, we enforce that the src_reg == BPF_REG_0. We should also allow
> > BPF_REG_X since that's what the converter generates; this enables us to
> > load eBPF programs that were generated by the converter.
> 
> good catch. classic->extended converter is just being untidy.
> It shouldn't be populating unused 'src_reg' field when BPF_SRC == BPF_K
> verifier is doing the right thing. It's rejecting instructions that
> have junk in unused fields to make sure that someday we can extend it
> with something useful.
> The fix should be something like this:
> diff --git a/net/core/filter.c b/net/core/filter.c
> index 13079f03902e..05a04ea87172 100644
> --- a/net/core/filter.c
> +++ b/net/core/filter.c
> @@ -478,9 +478,9 @@ do_pass:
>                                 bpf_src = BPF_X;
>                         } else {
>                                 insn->dst_reg = BPF_REG_A;
> -                               insn->src_reg = BPF_REG_X;
>                                 insn->imm = fp->k;
>                                 bpf_src = BPF_SRC(fp->code);
> +                               insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
>                         }

Yep, I just tested this and it works for me. Do you want to manage it
or should I carry it as part of this set?

Tycho