Received-SPF: pass (google.com: domain of linux-crypto-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20201207134309.16762-1-phil@nwl.cc>
In-Reply-To: <20201207134309.16762-1-phil@nwl.cc>
From:   Eyal Birger <eyal.birger@gmail.com>
Date:   Tue, 8 Dec 2020 16:47:02 +0200
Message-ID: <CAHsH6Gupw7o96e5hOmaLBCZtqgoV0LZ4L7h-Y+2oROtXSXvTxw@mail.gmail.com>
Subject: Re: [PATCH v2] xfrm: interface: Don't hide plain packets from netfilter
To:     Phil Sutter <phil@nwl.cc>
Cc:     Steffen Klassert <steffen.klassert@secunet.com>,
        linux-crypto@vger.kernel.org, netfilter-devel@vger.kernel.org,
        Linux Kernel Network Developers <netdev@vger.kernel.org>,
        Nicolas Dichtel <nicolas.dichtel@6wind.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Phil,

On Mon, Dec 7, 2020 at 4:07 PM Phil Sutter <phil@nwl.cc> wrote:
>
> With an IPsec tunnel without dedicated interface, netfilter sees locally
> generated packets twice as they exit the physical interface: Once as "the
> inner packet" with IPsec context attached and once as the encrypted
> (ESP) packet.
>
> With xfrm_interface, the inner packet did not traverse NF_INET_LOCAL_OUT
> hook anymore, making it impossible to match on both inner header values
> and associated IPsec data from that hook.
>

Why wouldn't locally generated traffic not traverse the
NF_INET_LOCAL_OUT hook via e.g. __ip_local_out() when xmitted on an xfrmi?
I would expect it to appear in netfilter, but without the IPsec
context, as it's not
there yet.

> Fix this by looping packets transmitted from xfrm_interface through
> NF_INET_LOCAL_OUT before passing them on to dst_output(), which makes
> behaviour consistent again from netfilter's point of view.

When an XFRM interface is used when forwarding, why would it be correct
for NF_INET_LOCAL_OUT to observe the inner packet?

What am I missing?

Thanks!
Eyal.